ProSoundWeb Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: 1 2 [3]   Go Down

Author Topic: Smaart or?  (Read 1291 times)

Mark Wilkinson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1019
Re: FFT Windowing
« Reply #20 on: September 09, 2020, 05:11:42 pm »

In broad, non-mathematical strokes:
Shorter time window = lower frequency resolution, suppresses influence of late arriving energy.
Frequency domain smoothing = lower frequency resolution, suppresses influence of late arriving energy.

Fine print: it must be complex smoothing, and it must be linear frequency smoothing, not fractional octave smoothing as is far more common. Without going too far down a very tedious rabbit hole, frequency-domain smoothing can give you a comparable result at HF while avoiding the LF data point loss that comes with time-domain windowing. There's actually a "hidden" feature in v8 that exists pretty much only to demonstrate this point: in Transfer Function options, check "Enable FTW" under "Advanced." That enables "Frequency-Domain Time Windowing", which is a frequency-domain linear complex smoothing function that's mathematically equivalent to windowing in the time domain. Input the half-window length in ms into the FTW field in the TF control bar, and you can compare that against a measurement taken without the setting but using "regular" smoothing, and also to a time-windowed measurement taken in IR mode if you choose. Details here, pages 6-8.

Michael, I think of gating and widowing as two different animals.
Windowing, i think of as how to handle the FFT block sizes being used for the mathematical computations...the stuff Calvert does that makes my head swim.
Gating, on the other hand, i think of as 'how long a time period will i let the measurement analyze'......and watch Jamie want to start cussing when it's brought up lol.

In the end, i guess frequency smoothing, gating, widowing .....all throw away data and look smoother.  Hopefully for good reasons.
Pls feel free to correct my thinking  :)
Logged

Russell Ault

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 825
  • Edmonton, AB
Re: FFT Windowing
« Reply #21 on: September 09, 2020, 05:13:40 pm »

[...] Fine print: it must be complex smoothing, and it must be linear frequency smoothing, not fractional octave smoothing as is far more common. [...]

...that was the part I was missing. Makes sense!

-Russ
Logged

Mark Wilkinson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1019
Re: FFT Windowing
« Reply #22 on: September 09, 2020, 05:31:13 pm »


I believe what Smaart does in "MTW" mode is to apply longer windows to lower frequencies to allow greater frequency resolution (in Hz, not octaves) at those frequencies at the expense of including more reflected sound.

Smaart is great for quick-and-dirty measurement in the field -- system health, sub alignment, and so forth. It's also nice for what I might call acoustic exploration where you move things around and can immediately see the effect. For detailed speaker measurements under more controlled conditions, as well as room acoustic measurements, where you really want to know what you're getting, I think there are better tools.

--Frank

Hi Frank, i think the MTW mode does allow greater frequency resolution down low, but at the same time minimizing reflected noise as frequency increases....iow, different time windows (perhaps i should say different effective gating vs freq).

I used to think better detailed speaker measurements could be gained with better tools than Smaart.
But speakers and room acoustics appear to be so coarse and so variable ..orders of magnitude grosser in their ability to perform or provide repeatable measurements .....
....than Smaart or any decent measurement program can provide given a constant stimulus ..
....that i'm being led to believe very fine measurements are pretty much point-in-time, point-in-space.... BS.
Logged

Frank Koenig

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1083
Re: FFT Windowing
« Reply #23 on: September 09, 2020, 05:33:09 pm »

More generally, we can think of there being three parts to this problem.

First is getting the best possible estimate of the system function (transfer function) subject to the noise and non-stationarity of the given measurement environment. This affects the choice of test signal (sine-sweep, pseudo-noise, easy jazz) and the estimation algorithm. The result is an impulse or (equivalently) complex frequency response.

Second is extracting the relevant parts, such as windowing out reflections, averaging multiple measurements, and smoothing. This blends into the third, which is representing the data in a perceptually meaningful way for the problem at hand, such as log-magnitude and phase plots, minimum phase and excess phase parts, group delay, and various time-frequency-energy representations. This last bit is very open ended.

--Frank
Logged
"Nature abhors a vacuum tube." -- John Pierce, Bell Labs

Frank Koenig

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1083
Re: FFT Windowing
« Reply #24 on: September 09, 2020, 05:50:35 pm »

Hi Frank, i think the MTW mode does allow greater frequency resolution down low, but at the same time minimizing reflected noise as frequency increases....iow, different time windows (perhaps i should say different effective gating vs freq).

