Noisy results from white noise input

[imperial-aero]

Hi,

I'm trying to send in white noise as input (using 8 loudspeakers) and measure the response from 8 microphones mounted around an annular tube. I carry out the FFT of the inputs (loudspeaker signals) and outputs(microphone signals) of the system and then find the transfer function. I use FFT averaging to reduce the noise variation. My problem is that my magnitude and phase plots are too noisy (pic attached), and was wondering whether it is something to do with the way I'm generating my white noise (VI attached).

I am expecting to observe a certain amount of noise in the signals but not as much as seen on the picture. I increased the amplitude (standard deviation of my Gaussian noise VI) but it did not improve the results and sending in Uniform White noise did not improve the results either nor did increasing the acquisition time.  Has anyone had similar issues with noise or have any suggestions to improve my results?

Thank you. 

[preston_johnson]

what daq input board are you using?  do you have low-pass anti-aliasing filters in place?  It appears your frequency of interest is in the 100's of hz range and your sample rate is 5kHz.  It is possible the noise comes from electromagnetic noise picked up by your sensing wires.  Also what microphones are you using?

[imperial-aero]

Thank you for the reply Preston,

I'm using a PCI-6229 board, with BNC-2090A terminal and SCC-68 for analog inputs and outputs.

I don't think the noise is actually from the devices because I first tested the whole set up with pure sinusoidal inputs and did not observe any noise. Now I'm sending in white noise because I want to excite the system at all frequencies below the Nyquist frequency (which is 2500 Hz as my sampling rate is 5000 Hz). Also, I'm interested in the behaviour of the system in the frequency range 0Hz to 700Hz.

The noise is probably due to the white noise input, but I expected them to be smaller than they are now. Also, like I said earlier, increasing the number of samples did not reduce the amount of noise. The microphones I'm using are just simple tie-clip ones but they are supposed to be good enough for my application. And I don't have any filters in place.

[johnsold]

It looks as though your excitation includes frequencies well above the Nyquist limit. Your VI had default fs on the AO clock and the noise generator set to 15000 Hz. I think to get meaningful results you may need to filter the noise before sending it to the AO Write.

If you look at the frequency response of filters, you will see that you need a high order filter or wider separation between the sampling frequency and the highest frequency you wish to excite. You indicated that the system response to 700 Hz is of interest. How much above 700 Hz do you need to excite with the noise to get good results? That will determine where to set the filter cutoff frequency. With a 4 pole Butterworth filter and a 700 Hz cutoff the response is down about 20 dB at 2500 Hz and has some visible attenuation at 500 Hz.

Lynn

[imperial-aero]

Thank you for the reply Lynn.

I've just been reading on Experimental input design and I found that (like you've mentioned) I need to filter my Gaussian white noise for the range I'm interested in. However, I don't know what the maximum frequency should be to get accurate results.

Also, I'm new to DSP and have never used filters. Do you know any useful references where I can learn the basics.

Cheers.

[imperial-aero]

I have modified my VI to filter the Gaussian noise I'm generating. I use a 5th order, low-pass Butterworth filter and I set my cut-off frequency to 700Hz. Like you mentioned I do observe some attenuation around 500Hz, but I'm not sure whether this affects my results. I have attached a pic of my results and my VI. As you can see from my Bode plots, the noise has been significantly reduced when compared to what I had before, however I still do obtain some noise in the results and I haven't been able to reduce it. I'd like some help on the following questions about my results

1. I'm not sure what the Sampling Rate of the filter does to my results. When it's at 15000Hz, I can see that the attenuation starts at around 500Hz. However, if I set the sampling rate to 5000Hz the attenuation starts at 2500Hz (same cut-off frequency of 700Hz was entered in both cases). However, even though these two sampling rates seems to affect the frequency at which singal attenuation occurs, it does not affect my results too much.

2. Like you've mentioned, I'm still not sure what cut-off frequency I should enter for my filter. I tried cut-off frequencies from 500Hz to 1200Hz. Again  this didn't affect the results much (there were only small differences in the Bode plots).

3. I tried changing the signal amplitude, the data acquisition time, order of the filter (from 5 to 4) without observing any improvements in the results.

4. I don't understand why the results are noisier for frequencies less than 50Hz and for frequencies greater than 450Hz. This was observed for all the tests I did. The only reason I can think of is that this might be a characteristic of the system.

5. You mentioned in your previous post that my fs on the AO clock is set to default. However, I have actually set it to 15000Hz. Am I missing something here?

I have changed various parameters of my VI (amplitude, acquistion time, order of filter, sampling rate of filter, type of noise (gaussian and uniform white noise)) but haven't been able to observe any significnat improvements in the results. Would be grateful if you could spot something I'm doing wrong or any further suggestions.

Cheers.

Message Edited by imperial-aero on 04-13-2008 10:43 AM

[johnsold]

1. The filter VIs work on arrays of numbers and do not actually use the frequency information for the filtering process. This is different from a physical filter where capacitor and inductor values must be changed to change the frequency. The sampling frequency information is used to scale the cutoff frequencies for the output display. The Sample Clock: Rate property node output should be wired to the noise generator and to the filter VIs.

2. Ideally the excitation signal spectrum should be flat over the entire bandwidth to which the device being tested can respond. Often this is impractical and a judgment call must be made as to how much is good enough.

3. 4. At low frequencies you may be seeing the effects of AC coupling in the microphone preamp. Try exciting the system with a sine wave at some of the frequencies where the noisy results occur and see what you get. For example 5-20 Hz and 450-480 Hz. Or slowly sweep the frequency across the frequency range. Also try just exciting one speaker at a time. Differences in the speaker responses might cause the assumptions made in the transfer function calculations to be invalid. Specifically the assumption that the acoustic excitation uniformly tracks the electrical signal input. I have seen speakers of the same type exhibit widely different responses.

5. I don't see what made me say that.

The after filtering image seems to show several resonances in the response. Some of the resonances are close enough together to have some coupling. This can make analysis and interpretation difficult. The single speaker excitation suggested above may help sort out whether it is speaker or response issues.

Lynn