FSAF (Fast subband adaptive filtering) measurement

[Tikkidy]

OK so I'm going to take measurements with the microphone and the speaker in the chain. But I wonder why I am getting different results with the Stepped Sine and Log Sweep:

Both taken with microphone 31.6cm from the 5.25" mid-woofer, in vented speaker of 7L. The fundamental in green shows the port tuning frequency of around 45Hz. When taking Sweeps and distortion, the log sine sweep is smoothed at 1/24th octave, I got that from the manual.

On the other hand, the stepped sine is not smoothed. So is the stepped sine more authentic?
After all, the stepped sine seems to reject noise better and can resolve harmonics below the noise floor.

But then the log sweep looks smoother from 2KHz upwards My gate is 9ms, but I thought this only affects the SPL & Phase view... . Is this the result of graphical antialiasing? Or something else?

 

[John Mulcahy]

When taking Sweeps and distortion, the log sine sweep is smoothed at 1/24th octave, I got that from the manual.

Sweep distortion results have 1/24 octave smoothing. Distortion results can be affected by driver temperature (and prior motion of the driver) and hence by the history of the signals the driver has seen.

 

[Tikkidy]

OK I'm with you.

What about just the frequency response?

Does the 10ms gate create a smoother appearance eg. 2KHz to 10KHz?

Which one is more authentic?

 

[John Mulcahy]

Does the 10ms gate create a smoother appearance

Yes, that's why it says "Freq. resolution 100.00 Hz" below your 10 ms right window setting.

Which one is more authentic?

Meaning what? If you only want the driver response and not room contributions you would need to window out reflections and/or measure in an environment free of reflections and probably measure with the mic a lot closer than your "31.6 cm" annotation indicates. The stepped measurements can't exclude the effects of reflections, but again measuring a lot closer would help.

 

[Tikkidy]

Authentic being true to the original. I'm guessing all these measurements are just estimates of the true frequency response because they're affected by room reflections.

So am I correct then in understanding that a 100Hz resolution means that there are really only 10 data points between 2KHz and 3Khz then? And between 500Hz and 2KHz (2 octave span) only 15 data points? So the Farina exponential sweep shows a smoother response by very nature of Sweep, not because of any graphical interpolation or anti-aliasing by REW. Meanwhile the stepped sine shows a less smooth response due room reflections.

Certainly, I can move the mic closer to get more signal and less room, perhaps to the diameter of the driver cone?

I'm proceeding with caution because, you know, GIGO. When I can get a close match with frequency response with stepped sine and Sweep, then I'll do FSAF and we can the compare all three!

 

[John Mulcahy]

So am I correct then in understanding that a 100Hz resolution means that there are really only 10 data points between 2KHz and 3Khz then?

Absolutely not. It is more like applying a smoothing filter that is 100 Hz wide to a set of data with points every few tenths of a Hz.

Certainly, moving the mic closer helps, perhaps to the diameter of the driver cone?

Depends what you are trying to achieve. For low frequencies it would be common to do a "near field measurement". Results are valid to approx 11 kHz/effective cone diameter in cm, measurement distance not more than effective diameter in cm/20. For higher frequencies measuring distance would depend on whether you are trying to measure the combined effect of multiple drivers or a single driver. For single drivers common would be to place the driver in a large baffle and measure at 1m, with the measurement setup as far as possible from all room surfaces or outside at height.

 

[dcibel]

Hi John,
For this FSAF measurement process, would it be possible to name the saved WAV files with some meaningful names, rather than long string of numbers and letters? Just some naming based on measurement name would be great, and add -1, -2, -3 to the suffix for duplicates.

I also think the "set FSAF folder" option should be moved to the preferences window, main problem in its current location is that its an option that's unavailable until a measurement is made or loaded.

Optionally, some automation to compress wav files to FLAC for space saving would be great too.

 

[John Mulcahy]

The names use the UUID of the measurement they belong to, otherwise it would be difficult to ensure the WAV files could be unambiguously linked to their source.

 

[dcibel]

If this UUID is important, I believe that if you encode the files as FLAC, I believe you could embed the UUID string as metadata and use a human readable file name. Perhaps that makes things overly complicated on the programming end, but from an end user standpoint, if I've taken multiple measurements, it's quite difficult for me to determine which WAV file in the folder I am looking for when I want to review.

