<%BANNER%>

Doppler Phase Modulation Effect for Non-contact Accurate Measurement of Vital Signs and Other Periodic Movements - From ...

Permanent Link: http://ufdc.ufl.edu/UFE0024294/00001

Material Information

Title: Doppler Phase Modulation Effect for Non-contact Accurate Measurement of Vital Signs and Other Periodic Movements - From Theory to CMOS System On Chip Integrations
Physical Description: 1 online resource (129 p.)
Language: english
Creator: Li, Changzhi
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: body, cancellation, circuit, cmos, complex, detection, dopper, integrated, movement, non, phase, radar, random, ray, sign, vital
Electrical and Computer Engineering -- Dissertations, Academic -- UF
Genre: Electrical and Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This dissertation begins with the spectral analysis of the non-linear Doppler phase modulation mechanism on which the non-contact vital sign detection is based. The harmonic and intermodulation issues caused by the non-linear Doppler phase modulation will be discussed in detail. Analyses will illustrate the method for optimal design of radio frequency and radio architecture. The theory will be verified by both numerical simulations and experiments using a Ka-band bench top system. Making use of the non-linear Doppler phase modulation, a new method and system for non-contact measurement of frequency and amplitude of mechanical vibration is invented in this chapter. With very simple radio architecture, this method does not need the calibration of signal amplitude and has the ability of self-verification. The non-contact Doppler radar measurement system has been integrated in both printed circuit board and CMOS integrated chip levels. The design, implementation, and characterization of the integrated radar system will be introduced in the subsequent chapters. As a main source of noise, the random body movement has been limiting the wide application of this technology. In Chapter 6, a random body movement cancellation technique will be introduced to cancel out the noise caused by random body movement. Both the complex signal demodulation and the arctangent demodulation will be studied for random body movement. Theoretical and experimental results will demonstrate the advantages of complex signal demodulation for random body movement cancellation. As an advanced signal processing method for Doppler radar non-contact vital sign detection, the RELAX algorithm will be discussed and used in experiment. It will be shown that the RELAX algorithm in general succeeds in mitigating the effects of the smearing and leakage problems of the periodogram caused by limited data length. Finally, an application case study of infant vital sign monitor will be presented.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Changzhi Li.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Lin, Jenshan.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024294:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024294/00001

Material Information

Title: Doppler Phase Modulation Effect for Non-contact Accurate Measurement of Vital Signs and Other Periodic Movements - From Theory to CMOS System On Chip Integrations
Physical Description: 1 online resource (129 p.)
Language: english
Creator: Li, Changzhi
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: body, cancellation, circuit, cmos, complex, detection, dopper, integrated, movement, non, phase, radar, random, ray, sign, vital
Electrical and Computer Engineering -- Dissertations, Academic -- UF
Genre: Electrical and Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: This dissertation begins with the spectral analysis of the non-linear Doppler phase modulation mechanism on which the non-contact vital sign detection is based. The harmonic and intermodulation issues caused by the non-linear Doppler phase modulation will be discussed in detail. Analyses will illustrate the method for optimal design of radio frequency and radio architecture. The theory will be verified by both numerical simulations and experiments using a Ka-band bench top system. Making use of the non-linear Doppler phase modulation, a new method and system for non-contact measurement of frequency and amplitude of mechanical vibration is invented in this chapter. With very simple radio architecture, this method does not need the calibration of signal amplitude and has the ability of self-verification. The non-contact Doppler radar measurement system has been integrated in both printed circuit board and CMOS integrated chip levels. The design, implementation, and characterization of the integrated radar system will be introduced in the subsequent chapters. As a main source of noise, the random body movement has been limiting the wide application of this technology. In Chapter 6, a random body movement cancellation technique will be introduced to cancel out the noise caused by random body movement. Both the complex signal demodulation and the arctangent demodulation will be studied for random body movement. Theoretical and experimental results will demonstrate the advantages of complex signal demodulation for random body movement cancellation. As an advanced signal processing method for Doppler radar non-contact vital sign detection, the RELAX algorithm will be discussed and used in experiment. It will be shown that the RELAX algorithm in general succeeds in mitigating the effects of the smearing and leakage problems of the periodogram caused by limited data length. Finally, an application case study of infant vital sign monitor will be presented.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Changzhi Li.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Lin, Jenshan.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024294:00001


This item has the following downloads:


Full Text

PAGE 1

DOPPLER PHASE MODULATION EFFECT FOR NON-CONTACT ACCURATE MEASRUEMENT OF VITAL SIGNS AND OTHER PERIODIC MOVEMENTS FROM THEORY TO CMOS SYSTEM ON CHIP INTEGRATIONS By CHANGZHI LI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009 1

PAGE 2

2009 Changzhi Li 2

PAGE 3

To my parents 3

PAGE 4

ACKNOWLEDGMENTS I would like to express my sincere gratitude to my advisor Dr. Jenshan Lin for his advice, encouragement, and mentoring throughout my Ph D study. I have truly en joyed doing research under his guidance over the years. Moreover, Dr Lins dedication to the service of the microwave society and kindness to other people ar e things I will follow throughout my life. I would also like to thank my co mmittee members, Dr. Jian Li, Dr William Eisenstadt, and Dr. Z. Hugh Fan for their time a nd precious comments. I am also thankful to my colleagues (Gabri el Reyes, Xiaogang Yu, Yan Yan, Chien-Ming Lee, Austin Chen, Mingqi Chen, Lance Covert, Ching-Ku Liao, Zivin Park, Zhen Ning Low, Raul Chinga, and Ashley Trowell) in the Radio Frequency Circuits and Systems Research Group, for all the help and happiness they offered. My sp ecial thanks also go to my friend Jun Ling and his advisor Dr. Jian Li for their great help on the signal processi ng side of my research. I would like to mention my thanks to the University of Florida Integrated Product and Process Design Program (IPPD). It was my great pleasure to collaborate with the IPPD students (Julie Cummings, Jeffrey Lam, Eric Graves, and Stephanie Jimenez) and their coach Ms. Wenhsing Wu. I would like to thank my parents for thei r encouragement and unconditional support. I dedicate this Dissertation to my family, w hose love gives me the courage to my life. 4

PAGE 5

TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........8 LIST OF FIGURES.........................................................................................................................9 LIST OF ABBREVIATIONS........................................................................................................12 ABSTRACT...................................................................................................................................13 1 INTRODUCTION................................................................................................................. .15 1.1 Background..................................................................................................................15 1.2 Recent Progresses on Non-contact Vital Sign Detection.............................................16 2 EXPERIMENT AND SPECTRAL ANALYSIS FOR OPTIMUM DESIGN OF RADIO FREQUENCY AND RADIO ARCHITECTURE..................................................................19 2.1 Introduction..................................................................................................................1 9 2.2 Measurement from Four Sides of a Human Body.......................................................20 2.3 Spectral Analysis.........................................................................................................24 2.3.1 Spectral Analysis of Doppler Radar Sensor.......................................................25 2.3.2 Sensitivity..........................................................................................................25 2.3.3 Harmonic Interference.......................................................................................26 2.3.4 Residual Phase Optimum/Null Points Manipulation......................................29 2.4 Simulation....................................................................................................................3 1 2.4.1 Spectrum............................................................................................................31 2.4.2 Optimum Choice of Carrier Frequency.............................................................32 2.4.2.1 Absolute detected heartbeat strength........................................................36 2.4.2.2 Detected heartbeat compared with harmonics..........................................38 2.4.2.3 Detected heartbeat compared with intermodulation tone.........................38 2.4.2.4 Summary...................................................................................................40 2.5 Experiment...................................................................................................................40 2.5.1 Observation of Harmonics.................................................................................41 2.5.2 Even Order Harmonics Dominant Null Point.................................................42 2.5.3 Odd Order Harmonics Dominant Optimum Point..........................................42 2.6 Conclusion...................................................................................................................43 3 A NEW METHOD AND SYSTEM FOR NON-CONTACT MEASUREMENT OF FREQUENCY AND AMPLITUDE OF MECHANICAL VIBRATION..............................44 3.1 Introduction..................................................................................................................4 4 3.2 Theory and Analysis....................................................................................................45 3.3 Experimental Results...................................................................................................48 5

PAGE 6

3.4 Wide angle Incidence and the Effect on Accuracy......................................................50 3.5 Conclusion...................................................................................................................53 4 SYSTEM INTEGRATION ON BOARD LEVEL.................................................................54 4.1 Introduction..................................................................................................................5 4 4.2 Circuit Design and Implementation.............................................................................54 4.3 Measuremen t Results...................................................................................................56 5 SYSTEM INTEGRATION ON CHIP LEVEL......................................................................58 5.1 Introduction..................................................................................................................5 8 5.2 A 5 GHz Indirect Conversion Double Sideband Transmission Radar Sensor.............58 5.3.1 CMOS 5 GHz Indirect Conve rsion Radar Transceiver.....................................58 5.3.2 Baseband Circuit................................................................................................62 5.3.3 Antenna..............................................................................................................63 5.3.4 Experimental Results.........................................................................................63 5.3 A 5 GHz Direct Conversion Softwa re Configurable Radar Sensor.............................65 5.3.1 CMOS 5 GHz Direct Conversion Radar Receiver............................................65 5.3.2 System Design Considerations..........................................................................66 5.3.3 Detailed Circuit Design......................................................................................69 5.3.3.1 LNA and preamplifier...............................................................................69 5.3.3.2 Mixers........................................................................................................71 5.3.3.3 Variable gain amplifiers............................................................................71 5.3.3.4 Bias circuits...............................................................................................72 5.3.4 Experiments.......................................................................................................74 5.4 Conclusion...................................................................................................................77 6 COMPLEX SIGNAL DEMODULATION AND RANDOM BODY MOVEMENT CANCELLATION IN DOPPLER RA DAR VITAL SIGN DETECTION............................78 6.1 Introduction..................................................................................................................7 8 6.2 Complex Signal Demodulation and Arctangent demodulation...................................80 6.2.1 Complex Signal Demodulation..........................................................................82 6.2.2 Arctangent demodulation...................................................................................84 6.3 Effects of Phase Offset.................................................................................................86 6.4 Simulation....................................................................................................................8 8 6.4.1 Ray-tracing Model.............................................................................................88 6.4.2 Demodulation without Random Body Movement.............................................89 6.4.2.1 Example I: 5.8 GHz quadrature radar.......................................................89 6.4.2.2 Example II: 24 GHz quadrature radar.......................................................91 6.4.3 Random Body Movement Cancellation.............................................................93 6.5 Experiment...................................................................................................................96 6.5.1 DC Offset Estimation in Baseband....................................................................98 6.5.2 Random Body Movement Cancellation...........................................................100 6.6 Conclusion.................................................................................................................103 6

PAGE 7

7 ADVANCED SPECTRAL ESTIMATION ALGORITHM FOR DOPPLER RADAR NON-CONTACT VITAL SIGN DETECTION...................................................................104 7.1 Introduction................................................................................................................104 7.2 Spectral Estimation Using Periodogram....................................................................106 7.2.1 Problem Formulation.......................................................................................106 7.2.2 Periodogram.....................................................................................................108 7.3 Spectral Estimation Using RELAX...........................................................................110 7.4 Numerical and Experimental Results.........................................................................112 7.4.1 Example I.........................................................................................................113 7.4.2 Example II........................................................................................................116 7.5 Conclusion.................................................................................................................118 8 APPLICATION CASE STUDY INFANT VITAL SIGN MONITOR.............................119 8.1 Introduction................................................................................................................119 8.2 System Architecture...................................................................................................120 8.3 Features......................................................................................................................1 21 8.4 Conclusion.................................................................................................................123 9 SUMMARY..........................................................................................................................124 LIST OF REFERENCES.............................................................................................................125 BIOGRAPHICAL SKETCH.......................................................................................................129 7

PAGE 8

LIST OF TABLES Table page 2-1 Summary of heart ra te detection accuracy.........................................................................23 3-1 Detection range for diffe rent pairs of harmonics...............................................................48 4-1 Building blocks and specifications used in the 4-7 GHz radar..........................................55 5-1 System specifications (power referred to 50 Ohm)...........................................................68 7-1 The RELAX algorithm....................................................................................................111 8

PAGE 9

LIST OF FIGURES Figure page 2-1 Block diagram of the Ka-band Doppler-radar system using double side-band transmission................................................................................................................... ....21 2-2 Heart-rate comparis on at 2 m distance...............................................................................22 2-3 Normalized spectrum comparison at 2 m dist ance for the front case and the back case under power lever of 350 W............................................................................................23 2-4 Bessel coefficients........................................................................................................ .....27 2-5 Surface formed by the values of J4(4 mr/ )/J1(4 mh/ ) as a function of mr and mh..........28 2-6 Simulated normalized spectrum comparison.....................................................................32 2-7 Spectral intensity of breathing fundamental, breathi ng harmonics, and heartbeat fundamental........................................................................................................................33 2-8 Simulated spectrum of baseband signal.............................................................................35 2-9 Amplitude of detected heartbeat signal versus the carrier frequency................................37 2-10 Relative strength of detected heartbeat signal compared with the third order harmonic as the carrier frequency changes........................................................................38 2-11 Relative strength of detected heartbeat si gnal compared with the intermodulation as the carrier frequency changes............................................................................................39 2-12 Measured and calculated baseband spectrum when residual phase is found to be 51.770.................................................................................................................................41 2-13 Measured baseband spectrum when ev en order harmonics dominate, corresponding to a null detection point......................................................................................................42 2-14 Measured baseband spectrum when odd order harmonics dominate, corresponding to an optimum detection point...............................................................................................43 3-1 Block diagram and experimental setup..............................................................................45 3-2 Theoretical harmonic ratio as a f unction of the movement amplitude..............................47 3-3 Detected baseband signal and spectrum when carrier frequency is 40 GHz.....................49 3-4 Method to obtain the displacement and to check the validity of the measurement...........49 3-5 Baseband spectrum detected with the target placed 0.25m in front of the antenna...........51 9

PAGE 10

3-6 Radiation pattern of the horn antenna used in experiment and a 10 antenna array.....52 3-7 Measurement error of 3rd order to 1st order ratio and 4th order to 2nd order ratio by different antennas...............................................................................................................52 4-1 The 4-7 GHz vital sign detector.........................................................................................55 4-2 Measurement results of complex demodulation................................................................57 5-1 Block diagram of the 5 GHz radar tran sceiver with the input/output interface.................59 5-2 Simplified circuit diagram of the 5 GHz radar transceiver................................................61 5-3 Chip microphotograph of th e 5 GHz radar transceiver......................................................62 5-4 Baseband circuit for signal amplification and differential to single-ended transformation....................................................................................................................62 5-5 Transmitted spectrum when RF-VCO tuned at the highest frequency and the lowest frequency............................................................................................................................64 5-6 Detected baseband spectrum..............................................................................................65 5-7 Block diagram of the software conf igurable 5.8 GHz radar sensor chip...........................67 5-8 Schematic of the receiver chain.........................................................................................70 5-9 Bandgap bias circuit for the mixer core and mixer buffer.................................................72 5-10 Constant-Gm bias circuit for the variable gain amplifier..................................................73 5-11 Schematic of the Gm-b oosted bias circuit.........................................................................73 5-12 Chip microphotograph of the 5.8 GHz radar sensor receiver............................................74 5-13 Experiment setup for non-contact vital sign detection......................................................75 5-14 Detection from the back of a human subject at 0.5m away...............................................76 5-15 Detection from the front of a human subject at 1.5m away...............................................76 6-1 Simplified block diagram for Doppler radar non-contact vital sign detection..................80 6-2 Demodulation block diagram.............................................................................................83 6-3 Ray-tracing model for 5.8 GHz radar application.............................................................87 6-4 Ray-tracing model for ra ndom body movement cancellation............................................88 10

PAGE 11

6-5 Demodulation for a 5.8 GHz radar....................................................................................90 6-6 Demodulation for a 24 GHz radar.....................................................................................92 6-7 Baseband spectrum detected by the I and the Q channels with a carrier frequency of 24 GHz...............................................................................................................................92 6-8 Baseband spectra obtained when random body movement is present...............................94 6-9 Random body movement cancellation us ing the two demodulation techniques...............95 6-10 Random body movement cancellation technique..............................................................96 6-11 Two identical transceivers used for random body movement cancellation.......................97 6-12 DC offset estimation...................................................................................................... ....99 6-13 Time domain signals....................................................................................................... .101 6-14 Random body movement cancellation using arctangent demodulation..........................101 6-15 Random body movement cancellation using complex signal demodulation...................102 7-1 Four complex-valued sinusoids with frequencies of -0.115, -0.1, 0.35 and 0.412 Hz....109 7-2 Experiment setup with a 4-7 GHz radar, a data acquisition module, and a laptop for signal processing.............................................................................................................. 112 7-3 A set of 12.85 seconds' baseband signal measured by the 4-7 GHz radar when detecting from the back of the human body.....................................................................113 7-4 Some intermediate stages are show n when applying RELAX for analyzing the baseband signal spectra....................................................................................................114 7-5 Final result of RELAX es timation to measured baseband signal when detecting from the back of the human body.............................................................................................115 7-6 Ray-tracing model.......................................................................................................... ..116 7-7 RELAX estimation to computer-generated baseband signal when detecting from the front of the human body...................................................................................................117 8-1 Infant monitoring system.................................................................................................119 8-2 Block diagram of the implemented infant monitoring system hardware.........................120 8-3 The prototype system designed and fabricated................................................................122 11

PAGE 12

LIST OF ABBREVIATIONS f Frequency fc Carrier frequency VCO Voltage controlled oscillator CW Continuous wave TX Transmitter RX Receiver RF Radio frequency IF Intermediate frequency LO Local oscillator LNA Low noise amplifier BPF Band-pass filter FFT Fast Fourier transform AT Attenuator BPM Beats per minute SP4T Single-pole four-throw RBMC Random body movement cancellation ML Maximum likelihood 12

PAGE 13

Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy DOPPLER PHASE MODULATION EFFECT FOR NON-CONTACT ACCURATE MEASRUEMENT OF VITAL SIGNS AND OTHER PERIODIC MOVEMENTS FROM THEORY TO CMOS SYSTEM ON CHIP INTEGRATIONS By Changzhi Li August 2009 Chair: Jenshan Lin Major: Electrical and Computer Engineering This dissertation begins with the spectral analysis of the non-linear Doppler phase modulation mechanism on which the non-contact vi tal sign detection is based. The harmonic and intermodulation issues caused by the non-linear D oppler phase modulation will be discussed in detail. Analyses will illustrate the method fo r optimal design of radio frequency and radio architecture. The theory will be verified by both numerical simulations and experiments using a Ka-band bench top system. Making use of the non-linear Doppler phase modulation, a new method and system for non-contact measurement of frequency and amplitude of mechanical vibration is invented in this chapter. With very simple radio architecture, this method does not need the calibration of signal amplitude and has the ability of self-verification. The non-contact Doppler radar measurement syst em has been integrated in both printed circuit board and CMOS integrated chip levels. The design, implementation, and characterization of the integrated radar system will be introduced in the subsequent chapters. As a main source of noise, the random body movement has been limiting the wide application of this technology. In Chapter 6, a random body movement cancellation technique 13

PAGE 14

will be introduced to cancel out the noise cau sed by random body movement. Both the complex signal demodulation and the arctangent dem odulation will be st udied for random body movement. Theoretical and experimental results will demonstrate the advantages of complex signal demodulation for random body movement cancellation. As an advanced signal processing method for Doppler radar non-contact vital sign detection, the RELAX algorithm will be discussed and used in experiment. It will be shown that the RELAX algorithm in general su cceeds in mitigating the effect s of the smearing and leakage problems of the periodogram caused by limited data length. Finally, an application case study of infa nt vital sign monitor will be presented. 14

PAGE 15

CHAPTER 1 INTRODUCTION 1.1 Background Using microwave Doppler radar to detect physiological movements can be traced back to the early 1970s [1]. The microwave sensing syst ems [1]-[10] transmit a radio frequency, singletone continuous-wave (CW) signal, which is refl ected off a target and then demodulated in the receiver. CW radar with the human body as the target will receive a signal with its phase modulated by the time-varying chest-wall positio n. Demodulating the phase will then give a signal proportional to the chest-wall position that contains information about movement due to heartbeat and respiration. This technique enable d non-contact detection of vital signs of humans or animals from a distance away, with out any sensor attached to the body. There are several advantages to a non-contact solution: physi cally, it neither confines nor inhibits the subject, and does not cause discomfort or skin irritation like methods using electrodes or straps. This is especially impor tant over extended periods of time, making the detector ideal for long-term continuous monitori ng applications. Also, the reliability can be increased since the patient is unaware of the me asurement and therefore is less likely to alter their respiration. Additionally, accuracy is enhanc ed due to the lack of surface loading effects that have been shown to reduce the accuracy of some other measurement methods. Compared to either infrared or visible light, microwave has greater penetration capability through the building materials, which brings unique property to many civilian and military applications. This technology will most likely have an impact on the growing home healthcare monitoring market, such as the diagnoses and obs ervation of sleep apnea and other respiratory ailments, as well as infant monitoring for resp iratory or cardiac concer ns. Detecting a persons presence by measuring heartbeat, this technology is also a good candidate for applications such 15

PAGE 16

as search-and-rescue for earthquake or fire victim s, border patrol, and entrance security. It has also drawn interests from departments requiring high security such as see-through-wall radar and airport TSA monitoring. 1.2 Recent Progresses on Non-contact Vital Sign Detection Recent developments in the early 2000s have demonstrated the feasib ility of integrating this function into modern wireless communicatio n devices operating in L and S bands [6]. The first integrated vital sign radar sensor chip using silicon CMOS was also demonstrated [7]. The chip integrates all the radio frequency (RF) ci rcuits including a free-r unning oscillator that provides transmission signal and se rvers as the reference [8]. Recently, the detection of vital signs using higher microwave frequency near millimeterwave was proposed [9]. It shows the advantag e of increasing sensitiv ity while reducing the transmitted power. An architecture using double-sideband transmission and detection was also proposed [10]. It replaces the quadr ature receiver architecture prev iously used to alleviate nullpoint detection problem. By using these two ke y techniques, a system was built and several experiments were conducted. The results show imp rovements in detection range and accuracy. It also shows the improved sensitivity to detect small physiological movements. In a demonstration, the system was able to detect bot h heartbeat and respirat ion signals from four sides of a human body [11]. Surpri singly, the detection of heartb eat from the back of human body shows the best accuracy among all four cases. Based on this finding, an experiment of measuring a sleeping subject's vital signs overnig ht was carried out by pl acing the radar sensor on the ground underneath the bed [ 12]. These results were further analyzed and an interesting phenomenon of nonlinear Doppler phase modula tion was discovered. A theoretical model was developed to explain the results [ 13]. The model can be used to pr edict the best carrier frequency for vital sign detection, which depends on th e body type of subject under test [14]. 16

