convolution-dsp

1-dimensional convolution library for Rust intended for use in DSP applications. Uses the overlap-add FFT method.

Planned features

• Input signal
  • ☑ Complex32 signals
  • ☐ f32 signals
• Filter kernels
  • ☑ f32 filter kernels
  • ☐ Complex32 filter kernels
• ☐ Use realfft when signal and kernel are both f32
• ☑ f64 support
• ☐ Fallbacks to non-fft convolution when it is faster
• ☑ Full or same length output mode, similar to numpy/scipy
• ☐ Threading
• ☐ Minimize memory allocations
• ☐ Faster than numpy/scipy
• ☐ Re-export num_complex?

References

Stephen W. Smith, Ph.D., The Scientist and Engineer's Guide to Digital Signal Processing, Chapter 18.

lib.rs:

use convolution_dsp::{ConvMode, Conv1dPlanner};
use num_complex::Complex32;

let filter = vec![0., 0., 0., 1., 0., 0., 0.];

let planner = Conv1dPlanner::new();
let mut conv = planner.plan_conv1d(&filter, ConvMode::Same);

let signal = vec![Complex32::ONE; 1_000_000];

let output = conv.process(signal);

assert_eq!(output.len(), 1_000_000);