How do I remove white noise in RT?

[ESengbus]

Hello, I am looking for a way to remove white noise from an input signal coming from a PXI board in real time.  My sample rate is 50 kHz.  Are there any functions built into LabVIEW to perform this operation.  I have not been able to find anything other than different low, high, or bandpass filters, and of course the white noise by definition has a nearly flat power spectrum and covers all frequencies.  If anyone has any suggestions, they would be greatly appreciated.  Thank you.

Evan Sengbusch

[TerraNova]

Well, measurements of this kind is often done by correlation functions.

Would you explain your problem in details - signal form, spectrum ets?

Best regards.

[Doug M]

Hi Evan,

When you say you are trying to remove noise in real-time, do you mean that you are actually running on a the ETS real-time OS using LabVIEW Real-Time module? If so, what specific hardware are you using?

[ESengbus]

The signal I am trying to monitor is mosty flat (with noise of course) inturrupted by spikes with multiple dominant frequencies (~kHz) that are of interest to me.  The noise is mostly white, but not entirely, there are a few large frequency spikes on the power spectrum.

I am using two PXI 6289 A/D cards.  The PXI system has a 2.4 GHz Pentium IV Processor, 40 GB hard drive space, and its buffer size is 96 Mb.  Yes, I am using LabVIEW RT module on the PXI.

My aim is to remove noise in real time in order to compress the amount of data written to the buffer.  Currently we are able to record 12 differential voltage channels on the PXI at 50 kHz without overwriting data on the buffer.  We want to be able to record approximately 30 differential channels.  Since only a small portion of the signal is important (spikes), the idea is to employ some type of real time noise removal to flatten the signal that is not important and then only record data points above a certain threshold in order to reduce the amount of data written to the buffer to allow the acquisition of more channels.  A threshold method does not work with our current setup of because the noise has enough amplitude that a threshold level that would sufficiently compress the amount of data written would cause us to lose important information about the spikes.

Thank you very much for your time and help.

Evan

[TerraNova]

Looks like you deal with ultrasound probing, doesn't?

Suppose you've done the best you could to reject the noise before it is going to be measured.

As far as I can see you need something like "pre-trigger" reading-sample way of buffer filling. This feature presents in NI digital oscilloscopes (in NIScope), as far as I can remember, but I just can't tell you wether it does in DAQ :(.
Anyway it needs a triggering level so you should choose it - the spectrum analysis won't help you with your problem coz it would take all the (50 kHz)x(30 chans)=(1500 S/s)= (THREExPXI6289) and you only have TWO (doesn't you) :).

Also you can offcourse just reject your noise before it enters DAQ and return to looking for triggering level again.

Best regards.

[ESengbus]

OK, I have just found out that some of the information I have been given regarding my PXI Controller and the way the software is running was incorrect (I have recently inherited this project from somebody else).  As it turns out, the program was being run with Windows on a laptop, not embedded in the PXI controller, and I believe this caused the buffer space issues.  I think I may be able to avoid all noise removal techniques and monitor all 30 channels simply by running the program in RT embedded on our PXI (which has 1GB RAM).  So, thank you very much for your input, but do not look too much further into the issue as I think I may be able to resolve my problem in a much simpler manner.  Of course, if I continue to have issues I will come back and post and your feedback would again be much appreciated.

Evan