AU2011265365B2 - Electrical nerve stimulation based on channel specific sampling sequences - Google Patents

Abstract

Abstract A method of activating electrodes in a multichannel electrode array using channel specific sampling sequences is presented. A channel specific sampling sequence is defined for each electrode, the sequence having a 1o particular duration, amplitude, and number of pulses. A weighting factor is applied to the channel specific sampling sequence. Each electrode in the multichannel electrode array is then simultaneously activated using sign correlated pulses, the sign-correlated pulses based on parameters of spatial channel interaction, non-linear compression, and each electrode's weighted 15 channel specific sampling sequence. 3038178_1 (GHMatters) P44928.AU.4 <2/7 0 U d~ d~ LO ) to C)t ~~~~~. If t 0 iO-If LO 8 o 0 L L ) L Ci c-ci

Description

AUSTRALIA Patents Act 1990 COMPLETE SPECIFICATION Standard Patent Applicant: MED-EL ELEKTROMEDIZINISCHE GERAETE GMBH Invention Title: ELECTRICAL NERVE STIMULATION BASED ON CHANNEL SPECIFIC SAMPLING SEQUENCES The following statement is a full description of this invention, including the best method for performing it known to me/us: 2 5 ELECTRICAL NERVE STIMULATION BASED ON CHANNEL SPECIFIC SAMPLING SEQUENCES Related Application 10 This application is a divisional application of Australian application no. 2008255249, the disclosure of which is incorporated herein by reference. Most of the disclosure of that application is also included herein, however, reference may be made to the specification of application no. 2008255249 as filed or accepted to gain further understanding of the invention claimed 15 herein. Field of the Invention The present invention relates to electrical nerve stimulation, and more particularly, electrostimulation of the nerve based on channel specific sampling sequences. 20 Background Cochlear implants (inner ear prostheses) are a possibility to help profoundly deaf or severely hearing impaired persons. Unlike conventional hearing aids, which just apply an amplified and modified sound signal, a 25 cochlear implant is based on direct electrical stimulation of the acoustic nerve. The intention of a cochlear implant is to stimulate nervous structures in the inner ear electrically in such a way that hearing impressions most similar to normal hearing are obtained. A cochlear prosthesis essentially consists of two parts, the speech 30 processor and the implanted stimulator. The speech processor contains the power supply (batteries) of the overall system and is used to perform signal processing of the acoustic signal to extract the stimulation parameters. The stimulator generates the stimulation patterns and conducts them to the nervous tissue by means of an electrode array which usually is positioned in 35 the scala tympani in the inner ear. The connection between speech processor 3038176_1 (GHMatters) P44928.AU.4 3 5 and stimulator is established either by means of a radio frequency link (transcutaneous) or by means of a plug in the skin (percutaneous). At present, the most successful stimulation strategy is the so called "continuous-interleaved-sampling strategy" (CIS), as described by Wilson B. S., Finley C. C., Lawson D. T., Wolford R. D., Eddington D. K., Rabinowitz W. 10 M., "Better speech recognition with cochlear implants," Nature, vol. 352, 236 238 (July 1991) [hereinafter Wilson et al., 1991], which is incorporated herein by reference. Signal processing for CIS in the speech processor involves the following steps: (1) splitting up of the audio frequency range into spectral bands by means 15 of a filter bank, (2) envelope detection of each filter output signal, (3) instantaneous nonlinear compression of the envelope signal (map law). According to the tonotopic organization of the cochlea, each stimulation electrode in the scala tympani is associated with a band pass filter 20 of the external filter bank. For stimulation, symmetrical biphasic current pulses are applied. The amplitudes of the stimulation pulses are directly obtained from the compressed envelope signals (step (3) of above). These signals are sampled sequentially, and the stimulation pulses are applied in a strictly non-overlapping sequence. Thus, as a typical CIS-feature, only one 25 stimulation channel is active at one time. The overall stimulation rate is comparatively high. For example, assuming an overall stimulation rate of 18kpps, and using an 12 channel filter bank, the stimulation rate per channel is 1.5kpps. Such a stimulation rate per channel usually is sufficient for adequate temporal representation of the envelope signal. 30 The maximum overall stimulation rate is limited by the minimum phase duration per pulse. The phase duration cannot be chosen arbitrarily short, because the shorter the pulses, the higher the current amplitudes have to be to elicit action potentials in neurons, and current amplitudes are limited for various practical reasons. For an overall stimulation rate of 18kpps, the 35 phase duration is 27ps, which is at the lower limit. 303617a.1 (GHMatters) P44928.AU.4 4 5 Each output of the CIS band pass filters can roughly be regarded as a sinusoid at the center frequency of the band pass filter, which is modulated by the envelope signal. This is due to the quality factor Q ~ 3 of the filters. In case of a voiced speech segment, this envelope is approximately periodic, and the repetition rate is equal to the pitch frequency. 10 In the current CIS-strategy, the envelope signals only are used for further processing, i.e., they contain the entire stimulation information. For each channel, the envelope is represented as a sequence of biphasic pulses at constant repetition rate. As a characteristic feature of CIS, this repetition rate (typically 1.5kpps) is equal for all channels, and there is no relation to the 15 center frequencies of the individual channels. It is intended that the repetition rate is not a temporal cue for the patient, i.e., it should be sufficiently high, so that the patient does not percept tones with a frequency equal to the repetition rate. The repetition rate is usually chosen greater than at twice the bandwidth of the envelope signals (Nyquist theorem). 