PAGE 17

The theoretical model brought another impor tant finding. The nonlinear Doppler phase modulation produces harmonics of the periodic movement being de tected. The signal levels of these harmonics maintain a fixed ratio as a func tion of the absolute amp litude of the periodic movement, regardless of the overall signal streng th received. This suggests that an accurate detection of periodic movement amplitude without calibration can be achieved in an inexpensive way [15]. The chest movements due to heartbeat a nd respiration can also be estimated using this technology. Continuing from the pioneering integration wo rks in the early 2000s, more circuits are designed and fabricated on both pr inted circuit board level [16] and integrated microchip level [17], with different level of in tegration featuring different appl ications. Achievements include successful integration of the radio front-end and th e baseband circuit together on a single printed circuit board, as well as the in tegration of the double sideband radio architecture on low cost CMOS chip [18]. The integrated systems and circuits have been used for various experiments to support the theoretical research. In the meantime, advanced models have been constructed for numerical simulation, illustrati ng the design criteria for optim al radio frequency and radio architecture [16]. As one of the main challenges for non-contact vital sign detection, the noise caused by the random body movement presents severe interference for accurate detection of respiration and heartbeat signal in practical applications. Sin ce random body movement is comparable or even stronger than the weak vital sign signal, to so me extent it is the main factor limiting broad applications of non-contact vital sensors. It has been shown recen tly that the different movement patterns of random body movement and physiological movement make it possible to remove the unwanted signal from the vita l signs [20]. The random body m ovement cancellation technique 17

PAGE 18

uses multiple antennas and transceivers to detect from the front and the back of the human body. Based on polarization and frequency multiplexi ng, signals detected from different body orientations are combined without interfering each other and noise caused by random body movement can be cancelled out. This dissertation will present the works on th e spectral analysis of the non-linear Doppler phase modulation effect, the new method of m easuring both the amplit ude and frequency of periodic movements based on non-linear phase modulation, the printed ci rcuit board level and the CMOS integrated chip leve l integrations of this tec hnology, the random body movement cancellation technology, and the RELAX spectral es timation for Doppler ra dar non-contact vital sign detection. 18

PAGE 19

CHAPTER 2 EXPERIMENT AND SPECTRAL ANALYSIS FOR OPTIMUM DESIGN OF RADIO FREQUENCY AND RADIO ARCHITECTURE 2.1 Introduction Recently, a Ka-band bench top system using double-sideband transmission was reported [9]. It has been shown that the short wavelength of the Ka-band sensor is more sensitive to small chest-wall motion. Meantime, double-sideband transm ission can avoid severe null point problem at short-wavelength without usi ng quadrature demodula tion. The system configuration is also simplified without using image-reject filters [ 10]. Using the Ka-band detector, measurements have been performed using different transmitting pow er levels at different distances from four sides of a body [11], showing that this detector achieved high detection accuracy. It is also found that measuring from the back of the body gives better accuracy than measuring from all other sides. While showing advantages over previously re ported lower frequency single-tone sensing systems, the Ka-band physiological movement sensor brings a few new questions to be answered: Firstly, as the frequency moves into Ka-band with wavelength in centimeter or even millimeter range, chest-wall movement due to resp iration may go out of the range of small-angle approximation [8], rendering the previously well-adopted model to no avail. Thus how to model the radar system more properly to guide future design? Secondly, as observed in experiments, while its very easy to accurately detect he artbeat when holding the breath, sometimes the detection accuracy seems to be reversely affect ed when breathing is added. How to explain this and how to avoid this phenomenon? Thirdly, what is the reason for bette r heart-rate accuracy when detected from the back of the body? In this chapter, in addition to pres enting the results measured on human body under different experimental conditions a rigorous spectral analysis of Doppler radar sensing of 19

PAGE 20

physiological movement has been performed. While in accordance with the results obtained from the simple small-angle approximation model [8]-[ 10] for low frequency operation, the theory is readily to be used for more general cases su ch as high frequency operation with large-angle phase modulation. Based on the theory to be pres ented, the aforementioned questions have been answered. The theory also explains from a spectral viewpoint the reason double-sideband transmission can be used to increase the de tection accuracy. Furthermore, harmonic effects caused by the physical nature of phase modulat ed Doppler radar are analyzed in detail. Numerical simulations by MATLAB and system -level simulations by Agilent ADS were performed to illustrate the theory and provide design tips for Ka-band sensors. The harmonics and the null/optimum point were observed and verified by experiment. 2.2 Measurement from Four Sides of a Human Body The block diagram of the Ka-band physiological movement detector and the orientation of the body in measurement are illustrated in Fig. 2-1. The transmitter chain contains a transmitting antenna (labeled as TXA) and an up-converter. The receiver chain includes a receiving antenna (RXA), a low noise amplifier (LNA), two down-conve rters, and an IF am plifier (IF_AMP). The baseband circuit contains a preamplifier (PreAMP), a band pass filter (BPF), and a low frequency amplifier (LF_AMP). Two 3 dB power splitters are used to divide the power generated by the radio frequency local oscillator (RF_LO) and the intermediate frequency local oscillator (IF_LO), with half of the power sent to the transmitter chain and the other half sent to the receiver chain. The output of the transmitting antenna has two main frequency components: lower sideband (LSB) fL = f2 f1 and upper sideband (USB) fU = f2 + f1. The LSB and USB frequencies are 26.54 GHz and 27.66 GHz ( f2 = 27.1 GHz and f1 = 560 MHz) with combined power being either 350 W or 14.2 W. The power is switched from 350 W to 14.2 W when an attenuator 20

PAGE 21

(AT) is inserted in the transmitter chain between the intermediate frequency local oscillator and the up-converter. The components of the transceiver are all operating in the linear region, so that the received signal would not cause any compressi on in any stage to produce strong harmonics that will be observed in experiment. Therefor e, the strong harmonics sometimes observed in experiment must be due to some other reason. DAQ IF_AMPLNARF Transceiver PowerSplitterPowerSplitter LF_AMPBPFPreAMP BasebandIF_LORF_LO TXA RXA Reference1 f 2 f case Right case Leftcase Frontcase Back AT Figure 2-1. Block diagram of the Ka-band Doppler-radar system using double side-band transmission. The target has four different orientations. The remote detection of physiological moveme nt was measured from four sides of the body, as indicated in Fig. 2-1. The four measurement angles are defined as the front back, left, and right cases. The subject, breathing normally, was s eated at a distance away from the antenna. A wired fingertip pulse sensor (UFI_1010 pulse transducer) was attached to the index finger during the measurement to provide the reference heartbeat. The e xperimental conditions were designed as combinations of the following pa rameters: two power le vels of 350 W and 14.2 W; five different distances from the antenna: 0. 5 m, 1 m, 1.5 m, 2 m, and 2.5 m; and measuring from four sides of the body. The signal processing part is similar to that in [9] [10]. The heartbeat signal was first separated from the respiration signal by a Butterwor th band-pass filter (BPF) with passband from 0.7-Hz to 3-Hz. The filtered signal was then wind owed and auto-correlated. After that, fast 21

PAGE 22

Fourier transform (FFT) was applied to the auto-correlated signal to obtain the heartbeat rate. Finally, the measured heartbeat rate was evaluate d by heart-rate accuracy. Heart-rate accuracy is calculated as the percentage of time the detected rate is within 2% of the reference rate [21] [22]. As an example, Fig. 2-2 shows 27 sec onds of heart-rate measured in the front case and in the back case at 2-m distance using 350-W power. The black solid curve shows the detected heartbeat rate in beats per minute (BPM), the gr ey solid curve shows the referenced heartbeat rate in BPM. Two grey dotted lines show the upper and lower lim its of the acceptable heartbeat rate, which is 2% variation from the referenc ed heartbeat rate. The re gion defined by the two limits is called the confidence interval When the detected heart rate falls into this confidence interval it is considered accurate. 0 5 10 15 20 25 80 85 90 95 Time (Sec)Beats/Min (Back) 0 5 10 15 20 25 80 85 90 95 Beats/Min (Front) Detected heart beat Reference heart beat Detected heart beat Reference heart beat 2% Higher than Reference 2% Lower than Reference 2% Higher than Reference 2% Lower than Reference (a) (b) A B Figure 2-2. Heart-rate comparis on at 2 m distance. The output pow er of the detector is 350 W. A) Measured in the front case. B) Measured in the back case. The measured results of the heart-rate accur acy for all the combinational experimental conditions are listed in Table 21. In experiment, the 27-second de tection accuracy from any side of the body and at any of the 5 tested distances is better than 80%. In addition, the measurement 22

PAGE 23

in the back case shows the best performance. The result s also indicate that better accuracy can be achieved with higher power, as expected. Table 2-1. Summary of heart rate detection accuracy Distance (m) Front Left Right Back 14.2 W 0.5 99.1% 96.3% 100% 97.6% 1 89.8% 89.8% 93.2% 100% 1.5 98.9% 89% 93.8% 94.3% 2 85.2% 80.5% 97.4% 93.6% 2.5 83.3% 85.7% 85.1% 85.5% 350 W 0.5 100% 100% 100% 100% 1 94.8% 94.7% 93.2% 100% 1.5 98.1% 97.6% 100% 100% 2 100% 100% 100% 100% 2.5 95.1% 100% 95.2% 97.2% 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 Beats/MinNormalized SpectrumBreathing Fundamental Breathing 2nd Harmonic Breathing 3rd Harmonic Heartbeat A 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 Beats/MinNormalized SpectrumBreathing Fundamental Breathing 2nd Harmonic Breathing 3rd Harmonic Heartbeat B Figure 2-3. Normalized spectrum comparison at 2 m distance for the front case and the back case under power lever of 350 W. A) Front cas e. B) Back case. (Insets: corresponding real time signals, with time span of 27 seconds). 23

PAGE 24

The unexpected result that the heart-rate accuracy is better in the back case than in other cases is related to the spectra of different cases. For example, Fig. 2-3 shows the normalized spectra and the baseband signals detected in the front case and in the back case at 2 m distance and under the power level of 350 W It is observed that besides the respiration and heartbeat tones, other tones exist in the spectra. Those frequency components, known as harmonics, are relatively stronger in the front case than in the back case. In the next section, we will apply spectral analysis to explain the cause of those harmonics and their effects on detection accuracy. 2.3 Spectral Analysis Consider one tone of the double-sideband signa l first. Neglecting amplitude variations, each tone transmitted by the CW radar is: ()cos(2()) Ttftt (2-1) where the transmitting frequency f is either fL or fU, t is the elapsed time, and (t) is the phase noise of the oscillator. If this signal is reflected by a target at a nominal distance d0 with a timevarying displacement given by x ( t ), the total distance traveled between the transmitter and receiver is 2 d( t ) = 2 d0 + 2x ( t ). Therefore, the received signal can be approximated as: 04 4() ()cos2 () d xt Rtft t c 02 d (2-2) where c is the signals propagation velocity (the speed of light), is the signals wavelength in air, which equals to c / f The baseband signal B ( t ) after two-step down-conversion is approximated as 4() ()cos () xt Bt t (2-3) where is the constant phase shift due to the distance to the target d0 and reflections at the surface, and)( t is the total residual phase noise [8]. 24

PAGE 25

2.3.1 Spectral Analysis of Doppler Radar Sensor From the theory of Fourier series, any time-varying peri odic displacement x ( t ) can be viewed as the combination of a series of single-t one signals. Therefore, for the ease of analysis and without loss of generality, x ( t ) is assumed to be single tone, i.e. x ( t ) = m sin t In this case, the detected phase-modulated si gnal can be represented as: 4sin() ()4sin() ()cos () Reee mt j jtmt B tt (2-4) The exponential term can be expand ed using Fourier series [23]: 4sin()4 e() e mt j j nt n nm J (2-5) where Jn( x ) is the nth order Bessel function of the first kind. Therefore, the Fourier series representation of the phase modulated signal in (2-4) is: ()4 ()Re()ee 4 ()cos() jt jnt n n n nm BtJ m Jnt (2-6) where () t is the total residual phase. Based on the above Fourier expansion, the phasemodulated baseband signal is decomposed into frequency components with n times the basic frequency of the periodic moveme nt. The baseband signal can thus be analyzed in frequency domain. 2.3.2 Sensitivity The body movement x ( t ) may consist of xr( t ) due to respiration and xh( t ) due to heartbeat. Since usually xr( t ) = mrsin rt is much larger than xh( t ) = mhsin ht the sensitivity is mainly the issue for heartbeat detection. Equation (2-6) show s that the sensitivity for heartbeat detection, 25

PAGE 26

dependent on the frequency component with n = 1, is decided by the value of J1(4 mh/ ), where mh is the amplitude of xh( t ). For body movement due to heartbeat, the amplitude mh is normally in the range of 0.01 mm [24], which corresponds to a very small value of 4 mh/ for transmitting frequency up to 50 GHz (e.g., for f = 50 GHz and mh = 0.01 mm, = 6mm, then 4 mh/ =0.02), thus a linear approximation of the Bessel function is applicable: 1 (),1. 2!n n nJxxx n (2-7) When x is small, Jn( x ) decreases rapidly as n increases. If we take n = 1 into consideration, the expression of the received baseband signal is es sentially equivalent to the widely-used smallangle approximation [8] for low radio frequency radars. Based on the small signal linear approximation in (2-7), the h eart-rate detect ion sensitivity is proportional to 2 mh/ which is in turn proportional to the working frequency. Therefore, more sensitivity is gained as we increase the frequency. This is the reas on for our interest to increase the transmitting frequency to Ka-band. 2.3.3 Harmonic Interference For respiration, it can be observed direct ly from chest wall movement that xr( t ) can be as large as several millimeters. Taking mr = 1.5 mm and transmitting frequency f = 20 GHz for example, the corresponding wavelength is = 15mm, and 4 mr/ = 1.26, which is out of the safety range for linear approximation, or, equiva lently, the small-angle approximation. In this case, a good estimation of the rece ived baseband spectrum is to directly investigate the Bessel coefficients, which are plotted in Fig. 2-4. 26

PAGE 27

0 2 4 6 8 -0.5 0 0.5 1 m/radJn(m) n = 1 n = 2 n = 3 n = 4 n = 5 n = 0 Figure 2-4. Bessel coefficients. Investigating (2-6) with the he lp of Fig. 2-4, we can see that although the periodic signal x ( t ) to be detected has a single tone 0, the actually received baseband signal B ( t ) may contain frequency components at any harmonic position of 0, i.e. = n0. For the convenience of discussion, we will refer the term with n = in (2-6) as fundament al, since they contains the signal at the desired frequency. We refer other terms with |n| 2 as the n th order harmonic. For n = 0, it corresponds to DC in baseband and will be neglected in the following discussion since it can be removed by baseband filter. The amplitude of the nth order harmonic is determined by the nth order Bessel function, with 4 mr/ as the variable. Therefore, when two-tone signals exist simultaneously, the phase modulation nature of Doppler radar sensor brings inevitably another effect which is not desired, i.e. harmonic inte rference generated by the signal with a lower frequency and larger amplitude. This effect plays a destructive role when any of the harmonic coincides with the desired signal with a higher frequency, especially when the latter is weaker compared with the former. 27

PAGE 28

1 1.5 2 0.05 0.1 0.15 0 0.5 1 1.5 2 m r (mm) mh (mm) J4(4mr/) / J1(4mh/)0.2 0.4 0.6 0.8 1.0 1.2 1.41.6 414 4 ()/()h rm m JJ (m i l l i m e t e r ) r m ( m i l l i m e t e r ) h m Figure 2-5. Surface formed by the values of J4(4 mr/ )/J1(4 mh/ ) as a function of mr and mh. A contour plot is drawn beneath the surface on the z = 0 plane. Inset: frequency allocation of the breathing fundamental, th e second to the forth order harmonics (spikes with circular end), and the heartbeat fundamental (spike with arrow end). As an illustration of this scenario, the inset of Fig. 2-5 plots an example of the frequency allocation of the breathing fundamental, its seco nd to forth order harmonics, as well as the heartbeat fundamental. Since normally the frequency of the forth order breathing harmonic is most likely to be close to that of the h eartbeat fundamental, the relative strength of J4(4 mr/ ) to J1(4 mh/ ) is used as an estimation of the harm onics effect. In Fig. 2-5, the value of J4(4 mr/ )/J1(4 mh/ ) is calculated for the 27 GHz system, as mr varies from 1mm to 2 mm and mh varies from 0.05 mm to 0.15 mm. A contour plot of this value is also drawn beneath the 3-D plot on the z = 0 plane. In Fig. 2-5, as mr increases and mh decreases, the ratio of the forth order breathing harmonic to the heartbeat fundamental becomes larger, which means harmonic interference is more likely to occur. To reduce the impact of harmonic interferen ce, it is desirable to reduce the amplitude mr or not to use very high transmitting frequency. By comparing the measured spectra in the front case and in the back case, it is seen that the movement due to respiration is reduced relative to that 28

PAGE 29

due to heartbeat. Therefore, harmonic interference is reduced and better accuracy is obtained in the back case. 2.3.4 Residual Phase Optimum/Null Points Manipulation Equation (2-6) shows that Doppler radar in evitably brings the effect of Harmonic interference, whose stre ngth is decided by 4 mr/ Meanwhile, the detection accuracy is also controlled by the residual phase To address this issue, the n < 0 and n > 0 terms in (2-6) should be combined. The negative-order and pos itive-order Bessel functions are related by: () for even () () for odd n n nJx n Jx Jxn (2-8) Therefore, (2-6) can be reduced to: 21 2 00 22 1 1044 ()()cos()-cos()()cos()+cos() 44 2()cos2cos2()sin(21)sin. kk kk kk kkmm B tJ ktktJktkt mm Jk tJkt (2-9) Note that the term 0(4/)cos Jm is neglected in (2-9) since this is the DC term and has nothing to do with successful detection. The last two terms of (2-9) co rrespond to the odd order and even order harmonics in baseband spectra respectively. Based on this equation, successful detection of the periodic movement, which is related to the fundamental frequency ( k = 0 in the last term of (2-9)), is also dependent on the residual phase Taking two extreme cases as an example: When is equal to odd order of 900, the even order frequencies in (2-9) vanish, thus desired fundamental frequency is emphasized wh ile even order harmonics are minimized. When is equal to even order of 900, the odd order frequencies in (2-9) vanish, thus desired fundamental frequency is minimized. 29

PAGE 30

The residual phase is contributed by two factors: (2-1 ) the constant phase shift due to the distance to the target d0 and reflections at the target surf ace; (2-2) the total phase noise. Because of the range correlation effect [25] the phase noise plays a minor role and is to a large extent determined by the distance to th e target d0: 04 () d t (2-10) where can be treated as unchanged during experime nt. Therefore, a series of optimum points and null points for accurate detection exis t along the path away from the radar. For single-tone transmission, the distan ce between null points is decided by: 4 4 d d (2-11) Intuitively, if two tones are used for tran smission with a slightly difference in d due to wavelength difference, the occurrence of globa l null point, where both tones are at their respective null points, is largely reduced. Our previous work [10] based on the small-angle approximation model has demonstrated that the global null points for double sideband transmission are encountered every 1/8, where 1 is the wavelength corresponding to the intermediate frequency local oscillator with a frequency much lower than RF carrier frequency. Same result can be obtained by simultaneously c onsidering two tones, each with null point distance decided by (2-11) based on spectral an alysis. Therefore, the distance between null points is made much larger by double sideba nd transmission. Furthermore, double-sideband transmission can always make an accurate de tection by slightly tuning the intermediate frequency local oscillator to change an arbitrary pos ition into an optimum point. 30

PAGE 31

2.4 Simulation The above analysis was implemented in simula tion in two ways: (1) numerically analyzing the spectra of the modeled time domain signals constructed in MATLAB; (2) investigating the system performance using envelope simulation in Agilent ADS. Both of them achieved the same result, demonstrating the theory in Sect ion II and providing some guidelines for the frequency planning of the physiological movement sensor. 2.4.1 Spectrum In the front case, the amplitude of xh( t ) is on the order of 0.01mm [24], but that of xr( t ) is on the millimeter range. However, precise decision of mr is difficult because of the complex pattern of chest wall movement, different body status, and different subjects. Therefore, the value of mr is estimated for matching the simulated spectra with experimental result on the subject under test. For our subjec t (a 1.75 m tall man) seating stil l and completely relaxed in a chair, the mr is estimated to be 1mm in the front case and 0.2 mm in the back case. Fig. 2-6 (a) shows the simulated baseband spectrum of B ( t ) under mr = 1mm and mh = 0.08mm, when the RF frequency is 27 GHz. The residual phase is assumed to be 450 for both sidebands, which means the target is seated half way between the optim um point and the null point. The second order and third order harmonics of respiration are clearly discernable in the spectrum. It can be inferred that, two effects may lead to destructive interfer ence to heartbeat signal: Fist, as the subjects respiration increases in frequency, the third orde r harmonic may move closer to the location of the heartbeat in the spectrum. Second, as the subject breathes more he avily, the forth order harmonic may grow larger, and b ecome a source of interference. On the other hand, experimental results in Section II imply that the amplitude of xr( t ) on the back is reduced relative to that of xh( t ). In this case, the problem of harmonics is significantly 31

PAGE 32

reduced. Fig. 2-6 (b) shows the simulated sp ectrum when measured from the back ( mr = 0.2mm and mh = 0.08mm), when the RF frequency is 27 GHz. 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 Beats/MinNormalized SpectrumBreathing Fundamental Breathing 2nd Harmonic Breathing 3rd HarmonicHeartbeat A 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 Beats/MinNormalized SpectrumBreathing Fundamental Heartbeat Breathing 2nd Harmonic B Figure 2-6. Simulated normalized spectrum comp arison. Residual phase is assumed to be 450. A) Detecting from the front. B) Detecting from the back. 2.4.2 Optimum Choice of Carrier Frequency As discussed in Section II, a significant fact or influencing the syst em performance is the RF carrier frequency. As the RF frequency in creases from 2 GHz to 30 GHz, the breathing fundamental, the second order to the forth order breathing harmonics, and the heartbeat fundamental are simulated. The results are plotted in Fig. 2-7 (a) and Fig. 2-7 (b), for detection in the front case and detection in the back case, respectively. The amplitudes of chest wall movement due to respira tion and heartbeat are: mr = 1mm, mh = 0.08mm for the front case, and mr = 0.2 mm, mh = 0.08 mm for the back case. The residual phase is assumed to be 450. For 32

