A design tool for cochlear implants, the surgically implanted auditory prostheses that electrically stimulate the auditory nerve in people with severe-to-profound sensorineural hearing loss. Adjust electrode count, pulse rate, coding strategy and insertion angle to see how sentence intelligibility in quiet and noise, channel bandwidth and cochlear coverage respond — and benchmark different fitting strategies.
Parameters
Active electrodes N
ch
Electrodes actually used for stimulation in the patient's MAP
Pulse rate
Hz/ch
Biphasic pulses per second on each electrode
Stimulation level
dB SPL
Coding strategy
How acoustic input is mapped to electrode pulses
Insertion angle
°
Angle of the cochlea (540°) covered by the array
Lower frequency
Hz
Upper frequency
Hz
Results
—
Active electrodes (ch)
—
Total stim rate (pps)
—
Channel bandwidth (Hz)
—
Sentence acc. in quiet (%)
—
Sentence acc. in noise (%)
—
Cochlear coverage (%)
—
Cochlea cross-section — electrode array & frequency map
The 540° spiral cochlea with the inserted electrode array. Each electrode covers a frequency band via the Greenwood place-frequency function; pulses are interleaved in time according to the coding strategy.
Sentence intelligibility vs. channel count (quiet & noise)
f: Greenwood place-frequency function (Hz); d: distance from the base (mm); L: total cochlear length (~35 mm); N: independent spectral channels. Friesen & Shannon (2001) found sentence intelligibility plateaus around 70% at N=8 and 90% at N=16.
Total stimulation rate (pps), channel bandwidth and cochlear coverage. Note the electrical dynamic range of a CI is only about 15 dB, compared with 120 dB for normal acoustic hearing.
Cochlear implant channel design and speech intelligibility
🙋
I've heard cochlear implants are completely different from hearing aids. What's actually going on inside one?
🎓
Exactly — a hearing aid is basically a smart amplifier that pushes louder sound onto the eardrum, but a cochlear implant (CI) is for people whose hair cells no longer work, so it stimulates the auditory nerve electrically. A receiver is placed under the skin behind the ear, and a silicone electrode array with 12-22 contacts is threaded into the cochlea, that snail-shaped fluid-filled organ in the inner ear. An external sound processor analyses sound, splits it by frequency, and routes pulses to the electrode at the matching cochlear position. The cochlea has tonotopy — the base codes high frequencies, the apex codes low — and it's described nicely by Greenwood's f=165.4(10^(2.1d/L)−1). That's why the tool's frequency range is set to 250-8000 Hz by default.
🙋
So more electrodes should always be better, right? But when I push N up to 32 the intelligibility flattens out instead of climbing to 100%.
🎓
Welcome to the fundamental puzzle of CI design. In 2001 Friesen & Shannon published a landmark study showing sentence intelligibility follows a 1−exp(−N/4) shape — about 70% at N=8 and 90% at N=16, then it plateaus. The reason is "channel interaction": current from neighbouring electrodes spreads through the conductive perilymph and overlaps, so the effective number of independent channels caps out around 8-10 even with 22 physical contacts. This is why manufacturers compete on coding strategy (CIS/ACE/HiRes) and insertion depth rather than raw electrode count.
🙋
Switching the coding strategy dropdown changes the score a bit. What's really different between CIS, ACE and HiRes?
🎓
CIS (Continuous Interleaved Sampling) was Wilson and colleagues' 1991 breakthrough at Research Triangle Park — stimulating electrodes one at a time in non-overlapping time slots, which kills the cross-electrode current interaction. Nearly every modern CI still rests on CIS. ACE (Advanced Combination Encoders) is Cochlear's n-of-m strategy: in each frame, only the n loudest electrodes fire, which boosts SNR in noisy environments. HiRes Fidelity 120 is Advanced Bionics' "virtual channels" trick: stimulating pairs of electrodes simultaneously with current steering creates intermediate spectral percepts, giving up to 120 bands. This tool applies empirical multipliers CIS=1.0, ACE=1.10, HiRes=1.15.
