"Vertical Spectrogram" Rescaling Issues

[Chris A]

There seems to be an issue with posting spectrograms in Windows 10 using the Y-axis as the frequency axis. When comparing spectrograms to various measurements, after I rescale the vertical axis for each of the measurements, when I return to prior measurements that I rescaled, the vertical axis seems to expand back to a more zoomed in scale. When I go to the "Limits" plot menu, when I select "fit to data" on the second measurement I've opened, it doesn't rescale the vertical axis. Instead, it remains in the uncommanded expanded vertical scale (zoomed in).

This behavior continues every time I return to prior measurements, making the job of comparing decay time responses vs. frequency via spectrograms virtually impossible. I have to do screen shots of the spectrograms to flip back and forth between them without the vertical scales going off again.

WHen I move to X-axis as the frequency axis, this problem seems to largely go away.

Chris

 

[John Mulcahy]

Thanks, I've fixed that for the next build.

 

[Chris A]

I downloaded the Beta version recently, but had to go back to the latest non-Beta version when I discovered that the data is still clipped on the vertical spectrogram axis, while the vertical scale itself seems to be okay now. This problem of course only occurs after rescaling the vertical axis to zoom in on a problem area, then try to zoom out again.

 

[John Mulcahy]

the data is still clipped on the vertical spectrogram axis

Please explain what you mean by that.

 

[Chris A]

Like this:

Here's the SPL & phase (transfer function) plot of that same measurement:

Apparently, the symptom appears now on the first time that the spectrogram is plotted.

Sorry for the delay--I had to re-install the Beta version (REW_windows-x64_5_20_14ea60.exe) once again to show the issue.

Chris

 

[John Mulcahy]

The plot has been generated for a total span of only 14 ms, there's nothing meaningful to show for lower frequencies.

 

[Chris A]

I see. This is, however, a change in behavior from the prior (released) version. So you do not plot anything below one wavelength now for the entire plot window...and you only calculate (or show) data for the plot window now, not the entire measurement...?

Chris

 

[John Mulcahy]

Wavelet data has always been generated only for the selected time range.

 

[Chris A]

...and yet something changed in the minimum plot scale from prior versions...

I guess that I can window in on the plot by generating the plot for, say, 100 ms, then letting it span two screens worth of real estate, then look only at one screen worth of spectrogram. It seems kind of clunky, however. (Just sayin'.)

Clearly I use the spectrogram view often enough to ask these questions, in order to see both excess phase or excess group delay and amplitude (SPL) with peak time curve on the same plot, and then zoom in on the data to see more than just excess group delay issues.

Thanks again,

Chris

 

[John Mulcahy]

Plot data could be generated for lower frequencies, but I'm not following what actionable information it is you feel resides there on a plot that has been generated for a 14 ms time span?

 

[Chris A]

I believe that you're equating "plot" with calculating limits of the spectrogram--i.e., you're discounting the need for visually zooming-in on the data by setting the current plot limits to be less than the time limit of data used for calculation. Those are two different quantities.

 

[John Mulcahy]

No, I'm considering the implementation of the various wavelet transforms in REW. The original wavelet mode has a somewhat novel approach using complex smoothing and the time shift properties of the FFT. It makes use of the entire windowed IR as a starting point and could show content down to the limit of that windowed span, though the utility of displaying the behaviour of the response over fractions of a wavelength eludes me. The Airy and Morlet implementations use a form of fast convolution. They take a segment of the IR according to the chosen spectrogram span. Using the entire windowed IR would be extremely inefficient as most of the convolution results would be discarded. They are not valid below the current lower frequency shown on the plot.

 

[Chris A]

Sorry - i found the horizontal scale buttons (zoom in/out) on the bottom right of the plot. For whatever reason, I had not considered that there was scaling on spectrogram plots--like the other plot types. I'm not sure where I got that notion, but the fact remains that it was the remaining piece of the puzzle for my use of spectrograms to better visualize what's occurring.

That solves my issue with zooming in on the first 10 ms or so of the horizontal axis, while still being able to see the shape of the peak energy time curve and any phasing trends of SPL occurring at low frequencies, while also being able to read whether the highest octaves (from 6-20 kHz) are also time aligned, and early reflections in and around the horn (in a full-range MEH or two-way/two-horn loudspeaker, for instance) are handled well or not (even if the extreme low frequencies are not really accurate, per se). It's important to being able to see the bigger picture--in one plot. While it may seem odd to you, this has helped me work around issues by avoiding using EQ, delay, etc. in DSP to try to correct the SPL vs. frequency, and to see when I'm done with a loudspeaker's dialing-in.

Chris