psychophysical and neurophysiological considerations …mark-white.com/literature/italy84.pdf ·...

PSYCHOPHYSICAL AND NEUROPHYSIOLOGICALCONSIDERATIONS IN THE DESIGNOF A COCHLEAR PROSTHESIS

Audiol. Ital. 1, 77-117, 1984

PSYCHOPHYSICAL AND NEUROPHYSIOLOGICALCONSIDERATIONS IN THE DESIGNOF A COCHLEAR PROSTHESIS

M. W. WHITE

Psychophysical and' neuropsychophisical considerations in the design ofof a cochlear prosthesis.

The Authors carry out an analysis of threshold, loudness and intensity dis-crimination functions for a wide range of stimuli. Phenomenological models ofthese functions are developed and neurophysiological concepts, such as temporalintegration processes, are used to interpret the behavioural data presented. Com-parative data are presented for single and multichannel prostheses in order to pro-vide the designer with a more effective understanding of the consequences ofelectrical stimulation of the auditory nerve.

KEy WORDS: Behavioral threshold, cochlear prosthesis, stimulation of auditorynerve.

Introduction

Direct electrical stimulation of surviv-ing auditory nerve fibers or ganglion cellsis now being employed to restore someaspects of hearing in profoundly deaf pa-tients (Simmons, 1964; Hochmair, 1980;Fourcin et al., 1979; Tong, 1979; Merze-nich, 1973). Psychophysical studies of thesensations of such patients has indicatedthat presumably localized activation ofgroups of fibers from a given area of thebasilar membrane gives rise to «auditory»sensations. In general, the magnitude ofthese sensations increases with stimulus am-plitude. In this study we examine in detailthreshold, loudness, and intensity discrimi-nation functions for a wide range of stimuli.

School of Medicine, Department of Otolaryn-gology University of California, San Francisco, Ca-lifornia 94243, U.S.A.

Lavoro presentato al Premio Internazionale diAudiologia «Stelio Crifo ».

Phenomenological models of these functionsare developed and neurophysiological con-cepts are used to interpret the behavioraldata presented. The purpose of this study isto develop a set of conceptual « tools » thatwill allow the prosthesis designer to moreeffectively understand the consequences ofelectrical stimulation of the auditory nerve.It is hoped that the use of this knowledgewill help us design more effective auditoryprostheses.

This study draws a great deal from thework of others in the fields of neurophysio-logy and psychophysics. Many concepts de-veloped by these investigators will be ap-plied to the interpretation of the data to bepresented here. Studies of cochlear nerveresponses to electrical stimuli have beenreported by Kiang and Moxon (1972). Theseauthors compared eighth nerve firing pat-terns to a limited set of both electrical andacoustic stimuli. Firing rates as a functionof stimulus intensity and threshold as a

78

. function of frequency were defined. Phase-locking of responses evoked by a 200 Hzsinusoidal electrical stimulus was also stu-died by Kiang and Moxon (1972). Cloptonet at. (1980) have studied threshold anddynamic ranges for different stimulatingelectrode configurations and waveforms.Frankenhaeuser and Huxley (1964), McNeal(1976), and Teicher and McNeal (1978)have developed general models describingthe electrical excitation of myelinated nervefibers such as those which comprise the pre-dominant afferent population in the auditorynerve. See Loeb et al. (1983) for a discus-sion of the biophysics of intracochlear elec-trical stimulation of these neurons.

The primary goal of this research is tobetter understand how we can utilize elec-trical stimulation of the auditory nerve tohelp the totally deaf to communicate. Inthis study, the communication (via the co-chlear implant) of information related tothe intensity of the stimulus is examined.This work is intended to aid the design ofboth multi-channel and single channel co-chlear prosthesis. In a single or multi-chan-nel cochlear prosthesis we must at leastminimally understand how to generate sti-muli that are audible, not too loud, andthat allow the subject to discriminate differ-ences in stimulus intensity. This study wasdesigned to enable the developers of suchprosthetic systems to understand many ofthe basic principles necessary to generateelectrical stimuli that meet these require-ments. This study is particularly useful inthat data for a large range of stimuli arepresented.

GOALS OF COCHLEAR PROSTHESISRESEARCH

A sound processor should minimally becapable of « adjusting» the amplitude of the

M. W. White

stimulus such that the stimulus remainswithin the dynamic range of the subjectand yet does not so restrict amplitude ex-cursions that little or no intensity informa-tion is conveyed. Although this should beconsidered a minimal requirement, and notat all a «sufficient» requirement, it is notreadily achieved. Indeed, many current pro-cessors do not meet this minimal require-ment. The exact placement of the stimuluswithin the dynamic range is a further refi-nement of a processor's design specification.Both intensity descrimination functions andloudness functions have been suggested indetermining the placement of the stimuluswithin the dynamic range.

We designate the highest stimulus levelthat is still comfortable to the subject as the«maximum comfortable loudness level» orsimply «MCL ». Threshold and MCL arevery important functions, for they definethe widest possible limits of audible and to-lerable stimulation. We define the ratio ofthese two limits as the subject's «dynamicrange ». The ratio is most often expressedin dB.

Threshold, maximum comfortable loud-ness (MCL), and dynamic range functionshave been measured for biphasic pulse sti-muli. For comparison, some sinusoidal sti-muli were also investigated. Pulsatile sti-muli were primarily used in these studies.

Pulse width can be varied independentlyof the pulse frequency. Pulse width is defi-ned as the total duration of the biphasic pul-se. With sinusoidal stimulation the durationof a sinusoidal cycle varies inversely withthe frequency of stimulation. In contrast,with pulsatile stimulation, these two va-riables can be varied independently of eachother. Furthermore, pulse width and pulsefrequency (i.e. pulse rate) have distinctlydifferent effects when considered in terms

Design of a cochlear prosthesis

of the most common types of neural exci-tation models.

MODELS OF NEURAL EXCITATION

In order to describe and interpret thedata to be presented, it is useful to reviewseveral models of threshold to electrical sti-mulation that may be useful in the interpre-tation of the data. One of the simplest be-havioral threshold models incorporates a mo-del of neural membrane excitation. In usingneural membrane models to approximatebehavioral threshold functions, a number ofassumptions are often invoked. It is as-sumed that the medium between the stimu-lating electrodes and the excitable membra-ne acts only as a purely resistive attenuator(Spellman, 1982). It is also assumed thatthe stimulator is an ideal current source(Vurek et at.) 1981). If this later assump-tion is approximately valid, there will belittle effect on neural threshold due to non-linear and frequency dependent electrodeimpedance. Other assumptions are com-monly made about the relationship betweenbehavioral threshold and the excitation ofnerve fibers. For example, a working as-sumption might be that behavioral thre-

79

shold occurs when «N» neural dischargeshave been elicited within a given time win-dow, from one or more fibers. For example,if there is no spontaneous activity and thethreshold detector is ideal, only one actionpotential would be necessary to elicit be-havioral threshold.

Hill's model (Hill, 1936) is schematic-ally depicted in figure 1. The stimulatoris represented by the current source. Thepassive membrane capacitance and resistanceare represented by the capacitor « Cm » andthe resistor «Rm» in parallel. Accomoda-tive effects are approximated by the capaci-tor «Ca» and the resistor « Ra» in series.The voltage across «Ra» is monitored bythe «threshold voltage detector ». Whenthe voltage across this resistor goes abovea present threshold voltage «Vt », the de-tector generates a simulated action poten-tial. This model mimicks certain generalfeatures of neural excitation. For example,the model predicts that a monophasic pul-se will elicit an action potential if the char-ge within the pulse is above some fixedthreshold value and if the pulse width is suf-ficiently within the time constant of themembrane (tm=CmRm). Likewise, at thehigher sinusoidal stimulus frequencies, this

extracellular

Stimulating tCurrent I(t)

intracellular

VoltageFollower

ThresholdVoltageDetector

Fig. 1. - Hill's lumped circuit neural excitation model.

80 M. W. White

"CD

",.5~"'J

~....z......1-"o:r(l.

.....zC0:.CD":eN'"c

STIMULUS

I_I I_I•.'~.---~~:'---B~:-'--l~~~.-'--ll;rl.-'--2" .•

TII1£: C111LLlSECONDSJ

Fig. 2.

511NULUS

.~~,--''1~~.--.~:-.--l~'-.'--I~&.-'--Zt.•TI1"£ CI1ILUSECONOS)

Fig. 3.

Fig. 2. - Simulated membrane potential as a function of time for a square wave stimulus ini-tiated at t=O. The stimulus waveform is displayed on the same time scale as that for themembrane potential function. A modified Hodgkin-Huxley model was used in this simulation.Beta-h was increased by a factor of four over that in the H-H model.

Fig. 3. - Hodgkin-Huxley variable «h» as a function of time for a square wave stimulus ini-tiated at t=O. The stimulus waveform is displayed on the same time scale as that for theHodgkin-Huxley state variable «h ». A modified Hodgkin-Huxley model was used in this si-mulation. Beta-h was increased by a factor of four over that in the H-H model.

model predicts that threshold will increaseat a rate of 6 dB! oct as the sinusoidal sti-mulus frequency is increased. The modelalso predicts another commonly observedphenomenon: that threshold is never reach-ed if a stimulus ramp rises at too slow arate. In the model, this is due to «Ra»and «Ca », which represent accomodativephenomena. For sinusoidal stimulation, ac-

comodation is evident at the lower fre-quencies (below 50-100 Hz), where thethreshold no longer decreases as the fre-quency is decreased.

Hill's model is a useful starting point fordescribing and understanding certain aspectsof neural excitation. However, other as-pects of neural excitation are not well re-presented by Hill's model. For example,


Hill's model does not simulate any post-excitatory phenomena. Hill's model is pri-marily a linear model. It is non-linear onlyin its use of the «voltage threshold detec-tor ». More complicated, non-linear modelssuch as the Hodgkin-Huxley model (H-Hmodel) and the Frankenhauser-Huxley mo-del (F-H model) more accurately representneural excitatory behavior (Frankenhauserand Huxley, 1964). These models simulatemany more features of neural responses thandoes Hill's model. Two of these featuresare discussed below: (1 ) sensi tiza tion dueto prior hyperpolarization of the membrane;and (2) temporal integration of sub-thresh-old, charge-balanced stimuli.

SENSITIZATION DUE TO PRIORHYPERPOLARIZATION OF THE MEMBRANE

Figure 2 illustrates one feature of neu-ral excitation. If one phase of the stimu-lus hyperpolarizes the neural membrane,the membrane may become more sensitiveto the stimulus subsequent to this hyper-polarization. A modified Hodgkin-Huxleymodel (1952) was used for this simulation.The first depolarizing phase does not ex-cite the model nerve. However, after thehyperpolarizing phase of the stimulus. Inthis example, both the hyperpolarizing phaseand the subsequent depolarizing phase werenecessary to excite the model nerve at thisstimulus level. Figure 3 displays how theHodgkin-Huxley variable «h» (i.e. the so-dium deactivation variable) varies during thestimulation. As «h» increases, the excita-bility of the membrane increases. The valueof «h» increases to a higher value duringand shortly after the hyperpolarizing phase,than its maximum value during the firstphase of depolarization. This simulation illu-strates how a nerve membrane become more

81

sensitive after a hyperpolarizing stimulus. Theduration of the hyperpolarization phase iscritical in determining how effective it is inincreasing the nerve's sensitivity. General-ly, if the hyperpolarizing phase is too shortor too long the nerve's excitability will notbe increased. The graph in figure 4 wasconstructed using the same nerve mem-brane model used in the previous example.The model is identical to the H-H model;except that «beta h» has been increasedby a factor of four. This has the effect ofincreasing the rate of response of the de-activation variable «h ». The graph in figu-re 4 was constructed by determining the re-lative current level at which a simulatedaction potential was generated for a set offrequencies. Figure 5 indicates how manystimulus cycles were necessary to excitethe model membrane at threshold currentlevels. Over the limited frequency rangeillustrated, this model approximates someof the behavioral threshold features obser-ved. The major effect on the nerve's «fre-quency response» is the increased slope (i.e.greater than the Hill model's 6 dB/octave)over the 90 to 200 Hz frequency region.

A note of caution should be insertedhere. Such highly nonlinear simulation mo-dels can be used to generate a relativelywide range of behavior by adjusting the mo·del's parameters. Therefore, a very good« curve-fit» could be achieved on relativelycomplex functions even though the actualmechanism(s) is not at all related to the si-mulation model used.

TEMPORAL INTEGRATION

AT THE NERVE MEMBRANE

Figure 6 (from Butikopfer and Lawren-ce, 1979) illustrates «temporal summa-tion» at the simulated nerve membrane

82 M. W. White

(simulated with an F-H model). Butikopferand Lawrence (1979) also present empiri-cal evidence from a study of the electricalstimulation of mylenated nerve fibers thatcorroborates their simulations using the F-Hmodel. In their F-H model simulation, themembrane voltage does not return to zeroafter each charge-balanced biphasic pulse.Each biphasic pulse increases the depolari-zation of the membrane. The F-H modelpredicts that closely-spaced biphasic pul-ses would be more effective at exciting thenerve than just one biphasic pulse. In con-trast, Hill's model predicts that multiplecharge-balanced biphasic pulses would beno better at exciting the nerve than just onepulse.

The simulations of membrane hyperpo-larization and temporal integration may of-

fer insights useful for interpreting the psy-chophysical data to be presented.

RESEARCH STRATEGY

This paper is divided in two sections.The first section describes the behavior ofthreshold and MCL functions for a widerange of stimuli. Threshold and MCL func-tions delineate the extremes of the sub-ject's dynamic range. The second sectiondeals with the mid-range functions of in-tensity discrimination and loudness. Theexperiments were designed to help us de-velop phenomenological models (and in so-me cases, evidence for specific mechanisms)useful in understanding the subjects res-ponses to a wide range of stimuli. Certainstimuli were particularly useful in separating

'.5I I

0.1 0.3

,REQUENCY (KHZ)

Fig. 5.

