**4. Operation of the cochlear implant system**

The main cochlear implant system functions are shown schematically in **Figure8**. Sound from the microphone is compressed by a single-channel automatic gain control (AGC) system. Compression ratios in CI systems tend to be substantially higher than those in acoustic hearing instruments: six to infinity, compared to two to three respectively. This reflects both the small electrical dynamic range of typically 10 dB [20] and the exponential like increase in loudness found for electrical hearing [21]. Both considerations require tight control of the stimulation current's amplitude to avoid discomfort. Research with different implant types shows a consistent advantage for slow-acting AGC, the benefit being a reduced compression of the information rich temporal modulations of speech [22–24], as well as a reduced co-modulation effect [25] associated with the single channel AGC.

Following AGC, the sound is broken into a number of frequency channels, this number varying between 12 and 24 channels, reflecting the number of intra-cochlear electrode contacts available in the implant model. In **Figure 8** only four channels are used to illustrate the principle. Today a Fast Fourier Transform (FFT) algorithm is often used to separate the incoming sound into discrete frequency channels. The amount of energy in each channel is then estimated by a rectification and low-pass filtering process. While the average energy is calculated over a period of perhaps 10 ms (milliseconds) or more, stimulation pulses will be delivered much more rapidly, typically once every millisecond. Hence calculations will be made that overlap in time in an attempt to follow the changes in speech energy over time. Next the acoustic energy in each channel is mapped to an electrical current amplitude that takes account of the CI user's sensitivity to electrical stimulation. The goal is to use smaller currents that barely produce a perception of electrical stimulation to represent low-intensity acoustic activity and larger currents that are perceived as loud to represent very intense acoustic events. This

#### **Figure 8.**

*A schematic of the sound processor system where sound is collected by the microphone, compressed by an automatic gain control, broken into discrete frequency channels, which have their energy assessed and mapped to the user's requirements. This information is then combined into a digital stream, transmitted by radio frequency to the implant where stimulation currents are generated.*

**93**

*Electrical Stimulation of the Auditory System DOI: http://dx.doi.org/10.5772/intechopen.85285*

needs to be managed separately for each channel, resulting in the continuous output of a stream of stimulation amplitudes for each channel. As shown schematically, these amplitudes are then combined together for transmission by the RF signal across the skin to the implant. Electronics inside the implant extract the digitally transmitted amplitudes, convert them to analogue values and then drive the implant's current source(s), resulting in stimulating currents being delivered by the intra-cochlear electrode contacts. For virtually all of today's clinical systems only one channel will be stimulated at a time. This approach avoids the channel interactions that would occur were channels presented simultaneously within the conductive scala tympani [26]. The disadvantage of this approach can be seen in **Figure 9** where a channel is only updated during its own time period, therefore, must wait until all the other channels have been updated until new information can be transmitted. Deliberately, very brief current pulses each of around 40–50 μs duration (20–25 μs/phase) are used, so that it is still possible to update each channel rapidly enough to keep up with the changes in acoustic energy over time. This often means stimulation at more than 1000 pps/ch. Such an approach generally leads to higher levels of speech understanding than where

*been stimulated no more information may be delivered until the other channels have been updated.*

*An illustration of now non-simultaneous waveforms delivers information for each channel. Once a channel has* 

Over the years the sound coding strategy, a software algorithm that relates audio from the sound processor microphones to the electrical patterns appearing at the electrode contacts, has changed. Initially it was believed that the damaged auditory system was not capable of transmitting much information, hence the most useful information was extracted from the speech and directly coded on sets of electrodes. For example an early feature extraction strategy F0F1F2 [29] extracted the first two formants of speech, F1 and F2, from which it is possible to estimate the vowel being articulated. Each formant range had a set of electrode contacts allocated, such that higher or lower frequencies for each formant lead to stimulation on more basal or more apical electrode contacts in that formant's electrode set. The rate at which pulses were delivered was related to the fundamental frequency (Fo) driving the vocal tract, leading to the

simultaneous stimulation is delivered [27, 28].

**4.1 Sound coding strategies**

**Figure 9.**

*Electrical Stimulation of the Auditory System DOI: http://dx.doi.org/10.5772/intechopen.85285*

#### **Figure 9.**

*The Human Auditory System - Basic Features and Updates on Audiological Diagnosis and Therapy*

of stimulation are interpreted as sound input by the higher levels of the auditory system, leading to the sense of electrical hearing. The next section will describe how sounds detected by the CI system's microphone will result in the generation of

The main cochlear implant system functions are shown schematically in **Figure8**.

