I have problems with computing global, segmental, and frequency weighted segmental SNR. Do you have samples?
Answer:
Each of these is just a way of looking at SNR over a different spectrum, selected for its importance in the design at hand.
For example, global SNR is the total signal divided by the total noise across the full spectrum of interest. For audio, that might be 10Hz to 20KHz (just picking arbitrary numbers here).
Segmental is sampled only across a certain spectrum of interest, ignoring data outside that frequency range. This is often expressed as a segment of "n" samples at "x" width samples. For example, if we're only interested in noise that can be reproduced by the speaker in a cellphone, while the actual audio signal might exist from 0Hz to 15KHz based upon the technology, we might be interested only in the 200Hz to 8KHz portion of that signal that can be reproduced in the handset speaker. So you would be evaluating some number of data points within the full spectrum as a segment of that full spectrum. Like I say, just an example. All this requires is a definition of the spectrum of interest, and you apply the same methods for measurement that you'd apply for global.
Frequency weighted places more emphasis on noise within a certain portion (or portions) of the spectrum in which we are interested, and requires a good bit more definition. For this one, I want you to think about a db meter and how it weights sound at different frequencies to create a "total" value. dbA is an "A weighted" measurement that used one curve (a "weighted filter") and dbC is a "C weighted" measurement. These two curves place a different amount of emphasis on the sound content based upon the frequency within the spectrum of interest. Your "frequency weighted segmental" must have some curve associated with the spectrum of interest so that you can assign more weight to noise in one part of the spectrum of interest than others. That must be established before you can begin to do any computation.
So for global, it's a simple calculation:
SNR (in db) = 10 log10 (signal power / noise power)
or
SNR (in db) = 20 log10 (signal amplitude / noise amplitude) ^ 2
For segmental, just use the same computation against the the narrow subsets of datapoints that are of interest.
For frequency weighted, you've got a knottier problem, since you need to apply the above based upon a frequency sensitive filter using some curve or another (the question is ... weighted HOW?). You'll need to overlay your emphasis curve with the sample data in order to place the desired emphasis on the various frequency components of the measured signal and noise. Once you've done that, you can use the standard method for integration and compute using the log method just as you would for a normal segmented signal measurement.
Without more specifics, it's hard to set up examples for you.
The answers post by the user, for information only, FunQA.com does not guarantee the right.
More Questions and Answers: