• TABLE OF CONTENTS
HIDE
 Title Page
 Dedication
 Acknowledgement
 Table of Contents
 Abstract
 Introduction
 The sleep analyzing hybrid...
 Software for processing of periodicity...
 Normal adult patterns
 Ontogenetic trends
 Appendix 1: Program listings
 Appendix 2: The use of spectral...
 Appendix 3: Data from all...
 Bibliography
 Biographical sketch














Title: Automated analysis of biological rhythms in the human electroencephalogram /
CITATION PDF VIEWER THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00098145/00001
 Material Information
Title: Automated analysis of biological rhythms in the human electroencephalogram /
Physical Description: vii, 247 leaves : ill. ; 28cm.
Language: English
Creator: Keane, Barry Patrick, 1949-
Publication Date: 1975
Copyright Date: 1975
 Subjects
Subject: Biological rhythms   ( lcsh )
Electroencephalography   ( lcsh )
Electrical Engineering thesis Ph. D
Dissertations, Academic -- Electrical Engineering -- UF
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
 Notes
Thesis: Thesis--University of Florida.
Bibliography: Bibliography: leaves 243-246.
General Note: Typescript.
General Note: Vita.
Statement of Responsibility: by Barry Patrick Keane.
 Record Information
Bibliographic ID: UF00098145
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: alephbibnum - 000161995
oclc - 02692739
notis - AAS8340

Downloads

This item has the following downloads:

PDF ( 5 MBs ) ( PDF )


Table of Contents
    Title Page
        Page i
        Page i-a
    Dedication
        Page ii
    Acknowledgement
        Page iii
    Table of Contents
        Page iv
        Page v
    Abstract
        Page vi
        Page vii
    Introduction
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
    The sleep analyzing hybrid computer
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
    Software for processing of periodicity data
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
        Page 24
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
    Normal adult patterns
        Page 41
        Page 42
        Page 43
        Page 44
        Page 45
        Page 46
        Page 47
        Page 48
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
        Page 68
        Page 69
        Page 70
        Page 71
        Page 72
        Page 73
        Page 74
        Page 75
        Page 76
        Page 77
        Page 78
        Page 79
        Page 80
        Page 81
        Page 82
        Page 83
        Page 84
    Ontogenetic trends
        Page 85
        Page 86
        Page 87
        Page 88
        Page 89
        Page 90
        Page 91
        Page 92
        Page 93
        Page 94
        Page 95
        Page 96
        Page 97
        Page 98
        Page 99
        Page 100
        Page 101
        Page 102
        Page 103
        Page 104
        Page 105
        Page 106
        Page 107
        Page 108
        Page 109
        Page 110
        Page 111
        Page 112
        Page 113
        Page 114
        Page 115
        Page 116
        Page 117
        Page 118
        Page 119
        Page 120
        Page 121
        Page 122
        Page 123
        Page 124
        Page 125
        Page 126
    Appendix 1: Program listings
        Page 127
        Page 128
        Page 129
        Page 130
        Page 131
        Page 132
        Page 133
        Page 134
        Page 135
        Page 136
        Page 137
        Page 138
        Page 139
        Page 140
        Page 141
        Page 142
        Page 143
        Page 144
        Page 145
        Page 146
        Page 147
        Page 148
        Page 149
        Page 150
        Page 151
        Page 152
        Page 153
        Page 154
        Page 155
        Page 156
        Page 157
        Page 158
        Page 159
        Page 160
        Page 161
        Page 162
        Page 163
        Page 164
        Page 165
        Page 166
        Page 167
        Page 168
        Page 169
        Page 170
        Page 171
        Page 172
        Page 173
        Page 174
        Page 175
        Page 176
    Appendix 2: The use of spectral techniques in the analysis of ultradian biological rhythms
        Page 177
        Page 178
        Page 179
        Page 180
        Page 181
        Page 182
        Page 183
        Page 184
        Page 185
        Page 186
        Page 187
        Page 188
        Page 189
        Page 190
        Page 191
        Page 192
        Page 193
        Page 194
        Page 195
        Page 196
    Appendix 3: Data from all age groups
        Page 197
        Page 198
        Page 199
        Page 200
        Page 201
        Page 202
        Page 203
        Page 204
        Page 205
        Page 206
        Page 207
        Page 208
        Page 209
        Page 210
        Page 211
        Page 212
        Page 213
        Page 214
        Page 215
        Page 216
        Page 217
        Page 218
        Page 219
        Page 220
        Page 221
        Page 222
        Page 223
        Page 224
        Page 225
        Page 226
        Page 227
        Page 228
        Page 229
        Page 230
        Page 231
        Page 232
        Page 233
        Page 234
        Page 235
        Page 236
        Page 237
        Page 238
        Page 239
        Page 240
        Page 241
        Page 242
    Bibliography
        Page 243
        Page 244
        Page 245
        Page 246
    Biographical sketch
        Page 247
        Page 248
        Page 249
Full Text











AUTOMATED ANALYSIS OF BIOLOGICAL
RHYTHMS IN THE HUMAN ELECTROENCEPHALOGRAM














BY

BARRY PATRICK KEANE


















A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF
THE UNIVERSITY OF FLORIDA IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY


UNIVERSITY OF FLORIDA
19757























































UNIVERSITY OF FLORIDA


3 1262 08666 420 7






























TO

MY WIFE

DEBBIE
















ACKNOWLEDGMENTS


The author would like to thank his supervisory

committee for their guidance and assistance in doing

this work. In particular, the author wishes to thank

Dr. Jack Smith for his close supervision, support, and

continued assistance while working in his laboratory.

Dr. Ismet Karacan, Baylor College of Medicine, provided

the data analyzed in this dissertation. The author would

like to thank Dr. Karacan for providing the data and for

his encouragement and suggestions concerning this research.

In addition, at various times, each person in the

laboratory made contributions or gave advice for which

the author is very greatful--these persons all created

a congenial atmosphere which made it a pleasure to work in

the laboratory and most certainly are in this way partially

responsible for the productivity which this work represents.

This work was supported by Mental Health Grant

MH 16960.
















TABLE OF CONTENTS


ACKNOWLEDGMENTS . . . . . . .

ABSTRACT . . . . . . . . .

CHAPTER

1 INTRODUCTION . . .


The Importance of Biological Rhythms . .
The Need for an Automatic System . . .

2 THE SLEEP ANALYZING HYBRID COMPUTER . .

Introduction . . . . . . . .
Operation of a Typical SAHC Subsystem .
SAHC Organization . . . . . . .
Comments on Zero Crossing Analysis . . .

3 SOFTWARE FOR PROCESSING OF PERIODICITY
DATA . . . . . . . .. .

Introduction . . . . . . . .
Digital Filtering of Periodicity Data . .
Low-Pass Filtering . . . . . . .
Running Average Low-Pass Filtering
of SAHC Data . . . .
High-Pass and Band-Pass Filtering . . .
Determination of Binary Ultradian
Patterns
Patterns . . . . . . . . .
Location of Peaks and Interval
Measurements . . . . . . . .
Treatment of REM . . . . . . .
General . . . . . . . . .

4 NORMAL ADULT PATTERNS . . . . . .


Introduction . . . . . . .
Five-Minute Running Averages . . .
Distribution Over the Night . .
Binary Ultradian Patterns, Cycle Lengths,
etc. . . . . . . . . .
Temporal Relationships ....


Page

. . iii

. . vi


1

1
3

7

7
7
12
16


18

18
21
21

28
30

34

37
38
40

41

41
44


. 49

S. 53
. 66


. .
. .









TABLE OF CONTENTS (Continued)


Page


Correlations, Autocorrelations, Indicated
Periods . . . . . . . . ... 74
Interrelationships Between Different
Activities . . . . . . . .. 77
Summary . . . . . . . . .. . 84

5 ONTOGENETIC TRENDS . . . . . ... 85

Introduction . . . . . . .... .. 85
Qualitative Observations .. . . . 85
Distribution Over the Night . . . .. 86
Binary Patterns, Cycle Length, Period
Length . . . . . . . . . 95
Temporal Relationships . . . . . 110
Correlations, Autocorrelations, Indicated
Periods . . . ..... .... ........ 123
Summary .. . . . . . . . . .123

APPENDIX

1 PROGRAM LISTINGS . . . . . ... 127

2 THE USE OF SPECTRAL TECHNIQUES IN THE
ANALYSIS OF ULTRADIAN BIOLOGICAL RHYTHMS 177

3 DATA FROM ALL AGE GROUPS . . .. . . 197

BIBLIOGRAPHY . . . . . . . . . 243

BIOGRAPHICAL SKETCH . . . . . . ... 247









Abstract of Dissertation Presented to the Graduate Council
of the University of Florida in Partial Fulfillment
of the Requirements for the Degree of Doctor of Philosophy


AUTOMATED ANALYSIS OF BIOLOGICAL
RHYTHMS IN THE HUMAN ELECTROENCEPHALOGRAM

By

Barry Patrick Keane

June, 1975


Chairman: Dr. Jack R. Smith
Major Department: Electrical Engineering


This dissertation reports on the analysis of biological

rhythms in the human electroencephalogram (EEG) using

modern hardware and software techniques developed expressly

for this type of analysis. Based essentially on their

frequency range, several types of EEG activity are defined,

ranging from delta (.5 to 2.0 Hz.) to beta (15 to 33 Hz.).

