View source for Talk:Signal-to-noise ratio

{{subpages}}

== Improving SNR in specialized fields ==

Thanks for the spectroscopic example, especially the way you stated the problem. There are an assortment of techniques, in communications engineering, that variously use multiple channels in what may be a similar way, but, in digital communications, we tend not to think of SNR, but rather of information loss.  This may not be good thinking; let me throw out a few examples of why we think of loss but could very well describe the phenomenon as improving SNR.  This may well go into the article, or link to a different page on error handling in digital communications -- which is a subset of fault tolerant. 

When sending digital information, it's usually a significant number of bits bundled into frames or packets. Along with the information of interest, there is control information (e.g., destination address), but also error control information. The latter is minimally a checksum, and then an increasingly powerful set of computationally intensive checking fields. 

Most commonly, the receiver recomputes the error checking information, and, if the transmitted checksum differs from the receiver-computed one, the receiver decides the information is in error. The action it takes varies with the particular application. Some "streaming" applications, such as voice or video, are more tolerant to sample loss than either to delay or errors in the data. These drop the frame and wait for the next one in the stream. 

In other applications, where errors are more important than delay, have mechanisms that will cause the data to be retransmitted until a correct copy is received or it can be concluded that the channel is broken.  Where retransmission is impractical but loss is a problem, as in telemetry from deep space, the error-checking field uses a more powerful polynomial, containing redundant information, which can not only detect errors but reconstruct the original data -- typically, the algorithm can detect errors of N bits and correct errors of (N-1) bits. 

Another approach is to send the same information over multiple paths, with the caveat that the receiver either can tolerate duplication (e.g., multiple copies of a faxed weather map) or retains state of reception (so it ignores duplicates, as in withdrawals from an ATM). The extreme case I encoutered were in some command and control networks for nuclear warfare, where some messages went out over tens of not only redundant, but physically different communications paths.

So, are these properly SNR examples? They do share properties with what you described, but communications engineers tend to think of SNR as more an analog than a digital property. [[User:Howard C. Berkowitz|Howard C. Berkowitz]] 10:17, 25 May 2008 (CDT)