I used to think better detailed speaker measurements could be gained with better tools than Smaart.
But speakers and room acoustics appear to be so coarse and so variable ..orders of magnitude grosser in their ability to perform or provide repeatable measurements .....
....than Smaart or any decent measurement program can provide given a constant stimulus ..
....that i'm being led to believe very fine measurements are pretty much point-in-time, point-in-space.... BS.

Not disagreeing. And Smaart does a good job. I just have a little trouble with their lack of transparency about what it's doing. It strikes me a bit dumbed-down. At least with ARTA it's all spelled out, it's easy to export the data, and the built-in analysis tools are good and I can check my own stuff against them. (And I've plagiarized from their manual ;) ) --Frank


Logged
"Nature abhors a vacuum tube." -- John Pierce, Bell Labs

Michael Lawrence

  • Full Member
  • ***
  • Offline Offline
  • Posts: 114
Re: FFT Windowing
« Reply #25 on: September 09, 2020, 06:35:37 pm »

I just have a little trouble with their lack of transparency about what it's doing. It strikes me a bit dumbed-down.
Frank, can I ask what you feel is not transparent? Happy to help shed some light. MTW is no trade secret, Jamie spends quite a bit of time on it in class so people know exactly what it does (and thanks to COVID-19 that is on YouTube as well). See the attached image, courtesy of Merlijn, for a run-down on the time records in MTW.

Since you're more of an under-the-hood guy, you can also right-click any saved measurement to dump the measurement data into ASCII, drop that into Excel or program-of-your-choice and confirm the time windows yourself if you like.

But seriously, if you have questions, feel free to give Calvert a buzz - he loves to talk about that stuff and now that we're all working remotely, he's got no one to talk to about it  ;D  Happy to connect you.
Logged

Frank Koenig

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1083
Re: FFT Windowing
« Reply #26 on: September 09, 2020, 07:16:28 pm »

Frank, can I ask what you feel is not transparent? Happy to help shed some light. MTW is no trade secret, Jamie spends quite a bit of time on it in class so people know exactly what it does (and thanks to COVID-19 that is on YouTube as well). See the attached image, courtesy of Merlijn, for a run-down on the time records in MTW.

Since you're more of an under-the-hood guy, you can also right-click any saved measurement to dump the measurement data into ASCII, drop that into Excel or program-of-your-choice and confirm the time windows yourself if you like.

But seriously, if you have questions, feel free to give Calvert a buzz - he loves to talk about that stuff and now that we're all working remotely, he's got no one to talk to about it  ;D  Happy to connect you.

Michael, thank you so much. I clearly need to revisit the Smaart material that's available now.

I never saw that table before but maybe I just didn't look hard enough. It certainly shows the bands, much as I suspected. I bet they correspond to the steps you see in the coherence plot when you have the delay setting way off.

In fairness, I suppose Rational makes money off the training so they don't want to give away all the jewels in a super-complete manual. And I appreciate the easy-to-use aspect. There's no reason for most folks to have to worry about a lot of settings that don't make much difference in their day-to-day audio testing.

--Frank


Logged
"Nature abhors a vacuum tube." -- John Pierce, Bell Labs

Michael Lawrence

  • Full Member
  • ***
  • Offline Offline
  • Posts: 114
Re: FFT Windowing
« Reply #27 on: September 09, 2020, 07:51:45 pm »

Michael, thank you so much. I clearly need to revisit the Smaart material that's available now.

I never saw that table before but maybe I just didn't look hard enough. It certainly shows the bands, much as I suspected. I bet they correspond to the steps you see in the coherence plot when you have the delay setting way off.

In fairness, I suppose Rational makes money off the training so they don't want to give away all the jewels in a super-complete manual. And I appreciate the easy-to-use aspect. There's no reason for most folks to have to worry about a lot of settings that don't make much difference in their day-to-day audio testing.

--Frank

No problem, Frank. Most of this info is 'free for the asking' to interested parties - very, very little is actually proprietary in terms of number crunching. It's just that, exactly as you stated - for the vast majority of users, there is no interest or desire (or need) to learn it at that level. It's about getting actionable results quickly - how to use the MRI machine rather than how to build one. That being said, there are a lot of hidden "tweakhead" things in there if you know where you look ;)

m

EDIT: And, yes, exactly, the stairstep drops in coherence with wrong measurement delay show the "edges" of the time records. If you dump the data into Excel and subtract adjacent frequency points, you can see exactly where the seams are.
« Last Edit: September 09, 2020, 07:54:07 pm by Michael Lawrence »
Logged

ProSoundWeb Community

Re: FFT Windowing
« Reply #27 on: September 09, 2020, 07:51:45 pm »


Pages: 1 2 [3]   Go Up
 



Page created in 0.046 seconds with 23 queries.