20 Summary of the Invention In accordance with one aspect of the invention, there is provided a system for activating electrodes, the system including speech processor means including a filter bank for receiving an acoustic representative electrical signal, each 25 filter in the bank of filters associated with one of the electrodes in the multi-channel electrode array; a multi-channel electrode array; and stimulator means for activating electrodes in the electrode array using channel specific sampling sequences, wherein the system associates each electrode in the multi-channel electrode array with a different channel specific sampling sequence having a predefined duration, amplitude, 30 plurality of pulses, and constant pulse rate, the duration and number of pulses in each channel specific sampling sequence derived from the center frequency of the associated band pass filter, and wherein the system applies a weighting factor to each channel specific sampling sequence so as to create a weighted channel specific sampling sequence for each electrode in the electrode array, the weighting factor 35 based, at least in part, on an acoustic representative electrical signal. 30381781 (GHMalters) P44925.AU.4 5 5 In an embodiment, the channel specific sampling sequence pulse amplitude is derived by sampling a signal waveform, the signal waveform different from the acoustic representative electrical signal. In an embodiment, the signal waveform is of a half period of a sinusoid between 0 and 10 71. In an embodiment, the signal waveform is a quarter period of a sinusoid between 0 and 7T/2, such that pulse amplitude distribution monotonically increases. 15 In an embodiment, the channel specific sampling sequence has a pulse rate between 5 10 kpps. In an embodiment, the stimulator means simultaneously activates each electrode using sign-correlated pulses. 20 In an embodiment, wherein the system calculates the amplitudes of the sign-correlated pulses based, at least in part, on parameters of spatial channel interaction, wherein calculating the amplitudes of the sign-correlated pulses includes compensating for geometric overlapping of electric fields from each electrode. 25 In an embodiment, the speech processor means determines the weighting factor for each channel specific sampling sequence. In an embodiment, the system comprises a radio frequency link between the speech 30 processor means and the stimulator means. In an embodiment, the speech processor means derives the weighting factor for each electrode in the multi-channel electrode array from an associated channel filter. 35 In an embodiment, the system comprises a rectifier for rectifying the output of each filter to create a half-wave rectified signal, wherein the speech processor means determines a maximum amplitude of each half-wave in the half-wave rectified signal. 3036178 1 (GHMattelsl P44928.AU.4 6 5 In an embodiment, each half-wave is replaced by the channel specific sampling sequence associated with the electrode, with the amplitude of the maximum pulse within the channel specific sampling sequence associated with the electrode equal to the maximum amplitude. 10 In an embodiment, each filter is a bandpass filter. In an embodiment, the duration of the channel specific sampling sequence is one half of the period of the bandpass filter's center frequency. 15 In an embodiment, the system includes a cochlear implant capable of stimulating the acoustic nerve. Brief Description of the Drawings 20 The foregoing features of the invention will be more readily understood by reference to the following detailed description, taken with reference to the accompanying drawings, in which: Figure 1 shows channel specific sampling sequences (CSSS) for two 6-channel systems utilizing biphasic pulses at 1Okpp/s and phase duration of 25ps 25 a. derived from a sinusoid within [0 7] b. derived from a sinusoid within [0 n/2], amplitudes monotonically increasing Figure 2 shows stimulation with channel specific sampling sequences (CSSS) a. Band pass filter output signal (653Hz - 876Hz) 30 b. Half wave rectified band pass filter output c. Associated CSSS stimulation sequence Figure 3 shows stimulation with channel specific sampling sequences (CSSS) a. Bandpass filter output signal (3457Hz -5500Hz) b. Half wave rectified band pass filter output 35 c. Associated CSSS stimulation sequence Figure 4 shows a comparison of stimulation patterns between CSSS and CIS 3036178_1 (GHMatters) P44928AU.4 7 5 a. Band pass filter output signal (653Hz - 876Hz) b. CSSS stimulation sequence c. CIS stimulation sequence (envelope sampling with single pulses at 1.5 kpps) Figure 5 shows a comparison of stimulation patterns between CSSS and CIS 10 a. Band pass filter output signal (3457Hz - 5500Hz) b. CSSS stimulation sequence c. CIS stimulation sequence (envelope sampling with single pulses at 1.5 kpps) Figure 6 shows estimated potential distributions in the scala tympani (X 15 3.6mm, d = 2.8mm) a. Responses to single channel activation b. Effective Potential Distributions (asterisks for CIS, circles for CSSS) Figure 7 shows a comparison of overall stimulation patterns between CSSS 20 and CIS (electrode distance: d = 2.8mm, space constant: X =3.6 mm) a. 6-channel CSSS b. 6-channel CIS Detailed Description of the Invention 25 A cochlear implant with stimulation patterns containing enhanced temporal information, especially in the low frequency range up to 1kHz, is described. It is known from literature that the neurons are able to track analogue electrical sinusoidals up to about 1kHz. This ability is not exploited in the present CIS strategy, since the sampling rate is too low to represent 30 high frequency