Biofeedback and Self-Regulation, VoL 4, No. 2, 1979
Feedback Delay and Amplitude Threshold and Control of the Occipital EEG l Thomas Mulholland, Robert Boudrot, and Anne Davidson
Psychophysiology Laboratory, VeteransAdministration Hospital, Bedford, Massachusetts
Six normal adults looked at 12 color transparencies that were presented 30 times each in response to the occurrence o f alpha in the occipital EEG and that were not presented when no-alpha occurred. The time delay between the occurrence of recorded EEG alpha and the feedback stimulus was systematically varied, as was the threshold value o f alpha amplitude required for feedback stimulus to occur. As time delay increased, the replicative reliability o f the temporal association between alpha and the visual stimulus decreased. The measures o f the EEG resnonse were: the mean durations o f alpha and of no-alpha durations (,~); an estimate of random "'scatter" or variability in a single trial (Se); and the ratio o f these (X/Se), which is the reciprocal of the coefficient o f variation. The latter ratio defines "'control" of the EEG response. Control was greatest for the minimum feedback delay (.25 sec) and a feedback threshoM o f 25°7o o f the maximum peak alpha amplitude recorded under a prior eyes-closed, "'resting'" condition. Control o f the EEG decreased as time delays increased and at lower and higher feedback thresholds. It was proposed that the measurement o f control (X/Se) for various combinations of values o f relevant experimental parameters couM be used to select that combination which gave the best control This "'optimum" combination o f values o f parameters couM be chosen for a standardized method. Previous studies have shown that the control of time durations of alpha and o f intervals o f little or no-alpha can be increased by a feedback f r o m the E E G to a visual stimulus (Mulholland & Eberlin, 1977). Control is measured as the ratio of the mean (X') duration o f alpha or no-alpha to the best
'This research was supported in its entirety by the Veterans Administration Medical Research Program. 93 0363-3586/79/0600-0093503.00/0 © 3.979 Plenum Publishing Corporation
94
Mulholland, Boudrot, and Davidson
estimate of sample random variance (s). In statistical terms this ratio is the reciprocal of the coefficient of variation. The duration and variability of the durations of alpha and no-alpha depend on many factors acting simultaneously, e.g., (1) the location of the EEG recording electrode (Mulholland, 1977b), (2) the reliability of the temporal association between the occurrence of alpha and the occurrence of the stimulus (Eberlin & Mulholland, 1976; Mulholland & Eberlin, 1977), (3) the amplitude of alpha required for the stimulus to occur (Boudrot, 1972), and (4) the time delay between the EEG and the stimulus (Mulholland, 1973), which indirectly affects (2). In the experiment reported here, alpha amplitude threshold and time delay were studied with regard to the control ratio (X/s). The aim of the study was to identify that combination of threshold and time delay which was associated with the largest value of the control ratio.
METHOD
Subjects Six normal adult volunteers (5 men, 1 woman) ages 22-50 were tested.
Apparatus Bipolar EEGs were recorded from silver/silver chloride electrodes located at O1-P3, O2-P,, and O1-O2 in the standard clinical EEG nomenclature, using a Grass Model 7 polygraph. Feedback was from O1-O2. The o.utput of the polygraph driver amplifiers went to filters, threshold-setting devices, and time-delay circuits (Boudrot, 1972). Alpha detection occurred when the recorded EEG signal was between 8 and 13 cps and greater than 5-8/aV. Detection was the closure of a relay. Minimum closure time was .05 sec with a delay of .25 sec. Maximum peak amplitude of the filter output recorded in the dark with eyes closed was considered to be "maximum alpha amplitude." A settable threshold permitted the triggering of the visual stimulus when alpha amplitude exceeded any preset percent of maximum amplitude. In this experiment, three feedback thresholds were tested: 10°70, 25°70, and 40070 of maximum amplitude. The time delay between alpha at or above criterion in the EEG recording and the onset of stimulation was controlled by a computer PDP-12.
