Channelling Claude Shannon

I'm playing with naive multilevel FSK sound tracks (generated with python) and Kdenlive capability to sync them with FFT convolution (for now it's hit and miss). Progressively I realized the questions popping in my mind about the task I was trying to accomplish converged toward information theory... 8-)

What information am I trying to transmit? How much? Do I have enough time? How will it be received and demodulated? With enough precision? Is there noise? etc etc...

And what precisely is the scope statement ?

The answer:

I want to be able (using Kdenlive FFT convolution) to sync a 0.5 sec sub-take into a larger one (say, one full day) within a subframe certainty of ± 1 ms, encoding the necessary information through a 10 bits@48 ksps channel.

Voilà. Is that mathematically possible? Yes. A full day is 86e6 slices of 1 ms: uniquely tagging each one of them would need 27 bits; during a 0.5 sec clip I could transmit 24000 samples of 10 bits, ie 240 kbits!

So there's enough headroom even considering noise presence and demodulation errors. That's why I won't bother grokking online digital signal processing courses to implement best practice modulation techniques: it'll be crude and naive, but working.