I think folks like N4HY and SM5BSZ have been working on
the rudiminents of what you describe, Jim. I recall N4HY
describing a DSP noise blanker whereby you would look
at the time samples and then essentially erase and then
interpolate across large impulses that didn't fit with the pattern.
With the very fast DSPs on the market today it would seem like
there should be a whole host of things that could be done
where signals are sorted out on the basis on spectral signature.
It probably won't ever get so good that my urban QTH can
be made as quiet as the rural countryside, but there is
certainly a lot of room for improvement for us city dwellers.
Switching power supply and motor impulse noise for instance
probably have enough repetition and pattern in them that it can
be exploited for noise reduction purposes. Noise reduction
that actually improves SNR - wouldn't that be a hoot :)
Mike, W4EF.........................................................
> It can distinguish it the same way you can. As you've noticed, there are
> certain systematic features or regularities to desired signals. Such
> systematic features can be detected by an algorithm. In the case of CW,
for
> instance, the signal is quite narrow band and has certain regular features
> (on and off keying). In the case of SSB voice, there are the voice
formants
> in vowels, which are fairly narrow band, and vary slowly (at least in a
tens
> of millisecond sense).
>
> Essentially, anything YOU can detect, a sufficiently sophisticated DSP
> algorithm can detect.
>
> Noise (in the atmospheric noise sense, or power line related) has very
> different spectral properties (atmospheric noise is impulsive, and
> spectrally white, power line noise has a fairly strong harmonic content
> related to the power line frequency) than the desired signal.
>
> And this is just in a single channel sense. If there's any (potentially
> time-varying) spatial or polarization characteristics that differ, then a
> good algorithm can find them.
> >
> > There are many techniques that would work with long time
> > exposure to signals. I never could find anything that would
> > work for a case like random DX. I'd be somewhat interested,
> > since I could combine my antennas into a very large array
> > that would work substantially in real-time.
>
> There's nothing "off the shelf" that can do it. You could start with
> searching for Matlab code that implements the MUSIC or ESPRIT algorithms.
> Octave is a opensource variant of Matlab. I don't know if Matlab will be
> fast enough though, but at least you could do some non-real time tuning of
> the algorithms.
>
> Maybe the guys in France who have been doing the real time stuff with
> colocated loops have some useful versions of their software.
>
> This is a definitely non-trivial job. It's probably a multi-hundred
thousand
> dollar problem, if someone was paying for it. Rather than yet another
PSK31
> demodulator, this is what the ham software community should be working on.
> Think of it.... even in the single channel implementation, a piece of
> software that could take the audio stream from a standard radio and split
> the pileup into separate channels.
>
_______________________________________________
See: http://www.mscomputer.com for "Self Supporting Towers", "Wireless Weather
Stations", and lot's more. Call Toll Free, 1-800-333-9041 with any questions
and ask for Sherman, W2FLA.
_______________________________________________
TowerTalk mailing list
TowerTalk@contesting.com
http://lists.contesting.com/mailman/listinfo/towertalk
|