PAGE 33

comparison, the amplitudes of B(t) for detection from the front and from the back are all normalized to unity, as expressed in (2-3). 5 10 15 20 25 30 10-8 10-6 10-4 10-2 100 RF fre q uenc y ( GHz ) Normalized amplitudes Breath fundamental Breath 2nd harmonic Breath 3rd harmonic Breath 4th harmonic Heartbeat fundamental A 5 10 15 20 25 30 10-10 10-8 10-6 10-4 10-2 100 RF fre q uenc y ( GHz ) Normalized amplitudes Breath fundamental Breath 2nd harmonic Breath 3rd harmonic Breath 4th harmonic Heartbeat fundamental B Figure 2-7. Spectral intensity of breathing fundamental, breath ing harmonics, and heartbeat fundamental. A) Measuring from the front. B) Measuring from the back. It is clearly shown from the simulation result that, as the frequency increases, the amplitudes of detected heartbeat spectra increase acco rdingly, thus better sensitivity is gained for small heartbeat signal detection. However, the harmonics of respiration signal increase at the same time. It is also shown by comparing Fig. 2-7 (a) and Fig. 2-7 (b) that, the respiration harmonics are reduced in the back case, identifying the advantage of detection from the back. In the front case, the second order respiration harmonic is larger than hear tbeat fundamental at f > 10 GHz, and the third order harmonic has almost similar amplitude as the heartbeat fundamental 33

PAGE 34

at f ~ 30 GHz. Therefore, if the body is breathing fast so that one of those harmonics moves close to the desired heartbeat signal in the spectrum, it would desensitize the heart-rate detection accuracy. On the other hand, in the back case, all the breathing harmonics except for the fundamental have amplitudes smaller than that of heartbeat fundamental, thus eliminating the potential hazard of harmonics interference. The above discussions are focused on the ha rmonics and desired signal components, and the simulation results presented in Fig. 2-7 are based on a fixed amplitudes of chest wall movement due to respiration and heartbeat, i.e. mr = 1mm, mh = 0.08mm for the front case, and mr = 0.2 mm, mh = 0.08 mm for the back case. However, in real applications, not only the harmonics, but also the intermodulation tones affect the detection, since there are two periodic signals (respiration and heartbeat) phase modulati ng the radio frequency carrier simultaneously. A more accurate mathematical description of th e Doppler radar non-contact vital sign detection baseband signal is thus: 4() 4() ()cos( )h rxt xt Bt (2-12) where xr( t ) = mr sin rt and xh( t ) = mh sin ht represent body movement due to respiration and heartbeat, respectively. Th e corresponding Fourier se ries of (2-12) is: 4 4 ()Re()e()ee 4 4 ()()cos( ) h rjlt jkt h r kl kl h r lkrh klm m BtJ J m m JJktlt (2-13) It is shown in (2-13) that th e non-linear property of cosine tr ansfer function not only causes the undesired effect of harmonics interference, but also causes intermodulation between respiration signal and heartbeat signal. Therefor e the detected strength of a desired signal (respiration or heartbeat) is determined by both the signal itse lf and the other signal (heartbeat or 34

PAGE 35

respiration). For example, the detected heartbeat signal is determined by the ( l = k = 0) terms, and its amplitude 10(4/)(4/) hJmJmr is dependent on both mr and mh. More in-depth simulations were performed in ADS. Typically, relaxed human beings have mh on the order of 0.01 mm [24], and mr varying from the order of 0.1mm to several millimeters for different people. As an illustration of the nonlinear property in Doppler radar vital sign detection, Fig. 2-8 presents the simulated baseband spectrum when mr = 0.8mm and mh = 0.08mm. 0 20 40 60 80 100 120 10-2 10-1 100 Beats/MinNormalized SpectrumB1 B2 H1 B3 C1a C1b Figure 2-8.Simulated spectrum of baseband signa l. B1, B2, and B3: breathing fundamental, the 2nd harmonic, and the 3rd harmonic; H1: Heartbeat funda mental; C1a and C1b: lower sideband and upper sideband caused by th e intermodulation of B1 and H1; As shown in Fig. 2-8, besides the desired breat hing signal (B1) and heartbeat signal (H1), harmonics of respiration (B2, and B3) as well as the intermodulation term s (C1a and C1b) were observed. Since the respiration signal (B1) is us ually the lowest frequency component in the baseband spectrum and respiration usually causes a larger chest wall movement than heartbeat does, it is easy to extract respiration signal while accurate detection of he art rate presents the 35

PAGE 36

main challenge for vital sign detection. The following simulations and discussions will be focused on the detection of heartbeat signal H1. Although C1a and C1b are always separated to H1 by a frequency of B1, the variation in human respiration rate and hear t rate, both among different peopl e and within the same person under different physical conditions, makes it possible that C1a and C 1b falls within the range of heart rate variation. Thus intermodulation represents a kind of interference for correct heart rate detection. Moreover, the harmonics of respiration (B3, B4, etc.) have frequencies varying with the respiration rate, thus may be very close to heartbeat fr equency and significantly deteriorate the detection accuracy. From this point of view, the ab solute strength of detected heartbeat signal, and the strength of detected heartbeat signal compared with both intermodulation tones and harmonics are important factors influencing the performance of the vital sign detector The three factors are investigated as follows. 2.4.2.1 Absolute Detected Heartbeat Strength The theoretical (solid line) and the ADS-simulat ed (markers) detected heartbeat strength as a function of the carrier freque ncy is shown in Fig. 2-9. The am plitude of heartbeat-induced body movement was assumed to be mh = 0.08 mm, which is a typical va lue of ordinary people [24]. And the amplitude of respiration-induced body movement was considered ranging from 0.8 mm to 1.8 mm, assuming the most challenging case wh en detecting from the front of the human body [11]. As shown in the result, the amplitude of detect ed heartbeat signal increases with the carrier frequency when the carrier frequency is lo wer than 17 GHz. However, when the carrier frequency becomes large enough, the signal amplitude begins to decrease, especially for large 36

PAGE 37

mr. According to (2-13), this can be explained as follows: as the carrier frequency increases, the decreasing speed of 0(4/)rJm begins to exceed the increasing speed of 1(4/) hJm. 0 10 20 30 40 0 0.01 0.02 0.03 0.04 Carrier Frequency (GHz)Detected Heartbeat Strengthrm=1.2rm=1.0rm=0.8Line: Theoretical value Marker: ADS simulation rm=1.8 rm=1.4 rm=1.6 Figure 2-9.Amplitude of detected heartbeat signal versus the carrier freq uency. The amplitude of respiration-induced body movement mr ranges from 0.8 mm to 1.8 mm; the amplitude of heartbeat-induced body m ovement was assumed to be mh = 0.08 mm; Moreover, the detected heartbeat signal may even disappear at cert ain frequencies for a fixed mr, producing a detection absolute-null-point This detection absolute-null-point is caused by the null point of Bessel functi on. It should be noted that the absolute-null-point can not be eliminated by adjusting the phase shift and is different from the null points discussed in [8] and [10], where those null poi nts for certain phase shift can be eliminated by either doublesideband transmission or quadrature modulation. Therefore, for the seek of maximizing the det ected heartbeat strength, there is an optimum carrier frequency for a fixed value of mr. For instance, the optimum carrier frequency when mr = 1.2 mm is approximately 27 GHz. This is the reason that, by trial and error, the carrier frequency in [11] was chosen to be around 27 GHz even though the Ka-band transceiver implemented can work up to 40 GHz. 37

PAGE 38

2.4.2.2 Detected Heartbeat Compared with Harmonics Figure 2-10 shows the theoretica l (solid line) and simulated (markers) detected heartbeat signal strength compared with the third order brea thing harmonic. It is sh own that as the carrier frequency increases, the relative strength decreases until it reac hes a null point, which is caused by the absolute-null-point of detection. 0 10 20 30 40 0 0.5 1 1.5 2 Carrier Frequency (GHz)Heartbeat / 3rd Harmonic rm=1.0rm=0.8Line: Theoretical value Marker: ADS simulationrm=1.8 rm=1.6 rm=1.4 rm=1.2 Figure 2-10.Relative strengt h of detected heartbeat signal compared with the third order harmonic as the carrier frequency changes. The amplitude of respiration-induced body movement mr ranges from 0.8 mm to 1.8 mm; the amplitude of heartbeatinduced body movement was assumed to be mh = 0.08 mm. To reduce the harmonics interference, the carri er frequency should be confined to a range defined by a critical value. Assuming this critic al frequency is set wher e the heartbeat signal has equal strength as the third order harmonic, a dashed line was drawn in Fig. 2-10. The intersections indicate the cri tical frequencies for different mr values. For example, when mr is equal to 1.0 mm, the maximum carrier frequency is set around 29 GHz. 2.4.2.3 Detected Heartbeat Compared with Intermodulation Tone Figure 2-11 shows the theoretica l (solid line) and simulated (markers) detected heartbeat signal strength compared with the intermodulatio n tone caused by the respiration signal and the 38

PAGE 39

heartbeat signal. Similar to the effect of harmoni cs, the relative strength decreases until it reaches a null point, and thus the carrier frequency should be limited to a critical frequency indicated by the intersections with the dashed line. 0 10 20 30 40 0 0.5 1 1.5 2 Carrier Frequency (GHz)Heartbeat / Intermodulation rm=1.0rm=0.8Line: Theoretical value Marker: ADS simulationrm=1.8 rm=1.2 rm=1.4 rm=1.6 Figure 2-11.Relative strengt h of detected heartbeat signal compared with the intermodulation as the carrier frequency changes. The amplit ude of respiration-induced body movement mr ranges from 0.8 mm to 1.8 mm; the amplitude of heartbeat-induced body movement was assumed to be mh = 0.08 mm; Comparing Fig. 2-10 and Fig. 211, it can be observed that th e problem of harmonics is more serious than that of intermodulation. Mean time, it is theoretically possible to solve the problem of intermodulation by discerning the inte rmodulation tones based on the fixed frequency separation, i.e. the frequency difference between the intermodulation tone and the heartbeat signal is the frequency of respir ation. Also, it is observed from Fig. 2-9, Fig. 2-10, and Fig. 2-11 that, for the same amplitude of heartbeat-induced body movement mh, the optimum carrier frequency is higher for smaller amplitude of respiration-induced body movement mr. For example, the optimum carrier frequency for mr = 0.8 mm is as high as 36 GHz, and it could be even higher when detecting from the back of the body. However, to avoid significant 39

PAGE 40

performance deterioration as mr increases, it is recommende d to limit the carrier frequency within the lower region of the Ka-band. 2.4.2.4 Summary Harmonics and intermodulation issues of D oppler radar vital sign detection have been modeled and simulated in Agilent ADS. It has been shown that in contrast to the common sense that detection accuracy can always be increased by increasing the carrier frequency, there is an optimum choice of carrier frequency. At the optimum carrier frequency, the heartbeat signal component can be maximized in the premis e that the harmonics interference and the intermodulation interference are not so large as to affect the detection accuracy. According to simulation, the carrier frequency can be increas ed up to the lower region of the Ka-band to improve detection accuracy. 2.5 Experiment Because the artifacts around human body bri ng noise and uncertainty for quantitative analysis, we used a mechanical device for ha rmonics and null/optimum point observation. The measured result is compared with the prediction by the theory based on the spectral analysis in Section II. A mechanical device with a flat metal reflector swinging at a frequency around 1.8 Hertz and amplitude of a few millimeters was used as th e target to test the theory of harmonics and null/optimum points. The transmitting frequency of the Ka-band detector was chosen to be 27.1 GHz. The harmonic interference effect was obser ved. Then, by adjusting the distance from the subject to the radar, different spectra were observed at positions close to the null point and the optimum point. Since the swinging frequency of the subject does not affect the relative strength of harmonics in the spectrum, the frequency is normalized to the fundamental frequency (the 40

PAGE 41

actual swinging frequency) in the following spect rum graphs and the amplitude of the spectrum is normalized to unity. 2.5.1 Observation of Harmonics When the residual phase is around 450, or the two tones have a nearly 900 difference in residual phase, both even order and odd order harm onics exist in the base band signal. This is observed in experiment by adjusting the posi tion of the subject. The measured baseband spectrum is shown as the solid line in Fig. 2-12. 0 1 2 3 4 5 0 0.2 0.4 0.6 0.8 1 Normalized Frequenc y Normalized Spectrum Harmonics Location Measured Spectrum Figure 2-12. Measured and calcula ted baseband spectrum when residual phase is found to be 51.770. Both odd order and even order harmonics exist. The swinging amplitude is calculated to be m = 2.4 mm. The frequency is normalized to the fundamental frequency. A powerful aspect of the Fourier expansion spec tral analysis is, the amplitude and residual phase of the periodic movement can be derived from the measured spectrum, as long as the RF frequency is known. By comparing the strength of the third order harmonics with the fundamental, the swinging amp litude is calculated as m = 2.4 mm to satisfy the J3(4 m / ) ~ 0.61 J1(4 m / ) relation from measurement. Using the value of m and evaluating the relative strength between odd order harmonics and even order ha rmonics, the residual phase is found to be 51.770 according to (2-9). Then, plugging these values back into (2-9), the theoretical baseband 41

PAGE 42

spectrum is obtained and shown as the dashed stems in Fig. 2-12. Comparing the two spectra, the theoretical value matches well with the measurement. 2.5.2 Even Order Harmonics Dominant Null Point As predicted by (2-9) in Section II, when the di stance from the radar to the target produces a residual phase equal to even orders of 900, odd order harmonics will vanish, corresponding to a null detection point. Simila r effect is observed in the experiment and the resulting baseband spectrum is recorded in Fig. 2-13. However, sinc e double-side band transmission is used in our experiment, its difficult to completely eliminate all the odd order harmonics. Therefore, there is still small fundamental frequency discernable in Fig. 2-13. The difficulty in observing a complete null point also reflects the effec tiveness of double sideband transmission scheme. 0 1 2 3 4 5 0 0.2 0.4 0.6 0.8 1 Normalized FrequencyNormalized Spectrum Harmonics Location Measured Spectrum Figure 2-13. Measured baseband spectrum when even order harmonics dominate, corresponding to a null detection point. 2.5.3 Odd Order Harmonics Dominant Optimum Point Similarly, when the distance from the rada r to the target produces a residual phase equal to odd orders of 900, even order harmonics will vanish, corresponding to an optimum detection point since fundamental tone is maximized while even order harmonics are minimized in the spectrum. This was verified experimentally and the resulting baseband sp ectrum is shown in Fig. 2-14. Once again, however, since double-side band tr ansmission is used in our experiment and 42

PAGE 43

there could be small nonlinear in ter-modulation in the RF receiver circuit, the even order harmonics are not completely eliminated and they are still discerna ble in the spectrum. 0 1 2 3 4 5 0 0.2 0.4 0.6 0.8 1 Normalized FrequencyNormalized Spectrum Harmonics Location Measured Spectrum Figure 2-14. Measured baseband spectrum when odd order harmonics dominate, corresponding to an optimum detection point. 2.6 Conclusion A low-power Ka-band non-contact heartbeat detection system has been demonstrated. Experiments on human body show that the Ka-ba nd detector achieves high accuracy and has better performance when measuring from the b ack of the body than from all other sides. A rigorous spectral analysis technique is used to analyze the harmonic effect in Ka-band sensor, showing that minimizing the harmonics interference leads to the advantage of detecting from the back. The theory also illustrates from the sp ectral analysis that doubl e side-band transmission can avoid severe null point problem for short wavelength detection. Si mulations have been performed to demonstrate this theory, providing guidelines for system design. Experiments have successfully verified this theory by the observation of harmonics that match with the theoretical value, and the observation of the null/optimum point. 43

PAGE 44

CHAPTER 3 A NEW METHOD AND SYSTEM FOR NON-CONTACT MEASUREMENT OF FREQUENCY AND AMPLITUDE OF MECHANICAL VIBRATION 3.1 Introduction Microwave and millimeter-wave have been widely used for position sensing [26], precision noise measurement [27], and displacement measurement [28]. Th e mechanism of most of the microwave displacement-related measuremen t systems is the detection of the phase shift caused by the movement of the target. Based on this, the Doppler radar non-contact vital sign detector has been developed to monitor period ic vital sign, i.e. resp iration and heartbeat, movements. However, the system described in Chapter 2 was designed to only detect the frequency of movement, not the amplitude. In this chapter, a detection method based on the measurement of the nonlinear effect inherent in phase modulation mechanism is inve stigated. By identifying different orders of harmonics caused by the nonlinear property of pha se modulation, it is possible to precisely measure not only the frequency of the periodic movement, but also the amplitude. This method offers several advantages: no calibration of signal amplitude required, self-verification to ensure accurate measurement, large detection range of movement frequency and amplitude, and a very simple hardware architecture. Furthermore, the near field and wide angle incidence eff ects are studied. The study, which is applicable to most of the microwave displacement sensing system, shows that proper choice of antenna radiation pattern and m easurement distance is critical for precision measurement. The study also provides guidelines for the design of measurement systems used for actuator calibration, Doppler radar vi tal sign monitoring, etc. 44

PAGE 45

3.2 Theory and Analysis The system architecture is similar to those pr eviously used to mon itor periodic movements, but it works under different modul ation conditions. The block diagram of the system and the experimental setup are shown in Fig. 3-1. xy d LO f PowerSplitterMixerLNARF TransceiverOutIn Baseband Actuato r ( ) x y ( ) xy Figure 3-1. Block diagram and experimental setup. The RF transceiver operates from 22 GHz to 40 GHz. A 3-dB power splitter is used to divide the generated signal into two, with half of the power sent to the transmitting antenna and half of the power sent to the receiver mixer as the reference. When the transmitted signal reaches the target, it is reflected and captured by the recei ving antenna. After a low noise amplifier, the signal is down-converted to baseband and furthe r amplified by the baseband circuit before it is digitized and fed into the lapt op for real time signal processing. In the experiment testing the performance of the system, the target is contro lled by a precision linear actuator through another laptop to produce desired periodic movements. As shown in Chapter 2, for ideal Doppler radar sensing of physical movements, the normalized detected baseband signal is: 4() ()cos () xt Bt t (3-1) where is the constant phase shift created on the transmission path and at the surface of the target, () t is the total residual phase noise, is the carrier wavelength, and x ( t ) is the time 45

PAGE 46

varying displacement of the target. Fo r a single tone periodic movement, x ( t ) = m sin t ; for a more complex movement, it can be decomposed into a series of single tone movements. When the movement amplitude m is much smaller than the wavelength a linear approximation can be applied [8]. However, when small wavelength is used such that x ( t ) is comparable to a rigorous spectral analysis should be applied, and the received signal can be expanded as [13]: 22 1 1044 ()2()cos2cos2()sin(21)sin kk kkmm BtJ kt J kt (3-2) where () t is the total re sidual phase, and Jn( x ) is the nth-order Bessel function of the first kind. Therefore, the phase-modulated baseband signal is decomposed into a number of harmonics of the fundamental frequency. While the movement frequency is readily obtained from th e fundamental frequency of B ( t ), (3-2) also shows that fo r a certain carrier frequency, the relative strength among the harmonics is decided by the movement amplitude m and the residual phase and is not a function of signal level determined by receiver ga in and measurement distance. For example, the absolute value of ratio among the 1st, 2nd, 3rd, and 4th order harmonics is: 12341 2 3 44444 :::|()cos|:|()sin|:|()cos|:|()sin|. mmmm HHHHJ J J J (3-3) Moreover, if separating the harmonics into gr oups of even order a nd odd order, the ratio inside each group is only decided by m Therefore, m can be found by fitting the measured harmonic ratio to the theoretical value from Bessel function, which can be performed for either odd order or even order harmonics. This analysis leads to a very important application the amplitude of the movement can be accurately determined in remote non-contact measurement without calibrating the signal le vel vs. distance, provided that the wavelength is accurately 46

PAGE 47

determined. If using linear modul ation method, the signal level which is affect ed by the receiver gain and the distance to the ta rget would need to be calibra ted to determine the movement amplitude. If both the even order and the odd order harmoni c ratios are used to extract the amplitude of movement, there will be more than a single result for the same measur ement. The accuracy of the measurement can thus be verified by checking whether the two results agree with each other. Therefore, a pair of harmonic ratios (i.e. an even order ratio and an odd order ratio) will be used simultaneously for a single measurement. For example, Fig. 3-2 shows up to the 6th-to-4th harmonic ratio as a function of normalized moveme nt amplitude. The plot shows that when the displacement is too small compared with the wavelength, the harmonics are too week to be observed, corresponding to the linear approximation region in [8]. As the displacement increasess, harmonics become observable, maki ng the invented measurement method possible. 0 0.2 0.4 0.6 0.8 1 0 5 10 Movement Amplitude (m/)Harmonic Ratio H3/H1 ratio H4/H2 ratio 0 0.2 0.4 0.6 0.8 1 0 5 10 Movement Amplitude (m/ )Harmonic Ratio H5/H3 ratio H6/H4 ratio BA Figure 3-2. Theoretical harmonic ratio as a func tion of the movement amplitude. The movement amplitude m is normalized to the carrier wavelength A) H3/H1 and H4/H2 ratio. B) H5/H3 and H6/H4 ratio. However, it should be noted that there are mu lti-solutions of movement amplitude because of the non-linear property, and it is impractical to accurately m easure either too small or too large a harmonic ratio. Therefore, a detection range is defined for a pair of harmonic ratios as 47

