Musings about communication in the 900 nanometer band...

Mike O'Dell mo at
Sat Jul 8 10:40:31 CDT 2006

i read through several of the sections of the web site
describing their experiments and they argue (pretty convincingly)
that "AM will always win in weak-signal reception."

however, the discussion of modulating the Luxeon took a somewhat
narrow view of the problem.

from my reading, the way to get the brightest light from a Luxeon-style device
is to drive it in pulse-mode driver - high current with very low duty cycle.
the average optical power is equivalent but the peak power is much higher.
(the electrical consumption can actually be lower in some cases.)

it makes one wonder whether revisiting FM might not be worthwhile.

simply FM the pulse rate with the audio. i believe off-the-shelf pulse-mode
LED drivers (designed for this new generation of LEDs) run with
50-100KHz rep rates.  there's a limit to the deviation you'd
want to run to keep the duty cycle (and hence device dissipation)
under control, but i think that's workable although i haven't
done the sums to compute the modulation index vs pulse rep-rate curve
vs device dissipation curves.

this also questions whether the limitations of an avalanche PIN diode
detector would be as problematic.  when biased with relatively HV
(80-100VDC) a PIN diode operates much like a photomultiplier, but
it is very noisy, non-linear, and the response isn't terribly fast.
however, the optical input is a narrow pulse so the PIN acts as
a low-pass filter - it should smooth the pulse train into something
much closer to a sinusoid at the pulse rep-rate fundamental.

so treat the PIN as a source for a low-level FM carrier at the pulse rep-rate.

amplify, bandpass filter at the center frequency, amplify to clipping
(ie, a limiter stage) and then demod with a PLL to recover the audio.
for the center frequencies of interest, the bandpass filter could be
implemented as a state-variable filter with fast op-amps so the frequency
and Q would be tunable to make it easier to experiment with the rep-rate.
(and getting a state-variable filter to clip is not hard (grin) so you
might subsume the limiter function in that stage as well.)

this analysis is probably completely bogus because a PIN turns out to be
way too slow or some of the other sums don't work out right when you really do them.

so i offer it as no more than an amusing speculation.


More information about the Tacos mailing list