Several methods of detection of the above activities

are discussed, including the technique used for this study,

based on zero crossing and pattern detection.

Software for preliminary processing of the above

mentioned raw data is described, including various digital

filters and special purpose software algorithms. Additional

software is described for quantitatively measuring the

tendency of each type of EEG activity to occur rhythmically.

In applying the system to sleep recordings, a care-

fully selected group of normal young adults is analyzed

in order to establish a norm by which to measure older








and younger groups. Next, the same measurements are made

on carefully selected subjects from other age groups,

ranging from 3 to 79 years of age, in order to show

quantitatively the evolution of the various ultradian

rhythms with age.

Results show a number of previously unknown onto-

genetic patterns in the ultradian occurences of alpha,

beta delta, sigma, and rapid eye movement (REM) activity.

The distribution over the night of many types of activity

is found to vary significantly (and predictably) with

age.

The amount of man-hours involved in doing such

studies has been reduced via automation to a small fraction

of what had previously been a prohibitively large value.















CHAPTER ONE
INTRODUCTION


The Importance of Biological Rhythms

From the moment of conception until death,
rhythm is as much part of our structure as
our bones and flesh.....Through studies of bio-
logical rhythms, many aspects of human vari-
ability--in symptoms of illness, our peaks
of strength and productivity, can be antic-
ipated. Moreover, by the end of this decade,
much that is still considered unpredictable in
health and human performance may become fore-
seeable through research into the nature of
biological time cycles. As a result, timing
promises to become an important factor in
preventive health programs and medicine.
(Luce, 1970, p.iv)

This quote by the director of the National Institute of

Mental Health emphasizes just how important biological

rhythms are in our daily lives.

Hormone levels fluctuate, moods change, strengths

and weaknesses vary, in fact "...Life and death may

hang in the balance of timing. Mortality has been

decided, experimentally, not by the amount, but by the

time of day that a rodent received X-rays or was injected

with pneumonia virus, bacteria, or drugs" (Luce, 1970,.p. 1).

In pharmacology, the most effective time to administer a

drug, or the time when it is most needed, may depend

critically upon the patient's biological clock phase.

Replication of experimental data may depend, again crit-








ically, upon the timing of the experiment.

Our emotional and physiological state at any given

time depends upon the relative phases of a number of bio-

logical oscillators. The importance of overall harmony

of these biological rhythms is expressed by Dr. Erwin

Bunning as follows:

...glandular tissue may be in the phase of
hormone production while another organ, being
in a different phase cannot make use of the
hormone; or an enzyme may be active in a
particular time when its substrate is not
available. Every transatlantic air traveler
knows the physiological discomforts that
may arise from such a lack of cooperation.
(Luce, 1970, p.11)

Neuroendocrine rhythms, for example, seem to exhibit

varying degrees of independence, as was shown by Weitzman et

al. in an experiment where the sleep-waking cycle was shifted

by 180 degrees (Weitzman et al., 1968). Some rhythms

adapted immediately to the new schedule, while others

took perhaps two or three weeks to readjust. The proper

phase-locking of these rhythms is then a steady-state

phenomenon; therefore, any abrupt perturbation of the

circadian rhythm will have associated with it a tran-

sient response or recovery period (at least greater

than two or three weeks) during which certain subsystems

are out of phase with each other. It is the steady-

state, phase-locked functioning of these rhythms that

keeps us running on an even keel. A number of studies

have been done showing the normal phase relationships

of concentrations of growth hormone, prolactin, lutein-









izing hormone, 17-OHCS, etc., both during sleep and during

a full 24 hour period (Chase, 1972; Pawel et al., 1972;

Honda et al., 1969; Sassin et al., 1972; Boyar et al., 1972;

Weitzman et al., 1966).

Although the majority of studies involving biological

rhythms have been concerned with neuroendocrine fluctuations,

serious difficulties arise in sampling these rhythms,

as will be explained in the following section of this

chapter.

The critical importance of biological rhythms is

more than enough motivation for looking into new and

more efficient means of analysis.


The Need for an Automatic System

Techniques presently used in analyzing biological

rhythms almost invariably involve an overwhelming amount

of work to produce a time series that adequately shows

the cycling of a given rhythm. If a completely automatic

system could be developed for doing this type of analysis,

a great deal more work could be done in a much shorter time,

thus making it possible to establish a body of normative

data.

Of all the possible parameters that exhibit rhythmical

fluctuations, electrical phenomena are probably the easiest

to monitor. For example, sampling of hormone concentrations

in blood plasma or urine at sufficiently short intervals

requires a tremendous amount of time and effort to collect









and assay the samples. Common practice in such cases has

been to sample at approximately 20-minute intervals

(Chase, 1972; Weitzman et al., 1973). Physical limit-

ations prevent sampling at much shorter intervals, yet

it has been shown in some cases that significant changes

can occur in less than 5 minutes, indicating that the

methods presently used may very well miss major changes

(Chase, 1972).

In addition to this drawback, it would seem that in

measuring hormone levels via the peripheral blood supply

there would necessarily be some integration of the rhythm,

thereby making this method a somewhat questionable indicator

of the actual rhythm of hormone production.

Data collection can be just as cumbersome for EEG

studies when done entirely by hand, for example, counting

REM's or adding up alpha time, etc. Nonautomatic tech-

niques are generally quite accurate for counting phasic

events, but a tremendous amount of time can be spent in

analyzing or marking just one record, let alone an entire

group for an experimental study. Introduction of human

biases that defy replication is also a problem with this

approach, but can be minimized by using multiple scoring.

and cross-checking (Agnew and Webb, 1972).

A number of semiautomatic techniques have been used

which reduce the required man-hours considerably, but

these usually involve a tremendous amount of computer

time. For example, some investigators have used ;!









the technique of doing a direct A/D conversion of the EEG

signal, then doing a spectral analysis using a Fast Fourier

Transform (FFT) program to find the power spectrum of short

epochs, typically 1 or 2 seconds in duration (Lubin et al.,

1973). Each spectrum is then integrated over the desired

frequency range to give the power in that range. In order

to cover a reasonable frequency range, each 2-second epoch

must consist of roughly 500 points. A 512 point FFT must

be calculated for every 1 or 2 seconds of EEG--this requires

a truly spectacular amount of computation time to process

an 8-hour record (Otnes and Enochson, 1972). In addition,

only delta, due to its high amplitude and power, has been

reduced effectively by this method to a series that clearly

depicts a well-defined nightly pattern.

A fully automatic system that could reduce the EEG

record to a relatively small number of values in a

reasonable amount of time would be an invaluable research

tool for analyzing rhythmical EEG phenomena. This need,

along with the critical importance of biological rhythms

in our daily lives, has been the major motivation for

this research.

Chapter Two describes a fully automatic sleep analyzing

computer system and related detection techniques. Also

discussed is the nature of the output data from the com-

puter.

Included in Chapter Three is a detailed description

of a software system for reducing the raw data from the






6


sleep analyzing computer to a relatively small number of

meaningful and descriptive parameters.

Chapters Four and Five are devoted to actual data

collection using the system. Chapter Four is concerned

with only normal young adult subjects, thus establishing

what can be considered to be normal mature patterns. Chap-

ter Five then describes how these patterns vary from age

group to age group, dealing with subjects ranging in age

from three to seventy-nine years.
















CHAPTER TWO
THE SLEEP ANALYZING HYBRID COMPUTER


Introduction

The system described in this dissertation handles

information extracted from EEG recordings by the Sleep

Analyzing Hybrid Computer (SAHC), the product of several

years research by several people working in the lab of

Dr. J. R. Smith at the University of Florida. This

chapter explains the basic operation and organization of

the SAHC and explains briefly why certain techniques of

analysis offer advantages over other methods.


Operation of a Typical SAHC Subsystem

The SAHC contains a number of "event detectors" as

subsystems used to detect the presence of various types

of EEG activity (Smith et al., 1975; Bremer, Smith, and

Karacan, 1970; Smith and Keane, 1973; Keane, 1972). Figure

2.1 is a functional block diagram of a typical event

detector depicting the various decision processes involved

in detection of a particular EEG waveform.

The first step in the processing of the signal is

band-pass filtering to attenuate lower frequency baseline

drifts and higher frequency activity superimposed on the

waveform of interest. A wide-band filter with 12 db/octave




































DETECTION


Figure 2.1 Block Diagram of a SAHC Event
Detector.








roll off is used. Table 2.1 lists the cutoff points for

each SAHC filter. The filter output is connected to both

the A/D circuit and the amplitude measuring circuit. The

amplitude measuring circuit is optional and presently is

only included in the delta detection circuitry. During

each cycle, this circuit determines if a certain amplitude

threshold has been exceeded.

The A/D circuit detects each negative to positive

zero crossing, sending out an indicative pulse that is

synchronized with the system clock cycle. Hysteresis is

included at this point in most of the zero crossing detec-

tors to reject "chatter" due to low amplitude high frequency

noise. Only negative to positive zero crossings are marked,

so that the interval examined between zero crossings is a

full cycle. The full cycle measurement (as opposed to

half cycle measurement) is used since it is not sensitive

to baseline drifts, and it exhibits improved noise charac-

teristics (Gondeck, 1973).

