I need to estimate the SNR (or, even better, the actual signal and noise powers separately) at the output of a matched filter. I have seen various implementations for communications systems using maximum likelihood or based on particular types of signal coding. However my application is a radar type one - the signal transmitted is known (e.g. chirp or polyphase pulse), but after the matched filter, the received signal is 'compressed' to only one or a few samples. Also, the SNR is not constant for the received signal array, as each sample corresponds to a range bin increasingly further away from the receiver. Essentially I wish to perform a MRC (maximal ratio combination) type algorithm for several receivers for each point of the array, where the SNR of any signal received at each receiver may be different. It is possible to estimate the noise power in any receiver by recording data when the transmitter is off, but I don't know how to estimate the signal power. I could do so rather more simply by analysing the data prior to the matched filter where the received signal can be seen as a finite length pulse, but in (frequently) poor SNR conditions this may not be possible. I would be grateful for any suggestions or pointers Thanks Tom
SNR estimation
Started by ●February 12, 2005
Reply by ●February 12, 20052005-02-12
Tom Derham wrote:> I need to estimate the SNR (or, even better, the actual signal and noise > powers separately) at the output of a matched filter. > I have seen various implementations for communications systems using maximum > likelihood or based on particular types of signal coding. > However my application is a radar type one - the signal transmitted is known > (e.g. chirp or polyphase pulse), but after the matched filter, the received > signal is 'compressed' to only one or a few samples. Also, the SNR is not > constant for the received signal array, as each sample corresponds to a > range bin increasingly further away from the receiver. > Essentially I wish to perform a MRC (maximal ratio combination) type > algorithm for several receivers for each point of the array, where the SNR > of any signal received at each receiver may be different. > It is possible to estimate the noise power in any receiver by recording data > when the transmitter is off, but I don't know how to estimate the signal > power. I could do so rather more simply by analysing the data prior to the > matched filter where the received signal can be seen as a finite length > pulse, but in (frequently) poor SNR conditions this may not be possible. > I would be grateful for any suggestions or pointers >If you know the noise power and the signal+noise power then the noise power is just signal+noise - noise. Your big trouble will be that while you can measure the noise itself repeatedly you'll only have one sample of the signal+noise, so you'll have some error in your signal estimates. None the less if you have some receivers that are dramatically better it should stand out. I'm sure you could pour all of this into the relevant math hopper and turn the crank enough times to get an estimate of the composite receiver performance. If you're lucky you'll get the rx performance in a closed-form that will let you turn the crank some more and get an optimum composite receiver processor -- wouldn't that be nice? -- Tim Wescott Wescott Design Services http://www.wescottdesign.com