•0.05•• S0. 1 0.3

FREQUENCY (KHZ)

Fig. 4.

IS0.05

3S

Fig. 4. - Neural threshold to membrane excitation as a function of sinewave stimulus frequen-cy using a model membrane simulation. A modified Hodgkin-Huxley model was used in whichbeta-h was a factor of four greater than that in the original Hodgkin-Huxley membrane model.The sinusoidal stimulus was initiated at a zero crossing with the first phase depolarizing themembrane model. The first phase of the sinusoidal stimulus depolarized the simulated membrane.

Fig. 5. - This graph displays the point in time at which an action potential was generated inthe simulation represented by figure 4. Time of occurance was recorded in terms of the timethat the membrane voltage reached 90% of maximum. «Time of occurance» was normalizedby dividing the actual time at which an action potential occured by the period of the sinewavestimulus.


~~ ------------------------------------------------------------------

i ------------------

i;;;

~~~ ------------------ ------------ ----------------------------------.-'E>0

Iii

D

D

83

,,.. ,3.2

,3.'

,4.'

Fig. 6. - Simulated membrane voltage as a function of time for multiple biphasic pulse sti-muli. The Frankenhauser-Huxley model was used in this simulation. Figure courtesy of Buti-kopfer and Lawrence (1979).

the effects of two or more mechanisms. Forexample, single biphasic pulse and singlesinusoidal cycle stimuli were used to obtainthreshold and MCL functions that weresubstantially unaffected by temporal integra-tion mechanisms. By comparing thresholdand MCL functions for single pulses withthose for multipulse stimuli, the significan-ce contribution of any temporal integrationmechanism(s) was estimated. If the thre-shold or the MCL function for multiplecharge-balanced biphasic pulses was signifi-cantly lower than those for single pulses,some form (peripheral and/or central) oftemporal integration was likely responsible.

Methods

Thresholds were measured with a modified Be-kesey tracking procedure using a minimum of 6threshold crossings for each threshold estimate. Theaverage of the stimulus minima· and maxima wascomputed to determine the estimated threshold sti-mulus current. The subject pressed a button whenhe or she heard the stimulus and released the but-ton when the stimulus was no longer audible. 600milliseconds elapsed between the end of a stimu-lus burst and the onset of the next stimulus burst.

The maximum comfortable loudness level wasdetermined by initiating the stimulus at an ampli-tude slightly above threshold and gradually increa-sing the stimulus level on a linear amplitude scaleuntil the subject pressed a button to indicate thatthe stimulus had reached the «maximum conifor-

84

table loudness ». The computer-generated stimulussequence was then immediately halted and the next« MCL» series initiated after a three second delay.600 milliseconds elapsed between the end of a sti-mulus burst and the onset of the next stimulusburst.

Intensity discrimination was measured with a2A-2IFC adaptive procedure which converged to a70.7% correct level. If the subject made two cor-rect responses in a row, the discrimination task wasmade more difficult. For every incorrect response,the task was made less difficult. A light markedeach interval in the two interval task. At leasteight threshold crossing occurred before the intensi-ty dl was calculated. The first two threshold cros-sing were not used in the calculation of the esti-mated intensity dl. This adaptive procedure wasdescribed by Levitt (1970).

During all tests, the subject could immediatelyterminate stimulation by disengaging a «master»switch which would immediately disconnect allelectrodes from the optically-isolated current dri-vers (Vurek, 1981). In addition, the computer-ge-nerated stimulus program utilized a set of on-going«consistency checks» which verified that the sys-tem components were operating properly. If anyone of these «consistency checks» proved invalid,stimulation immediately ceased within 1 msec orless.

All stimuli were generated with an opticallv-solated, controlled current source (Vurek. 1981) anddelivered directly to the subject's electrode contactsvia a subcutaneous cable. This cable excited throughthe skin and was connected through a set of re-lays to the described stimulators. Stimuli were ge-nerated from a c1ilYital-to-analogconverter at a sam-pling rate of 20 KHz. Both pulsatile and sinusoidalstimuli were generated. Where appropriate, an anti-aliasing filter was utilized.

Subiects were implanted with scala tympani in-tracochlear electrode arravs of sixteen wires. Theelectrode arrav and the implantation orocedure aredescribed in detail in Loeb et a1.. 1983. The aoical-most electrode is inserted approximately 21 to 26mm into the scala. Each electrode contact is mush-room shaped in order to increase its surface area.The eight bipolar electrode pairs are soaced at 2mm intervals along the silastic intracochlear insert.The inter-contact spacing between bipolar pairs isanorox;metelv 700 microns. center-to-center. Theelectrode pairs are oriented approximately radiallyrelative to the axis of the cochlea. Electrode num-

M. W. White

bering begins at the apical--most part of the arrayand progresses basally, such that the apical-most bi-polar pair is labeled «( 1, 2)>> and the basal-mostbipolar pair is labeled « (15, 16) ». An odd numbe-red electrode represents an electrode contact placedmore towards the modiolus. Both monopolar elec-trode configurations as well as bipolar electrodeconfigurations were used in this study. In the mo-nopolar configuration, only one electrode contactwas stimulated and the "return» path was anear-clip located on the ear lobe nearest the implan-ted cochlea. With monopolar stimulation, the samenumbering system is used, but only one numberdisplayed to indicate which electrode contact isstimulated.

Subject A, age 60, had a gradual onset of hea-ring loss starting after age 10 gradually increasinguntil her loss was profound about thirty years la-ter. There is no family history of deafness nor anyother known reason for this subject's hearing loss.

Subject B, age 61, had a gradual onset of hea-ring loss after measles at age 8 gradually increasinguntil the loss was profound in her twenties. Bothsubjects A and B participated in tests over a threeto four month period.

Subject C, age 51, had a sudden loss of hearingafter an automobile accident (at age 20) in whichhe sustained a bilateral fracture. Subject C under-went psychophysical and speech testing over a pe-riod of approximately a month and a half.

All subjects exhibited greater than 110 dB lossacross the frequency range. All subjects were post-lingual. Each of the three subjects was unable toutilize even specialized high-power body hearing-aids in standard speech discrimination tests. A tho-rough psychological evaluation was conducted toestimate how the subjects would respond to conse-quences of the implantation.

Results

RESPONSES TO SINGLE PULSESAND CYCLES

Figure 7 illustrates threshold and MCLresponses as a function of pulse width forsingle biphasic pulses. The slope of thethreshold function significantly increases for


Ieee. 34.

:~, 28.,...

" 189.+~+

~~

~+ i"'.... 19.

i

~

I+

18. I-*- nfl£SHCUl *-+- to..-----*

1- 9.189. 1_. 1888a. lee. 1_.

PU..SE WIImi (US) PU..SE WIDlM (US)

Fig. 7. Fig. 8.

85

+!

1_.

Fig. 7. - Threshold and MCL for single pulses as a function of pulse width. Data from sub-ject B with bipolar electrode pair (7, 8) stimulated.

Fig. 8. - Dynamic range (derived from data in figure 7) for singel pulses as a function of pul-se width. Data from subject B with bipolar electrode pair (7,8) stimulated.

pulse widths greater than 1000 l-Lsec(goingfrom a 4-6 dB/octave slope to an appro-ximately 9 dB/octave slope function, where-as the slope of the MCL function remainsrelatively constant and less steep as a func-tion of pulse width.

Figure 8 illustrates how the dynamicrange changes as a function of pulse widthfor single biphasic pulses. The dynamic ran-ge is quite narrow (8-12 dB) for pulsewidths of 200 to 1000 l-Lsec, but increasesvery significantly at the larger pulse widths.At a pulse width of 10 msec (i.e. a pulsewidth comparable to the duration of a cycleof a 100 Hz sinusoidal stimulus) the dyna-mic range is nearly 32 dB. This dynamicrange is approximately that of a 300 msec,100 Hz sinewave stimulus.

Figure 9 displays threshold and MCL(maximum comfortable level) as a function

of sinusoidal stimulus frequency. The sinu-soidal stimulus was either a burst with a300 msec duration or a single sinusoidalcycle. The single cycle threshold curve isonly approximately represented by a singleslope of about 6 dB per octave. Thresholddoes not increase quite as much as 6 dB/octave at the higher stimulus frequencies.An average increase in threshold of 5 dB/octave over the 1000-4000 Hz range forsingle cycles is typical. Between 100 and1000 Hz, the average increase in thresholdis 9 dB/octave for single cycles.

The spectrum of single pulses and sinecycles is a very significant function of thepulse width cycle duration. Threshold, MCL,and dynamic range functions are functionsof pulse width and as a consequence theyare also functions of the spectrum of thesestimuli.

86 M. W. White

600.

60.

SI-

!6.

.610.

-<>- .3 5£CJ T~SKUl-.•.. .3 SEC. tt:L 34.- SItG..E CYa..E. TIflESKUl- 51 MGlE CYa..E. tt:L 0

30.~

~~D--.....g

~~ ~ ~o• 0-

~/ / i29.

--A )~~

\/I 010. 0,,,,,--

-<>- 300 I1SEC- SItG..E CYa..E

0-0

O .100. 1000. ooסס1 10. 100. 1_. .ooסס1FREQUENCY(HZ) FREClJE~Y (HZ)

Fig. 9. Fig. 10.

Fig. 9. - Threshold and MCL as a function of frequency for single sinewave cycles and a burst,300 msec in duration. Data is from subject B with bipolar electrode pair (7,8) stimulated. Thesinusoidal bursts were generated without onset or offset envelopes.

Fig. 10. - Dynamic range as a function of frequency for single sinewave cycles and a burstduration 300 msec. Data is from subject B with bipolar electrode pair (7,8) stimulated.

RESPONSES TO REPETITIVE

BIPHASIC STIMULI

Responses to multiple cyclesinewave stimuli

As outlined in the strategy of researchsection (section 5), responses to multiplecycles and pulses have been examined inorder to understand the effects of any hypo-thetical temporal integration processes.

In figure 9, the threshold curve for 300msec bursts exhibits a relatively flat por-tion between 50 Hz and 100 Hz and a steepincrease in threshold as the stimulus fre-quency is increased from 100 Hz to 200 Hz.Above 200 to 400 Hz threshold increasesat a relatively shallow rate of 3-4 dB peroctave. The major difference between thesingle cycle and 300 msec MCL curves isa small vertical shift. For single cycles, asomewhat larger stimulus amplitude is re-quired to elicit the maximum comfortableloudness.

For the 300 msec stimulus, thresholdand MCL curves are roughly parallel for sti-mulus frequencies above 200 to 400 Hz.As a consequence, the dynamic range (seeFigure 10) remains approximately constantfor short duration sinusoidal stimuli. ~IOBger GyratieD l>timyh.ls. This is not truefor short duration sinusoidal stimuli. Thre-shold increases at a rate greater than thatof A g th@ ~ti.m.l11"1! £r~qll~U(,Y ii 1n,rpol;1CPrl,

..t~e d~rl;\r.lmkraJ.:l.gli:MCL vs frequency overthe entire frequency range that was mea-sured (50-4000 Hz). Qe~til;\laill>to d~cr~aHItgr thill>il l>hgrt l>timyli The reduction inthe dynamic range for these short stimuli,as compared to the approximately constantdynamic range of the 300 msec stimuli, isprimarily due to the difference in the shapeof the two threshold curves.

Figure 10 is a plot of dynamic range asa function of the stimulus frequency for thedata presented in Figure 9. The dynamic


range for short and long duration sinusoidsis roughly the same for frequencies belovT400 Hz. For frequencies higher than 400Hz, figure 10 illustrates that the dynamicrange for the short duration stimuli isconsiderably less than that for 300 msecstimulus bursts. Above 400 Hz, the dyna-mic range for the short duration stimuli isconsiderably less than that for 300 msecstimulus bursts. Above 400 Hz, the dyna-mic range remains approximately constantat about 17 to 18 dB for the 300 msecbursts. However, for the single cycle sinu-soids, the dynamic range decreased fromabout 18 dB at 400 Hz to less than 8 dBat 4 KHz. Such changes in dynamic rangeas a function of stimulus duration can havea profound effect on speech processing stra-tegies.

87

The differences in the two thresholdfunctions is the primary cause of the largedifferences in the dynamic range for thetwo types of stimuli. Threshold functionsfor 300 msec sine bursts are quite differentthan threshold functions for single cycles,probably because of one or more temporalintegration mechanisms.

Responses to multiple pulses

Graph 11 illustrates how threshold cur-rent changes as the number of pulses in-crease. As the number of pulses increase thethreshold decreases. The greatest change inthreshold occurs at the shorter interpulseintervals; particularly for interpulse dura-tions less than 5 msec. For example, the 1msec inter-pulse interval curve exhibits a

v

100

90

r.<I:~ 80v

S0o 2 9t;t OF PULSES

Fig. 11.

'00

C8: MONOPOLAR (DPW=200 US

90 PULSE'" CONSTANT

~X

80r. J~<I: ,/Y~v

70t-ZW~ 60

~V

S0 1

10 0 '0

INTERPULSE DUR MSFig. 12.

Fig. 11. - Threshold current as a function of the number of 200 usec biphasic stimulus pul-ses. Data is from subject B with monopolar electrode (1) stimulated. Circles represent thresholdsto pulses with an inter-pulse dew of 10 msec. Squares r7J'resent thresholds to pulses with aninter-pulse delay ofS mses.: )( ~.Sffls·ec,; A-.Lh1S·ec~Fig. 12. - Threshold current as a function of the inter-pulse duration (i.e., inter-pulse) delayin msec. Pulses were biphasic and 200 llsec in duration. Data is from subject B with monopo-lar electrode (1) stimulated. Circles represent stimuli with two pulses. Squares represent sti-muli with four pulses. Triangles represent stimuli with eight pulses. The X's represent stimuliwith only one pulse.

88 M. W. White

1000.

1000.

a -aD

10. ~~~~"""'-~~~""""-'~~~I""~10. 100. 1000.

FJlEQ.EHCY (PPS)

Fig. 13.

10. 1--~~""",_~~""",~~""""""10. 100. 1_. 1_.

FRElll.£I«:Y(PPS)

Fig. 14.

Fig. 13. - Threshold and MCL as a functin of pulse rate for pulse trains 300 msec in durationand pulse widths of 200 usee. Data is from subject B with bipolar electrode pair (15, 16) sti-mulated.