Sound from the microphone is compressed by a single-channel automatic gain control (AGC) system. Compression ratios in CI systems tend to be substantially higher than those in acoustic hearing instruments: six to infinity, compared to two to three respectively. This reflects both the small electrical dynamic range of typically 10 dB [20] and the exponential like increase in loudness found for electrical hearing [21]. Both considerations require tight control of the stimulation current's amplitude to avoid discomfort. Research with different implant types shows a consistent advantage for slow-acting AGC, the benefit being a reduced compression of the information rich temporal modulations of speech [22–24], as well as a reduced

Following AGC, the sound is broken into a number of frequency channels, this number varying between 12 and 24 channels, reflecting the number of intra-cochlear electrode contacts available in the implant model. In **Figure 8** only four channels are used to illustrate the principle. Today a Fast Fourier Transform (FFT) algorithm is often used to separate the incoming sound into discrete frequency channels. The amount of energy in each channel is then estimated by a rectification and low-pass filtering process. While the average energy is calculated over a period of perhaps 10 ms (milliseconds) or more, stimulation pulses will be delivered much more rapidly, typically once every millisecond. Hence calculations will be made that overlap in time in an attempt to follow the changes in speech energy over time. Next the acoustic energy in each channel is mapped to an electrical current amplitude that takes account of the CI user's sensitivity to electrical stimulation. The goal is to use smaller currents that barely produce a perception of electrical stimulation to represent low-intensity acoustic activity and larger currents that are perceived as loud to represent very intense acoustic events. This

*A schematic of the sound processor system where sound is collected by the microphone, compressed by an automatic gain control, broken into discrete frequency channels, which have their energy assessed and mapped to the user's requirements. This information is then combined into a digital stream, transmitted by radio* 

*frequency to the implant where stimulation currents are generated.*

co-modulation effect [25] associated with the single channel AGC.

electrical stimulation patterns.

**4. Operation of the cochlear implant system**

**92**

**Figure 8.**

*An illustration of now non-simultaneous waveforms delivers information for each channel. Once a channel has been stimulated no more information may be delivered until the other channels have been updated.*

needs to be managed separately for each channel, resulting in the continuous output of a stream of stimulation amplitudes for each channel. As shown schematically, these amplitudes are then combined together for transmission by the RF signal across the skin to the implant. Electronics inside the implant extract the digitally transmitted amplitudes, convert them to analogue values and then drive the implant's current source(s), resulting in stimulating currents being delivered by the intra-cochlear electrode contacts. For virtually all of today's clinical systems only one channel will be stimulated at a time. This approach avoids the channel interactions that would occur were channels presented simultaneously within the conductive scala tympani [26]. The disadvantage of this approach can be seen in **Figure 9** where a channel is only updated during its own time period, therefore, must wait until all the other channels have been updated until new information can be transmitted. Deliberately, very brief current pulses each of around 40–50 μs duration (20–25 μs/phase) are used, so that it is still possible to update each channel rapidly enough to keep up with the changes in acoustic energy over time. This often means stimulation at more than 1000 pps/ch. Such an approach generally leads to higher levels of speech understanding than where simultaneous stimulation is delivered [27, 28].

#### **4.1 Sound coding strategies**

Over the years the sound coding strategy, a software algorithm that relates audio from the sound processor microphones to the electrical patterns appearing at the electrode contacts, has changed. Initially it was believed that the damaged auditory system was not capable of transmitting much information, hence the most useful information was extracted from the speech and directly coded on sets of electrodes. For example an early feature extraction strategy F0F1F2 [29] extracted the first two formants of speech, F1 and F2, from which it is possible to estimate the vowel being articulated. Each formant range had a set of electrode contacts allocated, such that higher or lower frequencies for each formant lead to stimulation on more basal or more apical electrode contacts in that formant's electrode set. The rate at which pulses were delivered was related to the fundamental frequency (Fo) driving the vocal tract, leading to the

strategies name. Such a strategy supported only very modest levels of speech understanding, around 8% correct for monosyllabic words presented in quiet [30]. The information extracted was limited to begin with and further reduced through errors generated in real life listening situations where background noise, reverberation and intensity and frequency response variations led to the algorithm making mistakes in both the extraction of formant frequencies and in the estimation of Fo.

It was eventually recognized that the brain was better at extracting information than the feature extraction algorithms and hence "whole-speech strategies" replaced feature extraction. Today's sound coding strategies simply average the energy in each channel's frequency range and generate levels of stimulation that represent this. In some cases a so-called n-or-m strategy will work out which subset (n) channels from the total (m) number available have the highest energy and then only stimulate this reduced set. Refinements to this may neglect adjacent channels on the basis that stimulating both will not add anything, so select a more distant lower amplitude electrode to transmit more information [31].
