From sound to interleaved pulses
The workhorse strategy is Continuous Interleaved Sampling (CIS). Sound passes through a bank of bandpass filters, one per electrode. From each band the slow envelope is extracted, logarithmically compressed onto the user's electric dynamic range (threshold to comfort level), and used to amplitude-modulate a train of charge-balanced biphasic pulses. Crucially the electrodes fire non-simultaneously, in an interleaved sequence, so their electric fields never sum at the same instant. n-of-m and ACE strategies add a peak-picking step — of the m analysis bands, only the n with the most energy are stimulated each cycle — which raises the effective rate on the bands that matter and reduces overlap; most current clinical processors use them.
Any narrowband signal factors into a slow envelope E(t) and a fast temporal fine structure. Standard CIS transmits only E(t); the phase term is discarded.
Why 22 electrodes are not 22 channels
The array sits in the fluid-filled scala tympani, millimetres from the neurons it targets. A current injected at one contact spreads through this conductive bath. For an idealized point source in a homogeneous medium the extracellular potential falls off only as 1/r — so a single electrode excites a broad, overlapping swath of the tonotopic map.
Extracellular potential from a point current source I in a medium of conductivity σ. The slow 1/r decay is the physical root of current spread and channel interaction.
Because neighbouring electrodes recruit overlapping neural populations, they are not independent. Psychophysical and spectral-ripple estimates repeatedly put the number of truly independent channels at only about four to eight, regardless of whether the array has 12 or 22 physical contacts. A crude but instructive model treats overlap as an inter-channel correlation ρ.
Effective independent channels given N physical electrodes and excitation overlap ρ (illustrative). As ρ rises, adding electrodes yields sharply diminishing returns.
This is why the field invests so heavily in current focusing and steering: tripolar and phased-array configurations that sculpt the field to be narrower, and simultaneous weighted stimulation of electrode pairs to create 'virtual channels' between physical contacts. The gains are real but modest — you are fighting 1/r.
What the envelope leaves behind
Standard CIS deliberately discards temporal fine structure — the fast phase term \cos\phi(t) above. In quiet, the envelope carries most of the information needed for speech, which is why quiet sentences work so well. But fine structure carries pitch, the cues that let you follow one voice in a crowd, and the melodic content of music. Discard it and you have explained, in one line, why music and speech-in-noise are hard.