🙋
When I drop the insertion angle to 270° the coverage falls to 50% and the verdict warns me. How do surgeons actually get the array deeper?
🎓
Sharp eye. The cochlea is about 2.5 turns = 540°. Shallow insertion misses the apex, so the <500 Hz vowel and pitch information is gone — patients complain that music sounds metallic and male voices are hard to follow. To get deeper, surgeons use soft "perimodiolar" arrays like Cochlear's Contour Advance, or long straight arrays like MED-EL's 31.5 mm Standard. Recent best practice adds intraoperative fluoroscopy to track the tip, and robotic insertion at 0.1 mm/s to preserve any remaining residual hearing. Around 450° (83% coverage) is the modern clinical target.
🙋
Intelligibility in noise is ~20% lower than in quiet. How does that play out in real life?
🎓
This is the biggest day-to-day complaint of CI users. A CI throws away the temporal fine structure of sound and transmits only the envelope of each channel. Normal listeners use interaural phase and pitch cues to perform the cocktail-party trick of separating one voice from a crowd — CI users lose those cues. A patient who scores 95% on sentences in a quiet clinic often drops to 50-70% in a restaurant at ~10 dB SNR. It's worse for tonal languages like Mandarin, where pitch IS the meaning, and CI listeners typically achieve less than half the tone-recognition accuracy of normal-hearing listeners. Bilateral implantation and remote FM systems are the main mitigations today.
FAQ
Friesen & Shannon (2001) showed that sentence intelligibility follows P_sentence ≈ 1 − exp(−N/4), reaching about 70% at N=8 and 90% at N=16 before plateauing. Even with 22 physical electrodes, overlapping current fields (channel interaction) typically limit the effective number of independent channels to 8-10. This tool reproduces that saturation behaviour.
CIS (Continuous Interleaved Sampling, Wilson 1991) is the foundational scheme that stimulates electrodes sequentially in non-overlapping time slots to minimise electrical interference. ACE (Advanced Combination Encoders) is Cochlear's n-of-m strategy that fires only the n highest-amplitude electrodes per frame, improving noise robustness. HiRes Fidelity 120 (Advanced Bionics) uses virtual channels by simultaneously stimulating pairs to create up to 120 spectral bands. This tool applies empirical efficiency multipliers CIS=1.0, ACE=1.10, HiRes=1.15 to the baseline intelligibility.
The cochlea is a ~540° spiral; a deeper insertion reaches the apical (low-frequency) region. With only 360° insertion, electrodes stimulate basal and middle turns only, dropping the <500 Hz vowel cues and producing complaints like 'tinny music' or 'male voices hard to distinguish'. This tool computes coverage = angle/540° and recommends 450° or more (83% coverage). MED-EL's 31.5 mm Standard array is a typical deep-insertion design.
Cochlear implants discard the temporal fine structure and transmit only the envelope of each channel. Normal-hearing listeners use interaural phase differences and fundamental-frequency cues to segregate speech from noise; CI users lose these. At 10 dB SNR, intelligibility typically falls to about 0.8× the quiet score, and the gap is even worse for tonal languages (Mandarin) and music. This tool approximates the noise score as 80% of the quiet score.
Real-world applications
Clinical fitting (MAPping): Audiologists personalise each CI MAP by sweeping T-levels (the just-audible current) and C/M-levels (the maximum-comfortable current) on every active electrode. The channel-count, pulse-rate and coding-strategy trade-offs explored in this tool are exactly what is iterated inside clinical software like SoundWave (AB), Custom Sound (Cochlear) and Maestro (MED-EL) during the 2-3 years of follow-up after implantation.
Newborn screening and early implantation: Niparko et al. (2010) showed that children implanted before 12 months of age can reach age-appropriate spoken-language development. This drove the worldwide adoption of newborn hearing screening with ABR/OAE; in Japan around 90% of newborns are tested, and bilateral profound losses are typically implanted around their first birthday. This tool is helpful for explaining long-term outcomes to families considering early surgery.