envelope waveforms. The stimulation strategy utilized is based on channel specific sampling sequences (CSSS). The basic idea is to a apply a stimulation pattern, where a particular relationship to the center frequencies of the filter channels is preserved, i.e., the center frequencies are represented in the temporal 35 waveforms of the stimulation patterns, and are not fully removed, as in CIS. 3038178_1 (GHMattes) P44928.AU.4 8 5 Each stimulation channel is associated with a particular CSSS, which is a sequence of ultra-high-rate biphasic pulses (typically 5-10kpps). Each CSSS has a distinct length (number of pulses) and distinct amplitude distribution. The length of a CSSS is derived from the center frequency of the associated band pass filter. A CSSS associated with a lower filter channel is longer than a 10 CSSS associated with a higher filter channel. Typically, it is one half of the period of the center frequency. The amplitude distribution can be adjusted to patient specific requirements. For convenience, the amplitude of the maximum biphasic pulse within a CSSS is normalized to one. For illustration, two examples for a 6-channel system are shown. In Fig.1(a), the CSSS's are 15 derived by sampling one half of a period of a sinusoid, whose frequency is equal to the center frequency of the band pass filter (center frequencies at 440Hz, 696Hz, 1103Hz, 1745Hz, 2762Hz, and 4372Hz). Sampling is achieved by means of biphasic pulses at a rate of 10kpps and a phase duration of 25ps. For channels #5 and #6, one half of a period of the center frequencies is too 20 short to give space for more than one stimulation pulse, i.e., the "sequences" consist of only one pulse, respectively. In Fig.1(b), the sequences are derived by sampling one quarter of a sinusoid with a frequency, which is half the center frequency of the band pass filters. These CSSS's have about the same durations as the CSSS's in Fig.1(a), respectively, but the amplitude 25 distribution is monotonically increasing. Such monotonic distributions might be advantageous, because each pulse of the sequence can theoretically stimulate neurons at sites which cannot be reached by its predecessors. This is a pure "geometric" effect, and could possibly result in a broader temporal distribution of the firing pattern of the neurons. 30 An example of a stimulation pattern based on CSSS is depicted in Fig.2 for a voiced speech segment. For reasons of clarity, the influence of spatial channel interaction is neglected here. In addition, and in the following text, the instantaneous non-linear compression is omitted for convenience, however it is realized that such conversion is required for actual stimulation 35 patterns. Fig. 2(a) shows the output of a band pass filter (cut off frequencies 3038178_1 (GHMatters) P44928 AU.4 9 5 at 553Hz and 876Hz). Fig. 2(b) illustrates the half-wave rectified version of the signal. In Fig. 2(c), each half-wave-pulse is replaced by a CSSS, where the amplitude of the maximum pulse within each CSSS is equal to the maximum of the associated half-wave-pulse. Thus, Fig. 3 represents a sequence of weighted and time-shifted CSSS's. The CSSS used for this example is equal to 10 the CSSS in Fig. 1(a) for channel CH2, and for convenience, each biphasic pulse is represented as a single vertical line. An example of a stimulation pattern based on CSSS for a higher frequency channel is shown in Fig. 3 (the input speech segment is the same as for Fig. 2, spatial channel interaction is neglected again). The band pass filter 15 here selects a range between 3475Hz and 5500Hz. With a center frequency of 4273Hz, the period is 229ps, and sampling one half of this period gives space for only one pulse (cf. CSSS as shown in Fig. 1 for channel CH5). Here, the envelope sampling is reduced to a sampling with single pulses at a rate equal to about the center frequency of 4273Hz. 20 In Fig. 4 stimulation sequences of the new approach are directly compared to the corresponding CIS-sequences at 1.5kpps. The CSSS-based sequence in Fig. 4(b) clearly represents the temporal fine structure plus the envelope information of the band pass output shown in Fig. 4(a), whereas the CIS-pattern in Fig. 4(c) is obtained by sampling the envelope, and thus any 25 temporal fine structure is removed. At a stimulation channel at higher frequencies, Figs. 5(b) and (c) are derived by envelope sampling with single pulses. However, in this frequency range, neurons are only able to track the envelope signals, but cannot follow the stimulation frequency itself. The difference between traces 2 and 3 is the sampling rate, which is considerably 30 lower for CIS. For the practical implementation of the new stimulation approach as described above it is necessary to utilize simultaneous stimulation techniques. Interleaved sampling as employed in CIS is impractical here, since this would require stimulation pulses with a phase duration of only few microseconds. 3036178_1 (GHMatters) P44928.AUA 10 5 In the following, the most important mechanisms of channel interaction involved with cochlear implants are summarized. Channel interaction In principle, channel interaction in connection with pulsatile 10 stimulation strategies occurs as a spatial and as a temporal effect. Temporal interaction could be further separated into "physical" and "physiological" interaction. a. Spatial channel interaction Spatial channel interaction means that there is considerable geometric 15 overlapping of electrical fields at the location of the excitable nervous tissue, if different stimulation electrodes (positioned in the scala tympani) are activated. Thus, neglecting temporal channel interaction, the same neurons can be activated if different electrodes are stimulated. Stimulation of a particular electrode against a remote ground electrode (monopolar 20 stimulation) causes an electrical potential within the scala tympani which can roughly be described by two decaying exponentials at