Separating Acoustic Waves from Background Reflections

[jcarsw04]

I'm trying to classify materials based on the reflected acoustic wave they produce. One transducer sends out an ultrasonic pulse to the material of interest and another transducer picks up the reflected waveform. Some of the wave is transmitted thru the material and reflected off of the background surface. How can I separate the reflected wave from the material of interest and the background reflections? Thank you for any help on this topic.

 

[DragonPetter]

I'll pass along something I picked up here a while ago.

Try researching the cepstrum technique and comb filters.

In the time domain, reflections are ideally just time delayed copies of the original signal with some distortion or attenuation, and they translate into the frequency domain as a periodic interference with the original signal's spectrum. The time delay of the reflection has a periodicity associated with it that is dependent on the length of the time delay, and you can see it in the frequency domain. You can extract these interferences with the inverse Fourier or with the cepstrum.

Between the time delay, attenuation, phase, and distortion information of the reflections you should be able to sort between original signals and different types of reflections.

There are time domain techniques that use cross correlation as well, but I don't know what would be best for what you're really trying to do, hence I just suggest researching those techniques to evaluate if they work for you.

 

[Studiot]

What is the wavelength and beamwidth of your pulse carrier?

If your angle of incidence is near 90° then the front and back reflected signals will follow the same path in space, but separated in time.

The greater the obliqueness of the incidence the more the two paths with diverge in space.
If you use highly oblique angles you may be then able to position the receiver to avoid the back reflected path.

 

[rbj]

first make sure that the spacing of the reflecting wall is sufficient that there is space in time between the impulse response from the reflection offa the material you are testing and the reflection offa the rear wall. if you cannot make that separation there is no way to separate the contribution of one from the other.

now assuming that the two impulse responses are short enough that they do not overlap, then using whatever method you use; perhaps a linear-frequency-swept-sinusoid (what we sometimes call a "chirp"), or perhaps some broadbanded driving signal and a 2-channel FFT, or perhaps a specifically generated noise with linear feedback shift register (LFSR), sometimes called "Maximum Length Sequences" or binary Galois sequence.

use any of these methods to get an impulse response (or a frequency response and then inverse DFT to get the impulse response), and then separate out the later reflection you don't want with a nice window, maybe even a rectangular window. then you have a sole and clean impulse response of the reflection of the material you're measuring. convert that to frequency response or whatever analysis you want. just treat it like a linear time-invariant system and drive the hypothetical system with whatever input you want.

 

[jcarsw04]

I like the idea of using the comb filter to cause destructive interference with the background reflection. It looks like there is still a lot I have to research. Is there a way for my receiver to automatically detect the start and end of the echo in software? If so, what techniques are typically used? Thank you again for your help.