11Front-end pre-processing
Not all of the implant's intelligence is in the coding strategy. A great deal of it sits earlier, in the front end — the stages between the microphone and the filter bank that condition the signal before any channel is chosen. Automatic gain control fits the enormous range of real-world sound levels into the processor's modest input window; directional microphones and noise-reduction algorithms raise the signal-to-noise ratio of what reaches the coder. This matters more than it might seem, because once speech is buried in noise no downstream cleverness can dig it back out — the information is simply gone. So the front end is where much of the modern progress in hearing-in-noise has been won, and where deep learning is now making its first big difference. This module covers the pre-processing that frames everything else.
TThe work done before coding
The coding strategy can only ever work with the signal the front end hands it. If that signal is clipped, too quiet, or swamped by noise, no filter bank or stimulation strategy can repair it. So the pre-processing stages — gain control, directionality, noise reduction — are not a footnote to coding; they are half the battle, and often the more decisive half in real listening environments.
CAutomatic gain control
Natural sound spans an enormous range of levels, but the implant's electric dynamic range — between threshold and comfort — is tiny by comparison. Automatic gain control (AGC) continuously adjusts the input gain so that both a whisper and a shout map into that narrow window without the soft sounds vanishing or the loud ones becoming uncomfortable. Good AGC is essential simply to make everyday sound usable, before any coding begins.
CDirectionality & noise reduction
To help in noise, processors use directional microphones (often adaptive) that favour sound arriving from the front, where a conversation partner usually is, and attenuate sound from the sides and back. Noise-reduction algorithms further suppress steady background noise, and wind-noise managers handle the outdoors. These front-end gains are among the most reliable real-world benefits an implant offers, and they stack with the coding strategy rather than replacing it.[2008]
CWhy the front end is such a lever
The leverage is a direct consequence of channel interaction (Module 7): because the implant resolves so little spectral detail, it cannot use the fine cues a normal-hearing listener exploits to separate a voice from noise. So the most effective place to fight noise is before the signal is degraded — raising the signal-to-noise ratio at the microphone. A few decibels gained here typically outweigh most changes to the coding strategy, which is why deep-learning noise reduction on the front end (Module 12) is one of the most promising near-term advances.[2001]
Why did a front-end change help so much?
What does automatic gain control (AGC) do in the front end?
Why is the front end such a powerful lever for hearing in noise?