[TowerTalk] Idealized low band system designs

[Jim Lux]

At 04:13 PM 2/16/2005, Tom Rauch wrote:
> > Bringing up a very interesting point... Is it
>theoretically possible (with
> > unlimited processing power and perhaps a lot of receiving
>elements
> > available) to cancel the noise?  After all, a gain antenna
>helps a bit
> > (because it "sees" less of the noise), but that's a
>directivity thing, not
> > a "gain" thing.
>
>I started trying this in the 70's Jim. Signal phase and
>level varies rapidly even at distances as close as several
>hundred feet on 160 meters. That means when combining
>elements spaced over a distance of a few wavelengths, the
>combining system would have to recognize and track a desired
>signal. That's a big technical problem for random signals on
>random frequencies.
>
>I can't directly combine arrays spaced more than 1000 feet
>or so on 160 meters with manual or fixed phasing, and
>automatic phasing is not practical with noise floor random
>signals.
I would agree that doing by hand is impossible, since you would need to 
update your weights on a millisecond by millisecond basis.

But these days, you CAN do the automatic tracking of random signals (albeit 
with a fair amount of real time computation). Algorithms like MUSIC or 
ESPRIT do just this, and, if you can tolerate some short delay, can even 
use what is coming in "now" to process what came in a short time ago.

Here's one shortK(but kind of math-heavy, unfortunately) summary of this 
sort of thing:
http://www.ece.utexas.edu/~bevans/courses/ee381k/lectures/13_Array_Processing/lecture13/lecture13.pdf

This one's a bit better and approaches the problem as a "blind beamforming"
http://cas.et.tudelft.nl/education/et4147/sheets.html
Start with #1, and then look at the ESPRIT one.

here's 50 pages from the IEEE Proceedings
http://wsl.stanford.edu/~ee359/godara2.pdf
(the first part of the article is at
http://wsl.stanford.edu/~ee359/godara1.pdf
)

Essentially, these techniques rely on looking at the data and doing 
direction finding on all the signals that are present, then, using that 
information to create an optimum filter for each of the desired 
signals.  Narrow band signals (like CW or PSK31) or signals with a lot of 
structure (like vowel sounds in SSB speech) should be almost ideal for 
these algorithms.



Obviously, there hasn't been a lot of research in the professional world 
into CW or SSB speech.. they're more interested in various and sundry 
digital coded signals.

In the acoustic world, there IS a fair amount of research in doing what's 
called "speaker localization" where you try to separate multiple people's 
voices out of a combination recorded by a set of microphones.  (It's also 
called the "cocktail party" problem.) This is fairly similar to the ham 
radio problem.

Ham radio is somewhat unique in that we use many-to-many links that share a 
channel (a "party-line" in analog telephone terms), while most digital 
links are one at a time, with sharing done by creating separate channels 
(either by time, code, or frequency division, or some combination of 
them).  There is no such thing as a "pileup" in the digital world.



>What I settled on is using the largest possible size antenna
>array, and then picking wider spaced cells as "pairs" to run
>in stereo on phased locked receivers. The more directivity
>you can get in a limited area, the better it will work.

Wherein you are using a very high quality adaptive speaker localizer (e.g. 
the original cocktail party adaptive filter) to do the work.  However, your 
processing isn't easily usable by another ham (Vulcan mind-melds being only 
fictional to date), nor is it necessarily well suited to filtering out a 
new and novel form of interference (at least not without a lengthy training 
period), such as that from BPL.

For the ham world, the big hangup right now is not the availability of the 
processing power (since it exists in cheap desktop computers), it's the 
peripherals.  You'd need some sort of fairly high dynamic range 
multichannel receiver that produces multiple digital audio streams for 
processing.  I don't know.. maybe something in the music business might 
serve as a high dynamic range multichannel A/D, and then, all you'd need is 
an array of appropriate RF front ends to downconvert to audio.

Today, it would be expensive (perhaps $1K/channel, by the time you're 
done), but clearly, there ARE people who have that kind of cash to 
spend.  Maybe some Silicon Valley millionaire is looking for a new 
challenging hobby, and multichannel HF receivers for busting pileups on the 
lowbands is it...


>73 Tom