Fig. 14. - Threshold and MCL as a function of pulse rate for pulse trains of 20 msec and300 msec in duration and a pulse width of 200 usee. Data is from subject B with bipolar elec-trode pair (7, 8) stimulated.

very large decrease in threshold as the num-ber of pulses is increased. In contrast, thethresholds of pulse trains with 10 mscc in.ter-pulse intervals exhibit only small decrea-ses as the number of pulses is increased.

Figure 12 is derived from the same dataillustrated in figure 11. Threshold is plottedas a function of the inter-pulse duration,with each curve representing a differentnumber of pulses. As the inter-pulse inter-val increases from 1 msec to 10 msec, thethreshold increases. A large increase inthreshold occurs as the interpulse delay isincreased from 2.5 msec to 5 msec. Thecurve with highest thresholds representsthresholds obtained with a stimulus of onlytwo, 200 f.l.sec biphasic. The dashed curverepresents thresholds with four pulses, andthe curve with lowest thresholds is for sti-muli of eight pulses. The «X's» in theupper right for graph 12 represent thre-sholds for single biphasic pulses.

Figure 13 is a plot of threshold and MCLfor 200 f.l.sec (100 f.l.secper phase) biphasicpulse trains applied to two bipolar electro-des (15,16) in subject B. Threshold decrea-ses as a function of pulse rate, particularlyfrom 200-400 to 2000 pps. Stated in otherterms, the threshold decreases very signifi-cantly for interpulse duration less than 5msec. In contrast, the MCL function exhi-bits only a very small decrease slope as thefrequency is increased. Figure 14 illustratesa similar function for a 20 msec and a 300msec stimulus driving bipolar electrode pair(7,8) in subject B.

Figure 15 illustrates how dynamic rangeincreases as a function of pulse rate for thedata illustrated in figure 13. Because thre-shold decreases significantly with pulse ratecompared with MCL function, dynamic range increases quite significantly, particularlyfor pulse rates greater than 200 pps.

Figure 16 illustrates how threshold andMCL vary as a function of burst duration


20.

18. /0~B 0

16. /i14.

I 012. /18.

~8.

~._. _.

FRfQl£tcY (PPS)

Fig. 15. - Dynamic range as a function of pulserate for pulse trains 300 msec in duration and pul-se widths of 200 usee. Data is from subject B withpolar electrode pair (15, 16) stimulated. Derivedfrom data illustrated in figure 9.

for a 2000 Hz sinusoidal stimulus. Thre-shold decreases at a greater rate than doesthe MCL function when the burst durationis increased. As a consequence, the dynamic

89

range increases as a function of the burstduration (Fig. 17).

Additional evidence for temporal inte-gration is observed in measures of thresholdas a function of burst duration. Figures 18,19 and 20 illustrate threshold as a func-tion of burst duration. In these illustrations,pulse width was 200 f1.sec(100 f1.sec/phase).In two subjects (A and B), thresholds wereobtained at 100, 200, 400, and 1000 ppsstimulus bursts except in figure 20 wherethe 200 pps data was not obtained. Thegreatest change in threshold vs burst dura-tion occurs at the highest frequency (i.e.pulse rate) tested: 1000 pps. At 1000 pps,the greatest change in threshold occurs with-in the first 20 msec. Because MCL vs burstduration changes relatively little in thesetwo subjects as a function of burst du-ration, this would imply that the dynamicrange changes significantly within the first20 msec of these stimuli: the dynamic rangebeing at a minimum for the very shortest

Fig. 16. - Threshold and MCL for 2000Hz sinusoidal stimuli as a function of burst durationfor a bipolar (13, 14)~nd a monopolar (14)l\electrode configuration in subject A.

Fig. 17. - Dynamic gnge for 2000 Hz sinu'?o~al stimuli as a function of burst duration for abipolar (13, 14) and a monopolar (14) electrode configuration in subject A. Derived from dataillustrated in figure 16.

90 M. W. White

burst durations, and increasing considerablywithin the first 20 msec.

For 100 and 200 pps pulse stimuli onlyvery modest decreases in threshold occur forvery large changes in the burst duration.These small thres~old changes could be theresult of at least~tV:~ factors: (1) A centraldetection mechanism may integrate over a

considerable time interval to detect a signal.In the normal auditory system there isstrong evidence that a more central tem-poral integration mechanism operates overat least a 50-100 msec interval; (2) Althoughmost presumed « temporal integration at thenerve» occurs during a 2-5 msec time win-dow, a small but significant peripheral tem-

IZS.-0- 1_ PPS

IZS.-0- 400 PPS-A- 200 PPS-+- 100 PPS

\\~-------~ a A A

"- 0 0

S S a"j

a "----0-\ j a

38.8. .82 .IM .!l6 .88 .1 .12 .14 .16

IlUlST D..RATIOO < SEe)

3.1.8. .81 .82 .83

BUlST OI.IlATlOO < SEC).84

Fig. 18. - Threshold as a function of burst duration for a set of pulse rate rates (100, 200,400, and 1000 PPS) and a pulse width of 200 (.tsec. The vertical scale is logarithmic. Data isfrom subject B with bipolar electrode pair (1, 2) stimulated. A 40 msec and a 160 msec timescale are used to display the data over the full range of measure.

188.-0- 1_ PPS-0- 400 Pf'S-A- 200 PPS-+- 108 PPS

188.

ZS. ZS.8. .02 .IM .!l6 .Be .1 .12 .14 .16 8. .81 .82 .83 .04

IlUlST WlATlOO (SEe) BURST lJ..RATlOO (SEC)

Fig. 19. - Threshold as a function of burst duration for a set of pulse tate rates (100, 200,400, and 1000 PPS) and a pulse width of 200 (.tsec. The vertical scale is logarithmic. Data isfrom subject B with monopolar electrode (1) stimulated. A 40 msec and a 160 msec time scaleare used to display the data over the full range of measure.


-c- 1000-0- 400 PPS~ 100 PPS0............ -

. "-\c'c ---.0>-------_0

\~

91

~. +-'-+~-~-+-~ •........•-+~-+-~-'-+0. .02 .04 .06 .08 .1 .12 .14 .16

euRST C'P-ATlC7.'l (SEC>

50.0. .01 .02 .03

BURST OLIlATlIlH (SEe).04

Fig. 20. - Threshold as a function of burst duration for a set of pulse rate rates (100, 200,400, and 1000 PPS) and a pulse width of 200 usec. The vertical scale is logarithmic. Data isfrom subject A with electrode pair (1, 16) stimulated. A 40 msec and a 160 msec time scaleare used to display the data over the full range of measure.

poral integration might occur over a consi·derably longer interval; (3) Over the stimu-lus interval, the extracellular and intracellu-lar ionic concentrations may change, andthereby change the nerve's excitability (Mo-ran and Palti, 1980; White, 1983).

The data for the 400 pps pulse trainalso exhibits a decline in threshold as theburst duration is increased. In figure 18,the threshold decreases in a manner simi-lar to the 100 and 200 pps pulse trains.However, in figures 19 and 20, the thre-shold decreases significantly at durationsgreater than 20 to 40 msec.

Figure 21 illustrates thresholds for bi-polar electrodes (1,2) in subject B and alsoillustrates thresholds for monopolar electro-de (1) in the same subject. For 1000 ppsstimuli, threshold drops by a factor of about3.5 during the first 20 msec with monopo·lar stimulation as compared to only a factorof two for the same interval with bipolarstimulation. If the interpulse duration isless than 2-5 msec, monopolar (as compa-red to closely-spaced bipolar electrode pairs)

stimulation consistently causes a larger per-centage decrease in threshold as the burstduration is increased over the first 0-20msec. Also, threshold declines at a greaterrate with monopolar stimulation (as oppo-sed to bipolar stimulation) when the pulserate is increased above 200-400 pps.

WI.

\ -4- HOHOPOLAR(I)

40-0- BIPOLAR (1,2)

00. \"-0",~ 60. \4 0______ 0

0

g 40. \-.*

420.

0. +----~~'_+~ .•.....•-'-_+__'~~'-+~-'--'__'__l0. .01 .02 .03 .&4

BURST 0UlA TI Qj (SEe)

Fig. 21. - Threshold as a function of burst dura-tion for monopolar (1) and bipolar (1, 2) stimula-tion. Pulse widths of 200 [tsec and repetition ratesof 1000 PPS were used in these experiments withsubject B. Data from figures 18 and 19.

92 M. W. White

INTERACTION OF PULSE WIDTH

AND PULSE RATE FACTORS

leee.

1 . +---- ..•....•..•..•...••..•....---'--"-•....•...•.•.•••.- •.......•.--.....•.•.•.••le. lee. leee. leee0.

FRfQte(;Y (PPS)

-*- T~SHa..OI PIoI •• 20e US-4l- 1'0. i P\oI •• 2ee US-11- Mct.i PW •• 1800 US-+- Tff1ESHCl..Di PW •• 1800 US

Fig. 23. - Threshold and MCL as a function ofpulse rate for pulse trains 300 msec in duration andpulse widths of 200 J.l.secand 1800 J.l.sec.Data isfrom subject B with bipolar electrode pair (7, 8)stimulated.

+, +-++/

" lee~ .v

Figure 22 displays dynamic range asa function of the pulse rate and pulse widthfor biphasic pulse trains. The burst dura-tion of all stimuli was 300 msec. The dyna-mic range is greatest for larger pulse widths.Also, the dynamic range increases as thepulse rate is increased above 200 pps.

Another significant response feature isevident in figure 22. For pulse widths of900 and 1800 fJ.sec, the dynamic range at100 pps is greater than the dynamic rangeat 50 pps or at 200 pps. This is due to thelower threshold at the 100 pps rate as com-pared to the thresholds at 50 pps and 200pps, for pulse widths of 900 and 1800fJ.sec. Figure 23 illustrates how the thre-shold dips at 100 pps for the 1800 fJ.spulse

Fig. 22. - Dynamic range as a function of pulserate and pulse width for pulse trains 300 msec induration. Data is from subject B with bipolar elec-trode pair (7, 8) stimulated. Data from that illus-trated in figure 23.

21.

width stimulus, but does not dip when 200fJ.spulses are used.

Figures 24 and 25 illustrate how thre-shold, MCL, and dynamic range change asa function of pulse width for pulse trains of300 msec in duration and pulse rates of100, 400, and 1000 pps. Increasing pulsewidths consistently extend the dynamic ran-ge for these pulse trains; just as in the singlepulse case illustrated by figure 7 and figure8. As previously described, threshold tendsto be a decreasing function of pulse rate;most notably for pulse rates above 200 to400 pps. In figure 24, threshold decreaseswith increasing pulse rate except at a pulsewidth of 1800 fJ.sec,where the threshold forthe 100 pps stimulus is lower than that forthe 400 pps stimulus .. For the 100 pps data,more than a vertical shift is required to align

leee8.lee. leee.FRECl.SCY (PPS)

-*- PU..SE WIDTH" 290 USEe-+- PU..SE WIDTH" gee USEe-.- Pll.SE WIDTH •• leae USEe

7.18.

"t!l 17.

19.

Design of a cochlear prosthesis 93

but the largest dynamic range at the largepulse widths.

TEMPORAL INTEGRATION

OF TWO-CHANNEL STIMULI

The previously described results indi-cate that some form of short-term temporalintegration may be present in subjects Aand B. This section presents evidence forone type of short-term temporal integra-tion process: «temporal integration at thenerve membrane ». One peripheral nervemechanism that might produce temporal in-tegration was described by Butikopfer andLawrence (1979) and was briefly reviewedin section 4.

When the current generated by two ormore stimulus pulses enters the nerve mem-brane, some of the « residual» or «remain-ing» excitation due to the first pulse mayreduce the nerve's threshold such that thesecond pulse may elicit an action potential,even though the second pulse alone, couldnot have elicited an action potential (Buti~kopfer and Lawrence, 1979). The locationof the generator's of these two current pul-ses is not significant in this model. Onlythe stimulus magnitude of the current reach-ing the nerve and the relative timing ofthese pulses is significant. In fact, the twostimulus pulses could originate from entirelydifferent location as long as sufficient cur-rent from both electrodes was developed atthe point of excitation and as long as thepulses were temporally «close to eachother ».

Figure 26 illustrates the loudness esti-mates that subject A made when the twobiphasic pulses came from separate electro-de channels plotted as a function of thedelay between the two pulses. Loudnessincreases significantly when the inter-pulsedelay (or duration) is reduced below 5 msec.

1_.

1SJ8. 1888.

-*- lee PP8 llf£SH(l..D--+- 488 PP8 TH<E9O...D

~888PP8~

+~- •

~ ~,,+*

*

~~.+7 -*- 199 PP8-+- 488 PPS4>- Ul99 PPS

8.I.

18.

21.

1888.

" 189.~

~ill.

the threshold curves in figure N. There ap-pears to be a significant interaction betweenpulse width and pulse rate at 100 pps. Ofthe stimulus set measured, the 100 pps sti-mulus exhibits the « narrowest» or smallestdynamic range at the small pulse widths,

l..J--'--~-'-'~&.f--.&.-...L...-'-'-"""""..I.f189. 1888.

PU..8E WIont (US >Fig. 24.. Threshold and MCL as a function of pul·se width at 100, 400, 1000 PPS and a burst dura·tipn of 300 msec. Data is from subject B with bi·polar electrode pair (7, 8) stimulated.

~. 1888.PU..8E WIDTH (US)

Fig. 25.. Dynamic range as a function of pulsewidth at 100, 400, 1000 PPS and a burst durationof 300 msec. Data i< from subject B with bipolarel~ctrode pair (7, 8) stimulated.

94

lJ)(fJ .

\i~'~o::Jo-' l~

, I , I , I , I , I , I , I , I , I , It'1l'1l0 2'11'11'11 '3'11'11'11 '-i00~ 5'110'11 60'11'11 7'1100 8'11'1113 9'11'1113 t0000

INTER-CHAN DELAY (US)

Fig. 26. - Loudness as a function of the delaybetween the initiation of single 200 IJ.sec biphasicpulses on two channels. The first channel was mo-nopolar electrode (1) and the second channel wasmonopolar electrode (7). Subject A made 9 loudnessestimates for each of the interchannel delays.

Two factors are necessary to cause achange in the subject's loudness estimateswhen stimulating two electrode channelswith fixed amplitude and fixed pulse-widthbiphasic pulses. First, each of two electrodechannels must produce a significant currentdensity at one or more common nervemembrane locations. Second, the delay be-tween pulses should be below 1-2 msec toelicit the loudest percepts; and the delaybetween pulses should be greater than 4 to5 msec to elicit the lowest loudness estima-tes. We would expect to observe an in-crease in loudness when we change the in-terpulse delay from 5 msec to 1 msec, ifthere were enough fibers that received a si-girificant amount of current from both ofthe .'two channels. In contrast, if none ofthe fibers received a significant amount ofcurrent from both electrode channels, wewould not expect to observe a significantchange in loudness as the inter-pulse delaywas changed from 5 msec to 1 msec.

M. W. White

