3 ms·

TL/DR: Simulating a low noise environment is lot easier than finding one in the real world. And that's not news. Researchers simulated a 1Tb/s transfer rate wi

TL/DR: Simulating a low noise environment is lot easier than finding one in the real world. And that's not news. Researchers simulated a 1Tb/s transfer rate within a 100Mhz channel. Shannon's limit[1] explains how: set the noise term arbitrarily small, and your transfer rate can be arbitrarily large. Bandwidth is popularly misunderstood to be some fixed range of spectrum, as if signals outside the range are attenuated to zero. But you can't attenuate a signal to zero. "Bandwidth" is measured[2] as the range over which attenuation is less than (say) 3db. If not for noise, you could utilize a limited-bandwidth channel as if it had infinite bandwidth merely by amplifying frequencies outside the range in the exact proportion they are attenuated. But every real system has noise. Overcoming this noise is the challenge of high speed transfers. There's no indication these "researchers" have accomplished anything whatsoever. [1] http://en.wikipedia.org/wiki/Noisy-channel_coding_theorem http://en.wikipedia.org/wiki/Noisy-channel_coding_theorem [2] http://en.wikipedia.org/wiki/Cutoff_frequency http://en.wikipedia.org/wiki/Cutoff_frequency

frozenport 12y ago
>>by amplifying frequencies outside the range in the exact proportion they are attenuated. This is true in the sense you know a function everywhere if you can expand it for example using the Taylor series with inifnite terms. This is not true for discrete data which is limited by the niquist rate and the DTFT of the data will clearly repeat, meaning you literally can't get at higher frequencies to amplify. See the following figure, notice how higher frequencies are lower frequencies repeated: http://upload.wikimedia.org/wikipedia/commons/thumb/1/1f/ReconstructFilter.png/400px-ReconstructFilter.png http://upload.wikimedia.org/wikipedia/commons/thumb/1/1f/Rec... >>But every real system has noise. Indeed, but in many channels there are physical limitations on the propagation which is slightly different then noise....
- paulsutter 12y ago
  I'm not sure what you mean - any real world test wouldn't be a discrete signal. And yes, real physical channels are imperfect in other ways, for example not perfectly linear, etc. Could you dig into that a little? You think these effects are more significant than noise? Love to hear about it.
  - frozenport 12y ago
    >>any real world test wouldn't be a discrete signal. Perhaps band-limited, because they have a clear start and stop. If they were infinite you wouldn't be able to encode data as you could measure them at t=+/- infinity and retrieve all the signal information. Perhaps HN isn't the best place to have this discussion :-) >>You think these effects are more significant than noise? Group velocity dispersion in optical channels, people called it `noise` was at one point treated as noise before the appropriate transforms were worked out, I believe Infinera did this. In the case of remote sensing, ewald's limiting sphere represents a limit on the scattering meaning that you can't simply deconvolve and get better resolution.

frozenport 12y ago

>>by amplifying frequencies outside the range in the exact proportion they are attenuated. This is true in the sense you know a function everywhere if you can expand it for example using the Taylor series with inifnite terms. This is not true for discrete data which is limited by the niquist rate and the DTFT of the data will clearly repeat, meaning you literally can't get at higher frequencies to amplify. See the following figure, notice how higher frequencies are lower frequencies repeated: http://upload.wikimedia.org/wikipedia/commons/thumb/1/1f/ReconstructFilter.png/400px-ReconstructFilter.png http://upload.wikimedia.org/wikipedia/commons/thumb/1/1f/Rec... >>But every real system has noise. Indeed, but in many channels there are physical limitations on the propagation which is slightly different then noise....

paulsutter 12y ago
I'm not sure what you mean - any real world test wouldn't be a discrete signal. And yes, real physical channels are imperfect in other ways, for example not perfectly linear, etc. Could you dig into that a little? You think these effects are more significant than noise? Love to hear about it.
- frozenport 12y ago
  >>any real world test wouldn't be a discrete signal. Perhaps band-limited, because they have a clear start and stop. If they were infinite you wouldn't be able to encode data as you could measure them at t=+/- infinity and retrieve all the signal information. Perhaps HN isn't the best place to have this discussion :-) >>You think these effects are more significant than noise? Group velocity dispersion in optical channels, people called it `noise` was at one point treated as noise before the appropriate transforms were worked out, I believe Infinera did this. In the case of remote sensing, ewald's limiting sphere represents a limit on the scattering meaning that you can't simply deconvolve and get better resolution.