The frequency discriminator circuitry measures the

time interval between consecutive zero crossings and

decides whether a given interval is within preprogrammed

(hardwired) limits. The time interval and corresponding

frequency limits are listed in Table 2.2.

The pattern recognition circuit monitors the activity

of the zero crossing and frequency discriminator circuits

and makes the ultimate decision as to whether or not a

given activity is present. The actual circuitry for real-


























Table 2.1 Corner Frequencies (-3 db) of the
SAHC Analog Filters.


LOWER CUTOFF
FREQUENCY

7.00 Hz.
13.00
0.15
11.00
8.00


UPPER CUTOFF
FREQUENCY

17.00
40.00
3.00
29.00
120.00


ALPHA
BETA
DELTA
SIGMA
ARTIFACT























Frequency and Period Limits of Digital
Frequency Discriminator Circuitry in
the SAHC.


12.00 Hz.
33.00
2.00
16.00


0.0833 sec.
0.0303
0.5000
0.0625


Table 2.2


ALPHA
BETA
DELTA
SIGMA


8.00 Hz.
15.50
0.50
11.75


0.125 sec.
0.645
2.000
0.0851








ization of this decision process varies from detector to

detector, but all perform basically the same function--If

a high percentage (typically 75-80 percent) of the cycle

lengths are within the prescribed limits, then the pattern

recognition circuit signals that the activity is present.


SAHC Organization

Figure 2.2 is a functional block diagram showing the

organization of switching units and subsystems within the

SAHC.

Inputs can be external and on-line if desired, but

usually are taken from the tape recorder, a SANGAMO 3500

14-channel FM recorder. Five channels are selected for

monitoring, and are filtered by the 60 Hz. notch filters

to remove 60-cycle noise. Timing can be selected from

any tape channel, is notch filtered also, and sent to the

time code reader.

Inputs to each of the SAHC detectors may be dialed to

any of the 5 monitored channels, but are connected to the

channels indicated in Table 2.3 except in special cases.

The electrode placements are based on the international

(10-20) electrode placement system.(Strong, 1970; Laidlaw

and Stanton, 1966).

The time code reader reads the elapsed time from the

appropriate tape channel or from the interval timing unit.

The internal timing unit is particularly useful for tapes

with noisy timing channels or for those that have no time

code.
























ENT.
INPUTS


p-'
EXTERNWA CONTROL FROMt POPS W
CASSETTE OUTrPUT BUFFERS
OUTPUT TO PDP 8 UNIT




Figure 2.2 SAHC Architecture.





















Table 2.3 Channel Selection for SAHC Detectors.


GENERAL
CORTICAL ELECTRODE
DETECTOR CHANNEL AREA LEADS

ALPHA 3 OCCIPITAL 03--OZPZ
BETA 1 FRONTAL F1--F7
DELTA 1 FRONTAL F1--F7
SIGMA 2 PARIETAL-TEMPORAL* Pl--T5*
ARTIFACT 1 FRONTAL Fl--F7

*Sigma is sometimes recorded from C3--A2.





15


The total amount of each type of activity is summed

up over an entire minute, then strobed into the output

buffers. Then, while the data for the next minute are

being measured, the information in the buffers is entered

serially onto the cassette tape. The SAHC can operate

real time or, by speeding up the tape recorder, can be

run at 32 times real time, making it possible to run an

entire 8-hour record in about 15 minutes. For alpha, beta,

delta, and artifact, the sums represent (are proportional

to) the total time that each activity is present during

the minute. For REM and sigma, the sums are the total

number of REM's and sigma sleep spindles detected during

that minute.

By monitoring the detectors and summing circuits during

each minute, the sleep stage logic circuits indicate, at

the end of the minute, the appropriate sleep stage for that

minute, based on the sleep scoring rules of Rechtschaffen

and Kales (Rechtschaffen and Kales, 1968). This information

and also the elapsed time from the time code reader are

strobed into the output buffers along with the detector

sums and are also entered onto the cassette tape.

After an entire record has been processed, the cassette

tape may be removed and stored for later processing, or it

can be read immediately, via PDP8 software, onto a DECtape

for permanent storage (Digital Equipment Corporation, 1972).

During the write swquence (or storage sequence) the cassette

unit is controlled by timing signals from the SAHC, but








during the read mode, as information is transferred

onto a DECtape, the unit receives controlling signals

from the PDP8. Once entered on DECtape, the data can

be stored for an indefinite amount of time and may be

referred to or processed at any time.


Comments on Zero Crossing Analysis

Zero crossing analysis has been shown to offer a

number of advantages over frequency analysis or spectral

analysis (Keane, 1972). The most objectionable aspect of

zero crossing analysis seems to be that, intuitively

speaking, a great deal of information appears to be lost

in the binary quantization of the raw data (Evans and

Mulholland, 1969). Further insight would indicate, however,

that such is not necessarily the case. In most EEG wave-

forms, the shape of the waveform between zero crossings

is not usually the same from one cycle to another, but

varies somewhat randomly. This randomness in shape from

cycle to cycle tends to cancel additional information

from amplitude variations between zero crossings--the greater

the randomness, the less the information loss.

It can be shown that for a wide sense stationary,

gaussian process with zero mean and normalized autocorre-

lation function, P (T), a binary quantization or "hard

limiting" yields a process y(t) with autocorrelation

Ry(T) 2 sin- p(t) (Thomas, 1969).

This is to say, the resulting autocorrelation falls off









faster with increasing T--the resulting spectrum has there-

fore been broadened. Working with the zero crossing pattern

of the waveform does not therefore necessarily mean that

such a technique is to be rendered significantly less

effective by a loss of information content.

Perhaps the most significant advantage associated

with this technique of analysis is its insensitivity to

amplitude variations. Amplitude varies widely from sub-

ject to subject, therefore discouraging the use of ampli-

tude sensitive analysis techniques such as spectral analysis

and frequency analysis. In addition, zero crossing is

more easily interpreted for nonstationary data since,

unlike the other techniques mentioned, stationarity is

not a primary prerequisite to this method of analysis.

The incorporation of pattern recognition hardware as

well as period discrimination yields a system that is

quite effective in detecting EEG waveforms for subjects

with vastly different physiological characteristics. This

technique has been applied in a number of EEG studies and

has been demonstrated to produce results showing consid-

erable improvement over other automatic techniques (Smith

and Karacan, 1973).















CHAPTER THREE
SOFTWARE FOR PROCESSING OF PERIODICITY DATA


Introduction

Several different types of information are extracted

from the raw SAHC data and stored on a system data tape.

Figure 3.1 depicts schematically the mechanics of the

software involved in the process. All software is imple-

mented on a Digital Equipment Corporation (DEC) PDP8/e

minicomputer. All stored data, including the raw SAHC

data, are stored on DEC's DECtapes, each of which has

a maximum storage capacity of 1471 blocks, each block

consisting of 128 12-bit words (Digital Equipment Corpor-

ation, 1973).

The top line of boxes in Figure 3.1 represents

information extracted from one night of recorded EEG.

The number above each box gives the relative position of

each section of data on the data tape. For instance,

if information for a particular record begins at block N,

then statistical information concerning that record is

stored in block N+16. Note that each of the sections

of data requires blocks of the DECtape.

Blocks 0 through 4 are set aside for use as a

directory. Information for the first record begins in

block 5.










Each of the programs shown in the lower row of

blocks in Figure 3.1 will be explained in detail in

following sections of this chapter, but their main

functions can be summarized as follows:

RNAV5--Filters raw SAHC data by implementing
a running average low-pass and band-pass
digital filter. Low-passed data go into
block N, band-passed into block N+4.

RNAV6--Does additional smoothing of the data
from block N via a 2-pass running average
filter. Output (to be used by PEAK program)
goes into block N+20.

GATE--Generates binary ultradian pattern by
locating "active" portions of record.
Results go into block N+12.

ULTRAD--Cleans up ultradian pattern (from
block N+12) by removing short duration
transitions. Results are returned to
block N+12.

ULTRD2--Further processes the ultradian pattern
by removing pulses that are not signifi-
cantly higher than surrounding data points.
Results go back into block N+12.

PEAK--Locates peaks within active pulses in
data from block N+20, puts peak locations
into block N+8. Measures distance between
peaks, records this information filed
by hour of occurrence and number of
occurrences. Results go into block N+16,
the first 180 locations.

ULT3--Measures length of each ultradian period
from data in block N+12, stores length
and minute of onset in locations 200
through 240 of block N+16.

REM--Obtains REM binary ultradian pattern
directly from SAHC sleep stage data.
Enters directory information in block N
and binary ultradian sequence in block
N+12.








H424

D


0


Figure 3.1 Software Processing Sequence and Data
Tape Format.









REMULT--Does final processing of REM binary
ultradian sequence by removing short
(less than 5 minutes) bursts of l's and
O's.


Digital Filtering of Periodicity Data

The raw data from the SAHC, consist of 1-minute

sums proportional to the amount of activity present

during that minute. For example, if the ith value of

this time series for beta is n, then there were K times

n seconds of beta activity detected during minute i, where

K is a constant. For beta, K is 60/72, Initial processing

of this raw time series consists basically of 1) low-

passing the data to smooth out high frequency fluctuations,-

that is frequencies above the highest frequency of interest

in periodicity analysis, and 2) band-passing the series

to accentuate any fluctuations of 10-to-20 minute period

length.


