Is nonlinear distortion of (small) speakers unimportant??

[SIY]

@QMuse Standards notwithstanding, the Virtins MI software has a GedLee metric included. I've been playing with it extensively for an upcoming article, and as you know, it actually is based on audibility tests rather than, "Wow, look at this spectrum, it must be great/terrible/whatever!"

I am not comfortable with using nearfield for distortion; there are too many potential error sources. Extracting distortion from log-sin (Farina) sweeps strikes me as a much better method, albeit one that should be done in a quiet room (I have to put the AP equipment out in the hall for this because of the fan noise). Multitone is complicated because of room reflections- I've used it and published some, but at this point have concluded that it's one method that actually requires an anechoic environment or a quiet and wind-free outdoor location for useful results. In those circumstances, it may end up being a superb method, but there's no correlation to listening tests yet.

Other tests that can also be helpful and can be extracted from Farina sweeps include variations of rub/buzz and intelligibility tests like POLQA, ABC-MRT, and PESQ.

Of course, all of this assumes one is using a real lab microphone and mike preamp.

 

[QMuse]

@QMuse Standards notwithstanding, the Virtins MI software has a GedLee metric included. I've been playing with it extensively for an upcoming article, and as you know, it actually is based on audibility tests rather than, "Wow, look at this spectrum, it must be great/terrible/whatever!"

I didn't know that, nice to hear it, thank you! I will have a look at that software first time I get a chance.

I am not comfortable with using nearfield for distortion; there are too many potential error sources. Extracting distortion from log-sin (Farina) sweeps strikes me as a much better method, albeit one that should be done in a quiet room (I have to put the AP equipment out in the hall for this because of the fan noise). [/QUOTE]

I fully agree with you on that. Btw, wouldn't you think that stepped sine with RTA and appropriate windowing can be even more accurate than sweeps as each tone lasts longer so better chances for the in-room steady state to fully stabilise?

 

[SIY]

Btw, wouldn't you think that stepped sine with RTA and appropriate windowing can be even more accurate than sweeps as each tone lasts longer so better chances for the in-room steady state to fully stabilise?

I think that you don't want the room to be involved. So using Farina allows it to be removed by gating at midrange and higher. The tradeoff is bass, but the potential errors for nearfield at bass frequencies are less since the wavelengths are much longer, breakup modes are less important, and the mike perturbs the boundary air less compared to mid and treble.

 

[tuga]

I am not comfortable with using nearfield for distortion; there are too many potential error sources.

@I fully agree with you on that. Btw, wouldn't you think that stepped sine with RTA and appropriate windowing can be even more accurate than sweeps as each tone lasts longer so better chances for the in-room steady state to fully stabilise?

Does it matter if the in-room steady state stabilises if you are measuring a speaker? Wouldn't that involve "masking"?

 

[QMuse]

I think that you don't want the room to be involved. So using Farina allows it to be removed by gating at midrange and higher. The tradeoff is bass, but the potential errors for nearfield at bass frequencies are less since the wavelengths are much longer, breakup modes are less important, and the mike perturbs the boundary air less compared to mid and treble.

I was thinking of measuring the <300Hz range with stepped sine and that maybe precise EQ at he mic position would get rid of the room non-linearities. For the >300Hz range you are certainly right.

 

[thewas]

Here you can also nicely hear the simulated differences due to intermodulation of a bassmid driver:
https://purifi-audio.com/tech/
Click on the article of Dec 12th, 2019 with the title "Low frequency harmonic distortion is almost inaudible. So what’s the point of low distortion drivers?"

 

[pkane]

In your experience how long does ferro fluid generally last before it degrades?

Can't answer that as a general question, but I had both tweeters killed on 20+ year-old Thiel 3.6s. In performing the autopsy, the ferro fluid looked and felt completely normal. I couldn't test it any more than that, but the fluid seemed to be of very low viscosity, and didn't seem to have deteriorated in any way (no gunk, no discoloration, very even appearance, etc). Prior to their untimely demise, the tweeters performed as well as I could ever remember them, so no obvious deterioration subjectively. These were the metal dome tweeters.

 

[andreasmaaan]

Once a driver is pushed to a point of excessive excursion it will not be able to play anything without distorion and asking small woofer to produce low bass even at moderate levels will lead to excessive driver excursion. CEA-2034 spec distorion measurement doesn't stress speakers enough to cause that so it is not very usefull.