Feedback Time Delay
95
Four time delays were tested: .25 sec (the minimum delay possible with this apparatus), .45 sec, .65 sec, and .85 sec. When the detected alpha met threshold and time-delay criteria, the stimulus was presented. The time-delay variable was constrained so that the stimulus was delayed only if the alpha detector was on for a time duration equal to the delay. If the alpha detector went off before the delay had elapsed, the system reset and the stimulus was inhibited until the next alpha event met the criteria for feedback. When the EEG met the criteria for alpha feedback, the visual stimulus was turned ON until the EEG was suppressed below the criteria for alpha feedback; then it was turned OFF.
Procedure The subject was seated in a comfortable chair in a dark soundproof room and the EEG electrodes were connected to the polygraph. A brief resting record was obtained to determine the dominant alpha frequency and maximum peak amplitude, and to adjust and calibrate the polygraph. Thirty events (an alpha " b u r s t " and its accompanying no-alpha interval) were collected by the PDP-12 computer with eyes closed and with eyes open in the da/'k. To compensate for habituation and fatigue effects, each feedback amplitude-threshold combination was systematically varied in a modified Latin square. Within each threshold, the four time delays were assigned in a fixed sequence so that each subject received each combination of threshold and time delay only once. Each of these 12 conditions was tested with a different colored slide projected on a screen 6 feet in front of the subject. The slides were in the same sequence for each S; the systematic variation of the time-delay and threshold conditions was superimposed on the fixed sequence of slides. Slides were not matched for brightness or for interest. A "habituation" condition was run prior to the start of the visual stimulation trials in which a feedback-controlled visual stimulus was presented 30 times. This was done to minimize the orienting response to the remaining stimuli. Slides were photographs of people in typical settings, natural scenes, and familiar objects.
Data Collection and Analysis The data reported here from each experimental trial were derived from the 0,-02 recording and consisted of 30 alpha durations and 3 0 no-alpha durations as defined by the alpha state relay ON and OFF. These were measured to the nearest .02 sec and collected using a PDP-12 computer (Goodman, 1973). For each trial the 30 durations of alpha were fitted with a
96
Mulholland, Boudrot, and Davidson
best-fit function that described the within-trial "habituation." For alphas a straight line was fitted in each trial. (1)
At a = F~(N) = ( A N + B) +. S e
where N is the serial number of the alpha duration in the trial. Similarly, no-alphas were fitted with an hyperbola.
(2)
A t n a = F~(N) = ( C / N + 19) +_S e
These curve-fitting procedures gave the mean duration and the "scatter" or variability of the raw data points relative to the best-fitted functions, the standard error o f estimate (Se). A control ratio X / S e was computed in each trial. The ratio of number of stimulations to the total number of alphas detected ( N s / N t ) was also computed in each trial. The statistics X, S e, X / S e, and N s / N t were analyzed with ANOVA using a repeated-measures-on-independent-subjects design. Variance was partitioned among time delays, thresholds, subjects, and their interactions. F ratios were computed using the corresponding mean square error. The criterion for significance in all tests was p ~< .05.
RESULTS The major effect of feedback stimulation is an alternation between EEG alpha events and EEG events that do not meet the criteria for alpha. For selected portions of a subject's EEG, the regularity of the alternation between alpha and no-alpha can be appreciated with ordinary on-line observation of the recording (Mulholland, 1973). The durations of the alpha detector ON and OFF, which are measures of alpha and no-alpha, can be compressed into a graphical display (see Figure 1). The display shows the time duration (At) of alpha and no-alpha for each event numbered serially from first to last in the trial (Boudrot, Goodman, & Mulholland, 1978). In Figures 1 and 2, the computer's graphical display of alpha durations for a single trial is shown for the four time delays. Alpha threshold for feedback was 25% of maximum peak amplitude. As the stimulus ON delay relative to the onset of alpha is increased, both the average duration of alpha and the variability of the average alpha duration increase. These qualitative results were verified by a quantitative analysis of the time durations of alpha and of no-alpha events. As described before, in each trial a best-fit function was computed for alpha durations and for no-alpha durations.
97
Feedback Time Delay
¢D ~'-,
/J-
D
~J
,j la
-
bl
Z
~
'
•~
0,~
..-
i~
0 z
= ~ ~
--/ .,4
.I
o
d
,-',04
.o ~
ea~
.~ ~
_
0 ¢'1
::2
[-
-!
• ~
~al
:-
...............................
