DIAdem

cancel
Showing results for 
Search instead for 
Did you mean: 
Reply

joint time-frequency analysis tool help

Hi Brad,

Wow, your plot looks perfect!  It looks just as I would expect.  What did you do differently, exactly - was it the different choice of layout?  Am I just plotting the wrong thing?  I'll try playing around with the settings more.

Thanks so much for your help,

Myra

0 Kudos
Message 11 of 31
(1,266 Views)

Hi Myra,

 

I think it was just the settings, that's why I included the screenshot of the dialog for you to try the exact same settings and see if you get the same graph.  If you choose more mesh lines on the Time axis, that should indeed increase the time resolution, and if you choose more mesh lines on the Frequency axis, that will indeed increas the frequency resolution.  The thing to remember is that there's only so much resolution to go around, and the more you add to the X axis the more you take away from the Y axis (and vice versa).  Also, if you zoom into an X region or Y region with tighter Xmin/Xmax or Ymin/Ymax values, you will also see fewer mesh lines on the graph, though you don't lose any resolution that way.

 

Brad Turpin

DIAdem Product Support Engineer

National Instruments

0 Kudos
Message 12 of 31
(1,254 Views)

Hi Brad,

    I spent some more time changing around the settings, and am not sure I understand what I found.  I tried changing the values from mine to yours one at a time, but I could only reproduce your graph once I changed the time mesh from 60 to 73.  I don't understand why changing the time mesh by such a relatively small amount would produce such a large change.  How did you discover this value - was it trial and error, or do you have a way of calculating roughly what mesh would be best?  I imagine that there are some optimal values of the time and frequency meshes (for optimal time and frequency resolution), whereby increasing one necessarily means a decrease in the other - but this seems more like there is some sort of critical value?  Perhaps there is only a fairly small range of mesh sizes that return reasonable results, and using values beyond this range could describe the seemingly inconsistent results I found before (that increasing or decreasing the time mesh did not lead to a corresponding change in the time resolution).

    I ask because I want to compare the frequency dependence of different runs and of different engines, so I chose time meshes that were multiples of the total number of seconds - 60, 120, etc. to have a round number of intervals per unit time that I could use throughout the analysis.  Now I wonder if that is even necessary to keep that constant?  

    Do you think you can explain to me how the graph is put together - I'm sure that the way I described earlier is much too simplistic, as it doesn't even take into account the frequency mesh.  Finally, did anyone find out which method diadem uses to calculate this?  I'm guessing it's the short time Fourier transform, but it could be Gabor?

Thanks so much for your help, I really appreciate it! 

Myra

0 Kudos
Message 13 of 31
(1,247 Views)

Hi Myra,

 

This application was built originally in DIAdem 8.1 and has a lot of version dependence from back then.  That old version of DIAdem could only run FFTs that produced channels of lengths that were powers of 2, for instance.  If you select the desired X and Y mesh densities and X and Y start and stop values, then click on the grey (background) area of the dialog itself, you will see a % number appear in the lower left hand corner of the dialog.  This number tells you what % of the time series data set is being analyzed for that combination of mesh sizes and start and stop values.  You want that value to be as close to 100% as possible, so I often tweak the mesh sizes and click on the grey background until I can coax that number close to 100%.

 

I have been meaning to rewrite this application using features only available in modern DIAdem versions, such as the DFT that operates on any length time series data and always uses all of it.

 

This may just push me to do that,

Brad turpin

DIAdem Product Support Engineer

National Instruments

0 Kudos
Message 14 of 31
(1,238 Views)

Hi Brad,

   Thanks so much - I should be able to find my own time mesh values.  Now that that's settled, I can start working through all of my data sets.  I was very pleasantly surprised to find the JTFA in diadem because prior to that I was trying to write my own STFT script and that was not going so well.   I was wondering if you're familiar with the JTFA toolkit in Labview?  Do you think there would be much of an advantage in trying the other methods available in that package? It is supposed to come with an express analyzer vi which I hope would allow me to try out each method without having to do any programming.  Depending on the data set used, apparently different methods can provide higher frequency and time resolution, right?  That's why I was wondering which analysis method Diadem uses.

