13The Present Commercial Landscape: Four Makers, Five Philosophies

TCochlear: ACE and the psychoacoustic PACE/MP3000

Cochlear's default ACE selects 8-12 amplitude maxima of 10-12 (or up to 22) bands at a default 900 pps using CIS-principle interleaved pulses PACE / MP3000 operates like ACE (n-of-m) but replaces simple energy-maxima selection with a psychoacoustic masking model, stimulating the most perceptually relevant channels With PACE, reducing the selected bands from 8 to 4 could maintain speech-understanding performance ACE is comparable to CIS and superior to SPEAK in challenging conditions.[2005][2006]

CMED-EL and Advanced Bionics

MED-EL processors run a 12-channel implant with 12 logarithmically spaced band-pass filters and can generate up to 18,180 pulses/sec aggregate (about 1515 pps per channel in the 12-channel CIS implementation), supporting simultaneous 12-electrode stimulation and a high-rate spectral-maxima option MED-EL's FS4 adds apical temporal-fine-structure coding on the four most apical channels at roughly 1500-2000 pps Advanced Bionics' Clarion CII/HiRes uses 16 electrodes with half-wave rectification and no low-pass filter at about 2800 pps Advanced Bionics HiRes Fidelity 120 adds current steering for up to 120 virtual channels at 1800-2900 pps per electrode.[2006][2014]

COticon Medical Crystalis and Nurotron APS/Symphony

Oticon Medical's Crystalis uses an FFT filter bank regrouped into 20 channels with N-of-M selection (default N=8) but modulates pulse WIDTH (with fixed pulse height) rather than pulse amplitude, at an adjustable rate of 250-1000 pps and a default of 500 pps, using a distributed all-polar return Nurotron's 26-electrode system (24-channel intracochlear array) runs APS, an N-of-M peak-selection strategy with 6-8 active stimulation channels, integrating the c-tone algorithm Nurotron's Symphony adds current steering to create virtual intermediate channels with peak picking Both Nurotron strategies aim to represent temporal fine structure.[2015][2017]

CWhy incremental gains have plateaued

Despite differing strategies and electrodes, present commercial devices show no significant overall performance difference Increasing electrodes beyond about 4-8 effective channels, current steering and fine-structure coding have yielded only modest or unproven speech-in-noise gains Current spread, channel interaction, the ~300 Hz electric temporal-pitch ceiling and variable neural survival fundamentally cap information transfer Strategies based on spectral signal analysis outperformed explicit feature extraction, which is why all present strategies share the CIS/n-of-m core.[2004][2006]

TThe current commercial landscape and its honest evidence base

Modern devices pair high-rate envelope coding with current steering: HiRes 120 (Fidelity120) creates intermediate virtual channels between physical electrodes (Firszt 2009). HiRes 120 produced significant gains on CNC words (57%->63%, p=.023), HINT-in-noise (61%->69%, p=.043), and BKB-SIN (10.2->9.2 dB, p=.037), but not HINT-in-quiet, and 7/8 preferred it (Firszt 2009). Fidelity120 improves spectral-ripple resolution (3.42 vs 2.31 ripples/octave) but worsens a temporal fine-structure cue, all 9 listeners falling to chance on Schroeder-phase (Drennan 2010). A skeptical within-subjects study found no overall strategy effect on speech (small HiRes edge for consonants); only 3/10 improved any spectral cue with F120 (Donaldson 2011). In the MED-EL line, fine-structure FSP offers telephone and music benefits but is statistically equivalent to envelope HDCIS on standard speech tests in independent data (Dillon 2016; Magnusson 2011).[2009][2010]

TBy the numbers