both sides of the electrode, and the space constant (in humans) is typically k = 3.6mm, as described by Wilson B. S., Finley C. C., Zerbi M., and Lawson D. T., "Speech processors for auditory prostheses," Seventh Quarterly Progress Report, 25 Feb.1s t through April 30th, 1994, NIH Contract N01-DC-2-2401 [hereinafter Wilson et al., 1994], which is incorporated herein by reference. This type of channel interaction is first of all due to the conductive fluids and tissues surrounding the stimulation electrode array. A similar space constant is also obtained by simulation, if a simple model of a cochlea composed of 30 exclusively ohmic resistors is assumed, as described by Kral A., Hartmann R., Mortazavi D., and Klinke R., "Spatial resolution of cochlear implants: the electrical field and excitation of auditory afferents," Hearing Research 121, pp. 11-28, (1998), which is incorporated herein by reference. This model allows a rough quantitative computation of the electrical potentials within the 35 scala tympani, as well as at the position of excitable neurons. 3038178 1 (GHMatirs) P44928.AU.4 11 5 b. Physical temporal channel interaction Physical temporal channel interaction means that the electrical properties of a stimulation pulse in the nervous tissue are biased by its predecessor, e.g., due to residual charge stored in the tissue and in the membrane capacitances of the neurons. Physical temporal interaction is 10 suppressed to a great extent by using symmetrical, biphasic stimulation pulses. Most of the charge delivered to the excitable tissue during the first phase of a stimulation pulse is removed during the second. However, since the tissue shows some capacitative behavior, some residual charge remains after the end of the stimulation pulse and possibly may bias the subsequent is stimulation pulse. Theoretically, triphasic pulses (with zero net charge) would help to further reduce physical temporal channel interaction. c. Physiological temporal channel interaction Physiological interaction means effects associated with the refractory properties of the neurons. Following Wilson et al, 1994, a recovery function 20 r(t) can be defined as r(t) = 0, for t < a, and r(t)=1 - exp(- t ta ), for t > ta, (1) with an absolute refractory period ta ~ 700ts, and a time constant T ~ 1250ps for the relative refractory period. For example, if two supra 25 threshold stimulation pulses are applied, and the second pulse falls into the absolute refractory period after the first, no additional action potential can be elicited. If the second pulse occurs during the relative refractory period, an enhanced amplitude is necessary to generate an action potential. The influence of physiological temporal interaction on speech 30 understanding is currently investigated at various research centers worldwide. At the moment, it seems that the similarity between neural excitation patterns due to electrical stimulation and natural excitation patterns can be enhanced, if very high stimulation rates are employed (> 3kpps per channel, as described by Matsuoka A. J., "Compound action potentials 3038175_1 (GHMatters) P44928.AU.4 12 5 evoked by electrical pulse trains: effects of stimulus parameters on response patterns," thesis at University of Iowa, (July 1998), which is incorporated herein by reference. High rates may mimic membrane noise (spontaneous activity) and thereby keep different neurons in different refractory states. If this is the case, it can be expected that the ensemble spiking patterns can 10 reflect the envelope of amplitude modulated electrical pulse sequences up to considerably higher frequencies, and thus more temporal information can be provided to the brain. Consideration of spatial channel interaction In CIS strategy, the influence of spatial channel interaction is reduced 15 by employing pulses which are not overlapping in time (interleaved sampling). The conductivity in the scala tympani here leads to a considerable spread and a de-focusing of the electrical field at the site of the excitable tissue. However, an additional effect occurs, if simultaneous stimulation of two or more electrodes against a remote ground electrode is considered. Here 20 the conductivity represents a shunt conductance between active electrodes, which in general results in a temporal mixture of constructive and destructive superposition of electrical fields at the position of the neurons. For example, if two simultaneous stimulation channels produce currents with equal amplitudes, but different signs, most of the current will flow through the 25 shunt conductance and will not reach the intended neurons. This additional effect can be removed, if "sign-correlated" pulses are employed. Sign-correlation here means that the signs of the phases of simultaneous stimulation pulses are equal. This ensures that the sum of the magnitudes of the single stimulation currents is forced to flow into the reference electrode. 30 Thus, at the site of the excitable neurons, only constructive superposition of currents is possible. The injection of a current by means of a single active electrode into the scala tympani causes a particular voltage in the tissue just close to the electrode (measured against the remote reference electrode), and an 35 exponential decay at both sides of the electrode. The space constant typically 3038178_1 (GHMatters) P44928.AU.4 13 5 is k = 3.6mm, as described by Wilson et al, 1994. Assuming a linear and pure ohmic system, the injection of currents in more than one electrode causes a superposition of the potential distributions due to the single currents. The idea here is to modify stimulation currents such that at least the potentials at the position of the electrodes are equal as in the case of single io channel stimulation. Assuming N channels, the single channel (non simultaneous) current amplitudes Xn (n = 1-N) and the amplitudes yn (n = 1-N) for simultaneous channels are related via the following set of linear equations: 'x i ' 'y, X2 Y2 X3 Y3 15 ... =H ... , (2) XN-2 YN-2 XN-I YN-1 XN YN where Matrix H is -d -2d -(N-3)d -(N-2)d -(N-l)d 1 e ... e e e X -d -d -(N-4)d -(N-3)d -(N-2)d e1 ... e X e X e X -2d -d -(N-5)d -(N-4)d -(N-3)d e ... e e e X H =. , (3) -(N-)d (N-)d -N-5d - -2d e e e ... 