Module dsp::convolution

#include <std/dsp/convolution.h> source

Convolution

Linear (non-circular) convolution: y[n] = sum_k x[k] * h[n-k], output length len_x + len_h - 1. Two implementations with the same output for the same input, trading off differently:

conv_direct / fconv_direct — O(len_x * len_h), no allocation, exact
  (fconv_direct: exact up to Q8.24 rounding). Best for short filters
  (e.g. convolving with a small FIR kernel).
conv_fft — O(n log n) via fft.h, allocates a scratch buffer sized to
  the next power of two >= output length. Best for long inputs/long
  kernels, where direct convolution's O(len_x*len_h) cost dominates.

Functions

conv_direct	out must have space for (len_x + len_h - 1) samples.
fconv_direct
conv_fft	FFT-accelerated linear convolution — same output as conv_direct (up to floating-point rounding), computed via zero-padded forward FFTs, pointwise complex multiply, and one inverse FFT. out must have space for (len_x + len_h - 1) samples.
conv_direct_f32	Single precision: conv_direct / conv_fft on floats, same contracts.
conv_fft_f32