PSK versus FSK]
Thu, 18 Feb 1999 08:46:08 -0500
> I've been following today's exchange on PSK31 with some fascination. As Andy
> points out, "the maths behind communications theory such as this is
> horrendously complicated." But the bottom line (as we're fond of saying over
> here) is that it works.
> It's sometimes entirely acceptable to have the few-hundred-Hertz emission
> bandwidth that results from "plain" BPSK. Here, our 160-190 kHz license-free
> band can accomodate numerous such signals, even if some of the outer,
> unnecessary, sidebands overlap. Given the size of the 136 kHz band, though,
> one does begin to wish the transmitted bandwidth wouldn't have to exceed the
> bandwidth necessary to actually receive the signal.
> Suitably encoding the data then shaping the signal with a raised-cosine filter
> is a remarkably efficient method, even with the necessity for linear
> amplification afterward. Since many administrations specify limits on either
> transmitter output power or ERP, rather than limits on DC input, amplifier
> efficiency shouldn't be a big problem. But if it is a drawback, envelope
> elimination and recovery techniques in the power amplifier might be one
> relatively convenient solution.
> Just to give an idea how spectrum-efficient one can get with suitable
> application of communication theory, I hope you'll forgive me if I indulge in
> what may initially seem a far-fetched analogy from my own occupation:
> High definition digital television originates with data rates on the order of
> 1.5 gigabits per second. Our over-the-air DTV system in the States compresses
> the image data significantly, so that the combined vision and sound stream is
> 19.39 megabits/second when it reaches the input of the digital television
> transmitter. Before applying the bitstream to the modulator, though, we
> further pseudo-randomize it and add additional error correction bits,
> resulting in a bit rate of almost 30 Mb/s. All that has to squeeze into a 6
> MHz wide channel!
> To accomplish that marvel, the bits are grouped into blocks of three, which
> are treated as symbols, represented by eight discrete levels--two to the third
> power, of course. Thus, we're down to 10 megasymbols/second.
> Those symbols are raised-cosine-filtered by EXTREMELY fast DSP before being
> applied to the modulator. Almost none of the symbols involve transitions
> between widely separated values, thanks to all the pseudorandomization and
> convolution in the preceding steps. Thus, believe it or not, the resulting
> filtered waveform can carry mutiple symbols per single "cycle" of baseband.
> (Computer modems have been doing this for some time, of course.)
> One advantage to filtering in DSP is that the transmitter can be made to adapt
> itself to modest nonlinearities, frequency response ripples, and group delay
> errors in the power amplifier stages. Similar adaptive error correction takes
> place in the receiver, to compensate for exactly the same sort of problems in
> the receiving system, plus multipath distortion and NTSC-to-DTV co-channel
> interference. (Adaptive filtering at the transmitter would neither be
> necessary nor practical for LF amateur work, but can readily be implemented in
> software at the receiving end.)
> The end result is a signal that fits snugly but cleanly inside the alloted
> channel. With adaptive receiving techniques, it's also exceedingly robust.
> Just the thing one needs in the longwave environment, I'd say. It will be
> most interesting to see how the work on PSK31 progresses.
> John KD4IDY