PAGE 48

the lowest range of movement amplitude that induces the measurable harmonic ratio, which is 0.2~5 from experience. For exam ple, in Fig. 3-2 (a), the detection range of the 3rd-to-1st and the 4th-to-2nd harmonic ratio pair is 0.214 ~ 0.290 which correspond to 0.2 for H4/H2 and 5 for H3/H1 respectively. Similarly, the detection range of the 5th-to-3rd and the 6th-to-4th harmonic ratio pair in Fig. 3-2 (b) is 0.335 ~ 0.489 Based on the observation above, the applicable detection range normalized to for different harmonic pairs is calculated in Table 3-1. The HPair with index i means the (i+2)th-to-ith and the (i+3)th-to-(i+1)th harmonic ratio pair. Table 3-1. Detection range for different pairs of harmonics H-Pair i = 1 i = 2 i = 3 i = 4 i = 5 Lower Bound 0.214 0.335 0.455 0.575 0.694 Upper Bound 0.290 0.489 0.677 0.859 1.039 H-Pair i = 6 i = 7 i = 8 i = 9 i = 10 Lower Bound 0.813 0.932 1.051 1.170 1.288 Upper Bound 1.216 1.391 1.565 1.737 1.909 It is shown in Table 3-1 that except for a small gap between the i = 1 pair and the i = 2 pair, the nonlinear detection met hod can detect any movement amp litude that is larger than 0.214 By tuning the frequency for about 10% to chan ge the wavelength, the gap can also be covered. Therefore, the measurement range of the system is any movement amplitude larger than 0.335 for a fixed carrier frequency system and larger than 0.214 min for a frequency tunable system, where min is the minimum carrier wavelength. 3.3 Experimental Results Experiments have been performed using the system shown in Fig. 3-1. An example of typical measured baseband signal and spectrum is shown in Fig. 3-3. The carrier frequency is 40 48

PAGE 49

GHz, and the transmitted power is 50W. The target is 1.65 m away from the antenna. Up to the 5th order harmonic of movement is clearly discernable in the detected spectrum. 0 5 10 15 -0.5 0 0.5 1 Time (Second)Baseband Signal (V ) 0 1/3 2/3 1 4/3 5/3 2 0 0.5 1 Frequency (Hz)Normalized Spectrum A B Figure 3-3. Detected baseband signal and spectrum when carrier frequency is 40 GHz. A) Baseband signal. B) Baseband spectrum. 1 1.5 2 2.5 3 0 1 2 3 4 5 Displacement (mm)Harmonic Ratio H3/H1 ratio H4/H2 ratio Meaured H3/H1 Meaured H4/H2 Figure 3-4. Method to obtain the displacement and to check the validity of the measurement. It is observed from the spectrum that the period of the movement is 3 second. Given the carrier frequency f = 40 GHz, the theoretical harmonic ratios based on (3-3) can be calculated. Fig. 3-4 shows the 3rd-to-1st and the 4th-to-2nd harmonic ratios as the movement amplitude changes. From the spectrum of Fig. 3-3 (b), the measured 3rd-to-1st and 4th-to-2nd harmonic ratios are 2.3613 and 0.414, corresponding to the movement amplitude of 2.056 mm and 2.045 mm 49

PAGE 50

respectively. The measurement result agrees we ll with the programmed movement amplitude of 2 mm for the actuator. This example shows th e invented nonlinear phase modulation method can provide accurate result with a simple radio architecture. 3.4 Wide angle Incidence and the Effect on Accuracy In real applications, the antenna has finite directivity when radiating out the signal. The signal will thus reach at different parts of the target. Relative to the antenna position, different parts of the target have different movement di rection and speed. Suppose the antenna main beam is aligned to the center of target, th en the received signal from location ( x y ) on the target should be expressed as: 2 2 04sin()(,) (,,)(,)cos( ) mtdxy BtxyIxy (3-4) where I ( x y ) is the reflected electromagnetic intensity, d0 is the horizontal distance from the antenna to the target, and ( x y ) is the vertical distance from that location to the center of target, as shown in Fig. 3-1. The total re ceived signal is th e integral of B ( t x y ) over the whole target surface: ()(,,) s BtBtxyds (3-5) Since I ( x y ) is dependent on the antenna, we first cons ider the cosine term of (3-4). Due to wide angle incidence, different parts of the target induce di fferent baseband spectrum. For example, Fig. 3-5 compares the simulated baseband spectrum induced from the center and the corner of the target, which is a 14 cm by 14 cm square placed at 0.25 m in front of the antenna. From this point of view, when signals w ith different baseband spectrum are added together, it may badly deteriorate the meas urement accuracy. Fortunately, the impact of I ( x y ) can help resolve this problem. Intuitively, if the antenna beam is as narrow as a line, the detected 50

PAGE 51

signal contains only the information from the center of the target and thus produces ideal measurement result. Therefore, di rectional antenna can be viewed as a filter that blocks the unwanted signal from the edge of the target. 0 1 2 3 4 5 6 0 0.5 1 Normalized FrequencySpectrum 0 1 2 3 4 5 6 0 0.5 1 Normalized FrequencySpectrum A B Figure 3-5. Baseband spectrum detect ed with the target placed 0.25m in front of the antenna. A) Spectrum induced from the target center. B) Induced from the target corner. To qualitatively illustrate this effect, th ree types of antennas are considered: the omnidirectional antenna, the hor n antenna used in the experi ment, and a 10 by 10 array of omnidirectional antennas. Fig. 3-6 shows the radi ation pattern of the horn and the antenna array. Numerical simulation was performed to get th e spectrum of the total received baseband signal B ( t ) in (3-5). The detection error was obtai ned by comparing the re ceived spectrum with the spectrum induced from the center of the target, which is the ideal case using the antenna with infinite directivity. The error is defined as the deviation in the 3rd-to-1st and the 4th-to-2nd harmonic ratios compared with those in ideal case. The result is pr esented in Fig. 3-7. From the result, the measurement error is always less than 5 percent in any case. The error gets smaller as the radiation pattern gets more focused. The error also has a tendency to reduce as the distance increases, although a fluctuation is observed. The fluctuation, especially when the distance is relatively small, is caused by the near field effect. To be more specific, the difference in the distance from different parts of the target to the antenna leads to the difference in residual 51

PAGE 52

phase, causing out-of-phase signals canceling each other. When most of the undesired signals, i.e. signal produced far from the center of the target, cancel each other, the error is small; otherwise the error becomes large. 0o180o 90o90o 30o150o60o120o30o150o60o120o-10-20 -30 dB 0o 180o 90o90o 30o150o60o120o30o150o60o120o-10-20 -30 dB 0o 180o 90o90o 30o150o60o120o30o150o60o120o-10-20 -30 dB 0o 180o 90o90o 30o150o60o120o30o150o60o120o-10-20 -30 dB A B C D Figure 3-6. Radiation pattern of the horn antenna used in experi ment and a 10 antenna array. A) H-plane of the horn. B) E-plane of the horn. C) = 00 plane of the array. D) = 450 plane of the array. 0 0.5 1 1.5 2 2.5 -4 -2 0 2 Distance from Antenna (m)Error % Omnidirectional Ant Horn Ant 10 by 10 Array 0 0.5 1 1.5 2 2.5 -1 0 1 Distance from Antenna (m)Error % Omnidirectional Ant Horn Ant 10 by 10 Array A B Figure 3-7. Measurement error of 3rd order to 1st order ratio and 4th order to 2nd order ratio by different antennas. A) 3rd order to 1st order ratio. B) 4th order to 2nd order ratio. 52

PAGE 53

The above analysis shows that the wide angle incidence due to large target aperture does not cause significant degradation of measurement accuracy. 3.5 Conclusion A non-contact detection technique of measuri ng both the frequency and the amplitude of periodic movement using the property of nonlinear phase modulation is demonstrated by a 2240GHz radar sensor. It is shown that this method has the following advantag es: 1) no calibration of signal amplitude required for accurate measurem ent of movement amplitude; 2) ability of selfverification; 3) very simple architecture. The method can be used to detect movements with amplitudes larger than 0.335 of the carrier wave length for a fixed carrier frequency system and 0.214 of the minimum carrier wavelength for a frequency tunable system. By analyzing and simulating the wide angle incidence effect, it is shown that a proper choice of radio frequency, measurement distance, and antenna pattern can ensure accurate measurement of movement amplitudes. 53

PAGE 54

CHAPTER 4 SYSTEM INTEGRATION ON BOARD LEVEL 4.1 Introduction While the Ka -band radar system is the most sensitive, systems with lower carrier frequencies have lower costs and fewer harmonics issues, which is a significant advantage when the requirement on sensitivity is not as tight. Additionally, a direct-conversion quadrature transceiver architecture offers a competitive so lution: the complex signal demodulation [20] perfectly solves the null-detection point problem in software, and the DC offset is not a big concern for general purpose vital sign detecti on where the frequency com ponents of interest are not related to DC. 4.2 Circuit Design and Implementation A portable 4-7 GHz direct conversion quadr ature radar was designed and implemented. Fig. 4-1 shows the block diagram and photo of the radar, which integrates the quadrature transceiver, the two-stage base band amplifier, and the power management circuit on a single Rogers printed circuit board (R O4350B) with a size of 6.8 cm 7.5 cm. In the transceiver part, four voltage controlled oscillat ors (VCOs) are used to completely cover the frequency range from 4.4 GHz to 6.7 GHz. The four VCOs guarantee the phase noise to be always lower than 101 dBc/Hz at a 100 kHz offset, and the maximum output power to be more than 2 dBm over the entire frequency range. A DC 8.0 GHz single-pole four-throw (SP4 T) electronic switch is used to switch among the four VCOs. To change the op eration range (i.e. the distance between the radar and the subject to be monitored), a gain bl ock is used to adjust the gain after the switch. After the Wilkinson power divider, half of the power is transmitted through the transmitting antenna and the other half of the power is furthe r amplified and used to drive the mixer in the receiver chain. The receiver ch ain contains a 3.5 7 GHz low noi se amplifier (LNA), two stages 54

PAGE 55

of adjustable gain block, and the down-conversi on mixer, which is a comp act I/Q mixer utilizing two standard double balanced mixer cells and a 90 degree hybrid fabricated in a GaAs MESFET process. The radio frequency part of the receiv er chain has an adjustable 30 dB of dynamic range. The specifications and manufacturers of th e radio frequency components are listed in Table 4-1. DAQ Power Management To computerLNAGainBlockMixer GainBlockPower SplitterGain BlockSwitch1234VCOs Wall Plug/BatteryRFinRFout1C2C1R2R Attenuator (optional) A B Figure 4-1. The 4-7 GHz vital sign dete ctor. A) Block di agram. B) Photo. Table 4-1. Building blocks and specifi cations used in the 4-7 GHz radar Block Manufact urer Specifications VCO1 Hittite 4.46-5.0 GHz, -105dBc/Hz @100 kHz phase noise, 4 dBm output power VCO2 Hittite 5.0-5.5 GHz, -103dBc/Hz @100 kHz phase noise, 2 dBm output power VCO3 Hittite 5.6-6.1 GHz, -102dBc/Hz @100 kHz phase noise, 2 dBm output power VCO4 Hittite 6.1-6.72 GHz, -101dBc/Hz @100 kHz phase noise, 4.5 dBm output power Switch Hittite DC-8 GHz, 40 dB isolation @6 GHz, 1.8 dB insertion loss @6 GHz, SP4T Gain Block RFMD DC-8 GHz, 15.5 dB maximum gain, 14.5 dBm P1dB @6Ghz Mixer Hittite 4-8.5 GHz, 50 dB LO to RF isolation, 40 dB image rejection LNA Hittite 3.6-7.0 GHz, 16 dB gain, 2.5 dB NF 55

PAGE 56

The down-converted baseband quadrature signa ls are amplified by a two-channel twostage amplifier, which is real ized in a space-saving package with four unit-gain-stable op amplifiers. Except for the VCOs, the passive I/Q mixer, and the SP4T switch, all the other components have a single supply voltage of 5 V. The VCOs are 3 V supplied, and the SP4T switch needs a fixed bias of -5 V and negative c ontrol voltage of 0/-5 V. Meantime, the VCOs require a 0 to 10 V tuning voltage. Therefore, 5 V and 3 V fixed output voltage regulators are implemented, and an adjustable output regulator w ith up to 11 V output voltage is used to tune the frequency. A switched-capacitor voltage converter is used to generate a -5V logic supply. Either a 6 9 V wall plug or a 9 V batte ry can be used to power up the radar. Figure 4-1 (b) shows the photo of the integrated transceiver. For research purpose, four voltage controlled oscillators (VCO) were implemented for a wide tuning range to obtain different optimal carrier fre quencies under different environments [14]. When a specific application is known, only one VCO is needed and the SP4T switch can be eliminated to further reduce the cost. By removing thr ee of the on-board VCOs, simpler versions of the radar have been fabricated and packaged for baby monitor and vital sign detection robot prototypes, which were built to demonstrate the immediate real-lif e applications of non-cont act vital sign detection. 4.3 Measurement Results The integrated quadrature radar system was used to illustrate the ability of complex signal demodulation technique to eliminate the null detection point problem. Fig. 4-2 shows the measurement results obtained by slightly adjustin g the subject-to-antenna distance. Three typical cases are reported here as: (a) the I channel was at the optimal detection point while the Q channel was at the null detection point; (b) the Q channel at the optimal while the I channel at the null detection point; (c) bot h channels are at the midpoint between the null and the optimal. It is observed that although the I and the Q channel will change from one state to another (i.e. 56

PAGE 57

optimal/null detection) as the distance betw een the antenna and the subject changes, the reconstructed complex signal always has a stab le spectrum without th e null detection point problem. 0 2 4 6 8 10 12 -0.2 0 0.2 Time (Second) 0 20 40 60 80 100 120 0 0.5 1 Beats/Min 0 20 40 60 80 100 120 0 0.5 1 Beats/Min 0 2 4 6 8 10 12 -0.2 0 0.2 Time (Second) 0 20 40 60 80 100 120 0 0.5 1 Beats/Min 0 20 40 60 80 100 120 0 0.5 1 Beats/Min 0 2 4 6 8 10 12 -0.2 0 0.2 Time (Second) 0 20 40 60 80 100 120 0 0.5 1 Beats/Min 0 20 40 60 80 100 120 0 0.5 1 Beats/MinVoltSpectrumI I Q Q I Q Q II/Q SpectrumCSD A B C Figure 4-2.Measurement results of complex dem odulation. Plotted are th e time domain signal, the I/Q channel spectrum, and the spectrum from the complex signal demodulation (CSD). A) I: optimal detection, Q: null de tection. B) I: null detection, Q: optimal detection. C) I, Q: both at the midpoint between the optimal and the null detection points. 57

PAGE 58

CHAPTER 5 SYSTEM INTEGRATION ON CHIP LEVEL 5.1 Introduction In this chapter, two versions of CMOS chip level integration of the Doppler non-contact vital sign detection radar will be presented. The first version is a 5 GHz indirect conversion double sideband transmission radar sensor inte grated in UMC 0.18 m mixed-mode/RF CMOS process. Double sideband transmission and inte rmediate frequency (IF) tuning are used to eliminate the null detection point problem. This is the first integr ation of double sideband Doppler non-contact vital sign de tection radar, demonstrating the compatibility of this architecture with CMOS technology. The second ve rsion is a 5 GHz direct conversion quadrature radar sensor integrated in UM C 0.13 m mixed-mode/RF CMOS pro cess. This is a full systemon-chip integration with bias ci rcuitry, baseband circuitry, and digi tal control interface. The chip is designed to be software configurable for gain and operation points. Important design issues for direct conversion Doppler vital sign detection radar, e.g. baseba nd flicker noise and gain budget, will be discussed in detail. 5.2 A 5 GHz Indirect Conversion Double Si deband Transmission Radar Sensor 5.3.1 CMOS 5 GHz Indirect Conversion Radar Transceiver In Chapter 2, a Ka-band radar using doublesideband transmission for non-contact vital sign detection has been demonstrated using a benc h-top prototype system integrating commercial modules. It has also been shown that double-side band transmission can avoid null detection point problem at high frequency without using quadrature detection [10]. A fully integrated 5 GHz double-sideband radar chip designed and fabricated in UMC 0.18 m mixed-mode/RF CMOS process is presented. This frequency at C-band was chosen for its higher sensitivity to small physiological moveme nts than the previous UHF-band integrated 58

PAGE 59

direct-conversion system [8] and its better tolerance to harmonic interference than the Ka-band system [14]. This is the first time that double-sideband architectur e was implemented in low-cost CMOS chip for vital sign detec tion. Different from the previous Ka-band system, the 5 GHz chip uses a differential architecture. Electrical characterization verified the design and lab experiments demonstrated that the integrated CMOS double-sideband radar sensor chip can successfully detect the heartbeat and respirati on from a distance away. The differential design also effectively reduces LO leakage. Fig. 5-1 and Fig. 5-2 show the block diagram and schematic of the 5 GHz radar sensor chip. For vital sign detection, the transmitted ra dio frequency carrier is not modulated until it reaches the human body, where the periodic phy siological movement phase-modulates the carrier. Therefore the transmitter has two voltage controlled oscillators (VCO) as the signal sources, and a Gilbert double-bala nced mixer as the up-converter (Up-Conv). The receiver chain has a low noise amplifier (LNA), a Gilbert doublebalanced mixer as the down-converter (DownConv), an intermediate frequency amplifier (I F-Amp), and a second stage passive mixer (IFMixer). IF-AmpLNAIF-VCORF-VCO Baseband IF-Mixer Test Down-ConvUp-Conv G=18 dBG=23 dB1 f 2 f 5 GHz Radar ChipBalun Figure 5-1. Block diagram of the 5 GHz radar tr ansceiver with the input/output interface. ( f1 = 60~520 MHz; f2 = 4.6~5.7 GHz) The radio frequency voltage controlled oscilla tor (RF-VCO) shown in Fig. 5-2 (a) is a cross-coupled LC oscillator. PMOS transistors were used to achieve better phase noise 59

PAGE 60

performance. As shown in Chapter 2, the desired signal, the harmonics of respiration, as well as the intermodulation interference are all dependent on the carrier frequency and there is an optimum choice of the carrier frequency for di fferent people with different physiological movement amplitudes. Therefore, the RF-VCO was designed to ha ve a tuning range as large as possible. Two N-Well varactors (VARMIS_18_RF) were used to achieve a tuning range of 1.1 GHz around the center frequency, as the control vo ltage sweeps from 0 volt to 1.8 volt. The intermediate frequency voltage c ontrolled oscillator (IF-VCO) also needs to be tunable to avoid the null detection point. It was designed to be a two stage differential ring oscillator as shown in Fig. 5-2 (c). For double-sideband vital sign detection in 5 GHz band, an IF frequency of 100 MHz to 300 MHz is desirable, and the differentia l ring oscillator was designed to have a tuning range from 60 MHz to 520 MHz. The up-converter is realized by a Gilbert double-balanced mixer as shown in Fig. 5-2 (d). An inductor of L3 = 1.5 nH is used for source degenera tion to improve the linearity. Two 1.58 nH inductors, L4, are used at the mixer load to tune the output to the center frequency of 5.5 GHz. The mixer is driven by the RF and th e IF VCOs to generate a double-sideband transmission spectrum. For the receiver chain shown in Fig. 5-2 (b), single-ended cascode architecture was used for the LNA. The source degeneration and input matching inductors are realized by bond wires, with only the load inductor remained on-chip. Th e Gilbert double-balanced mixer transforms the singled-ended input to differential signal af ter the LNA stage. For the first stage down conversion, the Gilbert down-converter uses PMOS tr ansistors as the load si nce the output signal is at the IF frequency around 200 MHz. Followi ng the down-converter, a differential amplifier was used to boost the signal leve l. A resistor feedback loop was implemented at the IF amplifier 60

PAGE 61

load to set the DC operating point. A passive mixe r was used at the last stage to down-convert the IF signal to baseband. The high linearity of passive mixer effectivel y suppresses the circuitgenerated harmonics of respirati on signal, which is important for accurate detection of heartbeat [14]. 2L RFLO+ RFLOoutRX (aoutRX)((c)b) RFLO+RFLOoutTXoutTX (d)3L4L4L Down-ConverterIF-MixerIF-Amplifier RFctrlV 5L5L6L6LRFLORFLO 1V-2V+2VIFLO+IFLO1V+ IFctlV 1LinRF IFLO+IFLO LNA A B IFLO IFLO Figure 5-2. Simplified circuit diag ram of the 5 GHz radar transceiver (details of biasing not shown). A) Cross-coupled LC oscillator for RF-VCO. B) Receiver chain. C) Differential ring oscillator for IF-VCO. D) Up-converter. C D 61

PAGE 62

The microphotograph of the transceiver chip is s hown in Fig. 5-3. A differential test port is provided at the top side of the chip to measur e the RF-VCO output signal before packaging. Although indirect conversion archit ecture is used, no image reject filter is needed since both sidebands are required. The complete RF transceiver is integrated in one chip. Test OutputRF OutputRF In4RF-VCO Up-ConvDown-ConvLNA 123BasebandOutput Figure 5-3. Chip microphotograph of the 5 GHz radar transceiver (1: IF-VCO; 2: IF-Mixer; 3: IF-Amplifier; 4: Bond wire for LNA source degeneration). 5.3.2 Baseband Circuit OutIn-1C2C2R1R2R1RR RR10R(1~10)RIn+ 10R Figure 5-4. Baseband circuit for signal amp lification and differential to single-ended transformation. An external baseband circuit is needed to amplify the detected signal and transform the differential signal to single-ended for an external Data Acquisition Module (DAQ). As shown in Fig. 5-4, the differential input is amplified by 7.7 dB at the first st age, then further amplified by 62

PAGE 63

10 dB and transformed to single ended at the second stage. The optional third stage provides additional amplification of up to 10 dB for long-distance detection. 5.3.3 Antenna It is observed in experiment that the antenna gain should be higher than 4 dB for the sensor chip to have good signal-to-noise ratio at 2 m aw ay. Two 2-by-2 patch ante nna arrays with gain of 9 dB are used to transmit and receive signals. The antenna is designed to have the maximum directive gain at broadside, so that the vital si gns of the subject in front of the antenna will be picked up. The complete system including RF radar chip, baseband circuit board, and DAQ module was designed to be powered by a laptop throug h USB cable, which also sends the digitized baseband signal to computer for further processing. 5.3.4 Experimental Results The electrical characteristics of the 5 GHz radar chip were tested first. The functional test was then performed by detecting vital signs of a human subject seated at a distance of a few meters away from the 9 dB antenna array. A 4 4.5 5 5.5 6 6.5 -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 0 Freq. [GHz]Amp. [dBm]Res BW 100 kHz VBW 100 kHz Sweep 301.4 ms(601 pts) Atten 10dB Mkr1 5.467 GHz; Mkr2 5.683 GHz; Mkr3 5.900 GHz -7.16 dBm; -30.51dBm; -6.14 dBm 63

