I am working on colored noise (reverberation) rejection algorithms. One approach present in literature is through prewhitening filter. In this approach as per my understanding, first we estimate reverberation spectrum through the data set in which no target is present i.e. we estimate ar coefficients (through auto regression) from this data set. In the next step, we use these ar coefficients in inverse filter to remove the reverberation from the next incoming data which can contain both reverberation as well as target.
I have done some MATLAB coding for this like:
here single_ping_data is the data set containing both reverberation and target. In the first ten" pulse_samples" chunks, there is only reverberation present so I estimate ar coefficients from this data set as a_.
In the next step, i formulate a combine signal as:
and finally i applied inverse filter on combine signal as:
y = filter(1,a_,combine_signal);
I repeated this process in a loop up to complete length of " single_ping_data".
The problem is that when i applied match filter on this filtered data, i could not get better results as compared to ones i got without applying this filter.
Can any one kindly help me in this regard?
Perhaps the signal isn't present within the reverberation spectrum in any significant way? In that case any reverberation rejection would be rather useless - which is what you've described.
Thanks for reply. Target is present in the data for sure but after 10 no of data chunks. First 10 chunks contain reverberation only.
I didn't mean to ask if the target was present. I asked that you might consider if the target [when present] was within the reverberation spectrum [at the same time]. Decreasing reverberation does no good if the target isn't within it. It sounds like that could be your situation.
I have to ask the question. How can the first 10 chunks be reverberation only? Almost by definition, reverberation is delayed compared to the original signal.
Am I totally confused?
Another point. Reverberation does not only mean holes in the spectrum, it also means delayed copies of the source signal.
I remember going thorough what seemed like a very simple exercise related to reverb. Take a source signal and one delayed and attenuated copy of it. Even this simple case is very difficult to invert.
The first 10 chunks corresponds to a range (converted from time corresponding to 10 chunks) where no target was being placed.
Have you modified the matched filter also?
No, i implemented FFT based conventional match filter on reverberation filtered output
Do i need to modify the match filter and how to do it?
Yes, the matched filter should be modified. The same way you are filtering the signal you need to filter the waveform used to create the matched filter. May be there are some details that need to be considered but I do not have any reference book around. I can look at this again in a day or two.
Thanks for help.