..;
.4
,_-...
c;
-
d
~.~ ~.~,..~~
........
,,-~
0
98
Mulholland, Boudrot, and Davidson
G
g
~0..1
oj ,~Irr
z
Ul
o Q:
o
o~ ul O.
Q~ Z
-
¢,1 II •
if
0
(o
Q
•
.,I
• ~"
II
j i,r
I"
.
Z
g
g
-
n .r" Vl "z" I"
,4
.~ -!
D ~n"
0
:
Nrr
-
Q.
z
.,I
•
o
.
-r
-
,..:
U
-,
:
~,~
o
~,
0
~-
z
Feedback Time Delay
99
The threshold o f a l p h a a m p l i t u d e a n d the feedback t i m e delay o f the stimulus O N in r e l a t i o n to the recorded E E G a l p h a h a d a significant effect o n the a l p h a d u r a t i o n ()(), the variability o f the a l p h a d u r a t i o n (Se), the c o n t r o l of a l p h a d u r a t i o n s X / S o a n d the ratio o f n u m b e r o f s t i m u l a t i o n s to n u m b e r o f alphas detected (Ns/Nt). There were lesser effects o n the n o alpha d u r a t i o n s . See T a b l e I.
Amplitude Threshold The m e a n a l p h a d u r a t i o n increased with a n increased feedback a m p l i tude threshold (F2,1o = 19.1;MSE = . 16). The m e a n d u r a t i o n o f n o - a l p h a decreased b u t the effect was n o t significant. The s t a n d a r d error o f estimate (Se) for a l p h a d u r a t i o n s increased with increased threshold (F2,10 = 9.03; M S E = .028). N o significant differences o f Se were f o u n d for n o - a l p h a . C o n t r o l (X/Se) o f a l p h a d u r a t i o n s was least for the highest threshold b u t the result was not statistically significant. C o n t r o l o f n o - a l p h a d u r a t i o n s was least for the highest threshold (F2,10 = 7.64; M S E = .17). T h e ratio (Ns/Nt) decreased as t h r e s h o l d increased (F2,,0 = 22.3; M S E = .062). See T a b l e I. Table I. Control of Alpha and No-Alpha Durations for Time-Delayed Feedback Stimulation of EEG Occipital Alpha f
Feedback amplitude threshold Time delay (secs)
X/Se
.25
2.23 1.84 1.57 1.36 1.75
1.54 1.48 1.44 1.17 1.41
.25 .45 .65 .85
.43 .42 .45 .52 .45
.50 .56 .52 .56 .53
.53 .61 .74 .80 .67
.25 .45 .65 .85
.37 .29 .30 .33 .32 .99 .99 .76 .73 .87
.23 .30 .39 .39 .33 .84 .71 .42 .49 .63
.36 .43 .51 .70 .50 .53 .42 .37 .24 .39
Average
Average
Average
Ns/Nt
Average
10% 25% 40% 1.80 1.94 1.54 1.65 1.73
.45 .65 .85
Se
Alpha
.25 .45 .65 .85
No-alpha Average 1.86 1.75 1.52 1.39 1.63
10% 25% 40%
Average
1.46 1.36 1.65 1.49 1.49
1.78 1.15 1.50 1.13 1.10 .97 1.31 .96 1.42 1.05
1.46 1.33 1.24 1.25
.49 .53 .57 .63 .55 "
2.10 2.12 1.78 1.63 1.91
1.62 1.34 1.58 1.24 1.45
1.25 1.37 1.62 1.28 1.38
1.66 1.61 1.66 1.39
.32 .34 .40 .47 .38 .78 .70 .53 .62
1.63 1.59 1.14 1.35 1.41
1.16 1.06 1.49 1.07 1.20
1.70 1.46 2.15 1.50 1.70
1.49 1.37 1.60 1.31
100
Mulholland, Boudrot, and Davidson
Time Delay Mean alpha duration increased with increased time delay (F3,1~ = 5.69; MSE = .012). Mean no-alpha duration was least for the longest delay but this result was not statistically significant. Standard error of estimate for alpha durations increased with increased time delay (F3,,, = 4.2; MSE = .020). No significant differences among time delays with respect to S e was found for no-alpha S e. Control (X/Se) of alpha durations decreased with increased time delay (F3,1~ = 5.91; MSE = .136). No significant differences in control were found for no-alpha, although the greatest control was associated with the briefest time delay. Ns/Nt decreased as time delay increased (F3,,5 = 15.4; MSE = .02).