PAGE 64

B 4 4.5 5 5.5 6 6.5 -100 -90 -80 -70 -60 -50 -40 -30 -20 -10 0 Freq. [GHz]Amp. [dBm]Res BW 100 kHz VBW 100 kHz Sweep 301.4 ms(601 pts) Atten 10dB Mkr1 4.417 GHz; Mkr2 4.625 GHz; Mkr3 4.825 GHz -12.50 dBm; -29.95dBm; -12.40 dBm Figure 5-5. Transmitted spectrum when RF-VCO tune d at the highest frequency and the lowest frequency. A). Tuned at the highest frequency. B) Tuned at the lowest frequency. Fig. 5-5 shows the transmitted spectra when the RF-VCO was tuned at the highest output frequency of 5.683 GHz and at the lowest fre quency of 4.625 GHz, respectively. A tuning range of over 1 GHz was achieved for the ease of tuning the system for different subjects under test. The single-sideband transmitter output power at the antenna connector varies from -12.5 dBm to -6.1 dBm as the frequency increases from 4.6 GHz to 5.7 GHz. By comparing the transmitted spectra with those of the Ka-ba nd transceiver as shown in Fig. 23, it is found that the differential architecture effectively reduced the LO leakage. The ideal receive r test for vital sign detection requires feeding into the receiver input signals that have frequencies close to the transmitted upper and lower sidebands (with frequency offset of less than several hertz). However, since the signal generator can not be synchr onized with the free running onchip VCO, a weak signal with a frequency of 20 MHz higher than the transmitt ed upper sideband was fed into the receiver to verify its characteristics. The receiver gain is 40.8 dB. The current cons umption of the whole RF transceiver chip is 42 mA at 1.8 volt power supply. 64

PAGE 65

0 20 40 60 80 100 0 0.5 1 Beat/MinSpectrum Respiration Harmonic Heartbeat 0 3 6 9 12 15 -1 0 1 A 0 20 40 60 80 100 0 0.5 1 Beat/MinSpectrum Respiration Heartbeat Harmonic 0 3 6 9 12 15 -0.5 0 0.5 B Figure 5-6. Detected baseband sp ectrum. A) The subject was 0.5 m away from the antenna. B) The subject was 2 m away from the antenna (Insets: detected baseband signal [volt] versus time [sec]) The 5 GHz radar was tested in a lab envir onment for vital sign detection of a human subject of 1.75 m height. The measurement was perf ormed when the subject was seated at 0.5 m, 1 m, 1.5 m, and 2 m away and facing the antenna. As shown in Fig. 5-6, both the respiration and the heartbeat components were successfully dete cted in the baseband spectrum. By comparing the detected spectrum with those in Fig 2-3, it is verified that the 5 GHz transceiver has less harmonic interference than the Ka -band transceiver due to its l onger wavelength at lower carrier frequency. 5.3 A 5 GHz Direct Conversion Software Configurable Radar Sensor 5.3.1 CMOS 5 GHz Direct Conversion Radar Receiver To lower the cost and embed into portable devices such as cell phones, Doppler radar sensing function is also integr ated in BiCMOS and CMOS mi crochips. Examples include 2.4 GHz quadrature transceivers in 0.25 m BiCMOS and CMOS processes [8] and 5 GHz double sideband transceiver in 0.18 m CMOS process [ 18]. However, the reported works were limited 65

PAGE 66

to radio frequency integrations demonstrating the feasibility of on-chip Doppler radar for vital sign detection. External bias circuits, filters, and baseband amplifiers were required for detection. Also, it was difficult to configure the chips for environment variations including temperature and detection range changes. In this section, a software-c onfigured 5.8 GHz radar sensor integrated in UMC 0.13 m CMOS process is reported [19]. Powered by a 1.5 V AA battery, no external biasing or signal processing circuits is needed fo r operation. A 3-wire digital program /control unit is integrated in the receiver to configure it for different detection range. All th e operation points are also digitally controlled for the ease of embedding into portabl e devices such as a cell phone. The radar sensor chip has 1 GHz radio bandwidth, and can be tuned for optimal performance for people with different physiological moveme nt amplitudes. Important desi gn issues including the baseband flicker noise and gain budget will be discussed. 5.3.2 System Design Considerations The block diagram of the direct conversion 5.8 GH z radar sensor receiver chip is shown in Fig. 5-7. The receiver chain c onsists of a low noise amplif ier (LNA), a gain adjustable preamplifier (PreAmp), two passive mixers for I/ Q channels, and two vari able gain amplifiers (VGA). A Gm-boosted bias circuit combining co nstant-Gm and bandgap references performs temperature compensation for the LNA and preamp lifier. A bandgap voltage reference circuit is used to bias the mixer at the minimum flicker noise operation state. The VGAs are biased by a constant-Gm circuit. To make the receiver chip so ftware-configurable, a 3-wire digital program/control unit is integrated on-chip to decode th e digital control codes. The control bits ca n be programmed into the chip to set the operating point and receiver gain. A co mplex programmable logic device (CPLD) board is used to communicate with the chip. 66

PAGE 67

Gm Boosted Bias Bandgap Constant Gm LNAPreAmpMixerVGA 3 Wire Control Unit RFin ClockDataLoadQI DAQ CPLD 5.8 GHz Radar Receiver Figure 5-7. Block diagram of the software configurable 5.8 GHz radar sensor chip. For vital sign detection, a continuous wave unmodulated radio signal is transmitted toward the subject under test. The refl ected signal is modulated by the physiological movement caused by respiration and heartbeat, and is captured by the receiving antenna. The receiver chip downconverts the received signal with the transmitted carrier as the re ference, amplifies it, and sends it into the computer through a data acquisition module (DAQ). Applying the two-way Radar Ra nge equation to the non-contact vital sign detection, the received signal power Pr varies as the distance R between the subject and antenna changes: 2 34(4)rttrPPGG R (5-1) where Pt is the transmitted signal power, Gt and Gr are the gains of the transmit/receive antenna, is the radar cross secti on of the human subject, and is the wavelength. The received signal power is reduced by 12 dB as the distance betw een the radar detector and the subject doubles. For relatively constant signal amplitude at the ba seband output, adjustable receiver gain with large dynamic range is beneficial in real app lications. A digitally controlled variable gain amplifier is implemented to provide desired dynamic range. 67

PAGE 68

On the other hand, the unlicensed 5.8 GHz industr ial, scientific, and medical (ISM) band is popular for many wireless applicatio ns. The non-contact vital sign detector is designed for use in home or hospital environments with potential radi o interferers. It is necessary to avoid in-band interference from saturating the radio frequenc y front-end. Therefore, th e preamplifier between the low noise amplifier and the mixer was designe d to have high/low gain modes with different input referred P1dB. Another issue of non-contact v ital sign detection is the in terference to heartbeat signal caused by the harmonics of respiration signal. The harmonics are gene rated by the non-linear Doppler phase modulation [13] and the non-linearity of the electroni c circuit. Differential circuit can effectively reduce the even order harmonics. Therefore, the preamplifier was also designed as an active single-to -differential balun. Based on the above considerations, the receiver system specifications were determined and listed in Table 1. Because the down-converted vita l sign signal has a very small bandwidth of less than a few Hertz, the receiver noise figure is strongly affected by the transistor flicker noise. Since the receiver was designed to be directly sampled by a DAQ (NI-6008) with 12 bits ADC, the integrated baseband noise volta ge was designed to be less th an the rms quantization error, which is 0.12 V2. More on the specifications and design issues of low frequency baseband noise will be discussed in the next section. Table 5-1. System specifications (power referred to 50 Ohm) Current [mA] Gain [dB] P1dB [dBm] NF [dB] LNA 4.7 25 >-28 2.2 PreAmp LG 10 0 0.5 17 PreAmp HG 11 6 -9 7.8 Mixer+Buf 2 -5 2.8 VGA 0.6 6/12/18 0/-7/-14 RX Chain 20~21 26~44 >-29 68

PAGE 69

5.3.3 Detailed Circuit Design Fig. 5-8 is the simplified schema tics of the receiver chain without showing bias circuits and control units. 5.3.3.1 LNA and PreAmplifier Because the optimal carrier frequency varies for people with different body types and physiological movement strength, it is desirabl e to have a wide radi o frequency tuning range. Resistors in series with the loading inductors are used in the LNA and preamplifier to broaden the RF output bandwidth. At the input of the LNA, a pad with large parasitic capacitance (210 fF) is used to form a T matching network with the RF bonding wire Lb and part of the on-chip inductor Lg, achieving an input bandwidth of larger than 1 GHz. The single-ended cascode architecture is used for the LNA. The preamplifier is designed for two purposes First, it acts as an active single-todifferential balun. Second, it prov ides low/high gain options to extend dynamic range. When large in-band interference is present or the subject is close (<0.5m) to the antenna, the preamplifier switches to the lowgain mode with high input P1dB of 0.5 dBm. To realize this, two cascode amplifier stages are placed in parallel to drive th e output load. The switch between the high/low gain cascode is controlled by bits D1 and D2 in Fig. 2. Feedback capacitors Cb1 and Cb2 are added in the cascode amplifier to produc e differential output [29] For the high gain cascode, a fine-tuned Cb2 can achieve better than 0.4 dB amp litude imbalance. For the low gain cascode, however, it is difficult to achieve desi red performance with only a feedback capacitor due to the low gain of the input tran sistor. Therefore, a parallel LC tank (L3C3) is implemented to provide high resonant impedan ce at the source of the low-gain cascode. Source degeneration resistors R4 and R5 are used to improve the linearity of the low gain cascode stage. 69

PAGE 70

70sLgLLR b 1C b 2C3C3L1D2D4R5R RFin1D2Dint int LOI+LOI LOQ LOQ LNAPre-AmplifierMixer1B3B2B4B 5BBuffer,VGA Buffer,VGA1 2High Gain Enable Low Gain Enable : : D D Low Gain High Gain3D4D Out-Out+In-In+ b ufB b ufB s RaRbR VGA L0P b VGA Figure 5-8. Schematic of the receiver chain. D1-D4: gain control bits. B1-B5: voltage/c urrent bias generated by the on-chip bia s circuitries.

PAGE 71

A source follower buffer is used to drive the passive mixer in voltage-driven mode since the mixer uses large transistors to achieve bette r noise performance, as discussed in the next section. The source follower was chosen because it improves the isolation and is robust against load variation and process-voltage-temperature variations. 5.3.3.2 Mixers For Doppler radar detection of human vita l signs, the baseband signal bandwidth is typically less than a few Hertz. Flicker noise do minates at these frequencies, posing a challenge for the noise performance of the mixer in the direct conversion receiv er. Conventional GilbertType active mixers contain severa l noise sources: the transconducto r noise, the LO noise, and the noise from the switching transistors. These noise sources establish an unacceptable noise figure in the interested baseband spectrum. Therefore, the Gilbert-Type active mixers are not suitable for the vital sign detection rece iver. Passive mixers avoid the transconductor stage in the active mixers and have no DC bias current. This featur e minimizes the flicker noise at the mixer output. Therefore, passive mixer was chosen for direct conversion vital sign detections. In order to minimize the flicke r noise of the passive mixer, the gate-source voltage of the switching transistor should be close to Vth. This break-before-make bias technique will minimize the DC bias current in the switching transistors. Fo r the switching transistor size, there is a tradeoff between noise figure and capacitive load to the preamplifier stage. To provide appropriate noise figure in the interested vi tal sign bandwidth, large switchi ng transistors are used and they produce relatively large cap acitive loads to the preamplifier. The LO signal power is set to 1 dBm for maximum gain based on simulation. 5.3.3.3 Variable Gain Amplifiers The variable gain amplifier s hown in Fig. 2 is designed to pr ovide three-stage gain from 6 dB to 18 dB. To reduce the baseband flicker noise, large differential pair with resistor loads is 71

PAGE 72

used. The maximum gain is obtained when both D3 and D4 are high. When D4 is low, Rs degenerates the differential-mode signal and redu ces the gain by 6 dB, while improving the input P1dB by 7 dB. When D3 is low, the load resistors Ra and Rb are shorted and the gain is further reduced by 6 dB. The VGA outputs are sampled by a 12-bit DAQ (NI-6008) through the differential sampling channels. A source follower is used as buffer and le vel shifter at the VGA output. The VGA input referred noise voltage is designed to be le ss than 744.1 nV/sqrt(Hz) at the frequency of 1 Hz, so that the output noise in tegrated over the vital sign signal bandwidth is much smaller than the DAQ quantization error. 5.3.3.4 Bias Circuits All the bias circuits are on-chip and so ftware configurable through the 3-wire program/control unit. Typically, each bias curren t/voltage has 4 optional values so that the receiver chip can be digitally tuned for best performance under process variation. A bandgap bias circuit was designed for the mixers, as shown in Fig. 5-9. And a constant-Gm circuit shown in Fig. 5-10 was designed for the VGA. out1Vout2V1D2D Figure 5-9. Bandgap bias circuit for the mixer core and mixer buffer. 72

PAGE 73

Iout Figure 5-10. Constant-Gm bias circuit for the variable gain amplifier. IoutPTATConst-Gm Figure 5-11. Schematic of th e Gm-boosted bias circuit. The bias of the radio frequency front-end is very important for the performance of the radar chip. The LNA and preamplif ier share the same Gm-boosted bi as circuit, as shown in Fig. 5-11. A straightforward choice is the classic constant-Gm circuit, whose transconductance depends on the reference resistor only. However, the sheet resistances of on-chip resistors also have a temperature coefficient and may reduce the transconductance as temperature increases. To compensate for this temperature dependen ce, bandgap and constant-Gm circuits are used 73

PAGE 74

simultaneously, and a current subtra ctor is used to subtract a d ecreased-with-temperature (DWT) current from the current of the constant-G m circuit. As a result, the transconductance corresponding to the output re ference current is boosted to achieve better temperature performance for the radio front-end. 5.3.4 Experiments The software configurable radar sensor re ceiver chip was fabricated in UMC 0.13 m process. Fig. 5-12 shows the microphotograph of the fabricated chip. A circuit board with a Xilinx XC9536 CPLD was built to program the chip. The high-speed XC9536 can program the radar chip at 10 Mbps, minimizing the configur ation time. Fig. 5-13 shows the experimental setup. The sensor chip can be powered by a commercial 1.5 V battery. inRFoutI+outI-outQ+outQ-LoadClkDataQLO+QLO ILO+ILO LNA Pre-Amp MixerI MixerQ VGA I VGA Q Gm-boosted bias Bandgap Const-Gm 3-wire control Figure 5-12. Chip microphotograph of the 5.8 GHz radar sensor receiver (1.2mm 1.2 mm). Experiments have been performed in lab enviro nment to detect the vital signs of seated people. Without any external baseband amplifier, the radar sensor chip can detect the vital sign of human subjects at a distance of up to 1.5 m. Th e detection range can be further increased if 74

PAGE 75

higher transmitted power, external baseband amplif ier or DAQ with more ADC bits are used to increase signal noise ratio. DAQ Sensor ICCPLDBatter y Tx AntRx AntLabVIEW Interface Figure 5-13. Experiment setup for non-contact vital sign detection. Because the radar chip has I/Q channels, co mplex signal demodulation [20] was used for the baseband signal processing, which helped eliminate the null detection point problem while reducing the cost by demodulati ng in software. Figs. 5-14 and 5-15 show the baseband signal spectrum detected at 0.5 m and 1.5 m from the b ack/front of the human subject respectively. Clean baseband time domain signals were obser ved. The respiration a nd heartbeat signal components can be picked up from the spectrum, indicating a respirat ion rate of around 17 beats/min and a heartbeat rate of around 65 beats/min. 75

PAGE 76

0 5 10 15 20 25 -20 0 20 Time (Second)I/Q Signal [mV] 0 20 40 60 80 100 120 0 0.5 1 Beats/MinCSD Spectrum Respiration Heartbeat A B Figure 5-14. Detection from the back of a human subject at 0.5m away. A) Baseband signal. B) Complex signal demodulated (CSD) spectrum. 0 5 10 15 20 25 -40 -20 0 20 Time (Second)I/Q Signal [mV] 0 20 40 60 80 100 120 0 0.5 1 Beats/MinCSD Spectrum Respiration Heartbeat A B Figure 5-15. Detection from the front of a human subject at 1.5m away. A) Baseband signal. B) Complex signal demodulated (CSD) spectrum. Both the time domain waveform and the spectrum show that the heartbeat signal component is more pronounced when detecting from the back of the human body, which is consistent with the analysis in [13]. It is al so interesting to observe that the time domain 76

PAGE 77

waveforms present less noise th an those measured by the 5 GHz double sideband transceiver chip in [18]. This can be explained by the im proved designs of the mixer and variable gain amplifier for noise performance and the fully-int egrated on-chip bias a nd amplifier circuitry presented in this chapter. 5.4 Conclusion Two Doppler non-contact vital sign detection rada rs with different radio architectures have been integrated in single chip using UMC 0.18 m and UMC 0.13 m processes respectively. The first integrated double-side band vital sign radar sensor ch ip is demonstrated using 0.18 m CMOS process. The double-sideband transmissi on and detection architecture eliminates the need for image rejection. The differential design shows the advantages of reducing LO leakage. The 5 GHz double-sideband system has the advant age of avoiding null detection point by its wide frequency tuning range, and has less harmon ic interference than the previously reported Ka-band system. A software configurable 5.8 GHz radar sens or chip was fabricated in UMC 0.13 m CMOS process. Measurements demonstrated it can successfully detect h eartbeat and respiration signals without any external analog circuits. Important design issues such as the effects of baseband flicker noise and gain budget have been discussed. The reduced baseband noise compared to the results measured by previously re ported integrated chip indicates the advantages of system-on-chip integration and improved design for better flicker noise performance. 77

PAGE 78

CHAPTER 6 COMPLEX SIGNAL DEMODULATION AN D RANDOM BODY MOVEMENT CANCELLATION IN DOPPLER RA DAR VITAL SIGN DETECTION 6.1 Introduction As one of the main challenges for Doppler radar non-contact vital sign detection, the null detection point problem was solved by the freque ncy tuning technique [10] and the arctangent demodulation [30]. However, the frequency tuni ng technique requires tu ning the intermediate frequency once the distance between the antenna and the subject changes; and the reported arctangent demodulation requires ca libration of the baseband DC offset [31]. Unfortunately the DC offset for non-contact vital sign detection is not only produ ced by the electronic components, but also by the unmodulated reflected signal, i.e. signal reflected from stationary objects and other parts of the human body rather than the moving chest wall. Therefore, the DC offset calibration is more than just calibrating the recei ver chain. It is required once the experimental environment changes. Therefore, a complex signa l demodulation technique is invented in this chapter for non-contact vital si gn detection. This technique can reliably eliminate the null detection point problem at a low cost. Moreover, since the objective of vital sign detection is to identify the frequency of the desired signal components, which are not related to the DC component, the complex signal demodulation does not need DC offset information when no random body movement is present. Besides the demodulation tec hnique to eliminate the null detection point problem, the distinctive properties of non-cont act vital sign detecti on still require further attention in the following aspects. First, non-contact vital sign sensing is based on detecting the periodic phase change. Since the detection is typically carried out within a few meters, the phase offset accumulated in the transmission path varies significantly from one path to another, making the detection potentially sensitive to the near field effect [32] [33]. 78

PAGE 79

Moreover, since the non-contact vital sign detection is based on sensing the small physiological movement in millimeter or centim eter range, the presence of random body movement produces a significant source of noise for the accurate detection. Although successful Doppler radar non-contact vital sign detection under different environment for different applications has been reported in the past three de cades [1]-[10], a critical issue that prevents this technology from being widely used is the NOISE PRODUCED BY RANDOM BODY MOVEMENT. For example, when the radar was used for overnight monitoring of sleep apnea in clinical environment, the detected vital sign signa l will be interrupted when the subject under test is rolling over on the bed, producing false alarm. It will be pointed out in this chapter that the random body movement can be cancelled out by dete cting from both the front and the back of the human body using complex signal demodulati on. This cancellation technique can be widely adopted for applications such as sleep apnea mo nitor, lie detector, and baby monitor. In these applications, two identical portable radars can be conveniently implemented under and above the bed, or at a distance from the front and the back of a seated subject. In this chapter, the complex signal dem odulation and the arctangent demodulation are comparatively studied for random body movement cancellation [34]. It is shown that if the baseband DC offset information is known, both of the two demodulation techniques can be used for random body movement cancellation. When the DC offset cannot be accurately calibrated out, the complex signal demodulation is more favorable for random body movement cancellation. The ray-tracing model used in the st udy will also show the e ffects of constellation deformation and optimum/null detection ambiguity caused by the phase offset due to finite antenna directivity. 79

PAGE 80

The two demodulation techniques and thei r implementation for random body movement cancellation are studied in the next section. The effects of phase offset are discussed subsequently. Simulations have been performe d and the results are reported in the section thereafter. Supporting experimental results are presented followi ng the simulation results, and a conclusion is drawn in finally. 6.2 Complex Signal Demodulation and Arctangent demodulation Fig. 6-1 shows the block diagram for Doppler radar non-contact vital sign detection. The measurement can be performed from either th e front or the back of the human body. When random body movement is present and it affects accurate detection, the m easurement has to be performed simultaneously from both sides to cancel out the random frequency drift. Body ResHeart f I f Q 090 bIbQ DAQ DAQ Demodulation R es 090 Figure 6-1. Simplified block diagram for Doppl er radar non-contact vital sign detection. In the analysis of non-contact quadrature demodulation of vital sign, the single-beam model assumes an ideal antenna with infinite di rectivity focusing a beam at the location of the heart. When no random body movement is present, the normalized detected baseband signal in one of the baseband I/Q channels can be represen ted and analyzed by spectral analysis described in Chapter 2: 4() 4 4() 4 ()cos( )()()cos( )hh rr lkrh klxt m xt m Bt JJktlt (6-1) 80