Exactly. And not only that, by measuring only THD, it doesn't stress the driver in a way that gives meaningful results (since we'll never see the effects of higher excursion on midrange IMD).

 

[pma]

I am not comfortable with using nearfield for distortion; there are too many potential error sources. Extracting distortion from log-sin (Farina) sweeps strikes me as a much better method,

Used only for the frequencies below 0.4, see graph below. That's why I measured 100Hz HD and 200+300Hz IMD. Much less errors than from room reflections messing up the result.


Keele, D.B.: Low-Frequency Loudspeaker Assessment by Nearfield Sound-Pressure Measurement

 

[SIY]

Room reflections are easy to gate out. But doing nearfield close to a cone or dome at mid and treble may give misleading results because the driver is not necessarily pistonic, and near the surround, decidedly not, so depending on exactly where the mike is, you can get a misleadingly high or low result. Backing off a meter or two integrates the driver output.

 

[QMuse]

Room reflections are easy to gate out. But doing nearfield close to a cone or dome at mid and treble may give misleading results because the driver is not necessarily pistonic, and near the surround, decidedly not, so depending on exactly where the mike is, you can get a misleadingly high or low result. Backing off a meter or two integrates the driver output.

But aren't you loosing LF resolution with gating? How exactly do you measure <100Hz distorion?

 

[SIY]

But aren't you loosing LF resolution with gating? How exactly do you measure <100Hz distorion?

As I said before, the wavelengths there are quite a bit larger compared to cone, cap, and surround dimensions, and most cones still operate relatively pistonically at low frequencies. The mike itself is a more minor perturbation of the acoustic environment near the cone at LF. Yes, it's possible to miss something for a really floppy cone, but it's a lower probability. In reality, an anechoic environment is best, but nearfield is pretty good for bass.

 

[QMuse]

As I said before, the wavelengths there are quite a bit larger compared to cone, cap, and surround dimensions, and most cones still operate relatively pistonically at low frequencies. The mike itself is a more minor perturbation of the acoustic environment near the cone at LF. Yes, it's possible to miss something for a really floppy cone, but it's a lower probability. In reality, an anechoic environment is best, but nearfield is pretty good for bass.

Well, they do operate pistonic untill you push them, and then they don't. When measuring distortion my guess would be that the most interesting thing to find out would be the SPL limit (and slightly above it) where non-pistonic behavior is starting to happen and what is the distortion spectrum when it starts to happen. My guess would be that, as SPL starts to raise, woofers would be the first to exhibit non-pistonic behaviour, especially at lower end of the LF.

So you measure LF<300Hz distortion with nearfield measurement without gating? Are you measuring drivers and ports separately and then sum them like it is done when measuring FR?

Btw, if you're measuring LF distortion nearfield without gating why can't you use steeped sine RTA instead of sweeps?

 

[hardisj]

Just thinking out loud here...

I don't think there is one measurement that tells you everything.

We generally accept that a nearfield measurement will not net you accuracy when the wavelength produced is << effective diameter cone. There's plenty of proof available where people take a measurement at varying distances and you can see the HF perturbations become more pronounced the further the measurement distance is. Thus, the definition of nearfield measurements is defined by the drive-unit size. But does HF non-linear distortion (the typical % HD/IMD/etc) really even matter anyway? The linear distortion (frequency response) would be the dominant source of distortion information for higher frequencies, which is already done in the far-field. I view linear distortion as more useful than non-linear distortion for HF content. Especially given that non-linear distortion audibility is still not an exact science but we can draw direct correlations between an FR graph and what we hear. A farfield measurement will show HF break-up and the polars will help make it even more clear. Whereas, a NF measurement will show non-linear distortion caused by motor/suspension (near Fs) and low-to-mid IMD.

IMO, you need to use both linear distortion (frequency response) and non-linear distortion (the typical % HD/IMD/etc) to get any useful information here. Personally, once I have defined the pistonic range (which is approximately the same as the beaming point, for all intents and purposes) then I use nearfield data for low-to-mid non-linear distortion and farfield data for HF linear distortion. A 10" midwoofer will be non-directional (not beaming) up until about 675Hz*; below this nearfield for distortion and above this farfield FR for distortion. A 1" dome tweeter will not beam until about 6750hz; below this NF for distortion and above this I use farfield FR for distortion.

I'm not saying it's the only way. But I just weight importance here. Again, I can directly tie deviations in FR to audibility; I can't do that with % distortions until it's just "too much" and there's ambiguity there. So I use both sets of data to help me make a conclusion.

Just my $0.02.