 

[John Mulcahy]

REW knows and will load the relevant file for you, and you can see the UUID on the Info dialog for each measurement. I will look at the flac options. The main issue is that measurement names are often not unique.

 

[dcibel]

Thanks John, I will pull the UUID from the measurement info for now.

 

[sm52]

dcibel,​

hello. How does using FSAF help you? Can you show with an example?

 

[dcibel]

FSAF helps in the way that all measurements help. They provide insight to understand the characteristics of a driver or system, and provide metrics of comparison between any change made to that system. I like that the test can be run with either pink noise or any audio track, really any test that doesn't involve an annoying sine sweep is preferred for me. Forgetting about distortion components, FSAF allows REW to measure impulse response with noise stimulus, which is a great addition to see.

FSAF chart is nice in that it encompasses harmonic distortion, inter-modulation, and noise into a single result. I am still working on overall comprehension, it is not fully clear to me yet how the difference between high order and low order distortion components affect FSAF as an overall metric.

But the magic here isn't so much in the chart itself, you have the ability to run any audio through FSAF and listen to the "residual" component of distortion and noise. This is a highly useful tool to provide correlation between test result graphs to actual audible effects.

 

[sm52]

you have the ability to run any audio through FSAF

Yes. Now no one can say that music is not a sine wave, or music is not an impulse. Because it is music that passes through the measuring system. And the result of the passage is recorded. Maybe with the help of FSAF you can be more confident that the configured system will play better than other options.

 

[Tikkidy]

Dear John,

I notice that in the current beta release one may apply HP and LP filters to the test file for the FSAF testing.
In the Sweep testing, complete PEQ and complex filters can be used.

Is it possible to implement this type of testing with filters in place, for FSAF testing, in the future?

The reason I ask is... suppose I want to play a music sample through 2 drivers that I've equalized to have the same on-axis response...

 

[John Mulcahy]

I notice that in the current beta release one may apply HP and LP filters to the test file for the FSAF testing.
In the Sweep testing, complete PEQ and complex filters can be used.

Is it possible to implement this type of testing with filters in place, for FSAF testing, in the future?

I'll add that to the feature request list.

 

[dcibel]

I second this request. EQ "measure with these filters" is an excellent feature for either sweep or FSAF test. I believe it could replace the current simple filter drop-down option entirely.

 

[Tikkidy]

First up is the Focusrite Scarlett 2i2(3rd generation) in loopback mode.
ie. Output 1 -> Input 1.

The "residual" distortion of two musical clips, amplified by 36dB, for your listening pleasure.

What, you can’t hear anything?
Come-on, connect your headphones and listen again.

Next up, I insert an amplifier into the loop back measurement. So now the signal chain is:
Scarlett 2i2(3rd generation) output -> Hypex UcD400MP -> Scarlett 2i2(3rd gen) input.

Again, the residual" distortion, amplified by 36dB...

Have a listen.
What do you notice?

PS. Sorry about the files sizes. They are 32bit float…

 

[sm52]

connect your headphones and listen again.

I listened. In one case out of two there is only noise. in another there is a very quiet piece of music against a background of noise. What was the goal?

 

[Tikkidy]

Yes, that’s what I heard.

Is this the same as the "the audio interface is transparent” ?

The amplifier UCD400MP is not quite?

Here are the 2 same tracks with virtual audio cable loopback ie. Digital out to digital in…

 

[dcibel]

I think this FSAF really needs a "loading" or "processing" indicator on the measurement window, as it takes a fair bit of time when loading an external audio file, or changing filter settings.

 

[Tikkidy]

Yes. I was thinking I might have upgrade my 10W sipping silent computer.

MOAR POWAH!!

What CPU/GPU doth one need?

 

[dcibel]

If you're just looking for an excuse to upgrade...

One more request might be a decimal place on the test duration box. When I load a 5.5second audio file, it looks like it's going to cut off the last 0.5 second.

 

[Tikkidy]

I'm one of those minimum viable solution guys... a timer or spinning/thinking wheel to show my computer hasn't crashed would be nice. Donation coming @John Mulcahy

 

[John Mulcahy]

I have added a busy indication for the next build.