PAGE 81

where xh( t ) = mhsin ht xr( t ) = mrsin rt are the periodic body move ments due to heartbeat and respiration, is the wavelength of the wireless signal, is the total residual phase accumulated in the circuit and along the transmission path, and Jn is the Bessel function of the first kind. For a quadrature transceiver, the baseband output in the I / Q channel can be represented as B ( t ) and the quadrature of B ( t ). Meanwhile, the Bessel coefficient with a negative index number and a positive index number in (6-1) can be combined using the property: Jn ( x ) = J n ( x ) for even numbers of n and Jn ( x ) = -J-n ( x ) for odd numbers of n. Therefore, the baseband I/Q output can be represented as: I1001 20024() 4() ()cos( ) 2sin()+sin()+sin2cos(2)+cos(2)+cosh r rh rhxt xt It DCCtCt CtCt (6-2.a) Q1001 20024() 4() ()sin( ) 2sin()+sin()+cos2cos(2)+cos(2)+sinh r rh rhxt xt Qt DCCtCt CtCt (6-2.b) where Cij = Ji(4 mr/ ) Jj(4 mh/ ) determines the amplitude of every frequency component, DCI = J0(4 mr/ ) J0(4 mh/ )cos and DCQ = J0(4 mr/ ) J0(4 mh/ )sin are the DC components of the signals in I and Q channels, respectively. The ellipses in (6-2) represent hi gher order odd and even harmonics. It should be noted that DCI and DCQ are the DC information directly related to the physiological movement. As will be shown in the following sections, they are beneficial for the recovering of vital sign signals in some cases and thus can be defined as the desired DC information. In real experiment, due to the DC offset caused by the signals reflected from environmental stationary objects (c lutter) and the DC offset accumulate d in the electronic circuit, the measured baseband DC level is the DCI/ DCQ level plus an undesired DC offset. From (6-2), the ratio of cos and sin determines the relative strength between the even order and the odd order harmonics. Therefore, the optimal/null detection point is determined by 81

PAGE 82

the residue phase For example, when is close to 900, the fundamental frequency of respiration and heartbeat signals dominates in the I channel while the second order harmonic of desired signals dominates in the Q channel, thus I is close to the optimal detection point and Q is close to the null detectio n point. According to the single-beam model, when either one of the two quadrature channels is close to an optimal det ection point, the other one should be close to the null detection point. 6.2.1 Complex Signal Demodulation The complex signal demodulation can eliminat e the optimum/null det ection point problem by combining the I and Q signals in baseband. As shown in Fig. 6-2 (a), the complex signal is software-reconstructed in real time as: jj IQ 10 01 20 024() 4() ()()()exp 2sin()+sin()+e2cos(2)+cos(2)+eh r rh rhxt xt StItjQtj DCjCtCtCtCt (6-3) where DCIQ = DCI + j DCQ. Since ej has a constant envelope of one, the effect of on signal amplitude can be eliminated. Applying the complex Fourier transform to the signal S( t ) for spectral analysis, the residual phase will not affect the relative st rength between the odd order and the even order frequency components. The de sired signal components ( odd order tones) will always be present in the spectrum. Meanwhile, even though DC offsets exist in the I/Q channels and may lead to the error of measured DCI/ DCQ, they only affect the DC term of S( t ), i.e. DCIQ in (6-3). Therefore, the existence of DC offset does not affect obtaining the frequency of the de sired signal components. In practice, the residual baseband DC level, which is a sum of the physiological-movementrelated DCI/ DCQ and the DC offset, can be easily extrac ted as the average of signals in every time-domain sliding window and thus be safely removed. As a result, the complex signal 82

PAGE 83

demodulation greatly simplifies the demodulation procedure and is immune from DC offset when no random body movement is present. Howe ver, the complex signal demodulation is not able to completely eliminate th e higher even order harmonics. I DC Q DC ReIm 0 50 100 0 0.5 1 FFT Complex() St A IQ 0 50 100 0 0.5 1 FFT DCCalibration arctan Q I 0180 Fn () t B Figure 6-2. Demodulation block diagram. A) Complex signal demodulation. B) Arctangent demodulation. For random body movement cancellation, measuremen ts need to be performed from both sides of the human body. In this way, the signal detected from the two transceivers can be expressed as: 1 1 14() 4() 4() ()exp h r fxt xt yt Stj (6-4.a) 2 2 24() 4() 4() ()exph r bxt xt yt Stj (6-4.b) where xh1( t ) and xr1( t ) are the heartbeat-induced and the respiration-induced physiological movements on the front chest wall, xh2( t ) and xr2( t ) are the heartbeat-induced and the respirationinduced physiological movements on the back, 1, 2 are the residual phase of the transceivers in front of the body and behind the body, and y ( t ) is the random body movement. Note that the pairs of physiological movements on both sides of the body, e.g. xh1( t ) and xh2( t ), move in the same direction relative to their respec tive detecting radar. On the other hand, when the body is drifting 83

PAGE 84

toward one of the radars, it is moving away from the other one. Therefore, by multiplying Sf( t ) and Sb( t ), the y( t ) term in the baseband output Sfb( t ) = Sf( t ) Sb( t ) will be cancelled out, while the physiological movement terms are enhanced: 1212124()()4()()()exphhrrfbxtxtxtxtStj (6-5) The above operation can also be interpre ted as convolution and frequency shift in frequency domain, thus canceli ng the Doppler frequency drift and keeping only the periodic Doppler phase effects. Although it is shown that the complex signal de modulation itself does not require the DC information and is immune to the DC offs et, the performance of random body movement cancellation is affected by the DC offset. Proper estimation or ca libration of the DC offset is beneficial for successful cancellation of the noise due to random body movement. 6.2.2 Arctangent demodulation Another way to eliminate the optimum/null detection point problem in the quadrature demodulation system is to use arctangent de modulation [30] by calcula ting the total Doppler phase shift. Its block diagram is shown in Fi g. 6-2 (b). Taking into account the phase discontinuity when the signal trajectory cros ses the boundary of two adjacent quadrants, the arctangent demodulation calculate s the total angular information ( t ) as: 4() 4() () ()arctan ()h rxt xt Qt tF It (6-6) where F is a multiple of 1800 for the purpose of elimina ting the discontinuity when ( t ) crosses the boundary of two adjacent quadrants in the constellation graph. Because ( t ) is a linear combination of the desired signal xh( t ) and xr( t ), the information of the vital signs can be obtained with the non linear phase modulation effect eliminated. The 84

PAGE 85

advantage is the ability to eliminate the harm onic and intermodulation interference. However, previous demonstration [30] of this technique re quires accurate calibration of the DC offset in order to properly reconstruct the angular information. The difficulty of accurate DC offset calibration encountered in Doppler radar vital sign detection is that the DC offset is not only produced by the electronic circui t, but also by the unmodulated reflected signal, i.e. signal reflected from stationary object s and other parts of the human body rather than the moving chest wall. Therefore, the DC offset changes as the environment changes and needs to be calibrated when it is changed. On the other hand, the presence of baseba nd DC offset results in an error in DCI/ DCQ defined in (6-2). This leads to a shifted tr ajectory in the constellation graph. Although the angular information ( t ) will be changed significantly when the trajectory is shifted, the angular movement is still periodic. This implies that when analyzing the spectrum of ( t ) in the presence of a DC offset, the desired frequency component s still exist. The difference observed in the spectrum is a changed harmonic level. Therefore, if the DC offset can be properly estimated, it is still possible to extract the desired vital signs without a DC offset calibration. As will be demonstrated in Section V, a traj ectory-fitting procedure is adopted in this chapter for DC offset estimation in the baseband. Experiments will show that this procedure can be used for vital sign detection in the absence of random body movement. When random body movement is present, the an gular information recovered from the front ( f) and the back (b) of the human body can be expressed as: 1 1 14() 4()4() ()h r fxt xtyt t (6-7.a) 2 2 24() 4()4() ()h r bxt xtyt t (6-7.b) 85

PAGE 86

where xh1( t ), xr1( t ), xh2( t ), xr2( t ), 1, and 2 are the same as defined in Section II-A. Instead of multiplying the two signals as in the case of using complex signal demodulation, the random body movement can be cancelled out by adding the angular information of (6-7.a) and (6-7.b) together to obtain fb( t ) = f ( t ) + b( t ): 1212 14()()4()() ()hhrr fbxtxtxtxt t2 (6-8) As a result, the Doppler frequency dr ift caused by the random movement of y(t) is eliminated, leaving only the desired fre quency components in the spectrum of fb( t ). The critical issue of using arctange nt demodulation for random body movement cancellation is how the presence of the baseband DC offset will affect the detection. As will be shown in simulation and experiments, the comple x signal demodulation is more favorable for random body movement cancellation when the DC offset cannot be accurately determined and calibrated. 6.3 Effects of Phase Offset Since a real antenna with a cer tain radiation pattern does not have infinite directivity, signals are reflected and capture d from different parts of the body. When signals on different paths with different intensity and residual phases are received by the radar, they are simply summed together by the receiving antenna, either canceling out or enhancing the desired signal components. Therefore, a ray-tracing model [16] is developed to comp ensate for the shortage of the single-beam model As shown in Fig. 6-3 (a), th e actual received signal shoul d be represented from a raytracing point of view as: 1/2 2 2 04 ()(,)cos(,)(,)sin()(,)sin()hhrr s I tExy xydmxytmxytds (6-9.a) 86

PAGE 87

1/2 2 2 04 ()(,)sin(,)(,)sin()(,)sin()hhrr s QtExy xydmxytmxytds (6-9.b) where E ( x, y ) is the intensity of the electromagnetic wave reflected from the location ( x, y) of the human body, d0 is the distance between the an tenna and the subject under test, ( x, y) is the distance between location ( x, y) and the antennas projec tion on the chest wall, is the residual phase shift accumulated in the electronic circuit. The integration is carried out over the entire subject under test. (,) x y B A 0dd Z X Y I ChannelQ Channel A B A B Figure 6-3. Ray-tracing model for 5.8 GHz radar application. A) Human body radiated by an antenna. B) Angular information of signals reflected from point A and B on the body. The antenna is facing the body in the Z direction of the X Y Z coordinate. Assume the antenna is placed 1 m away in front of the heart center, and the locations of the heart center A and the body center B on the front chest wall are separated by 11 cm. The difference in the transmission path for signals from the antenna to the two points is 2210.1110.006m x which is replicated in the rece iving path and would produce a phase difference of 83.5 degree for a 5.8 GHz radar. M eanwhile, the radiation in tensity of the antenna on the body surface is different from point to poin t, depending on the antenna radiation pattern. This implies that the received baseband signals fr om the two points will have different locations and movement patterns in the constellation graph, as shown in Fig. 6-3 (b ). Therefore, the real 87

PAGE 88

case for vital sign detection using complex signa l demodulation and arctangent demodulation is complicated by the phase offset. Numerical simula tions are needed and will be presented in the following section. 6.4 Simulation Simulations have been performed based on the ray-tracing model The two demodulation techniques were applied to vital sign dete ction in the presence/absence of random body movement. 6.4.1 Ray-tracing Model A B C D Figure 6-4. Ray-tracing model for random body movement cancellation. A) The phase offset on the surface of human body radiated by a 5.8 GHz radar. B) A 7 by 7 elements antenna arrays radiation intensit y on the human body. C) Approximation of the normalized amplitude of body movement caused by re spiration. D) Approximation of the normalized amplitude of body movement caused by heartbeat. The body model for a subject of 1.8 m height is shown in Fig. 6-4. Assuming the antenna is 1 m in front of the heart center, the phase offset in different paths compared with the beam 88

PAGE 89

propagating to the center of the heart is show n in Fig. 6-4 (a) for a 5.8 GHz radar sensor. Dramatic change in phase offset is observed. Sh own in Fig. 6-4 (b) is the radiation intensity on the human body produced by an ideal 7 by 7 an tenna array comprised of omnidirectional antennas spaced by /2. Fig. 6-4 (c) and (d) are the approxi mation of the normalized amplitude of body movements caused by respiration and heartbeat, respectively. It can be inferred that when a carrier frequency of 24 GHz is used for the high er sensitivity at shorte r wavelengths, the phase change will be more significant. 6.4.2 Demodulation without Random Body Movement To demonstrate the properties of the two demodulation techniques, numerical simulations were first performed without ra ndom body movement present. Two examples are presented, i.e. a 5.8 GHz quadrature radar, and a 24 GHz quadr ature radar. Three types of signals were recorded and analyzed. Case I: a single-beam signal projected to the heart center, i.e. point A in Fig. 6-3. This is the case analyzed by the single-beam model Case II: a single-beam signal projected to the bo dy center, i.e. point B in Fig. 6-3. In this case, respiration signal was picked up wh ile heartbeat signal is almost absent. Case III: the actual signal transmitted and received by the radar. It should be noted that only Case III can be realized in the laboratory. Case I and II analyze signals carried by a hypotheti cal single beam radiated by an ante nna with a very high directivity radiation pattern. 6.4.2.1 Example I: 5.8 GHz Quadrature Radar Simulation results are shown in Fig. 6-5 for detection from the back of the human body. The residual phase produced in the electronic circuit was assumed to be 00, which means the Q 89

PAGE 90

channel was at the optimum detection point while the I channel was at the null detection point according to the single-beam model. A CaseICaseII -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 I ChannelQ Channel B I ChannelQ Channel CaseIII C 0 2 4 6 8 10 12 30 40 50 60 Time [sec](t) [deg] D 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum Arctangent Demodulation Complex Signal Demodulation Figure 6-5. Demodulation for a 5.8 GHz radar. A) Signal detected at heart center (Case I) and at body center (Case II); B) Actu al received signal (Case III); C) Angular information ( t ) of the received signal; D) Baseband spectra obtained by the complex signal demodulation and the arctangent demodulati on (the DC component is not shown in the baseband spectrum). Figs. 6-5 (a) and (b) show the si gnal trajectories in the constellation graph. As predicted in Section II, signals reflected from differen t parts of the human body are affected by two variations: the phase offset and the radiation intensity. The form er variation embodies itself as different angles of the trajectory shown in Fig. 6-5 (a), while the latter is demonstrated as 90

PAGE 91

different radii of the trajectory. As a result, when the receiver receives the vital sign signals, which is the superpositio n of all the signals reflected from different parts of the body, the total received signal trajectory is deformed from an id eal circle, as shown in Fi g. 6-5 (b). It should be noted that the constellation deformation is not caused by noise, which was not included in simulation. Although the foregoing discussi ons and simulation results appear undesirable, the recovered angular information ba sed on (6-6) is nonetheless period ic and not seriously disturbed by the phase offset problem, as shown in Fig. 65 (c). The spectrum of the complex signal and the recovered angular information were analy zed and plotted in Fig. 6-5 (d). Although the detection was made with one channel at the nu ll detection point and th e other at the optimum detection point, both of the two demodulation te chniques can successfully identify the respiration and heartbeat components. Therefore, the complex signal demodulation an d the arctangent demodulation for 5.8 GHz radar system are demonstrated to be effectiv e solutions to achieve reliable detection and eliminate the null detection point problem. 6.4.2.2 Example II: 24 GHz Quadrature Radar In this example, the carrier frequency was 24 GHz and the residual phase produced in the electronic circuit was assumed to be 450, which means the detection was performed at the middle between the null and the optimum detection points. The constellati on plots are shown in Fig. 6-6 (a) and (b). Due to fast variation of the phase offset on the surface of the human body, more severe trajectory deformation was observed. However, angular information recovered from (6-6) is still periodic, as shown in the inset of Fi g. 6-6 (b). The baseband spectra from the two demodulation techniques are shown in Fig. 6-6 (c). Again, the respiration and the heartbeat components can be identified from the sp ectrum by using both of the techniques. 91

PAGE 92

A -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 I ChannelQ Channel CaseICaseII B I ChannelQ ChannelCaseIII 0 5 10 0 100 200 Time [sec] () t C 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum Angular Demodulation Complex Signal Demodulation H2 Int Int Heartbeat Figure 6-6. Demodulation for a 24 GHz radar. A) Signal detected at heart center (Case I) and at body center (Case II). B) Actual received signal (Case III), with the recovered angular information shown in inset. C) Baseba nd spectra obtained by the complex signal demodulation and the arctangent demodulation (DC component not shown in the spectra). The complex signal demodulation causes harmonic (H2) and intermodulation (Int) interference. A 0 20 40 60 80 100 120 0 0.5 1 Normalized Spectrum I Channel Q Channel B 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum I Channel Q Channel Figure 6-7. Baseband spectrum detected by the I and the Q channels with a carrier frequency of 24 GHz. A) Spectrum of a single-beam signal projected to the heart center. B) Spectrum of the actually received signal. 92

PAGE 93

Furthermore, the result in Fig. 6-6 (c) veri fies that the arctangent demodulation can eliminate the harmonics and intermodulation terms caused by the nonlinear phase modulation effect, making the spectrum cleaner than th at obtained by complex signal demodulation. Another phenomenon to be noted is the optimum/null det ection ambiguity Shown in Fig. 6-7 (a) is the baseband spectrum of the I and the Q channels in Case I, i.e. the spectrum of the single beam signal projected to th e center of the heart. The p eaks of the respiration and the heartbeat components in I channel have the same amplitudes as those in Q channel, which is in accordance with (6-2) predicted by the single-beam model since the detection was performed at the mid-point between the null and the optimum detection points. However, the baseband spectrum of the actual received signal, as shown in Fig. 6-7 (b), shows that the I and the Q channels have significant differences in the heartbeat signal strength. While the I channel preserves the heartbeat signal, the Q channel shows strong harmonic and intermodulation components. This is because of the enhancem ent and cancellation among signals with different phase offsets. It demonstrates the necessity of effectively combining the two channels even when the detection is not carried out at the null detection point. 6.4.3 Random Body Movement Cancellation The random body movement cancellation technique was also simulated using the raytracing model with a carrier frequency of 5.8 GHz. The random roaming of the body was fully modeled in three dimensions ( X Y and Z ) which are defined in Fig. 6-3. Typically the subject under test has larger random body movements in tw o dimensions than the third dimension, e.g. the horizontal movements in the X and Z directions are more obvious than the vertical movement in the Y direction for a seated pers on. Therefore, the time-var iant velocity of random body movement was modeled as uniform distribution between 0 and a maximum value of 4 mm/s in the X and the Z directions. The amplitude of random body movement in the Y direction was 93

PAGE 94

modeled as 0.1 of that in the other two direc tions. The movement compon ents in each direction are shown in Fig. 6-8 (a), and th e baseband spectra detected from the front and the back using the two demodulation techniques are shown in Fig. 6-8 (b). When random body movement is present, the desired respiration and heartbeat signal components will be overwhelmed by the noise generated by random body movement. 0 2 4 6 8 10 12 -1 -0.5 0 0.5 1 Time (s)Body Move (cm) Z X Y A 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum Front: AD Front: CSD Back: AD Back: CSD B Figure 6-8. Baseband spectra obt ained when random body movement is present. A) The random body movement is shown in the Z, X, and Y di rections, which are defined in Fig. 6-3. B) Baseband spectra by arctangent de modulation (AD) and complex signal demodulation (CSD). If the system can successfully calibrate out the DC offset up to the baseband output, the recovered baseband angular information and the spectra obtained by random body movement cancellation were simulated and shown in Fig. 6-9. The respiration and heartbeat components were successfully recovered by both demodulation techniques, which showed similar performance in recovering the desired signal co mponents. It should be noted that although the random body movement can exist in the direction perpendicular to the radar direction, this 94

PAGE 95

technique still works reliably because only the moveme nt in the radar directi on is critical for the detection. A 0 2 4 6 8 10 12 0 50 100 150 200 Time [ sec ] f(t), b(t), fb(t) [deg] From Front From Back RBMC B 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum Angular Demodulation Complex Signal Demodulation Heartbeat Figure 6-9. Random body moveme nt cancellation using the two demodulation techniques. A) Recovered angular information. B) Rec overed baseband spectrum. Accurate DC information is used in demodulati on but not shown in the spectrum. If the DC offset cannot be perfectly calibra ted out up to the baseband output, however, the performance of random body movement cancel lation based on both of the demodulation techniques deteriorates. Shown in Fig. 6-10 is an example when DC offset was present at the baseband output of the two transceivers. For each transceiver, the baseband DC offset levels were modeled to be the same in the I / Q channels and were 30% of the maximum signal amplitude. In the simulation, the above DC offset level was added to the ideal I and Q channel signals. Then, both demodulation techniques we re applied to cancel out the random body movement. It is shown that the complex signal demodulation can still iden tify the respiration and heartbeat components, but the arct angent demodulation is unable to recover the heartbeat signal. The reason for this disadvantage of using ar ctangent demodulation in random body movement 95

PAGE 96

cancellation is, as shown in (6-8 ), the cancellation is based on the linear combination of the calculated phase, which is st rongly affected by the location of the constellation origin. A 0 2 4 6 8 10 12 -50 0 50 100 150 200 Time [sec]f(t), b(t), fb(t) [deg] From Front From Back RBMC B 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum Angular Demodulation Complex Signal Demodulation Respiration Heartbeat Figure 6-10. Random body movement cancellation tech nique. A) Recovered angular information. B) Recovered baseband spectrum. The random body movements are modeled in three dimensions, and the DC offset in each transceiver is 30% of the maximum signal amplitude. 6.5 Experiment Experiments have been performed in the labora tory to verify the theory and compare the performance of the two demodulation techniques for random body movement cancellation. For consumer applications of this technique, it is desirable to have portable radars that can detect vital signs from several meters away, radiate a power of lower than 0 dBm, and have all the hardware integrated together at an affordable price. Therefore, 4 GHz portable radar was designed for this purpose. The radar integrates the quadrature transc eiver, the two-stage baseband amplifier, and the power management circuit on a single printed circuit board (Rogers RO4350B substrate) with a size of 6.8 7.5 cm2. The block diagram, photo, and specifications of the radio frequency component s are described in Chapter 4. 96

PAGE 97

Since the vital sign has a frequency less than several Hertz, large coupling capacitors C1 and C2 of 10 F were used to isolate the DC voltage s of the mixer output and baseband amplifier input. Because the 10 F coupling capacitors block the DC signal in addition to isolating DC voltages of two different circu its, a variable DC offset was inevitably introduced into the measurement. The coupling capacitor ( C1, C2 = 10 F) and the baseband amplifier input resistor ( R1, R2 = 160 k ) were chosen such that for a heartbea t signal with a frequency around 1Hz, the voltage drop on the capacitor is no more than 1/10 of the signal amplitude. This leads to a time constant of approximately 1.6 seconds, which means that in the real-time signal processing software, a 2 second initiation time is needed. Figure 6-11. Two identical transceivers used for random body movement cancellation. Inset: the antenna used for each transceiver. Note that one transceiver uses vertically polarized antenna array, while the other uses horizontally polarized antenna array. For random body movement cancellation, measurements were performed by two identical radars. As shown in Fig. 6-11, patch antenna arrays with orthogonal polar ization were installed in the two transceivers to eliminate the interfer ence between the two units. It was observed in the experiment that the antenna gain should be highe r than 4 dB for the radar to have a good signalto-noise ratio from up to 2 m away. The antenna was designed to have a maximum directivity gain of 9 dB at broadside, so that the vital si gns of the subject in front of the antenna will be picked up. Free-running VCOs were used for the two transmitters so th at the actual carrier 97