A set of experiments were conductedin which the loudness was measured as afunction of three variables: the inter-pulsedelay, the distance between electrode chan-nels, and the electrode channel geometry.Two subjects were tested in this series ofexperiments. The electrodes were either mo-nopolar or bipolar (with contacts orientedradially and an intercontact separation ofapproximately .7 mm). As the distance be-tween electrode channels was increased, thechange in loudness (due to the change ininter-pulse delay) decreased very significant-ly. The type of electrode was a very signi-ficant factor in the loudness function. Bi-polar electrodes as compared to monopolarelectrode channels significantly reduced theextent of these two-channel interactions.

Figure 27 illustrates how two monopo-lar channels interact when the stimuli fromthe two channels are not temporally over-lapping. The subject was asked to adjust aslide pot in accordance with how loud thestimulus was. The slide pot was graduatedon a scale of 0 to 100; where « 0 » was justbelow threshold and «100» representedthe maximum comfortable loudness (MCL).

Loudness estimates were obtained for apseudo-randomized set of stimuli in whichthe delay between the pulses on the twochannels was varied. Inter-channel delayswere set to one of eight values. Delays of.5, 1, 2, 4, 6, 8, 9, 9.5 msec were used.The interaction was measured as a func-tion of interchannel delay and distance be-tween the two channels. Instead of usingone pulse per channel, each channel was sti-mulated at a rate of 100 pulses per second.This experimental modification i.e., leng-thening the burst duration) made the tasksimpler for the subject, but does not changethe basic interpretation of the experiment.Each biphasic pulse was 200 IJ.sec (100


Fig. 27. - Subject A's loudness estimates as a func-tion of the delay between pulse trains on two mo-nopolar electrodes and as a function of the distan-ce between the two electrodes. The loudness esti-mates have been shifted vertically on the loudnessscale (0-100) for readibility. One of the two stimu-lated electrodes was the most apical monopolarelectrode (ie electrode «1»). The other stimulatedchannel was a monopolar electrode located either2, 4,6, 8, 10, 12, or 14. mm from electrode (1).Each of the two channels was driven.by a 100 pps,30'0 msec, biphasic pulse train with pulse widths of200 !J.sec(100 !J.sec/phase). In" the topfourcurves,twelve loudness estimates. were obtained for eachof the eight interchannel delays. The subjel?t madeonly four to nine loudness estimates for· each in-channel delay iri the last four curves.

l-tsec/phase) in duration. The pulse trainwas presented for 300 msec.

In figure 27, monopolar electrodes wereused for both chanenls. Each curve illu-strates the interaction between two mono-polar channels at a different inter-channels'eparation. The curves are ordered such thatthe interchannel separation increases in 2mm steps, from a starting separation of 2mm (in the top curve), to an interchannelseparation of ·14 mm (in the bottom curve).As the distance between the two electro-des was increased, the amount of interactiondecreased. In this subject (A), interaction

UlUlW.za:Jo-'

8 mm

I14 mm

o ,I" I , I , I , I , I , I , I , I • Io 1000 2000 3000 "000 5000 6000 7000 8000 9000 \0000

DELAY CUSEO

between monopolar electrodes was quite si-gnificant at a distance of 6 mm. Interactiondecreased at an 8 mm interchannel separa-tion and was negligible at a 10 mm inter-channel separation. The difference betweenthe mean loudness for a 1 msec interchanneldelay and the mean loudness for a 5 msecinterchannel delay. The difference in loud-ness was divided by the average of thetwo means and then multipled by 100 toobtain the «percentage change in loud-ness ». This index is a measure of the in-teraction between two channels and will bereferred to as an «interaction index ».

Index = 100 * (Lmax - Lmin) /[(Lmax + Lmin)/2l

Figure 28 summarizes data obtainedfrom subjects A and B. The interaction in-dex is plotted as a function of inter-chan-nel distance and electrode geometry (bipolaror monopolar). The two subjects are dra-matically different in this function, withsubject B exhibiting considerably less in-teraction than subject A. As previously sta-ted, channel interaction «drops off» withinter-channel separation and is further re-duced with bipolar stimulation. When bi-polar electrodes were stimulated in subjectB no statistically significant changes in loud-ness were detected at even the smallest in-terchannel separation of 2 mm.

At the same interchannel separation,loudness changes significantly more (as theinterchannel delay is varied) with monopo-lar stimulation than with bipolar stimula-tion.

Interestingly, the fact that the electro-de configuration. has a strong effect on such«chapnel interactions» supports the con-cept that these temporal interactions are oc-curring at a peripheral rather than a centrallocation. (1) When stimulated alone, eachchannel (whether monopolar or bipolar)

96 M. W. White

Fig. 28. - Temporal interaction as a function of in-terchannel distance, electrode geometry, and sub-ject (A or B). The two subjects are dramaticallydifferent in this measure, with subject B exhibitingmuch less interaction than subject A. Monopolarelectrodes are represented with solid lines. SubjectA's bipolar electrodes are represented with a das-hed line. There was no measureable interaction withbipolar stimulation of even adjacent channels insubject B. The distance between the horizontaldashed line and the horizontal axis represents thestandard erro~ of the interaction index for subjectB with two channel bipolar stimulation. In bothsubjects, the same monopolar electrode configura-tions were used at each interchannel separation(see legend of figure 27). The bipolar electrodeswere located at the same «cochlear place» as themonopolar electrodes. For example, bipolar elec-trode (1, 2) is approximately «located» in the sa-me position as monopolar electrode (1).

xwoz

z,oo'"of-V<r::Ct:Wf-Z

o o

2-CHAN IN1ERAC11DNHONOPOLAR - LINEB [POLAR - DASHED

">--SuaJECT"

I<4.

\ 0

rNTERELECT DrST CMM)

course, is not certain but is quite plausible.Let us assume for the moment that thechanges in loudness were due to centralmechanisms sensitive to the delay betweenthe two channels. If this and the previousassumptions were correct, the monopolarand bipolar electrode configurations shouldgenerate nearly identical loudness changesfor any given change in interchannel delay.However, monopolar and bipolar configu-rations do not generate similar changes inloudness for a given change in «interchan-nel delay». If (1) and (2) are «sufficientlyaccurate », then it is likely that the sourceof the temporal interaction is peripheral.The peripheral-nerve, temporal integrationprocess discussed in section 4 may be a li-kely candidate. In these two channel ex-periments, each of the monopolar electrodechannels probably excite approximately thesame number of fibers as their bipolar coun-terparts. but it is likely that a greater num-ber of fibers are «just below» thresholdafter the first monopolar channel's stimuluspulse than in the bipolar case. This is likelythe situation, because monopolar electrodesgenerate a greater current spread.

Discussion

elicited approximately the same loudness.(2) For the same interchannel separation,monopolar and bipolar electrodes were atnearly the same cochlear locations. Underthese circumstances, we might assume thatany given· monopolar electrode would exci-te nearly the same population of fibers that« its» bipolar equivalent (i.e., same location- same loudness) would excite. This, of

Behavioral threshold is a function ofpulse rate, pulse width, and burst duration.Relatively simple phenomenological modelsof threshold behavior can be useful in des-cribing many characteristics of threshold be-havior. This section considers how .thesemodels can relatively accurately summarizethe experimental data. Many of these sim-ple phenomenological models directly sug-gest one or more alternative mechanisms.

Design of.a cochlear prosthesis

THE THRESHOLD FUNCTION

The threshold function will be describedin two parts. Thresholds to single pulsesand cycles will be considered first and thenextended to thresholds of multiple pulses

.and cycles.

THRESHOLDS TO SINGLE PULSESAND SINGLE SINE WAVE CYCLES

THreshold functions for single biphasicpulses and single cycle sine waves are simi-lar. Hill's model only roughly approximatesthe behavioral threshold function for singlecycle sine waves and single pulses (Figs. 9and 7). In a «best-fit» Hill's model, thre-shold increases above 50-100 Hz to 6 dB/octave. Behavioral threshold to single cyclesdeviates from this model in two frequencyregions. Threshold does not increase quiteas much as 6 dB/octave (perhaps an averageof 5 dB/octave over the 1000-4000 Hz ran-ge) at the higher frequencies (or at theshorter pulse widths). This data indicatesthat shorter cycles and pulses require slight-ly less charge for excitation. McNeal (Mc-Neal, 1976; McNeal, 1977) has shown thatthe shorter pulses are more effective in ex-citing the nerve than would be predictedfrom Hill's «constant-charge» model. Hill'smodel is most directly applicable to elec-trical stimulation of an excitable membranein which one electrode is intra-axonal andthe second electrode is located across themembrane in the extracellular medium.McNeal developed a more realistic electri-cal stimulation model of myleneated nerveby assuming that both electrodes are extra-cellular and at specified locations relativeto the nodes of Ranvier.

In Hill's model, all current that is ge-nerated at the electrodes must traverse the

97

membrane and therefore all the generatedcurrent contributes to the excitation of thenerve membrane. In McNeal's model, thecurrent generated at the electrodes is notconstrained to enter the nerve fiber. Somepercentage of the current may never enterthe nerve fiber and therefore not contri-bute to excitation. Part of the current ge-nerated at the electrodes, may simply beshunted around the nerve by the surround-ing ionic medium. The relative amount ofcurrent that passes through the nerve mem-brane depends on the relative impedance ofthe nerve pathway (composed of the mem-brane and inter-axonal impedances) compa-red to the impedances of the alternativecurrent paths. Here we assume that thesurrounding medium can be approximatelymodelled by a pure resistance (Spellman,1982). A greater percentage of the total sti-mulus charge will flow into the nerve whenthe pulses are shorter; because a greaterproportion of stimulus charge flows throughthe membrane capacitance during the. earlierportions of the stimulus pulse. In otherwords, the membrane capacitance offers alower impedance to the higher frequencystimulus components. Therefore, thresholdcurrent does not increase at a 6 dB/octaverate, but at a somewhat lower rate, becauseshorter pulses (cycles) require less chargeto excite the nerve.

This «frequency response» function islikely a second-order function of the distancebetween the neural process and the stimula-ting electrode( s) as well as the electrodeconfiguration.

As pulse width is increased above 1msec, hyperpolarization generated by onephase of the stimulus may increase the ner-ve's sensitivity to the following phase ofdepolarization. As a conseguence, the thre-

•

IJ

98

shold may decrease at a greater rate than6 dB/oct as the cycle duration is increasedabove 1 msec. Typically, the average slopebetween 1000 and 100 Hz is 9 dB/octavefor single cycles and single pulses. Sucha mechanism has been described in section 3.

A slightly more complex linear filterthan that in Hill's model would more accura-tely model threshold behavior for single cyclesine waves and single pulse stimuli. Thefilter would have a slope of approximately4-6 dB/octave for frequencies above 1000Hz and 9 db/octave for frequencies below1000 Hz. The frequency response would« flatten-out» for frequencies below 100Hz. Data for cycle durations longer than 20msec (or equivalently, 50 Hz) were notobtained because maximum charge densitylimits would often have been exceeded.

In summary, the nerve's threshold forsingle cycles or single pulse varies at roughly6 dB/octave for cycle duration less than10-20 msec. However, significant secondorder effects alter the threshold functionover certain frequency-pulse-width regions.For single pulses or cycles shorter thanabout 1000 tLsec, threshold increases at arate somewhat less than 6 dB/octave. Forsingle biphasic pulses or single sinewavecycles longer than 1000 tLsec and shorterthan 10-20 msec, threshold increases at ahigher rate than 6 dB/octave (about 9 dB/octave) as the stimulus frequency is in-creased.

11ULTIPLE PULSE AND CYCLE

THRESHOLDS: THE HIGHER PULSE RATES

By using single sinusoidal cycles or sin-gle biphasic pulses, a first-order estimate ofthe system's frequency response can be ob-tained with little «contamination» in thisfrequency response measure due to central

M.W. White

or peripheral temporal integration of exci-tation generated by multiple pulses or cycles.

In this section we compare the thre-sholds obtained with multiple cycle stimulito those thresholds obtained with singlecycle stimuli. Since the stimulus spectrum(when measured with a time window ofapproximately the same shape and dura-tion as the impulse response of the linearfilter within Hill's «best-fit» model) isaffected RQ.t; only to a second order by chang-ingthe burst duration or the pulse rate, thedifference between single-cycle and multi-cycle thresholds can be used as an esti-mate of the significance of, and the timeorder of, any temporal integration pro-cess(es). Exceptions to this generality willbe noted later in the text.

Threshold is a strong function of bothpulse rate and burst duration. Thresholdeffects that can be reasonably approxima-ted with only a simple linear filter modelwere discussed in the previous section deal-ing with single sine cycles and pulses. Toinclude the effects of multiple pulses (i.e.,the effects of pulse rate and burst duration),an addition to this filter model is required.When the stimulus contains more than onebiphasic, charge-balanced cycle the linearfilter model for threshold behavior shouldbe extended to include temporal integration.

Temporal integration mechanisms ap-pear to be useful in describing threshold be-haviour to stimuli with multiple pulses. Adescription of several of the most probabletemporal integration mechanisms presentedat the end of this section. The followingresponse characteristics are well modelledby temporal integration processes:

1) The threshold decreases as the num-ber of pulses increase (see figure 11). Cor-respondingly, threshold decreases as theburst duration increases (see figures 18, 19

Design of' a 'cochlear prosthesis

