Hi all I am working on low frequency sonar signal processing algorithms as my final thesis. I have a confusion b/w time domain implementation of matched filter (MF) process and frequency domain implementation of MF process. If i use LFM pulse replica of bandwidth W for target doppler (or speed) finding, i need multiple doppler shifted replicas/MFs according to multiple target speeds for time domain MF implemetation as i understand.i.e. I have to use bank of MFs and check for which MF, maximum peak occurs. But if use frequency domain MF , do i still need multiple replicas or single zero doppler shifted replica will be enough? Please help me . If nay one knows any helping material on DFT based sonar doppler finding, please let me know THANKS
sonar matched filter
Started by ●September 14, 2010
Reply by ●September 14, 20102010-09-14
On Sep 14, 8:16�am, "naumanm" <masudnauman@n_o_s_p_a_m.yahoo.com> wrote:> Hi all > > I am working on low frequency sonar signal processing algorithms as my > final thesis. I have a confusion b/w time domain implementation of matched > filter (MF) process and frequency domain implementation of MF process. > > If i use LFM pulse replica of bandwidth W for target doppler (or speed) > finding, �i need multiple doppler shifted replicas/MFs according to > multiple target speeds for time domain MF implemetation as i > understand.i.e. I have to use bank of MFs and check for which MF, maximum > peak occurs. But if use frequency domain MF , do i still need multiple > replicas or single zero doppler shifted replica will be enough? > > Please help me . > If nay one knows any helping material on DFT based sonar doppler finding, > please let me know > > THANKS � �Hello, Just to clarify a point - the Doppler effect produced by a moving target on a LFM signal will affect the BW of the pulse. Whether or not that effect is negligible is dependent on the range of parameters you are looking at e.g. sonar centre frequency, original BW, and range of expected target velocities. Back to your question: It depends on what you mean by a frequency domain matched filter. If you are just multiplying the Fourier transform of the MF by the Fourier transform of the signal then, yes you'll need a replica for each doppler velocity you are interested in. The only other way I know to process LFM signals is generally called "deramp" or "stretch" processing. It is commonly used in Radar to reduce the A/D requirements of the front end. In this case you mix (multiply) the time domain signal another LFM signal having a chirp in the opposite direction. You can think of this as a basebanding operation with a linear time varying frequency shift. The output for the LFM pulse (if the chirp rates match) will be a sinsusoid - so then you simply take the FFT the finish the pulse compression. Targets at different ranges will appear at different FFT frequency bins. I don't think I've ever seen a sonar system implemented this way. Also, for different target velocities you'll need to use different chirp rates. As I recall now - most of the decorrelation due to different velocities is due to the different chirp rate of the return signal. The only book I can recommend is Neilson's "Sonar Signal Processing". As I recall he discusses frequency matched filtering by I recall having problems with his explanation since it wasn't mathematical. I believe there is an older IEEE paper (1970's - 1990's) which does a review of Sonar signal processing. The author may have been Burdic, but I am not sure. Rune, who posts to this news group, may be able to help. Cheers, Dave
Reply by ●September 14, 20102010-09-14
On Sep 14, 8:54�am, Dave <dspg...@netscape.net> wrote:> On Sep 14, 8:16�am, "naumanm" <masudnauman@n_o_s_p_a_m.yahoo.com> > wrote: > > > > > Hi all > > > I am working on low frequency sonar signal processing algorithms as my > > final thesis. I have a confusion b/w time domain implementation of matched > > filter (MF) process and frequency domain implementation of MF process. > > > If i use LFM pulse replica of bandwidth W for target doppler (or speed) > > finding, �i need multiple doppler shifted replicas/MFs according to > > multiple target speeds for time domain MF implemetation as i > > understand.i.e. I have to use bank of MFs and check for which MF, maximum > > peak occurs. But if use frequency domain MF , do i still need multiple > > replicas or single zero doppler shifted replica will be enough? > > > Please help me . > > If nay one knows any helping material on DFT based sonar doppler finding, > > please let me know > > > THANKS � � > > Hello, > Just to clarify a point - the Doppler effect produced by a moving > target on a LFM signal will affect the BW of the pulse. Whether or not > that effect is negligible is dependent on the range of parameters you > are looking at e.g. sonar centre frequency, original BW, and range of > expected target velocities. > > Back to your question: It depends on what you mean by a frequency > domain matched filter. If you are just multiplying the Fourier > transform of the MF by the Fourier transform of the signal then, yes > you'll need a replica for each doppler velocity you are interested in. > > The only other way I know to process LFM signals is generally