4Continuous Interleaved Sampling: The Pulse That Changed Everything

TThe CIS signal chain

The chain is: input to pre-emphasis filter to a bank of band-pass filters to per-band envelope detection (full-wave rectifier plus low-pass filter, cutoff ~200-400 Hz) to nonlinear logarithmic compression to amplitude-modulation of interleaved biphasic pulse trains to non-simultaneous pulses on electrodes EL-1 to EL-n Envelope detection is full-wave rectification followed by a low-pass filter with cutoff typically 200-400 Hz, encompassing voicing fundamental and stop-consonant transitions Pre-emphasis attenuates components below about 1.2 kHz at 6 dB/octave so weak high-frequency consonants can compete with intense low-frequency vowels CIS makes no assumptions about speech production or perception and does not vary stimulation rate between voiced and unvoiced sounds.[1999][2004]

CWhy interleaving works

Non-simultaneous interleaved biphasic pulses eliminate the field-summation interaction that limits simultaneous analog strategies CIS was introduced in the landmark Wilson et al. 1991 Nature paper and showed much better speech perception than Compressed Analog Logarithmic compression maps the wide acoustic dynamic range (up to ~100 dB) into the narrow electric dynamic range (~10 dB) and yields near-normal loudness growth An AGC or volume control is needed to shift the ambient acoustic range into the processor's input window.[1991][1999]

CRate and the anti-aliasing rule

Stimulation rate per channel usually exceeds 800 pps and is typically about 1000 pps per electrode or higher The pulse rate must exceed about four to five times the envelope low-pass cutoff to avoid aliasing/distortion of the modulation waveforms (a 'four-times' oversampling rule) Between 4 and 22 channels have been used in CIS implementations; the strategy itself does not limit channel number The envelope cutoff is set at 200 Hz or higher so the voice fundamental F0 is represented in the modulation.[2002][1999]

CWhat CIS leaves on the table

CIS delivers only the band envelopes and discards the temporal fine structure within each band CIS presents a fixed set of waveform-derived channels with no explicit spectral-feature or peak selection The drive to extract perceptually important formants and to select spectral maxima motivated feature-extraction and n-of-m strategies CIS remains widely implemented across MED-EL, Advanced Bionics and Cochlear systems with many variations and is the conceptual basis for later strategies.[1991][2006]

TInside CIS and its envelope-strategy relatives

CIS delivers brief, charge-balanced pulses to each electrode in a strictly non-overlapping (interleaved) sequence, eliminating the channel interaction that limits simultaneous compressed-analog stimulation (Wilson 1993). The original Ineraid CIS research processors used 5-6 channels versus 4 for the standard CA processor; the change in stimulation paradigm, not channel count, drove the gain (Wilson 1993). Per-channel pulse rate was a free parameter ranging roughly 500-2941 pps; high performers tended to use higher rates (~817-2941 pps) versus ~100 pps in the earliest implants (Wilson 1993). Every subject scored higher, or repeated 100% correct, on every test with CIS, including poor CA performers (e.g., SR10 improved 0%->56% on spondees) (Wilson 1993). Later envelope variants (HDCIS, SAS) and the MED-EL fine-structure family all trace to this interleaved-pulse foundation; HDCIS at high rate scored 27.2% on monosyllables in a separate cohort (Riss 2016).[1993][1991]

TBy the numbers

FHear it