Low-Pass Filtering

The simplest method to implement that will produce

a low-pass or smoothing effect is a moving weighting

function or moving average filter (Steiglitz, 1974).

All nonrecursive digital filters are of this type since

they represent a weighted sum of past, present and

future values. There are some particularly appealing

advantages to using moving average or finite impulse

response (FIR) filters. Perhaps the most significant

advantages would be the following: 1) Moving average,








nonrecursive digital filters are always stable,

and 2) Quantization and roundoff problems are generally

negligible for nonrecursive filters (Rabiner, 1971).

This approach may be expressed as

N/2

yi=- akxi-k ,
k=-N/2

where the existence of non-zero ak's for k<0 indicates

a noncausal filter. This imposes some restrictions in

that it is not realizable in real time, but poses

no problems as far as implementing the filter using the

computer, or if the filter is modified to include a

delay so that

N/2

Yi' Yi-N/2= akXi-k-N/2.
k=-N/2

The most readily implemented low-pass filter of this

type would be the case where all the ak's are equal--

the moving average filter given by

N/2

yi= E akxi-k where ak=l/(N+l) for all k
k=-N/2

or

N/2

yi=1/(N+l) E Xi-k

k=-N/2

The weighting coefficients for this filter are given in









Figure 3.2.

This method may be extended from this simple one-pass

nonrecursive case to a two-pass, three-pass, etc. to get

increased attenuation at high frequencies. This would be

equivalent to cascading a number of the simple one-pass

versions. Note that these multiple-pass filters can be

combined to a single one-pass where the ak's are no longer

equal. For example, a two-pass version can be implemented

in either of the following two ways: 1) Cascade two one-

pass windows of lengths N+l (Figure 3.3a). 2) Combine to

a one-pass window of length 2N+1 (Figure 3.3b). For each

one-pass,

N/2

yi=l/(N+l) Xik

k=-N/2

N/2

and zi=l/(N+l) Yi-1

k=-N/2

N/2 N/2

E 1/(N+1) E Xi-l-k
1=-N/2 k--N/2

This is equivalent to the one-pass filter where

N

zi amxi-m

m-N

where the m coefficient, am, is given by letting xi-m

equal unity, all other x's equal zero, and finding the













l-1/(N+1)





a2 ----- aN/2


a-N/2 --- a-2


Figure 3.2 One-Pass Running Average Low-Pass Filter
Coefficients.




















I- iN+l
-ak= 1/(N+1)
ak= 1/(N+1)
4


-- N+ -

ak= 1/(N+1)


2N+1

ak=(1/(N+1)) 2(N+1-k)


Figure 3.3 Two-Pass Running Average Low-Pass Filter
and Its Single-Pass Equivalent.








value of zi (which under these conditions will equal

am) .

To find ao, let xi_0=l, all other x's = 0. Then

N/2 N/2 1
zi= 1/(N+l) 1
1=-N/2 k=-N/2 l+k=0

giving a0=(l/(N+l))(1/(N+1)) (N+) =1/(N+) .

For aN, let xiN=l, all other x's = 0. Then

Z N/2 1/(N+1) N/2 ]
SaN= I/(N+I) 1
1=-N/2 k=-N/2 k+l=N

giving aN=(1/(N+1)) (1/(N+1)) (1)

In this manner it can be shown that the am's are given by

am=(I/(N+1))2(N+l-m) .
These weighting coefficients form a triangular window as

shown in Figure 3.4. By the same approach, any multiple-

pass, nonrecursive filter can be found to be equivalent

to a one-pass with appropriately modified weighting

coefficients and window length.

Although modifying a multiple-pass to a 1-pass

moving average filter involves fewer steps and may appear

simpler, it generally involves considerable computation

time. For instance, if an evenly weighted average filter

(all ak's = 1/(N+1) ) is to be implemented three times

(i.e. three passes are to be made) on a series of 500

data points, the time required can be shown to be approx-

imately (1,000TA + 500TM) x 3, where TA time required to

implement addition, TM = multiplication time (Otnes and
















(1/(N+1)) 2(N+1)


_i ._ (1/--( (N+1))2


a-2 a-1 a0 al a2


Figure 3.4 Coefficients for Single-Pass Version of
Low-Pass Filter from Figure 3.2.


-_ A








Enochson, 1972). Note that this is independent of window

length. The one-pass version of the same filter requires

about 3 x (N+l) x 500(TA+TM) On a machine that uses

software multiplication, TM>>TA, so that the total execu-

tion time for these filters is Te3 1500 TM for the three-

pass and Tel 1500 x (N+1)TM for the one-pass, indicating
el m
that the multiple-pass filter is faster, becoming increas-

ingly superior with increasing N. Even with a hardware

multiplier giving a significant reduction in T the three-

pass version is still independent of window length and soon

becomes faster as N increases. In the limit as TM--*TA,

the execution times would be Te3 Z 4500TA and Tel 3000TA,

for the three-pass and one-pass, respectively, so that

the multiple-pass version still gives a significant decrease

in execution time as N increases. When multiple-pass

filtering is to be used in this processing system, it is

implemented as a multiple-pass and is not modified to a

single-pass.


Running Average Low-Pass Filtering of SAHC Data

As the raw data of 1-minute sums are read from the

SAHC data tapes, it is low-passed by a single-pass running

average filter with a five-minute averaging window. This

smoothes out the data and is effective in attenuating the

high frequency fluctuations present in the raw data without

significantly altering the background trends which are of

interest. Figure 3.5 shows typical plots of beta data









i.
1.0
Raw Beta Activitlr
(Normalized) r .5

Jl.iU I i. i ,
1 2 3 4 5 6 7 Hours
of Sleep




Filtered
Beta Activity A
(Normalized) *A V


1 2 3 4 5 6 7 Hours
of Sleep


Figure 3.5


Low-Pass Filtering (by RNAV5) of Raw
SAHC Data.









from the SAHC. The top plot is the raw data, 1-minute

sums generated by the SAHC. The lower plot is the smoothed

version which would be entered into data block N for this

particular record by the RNAV5 program. To insure that

minute fluctuations do not affect the peak detection algo-

rithm, additional smoothing is done by the RNAV6 program

which implements a two-pass filter using a 3-minute data

window on each pass. The results of two successive 3-

minute smoothings of the beta data from Figure 3.5 are

shown in Figure 3.6. Small high frequency fluctuations

that could disturb or slow down the peak detection program,

PEAK, have been removed, but the trends of interest have

not been significantly altered.

It should perhaps be mentioned at this point that the

RNAV5 program enters directory information into block N

at points above 500. This includes record number, data,

number of minutes of data, type code (alpha, beta, etc.),

scale factor, and lengths of filter windows. Directory

information stored in this manner is readily accessible

during processing without having to make long excursions

on the tape--that is, information concerning each record

is stored right along with all other data. A centralized

directory arrangement would require moving the tape from

each record to the central directory for each reference.


High-Pass and Band-Pass Filtering

A nonrecursive high-pass filter can be implemented













Filtered Beta
(First Pass)


.0


L-i r


1 2 3 4


(Second Pass)











(Third Pass)


5 6 7 Hours
of Sleep


1 2 3 4 5 6 7 Hours
of Sleep


7 Hours
of Sleep


Figure 3.6 Additional Smoothing of SAHC Data.









by the same basic procedure as the low-pass, i.e. by calcu-

lating a moving average of the data. The difference here

is that the filter output is the deviation of the current

input value from a background trend given by the average

of the current point and N surrounding points:

N/2
yi=xi 1/(N+l) E xik
k--N/2

In addition to being easily realized, this filtering

approach offers a number of design advantages over more

involved techniques. In the design of such filters, one

need only determine the proper window length to give a

local trend over the desired neighborhood. Again, he

need not concern himself with instability, frequency

warping and other such problems inherent in the design

of recursive digital filters (Rabiner, 1971).

A band-pass effect can be obtained by combining the

low-pass and high-pass procedures. That is, we can implement

a band-pass filter by taking the output as the difference

between a local average of the current point and N1 sur-

rounding points and a less local average of the current

point and N2 surrounding points where N2>NI:

Nl/2 N2/2

yil1/(Ni+l) 1 xi+k 1/(N2+1) 2 i+1
k=-N1/2 1=-N2/2

The data in Figure 3.7 have been band-pass filtered by

this method using a 5-minute local average and a 20-minute

extended average. Note that the 10-20-minute cycles have






















of Sleep


Figure 3.7 Bandpass Output of RNAV5.
(Same Record as Figures 3.5 and 3.6).








been noticeably accentuated. This band-pass procedure is

implemented by the RNAV5 program at the same time as the

low-pass. (Since the short 5-minute averages are already

available, all that remains is to measure the extended

20-minute averages and calculate the difference.) The

band-passed information is entered into the second data

block (N+4) for each record. This band-passed data can be

used to investigate tendencies to show peaks at 10-20-

minute intervals during active regions. This apparent

tendency has been noted (Aserinsky, 1971) but has never

been quantitatively evaluated.


Determination of Binary Ultradian Patterns

The GATE, ULTRAD, and ULTRD2 programs take the basic

activity information from block N and produce a binary

sequence that points out the active regions. (In this

dissertation, the regions of high beta activity will be

referred to as "active" beta regions or "beta periods".)

The same terminology will be applied to other activities.

