The benefits are that you can see the signal change during the 120 second acquisition time. You could see a 200 hz signal oscillate back and forth for example. This is the approach taken by the JTFA (joint time frequency analysis) toolkit. They use gabor spectrograms that take gaussian windows of the data. There is a tradeoff between response time (ie time to put that frequency on the screen) and accuaracy (ie long data sets). The overlapping FFT with exponential averaging is a tradeoff but it all depends on the requirements of the job.
If you want real time display then you can't wait for that long 120 second acquisition time. Also if you want information about frequency changes within that 120 seconds. -Scott At 13:56 -0500 02/04/2004, Bruce Ammons wrote: >I don't see any benefit to averaging parts of the signal. If you >average within a single revolution, you will lose your phase >information. There is no benefit of averaging sections of repeated >signals either. Just do a large FFT for each revolution and average >them. > >FYI, you could average the signals themselves or the complex FFTs for >repeated signals. I don't recommend averaging magnitude and phase in >this case. > >Bruce > >------------------------------------------ >Bruce Ammons >Ammons Engineering >www.ammonsengineering.com >(810) 687-4288 Phone >(810) 687-6202 Fax > > > >-----Original Message----- >From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On >Behalf Of Gerd Rech >Sent: Wednesday, February 04, 2004 10:26 AM >To: Bruce Ammons >Cc: Info LabVIEW >Subject: AW: [W6.1] RMS averaging with "FFT Spectrum (Mag-Phase).vi" > > >I have a phase reference pulse once per rev which starts the >acquisition. So I should get consistent phase readings for repeated >measurements. However, I was wondering if it would make sense to average >parts of the signal(using overlapping segments) for speeding up the >averaging process. The idea is to make as much use as possible out of >the data that I get into the PC. > >Gerd > > >-----Ursprungliche Nachricht----- >Von: Bruce Ammons [mailto:[EMAIL PROTECTED] >Gesendet: Mittwoch, 4. Februar 2004 15:27 >An: 'Gerd Rech'; [EMAIL PROTECTED] >Betreff: RE: [W6.1] RMS averaging with "FFT Spectrum (Mag-Phase).vi" > > >The only way to get meaningful phase is to start your acquisition at the >same phase for every segment. An example of this would be an encoder >pulse. If each acquisition starts at a once per rev pulse from the >encoder, they will all have the same phase components. Another >possibility is doing order tracking, where there is a fixed number of >samples per revolution. If you shift your window by the number of >samples per revolution, you will get the same phase again. > >Any time you have a steady stream of data with no reference point, the >phase data is essentially useless. If it is a repeating signal you >should be able to identify a reference point. I suppose if you don't >have a hardware signal, you could use convolution to identify repeating >cycles of data. If phase is important, you should have a reference >signal. > >Bruce > >------------------------------------------ >Bruce Ammons >Ammons Engineering >www.ammonsengineering.com >(810) 687-4288 Phone >(810) 687-6202 Fax > > > >-----Original Message----- >From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On >Behalf Of Gerd Rech >Sent: Wednesday, February 04, 2004 3:36 AM >To: [EMAIL PROTECTED] >Subject: Re: [W6.1] RMS averaging with "FFT Spectrum (Mag-Phase).vi" > > >Hi folks, > >I know the "overlapping average" technology from signal analysers. In >those cases I used them up to now, phase was not relevant. Now I have an >idea about a different use where phase would be important. Scott's >comment that phase would be meaningless if overlapping averaging is used >is making me thinking. > >What about this: >1. Cut the long waveform stream into a number of (overlapping) pieces. >2. Use the FFT vi for each piece, which will produce amplitudes and >phases for each frequency bin. 3. Average apmlitude and phase for each >frequency bin separately. > >Would this create meaningful phase? >I would guess yes, as all pieces were acquired in a consistant stream of >data originally. > >Seems that my mathematical understanding does not reach far enough to >understand this completely. > >Cheers > >Gerd
