Hi Geri, hi group,
encouraged by the "Mystery Signal" discussion, I had an enjoyable qso with
DK8KW today, sending out a couple of pictures around 137.73 kHz. I had been
experimenting with fourier image transmission for a while.
The modulation I used is similar to parallel multitone hell (MT Hell) as
described on ZLBPU's page. For n lines and a column duration of T only the
optimal total bandwidth of n/T is required. It's disadvantage is a linear
transmitter with a high crest factor (peak to average ratio), as for a bright
vertical line all the energy is concentrated in a narrow sin(t)/t pulse, only
T/n long.
As phase is not displayed in the received spectrograms, a chirp filter can be
used to smear out the same energy over the symbol time T, reducing peak power
by a factor of n for chirped multi-tone (CMT). Its easy and efficient to
apply a quadratic phase term to the image intensity before the FFT.
For a full white image, the resulting sweep signal looks similar to the SMT
(sequential multitone) mode Andy used in his experiments. But for a single
horizontal line, SMT would transmit a series of full-power pulses of short
duration T/n. To separate these spectrally, the rx would have to use a
bandwidth of n/T for each channel, or n*n/T for the full image height. Thus,
SMT is less bandwidth-efficient by that factor of n compared to MT or CMT.
This makes CMT interesting for fast "high resolution" grey-scale image
transfer like the 64*64 pixel portraits I have sent to Walter and Geri.
So, where's the catch, besides the linear tx? Theoretically, there could be a
"reversed chirp" luminosity pattern in the image, refocussing all the smeared
energy into a high peak. For random images, the crest factor would be on the
order of sqrt(n) regardless of any phase modulation scheme. But as pixel
values are always positive, the image has a strong spatial zero-frequency
component, and thus chirping still helps. Practically I found that an extra
6dB dynamic headroom was usually sufficient. When precalculating the sound
data, one could always scale them down or try a reversed chirp direction.
In the current implementation, a QBasic program transforms a bmp image to a
1MB wav file. I can play this to my SSB rig with a second instance of gram
using "Scan File", while leaving the "Scan Input" monitor running. Due to
limited FFT size (8192) and playback sample rate (11025 Hz), minimum
bandwidth is 1.3 Hz per line. Its fast, but too wide for the LF qrss band and
rather tall on an 86 Hz display, so I will try to use 64*0.33 Hz in the
future.
So much for now, 73 cu
Markus, DF6NM
|
