Sun Dec 9 16:22:43 CET 2007


let's see.. there's something wrong with my symbol rate. i thought it
was 512 samples, but it's 256. corrected for this, i can receive
signals. however, it seems the bandwidth is mismatched. so i have 2
calculations that are probably erroneous:

  * necessary bandwidth -> filter coeff
  * symbol rate at the transmitter

doing some manual experiments, i got the filter pole fixed at #x0100,
which gives very nice waveforms. making it bigger only increases the
noise, but doesn't seem to influence the shape too much.

so everything looks pretty good, but it seems there is too much
inter-symbol interference due to the assymetry of the receive
filter. i could try to hack around this by doubling each bit, but
keeping the envelope constant.

got now:
    * halved symbol rate (transition + stable)
    * 3 x bandwidth (100 -> 300)

now at least the filter makes it roundtrip from 0 to max amp.

now try to subtract startbit from each bit, then use sign.

this works!

reception seems quite robust. at least when it's not receving
bogus. so i need a way to eliminate the worst kinds of noise, which
are transients that trigger a start bit. these could then be used as
human input.

it's not very robust tho.. probably i need to compute the maximum, and
use half of that (or less..) as a threshold.

  looks like this is relatively robust:
        threshold = 1/4 of maximal power
        translated to amplitude this is 1/2