ICS

THE PROBLEM

The hidden cost of spatialization

When several speakers carry correlated material with different gains and delays, the signals sum at some frequencies and cancel at others — a repeating pattern of peaks and notches across the spectrum. A 0.5 ms delay difference puts the first notch at 1 kHz with notches every 2 kHz after it; 2 ms moves the first notch down to 250 Hz. Real systems usually create partial notches that hollow out the sound — often subjectively worse than complete cancellation, because the result sounds 'wrong' rather than 'missing'.

The more speakers, the worse it gets. A stereo pan combs mildly; an eight-speaker VBAP system creates eight interfering versions of each source; a WFS array makes an interference pattern that devastates timbre — especially for moving sources, where the comb pattern sweeps continuously.

The usual fixes fail structurally. EQ cannot work because moving a source moves every notch. Delay alignment cannot work because the delays that create the combing are the same delays that create the spatial impression — remove them and the image collapses. Yet the interference is fully predictable: the delays are known, the correlation is measurable, and everything needed to reverse it exists in the digital domain before the signal ever reaches a speaker.

HOW IT WORKS

How it works

UNDER THE HOOD

Going deeper

ICS treats comb filtering as a solved problem rather than an inherent limitation. The spatializer already knows everything that causes the interference: source positions, speaker positions, the gain matrix. From these, ICS derives the exact timing differences between every source and speaker — which determine precisely which frequencies will reinforce and which will cancel — and continuously tracks the correlation between the signals feeding each speaker, weighted by those timing differences.

From the delay and correlation data, ICS derives inverse filters. Where the acoustic combination would notch at 1 kHz, ICS boosts at 1 kHz — but only on the correlated component, and only in proportion to the predicted interference; where it would peak, ICS attenuates complementarily. After acoustic combination, the response is flat. This is not a 'make it sound better' processor — it is interference cancellation from physical principles.

The correction adapts sample by sample: a source moving left to right receives different correction at every point along its path; a sustained note and a percussive hit receive differently weighted treatment. The key insight that makes this robust: the acoustic comb pattern varies with listener position, but the correlation between speaker feeds does not — so correcting in the digital domain works everywhere in the room. And critically, the gains and delays that create the spatial perception pass through unchanged; only the timbral damage they would cause is removed.

With ISE the integration is automatic — positions, gains and correlation flow directly, zero calibration, and the pair behaves as a single spatial engine that simply sounds right. Any other renderer that exposes positions and a gain matrix can drive ICS standalone. Everything is parametric — no FFT, a few samples of latency, embedded-friendly.

AT A GLANCE

At a glance

PARAMETERS

Hands on the algorithm

POSITIONING

Compared to the alternatives

APPLICATIONS

Where it fits