   I thought I should tell you how impressed I have been with diadem.  I started out trying to do my analysis in Signal Express, but it is extremely limited. My data sets are so large that it would be impossible to use excel - so I started using diadem.  It was definitely difficult to transition - diadem is not always straightforward, but it is so powerful that I continue to be amazed at what I can accomplish with it.  I'm also impressed with how great the graphs look - I still clean them up a bit in photoshop afterwards, but I think that now that is more out of habit than necessity.  Thanks!

Myra

0 Kudos
Message 15 of 31
(1,223 Views)

Hi Myra,

 

I have started updating the application you are using for DIAdem 11.1 and later.  The new version will include the DFT that uses any FFT length (not just powers of 2).  It will also switch the JTFA to having the Time on the Y axis and the Frequency on the X axis.  This will avoid an interpolation step which was required to invert the graph's data matrix, and it will also be compatible with the Contour Plot display in VIEW.  So it should work faster and produce cleaner results that you can even zoom into interactively in VIEW.  I have most of this working already.  Please feel free to send me a picture of one of your cleaned graphs, it may be that everyone will prefer the touch-ups you're doing, and I could then just include that as part of the new version.  In the meantime, you could always change the REPORT layout file that the application loads each time so that you don't have to make the same touch-ups over and over again.

 

I'm glad to hear you're liking DIAdem,

Brad Turpin

DIAdem Product Support Engineer

National Instruments

0 Kudos
Message 16 of 31
(1,215 Views)

Hi Myra,

 

The Joint Time Frequency Analysis part of the application you are using employs a straightforward time interval FFT.  You are choosing the time intervals when you select the mesh lines and start and stop times in the lower part of the dialog.  The start and stop frequencies just determine which part of the frequency spectrum not to graph-- it still gets calculated every time.  The middle part of the dialog determines the details of the FFT calculation, such as amplitude type, frequency averaging, window function, etc.

 

Another benefit of the new version should be that it gives more consistent results with minor changes to the mesh lines and start and stop values.  This will be due to always using close to 100% of the time series data with the DFT as well as avoiding the matrix inversion and interpolation to put the Time on the X axis and Frequency on the Y axis.

 

Brad Turpin

DIAdem Product Support Engineer

0 Kudos
Message 17 of 31
(1,212 Views)

Hi Myra,

 

Here's a sneak preview of the updated "Time Frequency Order Analysis" application for DIAdem 11.1 and later versions.  When it's fully tested and documented on the web, it will replace the older version here:

 

http://zone.ni.com/devzone/cda/tut/p/id/3549

 

The two big changes are both in the Joint Time Frequency Analysis section-- the FFTs now run with no power of 2 length restriction, and the results can be automatically sent to a VIEW contour plot (first available in DIAdem 11.1).  The latter change caused the Time and Frequency axes to switch places on the graphs.

 

Brad

0 Kudos
Message 18 of 31
(1,185 Views)

Hi,

I am working on time frequency analysis. i have implemented two techniques of time frequency analysis namely STFT and Gabor transform. But, i dont know how to meaure the frequency resolution in terms of numerical value over the time frequency pane. I requst you please give me idea so that i can easly discriminate which technique is better on the basis of frequency resolution.

 

Best Regards,

 

Roshan Kumar

0 Kudos
Message 19 of 31
(962 Views)

Hello roshan.iit123,

 

It is a little unclear what measurements you are looking to make in order to make the comparisons that you are referring to between the two frequency transforms. Are you looking for, as this forum discussion is referring to, the joint time frequency and order analysis tool?

Ian M.
National Instruments
0 Kudos
Message 20 of 31
(942 Views)