From: JosephKK on
On Fri, 02 Jul 2010 09:41:40 -0700, Tim Wescott <tim(a)seemywebsite.com>
wrote:

>On 07/02/2010 04:14 AM, andy baxter wrote:
>> On Thu, 01 Jul 2010 14:44:25 -0700, Tim Wescott wrote:
>>
>>>
>>> Depending on how many transmitters you have, and how organic the network
>>> is, there are a number of things that might work.
>>>
>>> Vladimir's suggestion of just having each transmitter on a different
>>> frequency would work if you can assign enough channels and if
>>> single-tone signaling works in your environment (which it will if the
>>> noise is wideband, or has specific, predictable frequencies that must be
>>> avoided).
>>
>> The problem with this is that the number of transmitters could
>> potentially be quite large - several thousand at a guess - so giving each
>> one its own frequency might not be so easy.
>
>Argh. Indeed.
>
>>>
>>> Spread-spectrum (code division multiplexing, really) makes sense if you
>>> have single-tone interference that you can't trust to stay on frequency,
>>> but trying to find a scheme that keeps the receiver's cheap may be a
>>> challenge. Spread spectrum won't do you much good if the noise is
>>> completely random and is already swamping your receivers.
>>
>> This is something I'm trying to clear up in my own mind - do spread
>> spectrum techniques have inherently better noise rejection than
>> traditional methods, or does it depend on the situation?
>
>It depends on the situation. We're really used to thinking of the world
>in terms of frequency spectrum because it's so easy to make resonant
>filters that not only is that one of the first things we learned how to
>do with electronics, but the world is full of them. But if you can put
>that aside, the basic architecture of a spread spectrum receiver in
>cloud-cuckoo land is exactly the same as a regular old superheterodyne
>receiver like you find in all but the cheapest radios: the signal from
>the antenna is multiplied by a reference signal (the LO in a superhet,
>the despreading sequence in a spread spectrum RX) and low-pass filtered.
>
>Spread spectrum radios do better overall than narrow-band radios in
>situations where there is lots of narrowband interference. It is also
>much easier to multiplex an unknown number of transmitters with CDMA if
>you can synchronize them all: with frequency-division multiplexing when
>you fill up the spectrum you're done, and if you have 1000 empty
>channels and two transmitters that happen to be fixed at one frequency
>you're screwed. With CDMA you can make a practically infinite number of
>"channels" that are mutually orthogonal, and each active transmitter
>just raises the noise floor for all of the receivers. So not only is
>planning much easier, but the failure is more graceful.

No, not at all, it is quite as easy to fill the code space as it is to
fill up the frequency space. Especially if you want short transmissions.
Gold codes work well for GPS because the transmission is continuous. The
Viterbi patents are closer to what OP seems to need.
>
>> Also, how do noise rejection and data rate work together? I'm guessing
>> that you can get better noise rejection by lowering the data rate, but
>> this is only an intuitive guess, and I was thinking that there are
>> probably equations which have been worked out covering this sort of
>> question, which we could get from a decent book on the subject.
>
>This is a basic communications question, not really spread spectrum.
>Yes, you get better noise rejection by lowering the data rate -- with
>totally random noise, when you average a constant signal + noise the
>output noise level goes down by the square root of the averaging
>interval while the signal strength remains the same.

See Shannon's law. It relates BW, signaling rate, and S/N ratio in a
very useful way.
>
>There are fancier ways to achieve better results when the noise isn't
>white, or when it's bursty -- but it all boils down to the fact that
>making longer observations generally get you more reliable results at
>the expense of data rate.
>
>> Another question - what about chirp spread spectrum? The thing that
>> kicked all this off was my boss was talking about how he thought we ought
>> to be using spread spectrum techniques, and I offered to do some research
>> on the subject, and set up a demo for them of chirp spread spectrum in
>> the audio domain using supercollider (an audio synthesis programming
>> language). I chose to do the demo using CSS because it looked like it was
>> a simpler way of getting the benefits of spread spectrum than CDMA.
>
>I'm not familiar with it specifically, but _any_ spread spectrum
>technique is going to have similar advantages and disadvantages. Chirp
>spread spectrum may allow you to simplify the receiver, which would be a
>good thing.
>
>>> What about time-division multiplexing? Is there any reason you can't
>>> have the transmitters cooperate to send at different times, perhaps in
>>> round-robin order? Or are you already doing this?
>>
>> I don't think this would work - the transmitters have to be kept as
>> simple and low power as possible, so we want to avoid putting any
>> receiving capabilities on them. Something like ethernet where they each
>> broadcast a short id code at random intervals might work though, if we
>> make the receivers capable of detecting collisions.
>
>And if you make the duty cycles of the transmitters short enough, and
>the hold-off intervals random enough -- if you're really going to have
>thousands of transmitters then you could easily reach a situation where
>you'd never have a clear space long enough for a transmitter to send in
>the clear.
>
>Something that spread spectrum could do for you in theory is to let you
>decode transmissions that overlap, as long as they start at sufficiently
>different times. But trying to do so will seriously complicate the
>receiver. CSS looks like it would complicate such a receiver the least,
>which is nice. My brain boils with possible ways to do this that would
>(a) work, and (b) keep the receiver at least moderately practical.

Another issue is keeping the transmitters cost effective.

Come to think of it, OP has not told us whether or not freespace
propagation or bounded media is involved and how much support of existing
infrastructure will be required.
>
>>> Like Vladimir, I could help you with this. I think you should seriously
>>> consider getting an expert to at least help you kick things off -- there
>>> are a lot of ins and outs in this sort of thing, and I very much doubt
>>> that they're all going to be collected in one book for you. In
>>> particular, you don't want to get dazzled by some complicated technique
>>> when there's a simple expedient that would work well.
>>
>> We have considered this, but the point for us is that even if we do get
>> someone else in to work on this part of the project, it would be useful
>> to know a bit more about the subject ourselves first just to have a
>> better idea of whether this is justified, and to be able to assess the
>> work they are doing if we do take someone else on. Which is why I
>> suggested that we should look for a good book that would at least teach
>> us the language of the subject and give us an idea of what is what in
>> that field.
>
>Understood. A good expert would start by being your consultant -- i.e.
>they'd start by evaluating what you're trying, and let you know if what
>you want to do is possible, whether it's feasible, and whether it's
>practical given your budget. Only after they're done with that would
>they take your money to start actually designing the system.
>
>They'd also leave you as educated as they could, so that you'd at least
>understand how to use what they did.
>
>But yes, studying up on it a bit first would be good. The texts that I
>have sitting on my shelf that pertain to this, at the level that you're
>probably interested in, are:
>
>"Information Transmission, Modulation and Noise", Mischa Schwartz, 1980,
>McGraw-Hill
>
>"Principals of Communications Engineering", Wozencraft and Jacobs, 1965,
>Wiley
>
>"Principals of Data Transmission", A.P.Clark, 1983, Wiley
>
>Unfortunately, these will all give you lots of foundation for
>understanding your problem without actually connecting directly. To
>bracket your problem while still not hitting it on center, get a copy of
>the ARRL Handbook -- that has a ton of applied theory for wireless
>communications, but it (a) doesn't touch heavily on spread spectrum and
>(b) misses what you want in the other direction -- the Handbook tends to
>give solutions without tying things back to theory, while the three
>books I cited tend to give theory without tying it to solutions.
>
>If you're going to talk to a consultant, the theory books are probably
>the ones to go with -- but the Handbook is more fun, and a better door
>stop in a high wind. People with more up to date libraries than mine
>often recommend that people "Get the Proakis", and John Proakis has a
>number of titles out there. The "Digital Communications" book by
>Proakis that's on Amazon looks like a likely find -- it's table of
>contents looks like it covers what the Clark book I cited covers, and
>it's probably more up to date.