Bilateral and bimodal hearing: Compared to a single CI, bilateral implantation improves speech-in-noise scores by 15-25% and restores binaural cues (ITD and ILD) for source localisation. When useful residual hearing remains in the opposite ear, a bimodal fitting (CI + hearing aid) combines electric and acoustic stimulation — particularly effective for low-frequency music perception, where the acoustic ear adds the missing temporal fine structure.
Research frontiers: Optogenetic cochlear implants (Moser 2020) replace electrical pulses with light pulses delivered through micro-optical fibres, promising orders-of-magnitude better spatial selectivity. AI-based noise reduction such as Forward-Focus and DeepFilter is now appearing in commercial processors, expanding the menu of "coding strategy" choices that tools like this one need to model.
Common misconceptions and pitfalls
The biggest misconception is that more electrodes always equal more natural sound. As the Friesen-Shannon curve shows, modern transcutaneous CIs saturate at an effective 8-10 channels. Many studies report no significant difference in speech-in-noise scores between Cochlear Nucleus (22 electrodes) and AB HiRes (16 electrodes). The real levers are insertion depth, residual-hearing preservation, coding strategy and bilateral use — electrode count is secondary. This tool also shows almost identical intelligibility at N=16 and N=22.
Next: assuming Greenwood mapping is gospel. The Greenwood function describes the place-frequency map of a normal-hearing cochlea, but in a CI patient the actual perceived pitches shift because of insertion position, neuron survival and central reorganisation. Especially in children, perceptual learning over ~6 months rewires the cortex to a new frequency map, so initial reports of "Mickey Mouse voices" are common. Treat the frequency map in this tool as a starting point and let post-operative MAP adjustments accommodate this shift.
Finally, higher pulse rate is not always better for temporal resolution. Pushing the rate from 500 to 1500 pps improves envelope tracking, but above ~3000 pps each pulse falls inside the neural refractory period (~1 ms), and action potentials fire only stochastically. Vandali (2000) reported optimum performance in the 800-1200 pps range, which matches the modern default of ~900 pps used in this tool. Higher rates also drain the battery dramatically, so there is no reason to push them blindly.
How to Use
Set Active Channels (4–22 electrodes) to define the number of stimulation sites along the cochlear array; more channels improve spectral resolution but increase power consumption.
Configure Pulse Rate (500–5000 pps) to control temporal resolution; rates above 1200 pps typically plateau in speech perception gains for most users.
Adjust Stimulation Level (0–50 dB above threshold) to set loudness comfort; verify cochlear coverage percentage to ensure adequate insertion depth and electrode positioning.
Set Insertion Angle (0–360 degrees) to model array orientation; angles 360–540° indicate full or over-inserted arrays affecting apical frequency mapping.
Observe Channel Bandwidth output (typically 250–500 Hz per channel) and sentence recognition scores to validate coding strategy effectiveness.
Worked Example
A patient receives a 16-channel implant with insertion angle 450° (full insertion), pulse rate 1800 pps, and stimulation level 35 dB. Simulator outputs: channel bandwidth 312 Hz (4000 Hz ÷ 16 channels minus guard band), cochlear coverage 98%, sentence accuracy in quiet 72%, and in noise (SNR +10 dB) 48%. Increasing to 20 active channels reduces bandwidth to 187 Hz, raising quiet accuracy to 78% but decreasing battery life by approximately 12%, requiring patient trade-off counseling.
Practical Notes
Full insertion (450–540°) extends apical coverage but risks electrode-modiolus distance variability; simulate multiple angles to optimize frequency allocation for patient's residual hearing profile.
Pulse rates above 2400 pps show diminishing returns in open-set word recognition; prioritize channel count or stimulation level adjustments instead to conserve battery life in daily use.
Channel bandwidth below 150 Hz degrades consonant discrimination; if active channels exceed 24 on newer devices, verify manufacturing specifications to avoid computational aliasing errors.
Sentence accuracy in noise typically drops 20–30 percentage points from quiet baseline; validate simulator predictions against clinical measures (HINT, AzBio) before device programming.