Measuring compression/limiting?

[Multicore]

Some active speakers use processing to limit signals sent to drivers, and presumably such processing is engineered taking things like driver and amp characteristics into account (perhaps also subjective and/or psycho-acoustic knowledge).

If a design uses something like multi-band limiting (e.g. a band for each driver) with a slow enough attack and decay then it might not look non-linear in some distortion tests.
But it might be interesting to see it's behavior in a graph. For example, a graph with input signal level on X and SPL on Y might reveal onset of limiting/compression, useful information about the performance capacity of the product.

In Amir's standard reports so far we see distortion as a function of frequency at two or three different SPLs. These are among the most interesting graphs for me after the frequency response as they suggest how much acoustic energy the speakers can emit. But I wonder if some kinds of processing might hide compression so it doesn't look like distortion as I think Amir sometimes notes in the text.

Would such graphs interest you? Could Klippel hardware with suitable software do it as a standard test protocol?

 

[Blumlein 88]

I would like to see such testing added as well if it's doable.

 

[fpitas]

I've seen it done. The input is stepped up while the FR is measured. Obviously you have to be careful you don't saturate the mic etc.

 

[Multicore]

Does REW have this capability? I'm sure Klipple hw could do a much better job but ...

 

[Curvature]

Compression is basically comparing FR vs. level. Any limiting built into active monitors will show up at some point, once you push hard enough.

 

[amirm]

In Amir's standard reports so far we see distortion as a function of frequency at two or three different SPLs. These are among the most interesting graphs for me after the frequency response as they suggest how much acoustic energy the speakers can emit. But I wonder if some kinds of processing might hide compression so it doesn't look like distortion as I think Amir sometimes notes in the text.

Any compression shows up as frequency response droop:

The challenge here is doing something that is representative of real life usage. Above test for example is asking the tweeter to play at the same 106 dBSPL as the bass.

The limiting could also have state/history. I remember the Apple smart speaker limiting after N number of frequency sweeps. I suspect it was a thermal limit.

Klippel does have a module for SPL vs response but it follows a standard which I do not like per above.

 

[Multicore]

The challenge here is doing something that is representative of real life usage.

Yes, exactly.

Above test for example is asking the tweeter to play at the same 106 dBSPL as the bass.

Which is indeed not very realistic.

The art of compression (in the studio) is the tuning/adjustment of time constants and levels to musical dynamics. Automatic compression can be very sophisticated these days but in a speaker the program material is unknown and the latency has to be very low (so no look-ahead). I think we can assume it would be designed 1) for music signals, 2) to do nothing if the input is low enough, 3) to limit power averaged over some suitable time.

If the designer does a good job at this then we'd want to be able to see that and approve, nodding like sage audio engineers at the measurements, seeing that it is minimally intrusive to listening while effective in mitigating the ugly sound or potential damage of over driving the physical limits of the speaker's components. Or frowning and tut-tutting over how it should be better. What might such data look like?

If I were to research this, I might try choosing an actual music recording, something very dynamic with full frequency (perhaps orchestra with piano and percussion). Then find an input level that produces the speaker's max SPL for that music, which would have to be some peak (time windowed) distortion measure. Then step the input level down from there while measuring output power (peak?) divided into, say, three frequency bands, low, mid, and high, and draw them curves on a graph with input step level on the X axis.

I venture this for the sake of discussion knowing it may be foolish. If I were to experiment with REW and my UMIK1 then I'd need to find a time when my neighbors upstairs and downstairs, family and dogs were away. It's bad enough doing sweeps for room EQ.

 

[Curvature]

If I were to research this, I might try choosing an actual music recording, something very dynamic with full frequency (perhaps orchestra with piano and percussion). Then find an input level that produces the speaker's max SPL for that music, which would have to be some peak (time windowed) distortion measure. Then step the input level down from there while measuring output power (peak?) divided into, say, three frequency bands, low, mid, and high, and draw them curves on a graph with input step level on the X axis.

That would work for instantaneous compression. Long term thermal compression would be another issue.

A solution here would be to use M noise as the source signal.

 

[Multicore]

A solution here would be to use M noise as the source signal.

Very interesting.

M-Noise

M-Noise is a new test signal that promotes standardized measurement of a loudspeaker system’s maximum linear output.

m-noise.org

 

[gnarly]

Steady state thermal compression is pretty easy to measure just looking for frequency response regions that don't increase linearly with signal....pretty easy... albeit LOUD lol.
Meyer's M-Noise has a crest factor that increases as frequency increases, that is supposed to replicate crest factors found in music better than a constant crest factor.

I think drooping frequency response regions from thermal compression is a relatively long term issue, compared to short term transient limiting.
And I think short term transient limiting is most likely the bigger SQ problem in home audio, than compression.

Unfortunately, measuring that problem means peak SPL needs to be measured freq by freq, and the time period of peak integration has to be defined.
I mean, is peak either SPL slow, fast, or truly 1 sample peak.

Using tone burst pulses and SPL meters set to either slow, fast, or peak; it's pretty easy to see how SPL reads higher as meter speed increases, using the same burst for repeated measurement. True peak, single sample, always reads highest.
It's also easy to see when that true peak begins to fail to increase linearly with increased drive level. (but again, unfortunately only on a single frequency at a time)

I usually just choose a a few frequencies to test that are near the bottom range of each driver's passband. With the idea that those frequencies are where excursion is maximum, and amplifier demands the greatest. Seems they should be the first to loose peak linearity.
So far, I'd say higher SPL peak linearity correlates very well with perceived SQ improvements. Very well. My 2c...

 

[Multicore]

I think drooping frequency response regions from thermal compression is a relatively long term issue, compared to short term transient limiting.
And I think short term transient limiting is most likely the bigger SQ problem in home audio, than compression.

Yes. That's what I had in mind when writing the OP.

To the extent I understood the rest of your post, it seems it could be elaborated into an automatic test with the Klippel hw. Maybe it needs some help choosing the test tone frequencies but Amir already near-field scans individual drivers.

I like it.

 

[AwesomeSauce2015]

FWIW:
I know that Erin @ Erin'sAudioCorner has some form of compression test, which does show results. For example, his review of the ELAC Debut 2.0 B6.2, shows significant compression on the tweeter as level gets quite high.
However, Amir's method of including the FR in the distortion graphs has a similar effect, although he does not always test at "102db". (Which as a SPL nut, I would like to see more of).
But a long-term compression test... idk.

I do know, however, that I personally do find the information useful. I like having speakers that can hit high spl at distance, so that when I turn them down, they sound much cleaner than a speaker that is closer to its limit.