windytan / stereodemux Goto Github PK

The resulting amount of stereo separation depends on the sample rate. Especially, 256 kHz and 171 kHz sound vastly different.

• Filter length depends on sample rate. Maybe the gain of some of the filters changes based on filter length?
• Why is the mono channel's gain always correct?
• Why is there no difference between 256 kHz and 384 kHz?
• Does the delay line somehow change behavior with long buffers?

If you have access to a proper FM transmitter or stereo encoder, the MPX output could be useful for testing that the separation is correct. This stereo file could be used as the test audio (warning: sounds nasty to human ears) : https://oona.windytan.com/blogfiles/stereo-test-for-wfm.wav , and if you have results, please post below :)

• WAV I/O (libsndfile)
• Autotools or CMake
• Output & input rate conversion (resampling)
• IQ input (internal FM demod)? Though this might be a little out of scope if we want to keep it in Unix philosophy
• Blackbox testing with a known MPX file (need to use something with no copyright issues like this one encoded with some known commercial stereo encoder – can you help?)
• Test for the de-emphasis frequency response

Currently the pilot tone is isolated by first mixing it down, then low-pass filtering it, and mixing it back up again. This wastes cycles; instead we could use a band-pass filter. However, shifting up the coefficients doesn't produce the same result - why?