How to filter an underwater acoustic signal for TDOA localisation

Hi,

I'd like to filter some ultrasonic signals in order to improve SNR on received signals. I'm trying to do an algorithm for underwater localisation based on the TDOA (Time Différential of Arrival) method. Actually, I use three kind of pass-band filters (filtfitlt(), bandpass() and filter()) to do that. However, I don't think I use properly those filters since when I apply a cross-correlation (gccphat() function) to the signals in order to get the TDOA , I don't get good results (time delay is almost always 0 which is false) so I can't estimate the localisation of the source. I don't have much experience doing digital signal processing, so any help would be very appreciated.

Regards,

Jorge.

[Star Strider]

It is best not to use filter other than with linear-phase FIR filters, since it introduces phase distortion.

The bandpass function produces 2 outputs, the filtered signal (first output) and the discrete filter it designs (second output). If you want to use the discrete filter object, use it with filtfilt (as the first argument), and the signal you want to filter as the second argument.

I have no experience with ‘TDOA’, so I cannot help you with that aspect of your problem.

[Jorge Polo]

Hi Star Strider,

In underwater acoustics, TDOA is basically a method used to estimate the position (bearing and range) of a sound source. In order to do that, once computes the difference on time of arrival of the sound wave at each sensor (or hydrophone) of the system. So, if we have 4 hydrophones, the TDOA between a reference hydrophone and the others would be:

where represents the acoustic signal time of arrival to the n hydrophone. Normally, TDOAs () should be found by cross-correlation so I use the gccphat() function in my script. However, since the sea is a very noisy channel, I have to do some filtering before doing correlation. When I do that, I don't get the good TDOAs. I use a bandpass to create a FIR with cutoff frequencies at 31kHz and 33kHz followed by a filtfilt function as you said, which improved my algorithm, so I thank you for that. However, when running the algorithme, sometimes it doesn't give the expected results (TDOA gives 0 for exemple). Could you give me any clue of how I could improve the filtering stage? The frequency of interest is 32 kHz.

Regards.

[Star Strider]

The filtfilt function is phase-neutral for all filters, so there should not be any phase distortion. The output should be a filtered version of the input, so the arrival times should not be affected. As I mentioned, I have no experience with ‘TDOA’, so I am only addressing your filter implementation.

Your best option may be to filter your signal before recording it, using a hardware filter. The Bessel filter is a phase-neutral (linear phase) filter that is frequently used to avoid phase distortion in the filtered signal. The cutoff frequency in a lowpass Bessel filter must be no higher than the Nyquist frequency (½ the sampling frequuency), although it may be lower.

[Amit Galor]

Hi Jorge,

I'm working on a similar problem for the purpose of localizing underwater sources. I'm currently contemplating between the pros and cons of TDOA compared with the DOA algorithm for this purpose.

Can you post your script to help me compare the results with what I get with DOA for sample signals?

I might also be able to help you after seeing your noise filters implementation, we implemented de-noising with pca and bandpass filters.

Kind regards,

Amit