Interaction of Time Delay and Threshold For mean alpha durations, there was an interaction between threshold and time delay: Differences among time delays were greatest for the 4007o alpha threshold compared to the other threshold conditions (F6,30 = 2.74; MSE = .023). See Table I. The results are summarized in Figure 3.
DISCUSSION The two feedback parameters tested in this experiment, time delay between alpha detector ON and stimulus ON and the alpha amplitude required for presentation of the stimulus, are relevant to the control of alpha durations during feedback stimulation. Control, as measured by the mean duration divided by the best estimate of the random sample variability (X/S e) decreased as feedback time delay increased; control was more for the 25°70 alpha amplitude threshold compared to lower or higher thresholds. Increased time delay between alpha and the stimulus was necessarily associated with a decreased reliability of the temporal association between alpha at criterion and the occurrence of the stimulus. These results are consistent with an analogy between this kind of alpha feedback experiment and a simple control system proposed previously (Mulholland, 1977a,b). The increase of alpha with increased stimulus delay is analogous to the increased period of response of an O N - O F F control system with increased feedback delay (Rose, 1967). The fact that a particular threshold for feedback was better than the others with respect to control of
Feedback Time Delay
101
.80.60 Ns/Nr. 40".2O
.60 .4o.20 (/) <[ -tO. .J
,¢
.6O -
S, .40 .20 -
2.4
X/S o 1 . 6 -
.8"
o
.z's
,4'~
.6'~
,8~
TIME DELAY (sees) Fig. 3 . ChanJ~es in the statistics Ns/Nt, X, S e, a n d / Y / S e in r e l a t i o n t o the f o u r v a l u e s o f t i m e d e l a y .
the EEG is analogous to the relation between feedback gain and control seen with simple control systems. These results also suggest a procedure that, if generalized, could be applied to test the relevance of other experimental parameters to the control of the EEG by visual feedback stimulation. Moreover, the cofnbination of values of the relevant parameters that yielded the largest control ratio could be selected as the standard for the experimental study of the EEG component of the human orienting response to visual stimuli. Thus standardization would be based on criteria that were rational and less arbitrary. As it turned out in this experiment, the largest control ratio was obtained with a minimum .25-sec time delay and a 25°70 threshold for alpha feedback for alpha durations (2.23) and for no-alpha durations (1.78).
(X/Se)
102
Mulholland, Boudrot, and Davidson REFERENCES
Boudrot, R. An alpha detection and feedback control system. Psychophysiology, 1972, 9, 461-466. Boudrot, R., Goodman, D., & Mulholland, T. An EEG alpha detection feedback stimulation and data analyses system. Behavior Research Methods and Instrumentation, 1978, 10, 646-651. Eberlin, P., & Mulholland, T. Bilateral differences in parietal-occipital EEG induced by contingent visual feedback. Psychophysiology, 1976, 13, 212-218. Goodman, D. ALFIE: Collection of EEG alpha under feedback control using time series analysis. Psychophysiology, 1973, 10, 432-440. Mulholland, T. Objective EEG methods for studying covert shifts of visual attention. In F. J. McGuigan & J. Schoonover (Eds.), Thepsychophysiology of thinking. New York: Academic Press, 1973. Pp. 109-151. Mulholland, T. Feedback as scientific method. In G. Schwartz & J. Beatty (Eds.),Biofeedback: Theory and research. New York: Academic Press, 1977. Pp. 9-28. (a) Mulholland, T. Biofeedback method for locating the most controlled responses of EEG alpha to visual stimulation. In J. Beatty & H. Legewie (Eds.), Biofeedback and behavior. New York: PlenumPress, 1977. Pp. 95-106. (b) Mulholland, T., & Eberlin, P. The effect of feedback contingency on the control of occipital alpha. Biofeedback and Self-Regulation, 1977, 2, 43-57. Rose, J. Automation: Its anatomy and physiology. Edinburg: Oliver and Boyd, 1968. P. 125. (Revision received August 25, 1978)