zerophase

Zerophase, or zero-phase, describes a property of a filter or signal-processing system where the phase response is zero across all frequencies. In theory, this means the output waveform is not shifted in time relative to the input, aside from any amplitude changes. True zero phase requires a noncausal, symmetric impulse response centered at time zero, which cannot be realized in real-time systems that must process data causally.

In practice, zero-phase behavior is approximated through noncausal methods or by processing data in both forward and reverse time. A common approach is forward-backward filtering (as in the filtfilt method), which applies a causal filter in the forward direction and then again in reverse. The net result cancels phase distortion, producing a zero-phase output, but at the cost of doubling the effective filter order and potential edge artifacts due to signal extension.

Zero phase is often discussed in contrast with linear phase filters, which have a constant group delay but still introduce a uniform time shift. Achieving true zero phase typically requires noncausal processing, making it unsuitable for real-time applications but useful for offline analysis and applications where waveform shape must be preserved.

Applications and considerations include audio processing, deconvolution, and seismic data analysis, where preserving waveform shape and aligning phases across channels are important. Practitioners must balance the desire for zero-phase distortion against the realities of causality, latency, and edge effects.