Polyphase Filter Bank implementation in C++ by David Mittiga and published here in coordination with him.
Usage is simple. Create the PFB (Polyphase Filter Bank) with the first input specifying data input length as log2(N). e.g., for 4096 samples, log2(4096) = 12. The second argument specifies the filter order. 2x the number of channels is the minimum. Larger values result in sharper filters. The third argument specifies the scale factor. A value of 1 results in nominal unmodified filters. Larger values result in wider, overlapping filters, which is useful to remove scalloping loss, but results in more correlated output due to the overlap.
auto pfb = std::make_unique<pd::PolyPhaseFilterBank>(12,13,1);
It is expected that the input data is continguous. Pass each block of samples to Iterate.
std::vector<std::complex<double>> timeData; // assume we have some vector of time data
std::vector<std::complex<double>> freqData(timeVec.size(), 0);
pfb->Iterate(timeData.data(), freqData.data());
Screenshot of a real-time frequency domain display of amateur radio FT8 data, built using this code:
