5The Processor Must Do What the Cochlea No Longer Can

FA compression the implant can no longer borrow

In a healthy ear the active outer-hair-cell mechanism amplifies soft sounds and progressively withholds gain as level rises, producing an instantaneous compression with a ratio of roughly five to one at the basilar membrane. This is what lets a single nerve fibre report on a sound field spanning a hundred decibels. The implant electrode delivers current directly to the nerve and inherits none of this; the cochlea’s compression is simply absent.

The mismatch is stark. Everyday acoustic environments and speech itself can range over thirty to sixty decibels of useful level, set inside a hundred-decibel ambient span, yet electric hearing offers only a few decibels of current between threshold and comfort. Something must compress the input before it reaches the electrode, and in an implant that job falls entirely to the sound processor.[2020]

TTwo stages: gain control then instantaneous mapping

Processors solve the problem in two stages. First, an automatic gain control adjusts overall sensitivity with a fast attack and a slower release, so that soft, loud, near and distant sources are brought into a roughly fixed input window before any channel-by-channel processing. The chosen input dynamic range, often several tens of decibels, decides which acoustic levels are represented at all; sounds below it fall under threshold and sounds above it saturate.

Second, within each channel an instantaneous, memoryless mapping converts the extracted envelope amplitude into a current between the threshold and upper levels. Because the input span is large and the electric span tiny, this mapping is strongly compressive, typically logarithmic so that equal acoustic decibel steps produce roughly equal loudness steps. Manufacturers parameterise its curvature with a loudness-growth function or a Q value; lower Q gives a flatter, more compressed map, higher Q a steeper one.[2013][2020]

CWhy mapping choices shape what is heard

These settings are not cosmetic. A wider input dynamic range makes very soft speech audible but spends precious electric range on it, potentially flattening the contrast between conversational and loud sounds; a narrower range sharpens contrast but can drop soft consonants below threshold. AGC release time interacts with background noise, because a slow release after a loud transient can momentarily suppress the speech that follows, reducing intelligibility in a fluctuating masker.

The loudness-growth curvature likewise trades audibility against comfort. Strong compression raises soft sounds but compresses the loudness cues that mark stress and emphasis; weaker compression preserves those cues but may make loud sounds uncomfortable on the steep electric slope. Clinicians therefore tune input range, AGC behaviour and map curvature together, guided by the patient’s own reports of soft-speech audibility and loud-sound comfort.[2019][2025]

CReading a poor outcome through the mapping

Many disappointing maps are really compression problems in disguise. A patient who hears speech in quiet but loses it the moment a room becomes noisy may have an AGC and input-range combination that lets the noise floor consume the audible electric range. A patient who reports that quiet voices vanish may have too narrow an input range or too little low-level gain, while one who finds everything either silent or blaring may have a map that is too steep for the steep electric loudness function beneath it.

The remedy is to think of the processor as a prosthetic cochlea: it is the only place left to restore the compression that biology has lost. Adjusting input dynamic range, gain at low levels, AGC release and the loudness-growth or Q parameter is, in effect, redesigning that compression to fit one nerve, one survival pattern and one listening life.[2025][2019]