called > "deramp" or "stretch" processing. It is commonly used in Radar to > reduce the A/D requirements of the front end. In this case you mix > (multiply) the time domain signal another LFM signal having a chirp in > the opposite direction. You can think of this as a basebanding > operation with a linear time varying frequency shift. The output for > the LFM pulse (if the chirp rates match) will be a sinsusoid - so then > you simply take the FFT the finish the pulse compression. �Targets at > different ranges will appear at different FFT frequency bins. > > I don't think I've ever seen a sonar system implemented this way. > Also, for different target velocities you'll need to use different > chirp rates. As I recall now - most of the decorrelation due to > different velocities is due to the different chirp rate of the return > signal. > > The only book I can recommend is Neilson's "Sonar Signal Processing". > As I recall he discusses frequency matched filtering by I recall > having problems with his explanation since it wasn't mathematical. I > believe there is an older IEEE paper (1970's - 1990's) which does a > review of Sonar signal processing. The author may have been Burdic, > but I am not sure. > > Rune, who posts to this news group, may be able to help. > > Cheers, > DaveI just did a quick search. The article I mentioned is "Digital Signal Processing of Sonar" by Knight, Pridham and Kay, Proc of IEEE, Vol 69 No 11 Nov. 1981. Cheers, Dave
Reply by ●September 14, 20102010-09-14
On Sep 14, 2:54�pm, Dave <dspg...@netscape.net> wrote:> On Sep 14, 8:16�am, "naumanm" <masudnauman@n_o_s_p_a_m.yahoo.com> > wrote: > > > > > > > Hi all > > > I am working on low frequency sonar signal processing algorithms as my > > final thesis. I have a confusion b/w time domain implementation of matched > > filter (MF) process and frequency domain implementation of MF process. > > > If i use LFM pulse replica of bandwidth W for target doppler (or speed) > > finding, �i need multiple doppler shifted replicas/MFs according to > > multiple target speeds for time domain MF implemetation as i > > understand.i.e. I have to use bank of MFs and check for which MF, maximum > > peak occurs. But if use frequency domain MF , do i still need multiple > > replicas or single zero doppler shifted replica will be enough? > > > Please help me . > > If nay one knows any helping material on DFT based sonar doppler finding, > > please let me know > > > THANKS � � > > Hello, > Just to clarify a point - the Doppler effect produced by a moving > target on a LFM signal will affect the BW of the pulse. Whether or not > that effect is negligible is dependent on the range of parameters you > are looking at e.g. sonar centre frequency, original BW, and range of > expected target velocities. > > Back to your question: It depends on what you mean by a frequency > domain matched filter. If you are just multiplying the Fourier > transform of the MF by the Fourier transform of the signal then, yes > you'll need a replica for each doppler velocity you are interested in. > > The only other way I know to process LFM signals is generally called > "deramp" or "stretch" processing. It is commonly used in Radar to > reduce the A/D requirements of the front end. In this case you mix > (multiply) the time domain signal another LFM signal having a chirp in > the opposite direction. You can think of this as a basebanding > operation with a linear time varying frequency shift. The output for > the LFM pulse (if the chirp rates match) will be a sinsusoid - so then > you simply take the FFT the finish the pulse compression. �Targets at > different ranges will appear at different FFT frequency bins. > > I don't think I've ever seen a sonar system implemented this way. > Also, for different target velocities you'll need to use different > chirp rates. As I recall now - most of the decorrelation due to > different velocities is due to the different chirp rate of the return > signal. > > The only book I can recommend is Neilson's "Sonar Signal Processing". > As I recall he discusses frequency matched filtering by I recall > having problems with his explanation since it wasn't mathematical. I > believe there is an older IEEE paper (1970's - 1990's) which does a > review of Sonar signal processing. The author may have been Burdic, > but I am not sure. > > Rune, who posts to this news group, may be able to help.Hmmm... don't know 'bout that - my gut feeling is that there are other problems that mess the signal up far worse than Doppler, when bandwidths become large. In that case I'd expect various frequency- dependent factors like attenuation, scattering and dispersion to start to become significant, which would mean that there is little hope of getting a simple, generic answer anyway. Doppler would probably be the least of your worries. (I know, the OP asked for help with his thesis, but academia and the Real World ar two different places, entirely.) Rune
Reply by ●September 14, 20102010-09-14