The GATE program is essentially an adaptive, zero-

hysteresis comparator. The comparator level being adaptive

allows for widely varying activity levels from subject to

subject. The level is set at 20 percent of the nights

maximum value, then the portions of the record above

the 20 percent level are marked as active periods, and the

portions below are marked as inactive regions. Rapid

transitions or "chattering" at the beginning and end of









each active period are ignored at this time, and results

are put into block N+12. Sample data are shown in Figure

3.8 a and b.

The ULTRAD program removes the "chatter" from the

binary ultradian sequence produced by GATE. The procedure

is as follows: If the comparator output (from GATE) remains

in a given state for at least 10 minutes, then that level

is considered an established state. That is, if the com-

parator level remains high for 10 minutes then it is

established that this is an active region. Once the

state is established as either active or inactive, tran-

sitions from that state will be ignored until a new state

is established by remaining in that new state for at

least 10 minutes. Sample output from ULTRAD is shown in

Figure 3.8 c. The ULTRD2 program further processes the

"active" pulses by requiring each detected pulse to pro-

trude above the local background. This program examines

the binary ultradian sequence at the beginning and end of

each active region. The single-pass filtered data from

block N are examined at points 5 minutes before and after

each transition in the binary ultradian sequence. If

the point 5 minutes inside the active region is not sig-

nificantly higher than the point 5 minutes outside the

active region, the transition is rejected. The pulse is

then made narrower until these requirements are met or

until the active region becomes less than 10 minutes in

which case it is eliminated from the binary sequence. By













RNAV5
a) Output


r



b) GATE
Output
LJ



r-
c) ULTRAD
Output




I

ULTRD2
d) Output i


1 2 3 4 5 6 ,7 -HoaUrs
of Sleep







1 2 3 4 5 67



......}1 -( ^.4 4 1- -4 ,-! J_< I-! . .-1


.1.I


....2 ....-. 3 .... 4.


5 .. -6 ... -..7 -


2 3 4


5 6 7


Figure 3.8 Determining the Binary Ultradian Pattern.









"significantly higher" in this case is meant a difference

of greater than 20/150 of the maximum over the night, so

this measurement is again adaptive and is automatically

adjusted for widely varying subjects. The data processed

by ULTRD2 are shown in Figure 3.8 d.


Location of Peaks and Interval Measurements

Peaks within the active regions are detected in the

ultradian data by the PEAK program. This program finds

peaks in the active regions and measures the lengths between

the peaks. The output from this program is 1) the locations

of the peaks, stored in block N+8 for display purposes, and

2) the length of the interval between these peaks, stored

in block N+16. These interval measurements are grouped

together according to their hours of occurrence. That is,

all intervals occurring during hour 1 are in group 1,

those from hour 2 are in group 2 etc., so'that the data

can be analyzed for variations over the night, or all the

data can be grouped together as a unit.

Peaks are located by the program in the following

manner: Each point of the multiple-pass filtered data is

examined to determine if it is a local maximum. If

2 maxima occur within 2 minutes of each other, the second

is ignored, and only the first is recorded. Intervals

between peaks are next measured and recorded. Intervals

between 2 peaks occurring in different ultradian cycles

are ignored-that is, the entire interval must occur within









the same active period. Output of the PEAK program for

a typical subject is given in Figure 3.9.

In addition to the intervals between peaks within

active periods, ultradian cycle measurements are also

made. The ULT3 program measures the length of each active

period from the binary sequence (block N+12) and records

the length and onset time in the statistics block (block

N+16). With this information, statistics programs can

measure period length, cycle time, etc.


Treatment of REM

Inasmuch as sleep staging has already been computed

by the SAHC and stored on SAHC data tapes, very little

processing is necessary by the REM program to create a

binary ultradian sequence. The REM program merely reads

the sleep stage data from the SAHC data tape and creates

a binary sequence, entering a "1" for REM and a "0"

for all other stages. Directory information is also

read from the SAHC directory and entered in block N

at points above 500 so that the directory conforms to the

cataloguing set up for all other data.

The REMULT program produces the final binary pattern

by removing short bursts that would interfere with the

calculation of average period and cycle lengths etc.

During active regions, an inactive pulse of less than

5 minutes is removed. Similarly, during inactive periods,

a REM period shorter than 5 minutes is removed.















Normalized
Beta
Activity



Peaks


1.0

.5 .


1 2 3 4 5 6 7 Hours

of Sleep


1 2 3 4 5 6 7 Hours
Figure 3.9 Peak Detection Within Active Regions. of Sleep








General

With the exception of the RNAV5 program, the processing

sequence is run for one group of data at a time. If a

group contains 10 records, all 10 records are processed

at once. The RNAV5 program runs records as they are grouped

on the SAHC data tape. If a group of records have been

entered consecutively on the SAHC data tape, they may be

processed simultaneously by RNAV5. Each type of data

processed is stored on its own data tape: All alpha data

are stored on the alpha data tape, all beta data are

recorded on the beta tape, etc.

A number of statistical programs for calculating

means, correlations, distributions, etc. are used to

extract group characteristics from the data. Each of

these programs will be described in Chapters 4 and 5.















CHAPTER FOUR
NORMAL ADULT PATTERNS


Introduction

The different types of activity show varying degrees

of organization in their respective young adult patterns.

A separate chapter is devoted to the young adult patterns

since it is felt that these data best represent the normal

interrelationships of the various EEG activities. The

data for this age group are unobscured by either develop-

mental phenomena or aging.

The experimental design for the age group study,

shown graphically in Figure 4.1, is basically a nested

design of three or four levels (Mendenhall, 1968). At

level one, the data are divided into five age groups.

Table 4.1 lists the ages of various subjects in each

age group. At level two, five normal subjects have

been selected for each age group. This chapter is devoted

to describing the Group 2 data.

At level three, there are two nights of recorded

EEG for each subject. Neither of these recordings has

been taken from the subject's first night in the

laboratory in order to eliminate the "first night

effect" (Agnew et al., 1966).














GROUP 4


LEVEL 1
n1-5


~SUBJ. 5 LEVEL 2
n2-5






LEVEL 3
n3-2





LEVEL 4
n4--variable

Figure 4.1 Design Structure for
Age Group Study.

















Table 4.1 Ages of Subjects in Each Age Group.


GROUP 0

10994 Age 4
10998

10720 Age 4
10724

10350 Age 5
10356

10719 Age 3
10723

10995 Age 3
10999


GROUP 1

10104 Age 13
10114

10110 Age 13
10115

10133 Age 13
10140

10141 Age 13
10147

10164 Age 13
10159


GROUP 2

11742 Age 20
11747

11722 Age 27
11725

6538 Age 30
6546

10250 Age 34
10256

10837 Age 26
10843


GROUP 3

10067 Age 47
10071

10349 Age 43
10352

10927 Age 52
10931

10897 Age 52
10918

10889 Age 53
10896


GROUP 4

11673 Age 79
11677

11740 Age 70
11745

11777 Age 69
11781

11281 Age 78
11286

11365 Age 68
11370








At level four, the data are sampled differently

according to the type of data. For example, for the

temporal statistics, eight samples are taken from each

subject--one measurement for each hour of sleep.


Five-Minute Running Averages

Five-minute running averages (from program RNAV5)

for alpha, beta, delta, and sigma are given in Figures

4.2 through 4.5, respectively. These data were obtained

from SAHC data tapes via RNAV5, as described in Chapter 3.

Perhaps the most striking of these data sets are the

beta data. Beta shows a strong, very well-organized

ultradian pattern throughout the night. Although the

amount of beta appearing in the first half of the night

is clearly less than the amount appearing during the

second half of the night, it is interesting to note

that this is due to shorter period lengths and not due to

a lower amplitude. That is, within any beta period,

the amount of beta per minute does not vary significantly,

but period length does increase over the night. It is

therefore possible at this point to make a statement

concerning the nature of the beta modulation process-

namely that it is the period length that is modulated

in this case. The beta appears periodically over the

entire night and is never suppressed entirely. This

might be :interpreted as meaning that whatever the

physiological processes associated with the appearance

of the beta periods are, they are not scaled down during



















































1 2 3 4 5 6 Hours


Figure 4.2 Plots of Alpha Activity
for Group 2.






















































1 2 3 4 5 6 Hours


Figure 4.3 Plots of Beta Activity
for Group 2.





















































1 2. 3 4 5 6 Hours


Figure 4.4 Plots of Delta Activity
for Group 2.





















































Figure 4.5 Plots of Sigma Activity
for Group 2.






49


the first half of the night, but are proceeding "at full

speed," identically as they do during the second half of

the night.

The modulation of delta activity on the other hand

is of a different nature. During the later two-thirds

of the night, the delta activity is greatly attenuated,

obscuring what seems to be a tendency to continue to

appear periodically over the entire night. In this case

it would seem that the physiological processes themselves

are perhaps being suppressed or scaled down.


