Difference between revisions of "Cochlear Implants"

Latest revision as of 12:01, 19 August 2024

Introduction

A hearing aid with a cochlear implants translates incoming sound into electrical signals that are directly stimulating the nerves in the cochlea. The process of converting the sound waves into electrical current is done by a sound processor outside of the ear. The steps are the following:

Microphone picks up sound
ADC converts analog signal to digital signal
Pre-filtering is applied for speech emphasis
Automatic Gain Control (AGC) is applied
Signal is split into frequency bands by a filter bank
Envelopes are calculated per band
Signal amplitudes per band are mapped to current amplitudes
Current amplitudes per band are convoluted with spiking patterns
Current steering distributes the current of a single band over multiple electrodes
The resulting signals sent to the electrodes

Microphone

% todo sensitivity spectral response

Analog to digital conversion

% todo Sampling rate ADC bit depth Signal representation

Pre-emphasis filter

The pre-emphasis filter is designed to enhance speech recognition by attenuating lower frequencies in sound, which are less critical for understanding speech. This attenuation helps emphasize higher frequencies, where important speech information, like consonant sounds, is more prominent.

As a consequence, the overall loudness of the sound, measured in decibels (dB), is typically reduced after filtering. The degree of loudness reduction depends on both the filter's characteristics and the spectral content of the input sound.

For example, when pink noise is processed through the pre-emphasis filter, as applied in Advanced Bionics (AB) devices, the loudness is reduced by approximately 10 dB. However, when speech is processed, the reduction in loudness is less pronounced due to the different spectral content of speech, which has more energy in the higher frequency bands where the filter has less impact.

Automatic Gain Control

Automatic Gain Control (AGC) in cochlear implants serves two primary purposes:

It maintains speech levels near the most comfortable listening level for the CI user.
It rapidly reduces the gain when very loud sounds are detected to prevent discomfort.

The AGC achieves these goals through two processes:

Averagers (fast and slow): These use buffers to average sound levels over specific time windows. The fast averager responds quickly to sudden changes in sound level, while the slow averager handles more gradual changes. When the average level exceeds a predetermined threshold, often called the kneepoint, the averagers calculate the excess loudness and triggers the compression system.
Compression: When excess loudness is detected by the averagers, the compression system reduces the gain according to a pre-defined compression function. This function usually applies more compression to louder sounds, effectively narrowing the dynamic range in the output signal and ensuring that softer sounds remain audible while protecting the user from loud sounds.

See Dynamic range compression (wikipedia) for more background info on compression.

Band filtering

After the emphasis filtering and gain control, the signal is split into several frequency bands. The middle of each band corresponds roughly to the frequencies of the electrodes in the cochlea. When 16 electrodes are use, there will be 16 frequency bands.

Envelopes

After the signal is split into frequency bands, for each band an envelope function is calculated. See: Envelopes (Wikipedia)

Methods for calculating envelopes:

Moving window rms.
Rectification and Smoothing.
Hilbert Transform. See: Hilbert transform (Wikipedia)

The mapping function

The mapping function (also called map law) translates the envelope amplitudes for each channel to electrical signal amplitudes.

Types of mapping functions

Al CI mapping functions can be written in the form I = K * Y(X) + C1

I is the electrical output in current units.
X is the envelope signal
K is a proportionality factor or slope (sometimes incorrectly called 'gain').
Y(X) is some function of X that determines the mapping type. Different manufacturers use different functions for Y(X), see table below.
C1 is a constant and determines the offset on the vertical axis.

Manufacturer	Type of Mapping	Y(X)
Advanced Bionics	Logarithmic Function	Y(X) = log(X)
Cochlear	Power Law Function	Y(X) = X^α
MED-EL	Normalized Logarithmic Scaling Function	Y(X) = log(1 + C * X)/log(1 + C)

In the Normalized Logarithmic Scaling Function (NLSF) C is a scaling parameter that determines the curvature of the Y function.

In the limit of C is zero it simplifies into a linear relationship: Y(X) = X and thus I = K * X + C1.
In the limit of C is infinite it simplifies into Y(X) = 1 and thus I = K + C1 which is constant and not very useful.
When C is large but finite it simplifies into Y(X) = 1 + log(Signal)/log(C) and by absorbing 1 * K in C1 and ln(C) in K we get I = K * log(Signal) + C1 which is just a logarithmic mapping function. In practice high values for C are used, in which case the NLSF is very close to the logarithmic mapping type.

In case of the Power Law Function:

For small values of alpha (0<alpha<<1) the Power Law Function can be very close to the Logarithmic Function.
For alpha is one the function becomes a linear function I = K * X + C1.
In the limit of alpha is zero it simplifies to a constant value: I = K + C1. Which again is not useful.

Sometimes different constants are used below and above the kneepoint.

It is useful to replace the formula I = K * Y(X) + C1 by I = K * (Y(X) + C2) + C3, by substituting C1 = K * C2 + C3 , because in practice C2 and C3 are determined by different parameters with different units. E.g:

I  = (M-T)/IDR * (Y(X) - M + IDR + GAIN) + T

where K is replaced by (M-T)/IDR, C2 is by - M + IDR + GAIN and C3 by T

IDR is the input dynamic range,
M is the most comfortable level,
T is the threshold level,
GAIN is the gain of the individual channel.

Terminology of Critical Points in the Mapping Function

In scientific literature, three key levels are recognized: 1. hearing threshold, 2. (most) comfortable level, and 3. pain threshold, each denoted in various ways.

The following abbreviations are commonly used to describe specific levels of electrical stimulation:

T, THR, or THL: These stand for Threshold Level, or Threshold Hearing Level. This is the lowest level of electrical stimulation that the user can perceive.

M, MCL, or C: These abbreviations stand for Most Comfortable Level, Maximum Comfortable Level, or Comfortable Level. This level refers to the point at which sounds are comfortably perceived—neither too loud nor too soft.

USL, MSL, UCL, or LDL: These stand for Upper Stimulus Level, Maximum Stimulation Level, Uncomfortable Loudness Level, or Loudness Discomfort Level. This level indicates the point at which sound becomes painful or uncomfortable for the user.

The terminology can vary depending on the manufacturer. For example, for the comfort level in clinical software:

Advanced Bionics uses 'M-level',
Cochlear uses 'C-level',
Med-El uses 'MCL'.

There can also be some variation in the precise definitions of these terms.

The term "Maximum Comfortable Level" is less commonly used and can sometimes refer to the highest level that is still comfortable, just below the pain threshold. This usage can be confusing and should be avoided where possible.

Spike patterns

%todo

Current Steering