and 20). In general, threshold decreases asburst duration is increased for burst dura-tions as long as 80-160 msec. This suggeststhat at least one of the temporal integrationmechanisms contains a relatively long in-tegration window.

2) In figure 12, threshold decreases asthe temporal spacing between the pulsesdecrease. The decrease in threshold as theinterpulse interval is decreased is particu-larly significant for pulse trains with inter-pulse durations less than some critical value.This critical value may be a useful measureof at least one of the « time windows» ofintegration. In figure 12, when the inter-pulse interval drops below 5 msec, thresholddecreases at a considerably greater rate forequal increases in the burst duration. Si-milarly, in figures 13 and 14, thresholddrops at a much greater rate for pulse ratesabove 200 pps (i.e., the interpulse inter-val is less than 1/200 pps = 5 msec).

3) For sinusoidal stimuli above 400 Hz,thresholds for 300 msec bursts are lowerthan those of single cycles (see figure 9).The deviation between the single cyclethreshold curve and the 300 msec thre-shold curve increases as the frequency isincreased above 200 Hz. Thresholds forlong-duration, higher frequency stimuli areconsiderably lower than thresholds for sin-gle cycles of the same frequency, probablybecause more stimulus cycles occur withinthe integrating time window of the hypo-thetical temporal integrator(s).

4) Loudness changes as a function ofthe delay between pulscf,i on two nearbychannels (see figures 26 and 27) .. Becausethe extent and amount of this interactionis highly dependent on the electrode con-figuration (e.g., monopolar vs bipolar), thisdata lends support to the peripheral originof at least one of the temporal integration

99

processes (see Section 9). Since this twochannel temporal interaction occurs prima-rily over a 2-5 msec time window, this datasupports the peripheral origin of a short-term temporal integration mechanism withan integration window of approximately 2-5msec.

Partial integration of subthresohld de-polarization at the nerve membrane (seesection 4) is one process that could generateresponse features similar to some of thosefeatures described above (i.e., those featu-res asosciated with relatively short integra-tion intervals). This mechanism is parti-cularly attractive in describing, at least,the shorter term temporal integration pro-perties which appear in the threshold be-havior. A longer term integrative mecha-nism also appears to be active. At very lowpulse rates and at relatively long burstdurations, threshold decreases as the burstduration is increased (e.g. figures 18, 19and 20). This behavior may be the result ofa longer term integration process. Such anintegrative mechanism could be the resultof a central neural temporal integrationmechanism or a peripheral-nerve-process tem-poral integration mechanism. Certainly, acentral integration mechanism has appealbecause of the strong evidence for itsexistence in the normal auditory system.Some of the possible mechanisms for tem-poral integration are described in the nextsection.

Possible temporal integration mechanisms

Threshold changes V(lry significantlywith the duration of the stimulus for bothpulse trains and sinusoidal bursts. At leastsome combination of three mechanisms maybe involved:

(1) For the higher stimulus frequen-cies, peripheral nerve temporal summation

v

100

may cause the threshold to decrease as theduration of the stimulus is increased. Thismechanism was described in the introduc-tion. This effect is particularly apparent forthe 1000 pps pulse train thresholds display-ed in figures 18 and 19. This effect' is alsoillustrated in figure 11 where threshold de-creases significantly as the number of pulsesis increased. In general, threshold drops, asduration is increased, and most significantlyfor monopolar electrodes, and for those bi-polar electrodes with wider pair spacing.In figure 19, the 1000 pps monopolar thre-shold drops significantly more than does thebipolar (figure 18) threshold for 1000 ppsas the train duration is increased fromapproximately 1 msec to 20 msec. Figure 9displays thresholds to 300 msec sinusoidalbursts and single cycles. Above 200 Hz,the threshold decreases significantly as theduration is increased. For these higher fre-quencies, peripheral nerve temporal inte-gration may be at least one of the reasonsfor significant drops in threshold.

(2) Central temporal summation. Ifspontaneous activity is generated peripheralto the hypothetical «neural detection cen-ter », detection of a stimulus may requiremore than simply detecting the presence orabsence of neural activity. Some form oftemporal integration would likely be neces-sary. If the neural input contains both noiseand signal, then an ideal detector will exhi-bit lower thresholds for longer burst dura-tions (Green and Swet, 1964). As a con-sequence, such central temporal summationmay be responsible for some of the drop inthreshold as the burst duration is increased.

(3) Stochastic nature of the nerve's ex-citability. The nerve's threshold is notconstant. The nerve's threshold varies overtime (Verveen, 1959). The addition of noiseto a deterministic model of neural threshold

M. W. White

ten"-should fl.oem a more accurate neural thre-shold model (White, 1983). In such a mo-del, the probability of excitation is increas-ed as the burst duration is increased. In afirst order model of this behavior, each sti·mulus cycle could be conisdered an inde-pendent trial. As the burst duration is in-creased, the number of trials increases pro-portionately. The probability of at leastone or «N» firing(s) increases as a conse-quence of the increased number of « trials ».From (1) above, there is evidence that eachtrial is not necessarily independent of theprevious trials. Indeed, previous stimuluscycles may tend to increase the probabilityof firing to subsequent stimulus cycles.

13. MULTIPLE PULSE AND CYCLETHRESHOLDS: THE LOWER PULSE RATES

Up to this point, we have primarily con-sidered responses to stimuli with pulse ra-tes between 400 pps and 4000 pps andsinusoidal stimuli with frequencies between400 Hz and 4 kHz. For stimuli with pulserates between 50 pps and 200 pps a rathermore complex pattern of behavior exists.

Threshold responses are quite depen-dent on both stimulus' pulse width and pul-se rate. These two factors significantly «in-teract ». For pulse widths shorter than 500to 1000 [tsec, responses are similar to whatwould be predicted from the 2-integratormodel useful in describing threshold behavi-or for the higher pulse rate stimuli. Be-cause the pulse rates discussed here are re-latively small (i.e., 50-200 pps), the hypo-thetical short-term integrator will only playa minor role with such low-pulse-rate sti-muli. The effects of the hypothetical longer-term (integration windows on the order of80-160 msec) temporal integrator are bestillustrated in figures 18 and 19 for the 100pps and 200 pps, 200 [tsec pulse width


stimuli. Only small decreases in threshold(2-4 dB) occur for quite large increase inburst duration (from 200 (J.sec to 100 msec).

A threshold model for single pulsesneeds to be an integral part of such a multi-pulse model. The single pulse model is pro-bably best placed just prior to the « slow»integrating process in this multi-pulse model.

Thus, for short pulse width, low repe-tition rate stimuli the previously describedmodel for high-pulse rates should be quiteuseful. However, if the pulse width is grea-ter~ than 500-1000 (J.msec, the complex be-havior of neural excitation becomes consi-derably more apparent. Single pulse behavi-or for large pulse widths has been discus-sed in section 11. At least two mechanismsmay be involved in reducing thresholds forthese single, longer-duration pulses:

(1) The quantity of charge entering theneuron increases with the pulse width. Thiswould cause only a .6 dB / octave change inthreshold if a simple resistor-capacitor andvoltage detector membrane model is used.

(2) The nerve may become more sensi-tive to the depolarizing phase of the stimu-lus after being driven by the hyperpolarizingstimnlus phase (see section 3).

With multiple, long-duration pulses,threshold decreases most significantly forpulse rates near 100 pps and sinusoidal fre-quencies near 100 Hz (see the 1800 (J.secpulse width threshold curve in figure 23).In figure 9, threshold is significantly lowerfor multiple 100 Hz cycles than for single100 Hz cycles. The difference in thresholdsfor multiple cycles vs single cycles is con-siderably less for the « surrounding» 50 Hzand 200 Hz frequencies.

Perhaps most interesting is the data dis-played in figure 24. Here we examine thre-shold and MCL as functions of pulse widthand pulse rate. The 100 pps threshold cur-

101

ve deviates significantly from the patternof the 400 pps and 1000 pps thresholdcurves, particularly for pulse width greaterthan 1000 (J.msec. Most remarkable is theclear evidence for a lower threshold at an1800 (J.sec pulse width for the 100 pps, ascompared to the 400 pps threshold. Thisis remarkable because the 400 pps, .3 secstimulus, contains four times as many 1800(J.sec pulses as does the 100 pps stimulus!Furthermore, the 400 pps stimulus hasmuch more closely spaced (i.e., temporally)pulses than does the 100 pps stimulus. Itappears that longer pulses augment the sen-sitivity of the system to subsequent pulsesif the interpulse interval is near 10 msec. Ifthe interpulse interval is closer to 20 msecor 5 msec, prior pulses or cycles do not ge-nerate such significant increases in the sys-tem's sensitivity (see figure 9). Such beha-vior is not that which would be generatedby some form of temporal integration. Inparticular, the system's sensitivity is greaterfor 1800 fl.sec pulses temporally spaced 10msec apart than for 1800 fl.sec pulses spaced5 msec or 25 msec apart (Fig. 24). A tem-poral integrator should exhibit at the mi-nimum, the same sensitivity (and generallymore sensitivity) to more closely spacedpulses.

Figure 29 illustrates the significant dropin threshold for a 100 Hz sinusoidal stimu-lus as the burst duration is increased. Forshort (200 (J.sec) biphasic pulses with a 100pps repetition rates, behavioral thresholddoes not drop nearly as significantly as doesthreshold for 100 Hz sinusoidal stimulus(see figure 29). The longer duration of eachphase of the 100 Hz sinusoidal stimuli maybe necessary for the process which reducesthe threshold to this type of stimulus.

Perhaps in an analogous manner, r:achstimulus cycle appears to increase the pro-

102

bability of an AVCN cell's firing during theinitial 20 to 40 msec of 100 Hz stimulus(see figure 30). This has been a very consis-tent feature of presumptive AVCN largespherical cell responses to 100 Hz sinusoidalstimuli. Such slow response onsets werenearly non-existent for 200 Hz and higherfrequency sinewaves.

300.-6- 199 PP$, 290 USECP\.l.SES-0- 100 HZ SIHEWAVE (I NORMALIZED)

M. W. White

role in the slow «build-up» of firing. Du-ring electrical stimulation, extracellular ionicconcentrations (e.g. the Potassium concen-tration) could change enough to alter theexcitability of nearby neurons (Moran andPalit, 1980). This effect can occur over atime interval of tens of milliseconds. Duringexcitation or subthreshold stimulation, suf-ficient quantities of one or more ions couldflow into the extracellular medium and the-refore affect fiber excitability.

It is also possible that the build-up ef-fect may be due to neural processing in thebrainstem with excitatory feedback path-ways onto AVCN neurons. If his were thecase, the described slow onset of responsesto 100 Hz sinewave bursts would presum-ably not be present in eighth nerve respon-ses to this type of electrical stimulus.

Fig. 29. - Threshold stimulus current as a functionof burst duration for two types of stimuli: a 100Hz sinusoidal burst and a 100 pps train of 200usec biphasic pulses. Data is from subject A withelectrode pair (1, 2) stimulated. The 100 Hz thres-hold currents have been shifted vertically (by mul-tiplying each 100 Hz threshold value by a factor of13.3) on the log scale to obtain a better visualcomparison.

White, Merzenich and Loeb (& Whit)have discussed a set of alternative mecha-nisms for this slow-onset behavior: TheFrankenhaeuser-Huxley model describes ma-ny of the effects of the more quickly act-ing nerve membrane phenomenon. It is notclear whether the F-H model can accuratelypredict such long-term effects as those des-scribed in the previous paragraph. A slowerfeature of nerve excitation (which is notincorporated in the F-H model) may playa

100.B. .1 .2

BtRST llURATl OH (SEC).3 THE MCL FUNCTION

For subjects A and B, MCL decreasesonly slightly as the pulse rate is increased(see figures 13 and 23). For these subjects(A and B), maximum comfortable loudnessis not a strong function of burst duration(see figures 16 and 9). MCL is only a veryweak function of pulse rate. Pulse width,on the other hand, has a considerably stron-ger influence on MCL than does either pulserate or burst duration. MCL decreases atabout 3-4 dB/per octave of pulse width(e.g. see figures 7 and 24). Likewise, for si-nusoidal stimuli with frequencies above 200Hz, MCL increases at about 3-4.5 dB/octave(e.g. figure 9). MCL behaves similarly forboth pulse and sinusoidal stimuli as a func-tion of the duration of the cycle.

MCL varies significantly with the si-gnal's spectrum; but MCL varies relativelylittle with burst duration or pulse rate. Be-cause MCL is only weakly influenced by


380-J::> 11dll00.."E..%

~;;

..00

38

0-J::> 03dB00.."a:..z0en;;

00

···..n ••;1 I'

TIME (USE C)

........

76

0-J::> 15dll00.."a:..%0•.;;

12000

TIYE (USEC)

76

0-J::> 17"_00.. ·" · · ·a:.. ·% · · ·0 · · ·•. · · ·;; .. · · · ·.. · · · .... · · · .... · · . I I ••.. .. .. . ... .. . u.I.I.'1

00TillE ellSECI

Fig. 30. - Post stimulus time histogram series wich illustrates the gradual onset of a presu-mably large-spherical AVCN cell's response to an electrical stimulus of 100 Hz. As stimuluslevel is increased, the response «rise-time» is considerably shortened in these single unit re-cordings in cat. Note that the vertical scale changes from 38 spikes maximum to 76 spikes maxi-mum in the two histograms representing responses to the two highest stimulus levels. Illustra-tion from White, Merzenich, and Loeb, 1983.