naumanm wrote:> Hi all > > I am working on low frequency sonar signal processing algorithms as my > final thesis. I have a confusion b/w time domain implementation of matched > filter (MF) process and frequency domain implementation of MF process. > > If i use LFM pulse replica of bandwidth W for target doppler (or speed) > finding, i need multiple doppler shifted replicas/MFs according to > multiple target speeds for time domain MF implemetation as i > understand.i.e. I have to use bank of MFs and check for which MF, maximum > peak occurs. But if use frequency domain MF , do i still need multiple > replicas or single zero doppler shifted replica will be enough? > > Please help me . > If nay one knows any helping material on DFT based sonar doppler finding, > please let me knowThe ambiguity function of the linear chirp signal is no good for Doppler measurement. The practical way to avoid the computational nastiness is select the signal parameters so the Doppler shift x signal duration is small number, and get the target speed from multiple measurements of the distance. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●September 14, 20102010-09-14
>On Sep 14, 8:54=A0am, Dave <dspg...@netscape.net> wrote: >> On Sep 14, 8:16=A0am, "naumanm" <masudnauman@n_o_s_p_a_m.yahoo.com> >> wrote: >> >> >> >> > Hi all >> >> > I am working on low frequency sonar signal processing algorithms asmy>> > final thesis. I have a confusion b/w time domain implementation ofmatc=>hed >> > filter (MF) process and frequency domain implementation of MFprocess.>> >> > If i use LFM pulse replica of bandwidth W for target doppler (orspeed)>> > finding, =A0i need multiple doppler shifted replicas/MFs according to >> > multiple target speeds for time domain MF implemetation as i >> > understand.i.e. I have to use bank of MFs and check for which MF,maxim=>um >> > peak occurs. But if use frequency domain MF , do i still needmultiple>> > replicas or single zero doppler shifted replica will be enough? >> >> > Please help me . >> > If nay one knows any helping material on DFT based sonar dopplerfindin=>g, >> > please let me know >> >> > THANKS =A0 =A0 >> >> Hello, >> Just to clarify a point - the Doppler effect produced by a moving >> target on a LFM signal will affect the BW of the pulse. Whether or not >> that effect is negligible is dependent on the range of parameters you >> are looking at e.g. sonar centre frequency, original BW, and range of >> expected target velocities. >> >> Back to your question: It depends on what you mean by a frequency >> domain matched filter. If you are just multiplying the Fourier >> transform of the MF by the Fourier transform of the signal then, yes >> you'll need a replica for each doppler velocity you are interested in. >> >> The only other way I know to process LFM signals is generally called >> "deramp" or "stretch" processing. It is commonly used in Radar to >> reduce the A/D requirements of the front end. In this case you mix >> (multiply) the time domain signal another LFM signal having a chirp in >> the opposite direction. You can think of this as a basebanding >> operation with a linear time varying frequency shift. The output for >> the LFM pulse (if the chirp rates match) will be a sinsusoid - so then >> you simply take the FFT the finish the pulse compression. =A0Targets at >> different ranges will appear at different FFT frequency bins. >> >> I don't think I've ever seen a sonar system implemented this way. >> Also, for different target velocities you'll need to use different >> chirp rates. As I recall now - most of the decorrelation due to >> different velocities is due to the different chirp rate of the return >> signal. >> >> The only book I can recommend is Neilson's "Sonar Signal Processing". >> As I recall he discusses frequency matched filtering by I recall >> having problems with his explanation since it wasn't mathematical. I >> believe there is an older IEEE paper (1970's - 1990's) which does a >> review of Sonar signal processing. The author may have been Burdic, >> but I am not sure. >> >> Rune, who posts to this news group, may be able to help. >> >> Cheers, >> Dave > >I just did a quick search. The article I mentioned is "Digital Signal >Processing of Sonar" by Knight, Pridham and Kay, Proc of IEEE, Vol 69 >No 11 Nov. 1981. > >Cheers, >Dave >Thanks Dave and Rune for quick reply. Yes i am simply multiplying FFT of beam data signal with FFT of replica pulse.Ok if I use multiple replicas/MFs, at each MF output, i get power spectrum at a range of frequencies.How should i compare these spectrums (not manually but using some algorithm) to find replica having maximum output? Second if i have multiple beams (say 128 beams) then for each beam i have to use multiple replicas/MFs. Will not it be a very computationally intense process if implemented in real time? Any suggestions are highly appreciated. THANKS
Reply by ●September 14, 20102010-09-14
