And make some decent coin while you’re at it.
I’m doing some reading on information theory and Claude Shannon currently.
Our friend Nyquist for proper sampling without aliasing. Shannon covers the information channel limits e.g. signal to noise, the maximum number of symbols per second, SNR required to discriminate between symbol elements. All of that is good background before designing a hardware/software DSP solution. I prefer analog front end filters to limit input frequencies such that my A to D sampling doesn't alias. After that, it's all DSP "magic". My railcar bearing analysis code used a 16-bit A to D sampled at 100 KSPS for 20 seconds. Interesting info modulates an 11 KHz band. Harmonics up to the 5th are useful in characterizing the bearing defect type. My colleague had an extensive library of known defect types in a digital audio collection on DAT. There are characteristics attributable to cup, cone, cage and rollers. The basic collection covered 55 common types.