burst duration or pulse rate, there is evi-dence for only a relatively small amountof temporal integration at the periphery atthese high suprathreshold stimulus levels.However, a considerable amount of centraltemporal integration may exist. If only asmall increase in the stimulus level is neces-sary to excite a large number of additionalfibers, only small increases in MCL wouldbe necessary to generate a sufficient quanti-ty of « neural activity» which would totallycompensate for the neural activity that was« lost » because of a decrease in burst dura-tion and/or pulse rate. This explanationuses a simple «total-neural-activity» modelfor central temporal summation (see sec-tion 18).

To a first approximation, the MCL func-tion could be modelled by a simple linear

filter with a frequency response function ofapproximately 3-4.5 dB per octave overthe frequency range of 1"00-4000 Hz; MCLincreasing with frequency. From 50-100 Hz,the filter's slope is approximately zero (i.e.,there is little or no change in MCL). Inthis simple model, a «threshold detector»is placed after the filter. When the voltageacross the detector reaches a specified value,the maximum comfortable level is «signal-ed» or «elicited» in this model.

A temporal integrator is not included inthis first order model because MCL is onlya weak function of pulse rate and burst du-ration. In contrast to the threshold func-tion, MCL is only a weak function of thosestimulus variables (Le., pulse rate, andburst duration) which would cause significant

104

changes in MCL if a peripheral temporal in-tegrator was present. As previously discuss-ed, a central temporal integrator would notnecessarily cause even moderate changes inMCL as burst duration or pulse rate arechanged; particularly if small changes in sti-mulus amplitude recruit relatively largenumbers of fibers.

A filter model of the MCL function isonly a first-order approximation. For exam-ple, burst duration has a small but notice-able affect on the shape of the MCL vsfrequency curves in figure 9. At the higherfrequencies, the single cycle MCL functionincreases at a slightly greater rate than doesthe MCL function to the 300 msec stimuli.This small effect cannot be simulated by theproposed linear filter model alone. The smallamount of temporal integration that is ex-hibited in the MCL function is of a con-siderably smaller magnitude than that exhi-bited in threshold functions. There are atleast three possible explanations; any com-bination of which may be responsible: (1)Those fibers that are excited at, and near,threshold levels may exhibit a very signi-ficant amount of temporal integration; how-ever, the presumably many additional fibersthat are excited at MCL levels may exhibitlittle or no temporal integration (or mayhave a much shorter time window of inte-gration). This will be refered to as the«two-population model» and is discussedfurther in section 17. If we assume thatloudness and therefore MCL is a functionof total neural activity (see section 18), thenthe MCL function would reflect propertiesof both neural populations. Therefore, wewould expect some influence, although re-duced, from the temporal integration asso-

M. W. White

ciated with those fibers excited near thre-shold levels.

(2) A central temporal integration me-chanism may be at least partially responsiblefor the auditory systems added sensitivityto multiple cycles at the higher suprathre-shold (as well as at threshold levels). Ifsmall increments in stimulus amplitude (atMCL levels) cause relatively large increasesin the number of excited nerves in a givensubject, only a small increase in the stimuluslevel would be necessary to maintain iden-tical amounts of total neural activity evenif the burst duration was quite significantlydecreased.

(3) For that portion of temporal integra-tion that is due to a central temporal inte-gration mechanism, another or additionalfactor may be at least partially responsiblefor the relatively small changes that arerequired in stimulus level at MCL for verylarge changes in burst duration. Both eighthnerve (& Moxon) and AVCN (& White)responses to suprathreshold electrical sti-mulation of the auditory nerve exhibit si-gnificant amounts of adaption. Specifically,responses to stimuli significantly above neu-ral threshold respond at a higher firing rateduring the initial 5-20 msec segment of thestimulus (estimate of time course). If thisis also the case with implanted subjects,total «summed» neural activity would notbe proportional to the duration of the sti-mulus. For example, of a relatively longduration stimulus (say 20-40 msec) is doubl-ed in duration to 40-80 msec, the totalneural activity would be significantly lessthan double that for the 20-40 msec stimu-lus simply because both stimuli elicited arelatively large portion of total neural acti-vity during the first interval of the response.

Design 0/ a cochlear prosthesis

SUMMARY OF THE THRESHOLD

AND MCL FUCTIONS

This section reviews both threshold andMCL behavior and suggests model that maybe useful in the interpretation of «mid-range» functions such as loudness and in-tensity discrimination. In one sense, thre-shold and MCL functions are simply the twoextremes of an infinite set of equi-loudnessfunctions. By understanding what differen-ces exist in threshold and MCL behavior, weshould be able to synthesize «integrated»models which describe both threshold andMCL behavior. At least three response fea-tures are distinctly different for thresholdvs MCL functions. The following discussionconsiders these differences and describes al-ternative interpretations.

The MCL function can be approximatelyrepresented as a linear filter followed by anon-integrating threshold detector. For sin-gle cycles, the filter's frequency response isapproximately 3-4 dB/octave above 100 Hz(see figure E). At longer burst durations,the MCL function deviates slightly from aconstant slope of 3-4 dB/octave. This de-viation is probably not due to the smallspectral difference in the two stimuli, sincethe small spectral change would cause achan:;e of excitation in the opposite direc-tion to that observed. Because the shape ofthe MCL functions frequency response isdiffe:'ent, although only slightly, for the twodifferent stimulus durations, the simple fil-ter model can only be an approximation tothe MCL function. Temporal integrationprocess(es) may cause the MCL function todeviate from the linear filter function as thenumber of cycles is increased. In figure 9,at frequencies above 200 Hz, the shape ofthe two MCL curves differ from each otherin a qualitatively similar manner to howthe two threshold curves differ from each

105

other. This could indicate that the essentialmechanism responsible for the difference inthe two MCL curves is identical to thatresponsible for the two threshold curves.Both pairs of curves diverge as the fre-quency is increased above 200-400 Hz. Thisbehaviour is consistent with that behaviourwhich would be generated by the previous-ly discussed peripheral nerve temporal in-tegration mechanism.

SHORT-TERM TEMPORAL INTEGRATION:

THRESHOLD AND MCL FUNCTIONS

In two subjects (A and B), thresholdresponses demonstrated behaviour consis-tent with a very significant component ofshort-term temporal integration. In contrast,the MCL functions were only slightly af-fected by this presumed short-term tempo-ral integration process. In the third subject(C), neither the threshold nor MCL funtionsbehaved in a manner consistent with thattypically associated with a short-term tem-poral integration process.

Two POPULATION MODEL

One model that would integrate theshort-term temporal integration propertiesof both the threshold and MCL modelswould incorporate two or more functionalneural populations. In this model we donot need to assume that two or more dis-tinct types of fibers exist, only that two ormore types of peripheral neural responsescan exist to identical electrical stimuli. Dif-fer:::nt types of responses could originatefrom the same neuron, if the initial site ofexcitation (ISE) was different. In this mo-del, one of the neural populations (group A)would exhibit a significant amount of «pe-

106 M. W. White

Fig. 31. - Threshold as a function of pulse ratefor biphasic current pulses. These threshold func-tions wer:: obtained by Herndon (1981) using anelectrode array implanted in the modiolus. Pulsewidths were 200 [tsec. Figure courtesy of MattHerndon.

Fig. 32. - Threshold as a function of pulse ratefor very short-duration «biphasic-exponential» pul-ses, The majority of charge transfer occurs in lessthan 25-50 [tsec per phase. These threshold func-tions were obtained by Herndon (1981) using anelectrode array implanted in the modiolus. Figurecourtesy of Matt Herndon.

_E~ •.,._

SLs RECTANGULARCURRENTPULSES

Et

E4

1000

E2

E3

400

800

.300

600

200

400

L _BIPHASIC EXPONENTIAL- -Y PULSES

FREOUENCY, pp,

FREQUENCY, PPS

200

r~;:/.....~~.<::::,:.,<:>~~.~,,:~.~_E2

'"":'.>.(.?'.~.~<J~.~., ,TE4

40

30

~.g0:r 20<IlwCl:

~'0

40

30~g" ""---0:r r-<Il 20"'Cl::rf-

10

("-

100

ripheral-nerve, temporal-integration; theother population (group B) would exhibitlittle or no «peripheral temporal integra-tion» or would exhibit a considerably shor-ter temporal integration time window. Inthis model, we assume that loudness, andtherefore MCL, is monotonically related tothe overall amount of neural activity contri-buted by both populations. We will assumethat group A responses exhibit the lowestthresholds (because the fibers are presu-mably closer to the stimualting electrodesand/ or because the processes or ISE's areexceptionally sensitive). As a consequence,threshold functions would strongly exhibitbehaviour representative of a temporal in-tegration process. In contrast, MCL func-tions may only exhibit a relatively smallamount of temporal integration, since only afraction of the total number of excited fiberswould « temporally integrate» the stimulus.

Herndon (1981) found two groups ofbehavioral responses depending on whichof four modiolar electrodes were stimulated.Specifically, Herndon discovered that thre-sholds to biphasic pulse stimuli decreasedvery significantly as a function of pulse ratefor two monopolar electrodes, and did notdecrease noticeably for the other two mono-polar, modiolar electrodes (fig. 31). In hisstudy, the stimulating electrode array waslocated in the modiolus and therefore mightbe expected to sample a different, and per-haps larger distribution of nerve processesthan would be sampled by a scala tympanielectrode array. Interestingly, Herndon(1981) reports that thresholds were cons-tant with pulse rate for all four electrodesif quite short, exponentially-shaped biphasicstimulus pulses were used (see figure 32).This would indicate that only an amplitudeand time-dependent non-linearity (e.g., the


H-H or F-H model) could accurately modelthis behavior.

Van Den Honert (personal communica-tions) in single neuron recordings in eighthnerve has found evidence for two types ofneural responses to electrical stimulation. Atlow stimulus, a longer latency response iselicited. This response exhibits distinctly dif-ferent refractory behavior than that obser-ved for the shorter latency responses gene-rated at the higher stimulus levels. Theyhave been able to eliminate the longer la-tency response component by destroyingeighth nerve ganglion cells and dendrites,but leaving the axons intact. They believethat this behavior is due to two distinct si-tes of excitation. One site (with its own res-ponse characteristics) may be excited atlower stimulus levels and another site isexcited at the higher stimulus levels.

With scala tympani stimulation, onetype of threshold behavior has been observ-ed in subjects A and B. These subjects havedemonstrated threshold functions consistentwith a peripheral, short-term integrationprocess. Very recently, a third subject (C)has been tested psychophysically and doesnot exhibit behavior consistent with thepresence of this peripheral, short-term tem-poral integration process. This subject iscompletely devoid of two-channel temporalinteractions (see section 9) at inter-channeldelays greater than .5 msec and does notdisplay significant decreases in threshold asthe inter-pulse interval is decreased from 5msec to 1 msec if the number of pulses isheld constant. Observations during surgery,the subject's etiology, psychophysical fieldinteractions measures, and ABR thresholdand magnitude measures indicate that thissubject probably has a very low nerve sur-vival. It is possible that this recent sub-ject (C) is nearly or totally devoid of

107

group A processes and therefore does notexhibit the specific set of response featuresassociated with this type of peripheral,short-term integration porcess.

With all three subjects, there is signi-ficant evidence for another temporal inte-gration process which exhibits a consider-ably longer duration integration window(approximately 80-160 msec). It appearsthat the longer-term temporal integrationmechanism may be distinct and separatefrom the hypothetical shorter-term temporalintegrator, simply because only two of thethree subjects exhibited both forms of tem-poral integration.

Another model that could account fordifferences between these threshold andMCL responses could be proposed. This mo-del is very similar to that proposed in thenext section (section 18) to account for cer-tain features representative of long-term(80-160 msec) temporal integration. In thismodel a central integration process is pro-posed. It is proposed that threshold andMCL functions of burst duration and pulserate are different because the «recruit-ment » of excited fibers is a non-linear func-tion of stimulus intensity. It is proposedthat MCL functions may be more shallowthan threshold funtcions of burst durationand pulse rate in subjects A and B becausemany more additional fibers are excited fora given increase in stimulus intensity atMCL levels than at threshold levels.

However, this model is only useful indescribing behavior associated with a cen-tral temporal integration mechanism thatsums already-generated neural activity fromnumbers of fibers. Contrary to that model'sassumptions, the evidence presented in sec-tions 9 and 4 indicates that the short-termtemporal integration mechanism originatesprior to, or at the point of excitation along

108

the peripheral neural process. Spellman's da-ta (1982) indicates that only linear processesoccur within the tissue surrounding the ner-ve processes; and therefore it is highly li-kely that the short term temporal integra-tion process occurs at the highly non-linearnerve membrane during the process of neu-ral excitation.

LONG-TERMTEMPORAL INTEGRATION:THRESHOLDAND MCL FUNCTION