> > >naumanm wrote: >> Hi all >> >> I am working on low frequency sonar signal processing algorithms as my >> final thesis. I have a confusion b/w time domain implementation ofmatched>> filter (MF) process and frequency domain implementation of MF process. >> >> If i use LFM pulse replica of bandwidth W for target doppler (or speed) >> finding, i need multiple doppler shifted replicas/MFs according to >> multiple target speeds for time domain MF implemetation as i >> understand.i.e. I have to use bank of MFs and check for which MF,maximum>> peak occurs. But if use frequency domain MF , do i still need multiple >> replicas or single zero doppler shifted replica will be enough? >> >> Please help me . >> If nay one knows any helping material on DFT based sonar dopplerfinding,>> please let me know > >The ambiguity function of the linear chirp signal is no good for Doppler >measurement. The practical way to avoid the computational nastiness is >select the signal parameters so the Doppler shift x signal duration is >small number, and get the target speed from multiple measurements of the >distance. > > >Vladimir Vassilevsky >DSP and Mixed Signal Design Consultant >http://www.abvolt.com >Thanks Vladimir for reply. Can you kindly explain multiple measurements of the distance? Do you mean to transmit no of pings say 15-20, for each ping perform MF with zero doppler shifted replica to find target range and from analysis of 15-20 target distances measured, extract speed of target?? Best Regards nauman
Reply by ●September 14, 20102010-09-14
On 9/14/2010 7:48 AM, Vladimir Vassilevsky wrote:> > > naumanm wrote: >> Hi all >> >> I am working on low frequency sonar signal processing algorithms as my >> final thesis. I have a confusion b/w time domain implementation of >> matched >> filter (MF) process and frequency domain implementation of MF process. >> >> If i use LFM pulse replica of bandwidth W for target doppler (or speed) >> finding, i need multiple doppler shifted replicas/MFs according to >> multiple target speeds for time domain MF implemetation as i >> understand.i.e. I have to use bank of MFs and check for which MF, maximum >> peak occurs. But if use frequency domain MF , do i still need multiple >> replicas or single zero doppler shifted replica will be enough? >> >> Please help me . >> If nay one knows any helping material on DFT based sonar doppler finding, >> please let me know > > The ambiguity function of the linear chirp signal is no good for Doppler > measurement. The practical way to avoid the computational nastiness is > select the signal parameters so the Doppler shift x signal duration is > small number, and get the target speed from multiple measurements of the > distance. > > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultant > http://www.abvolt.comAs a practical matter, LFM in low frequency sonar isn't very likely to get you the results you hope for. But, one can always do a theoretical treatment of it. Usually the bandwidth isn't wide enough to spread the tails of the ambiguity function out so that target SNR (really SRR) is adequate - just as a matter of the waveform alone and not to mention the nasties in an underwater environment. The range-rate approach only works in radar - and it's "low frequency sonar" which implies large round-trip times and sometimes "single ping solutions" depending on the platform. Sonar round-trip times are generally too long to allow this in most practical circumstances. And, range resolution may or may not be good enough to do a good job of differentiation (always a noisy process) - which is what this approach requires. If the receiver bandwidth is narrow enough (as it often is just wide enough to admit expected target radial velocities) then a single replica may be OK as Doppler doesn't affect the waveform all that much. Consider: speed of sound around 1,000 fps and 50fps target velocity then the Doppler shift is not more than 5% - which doesn't decorrelate all that much using a single replica. That's easy to simulate. I think what you hope to gain in SRR is traded off against range-Doppler ambiguity. In fact, isn't that where "ambiguity function" got its name in history? One needs to ask (and answer): "Why LFM?" But, it's a good academic exercise nonetheless. Fred
Reply by ●September 15, 20102010-09-15
On Sep 14, 5:05�pm, "naumanm" <masudnauman@n_o_s_p_a_m.yahoo.com> wrote:> Second if i have multiple beams (say 128 beams) then for each beam i have > to use multiple replicas/MFs. Will not it be a very computationally intense > process if implemented in real time?Sure. But that's merely a question of $$$. There is no fundamental limit of physics, messy environment variables, unknown parameters or anything like that, to prevent you. Rune
Reply by ●September 18, 20102010-09-18
>Back to your question: It depends on what you mean by a frequency >domain matched filter. If you are just multiplying the Fourier >transform of the MF by the Fourier transform of the signal then, yes >you'll need a replica for each doppler velocity you are interested in. > >I have got an IEEE paper by H. Glisson, C. I. Black and A. P. Page, “On Digital Replica Correlation Algorithms with Applications to Active Sonar”.According to this paper as i understand, multiple replicas needed only when Wideband LFM pulse condition is satisfied,i.e. when target velocity > 2610/TW. if narrowband LFM pulse is used, then according to this paper single zero doppler replica is enough.But how in that case target velocity/doppler can be found? My real intention here is to know how to find target velocity using LFM pulse in SONAR?LFM pulse is chosen since it is more immune to reverberation as compared to CW pulses. If any one have read this paper, can he/she kindly clarify my confusion? THANKS