Distribution Over the Night

Figure 4.6 shows the calculated "moments about the

midpoint" for several types of activity. This value gives

an indication of how strong the tendency is for a given

type of activity to be greater in either the first or

second half of the night. A negative number (as for

delta and sigma in this case) indicates greater activity

in the first half of the night, whereas a positive

value (beta, alpha) indicates greater activity in the

second half of the night. The values are normalized so

that each record is evenly weighted regardless of the

overall level of activity. For some groups the amounts

of a given activity may be extremely low (i.e. delta

in Group 4 or sigma in Group 0). In these cases it

is felt that the record may not properly reflect the

nightly distribution, so these records are not included

in the calculation of the group moment. The criteria for





50







HOIEHTS BOUT THE MIDPOIHT--GROUP 2
ALPHA. BETA, DELTA, SIGHR


URLUE <== + ==>
-969.00 -480.80 8.80 488.66 968.88
4-----4-----4-----+-----+----------+--- ------ ----






ALPHA *





*BETA


-52. : : .------











: : : : :+-----------
DELTA : *
-1816. :---*---I *







-252 -- : :---------


Figure 4.6 Moments About the Midpoint--Group 2.









exclusion are given in Table 4.2. These criteria have

been established by examining the data to determine

placement of the cutoff points to exclude records

showing extremely low activity.

In order to calculate the moment about the mid-

point, each record is normalized by making the maximum

(in the 5-minute running average) over the night equal

to 150 and scaling all other points appropriately.

Each point then contributes to the overall moment a

value equal to the product of the normalized activity

level just described times the distance from the midpoint.

Points located before the midpoint have been arbitrarily

assigned a negative value, whereas points after the

midpoint contribute a positive value. A listing of this

system program, MOMENT, appears in Appendix 1.

A high positive value for beta and a high negative

value for delta reflect the widely ranging nightly modu-

lations previously described for these activities.

The sigma patterns show a well organized ultradian

rhythm whose modulation is less pronounced than that of

either delta or beta. The group moment for sigma is small,

but shows such wide variation that it is impossible to

make a general statement about the sigma modulation.

The alpha data show no clear ultradian pattern.

Although there seems to be a slight ultradian tendency,

it is often impossible to extract the active regions

from the background "noise." The group moment for alpha
















Table 4.2




SAHC Count

Equivalent
Seconds


Rejection Criteria


BETA DELTA

8 8

9.47 4.13


*SIGMA depends on
to time.


for Moment


SIGMA

5

*count, not


count, not


Calculation.


ALPHA

8

4.13


convertible









indicates that it is fairly evenly distributed. The

variation is small enough in this case that it is not

unreasonable to state that the alpha is distributed evenly

over the night in most cases.


Binary Ultradian Patterns, Cycle Lengths, etc.

In order to find cycle lengths, period lengths, number

of periods, etc., each minute of every record is con-

sidered as part of either an active or inactive region.

A value of 1 is assigned to active minutes, and zero is

assigned to inactive minutes. These active and inactive

regions are determined by a 3-stage process. In stage 1

(system program, GATE) the active regions are selected

as those regions where the 5-minute running averages

are at least 20 percent of the night's maximum. The

next step (system program, ULTRAD) removes short bursts

of inactivity during active regions and vice versa. In

order to be established as an active (or inactive)

region, the input data must remain at 1 (or 0) for 10

consecutive minutes. Shorter bursts within these regions

are reset to the dominant value for that region. The

last step (system program, ULTRD2) checks to be sure that

the active regions protrude significantly above the back-

ground noise. The criterion here is that the difference

between the activity levels five minutes inside the

active region and the level five minutes outside the

region must be greater than 40/150 of the night's maximum,

this value having been chosen heuristically. The program





54


then eliminates pulses that have thus been shortened to

less than 10 minutes.

The length and beginning minute of each active pulse

are measured by system program ULT3. The information is

then stored in block N+16 (see Chapter Three) in the format

indicated in Figure 4.7. Statistical information about

period length and cycle length is measured by system

program GSTATS. In this case "period length" is to

be defined as the width of the active pulse from the

binary ultradian pattern, and "cycle length" is to be

the time between the minutes of onset of two consecutive

active periods. In computing the mean cycle lengths,

the GSTATS program detects "missing" active periods

(missed REM period, beta period, etc.) by excluding from

the ensemble any cycle lengths greater than 140 minutes.

No restrictions are placed on period length; however

periods that extend to the very beginning or end of the

night are excluded from the ensemble since it is felt

that these periods may have been cut short by the temporal

observation window.

The mean cycle lengths for Group 2 are given in

Figure 4.8. It is interesting to note that the mean

cycle length of sigma is somewhat higher than the other

values. The cycle length data have been tested for

significant differences using a paired-difference t-test

(Mendenhall and Sheaffer, 1973). Results (Table 4.3)

indicate that the sigma cycle lengths are significantly
















PERIOD


MINUTES OF


LENGTHS ONSET
200 200+1 220 I 220+1 240


LENGTH
OF Ith
ULTRADIAN
PERIOD


BLANK


BEGINNING
MINUTE
OF Ith
ULTRADIAN
PERIOD


Figure 4.7 Format for Ultradian Period Data Storage
on DECtape.


NO. OF
ULTRADIAN
PERIODS













Average Period

(inns.)
.118.8





104.4






90.0





75.0


BETA DELTA SIGMA
(94.3) (94.2) (106.5)



Figure 4.8 Mean Cycle Lengths for
Group 2 Subjects.


REM
(94.3)




















Table 4.3 Significant Cycle
Group 2 Data.


BETA

*


DELTA

*


Length Differences for


SIGMA


*.

.01
.01


REM


NS

NS

NS
.01


*No ultradian rhythm for alpha.


ALPHA

BETA

DELTA

SIGMA









longer (p<.01) than the beta or delta cycle lengths.

This at first seems astounding, but closer scrutiny

reveals that this is a direct result of the sampling

procedure. For example, consider the 4 cycles of

simulated beta and sigma data in Figure 4.9. If the

beta cycle lengths are all considered to be T seconds

long, but with increased active time during each cycle

(so that P4 >P3 >P2> P) then each corresponding sigma

cycle length will be T seconds plus the next active

period length for beta, minus the last active period

length for beta. The increasing trend in beta periods

results in a net increase in the sigma cycle time for

each sample. For this reason, the sigma cycle time does

not properly reflect the actual ultradian cycle time,

but instead gives an inflated value.

The magnitude of the variance for most types of

activity is sufficiently large to discourage the use

of spectral analysis in analyzing these data. This

subject is discussed in detail in Appendix 2.

Figures 4.10 through 4.13 give mean period lengths

for beta, delta, sigma, and REM. As previously mentioned,

the alpha activity lacks any definite ultradian rhythm,

and therefore did not provide sufficient data for cal-

culating means of period or cycle length.

Other conspicuously absent data points have been

omitted due also to the small amount of data available.

For example, some plots only show periods 1, 2, and 3 or









PI


P2


H1


I: P3 P4
tc------~r c-------


T T T T






I. -I-r, i


T+P2-PI
(O.T. T)


T+P3-P2
(G.T. T)


T+P4-P3
(G.T. T)


Figure 4.9


Simulated Data Showing Cycle
Length Sampling Procedure for
Sigma and Beta.


BETA







SIGMA .





60










HERN LENGTH OF SUCCESSIVE BETR PERIODS
PERIODS 1,2, RND 3--GROUP 2


<== 4 ==>
5.00 15.80 25.00 35.88 45.86
+-----+-----+-----+-----+-----+-----+-----4----------------


S. . + .
+ .



; : a



; ; :
S a . .

I---------------------






+ a

I- -- ---1r---
a a a + S S
5 5 a


*-a----------a-a-,--------a---- ---


Figure 4.10


Mean Lengths of Successive Beta
Periods for Group 2.


URLUE


1st
18.5







2nd
25.1







3rd
37.9
















MERN LENGTH OF SUCCESSIVE DELTR PERIODS
PERIODS 1,2. RND 3--GROUP 2


<(== + =>
8.60 24.80 46.80 56.88 72.88
-----+----+-----+----------+---------+-----+----+-----+


: --*--I


: + :

: + : : :
: .+ : : :
: + : : :
: I---- -- -- ----
: + : :
: + :
: + .


: + :
4 .

:* 4. .
: +


: + :



: + : : :
: .+ :
: + :
4. ..


: + :
I------------------ I

*





: + :


* + *
* + ,
* + C
* 4 .
* + .


Mean Lengths of Successive Delta
Periods for Group 2,


Figure 4.11


VALUE

4







1st
46.8







2nd
46.6







3rd
17.5 :


.-----.-,,,-.-,---.,----.-----.-----.---




62










HEAH LENGTH OF SUCCESSIVE SIGMA PERIODS
PERIODS 2.3, RHD 4 GROUP 2


URLUE <== =
10.80 38.60 58.60 70.8S 96.88
+----------+-----+-----+-----+--------------- ---------






2nd : :
72.7 : + ------ --------+





3rd: +
9. -------------------
4th :
*: + : :












.- -------------- ---------
61.2:


--.2 I------------I- -


Figure 4.12


Mean Lengths of Successive Sigma
Periods for Group 2.





63









MEAN LENGTH OF SUCCESSIVE REM PERIODS
PERIODS 1,2, AND 3--GROUP 2


URLUE <== + ==>
8.80 24.60 48.80 56.88 72.86
+-----+-----+-----+-----+-----+-----+-------------- ---

S + : : : :
: :*
: : : : : : : :
+
: :
1st +
: : + : : : :

S : + : : : :
d : : : : + : : : : :

2915.1 : : ------ -- +
*. : :


3nd: +
: 2 : + .:


: : : : + - :
3rd : : : + : : : :

42.8 : .I---------I-----*--------------I.
... .. . . .. . .