1 e -(N-2)d -(N-3)d -(N-4)d -d -d e X e e ... I e -(N-I)d -(N-2)d -(N-3)d -2d -d e e e ... e e 20 The coefficients of matrix H reflect spatial channel interaction. A coefficient at row i and column j describes the fraction of the single channel potential caused by electrode #j at the position of electrode #i. 3038178_1 (GHMatters) P44928.AU.4 14 5 For given amplitudes xo, it follows Y2 X2 Y3 X3 ... = H- ... , (4) YN-2 XN-2 YN-1 XN YN XN where H- 1 is the inverse matrix of H. Fortunately, matrix H- 1 in general 10 is a tri-diagonal matrix with non-zero elements only in the main-, the upper and lower neighboring diagonals (see Appendix). An example is shown in Fig.6 for six electrodes (N = 6). The x-axis is normalized to a distance d = 2.8mm between the electrodes, i.e., the electrodes are at positions 1 to 6. A space constant k = 3.6mm is assumed. The y-axis is 15 normalized to the maximum potential of electrode #4 at position 4. Fig.6(a) depicts the single voltage distributions in the scala tympani as responses to single electrode currents at different amplitudes. For CIS, the electrodes are activated sequentially, and thus each of the single potential distribution applies for the duration of a pulse phase. 20 Assuming a pulse repetition rate of 1.5kppulses/s for each channel, the overall time necessary to present all six distributions is 666 ts, which is just about the duration of the absolute refractory period (ta 700ts). This allows the following rough approximation: for CIS, due to physiological channel interaction, the "effective" stimulation pattern is the contour of the single 25 potential distributions, as shown in Fig.6(b), (asterisks). For Fig.6(b) (circles), the amplitudes yn (n = 1-6) for simultaneous stimulation are computed by means of (4). As demanded, the potentials coincide at the electrode positions. Obviously, the peaks obtained by taking the contour of the non-simultaneous potential distributions CIS are more 30 pronounced than with CSSS. Unfortunately, not all amplitude distributions 3036178_1 (GHMatters) P44928.AU.4 15 5 xn > 0 yield solutions yn with positive elements for all n. This is in contradiction to the principle of "sign-correlation", and requires to compute a modified vector y'n, which contains only non-negative elements (see Appendix). For the CSSS system, Fig. 7(a), the envelope sampling sequences for 10 each channel are chosen as shown in Fig.1(a). To obtain the actual stimulation signals for each channel, the spatial channel interaction is taken into account. As expected, the stimulation pattern reflects the temporal fine structure. In particular, the center frequency of channel #2 is represented in the temporal waveform. A so-called "hole-effect" can be observed: if electrode #2 is not 15 active, i.e., if the output of filter channel #2 is negative, then other spectral maxima are not masked (due to spatial channel interaction) and appear in the waveform. The CIS system, Fig. 7(b) is based on an overall sampling rate of 10kpps, resulting in a rate of 1667pps per channel for the 6-channel system. 20 Envelope detection for each channel is achieved with a full-wave rectifier and a low pass filter with a cut off frequency of 400Hz (Butterworth filter of 2nd order), respectively. Obviously, the envelope signal is sampled and presented, but the temporal fine structure is lost. 25 Implementation of a cochlear implant system based on CSSS Although based on highly synchronous stimulation, the CSSS approach is well suited for an implementation in a practical cochlear implant system. The information transfer rate from the speech processor to the implant can be kept comparatively low. An example of a data word for a 30 12-channel CSSS system is shown in Tab.1. One data word consists of 16 bits, including START and STOP bits. The two special bits SPEC1 and SPECO represent a low rate information channel and are used for the initialization of the implant. The implant is permanently supplied with data defining the amplitude distributions and pules repetition 35 rate of the normalized CSSS's, as well as data defining the reference current 3036178_1 (GHMatters) P44928.AU.4 16 5 levels for each channel. These data are stored in particular implant memories. Besides, safety bits (e.g., for cyclic redundancy check (CRC)) are transmitted. Note that for proper operation of the implant, the information defining the normalized CSSS's and the reference current levels theoretically have to be transmitted 10 only once. The four address bits ADD3-ADDO define the channel address, and bits W7-WO the weighting factor of the associated CSSS. The repetition rate of the 15 Tab.1: Data word Bit # Definition 1 START 2 ADD3 3 ADD2 4 ADD1 5 ADDO 6 W7 7 W6 8 W5 9 W4 10 W3 11 W2 12 W1 13 WO 14 SPEC1 15 SPECO 16 STOP 3036178_1 (GHMatters) P44928.AU.4 17 5 CSSS's is comparatively low, especially at the low frequency channels. It is not necessary to transmit the amplitudes of the individual ultra-high rate pulses, since the amplitude distributions are already stored in the implant. Assuming an input analysis range between 350Hz - 5500Hz for a 12 channel system, and a logarithmic spacing of the band pass filter ranges, 10 results in center frequencies 393Hz, 494Hz, 622Hz, 782Hz, 983Hz, 1237Hz, 1556Hz, 1958Hz, 2463Hz, 3098Hz, 3898Hz, and 49036Hz. Thus, the average CSSS-repetition rate is equal to the sum of the center frequencies, i.e., Rcsss = 22386Hz. This is equal to the average data word repetition rate Rdataword. The resulting average overall bit rate is Rbit = 1 6 Rataword ~ 358kbit/s. 15 Thus, a bit rate of 600kbit/s for a practical cochlear implant is sufficient for complete information transfer. However, this is a moderate rate as compared to the case, if each stimulation pulse has to be defined independently. Here, assuming a frame-rate of 10kpps of simultaneous stimulation pulses and a data word of 16bit per pulse, an overall bit rate of 1920kbit/s results. Such a 20 bit rate is almost impossible to realize with a inductive link system at reasonable power consumption. Within the implant, the correction of the amplitudes due to spatial channel interaction has to be performed for each simultaneous stimulation frame. 