k-Wave

1. bastlwastl
   

   Member
   Joined: Apr '14
   Posts: 7

   Hi,

   I'm working on an ultrasonic simulation and i would like to compare the result to actual data from our experiments. The general impression seems to be correct, but in detail, the frequency bandwidth of the used transducer plays an important role.
   As far as i know, the manual addresses this issue only quickly and in this forum, i found these 2 topics:
   http://www.k-wave.org/forum/topic/simple-3d-simulation-received-signal-question
   http://www.k-wave.org/forum/topic/sensor-specifications

   The second suggest to use gaussianfilter which is applying a frequency filtering (as far as i see). What seems to be correct compared to the spectra of the transducer provided by the manufacturer, results in clearly incorrect signals. Have a look here:
   http://homepages.physik.uni-muenchen.de/~Sebastian.Lehrack/gaussian_filter_spectra_3.5MHz.png
   http://homepages.physik.uni-muenchen.de/~Sebastian.Lehrack/compared_signals_filtered.png
   The gaussian filter results in a signal which differs from the data we measured. It also produces a signal appearing before the actual signal. In a second step, we applied Matlab's standard butterworth low and high pass filter and the result agrees much better with the measured data but yet not perfectly. However, it seems to me that the gaussianfilter is to simple and to strict regarding higher frequencies. And although it is not indicated, a transducer can have higher resonances, at least theoretically.

   Arising from this, what is the intended or recommended way to simulate the detector bandwidth?

   Greetings,
   Sebastian

   Posted 9 years ago #

2. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,640

   Hi Sebastion,

   The gaussianFilter function applies a simple non-causal or zero-phase filter using the FFT. If you navigate to the "Filtering A Delta Function Input Signal Example" under the "Numerical Analysis" heading in the html help, then open the third m-file (under the "Filtering the input time series" title), and add the following code around line 65:

   case 4
       % filter the input signal
       freq = 10e6;
       bandwidth = 50;
       source_func_filtered = gaussianFilter(source_func, 1/dt, freq, bandwidth);

   You can then compare the output of gaussianFilter with other types of filtering, and hopefully get a handle on what's happening. In particular, you'll notice that when using gaussianFilter or applyFilter/filterTimeSeries with the option 'ZeroPhase', true, the main features of the signal will still be in roughly the same place, but there might be oscillations ahead of time, i.e., the filter is not causal.

   If you compare this to applyFilter/filterTimeSeries with the option 'ZeroPhase', false, you'll notice there is a translation of the main features of the signal to ensure nothing appears ahead of time, i.e., the causal nature of the filter causes phase distortion.

   As you've noticed, the exact nature of the filtered output will depend on the filter shape. The best filter to use depends largely on what you're trying to achieve.

   Hope that helps,

   Brad.

   Posted 9 years ago #

RSS feed for this topic

Reply

You must log in to post.