Wed Dec 5 15:42:12 CET 2007

weird LPF output

i think i need to focus on building some more debug tools
today.. something's going wrong and i can't find the cause. the
problem is amplitude modulation in the LPF power output, going from
100 -> 400, with a period of about 35 = 140Hz. and a component at 4 x
that frequency, not locked, which is probably the carrier.

i measure this with the modulated signal, and with an unmodulated

go one by one: it's probably best to try to eliminate the DC offset,
so at least that is not drowning the signal component, which is a lot
smaller.. EDIT: this is already happening: sample is converted to
signed then multiplied.

ok.. questions

 * why is there a 1/8 Hz component in the power output? i would expect
   the power to be smooth.. not modulated -> this is just noise. the
   level is really low, so it's the accumulation of (2^(-8) * u).

 * why is there a 1/64 Hz component in the power output?
   EDIT: the frequency is a mixer mismatch = 1/8 - 1/8', where 8' is
   the not quite =8 measured carrier frequency.

 * what does the filter input look like?

dit some input signal measurement, and the first thing i notice is
that the carrier frequency is quite off. i get 28/4 is T=7 instead of
T=8. which would give a beat at 1/56. that might explain a lot...

ok.. i get it. the convolution of these 2 spectra:

     |  .  |     cos(w1 t)
    |   .   |    cos(w2 t)


 |     |.|     | cos(wd t) cos(ws t)

with wd = w2-w1 and ws = w2+w1

the sinewave that gets folded near 0 will interfere with the signal
data! so this approach just doesn't work without synchronization!

it looks like the only way to do this is to either have proper
synchronization, or use a band pass filter, not a mixer.

AM: first order lowpass with complex coefficient, followed by output
    power computation.

PM: requires AGC or cartesian->polar conversion for properly scaled Q
    -> phase feedback.

the quick and dirty way is to just filter the absolute value of the
input. then add a more selective filter. hmm.. i still need to kick
out DC, so better go for the frequency-selective filter.