25 In summary, the CSSS stimulation approach may be summarized as follows. (1) For stimulation, a multichannel electrode array within the scala tympani and a remote ground electrode is used (monopolar electrode configuration). The basic stimulation waveform is a symmetrical, biphasic 30 pulse. (2) Stimulation involves simultaneous activation of electrodes in the scala tympani employing sign-correlated pulses. Sign-correlated means that if two or more pulses occur simultaneously at different electrodes, positive and negative phases are absolute synchronous in time. 3038178_1 (GHMatters) P44925.AU.4 18 5 (3) The amplitudes of the sign-correlated pulses are estimated by taking into account parameters of spatial channel interaction. Assuming that a single electrode causes exponential decays of the potentials at both sides of the electrode allows a computationally efficient calculation of the pulse amplitudes, since a tri-diagonal matrix is involved. 10 (4) Processing of the acoustic signal involves a filter bank for splitting up the audio frequency range (similar to CIS). According to the tonotopic organization of the scala tympani, each band pass filter is associated with a stimulation electrode. (5) Each stimulation channel is associated with a normalized, 15 channel specific sampling sequence (CSSS) of ultra-high-rate pulses. Typically, rates between 5-10kpps are employed. For each channel, the CSSS has different length and different amplitude distribution. The maximum amplitude of a normalized CSSS is one. (6) The length of a CSSS is derived from the center frequency of the 20 associated band pass filter. Typically, it is one half of the period of the center frequency. For example, a band pass center frequency of 500Hz results in a CSSS-length of 1ms comprising 10 pulses. (assuming a ultra-high-rate of 10kpps). (7) The amplitude distribution of a CSSS is chosen for optimum 25 performance with respect to mimicking membrane noise. As many neurons as possible shall be kept in different refractory states. Although various exemplary embodiment of the invention have been disclosed, it should be apparent to those skilled in the art that various changes and modifications can be made which will achieve some of the 30 advantages of the invention without departing from the true scope of the invention. These and other obvious modifications are intended to be covered by the claims that follow. In the claims which follow and in the preceding description of the invention, except where the context requires otherwise due to express 35 language or necessary implication, the word "comprise" or variations such as 3038178_1 (GHMatters) P44928.AU.4 19 5 "comprises" or "comprising" is used in an inclusive sense, i.e. to specify the presence of the stated features but not to preclude the presence or addition of further features in various embodiments of the invention. 10 3038178_1 (GHMalters) P44928.AU.4 20 5 Appendix The matrix product (2) can be regarded as a convolution product of an infinite sequence hn and a sequence yn, with non-zero elements only at positions n = 1, 2, ...N, i.e., 10 xn =hn*yn, (Al) where sequence hn is given by 15 hn =c nun +cC nu__ . (A2) Function un is the unit step, i.e., un = 1 for n > 0, and un = 0 for n < 0. 20 Sequence hn represents an infinite impulse response (IIR) with exponential decays at both sides of the origin (a I <1). The z-transform is given by 1 -1 H(z)= + , (A3) (1 -az ') (1 -a z') 25 which can be expressed as (a+-) H(z)= a (z*'1 -(aC+ )+z-') (A4) 30 Transformation of (Al) into the z-domain yields 3036178_1 (GHMatters) P44928AU.4 21 5 X(z)= H (z)Y(z), (A5) and thus 10 Y(z)= H '(z)X(z). (A6) Inserting (A4) yields 15 Y(z)= 1(z" -(a+ )+ z-')X(z). (a+-) a (A7) The inverse z-transform immediately yields 1 1 20 y,= + (a+-) a (A8) where 8n is the unit impulse, i.e., 6n = 1 for n = 0, and 8n = 0 elsewhere. The first term of the convolution product (A8) is a finite impulse response (FIR). Equation (A8) can be expressed as 25 1 1 y ( + (x,, -(a-+ -)x + -x_ 1 ), (cc+ I)a (A9) which is a set of linear equations. To calculate yn at positions n = 1 and 30 n = N requires to know amplitudes xo and XN+1. Since sequence Yn can have non-zero elements only at positions n = 1, 2, ...N, it follows with (Al) 30381781 (GHMatters) P44928.AU.4 22 5 x 0 =ya+y 2 c2 +...+YN N =(yI+Y2 '" N Cc N-1)._OX (AlO) and similarly 10 XN+1 )la N+y 2 aN-1+"'+YNX YN-1 +Y2 N-2 +"'+YN) aXN (All) Inserting xo and XN+1 in (A9) for n = 1 and n = N allows to write (A9) as 15 matrix equation, and the matrix necessarily has to be identical to the inverse matrix of H: y, '' ' Xi' Y2 X2 Y3 X 3 .. = H ~1 ... , YN-2 XN-2 YN-1 XN-1 YN XN (A12) 20 where matrix H- 1 is a tri-diagonal matrix given by bo -a 0 ... 0 0 0 -a b -a ... 0 0 0 0 -a b ... 0 0 0 H -1 ... ... ... ... ... ... ... 0 0 0 ... b - a 0 0 0 0 ... - a b -- a 0 0 0 ... 0 - a bo (A13) 25 3036178_1 (GHMatters) P44928.AU.4 23 5 with coefficients 1 1 b1 =a 1 1 b= (a +-), and (--a) a 10 (A14) It shall be mentioned that the analysis can simply be expanded to the case, if the infinite sequence hn (A2) is of the form 15 hn = ac nun + p-" u_-_, (A15) i.e., the exponential decays are different for n > 0 and n < 0 ( I a <1, 1 p <1). 20 In the following, it is assumed that for a given vector Xn with xn > 0 for all n (n = 1, 2, ... N), equation (3) yields a vector yn containing negative elements at positions k, i.e., yk < 0. Negative elements mean negative current amplitudes, which are in contradiction to the principle of sign-correlation, and therefore have to be avoided. 