With all three subjects (A, B and C),there is strong evidence for a second tem-poral integration mechanism which has aconsiderably longer integration window. Forexample, in figures 18, 19 and 20, all thre-shold fucntions decrease with increasingburst duration for burst durations as largeas 80-160 msec. MCL functions also de-crease with burst duration, even for burstdurations as large as 80-160 msec. How-ever, in general, the change in MCL (in dB)for a given change in burst duration is notas great as that seen in the threshold func-tions. For example, in figure 14, MCLchanges only slightly compared to changesin threshold when the burst duration ischanged from 20 to 300 msec (except at50 pps).

The duration over which temporal inte-gration occurs is very similar to that seenin the normal auditory system. The centralorigin of the normal auditory system's long-term temporal integration process is com-monly acknowledged. Let us assume forthe moment that a central temporal inte-grator is appropriate for modelling the long-term temporal integration process observedin these electrically stimulated patients. Onecommonly used model assumes that thecentral processor integrates neural activityover many fibers and over a time interval

M. W. White

of approximately 100 to 300 msec and thatthe loudness function is some simple mono-tonic function of a single value which re-presents this total neural activity.

In this model, threshold and MCL arevery sensitive functions of fiber recruit-ment. With fiber recruitment, we considerboth the initial excitation of the fiber andthe fiber's firing rate vs stimulus intensityfunction. Overall neural activity is determi-ned by the number of recruited fibers andthe firing rate of each of these recruitedfibers. If at MCL, a large number of addi-tional fibers are recruited (or perhaps equi-valently, if the firing rates greatly increase)with only a small increase in stimulus am-plitude, only a small increase in stimulusamplitude would be necessary to maintainthe same quantity of «neural activity» fora large decrease in burst duration. If nearthreshold, only a small number of additio-nal fibers are recruited for a relatively largeincrease in stimulus current, a relatively lar-ge incerase in stimulus current would benecessary to maintain the same «neural ac-tivity» for a given decrease in burst dura-tion. Thus, if the recruitment function grewat an increasing rate with stimulus ampli-tude, we might expect MCL to be a weak-er function of burst duration than thre-shold.

In contrast, if we assumed that the lon-ger-term temporal integration occurred pe-ripherally, then a two-population modelcould be proposed to account for the diffe-rences in threshold and MCL functions ofburst duration. Such a model is very similarto that proposed in the previous section.These two hypothetical neural populationsassociated with long-term temporal integra-tion would necessarily be distinctly differentfrom the two hypothetical populations dis-cussed in the previous section on short-


term temporal integration. The hypotheti-cal « population A » short-term temporal in-tegration neural processes are absent in sub-ject C. However, both populations associa-ted with long-term integration would bepresent in all three subjects.

stimulus durations). If the change in neuralactivity is less at 100 Hz and 200 Hz for agiven change in stimulus intensity, then onemight expect a more gradual loudnessgrowth stimulus intensity and therefore anMCL higher than would have otherwisebeen expected at these low frequencies.

6.

280.

1. 2. J. 4. 5.RELATIVE AtFLITlIJE (DB)

Fig. 33. - Spike counts for 100, 200, 400, and3200-6400 Hz sinusoidal stimuli as a function sti-mulus amplitude (in dB) relative to threshold am-plitude. Spike counts were obtained from the ini-tial 20 msec response interval of unit recording inthe large spherical cell region of the cat AVCN.Spike counts were obtained with 50 tri[lls.

~~ 100.

200.

STIMULUS DURATION

A number of investigators have focussedon threshold, loudness, and MCL functionsfor stimulus bursts of durations greater than100 msec. The use of constant amplitude,long-duration sinusoidal stimuli for thre-shold, loudness, MCL, and intensity discri-mination measurements is too limited a sti-mulus set for obtaining information usefulin speech processor design. Relatively shorttransitions within the speech signal normallyconvey a large portion of speech informa-tion. These transitions range from a fewmsec to tens of milliseconds. Even the re-

In all three subjects studied, thresholdsdropped precipitously as pulse width wasincreased above 1-2 msec; and simifarly,with sine wave stimuli, threshold decreaseddramatically as the stimulus frequency wasdecreased below 500-1000 Hz. In contrast,a considerably smaller decrease in the MCLlevel was observed over these same stimu-lus ranges (Figs. 9 and 7). As a consequen-ce, dynamic range is large for the largerpulse widths and for the lower frequencysine wave stimuli, particularly for sine wavestimuli in the 50-200 Hz range where dyna-mic ranges are 22-35 dB (see figure 10).Section 13 contains a discussion of someof the alternative mechanisms that may beresponsible for the significant drop in thre-shold at the lower frequencies and largerpulse widths. Why does MCL not also dropas si:nificantly as does threshold when thesinusoidal stimulus frequency is reduced be-low 200 Hz? Recording of single unitAVCN responses in cat to scala tympanielectrical stimulation (White, 1983) has sug-gested one possible interpretation. Neuralresponses (Fig. 33) to sinusoidal stimuli of100 and 200 Hz generally grew with sti-mulus amplitude at a considerably lowerrate than did unit response to higher fre-quencies (i.e. 400 Hz, 3200 Hz, and 6400Hz). This pattern was consistent for diffe-rent stimulating electrodes and for differentstimulus durations (i.e. 5, 20, and 80 msec

THRESHOLD AND MCL AS A FUNCTION

OF PULSE WIDTH AND CYCLE DURATION

110 M. W. White

1.4

Fig. 34. - Intensity difference limens as a func-tion of loudness for single pulses of 200 !J.secand1800 !J.sec pulse widths. Data is from subject Bwith bipolar electrode pair (7, 8) stimulated.

0. +--'--_+_~-+---'-_---~f__'--+_---_+_-'-_40. 10. 20. 30. 40. ~. 60. 70.

LOlO£SS <8-100>

a SIHGLE Ieee USEC PULSE4 SlinE 2ee USECPULSE

latively «steady-state» portion of a vowelis quite complex in its envelope fluctua-tions. If the envelope is measured usinganything but a very long integration period,the amplitude envelope of the vowel willvary considerably over time. The envelopeis approximately periodic at the glottal pe-riod. This example illustrates how even themost «steady» segments of speech deviateconsiderably from the constant stimulus am-plitude model that has often been used im-plicity. Stimulus is integrated over some in-terval of time. One component may be re-lated to the nerve's hypothetical temporalintegration. This component probably con-tributes significantly to changes in thresholdover the first 20 msec of the stimulus, par-ticularly when the pulse rate or stimulusfrequency is above 200-400 Hz.

1.2

~ 1.~d .8

~ .6

~f- .43

.2

a

~A • a

A --r---•.. _-.....----r•.- A

MID-RANGE FUNCTIONS

Intensity discrimination

In the normal auditory system, intensitydiscrimination as measured in dB is not astrong function of stimulus level. Sufficient-ly above threshold, intensity discrimination(measured in dB) may stay nearly constantor decrease by only a factor of 2 or 3 overan 80 dB change in stimulus level. IntensitydLs for normal hearing subjects may rangefrom approximately 2 dB to .25 dB (Green-wood, 1983).

Intensity discrimination was measuredas a funtion of burst duration, pulse rate,pulse amplitude,and pulse width for subjectB. These intensity discrimination functionsare a strong function of the subject. Preli-minary results from a third subject (C) willbe briefly reviewed and a comparison made.

Figure 34 illustrates how pulse widthcan affect the distribution of intensity dLs

as a function of loudness for single pulses.Similarly, Figure 35 illustrates how pulsewidth affects the distribution of intensitydLs as a function of loudness for low repe-tition rate stimuli with burst durations of300 msec. In both examples, the longerpulse width stimuli results in larger inten-sity dLs, particularly at the lower stimuluslevels. The narrow pulse width stimuli pro-duce relatively constant intensity dLs acrossthe measured loudness range. The averageintensity dL increases with pulse width asdoes the dynamic range (Figs. 8 and 25).

Figure 36 illustrates how pulse rate canaffect the distribution of intensity dLs acrossthe dynamic range. At low pulse rates, theintensity dL remains relatively constant withstimulus intensity. However, at high pulserates, a marked change occurs in the distri-bution of intensity dLs across the dynamicrange. At the low stimulus levels, a rela-tively large change in stimulus intensity is


1.••

1.2

1.~~ .9d1:.6

~ ...~

.2

e.0. 10. 20. 30. 40. !50. 60. 70.

LOl.O£SS

D 10 MSECCYCLE,lee HZ

Fig. 35. - Intensity difference limens as a functionof loudness for low repetition rate stimuli with acycle duration 10 msec. Please compare this to thatcurve in figure 36 representing intensity dls for asmall pulse width (200 fJ.sec) and a low repetitionrate (50 pps) with a burst duration of 300 msec.Burst duration was 300 msec. Data is from subjectB with bipolar electrode pair (7, 8) stimulated.

necessary for discrimination. However, atthe higher intensities, there is very little, ifany, difference in the intensity dLs for the50 pps and 2000 pps stimuli. The dynamicrange for the high pulse rate stimuli isgreater than the dynamic range for the lowpulse rate stimulus (see figures 15 and 22)and correspondingly the average intensitydL for the high pulse rate stimulus is largerthan the average intensity dL for the lowpulse rate stimulus.

How does the intensity discriminationfunction change if the burst duration ischanged? Figure 37 illustrates intensity dis-crimination functions for two burst dura-tions: a single 200 f-tsec pulse and a 2000pps, 300 msec pulse train. Clearly, the func-tions are very different. With the single200 f-tsec biphasic pulses, relatively constantand small intensity dLs are observed as com-

!l 2000 PPSJ .3 SEC, 200 USEe PU..SES6 ~0 PPS; .3 SEC; 200 USEC

6

D

-------- 6--6

1. ••

1.2

l!l1.

d .9

l: .6

~•••tz-.2

0.0.

I • I • I • I I' I • I

10. 20. 30. 40. !50. 60. ~.L.CUH:SS (0-100)

Fig. 37. - Intensity difference limens as a functionof loudness for long burst duration (.3 see, 2000pps) and very short burst duration (one 200 fJ.secbiphasic pulse) stimuli. Stimuli are composed ofbiphasic, 200 !J.secpulses. Data is from subjects Bwith bipolar electrode pair (7, 8) stimulated.

1."

1.2D

~ 1.l!l~d .9

~ .6

~ .43 6

.2 6 --;r--A_------r-&

0.0. 10. ze. 30. 40. :59. 60. 70.

LOl()I£SS (0-100)

D .3 SEC, 2000 PPSJ 200 USEe PULSES6 SINGLE200 USECPULSE

Fig. 36. - Intensity difference limens as a functionof loudness for high pulse rate (2000 pps) vs lowpulse rate (50 pps) biphasic, 200 fJ.secpulse trainsof 300 msec duration. Data is from subjects B withbipolar electrode pair (7, 8) stimulated.

112 M. W. White

Fig. 38. - Dynamic range as a function of pulserate for pulse trains of 20 msec and 300 msec induration and a pulse width of 200 llsec. Data isfrom subject B with bipolar electrode pair (7, 8)stimulated.

pared to the large dLs observed for the 300msec, 2000 pps stimuli at the lower stimulusintensities. As before, the stimulus whichexhibited higher average intensity dLs alsoexhibited wider dynamic ranges (fig. 38).

Burst duration has a significant effecton the intensity dL and dynamic range func-tions if the pulse rate is sufficiently high.However, longer burst duration do not al-ways increase intensity dLs. For example,Figure 39 illustrates just such a situation.In this case, the 300 msec stimulus is a50 pps pulse train with the same pulse widthas in the previous example. In this example,the distribution of intensity dLs across thedynamic range is relatively constant and ap-proximately equal for both the single pulseand the 300 msec pulse train. Similarly, thedynamic ranges for the two stimuli are ap-proximately equal (fig. 38).

o

a 300 HSEC, :50 PPSA SIHGLE 200 USEC PULSE

a

A Q.--------7!'- 8A------------ A.2

1.2

1.4

a .8>-~ .6%w3 .4

In summary, either high pulse rateor large pulse width stimuli generated con-siderably elevated intensity dLs particularlyat the lower stimulus intensities. For highpulse rate stimuli, as the burst duration wasdecreased the elevated intensity dLs (at thelower stimulus levels) decreased until theintensity dLs were equal to those of singlepulses. Dynamic ranges were correspond-ingly higher for high pulse rate stimuli andlarge pulse width stimuli. Also, similarlythe dynamic range for high pulse rate sti-muli decreased as the burst duration wasreduced.

Figure 40 illustrates how intensity dis-crimination and dynamic range co-vary fora range of stimuli in subjects A and B. Be-cause the intensity discrimination and dy-namic range measures co-vary in a reason-ably consistent manner, we might expectthat there are one or more common mecha-nisms involved in the generation of thesetwo behavioral measures. In this figure, dy-namic range has been calculated in a di£-

0. +--~+__~+__~I__'____<f___''___lf___''___l~~ _0. 10. 20. 30. 40. 50. 60. 75.

LOUJI£SS (0-100)

Fig. 39. - Intensity difference limens as a functionof loudness for long burst duration (.3 sec, 50 pps)and very short burst duration (one 200 llsec bipha-sic pulse) stimuli. Stimuli are composed of bipha-sic, 200 llsec pulses. Data is from subject B withbipolar electrode pair (7, 8) stimulated.

2000.200.FREGIl.E«:Y (PPS)

-0--4-

17.5

~ I~.

~ 14.

~ 13.

'" 12.ui 11.

610.9.8.

7.20.

16.


ferent manner than in the other illustra-tions. Dynamic range was calculated by sub-tracting the current level (in dB) requiredto obtain a loudness of 15 from that cur-rent level (in dB) required to obtain a loud-ness of 70. Current levels required to obtainspecified loudnesses were determined withthe «best-fit» equation described in sec-tion 2. (The cochlear prosthesis group atStanford has used a similar method for anumber of years.)

.85

.7 00

~ .6 0

d .5 0 00

0 0>- .4l- e~ .3 00w~ .2 •

.1

0.O. 2. 4. 6. 8. Ie. 12. 14. 16.

!l'l'I'Wl1C RANGE (DB)

Fig. 40. - Scatter plot of two subject's (A and B)average intensity dLs as a function of dynamic ran-ge for a range of stimuli including biphasic pulseand sinusoidal stimuli at a set of burst durations. A« best-fit» line passing through these data pointsindicates that the average number of discriminabledifference within the dynamic range may be in theorder of 20.

