Mon Dec 3 16:56:50 CET 2007


 * AGC: or the absolute value of a symbol buffer + compute shift count
 * INTEGRATE: sum the entire buffer (no sideband rejection)
 * sample at say 8 points per symbol

since there's no filter other than the analog 450-2kHz this should
perform pretty bad. but i guess it's time for a fail-safe.. use noise
modulation first :)

NM -> AM (async) -> PM (sync)

a genuine problem doing this al experimental is the program
sequencing.. there's a huge difference between being able to do
something per sample and having to store some for later..

EDIT: it doesn't make sense to write an AM demodulator without
thinking about the BPSK that will follow, so i need to do AM with a
separate mixer + LPF.

mixer seems straightforward. the remaining problem is the LPF. If i
can make that work with simple shifts, where's as good as there..