25 One method of handling such a case is to compute a new vector y'n, where the elements at the positions k are set to zero, i.e., Y'n=k = 0. This restriction requires a modified input vector X'n. In the proposed method, x'n differs from vector xn only at positions k and remains unchanged elsewhere, i.e., x'n;k = Xn, and x' x . 30 It is claimed that conditions 3038178_1 (GHMatters) P44928.AU.4 24 5 y'fl=k= 0, and X'flk = Xk (A16) yield a definite solution for vector x'n at all positions. To prove this claim for an arbitrary pattern of k, "zero-sequences" of 10 neighboring indices within k of length L are regarded. For convenience, the smallest index within each zero-sequence is designated as starting index kO. For example, for N = 12, and assuming a pattern k = [1, 2, 5, 6, 7, 10], three zero-sequences [1, 2], [5, 6, 7], and [10] with lengths L = 2, 3, and 1 can be identified, and the starting indices are 1, 5, and 10, respectively. A zero 15 sequence of length L = 1 is also designated as "sequence". Two categories have to be distinguished: Category (1): a zero-sequence does not contain indices 1 or N, and Category (2): a zero-sequence contains either index 1 or N. For the example of above, zero-sequence [1, 2] belongs to category (2), 20 zero-sequences [5, 6, 7], and [10] belong to category (1). ad Category (1): here, for a given zero-sequence, neighboring elements with positive yn at the lower- and the upper range of a zero-sequence do exist at positions n = k0-1 and n = kO+L, respectively. For example, for N = 12 and a zero-sequence [5, 6, 7], kO = 5, and L = 3, and thus the neighboring positions 25 aren=4andn=8. Setting y'4 = 0 yields the following set of equations: 0 = -axkOI_ + bx'ko - axkQ+ 0 = - ax' kO +bx'k0. -ax' 0.2 0= -ax kO+L-3 + bx' k+L-2ax kO+L 0= -ax'kO+L-2 +bx'kO.L_ -axkO. 30 (A17) 3038178_1 (GHMatters) P44928.AU.4 25 5 Elements XkO-1 and xkOL, and coefficients a and b are known, and thus for L > 1, (A17) can be written as X kO XkO-1 X'kO+l 0 X'kO+2 0 ... = QL -. a X kO+L-3 0 X'kO+L-2 0 xX'k+L-1 xXkO+L (A18) 10 with matrix square QL b - a 0 ... 0 0 0 -a b -a ... 0 0 0 0 -a b ... 0 0 0 Q L 2 ''' '' .'. .'' .'' ''. ''. 0 o 0 ... b -a 0 0 o 0 ... -a b -a 0 0 0 ... 0 -a b (A19) 15 The number of lines (and rows) of matrix QL is L (L > 1). Note that amplitudes X'k are fully determined by the "neighboring" amplitudes XkO. and XkO+L. In particular, amplitudes X'ko and X'kO+L-1 can be calculated with C(L) d(L) 20 X'k0 =- XkO-l +- XkO+L and a a d(L) c(L) X'kO+L-1 X kO-1 + - XkO+L, a a (A20) 3038178_1 (GHMatters) P44928.AU.4 26 5 where coefficients C(L) and d(L) are the elements at the left- and right upper corner of matrix Q-1, respectively, i.e., at matrix positions (1,1) and (1,L). For each length L, there exists one unique pair of coefficients c(L) and 2 d(L). For L = 1, evaluation of (A17) yields c') = c = a With (A20), the b amplitudes y'ko.1 and y'kO+L can be determined: 10 Yko- =-aX k-2+ bX kO -ax kO = -axkO- 2 +(b-c (L))XkO-I - d (L)XkO+L, and k+L = kO+L-I EXkO+L- aXkO+L+ 15 =-d (L)XkO-I +(b-c )XkO+L -aXk+L+ (A21) Thus, setting the amplitudes y'k = 0 for a zero-sequence results in a modification of the elements in yn only at positions, which are neighboring to 20 the zero-sequence. Note that other elements of yn are not concerned. Equation (A21) can be implemented by means of the following steps: (1) replace coefficients -a, b, and -a in line k0-1 by -a, b+c(L), and - d(4, (2) replace coefficients -a, b, and -a in line kO+L by - d(4, 25 b+c(L), and -a, (3) delete lines and rows with indices k from matrix H-1, and remove elements with indices k from vector xn. ad case (2): if a zero-sequence contains index 1, the modified 30 amplitudes are exponentials up to index L (cf. (A10)), and can be derived from amplitude xL+1: 3038178.1 (GHMallers) P44928.AU.4 27 Xl (LXL+l 5 x 2 CC L-I X+ 5 L L(XX L+l (A22) Regarding the matrix operation, the coefficient b of line L+1 of H-1 has to be replaced by coefficient bo. Then all lines and rows with indices k have to 10 be removed, and the elements of xn with indices k can be ignored. Similarly, if a zero-sequence contains index N, the modified amplitudes are exponentials for indices greater than kO+L and can be derived from amplitude xko-1: 15 X ko QXXk0-l X k0+l CC2 Xko-l xIN tL Xko-ly (A23) Regarding the matrix operation, the coefficient b of line k0-1 of H-1 has 20 to be replaced by coefficient bo. Then all lines and rows with indices k have to be removed, and the elements of Xn with indices k can be ignored. Theoretically, vector y'n can again contain negative elements, but the magnitude of the negative elements are comparatively small. A repetition of 25 the proposed procedure could remove them, but in many cases, it is sufficient to replace the negative elements by zeros and neglect the impact. Based on the analysis of above, the following computational efficient procedure for the consideration of channel interaction in an N channel system 30 can be applied. 3036178_1 (GHMatters) P44928.AU.4 28 5 (1) Compute yn by multiplication of H- 1 and Xn. (2) Select elements yn=k < 0 and set yk' = 0. (3) Modify elements of H- 1 according to (A21), (A22), (A23) (4) Remove all lines and rows of H- 1 with indices k, and remove all 10 elements Xk. (5) Compute elements yn', which are neighboring to zero-sequences. Example: Let the result of the matrix multiplication yn = H- 1 xn (matrix H- 1 defined 15 by coefficients bo, b, and a, for a 12-channel system (N =12) be a vector containing negative elements at positions k = [1, 2, 6, 7, 9, 10, 11]. Then the modified vector y'n is 0 0 y 3 y 4 y 5 0 0 Y'8 0 0 0 rY12 20 (A24) and the unknown elements are computed by 303178_1 (GHMatters) P44928AU.4 29 y'3 bo -a 0 0 0 x3 y'4 -a b -a 0 0 x4 5 y = 0 -a b-c( 2 ) -d(2) 0 x . Y'8 0 0 -d( b-c( 2 ) -c( -d 3 ) X 8