As pulse rate or burst duration is in-creased, intensity dLs increase significantlyparticularly at the lower loudness levels.This may indicate that a change in pulse rateand/or burst duration has its primary effecton the fibers that are most sensitive to thestimulation. In a similar manner, an increasein pulse rate or burst duration reduces thre-

113

shold considerably more (in dB) than theMCL (see Figures 13 and 9). In other words,in subject B intensity discrimination, thre-shold, and MCL measures indicate that chan-ges in pulse rate and burst duration havetheir greatest effect on the lower regions ofthe dynamic range.

There are a number of factors that mayinfluence intensity discrimination across thedynamic range. This is a partial list of thosefactors:

(1) The number of additional fibers ex-cited for a given increment in stimulus in-tensity. Such «recruitment» is a functionof the current spread around the stimulatingelectrodes and the density of excitable neu-ral processes. Such « recruitment» is a func-tion of «the current density as a functionof location» and «the density of excitableneural processes as a function of locationwithin the cochlea and modiolus ».

(2) The rate at which the firing rate in-creases for a given increment in stimulusintensity.

(3) If their are two or more popula-tions which demonstrate different respon-ses to a stimulus variable, then the inten-sity discrimination function could be signi-ficantly affected due to such a response dif-ferential.

Loudness functions

Figure 41 displays loudness as a func-tion of stimulus amplitude for pulses withshort and long burst durations. If the sti-mulus amplitude is displayed on a log scale,the loudness functions grow slowly at firstand then increase very significantly as thestimulus is increased further. If the stimulusamplitude is displayed on a linear scale, theloudness functions appear to grow morelinearly. Loudness functions were well fit

114

with a simple equation of the form:(a+bI)**c, in which a, b, and c are con-stants for a given stimulus and « I » is thestimulus amplitude in !-tamp. These curvefitting were used to obtain estimates ofthe current levels required to obtain aloudness of '15' and a loudness of '70' inorder to calculate a more conservative esti-mate of the subjects' dynamic range (seeprevious section) .

7a. A 0 0

63.

~ sa.:il~ 40.

~ 30.

~ 20 ..J

10. A '0

0. . I . . .a. 100.

A/'PL I Tl()E (UA)

Fig. 41. - Loudness as a function of stimulus am-plitude for three types of stimuli: a single 200 (J.secbiphasic pulse (circles); a 300 msec, 2000 pps pul-se train of 200 (J.secbiphasic pulses (squares); anda single 1800 (J.secbiphasic pulse (triangles). Loud-ness function were well fit with a simple equationof the form: (A+Bx)**C, where A, B, and Careconstants for a given stimulus. Data is from sub-ject B with bipolar electrode (7, 8) stimulated.

Variation of Loudness

In both subjects A and B, loudness esti-mates at the lower stimulus levels variedwith a standard deviation of about + 1-10 units on the 0 to 100 loudness scale. Thestandard deviation of the loudness estima-tes decreased as a function of the mean

M. W. White

loudness. This variability was particularlytroublesome, at loudness levels below 10 to20 on the loudness scale. In many instances,the stimulus that «on the average» wouldelicit a loudness estimate of «10» wouldbe inaudible to the subject. As a conse-quence, intensity discrimination and pitchmeasurements at average loudness levelsof « 10» and «20» on the loudness scalewere more difficult. More significant, is theeffect of such variability on speech andsound processor design. For example, onewould probably want to design processorsin such a manner that all significant stimuliare placed sufficiently above threshold toinsure that the sounds will nearly always beaudible. In effect, this means that the de-signer has less useful dynamic range to workwith.

Why is there such variability at theselow levels? There are probably a number ofreasons. One suggestion involves a tinnitusor head noise. These subject-described « headnoises» vary in their intensity over time andwere present both before and after the im-plantation of the electrode array. This formof « tinnitus» may act as an effective mask-er of the lower stimulus levels.

Subject A exhibited another, perhaps re-lated, characteristic in her loudness estima-tes. When listening to repetitive 300 msecstimulus bursts (with a 700 msec inter-sti-mulus interval), subject A reported thatthe loudness of these constant amplitudebursts varied over a considerable range du-ring a listening interval. More specifically,the loudness estimates would wax and waneover 30-40% of the dynamic range. Sud-den loudness changes did not occur; butthe bursts were said to slowly change inloudness. Both a midrange stimulus (appro-ximately 50 loudness units on the average)and a lower level stimulus (approximately

Design of. a cochlear prosthesis

20-30 units «on the average») were inve-stigated. Only an approximate description ofthis process can be attempted here. Themaximum rate of change of these loudnessestimates was approximately 10 loudnessunits for a 3-5 sec interval. More typically,subject A's estimates would change about10 loudness units within a 5 to 15 sec in-terval. This characteristic of subject A'sestimates was observed during three sepa-rate test sessions over a two week period.

In contrast, subject B heard the repeat-ing constant amplitude stimulus bursts as aset of constant loudness bursts.

Multichannel stimulation

Previous work on electrode interactionshas been reported. Suffice it to say, that alarge body of work is necessary to designmultichannel processors that efficiently mapperceptually significant acoustic informationinto electrical stimuli across a multichannelarray. Previous studies (White, 1983; Ed-dington, 1978; and Merzenich and White,1976) have presented evidence that elec-tric field interactions may occur during elec-trical stimulation of two or more channels.White (1982) and Herndon (1982) havepresented evidence that temporal interac-tions occur between two channels. One de-sign strategy has been to avoid these in-teractions by using a set of techniques des-cribed below. However, with a better un-derstanding of these interactions, it may bepossible to effectively utilize them to ex-pand and improve the transmission of use-ful information to the nervous system.

One method used to minimize thesechannel interactions, is to minimize the dis-tance between the electrodes and the ex-citable tissue. Also, the distance between theelectrode channels may be increased to avoidthese interactions. Still another technique

115

may be useful: that of temporally inter-lacing the stimuli on the channels. Thistechnique avoids the electric field interac-tions that occur when simultaneously stimu-lating two or more channels. Non-simulta-neous stimulation appears to very significant-ly reduce channel interactions. However, thistechnique alone does not eliminate « tempo-ral» interactions. The nerve membrane ap-pears to partially summate excitation over apost-stimulus time interval of 2-5 msec insome subjects. Stimuli from two channelsmay partially summate at the nerve evenwhen the stimuli are not temporally over-lapping. However, if the stimuli are sepa-rated in time by at least 4-10 msec, theseinteractions are significantly reduced. A com-bination of the above techniques may beuseful in reducing channel interactions.

In the design of proposed speech orsound processors, it is important to esti-mate the range of stimuli that will appearat the cochlear electrodes. Some processorsmay use only biphasic pulse stimuli over arestricted set of pulse widths, rates, and am-plitudes. Because such a stimulus set is res-tricted, stimulus processing may be corres-pondingly simpler. Other processors mayuse what is commonly refered to as «ana-log» stimuli. The term «analog» is ill-defined, but generally implies that the sti-mulus at the electrodes is continuously vary-ing in time and that the spectral content ofthe stimulus is generally within the typicalspeech range of 200 to 8 KHz. Generally,the « analog» stimulus is some derivative ofthe original acoustic speech signal.

After the range of stimuli have beenestimated, it is hoped that the prosthesis de-signer can use the concepts and data pre-sented in this manuscript to help in his de-sign of a cochlear prosthesis.

116

RIASSUNTO

L'A. effettua una analisi psicoacustica della fun-zione di discriminazione per soglia, 'loudness' e in-tens ita adoperando una vasta gamma di stimoli.

Per l'interpretazione dei dati comportamenta-Ii, sviluppa dei modelli fenomenologici per tali fun-zioni, servendosi di concetti neurofisiologici fonda-mentali tra cui i processi di integrazione temporale.

Present a dei dati comparativi sugli impianticocleari uni- e multicanale in modo che il progetti-sta possa approfondire meglio Ie conseguenze dellastimolazione elettrica del nervo acustico.

REFERENCES

BUTIKOFER, R., and LAWRENCE, P. D. (1979)Electrocutaneous nerve stimulation - II: stimu-lus waveform selection. IEEE Transactions Bio-medical Engineering, 26, 69.

CLOPTON, B. M., SPELMAN, F. A., and MILLERJ. M. (1980) Estimates of essential neural ele-ments for stimulation through a cochlear pros-thesis. Annals of Otology, Rhinology and La-ryngology, 89, supp!. 66, 5.

EDDINGTON, D. K., DOBELLE, W. H., MLA-DEJOVSKY, M. G., BRACKMANN, D. E.,and PARKIN, J. L. (1978) Auditory prosthe-sis research with multiple channel intracochlearstimulation in man. Annals of Otology, Rhino-logy and Laryngology, 87, supp!. 53, 5.

FOURCIN, A. J., ROSEN, S. M., MOORE B. C.J., DOUEK, E. E., CLARK, G. P., DODSON,H., and BANNESTER, L. H. (1979) Externalelectrical stimulation of the cochlea: clinical,psychophysical, speech-percentual, and histologi-cal findings. British Journal of Audiology, 13, 85.

FRANKENHAEUSER, B., and HUXLEY, A. F.(1964) The action potential in the myelinatednerve fiber of Xenopus Laevis as computed onthe basis of voltage clamp data. Journal of Phy-siology, 171, 302.

GREEN, D. M., and SWETS J. A. (1974) Signaldetection theory and psychophysics. John Wileyand Sons 1966, reprinted by R. E. Krieger Pub!.Co., New York.

HERNDON, M. K. (1981) Psychoacoustics andspeech processing for a modiolar auditory pros-thesis. Technical Report No. G906-5; Stanford

M. W. White

Electronics Laboratories, Stanford University,Stanford, CA.

HILL, A. V. (1936a) Excitation and accomodationin nerve. Proceedings of the Royal Society, 119,305.

HILL, A. V., KATZ, B., and SOANDT, D. Y.(1936b) Nerve excitation by alternating current.Proceedings of the Royal Society, 121, 74.

HOCHMAIR-DESOYER, I. J., HACHMAIR, E. S.,FISCHER, R. E., and BURIAN, K. (1980) Co-chlear prosthesis in use: recent speech compre-hension results. Archives of Otolaryngology,229, 81.

HODGKIN, A. L., and HUXLEY, A. F. (1952)A quantitative description of membrane currentand it application to conduction and excitationin nerve. Journal of Physiology, 117, 500.

KIANG, N. Y. S., and MOXON, E. C. (1972)Physiological considerations in artificial stimula-tion of the inner ear. Annals of Otology, Rhino-logy and Laryngology, 81, 714.

LEVITT, H. (1971) Transformed up-down methodsin psychoacoustics. Journal of the Acoustical So-ciety of America, 49, 467.

LOEB, G. E., WHITE, M. W., and JENKINS,W. M. (1983) Biophysical Considerations in Elec-trical Stimulation of the Auditory Nervous Sys-tem. Annals of the New York Academy of Scien-ces (in press).

LOEB, G. E., BYERS, C. L., REBSCHER, S. J.,CASEY, D. E., FONG, M. M., SCHINDLER,R. A., GRAY, R. F., and MERZENICH, M. M.(1983) Design and fabrication of an experimen-tal cochlear prosthesis. Medical and BiologicalEngineering and Computing, 21, 241.

LOEB, G. E., WHITE, M. W., and MERZENICH,M. M. (1983) A new theory of acoustic pitchperception. Biological Cybernetics (in press).

MCNEAL, D. R. (1976) Analysis of model for ex-citation of myelinated nerve. IEEE Transactionsof Biomedical Engineering, 23, 329.

MCNEAL, D. R., and TEICHER, D. A. (1977)Effect of electrode placement on threshold andinitial site of excitation of a mylenated nervefiber. In: Resnick, J., Hambrech, T.: Functio-nal electrical stimulation. New York, M. Dek.ker.

MERZENICH, M. M., MICHELSON, R. P., PET-TIT, C. R., SCHINDLER, R. A., and REID,M. (1973) Neural encoding of sound sensationevoked by electrical stimulation of the acousticnerve. Annals of Otology, Rhinology and Laryn-gology, .82, 486.


MORAN, N., and PALTI, Y. (19801 Potassium ionaccumulation at the external surface of the no-dal membrane i nfrog myelinated fibers. Biophy-sical Journal, 32, 939.

SIMMONS, F. B. (1966) Electrical stimulation ofthe auditory nerve in man. Archives of Otola-ryngology, 84, 1.

SIMMONS, F. B., WHITE, R. L., MATHEWS,R. G., and WALKER, M. G. (1981) Pitch cor-relates of direct auditory nerve electrical stimu-lation. Annals of Otology, Rhinology and La-ryngology, suppl. 82, 15.

SPELLMAN, F. A., CLOPTON, B. M., and PFIN-GST, B. E. (1982) Tissue impedance and cur-rent flow in the implanted ear. Annals of Oto-logy, Rhinology and Laryngology, suppl., 91, 3.

TEICHER, D. A., and MCNEAL, D. R. (1978)Comparison of a dynamic and steady-state mo-dels for determining nerve fiber threshold. IEEETransactions of Biomedical Engineering 25, 105.

VERVEEN, A. A. (1959) On the fluctuation ofthreshold of the ner~e fiber. In: Tower, D. B.,Schadde, J. J.: Structure and Function of theGeneral Cortex. Proceedings of the Second In-ternational Meeting of Neurobiologists, Ams-terdam.

117

VUREK, L. S., WHITE, M. W., FONG, M., andWALSH, S. M. (1981) Opto-isolated stimulatorsused for electrically evoked BSER. Annals Oto-logy Rhinology and Laryngology, suppl. 82, 21.

WHITE, M. (1978) Design Considerations of aProsthesis for the Profoundly Deaf. DoctoralDissertation, University of California, Berkeley.

WHITE, M. W., MERZENICH, M. M., andLOEB, G. E. (1983) Electrical stimulation ofthe eighth nerve in cat: Temporal propertiesof unit responses in the large spherical cell re-gion of the AVCN. Annals of Otology, Rhino-logy and Laryngology (submitted).

WHITE, M. W., MERZENICH, M. M., and GAR-DI, J. N. (1983) Multichannel electrical stimu-lation of the auditory nerve: channel interac-tion and processor design. Archives of Otola-ryngology (submitted).

M. W. WHITE,~.

School of MedicineDepartment of Otolaryngology

University of CaliforniaSan Francisco, California

94143 (USA)

psychophysical and neurophysiological considerations …mark-white.com/literature/italy84.pdf ·...

Documents