Figure 4.13


Mean Lengths of Successive REM Periods
for Group 2.









perhaps 2, 3, and 4. Points have not been plotted for

other periods in these cases because only a few points

remain in the ensemble after removing periods that

extend to the beginning or end of the record.

The beta plots in Figure 4.10 show a clearly increas-

ing trend in mean period length across the night. As

might be expected, this trend is paralleled by a corres-

ponding trend shown in the REM data. (The correlation

of beta and REM will be discussed later.) The REM data,

as shown in Table 4.4, compare reasonably well with data

from previous investigations.(Dement and Kleitman, 1957;

Verdone, 1968; Williams et al., 1974). The mean period

length of the third REM period is large but in this

case within one standard deviation of previous data. The

fourth and fifth REM periods were often interrupted by

the end of the sleep period. In this case the ensembles

for the fourth and fifth REM periods contained three

members and one member, respectively, and were therefore

omitted from the plots.

The sigma period plots in Figure 4.12 show a slight

decreasing trend over the night, having means ranging

from 72 to 61 minutes.

Figure 4.11 shows that the first two delta periods

have about the same mean value, but the third is signifi-

cantly smaller. Only one of the ten records for Group 2

showed a fourth delta period that stood out distinctly

enough above the background to be detected. Referring to























Table 4.4 Lengths of Successive REM Periods.



Period


Dement and Kleitman


Verdone

Williams et al.

Present Study


13

15.2

15


2

19

27

31.2

29


3

24

30

34.3

42


4

28

33

42.5

*


*Insufficient data available


5

34

31

25.8

*








Figure 4.4 however, it appears that several records had

a tendency to show a fourth delta period, but failed

to rise significantly above the background.

Average alpha, beta, delta, sigma, and REM times as a

percentage of total sleep times are given in Figure 4.14.

Table 4.5 and Table 4.6 show a comparison of these results

with data derived from that of Webb (Colquhoun, 1972).

Since beta occurs both during stage 1 and stage REM, the

sums of stage 1 times and REM times should be roughly

comparable to the active beta time (Keane, 1972). Like-

wise, total active sigma time can be compared with sums

of stages 2, 3, and 4, since considerable sigma activity

occurs during all of these stages (Gondeck, 1973; Silver-

stein, 1974). Finally, stages and 4 can be summed and

compared with total active delta time. REM time is of

course compared directly.


Temporal Relationships

Figures 4.15 through 4.19 show temporal character-

istics of the Group 2 data for alpha, beta, delta, sigma,

and REM. Each value represents the average amount of

active time (in minutes) for a given hour. The generally

low values for hour 8 are a result of sleep times less

than 8 hours.

Alpha time appears to be quite constant over the

night, being lowest during hour 2 and highest during

hour 5.





67












Table 4.5 Total Percent Active Time
for Various Group Data.


Delta

26.2


Sigma

58.6


REM

23.7


Table 4.6 Sleep Stage Percentages for
Young Adult Males (Age 20-29).


2, 3, or 4

70


Beta

35.4


1 or REM

29


3 or 4

20






68

RUERAGE PERCENT TIME PER NIGHT IN ALPHR,
EETAF,)ELIA5..SlIGI, RFD REM--GROUP 2


<== 4 ==>
18.00 38.86 5s0.6 70.8 90.8e
4---- 4 -----4----------------------4 4 --


I------*----I


:ALPHA








]--c--I








1-*-IDELTA:

















--*---J REM


BETA:

















:I---*---I


Figure 4.14 Percent Active Time per Night for
Alpha, Beta, Delta, Sigma, and REM.


:SIGMA




69



TOTRL RCTIUE MINUTES (RUERAGE) OF RLPHR
ERCH HOUR--GROUP 2


URLUE <== =">
-5.00 5.00 15.00 25.08 35.86
+-- ---------------- -- ---------4- 4---4
*
+
(1): : : : +
: : : + : : : : :


5.9 : ------*--- + : : : : :
: : : : : + : : : : :
: : : : + : : : : :
(2) : :
1.5 : --- --- +: :


(3): : : : :
5. 1: I-- ------- -- : : : :
+ : :

(4) :
5.1 ------------ + :
: : : : : : : :
: : : : : : : :
(5) .
8.3 ---------------*---------------
: : : : : : : :
: : : : + : : : :

(6) +
5.3 : ---------- ------------I
: : : : + : : :
: : : : :

(7) : :
5.8: :I-----------*- -------- : :




Figure 4.15 Temporal Distribution of Alpha Activity
for Group 2.









TOTAL RCT1VE MINUTES (RUERAGE) OF BETA
EACH HOUR--GROUP 2


URLUE
8.50
4-----+-----4



(1) :
12. 1-------



(2)
17.8



(3)
15.5




28.5



(5)
23.8



(6)
26.3



(7)
38.7:
a e a ,,,,


<== + ==>
11.56 22.58 33.58 44.56
-- 4---+------4------ ----+-----+-----+ 4-


: : + : : : :
: : : : :



I---------------*-------------
------------- + : : : :


: : : : : :
: : : : : :


I------------------: : : :


: *
: : : : : : :


: : + : : : : :

: : : : : : :
S-------------- : : :








I------------------*-----------------------I:
: : 4* : : : :

. 4 5. 5. .

-------------------I : :



:I----------------<*----------------2


Figure 4.16 Temporal Distribution of [.Beta Activity
for Group 2.










TOTRL RCTIVE MII1UTES (AUERFGE) OF DELTA
ERCH HOUR--GROUP 2


16.80


<== + = >
38.00


44.68


+-----+-----.-----+----- -----+-----+---- +---------- 4 -----


URLUE






(1)
44.5



(2)
23.6



(3)
28.9



(4)
2.7



(5)
3.5


:1----------1--


Figure 4.17 Temporal Distribution of Delta Activity
for Group 2.


2.80


58.e8


+
: : + : : :



:+ --------- ---------
+ :

: : + : : :
: : + : :

---------*----------::



+
+








*
: : + : : :
: : + : : :


I ------------*-----------I :
+*






**.
S + .





: : + : : :
I : : + : : :




72




TOTAL ACTIUE 1IHNUTES AVERAGEE) OF SIGMR
EACH HOUR--GROUP2


<== + ==>
8.80 24.00 48.00 56.86 72.86
-----+----+---------------+-------------------------


URLUE

4.



(1)
56.2



(2)
47.2



(3)
42.9



(4)
45.3



(5)
39.3



(6)
36.8



(7)


28.9 : I-------


Figure 4.18 Temporal Distribution of Sigma Activity
for Group 2.


4 : : : :. :

: : -----------*-----------I :
: : + : : : :


: : + : .
- -I-----*-------I
+ : : : :

: : : : :

------- -------*------------

: : + : :
: : + : : : :
+ : : : :

I: : -----*---- : : :

: : + : : : :


I--------------------I :



: 4 a a
*
------------ --------------I




---*----------I +:
aa





73


TOTAL ACTIVE MINUTES (RUERRGE) OF REM
EACH HOUR--GROUP 2


<== + = >
0.00 18.00 20.80 30. 0e 40.68
*-----+---------- ---+ -------- --- --- ------ 4-


(1)
1.7



(2)
18.1



(3)
7.7



(4)
28.2



(5)
14.7



(6)
23.1



(7)
23.6



(8)
5.4


Figure 4.19 Temporal Distribution of REM Activity for
Group 2.


URLUE


S : : : 4 : : :
S : : 4. : : :

: : : 4 : : :


" 4. : : :

:I----------*----------I+


: : : 4 : :
: : : : : :

: : :


: : :--------------------:


: : : : 4 : : :

I: -- -- -*- -------------I
: : 4. : :

. .


: I- -------------- ------------ -- -------
S a :. : : :



S : ----------- -------------

: : : 4+ : :
: : : : 4 : : :

I-- ------I :








Beta and REM increase markedly across the night as

time asleep increases. The REM data compare quite well

with the results of previous investigations by Webb and

Verdone (Colquhoun, 1972), as shown in Figure 4.20.

There is a sharp decline in the amount of delta as

sleep time increases. This decline in delta activity has

been demonstrated by many investigators, although the

results are most often given in terms of amounts of stage

3 and stage 4 sleep.

The sigma data show a slight decrease over the first

6 hours and a sharp decrease for the 7th hour. The

trend in this temporal data for sigma is much clearer than

for the data collected according to period number.


Correlations, Autocorrelations, Indicated Periods

Program BCOR1 and subroutine BCOR measure the corre-

lation between any two given binary ultradian patterns.

If the programmer so indicates, the program will measure

autocorrelation instead of correlation between two

different patterns. Results are given for values

of lag from zero to one-half the record length of the

shortest record. The correlation (or autocorrelation)

functions are then averaged over each group of 10 records.

Table 4.7 shows the implied period of the ultradian

pattern determined by noting the lag time for which the

group autocorrelation function peaks up, provided that it

shows a periodic component. The periods indicated are












32


28 /


24


20 A /
0

3 16
-4

S 12 Webb
I--

8 ----- Verdone

,'/ Present Study
4 ,/"


S I I
1 2 3 4 5 6 7

Hours of Sleep

Figure 4.20 Hourly Distribution of REM.


























Table 4.7 Ultradian Cycle Time Indicated by
Group 2 Autocorrelation Function.