Y

1 12 0 0 0 -d( 3 ) b 0 -c ( x 12 (A25) Note that element y'4 = y4., because position n = 4 is not neighboring to a zero-sequence. Element y's is neighboring to two zero-sequences. Therefore, 10 the corresponding element in the main diagonal is b-c( 2 )-c( 3 ), reflecting the influence of both zero-sequences. Coefficients c( 2 ), d( 2 ) and c( 3 ), d( 3 ) are computed by inverting matrices Q2, and Q3, which themselves only depend on coefficients a and b. 15 It is to be understood that, if any prior art is referred to herein, such reference does not constitute an admission that such prior art forms a part of the common general knowledge in the art, in Australia or any other country. In the claims that follow and in the preceding description of the 20 invention, except where the context requires otherwise due to express language or necessary implication, the word "comprise" or variations such as "comprises" or "comprising" is used in an inclusive sense, i.e. to specify the presence of the stated features but not to preclude the presence or addition of further features in various embodiments of the invention. 25

Claims (16)

1. A system for activating electrodes, the system including speech processor means including a filter bank for receiving an acoustic representative electrical signal, each filter in the bank of filters associated with one of the electrodes in the multi-channel 10 electrode array; a multi-channel electrode array; and stimulator means for activating electrodes in the electrode array using channel specific sampling sequences, wherein the system associates each electrode in the multi-channel electrode array with a different channel specific sampling sequence having a predefined duration, amplitude, plurality of pulses, and constant pulse rate, the duration and number of pulses in each 15 channel specific sampling sequence derived from the center frequency of the associated band pass filter, and wherein the system applies a weighting factor to each channel specific sampling sequence so as to create a weighted channel specific sampling sequence for each electrode in the electrode array, the weighting factor based, at least in part, on an acoustic representative electrical signal. 20

2. The system according to claim 1, wherein the channel specific sampling sequence pulse amplitude is derived by sampling a signal waveform, the signal waveform different from the acoustic representative electrical signal. 25

3. The system according to claim 2, wherein the signal waveform is of a half period of a sinusoid between 0 and 7r.

4. The system according to claim 2, wherein the signal waveform is a quarter period of a sinusoid between 0 and 7/2, such that pulse amplitude distribution monotonically 30 increases.

5. The system according to any one of claims 1 to 4, wherein the channel specific sampling sequence has a pulse rate between 5-10 kpps. 35

6. The system according to any one of claims I to 5, wherein the stimulator means simultaneously activates each electrode using sign-correlated pulses.

7. The system according to claim 6, wherein the system calculates the amplitudes of the sign-correlated pulses based, at least in part, on parameters of spatial channel 31 5 interaction, wherein calculating the amplitudes of the sign-correlated pulses includes compensating for geometric overlapping of electric fields from each electrode.

8. The system according to any one of claims 1 to 7, wherein the speech processor means determines the weighting factor for each channel specific sampling sequence. 10

9. The system according to claim 8, further comprising a radio frequency link between the speech processor means and the stimulator means.

10. The system according to claim 8, wherein the speech processor means derives the 15 weighting factor for each electrode in the multi-channel electrode array from an associated channel filter.

11. The system according to claim 10, further comprising a rectifier for rectifying the output of each filter to create a half-wave rectified signal, wherein the speech 20 processor means determines a maximum amplitude of each half-wave in the half wave rectified signal.

12. The system according to claim 11, wherein each half-wave is replaced by the channel specific sampling sequence associated with the electrode, with the amplitude 25 of the maximum pulse within the channel specific sampling sequence associated with the electrode equal to the maximum amplitude.

13. The system according to any one of claims 10 to 12, wherein each filter is a bandpass filter. 30

14. The system according to any one of claims I to 13, wherein the duration of the channel specific sampling sequence is one half of the period of the bandpass filter's center frequency. 35

15. The system according to any one of claims I to 14, wherein the system includes a cochlear implant capable of stimulating the acoustic nerve.

16. A system for activating electrodes substantially as described herein with reference to the accompanying drawings.