PAGE 98

wavelengths were close to each other but always had a slight difference in the absence of a phase-locked-loop. As a result, the signal from on e transceiver was further rejected by the other transceiver in the baseband because the small differ ence in the carrier frequency results in a large difference in baseband frequency compared to the vital sign frequencies. The phase noise reduction due to range correlation makes the free -running VCO adequate for vital sign detection [8]. To reduce the hardware cost and the requireme nt of signal processing speed, the amplified baseband signals were sampled by a 12-bit mu ltifunction data acquisition module (NI USB6008) with a low sampling rate of 20 Hertz, which is fast enough for the vital sign signal of typically less than 2.5 Hertz. The sampled da ta were fed into a laptop for real-time signal processing by LabVIEW. The sampling rate and resolution make it possible to implement the baseband signal processing in a low cost DSP mi crochip such as the TI C2000 family digital signal controllers for various applications in the future. To focus on the properties of demodula tion and random body movement cancellation techniques, no baseband filtering was implemented in either hardware or software. All the results presented are based on the original baseband signal. 6.5.1 DC Offset Estimation in Baseband Because of the coupling cap acitor in the radar between the receiver output and the baseband amplifier input and the variability of DC offset within the experimental environment, it is relatively difficult to accurately calibrate out the DC offset of the whole system. Instead, the DC offset was estimated by fitting the signal trajectory into a proper segment of circle in the constellation graph. 98

PAGE 99

A -1 0 1 -1.5 -1 -0.5 0 0.5 1 1.5 I ChannelQ Channel No DCWith DC B 0 20 40 60 80 100 120 0 0.5 1 Beats/MinNormalized Spectrum Complex Signal Demodulation Arctangent Demodulation Figure 6-12. DC offset estimation. A) Traject ory of detected baseba nd signal with no DC information and with estimated DC offset level subtracted. B) Sp ectra obtained by the two demodulation techniques. Signal with estimated DC offset subtracted was used for arctangent demodulation. For example, Fig. 6-12 (a) shows the constella tion graph of the baseband signal detected from the back of the human body when no random body movement was present. Because of the coexistence of the undesired DC offset and the desired DC information, i.e. DCI/ DCQ in (6-2), the original signal trajectory was located at the center of the constellation graph. After subtracting an estimated DC offs et level of 0.8 V for both the I and Q channels in the baseband, the trajectory was fitted into a circle. Fig. 6-12 (b) shows the baseband spectra obtained by the complex signal demodulation and the arctangent demodulation. As shown in both theory and experiment in [20], the DC offset does not affect complex signal demodulation when random body movement is absent. Therefore, the spectr um obtained by complex signal demodulation can 99

PAGE 100

be used as a reference to evaluate the reliabi lity of arctangent demodulation using the estimated DC offset information. The spectra of Fig. 6-12 (b) match well with each other, showing that the baseband DC offset estimation method is accura te enough for arctangent demodulation when no random body movement is present. Based on this DC offset estimation method, estimated DC offsets were subtracted from original detected data and used for random body movement cancellation. However, it should be noted that in the presence of the random body movement, the DC produced by the signals reflected from the bulk of the body always changes. Therefore, it is impossible to dynamically obtain the precise DC offset of the overall system: no matter whether the DC offset is calibrated out using the met hod proposed in [30] or estimated by the signal trajectory fitting method of this chapter, there will always be DC information error when the body position changes. It is of great interest to compare in real experiments that how robust the two demodulation techniques are in the pres ence of the inevitable DC offset error. 6.5.2 Random Body Movement Cancellation During the experiment, the subject under test was gently changing position in a chair, so that the noise of random body movement was emphasized. Fig. 6-13 shows the time domain signal detected from the front and the back of the human body when random body movement was present. Since the physiol ogical movement caused by respira tion and heartbeat has larger amplitude on the front chest wall than on the back the signal detected from the back is more severely affected by the random bod y drift. Note that the detected signal amplitude shown in Fig. 6-13 does not reflect the real physiological movement amplitude, since other factors such as distance and baseband amplifier gain also affect th e signal level. For example, in the experiment, the baseband amplifier gain of the radar detecting from the back is 3 dB higher than the other one detecting from the front. The two demodulation techniques were used to cancel out random body movement to recover the desired signal. 100

PAGE 101

A 0 2 4 6 8 10 12 -1 -0.5 0 0.5 1 Amplitude I Channel Q Channel B 0 2 4 6 8 10 12 -1 -0.5 0 0.5 Time (Second)Amplitude I Channel Q Channel Figure 6-13. Time domain signals. A) Detected from the front. B) Detected from the back. A 0 20 40 60 80 100 120 0 0.5 1 Spectrum Front Back B 0 20 40 60 80 100 120 0 0.5 1 Beats/MinCombined Spectrum Failed to recover heartbeat Figure 6-14. Random body movement cancellation using arctangent demodulation. A) Spectra measured from the front and the back of the human body. B) Spectrum from combining the two transceiver outputs, the heartbeat information cannot be recovered due to inaccurate DC offset information. The DC offset estimation based on signal trajec tory fitting was used here for arctangent demodulation. The baseband spectra detected from the front and the back of the human body are shown in Fig. 6-14 (a). The angular informati on from the two transceivers was combined as described in Section II-B, and the resulting baseba nd spectrum is shown in Fig. 6-14 (b). Due to 101

PAGE 102

the inaccuracy of DC offset estimation, the combined spectrum failed to recover the desired heartbeat signal component. On the other hand, the same signals have been processed by the complex signal demodulation. Fig. 6-15 (a) shows the baseband spectra of the complex signal detected from the front and the back of the human body. Since the p hysiological movement at the back is weaker than that at the front chest wall, the noise completely overwhelmed the physiological signals from the back and only overwhelmed the heartb eat signal from the front When the technique described in Section II-A was applied to combine the signals detected from the front and the back of the human body, the heartbeat signal wa s successfully recovered as shown in Fig. 6-15 (b). A 0 20 40 60 80 100 120 0 0.5 1 Spectrum Front Back B 0 20 40 60 80 100 120 0 0.5 1 Beats/MinCombined Spectrum Recovered heartbeat Figure 6-15. Random body moveme nt cancellation using comple x signal demodulation. A) Spectra measured from the front and the back of the human body. B) Output spectrum by the random body movement cancellation tech nique, the heartbeat information is recovered. The above comparative study verifies the simu lation in Section IV-C that the complex signal demodulation is more favorable in random body movement cancellation when the DC offset at baseband output cannot be accurately determined. 102

PAGE 103

6.6 Conclusion Simulations and experiments have been pe rformed to demonstrate the complex signal demodulation and the arctangent demodula tion for random body movement cancellation in Doppler radar vital sign detection. T modulation is easier to implement in that it n, or he complex signal de does not need an intermed iate signal processing stage to r ecover the angular informatio and it is robust when DC offset is present. Th e latter property also makes it more favorable f random body movement cancellation. On the other hand, the arctangent demodulation has the advantage of eliminating the harmonic and in termodulation interference at high frequencies using high gain antennas. The effects of constellation deformation and optimum/null detection ambiguity caused by the phase offset due to fini te antenna directivity are also discussed. 103

PAGE 104

CHAPTER 7 ADVANCED SPECTRAL ESTIMATION ALGOR ITHM FOR DOPPLER RADAR NONCONTACT VITAL SIGN DETECTION 7.1 Introduction When nonlinear Doppler phase modulation is employed to monitor vital signs without contact, one of the challenges that we encounter is the presence of undesired harmonic terms and intermodulations other than the sinusoids of interest. A spectral es timation algorithm is needed to accurately estimate the sinusoidal frequencies be fore identifying the heartbeat and respiration rates. The conventional periodogram can not reliably separa te the rich sinusoidal components since it suffers from smearing and leakage proble ms, especially for the case of limited data samples. A parametric and cyclic optimization a pproach, referred to as the RELAX algorithm, is suggested instead to mitigate these difficulties. In this chapter, both simulated and experimental results are provided to validate the superiority of using the RELAX algorithm for accurate noncontact vital sign detection. Let the (angular) frequencie s of the heartbeat and respiration be denoted as h and r, respectively. Then the received measurements consist of an infinite number of sinusoidal components at frequencies of p,q = p r + q h, where p and q are integers. Among them, 1 0 = r and 0 1 = h are of major interest; when p q 0, p,q's represent the frequencies of the intermodulation terms; while all the other components are the harm onics of either the heartbeat signal or the respiration signa l. The presence of the undesi red intermodulation and harmonic tones stems from the nonlinear na ture of the large-angle Doppler phase modulation. A detailed treatment of the effects of the nonlinear phase modulation from a sp ectral analysis perspective is presented in [13]. In addition, [14] provide s a guideline on selecting an optimum carrier frequency to enhance the desired frequency tones, and this optimization scheme greatly enhances the subsequent step of frequency estimation. 104

PAGE 105

The task of estimating si nusoidal frequencies (i.e., h and r, as well as the harmonic and intermodulation terms) from the received measurements can be accomplished by spectral estimation [35]. Once the frequencies are estimated, h and r can be identified out of them based on the prior knowledge about the power le vel and the linear relationship between the sinusoidal frequencies. Thus, a reliable spectral estimation algorithm is critical to the entire success of Doppler radar non-contact vital sign detection. Among the many spectral estimation algorithms, the Maximum Likelihood (ML) estima tor and the conventional periodogram are two important techniques. The existing frequency estimation approaches used for vital sign detection via Doppler radar are mainly periodogram based, which still can not provide sufficiently accurate frequency estimates needed for reliable vital sign detecti on. The nonparametric peri odogram can be readily derived from the Fourier transf orm, and the frequency estimate s correspond to th e locations of the dominant peaks of the peri odogram [35]. Periodogram, however, suffers from smearing and leakage problems. Smearing decreases the reso lution, making two closely spaced sinusoids irresolvable. In our experiment, 3 0 or 4 0 (i.e., the 3rd or 4th order harmonics of r) might be located near h, making it possible that due to smearing, their mainlobes merge into a single peak. Regarding leakage, the sidelobe of a str ong signal can bury the weak ones. Typically, the respiration signal dominates the spectra, and its harmonics may overwhelm the heartbeat signal, making the latter invisible in the periodogram. In either situation, the periodogram still cannot provide sufficiently accurate vital sign detection despite the a dvances in [13][14]. The parametric ML estimator can be much mo re accurate than the periodogram. However, when the number of the sinusoids is large (a situation encountered in the application considered 105

PAGE 106

herein), the ML estimator becomes computationally prohibitive, due to the need to search a large dimensional space. A parametric and cyclic optimization algorith m, referred to as the RELAX algorithm, has attracted our attention due to its simplicity. Proposed by Li and Stoica in 1996 [36], RELAX is an asymptotic ML approach [37] and outperforms the conventional periodogram significantly. RELAX iteratively estimates the parameters of each sinusoid in a super clean fashion. As long as the stronger sinusoids are accurately estimat ed, RELAX is able to reveal the weak ones by subtracting them out from the measurements. The rest of this chapter is organized as follo ws: Section 7.2 elaborates the theory and the nonlinear property of Doppler rada r vital sign detection. Section 7.3 formulates the problem of interest, and shows the drawback of the conve ntional periodogram. Section 7.4 details the RELAX algorithm, followed by the simulated and e xperimental results given in Section 7.5. Conclusions are provid ed in Section 7.6. 7.2 Spectral Estimation Using Periodogram In this section, the vital sign detection problem is formulated as a spectral estimation problem. The conventional periodogram as a spectral estimation algorithm will be elaborated, and its inherent drawback will be revealed by an example. 7.2.1 Problem Formulation As shown in Chapter 6, the baseband signal obtained by complex signal demodulation can be represented as: p,q,()()()() ee()jt j pq pqStItjQtvtCvt (7-1) where v( t ) represents the noise term. Since ej has a constantn unity modulus, the null detection point problem due to the effect of on the signal amplitude is eliminated [20]. 106

PAGE 107

A sampled version of the received complex signal can be written as: ,,() (tpqjn tp q pqyn evn )t (7-2) where n = 0, 1, N -1 with N being the sampled data length, and t is the sampling interval. The complex-valued amplitude p q is equal to Cp qej. By denoting the normalized frequency as p,q= tp,q, (7-2) can be simplified to: ,,pqjn np q pqye nv (7-3) As shown in the Introducti on, the data sequence in 1 0 N n ny consists of an infinite number of sinusoidal components, and it is impossible to estimate an infinite number of unknown parameters from a finite number of data sample s. Fortunately, the power of the harmonics decays very quickly, so we can safely neglect the weak sinusoidal compone nts and assume that 1 0 N n ny is dominated by K strongest ones. The value of K should be chosen so that the desired sinusoidal components (i.e., h and r) are not omitted. Mapping p q's and p,q's of the K strongest sinusoids to 1 K k k and 1 K k k, respectively, (7-3) can be rewritten as: 1kK jn nk kyenv (7-4) where is the sum of the noise term vn and the neglected weak sinusoidal components. nvHenceforth, vectors and matrices are denoted, respectively, by boldface lowercase and uppercase letters. ()T means the transpose, ()H means the conjugate tr anspose, and || || represents the Euclidean norm. The ith component of a vector b is written as bi. Denote the received symbol vector as y = [ y0, y1, yN-1]T, the steering vector a( k) = [1 k j e2 kje (1) kjNe]T, and the noise vector v = [011 Nvvv ]T. Then (7-4) can be expressed in a compact form as: 107

PAGE 108

1()K kk k y av (7-5) The problem of interest then becomes estimating 1,K kk k from the measurement vector y. Identifying the final targets (i.e., h and r) out of 1K k k is accomplished by taking advantage of the prior knowledge of the power le vels and the linear relationship of the sinusoidal frequencies. 7.2.2 Periodogram The periodogram is defined as: 2 2 1 0() 1 ()H N jn n nPye NN a y (7-6) The Fast Fourier Transform (FFT) speeds up the computation of the periodogram, and zero-padding y before conducting FFT is essential to achieve high frequency estimation accuracy. Ideally, with an infinite sample length (i.e., N ), the kth sinusoidal component contributes one vertical line in the spectrum at frequency k. Practically, however, the finite data length N causes smearing and leakage in the periodogram. An illustrating example is given in Fig. 7-1. The data comprises four complex-valued sinusoids without noise. The data length is N = 40 and the sampling frequency is 1 Hz. Stems represent the true line spectra, and the curve is for the periodogram of Due to the small number of sample s, smearing occurs and merges two closely spaced sinusoids into one peak. More speci fically, the sinusoids at -0.115 and -0.1 Hz are located within the re solution limit of the periodogram (i.e., 1/ N Hz), and as a result, their mainlobes superpose to form a single peak. 39 0 n ny 108

PAGE 109

-0.5 0 0.5 0 0.5 1 1.5 2 2.5 Frequency (Hz)Power (dB) Periodogram True Line Spectra Figure 7-1. Four complex-valued sinusoids with frequencies of -0.115, -0.1, 0.35 and 0.412 Hz. They are presented with data length N = 40 and sampling frequency of 1 Hz. The true line spectra (red stems) is plotted with the periodogram, which is obtained by performing FFT on the received sequence pa dded with 2008 zeros. Due to smearing, 0.115 and -0.1 Hz sinusoids form a single p eak in the periodogram; and due to power leakage, the weak 0.412 Hz signal is burie d in the sidelobe of the strong 0.35 Hz sinusoid. Leakage spreads the otherwise concentrated po wer to the entire frequency band, generating sidelobes. In this example, the sidelobe of th e strong signal (at 0.35 Hz) buries the weak one at 0.412 Hz, making it almost invisible in the periodogram. If K = 4 is known as the a priori information, from the periodogram plotted in Fig. 7-1, it is tempting to claim the frequency estimates as 0.35, -0.107, 0.314 and 0.386 Hz (the last two correspond to the sidelobe peaks of the 0.35 Hz signal). Obviously, these estimates are unacceptable. In the experiment of vital sign detec tion, the measured signals are noisy; N is in hundreds; the harmonics of 3 0 or 4 0 have a good chance of being close to h; in addition, the respiration signal is often much stronger th an the other sinusoidal components. Considering the problems of the periodogram, a more reliable spectr al estimation algorithm is desirable. 109

PAGE 110

7.3 Spectral Estimation Using RELAX The RELAX algorithm is detailed in this sect ion for spectral estimation. In the presence of the zero-mean white Gaussian noise, the ML es timator requires solving the following nonlinear least-squares fitting problem: 12 1 1 ,argmink K kk kK K jn kk nk k kye (7-7) The optimization problem in (7-7) can be tackled in a conceptually simple and computationally efficient manner by employing the RELAX algorithm. Its pseudo-code is outlined in Table I. The main idea of RELAX is that it postpones estimating a new sinusoidal component until the already determined ones are good enough. In this manner, the RELAX algorithm make an effort to provide excellent initial conditions for the next new sinusoidal parameter estimation [38]. It can be easily noticed from Table I that th e building block of the RELAX algorithm is to estimate one sinusoidal com ponent with the parameter ,ii from measurement vector i y by assuming that i y consists of this sinusoid and the noise term only. The ,ii can be estimated as: 2 ,argmin()iiii iii ya (7-8) Minimizing the right side of (7-8) with respect to i yields: 2 () ()H ii i i ay a (7-9) Replacing the i in (7-8) with its estimate in (7-9) yields: 110

PAGE 111

2 () 2 2 argmin () argmax ()i i iii H ii i aPy a y a (7-10) where () 2()() ()iH i ii aaa PI a denotes the orthogonal proj ection onto the null space of ()H i a Recalling 2()i a= N and (7-6), the last expression in (7 -10) is exactly the periodogram of i y Thus i can be readily obtained as the location of the dominant peak of the periodogram. Table 7-1. The RELAX algorithm. Initialize 0,1,2,,kkK for i = 1, 2, K 1, () yya K ik kki k 2 2 () argmax () a y a H i i 2 () () a y a H ii i i repeat for j = 1, 2, i 1, () yya i jm mmj m 2 2 () argmax () a y a H j j 2 () () a y a H jj j j end for until (convergence) end for How RELAX algorithm works will be illustrated in the first example in the next section. The convergence is practically determined by ch ecking the difference of the cost function in 111

PAGE 112

(7-7) between two successive iter ations. The RELAX algorithm ends if this difference is less than a predefined threshold t. 7.4 Numerical and Experimental Results Numerical and lab experiments have been performed to demonstrat e the advantages of using RELAX for the non-contact vital sign detection. For lab experiments, the 4-7 GHz broadband portable radar discussed in Chapter 4 was used. Fig. 7-2 shows the radar photo and the experimental set up. The amplified baseband output can be socket-connected to a data acquisition mo dule (DAQ) for A/D conversion before fed into the laptop for signal processing. To reduce the co sts of hardware and the requirement on signal processing speed, the amplified baseband signals were sampled by a 12-bit multifunction data acquisition module (NI USB-6008) w ith a low sampling rate of 20 Hz. The sampled data were fed into a laptop for real time signal processing by LabVIEW. The sampling rate and resolution make it possible to integrate the baseband signal processing in to low cost DSP microchips. Figure 7-2. Experiment setup w ith a 4-7 GHz radar, a data ac quisition module, and a laptop for signal processing. 112

PAGE 113

7.4.1 Example I Both of the periodogram and RELAX were a pplied to the measured data. The data sequence contains 257 samples, with a sampling fr equency of 20 Hz. The measured time domain baseband signals are shown in Fig. 7-3. 0 2 4 6 8 10 12 -0.4 -0.2 0 0.2 0.4 Time [ Sec ] Amplitude I Channel Q Channel Figure 7-3. A set of 12.85 s econds' baseband signal measured by the 4-7 GHz radar when detecting from the back of the human body. The RELAX algorithm was applied to the base band signal, and a few intermediate stages are shown in Fig. 7-4. Because y is complex-valued, both the positive and negative frequency axis should be taken into consideration. K was set to be 14 in this example. Figure 7-4(a) shows that the RELAX algorithm first estimated the strongest sinusoid with the {complex amplitude, frequency [beats/ min]} of {-0.0924+0.0540i, -19.7726} from the periodogram. By subtracting it out from the data sequence, the second strongest sinusoid was found to be {-0.0464+0.0768i, 20.0046} in Fig. 7-4( b). Then RELAX updated these two already estimated sinusoids. Update started by subtracti ng out the second strongest sinusoid from the original data sequence, the si nusoid estimated in (a) was then updated to be {-0.0901+0.0546i, 19.7915}, see Fig. 7-4(c). Then in Fig. 7-4(d) this newly updated sinusoid was removed from the original data sequence, and the other sinusoid estimated in (b) was updated to be {0.0464+0.0770i, 20.0066}. 113

PAGE 114

A -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB)B -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB)C -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB) D -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB) E -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB)F -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB) 114 G -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB)H -200 -150 -100 -50 0 50 100 150 200 0 0.02 0.04 0.06 0.08 0.1 Frequency (Beat/Min)Power (dB) Figure 7-4. Some intermediate stages ar e shown when applying RELAX for analyzing the baseband signal sp ectra. RELAX postpones estimating a new sinusoidal component until the already estimated signals ar e good enough, thus offering good initial conditions for the estimation of the new signa l. The estimates of the determined signa ls are updated by iter ating, as C) and D) show. When it is the term to estimate a new signal, it will appear in spectra because RELAX removes all the estimated stronger sinusoidal component from the spectra, as B), E)-H) show.

