Sample Rate Converter

Function input takes sampled signal, b, old sample rate Fs, and wanted new sample rate Fsnwant. Function outputs resampled signal, bn, and actual new sample rate Fsn (rounding to nearest integer sample):

[bn,Fsn] = sampleconverter(b,Fs,Fsnwant)

The algorithm works by strategically inserting blank frequencies or removing frequencies about the Nyquist frequency of B while preserving conjugate structure of the frequency domain (input is typically real, so I operate to fully preserve Hermitian Symmetry with as little signal distortion as possible).

Noting how Nyquist behaves differently for even vs odd signal lengths (explicit vs implicit mirroring), we yield 4 upsample cases: {even to odd, even to even, odd to even, odd to odd} and likewise 4 downsample cases: {even to odd, even to even, odd to even, odd to odd}. Analysis reveals upsample cases 1 and 2 are the same, and 3 and 4 are the same. Frequency domain diagrams may help you to distinguish the 6 remaining distinct cases. After resampling we get new length, unscaled Bn.

We rescale Bn by the ratio of the new sample length to the old sample length (Nn/N), take the ifft of Bn and we have our answer!