BETA DELTA SIGMA REM

Cycle Time, T (mins.) 107 101 113 106

Autocorrelation, R(T) .37 .31 .36 .36








slightly higher than those determined by averaging. This

is probably due to the fact that the larger periods will

dominate when using this method. That is, in calculating

the autocorrelation, as the series is shifted, the peaks

will occur where the longer active periods line up,

creating a tendency to favor these larger values and

peaking at slightly higher lag times. The method also

cannot detect "missed" active periods and therefore will

be somewhat sensitive to disturbances of this nature

(Lubin et al., 1973).

The autocorrelation of the alpha patterns showed no

periodicity and was therefore omitted from the table.

All other activities showed quite high autocorrelations

(.31 to .37) for lags of slightly greater than 100

minutes, indicating a strong ultradian component for the

group as a whole.

The correlations were lower and in general not as

clearly periodic for other groups, indicating that this

method is not as useful for the less stable rhythms.


Interrelationships Between Different Activities

A number of interesting correlations were observed

for Group 2. Using BCOR and BCOR1, it is possible to

determine group correlations of any pair of ultradian

patterns. Table 4.8 is a summary of a number of calculated

correlations. Phase relationships are indicated by the

lag times for which the maximum correlation occurred.


























Table 4.8









Beta-REM

Beta-Delta

Beta-Sigma

Delta-Sigma


Maximum and Minimum Correlations and
Periodicities of Group 2 Correlation
Functions.

Maximum Minimum
Correlation Correlation
and and
Associated Associated Indicated
Lag Time Lag Time Period

.76, 13 mins. -.316, 64 mins. 101 mins.

.148, 50 -.427, 3 98

.304, 52 -.75, 0 110

.268, 0 -.086, 42 90









The indicated ultradian period is also given--it has

been calculated by measuring the distance between the

first and second peaks in the correlation function when

a periodic component is clearly present.

A very high correlation of beta and REM (+.76)

occurs at a lag of 12 minutes, indicating a stable

phase relationship between beta and REM where the REM

lags behind the beta by approximately 12 minutes. There

is a strong periodic component of 93 minutes in the

correlation function. This relationship can be seen

very clearly in the Group 2 beta-REM ultradian patterns

shown in Figure 4.21.

Beta and delta are highly negatively correlated

for small lags, that is beta and delta are very emphat-

ically negatively correlated across the night. Notice

this relationship is apparent in Figure 4.22 in the

binary ultradian patterns. There was a very clear

periodicity here of about 98 minutes.

There is yet a stronger negative correlation (-.75)

between beta and sigma, the periodicity in this case

appears to be approximately 110 minutes. The beta and

sigma ultradian patterns are given in Figure 4.23.

Delta and sigma are positively correlated for zero

lag. Referring to the binary ultradian patterns for delta

and sigma (Figure 4.24), the overlapping of the delta

and sigma patterns is obvious. It seems that, were delta

to appear in the second half of the night, it would continue

























1 4





























REM

1 2 3 4' 5 6 7 Hoar


Figure 4.21 Binary Ultradian Patterns for
Beta and REM.













i, .. j n 1l L1- l-

I f 1 1 = + I I I





F J 1 0 1- 1F4



S1I I 1 1 t- I-

Beta
I Delta
1 2 3 4 5 6 7 Hours
Figure 4.22 Binary Ultradian Patterns for
Beta and Delta.


---flF-


nl- i























..t .. ...t h
















1 1 I AI 11 I, i l l



BETA
SIGMA

1 2 3 4 5 6 7 Hours


Figure 4.23 Binary Ultradian Patterns for
Beta and Sigma.


















I1 :1 i | I 1 1









...., ,F n .,
I- 7 .. .11


SIGMA
DELTA
1 2 3 4 5 6 .7 Hours
Figure 4.24 Binary Ultradian Patterns for
Delta and Sigma.








to exhibit the same phase relationship relative to

sigma. The two processes seem therefore to be phase-

locked, under the influence of the same timing cues,

but delta is somehow suppressed in the last two-thirds

of the night.


Summary

In the majority of cases the Group 2 patterns were

the most well organized, most stable, and most easily

defined. Younger age groups were generally in an

interim stage of development, and older groups showed

less stability in most cases. The following chapter will

describe many of the parameters dealt with in this chapter

and how these parameters vary with age.















CHAPTER FIVE
ONTOGENETIC TRENDS


Introduction

This chapter covers essentially the same topics as

described in Chapter 4, but in this case the emphasis

will be placed on the variations or trends across age

groups. Less emphasis will be placed on the actual data

collection procedures, since, for the most part, these

aspects were adequately explained in Chapters 3 and 4.


Qualitative Observations

Five-minute running averages for all age groups are

given in Appendix 3. A number of qualitative character-

istics can be seen in the running averages. These char-

acteristics will be measured quantitatively in the following

sections, but first, here are some general observations.

The young adult group (Group 2) as a whole shows

greater stability. Alpha seems to be the exception here.

As mentioned in Chapter 4, it shows no ultradian rhythm

in the young adults, but it does seem to show a fairly

well-organized ultradian rhythm in a number of the Group 4

subjects.

Beta activity is the best organized overall. Group 1

and Group 3 are both slightly less organized than Group 2.

Group 4 is again less organized than Group 1 or Group 3,








but better organized than Group 0 which still shows

ultradian rhythmicity.

Only in Group 2 does sigma show a clear ultradian

rhythm. The subjects in Group 0 have very little sigma

activity, rendering the 5-minute averages largely meaning-

less. Some records in Group 1 show an ultradian pattern,

but as a whole this group is considerably less organized

than Group 2. Only about 2 records in Group 3 show a

clear ultradian rhythm, but at least 4 records in Group 4

show a good ultradian rhythm.

As previous studies have also shown, delta progres-

sively decreases with age (Feinberg et al., 1967; Colquhoun,

1972) Group 0 shows constant delta throughout the night,

making it difficult to detect any ultradian rhythm. There

is a tendency to show less delta in the second half of

the night (Lubin et al., 1973). Group 1 subjects have a

good ultradian rhythm and frequently have pronounced

active regions or ultradian periods in the second half

of the night. Group 2 shows a well organized rhythm,

but shows fewer active regions in the last two-thirds

of the night, which is consistent with previous inves-

tigations (Lubin et al., 1973; Feinberg et al., 1967).

Degeneration of the ultradian rhythm continues for Group 3

and Group 4 as delta activity continues to fall off.


Distribution Over the Night

Figures 5.1 through 5.4 show moments about the midpoint





87





IOrlfrL I'rD HDrl.t.IITS CIBOUT 1THi MIDPOIHT
FOR tLPHrtlF---GRUPS 8, 1, 2, 3, frill 4


Ufi. tIE
-1208.600
+----..-.....-.--



















-235.6









-38. 6









-4.
^9. t*


< = .- + -->
*(.0c.00 C. 0. C0. % U% .200. 0
_..---4-----+-- --+- 4-----+-----+---_4-----+


Figure 5.1 Normalized Moments About the
Midpoint for Alpha.





















Table 5.1 Significant Group Differences for
Alpha Moment About the Midpoint.


GROUP 1 GROUP 2 GROUP 3 GROUP 4

GROUP 0 *

GROUP 1 .01 NS .05

GROUP 2 NS NS

GROUP 3 NS


*Insufficient data available.











NOrnnLZLlZ MHOll[HTS riBOUT THE MIDPOIHT
FOR F[:E1R--(iGROUlIF; 0, 12, 2, 3, AND 4


-1200.00 -680.600


==- + ==>
.00 c


6C.e. DO


VUALU:











1356.0








9-6.8













20. c




265. 0








-8 (.01


Figure 5.2 Normalized Moments About the
Midpoint for Beta.


1 (?,. 06


4- -...-.-.---..-.-.-..-----+----. _+---- + --- 4_---_4- . 4----- 4--
4 +
4 +
: +

: +
; +
+

+
4-




+
4-


: + C

4- C
S +
S +
S +


+


S+ .
+


C +
+

: + .

+
:: : : / +
+ C C


4


4 C

.C
S C 4





90











Table 5.2 Significant Group Differences for
Beta Moment About the Midpoint.


GROUP 1 GROUP 2 GROUP 3 GROUP 4

GROUP 0 .05 .01 .01 .01

GROUP 1 .05 .01 .01

GROUP 2 .01 .01

GROUP 3 .05





91




NORRnnl.IZrD HMOiNTt FaIOUT 1iHE MIDPOIN7
FOl' DELLf--GROUP'S 0, 1, 2, 3. RID) 4


fILLUE


-1200.00 -600. (0


(=- 4 tr>
0. ('0


600.00


-1006.0 o










-147t8.0















-I4e. C.





- 142.06










*.....;C.... 1


Figure 5.3 Normalized Moments About the
Midpoint for Delta.


I200t. CI


+
+
4-

4.
+
+
+
4
+

+
+
+

+
4

+
+
+
+

+
+

4
+
+
+
+
4





+
+
+


+I:
+
+
+
+
4





+

+

4
4
4
4
+
4


4-


+-----t-----t----- +-----+-----t-----t -----1----~-----~ -----+






92









Table 5.3 Significant Group Differences for
Delta Moment About the Midpoint.


GROUP 1 GROUP 2 GROUP 3 GROUP 4

GROUP 0 .05 NS .01 .01

GROUP 1 .01 .01 .01

GROUP 2 .01 .01

GROUP 3 .05




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs