DARPA Wants Extreme Wireless Interference Buster

coondoggie writes "This month the Defense Advanced Research Projects Agency will begin looking for technology that will let wireless communications work through the most extreme interference. From the article: 'The CommEx program will assess next generation and beyond jamming threats and then develop advanced interference suppression and avoidance technologies to successfully communicate in the presence of severe, traditional, and novel types of interference that are orders-of-magnitude more severe than what are currently addressed by the most advanced systems, DARPA stated.'"

Comments

distinct lack of content

[advocate_one]

in the article... anyone got a better link?

Ok, an interesting challenge.

[jd]

The conventional approach by NASA is to use Turbo Codes to handle burst errors and Reed-Solomon to handle randomly-distributed errors. You'd need to increase the error correction bits to handle really significant errors, but that seems like a good starting point. If you were to imagine the data as a cube, then produce the error-correction codes for each and every line you could draw through that cube, then each unit within that cube is represented by three sets of error-correction codes.

Re:Ok, an interesting challenge.

[jd]

Any given point in a cube exists along three lines - one parallel to X, one parallel to Y and one parallel to Z. If you wish to find a fourth orthogonal axis, please do so. Your TARDIS awaits.

Regardless, three is plenty. If you want to improve reliability still further, then if turbo codes are mapped onto X, use LDPC (low-density parity-check code) alone -X.

Since Reed-Solomon is for random errors, you'd probably want to use a conventional ten-bit system - 8 bits of data, 2 bits of ECC. A layout used by a lot of modern electronics. Since turbo codes and LDPC are intended to fix bursty errors, they're useless if the block size used is smaller than the size of burst that you can expect. In the scenario the DoD is interested in, it's extremely hard to say what to do here. If the block size is too big, then the error correction codes have an increased probability of being damaged by such a burst. Thus, not only is there a lower limit but there is also an upper limit.

NB: If you not only want to ECC each line but also each plane using turbo codes, then you have a further constraint. The line cannot exceed the minimum size but the plane cannot exceed the maximum size.

My thinking is that each line's turbo codes would also cover the Reed-Solomon codes, and each planar turbo code would also cover all of the turbo codes for each line within that plane.

Since the same object is mapped from three (or six, using LDPC) different perspectives and the probability of the noise being such that all three/six perspectives are damaged in mathematically identical ways is extremely low, you can generate a set of the most probable original messages at that level. This could be done by combining the probabilities generated from each of the algorithms, a voting system, or some sort of statistical analysis. How doesn't matter that much, so long as you can produce a bounded set of messages that could be described by the codes.

Here, you can use Reed-Solomon not as an error-correction code but as an error-detection code. You know for a fact that the original pattern will satisfy each and every one of the Reed-Solomon ECCs. Any one of the candidate original messages that does not meet this constraint is incorrect. It has to be.*

If the slices used are one bit thick and each line drawn has a diameter of one bit, then I see no possible way for any aliasing effect to occur, which means you'll end up with exactly one reconstituted message even if fairly extensively damaged in transit. That's probably still true even if working using byte-sized atoms, but bit-planes have some definite advantages for this kind of work.

*Ok, it may be possible to inject deliberate corruption such that the resultant turbo codes would produce a candidate solution which aliased onto the Reed-Solomon codes. It'd be better if you could use cryptographic hashes, but those tend to be big and therefore more subject to being corrupted via a wireless connection. You could break the data into fixed-sized chunks, cryptographically hash each chunk, append the hash to that chunk, then do the whole process as outlined for the composite message. If you did this, then you would test the turbo code candidates against the hashes and if they all failed then you'd use the Reed-Solomon to re-fix the messages before testing against the hashes again.

This method is by no means foolproof, it is not intended to be optimal, it has not been evaluated for every possible scenario to see if it's watertight, it is not FDA-approved and it's likely signals analysts and crypto experts would regard it skeptically. It is not intended as an actual submission to the DoD. What it is intended to do is show that the problem does not require any tools that do not already exist. Indeed, it does not require any tools that aren't already being used by industry to tackle some subset of the problem. There is no novelty in this problem, beyond what combination of tools it will take to defeat deliberate interference of arbitrary nature.

jamming != white noise

[tempmpi]

They are looking at intentional jamming, not at white noise. Your solution would be almost perfect for white noise channels but not for channels with jamming.
E.g.: No jammer will be able to distribute its noise evenly in both time and space. You should be able add a nice bit of performance if you are able to predict the behaviour of the jammer to some extend. So spread spectrum with non-uniform frequency distribution of the signal energy could be a topic. Some jammers might not even send real noise but pseudo random noise. Then you could try to subtract the jammer from your received signal.

Re:jamming != white noise

[Aceticon]

And then let's not forget the spatial dimensions in addition to the frequency dimension - jamming signals originate from one or more sources at specific locations so this could also be used in eliminating interference from jammers.

I remebers from reading a bit of the GSM spec that mobile phones includes adaptative antenna and algorithms that allow retrieving a usefull signal not just from the direct line of sight transmission (from the mobile tower) but also from multiple reflections with different path lengths. Could not the same techniques be used to, instead of boosting a signal, offset that signal?

Re:What can they hope for

[Trepidity]

I think that's probably true for algorithms, but it might not be for deployed systems. Current communications devices can't use all the available theoretical techniques over all possible frequencies in all possible configurations, so there might be some significant gains on that front with new transmitters/receivers/etc. For example, most spread-spectrum systems operate over relatively narrow portions of the spectrum, at least compared to the whole electromagnetic spectrum--- nobody is spreading over everything from radio waves to x-rays, or anything close to that. Using larger parts of it has both some technical and operational challenges, since if your spread is over very large parts of the spectrum, parts of your signal are being transmitted on frequencies with extremely different properties.

I've got one for them!!!

[syousef]

It's called cat5e.

Re:I've got one for them!!!

[L4t3r4lu5]

Wrap your Cat5 in a coil made of the power lead to your PC, then turn your PC on. Tell me it's immune to EM interference using that computer.

If it catches fire, that counts as "successfully blocked signal".

Frequency Hopping.... more?

[NouberNou]

Not really sure what they can advance on besides frequency hopping routines that are quicker and cover a larger spectrum. SINCGARS, HAVE QUICK I/II and SATURN are pretty good at counter-jamming already, as long as the sequence can not be easily predicted and the fills are updated regularly (every 24 hours or so) then jammers will have to invest quite a bit into the infrastructure of their broadcasting equipment. It certainly wouldn't be portable, and it'd be loud and easy enough to find and take out by more traditional means.

The only thing, like I mentioned above, is moving to waveforms that are spread across larger frequency ranges (which can be problematic) and are faster so the jamming equipment can't keep up with the normal signal. Beyond that, digital data over RF can be reinforced by better packet correction and error handling.

Of course you could always just overpower every other signal on your band, but I do not think battery tech has reached that level yet for portable radios, and well... Most soldiers prefer not to be cooked alive if they have a choice.

Re:What can they hope for

[AK Marc]

You need signal/noise of some level. Jamming acts like noise. You can't add power past some point, so that means you must subtract noise. How? Well, you can try MIMO techniques that try to essentially lock on to the jamming signal and subtract it out. Or lock on to the generated signal with rejection of the jamming signal. Right there are two possibilities that don't violate the theory and should be able to get real gains. Maybe not the best possible, as moving jammers or moving desired signals would be an issue. But that's just a napkin thought. There are more out there. Jamming may affect the signal like noise, but it isn't.