PAGE 115

This cyclic update is repeated until the co st function in (7-7) between two successive iterations was less than t=10-3. Now by subtracting out both of them from the original data sequence, the initial estimate for the third st rongest sinusoid was found to be {-0.0342+0.0542i, 65.8566} in Fig. 7-4(e). It then began a new round of iterative updates of the three strongest sinusoids. This time, one sinusoid was updated by s ubtracting out the other two from the original data sequence. After convergence, by subtracting out these three si nusoids from the original data sequence, the fourth sinu soid is initially estimated to be {-0.0427 + 0.0440i, -65.7289}, as shown in Fig. 7-4(f). Likewise, anot her round of iterative updates be gan and each sinusoid was updated by subtracting out the other three from the original data sequence. Figs. 7-4(g) and (h) show the initial estimates for the fifth and sixth sinusoids to be {-0. 0027 0.0291i, -24.3334} and {0.0233 0.0136i, 40.8858}, respectively. This procedure is repeated until the parameters of all K = 14 sinusoids were estimated and updated. 0 20 40 60 80 100 12 0 0 0.2 0.4 0.6 0.8 1 Frequency (Beat/Min)Normalized Spectrum Periodogram Estimates from RELAX Figure 7-5. Final result of RELAX estimation to measured baseband signal when detecting from the back of the human body. Data length N = 257, convergence threshold t = 1-3. K is set to be 14. Respiration and heartbeat rate s can by no means lay on the negative part of the frequency axis, so we narrowed our consideration to the range of 0~120 beats/min. Fig. 7-5 plots the 115

PAGE 116

estimates from RELAX in this range, together wi th the periodogram of the baseband signal. 5 out of the 14 sinusoids are in cluded in this range. The resp iration signal component at 20 beats/min has been identified. The RELAX show s that the respiration consists two closely located sinusoids at 19.1 beats/min and 23.5 beats/min, owning to the variation of respiration rate during the measurement period. Th e heartbeat signal component was identified as 65.9 beats/min. And the frequency components ne ar 40 beats/min were caused by the harmonics of respiration signal. 7.4.2 Example II To further demonstrate the advantage of the RELAX algorithm, a computer-simulated baseband signal reflecting a situation where the signal 3 0 is located very close to h, was processed. A ray-tracing model was used to si mulate the Doppler radar vital sign detection. Using this method, the antenna radiation patte rn, the physiological movement, and the phase offset among different signal tran smission paths can be properly m odeled as shown in Fig. 7-6. The baseband signal is the sum of all the signals reflected from different parts of the human body. A B C D Figure 7-6. Ray-tracing model. A) The phase offset on the surface of human body. B) A 7 by 7 elements antenna array's radiation intens ity on the human body. C) Approximation of the normalized amplitude of body move ment caused by respiration. D) Approximation of the normalized amplitude of body movement caused by heartbeat. 116

PAGE 117

The carrier frequency was set to be 20 GHz, which was adopted for higher sensitivity to small movement. To challenge the spectral esti mation algorithms, simulations were performed for the detection from the front of the subject, in which case the respiratio n power is much larger than the power of the heartbeat signal. For ba seband signal processing, the sampling period was 0.027 sec, and the data length N = 445, corresponding to a to tal sampling time of 445.027 = 12.015 sec. Therefore the peri odogram resolution limit was 1/12.015 = 0.0832 Hz, which was converted to the beats/min unit of 4.992. The respiration rate was 21.5 beats/min, while the heart rate was set as 67.5 beats/mi n. The frequency difference between h and 3 0 was 3 beats/min, which was less than the resolution limit of the periodogram. As a result, they formed a single peak and the periodogram cannot reso lve them. The RELAX algorithm, however, successfully located the two frequency components as indicated in Fig. 7-7. 0 20 40 60 80 100 120 0 0.2 0.4 0.6 0.8 1 Frequency (Beat/Min)Normalized Spectrum Periodogram Estimates from RELAX Figure 7-7. RELAX estimation to computer-generated baseband si gnal when detecting from the front of the human body. Data length N = 445, threshold t = 10-3. K is set to be 14. The RELAX estimates for the heartbeat a nd respiration rates were 21.5357 and 67.918 beats/min, respectively. And 3,0 near the heartbeat was found to be 64.6511, whose true value i 64.5 beats/min. It is shown that although there exist two closely located sinusoidal components around the heartbeat frequency, RELAX still successfully resolved them. s 117

PAGE 118

7.5 Conclusion The RELAX algorithm has been employed to process baseband signals for Doppler radar non-contact vital sign detection. It has been shown both theoretically and experiment ally that the RELAX algorithm in general succeeds in mitigat ing the effects of the smearing and leakage problems of the periodogram caused by limited data length. Therefore th e non-contact vital sign detector enjoys accurately estimated heartbeat and respiration frequency tones from the RELAX algorithm despite the possibility that the harmonics of the resp iration signal might be closely located to the heartbeat signal, causin g the traditional periodogram to fail. 118

PAGE 119

CHAPTER 8 APPLICATION CASE STUDY INFANT VITAL SIGN MONITOR 8.1 Introduction Infant monitoring is a prime application fo r non-contact vital sign detection technology. The fear of Sudden Infant Death Syndrome (SID S) and the prevalence of infant breathing problems have many parents searching for a way to monitor the health and well-being of their child, especially while sleeping unattended. Therefor e, it is expected to ha ve a prime market of first time parents with the income and the inc lination to buy baby monitors that provide more than just sound or video. A noncontact heartbeat/respiration m onitor would remotely watch over the health of the child and provide parents with peace of mind. The remainder of this chapter outlines a prototype monitoring system designed for the purpose of baby monitoring [39]. Fig. 8-1 illustrates the receiver uni t and the monitor unit in use. The monitor unit is designed to hang on the side of the infants crib and detect the infants breathing and heartbeat. It then communicates wirelessly with the receiver unit allowing the parent to monitor the child remotely. Alarms sound and red lights flash on both units if the childs respiration and heartbeat are too weak. A B Figure 8-1. Infant monitoring sy stem. A) The receiver unit mounted on crib. B) The monitor unit in operation. 119

PAGE 120

8.2 System Architecture The monitoring system consists of two main de vices: a monitor unit which is placed on the side of the infants crib and a receiver unit which is carried by th e parent. The monitor is divided into several sub-sections: RF (Radio Frequenc y) circuitry to produce and receive the radio signals for vital sign detecti on, a microcontroller for signal processing, an XBee wireless communication chip for communica tion with the receiver, a power management circuit, and a simple user interface consisting of bu ttons, switches, LEDs, and a speaker. Figure 8-2. Block diagram of the implemen ted infant monitoring system hardware. Figure 8-2 shows a block diagram of the mon itoring systems hardware. In the monitor unit, a 5.8 GHz single tone unmodulated carrier signal is generated by a voltage controlled oscillator (VCO). It is amplified and transmitted via microstrip antenna towards the infant. The reflected RF signal, which has been phase modulat ed by the movement of the infant, is captured 120

PAGE 121

by a receiver antenna and amplified by a low noi se amplifier (LNA) and a two-stage variable gain pre-amplifier. The local os cillator signal and th e received signal are then mixed together, amplified by a two-stage baseband amplifier, and sent through a low power microcontrollers A/D port for analysis. Voltage re gulators and battery management micro-chips are integrated into the system to support wall -plug power supply, battery powe r supply, and battery recharge functions. Since the monitor unit merely makes a determin ation whether heartbeat or respiration is present or not, signal processing is kept to a minimum and performed in the time domain for simplicity and speed. Since spectrum estimation methods such as fast Fourier transform (FFT) are not required, time domain pr ocessing also reduces cost and pow er by eliminating the need for a sophisticated digital signal processing (DSP) mi croprocessor. The DC of fset produced not only by electronic circuits, but also by the down-conversion of signals reflected from stationary objects in the surrounding environm ent, is no longer negligible as in the case of frequency domain signal processing. To address this issue, the time-dom ain signal processing supports automatic calibration to compensate for the DC o ffset of the system in different application environments. Because the radar circuitry is only implemen ted on the monitor side, the receiver carried by the parents has a much lower power level for its power management circuit. Wireless communication with the monitor unit via XBee wireless transcei vers keeps the receiver unit updated with the current alarm stat us. A vibrating motor was also a dded to the recei ver to ensure all possible methods of alerting the parent (light, sound, and movement) are utilized. 8.3 Features The main features of the baby monitor protot ype are design simplicity, range, reliability, and automatic calibration. The 5.8 GHz carrier frequency was chosen for two main reasons. 121

PAGE 122

First, this frequency has been successfully demo nstrated for non-contact vital sign detection in the previous chapters. Second, this frequency is in the unlicensed industrial, scientific, and medical (ISM) bands, where low-cost commercial components are available. The power level at the transmitter output can be adjusted from 10 dBm to 8 dBm. The prototype system, using a low cost, low power microcontroller, was able to accu rately detect respirati on with a range of at least 1.15 meters. This distance reaches to the furt hest corner of a standard crib and allows for consistent monitoring regardless of the infants position. Automatic calibration of the DC offset, performed on start up, allows the system main tain accuracy in a variety of environments. B A C Figure 8-3. The prototype system designed and fabricated. A) Exte rior of the prototype system; B) Interior of the monitor unit; C) Interi or of the receiver uni t. (Not to scale) To provide flexibility and ease of use, both the receiver and monitor units can be powered by batteries or plugged directly into wall power. The receiver un it can travel up to 50 meters away from the monitor while still being able to alert the parent. For operating range larger than 122

PAGE 123

50 meters, the receiver needs to be connected to a laptop which fo rwards the alarm to parents cell phone or PDA. The fabricated prototype is shown in Fig. 8-3. The cost for mass production of the monitoring system is targeted to be under $80 for the monitor and receiver pair; multiple receivers could be added to work with a singl e monitor. Limited by the package facilities available, the prototype has a relatively large size (see Fig. 8-3 for dimensions). With more refined packaging, the receiver could be reduced to the size of an ordinary cell phone. The monitor could also be significantly reduced, with the limiting factor being the two 6 cm by 6 cm patch antennas. 8.4 Conclusion An application-oriented design of infant vital sign monitor has been presented in this chapter. Using the radar presented in Chapte r 4 as the key monitoring device, the system provides a low-cost solution to monitoring the Sudden Infant Death Syndrome (SIDS) in real time. Three possible methods of alert (light, s ound, and movement) are remotely provided to parents through wireless link, providing parents with peace of mind. In the near future, there is expected to be a prime market of first time pa rents with the income a nd the inclination to buy this kind of baby monitor to watch over their ba by. Similar products for applications such as intruder detection radar [40], sleep apnea monitor, telemedicine, see-through-wall radar [41], and fire/earthquake rescue are also expect ed to emerge soon in our daily life. 123

PAGE 124

CHAPTER 9 SUMMARY The theory and implementation of wireless non-contact vital sign radar sensor is presented in this dissertation. The implemented system can detect and display the detected heartbeat and respiration in real time. With the Ka-band bench top system opera ting at around 27GHz, the short wavelength increases the sensitivity of de tection. The double-sideband transmission and detection method solves the nullpoint detection problem and simp lifies the system architecture. Long-term measurement results demonstrated the robustness of th is system. Nonlinear Doppler phase modulation was discovered and a model was de veloped to explain the experiment result. A method of accurately measuring periodic movement amplitude is described. A 4-7 GHz radar system fully integrated on printed circuit board and radar front-end chips integrated in 0.18 m and 0.13 m CMOS technologies have been demonstrated. The theory of random body movement cancellation has been el aborated, simulated with a raytracing model, and verified by lab experiments. On the signal processing si de, the RELAX algorithm has been employed to mitigate the effects of smearing and leakage problems of periodogram caused by limited data length. In summary, the radar sy stem and non-contact detection t echnology can be used in many applications in science, engineering, and medicine. 124

PAGE 125

LIST OF REFERENCES [1] J. C. Lin, Noninvasive microwave measuremen t of respiration, Proc. IEEE, vol. 63, no. 10, p. 1530, Oct. 1975. [2] K. M. Chen, D. Misra, H. Wang, H. R. Chuang, and E. Postow, An X-band microwave life-detection system, IEEE Trans. Biomed. Eng., vol. 33, pp. 697-702, July 1986. [3] H. R. Chuang, Y. F. Chen, and K. M. Chen Automatic clutter-c anceller for microwave life-detection system, IEEE Trans. Instru m. Meas., vol. 40, no. 4, pp. 747-750, Aug. 1991. [4] J. C. Lin, Microwave sensing of phys iological movement and volume change: A review, Bioelectromagnetics, vol. 13, pp. 557-565, 1992. [5] K. M. Chen, Y. Huang, J. Zhang, and A. Norman, Microwave life-detection systems for searching human subjects under earthquake r ubble and behind barrier, IEEE Trans. Biomed. Eng., vol. 47, pp. 106-114, Jan. 2000. [6] A. D. Droitcour, V. M. Lubecke, J. Lin, O. Boric-Lubecke, A microwave radio for Doppler radar sensing of vital signs, I EEE MTT-S Int. Microwave Symp. Dig., May 2001, pp. 176-178. [7] A. D. Droitcour, O. Boric-Lubecke, V. M. Lubecke, J. Lin, .25m CMOS and BiCMOS single chip direct conve rsion Doppler radars for remo te sensing of vital signs, IEEE Int. Solid State Circuits Conf., Dig., Feb. 2002, pp. 348-349. [8] A. D. Droitcour, O. Boric-Lubecke, V. M. Lub ecke, J. Lin, and G. T. A. Kovac, Range correlation and I/Q performance benefits in single-chip silicon Doppler radars for noncontact cardiopulmonary monitoring, IEEE Trans. Microw. Theory Tech., vol. 52, pp. 838-848, Mar. 2004. [9] Y. Xiao, J. Lin, O. Boric-Lubecke and V. M. Lubecke, A Ka-band low power Doppler radar system for remote detection of cardi opulmonary motion, Proc. 27th IEEE Ann. Int. Conf. Engineering in Me dicine and Biology Soc ., Sept. 2005, pp. 7151-7154. [10] Y. Xiao, J. Lin, O. Boric-Lubecke, and V. M. Lubecke, Frequency tuning technique for remote detection of heartbeat and re spiration using lowpower double-sideband transmission in Ka-band, IEEE Trans. Microw. Theory Tech., vol. 54, pp. 2023-2032, May, 2006. [11] Y. Xiao, C. Li, and J. Lin, Accuracy of a low-power Ka-band non-contact heartbeat detector measured from four sides of a human body, IEEE MTT-S Int. Microwave Symp. Dig., June 2006, pp. 1576-1579. [12] C. Li, J. Lin, Y. Xiao, Robust overnight monitoring of human vital signs by a noncontact respiration and heartb eat detector, Proc. 28th IEEE Ann. Int. Conf. Engineering in Medicine and Biology Soc., Sept. 2006, pp. 2236-2238. 125

PAGE 126

[13] C. Li, Y. Xiao, and J. Lin, Experiment a nd spectral analysis of a low-power Ka-band heartbeat detector measuri ng from four sides of a hu man body, IEEE Trans. Microw. Theory Tech., vol. 54, no. 12, pp. 4465-4471, Dec. 2006. [14] C. Li, and J. Lin, Optimal carrier freque ncy of non-contact vital sign detectors, Proc. IEEE Radio and Wireless Symp., pp. 281-284, Long Beach, Jan. 9-11, 2007. [15] C. Li, J. Lin, Non-contact measurement of periodic movements by a 22-40GHz radar sensor using nonlinear phase modulation, IEEE MTT-S Int. Micr owave Symp. Dig., June 2007, pp. 579-582. [16] C. Li, Y. Xiao, and J. Lin, Design guidelin es for radio frequency non-contact vital sign detection, Proc. 29th IEEE Ann. Int. Conf. Engineering in Medici ne and Biology Soc., Lyon, France, Aug. 23-26, 2007, pp. 1651-1654. [17] C. Li, J. Lin et al. Devel opment of non-contact physiol ogical motion sensor on CMOS chip and its potential applications, The 7th Int. Conf. Application-Specific Integrated Circuits (ASICON), vol. 2, pp. 1022-1027, Guiling, China, Oct. 26-29, 2007. [18] C. Li, Y. Xiao, J. Lin, A 5 GHz double-sideband radar se nsor chip in 0.18 m CMOS for non-contact vital sign de tection, IEEE Microw. Wirele ss Components Lett., vol. 18, issue 7, pp. 495-496, Jul. 2008. [19] C. Li, X. Yu, D. Li, L. Ran, and J. Lin, Software configurable 5.8 GHz radar sensor receiver chip in 0.13 m CMOS for non-cont act vital sign detection, IEEE RFIC Symp., June 2009, pp. 97-100. [20] C. Li, and J. Lin, Complex signal demodulation and random body movement cancellation techniques for non-contact vi tal sign detection, IEEE MTT-S Int. Microwave Symp. Dig., Jun. 2008, pp. 567-570. [21] B. Lohman, O. Boric-Lubecke, V. M. Lubecke, P. W. Ong, and M. M. Sondhi, A digital signal processor for Doppler radar sensing of vital signs, Proc. 23rd IEEE Annu. Engineering in Medicine and Biolog y Society Conf., Sept. 2001, pp. 3359-3362. [22] B. H. Yang and S Rhee, Development of th e ring sensor for healthcare automation, Robot. Autonom. Syst., vol. 30, pp. 273-281, 2000. [23] D. C. Champeney, Fourier transforms and th eir physical applications, Academic Press, 1973. [24] M. Singh and G. Ramachandran, Reconstruction of sequential cardiac in-plane displacement patterns on the chest wall by laser speckle in terferometry, IEEE Trans. Biomed. Eng., vol. 38, pp. 483-489, May 1991. [25] M. C. Budge, Jr. and M. P. Burt, Range correlation effects on phase and amplitude noise, Proc. IEEE Southeastcon, Charlotte, NC, 1993, p. 5. 126

PAGE 127

[26] A. Stezer, C. G. Diskus, K. Lubke, and H. W. Thim, Microwave position sensor with sub millimeter accuracy, IEEE Trans. Microw. Theory Tech., vol. 47, pp. 2621-2624, Dec. 1999. [27] E. N. Ivanov, M. E. Tobar, and R. A. Woode Microwave interferom etry: application to precision measurements and noise reducti on techniques, IEEE Trans. Ultrason., Ferroelect., Freq. Contr., vol 45, pp. 1526-1536, Nov. 1998. [28] S. Kim, and Cam Nguyen, On the development of a multifunc tion millimeter-wave sensor for displacement sensing and low-ve locity measurement, IEEE Trans. Microw. Theory Tech., vol. 52, pp. 2503-2512, Nov. 2004. [29] H. Ma, S. J. Fang, F. Lin, and H. Nakamura Novel active differential phase splitters in RFIC for wireless applications, IEEE Tran s. Microw. Theory Tech., vol. 46, no. 12, pp. 2597-2603, Dec. 1998. [30] B. Park, O. Boric-Lubecke, and V. M. L ubecke, Arctangent demodulation with DC offset compensation in quadrature Doppler radar receiver systems IEEE Trans. Microw. Theory Tech., vol. 55, pp. 1073-1079, May 2007. [31] B. Park, V. M. Lubecke, O. Boric-Lubeck e, and A. Host-Madsen, Center tracking quadrature demodulation for a Doppler ra dar motion detector, IEEE MTT-S Int. Microwave Symp. Dig., pp. 1323-1326, June 2007. [32] J. M. Taylor, and A. J. Terzuoli, On the c oncept of near field radar cross section, IEEE Ant. Prop. Soc. Int. Symp. Dig., vol. 2, Jul. 1997, pp. 1172-1175. [33] M. I. Skolnik, Radar Handbook, New York, McGraw-Hill, 1970. [34] C. Li, J. Lin, Random body movement cancellation in Doppler radar vital sign detection, IEEE Trans. Micr ow. Theory Tech., vol. 56, issue 12, pp. 3143-3152, Dec. 2008. [35] P. Stoica, and R. Moses, Spectral analys is of signals, Prentice-Hall, Upper Saddle River, NJ, 2005. [36] J. Li, and P. Stoica, Efficient mixed-spectru m estimation with applications to target feature extraction, IEEE Tran s. Signal Processing, vol. 44, no. 2, pp. 281-295, Feb. 1996. [37] P. Stoica, A. Jakobsson, and J. Li, Cisoid pa rameter estimation in the colored noise case: asymptotic Cramer-Rao bound, maximum likel ihood, and nonlinear le ast-squares, IEEE Trans. Signal Processing, vol. 45, no. 8, pp. 2048-2059, Aug. 1997. [38] J. Ling, P. Stoica, J. Li, and Yu. I. Abramovich, On using cyclic algorithms for sinusoidal parameter estimation, IET Elect ron. Lett., vol. 44, issue 19, Sep. 2008. 127

PAGE 128

[39] C. Li, J. Cummings, J. Lam, E. Graves, W. Wu, Radar remote monitoring of vital signs from science fiction to r eality, IEEE Microw. Magazine, vol. 10, issue 1, pp. 47-56, Feb. 2009. [40] M. Morinaga, T. Nagasaku, H. Shinoda, and H. Kondoh, GHz intruder detection radar with beam-switched area coverage, I EEE MTT-S Int. Microwave Symp. Dig., Jun. 2007, pp. 389-392. [41] O. Boric-Lubecke, J. Lin, B.-K. Park, C. Li, W. Massagram, V. M. Lubecke, A. HostMadsen, Battlefield triage life signs detec tion techniques, Proc. SPIE Defense Security Symp., vol. 6947 XII, no. 69470J, 10 pages, Apr. 2008. 128

PAGE 129

BIOGRAPHICAL SKETCH Mr. Changzhi Li received the B.S. degree in electrical engineering from Zhejiang University, Hangzhou, China, in 2004, the M.S. a nd Ph.D. degrees in electr ical engineering from the University of Florida, Gainesvi lle, FL, in 2007 and 2009, respectively. In the summers of 2007 and 2008, he worked at Alereon Inc., Aus tin, TX, first on the characterization of the worlds first 3.1-10.6 GHz worldwide ultrawideb and transceiver chip, then on the design of phase-locked loop in CMOS 65 nm process. In the summer of 2009, he worked at Coherent Logix Inc., Austin, TX, on software-defined radio. His research interests include biomedical applications of Microwave/RF, wireless sens or, frequency synthesizers, and Microwave/Millimeter-Wave Circuits. Mr. Li is a student member of the IEEE Microwave Theory and Techniques Society (IEEE MTT-S), and the IEEE Engineering in Medicine and Biology Society. He is the second place recipient of the Best Studen t Paper award in the 2007 IEEE Radio and Wireless Symposium (RWS), and is the finalist in the 2008 IEEE MTT-S Internatio nal Microwave Symposium (IMS) Student Paper Competition. He received IEEE MTT-S Graduate Fellowship Award in 2008. 129