Hacker News Reader

Comment by allegory | Hacker News Reader

"The receiver “listens” to determine the RF environment and uses algorithms to predict how the transmission will be altered by the environment and cancel it out from data received from the receiver. By doing so, the problem of self-interference goes away, and the receiver can “hear” the signal it is supposed to hear while the transmitter is active."

That's a little simplistic IMHO and akin to magic.

The 'RF environment' changes drastically as you move (depending in the propagation of your waves and reflections etc) and as other things operate so even with adaptive algorithms there is going go be a period of learning that will reduce this back to half duplex periodically. With the masses of noise and interference and harmonics around I'm not sure I believe the claims entirely. Maybe in ideal conditions it won't spend 100% of the time learning?

A link to a paper would be nice...

View on HN

I think a very good paper solving this problem very clearly with a cool execution is by Katti Group.

http://web.stanford.edu/~skatti/pubs/sigcomm13-fullduplex.pd...

It deals with all the issue and shows the 2x throughput as well.

Thanks - will read next :)

It's interesting to think of the side uses for this technology, should it become widespread. The future WIFI radio in your cellphone will end up being a precise measuring device for the RF environment. I mean you've basically got a TDR instrument in your phone sensitive to 110 db system loss..

For sure it will work as a motion detector (search for gunn diode motion detectors- long used by alarm systems).

I don't have time to skim the papers now, but here's the research project the CTO worked on while at Stanford that the tech appears to be based off:

http://sing.stanford.edu/fullduplex/ http://sing.stanford.edu/pubs/mobicom11-duplex.pdf

Good document, also, http://www.mit.edu/~viveck/code/Research_files/fbdof-1.pdf covers broader topics including FulL Dux.

Simple laymen's explanation, the transmitter wave form is fed back into the receiver but inverted so it cancels out the transmitters signal. You can think of this like noise cancelling headphones in reverse. With noise cancelling headphones you sample outside noise and invert it to cancel that. Here you sample your own signal and invert it and add it to the received signal and the inverted signal cancels out the transmitted signal. This technology has been around a while but the DSP technology is just now really getting good enough for it to be practical.

Do the papers address how it will deal with the analogue to multipath where the receiver hears a lower power reflected signal later?

No, I guess that is why they formed a company. This will be the secret source.

Perfect thanks - reading now!

I don't get it either. It seems like the receiver should be saturating if it's really receiving concurrently with the transmitter.

Well I can see how that bit works. If the TX and RX signal paths are short enough (remember EM energy doesn't move infinitely fast), you can subtract the TX's signal from the RX's signal and appropriately gain adjust it, then you will be left with the RX's signal only.

However the problem I have with this is that if your signal propagates 100m then is reflected back another 100m, how does the RX then discriminate between that signal and the signal from another TX on the basis that the signal is so far out of phase (670ns P-P approx assuming vacuum) having travelled such a distance and back.

You need buffers, software and all sorts of magic there as the RF carrier has gone a few cycles then.

In my previous comment I was considering only saturation at the ADC, however, I later realized that they were probably performing cancellation in the analog domain first, and then throwing some DSP magic at it later on in the digital domain to handle reflections, etc.

Good point. It comes down to how well the analog cancellation will attenuate the in-band blocker (cf. IIP3 spec of the RF LNA).

Reply on news.ycombinator.com