*Speed of Sound ~ 13500 in/s in most conditions. Using half-wave for beaming approximation. 13500/2/DriverDiameter. Again, this is an approximation. Don't bark at me if you use full-wave or 20*F and 80% Humidity for your speed of sound calculations or I didn't account for the surround in my calculations.

 

[KSTR]

Low frequency distortion in bass drivers is mostly motor+suspension, not the cone itself. Therefore I always liked to use nearfield miking, some 5...10cm away from cone, pointing at a point along the dustcap-cone seamline. Mostly with Farina Log-Sweep where you can apply FDW on the harmonic IR's as well to reduce remaining room effect, but stepped sine works also. At any rate, things rattling around in the room is a no-go.
Ported speakers are harder to measure (at levels where overall distortion is still within low limits), notably when the port is not located close to the woofer. Whenever possible, I used outdoor ground-plane measurement, or in a very big storage hall. Trying to combine near-field distortions from individual woofer/port measurements gave me headaches, didn't work out reliably.

 

[andreasmaaan]

To be pragmatic about it, generally if I want to get a quick and cheap idea of a driver's distortion performance, I'll measure it at a distance of around 20-30cm, well away from any reflective surfaces.

Near-field enough for the ratio of direct to reflected sound to be high (assuming 2m to the nearest reflective surface and a 3dB reduction in SPL per doubling of distance, this gives a ratio of around 8:1 for direct sound to first reflections, and around 4:1 for direct sound to "reverberant" field), and far-field enough to take into account most of the non-pistonic behaviour of the driver.

But in general, for the best results in-room, you need to take both nearfield and gated farfield measurements.

Groundplane (with the loudspeaker elevated and pointing at the microphone) is better than nearfield in all respects, since you are getting the full non-pistonic components of the drivers behaviour, and the effects of the baffle on the relative SPLs of fundamental(s) and harmonics/IM products. Also because it tends to be accurate up to a higher frequency (determined by the distance from the microphone to the ground as opposed to the longest dimention of the driver/baffle.

I would generally prefer groundplane for ported boxes in particular.

 

[QMuse]

But in general, for the best results in-room, you need to take both nearfield and gated farfield measurements.

But, as @KSTR said, neither with help to measure distortion of LFs of ported floorstander where port is not close to the woofers. I choosed to do it with mic at LP (where precise EQ has been done) with stepped sine RTA as I didn't see any better way.

Groundplane (with the loudspeaker elevated and pointing at the microphone) is better than nearfield in all respects, since you are getting the full non-pistonic components of the drivers behaviour, and the effects of the baffle on the relative SPLs of fundamental(s) and harmonics/IM products. Also because it tends to be accurate up to a higher frequency (determined by the distance from the microphone to the ground as opposed to the longest dimention of the driver/baffle.

I would generally prefer groundplane for ported boxes in particular.

But wouldn't you have reflections from ground and from other surfaces as well with frequencies<3000Hz? And those wouldn't be corrected for linear distortions at the point of your measurement..

 

[andreasmaaan]

But, as @KSTR said, neither with help to measure distortion of LF of ported floorstander where port is not close to the woofers. I choosed to do it with mic at LP (where precise EQ has been done) with stepped sine RTA as I didn't see any better way.

Groundplane is far more reliable for this type of measurement. Should be very accurate up into the 1000s of Hz, and it's accuracy range can clearly be inferred from the frequency response if you know the quasianechoic frequency response of the speaker you're measuring already (there will be a null at a certain frequency corresponding to the distance from the mic to the groundplane).

 

[QMuse]

Groundplane is far more reliable for this type of measurement. Should be very accurate up into the 1000s of Hz, and it's accuracy range can clearly be inferred from the frequency response if you know the quasianechoic frequency response of the speaker you're measuring already (there will be a null at a certain frequency corresponding to the distance from the mic to the groundplane).

I thought we were discussing measuring distortion for freqs<300. And even with groundplane measurement I don't see why stepped sine would be inferior to sweep?

Btw, do you put floorstander sideways on the ground? How do you handle the scenario where port is 1m away from the woofers?

 

[andreasmaaan]

I thought we were discussing measuring distortion for freqs<300. And even with groundplane measurement I don't see why stepped sine would be inferior to sweep?

Yes, groundplane is very accurate in the low frequencies. It's in the high frequencies that accuracy falls away as a result of cancellation between the groundplane reflection and the direct sound hitting the microphone. But for a typical measurement mic this won't occur until well into the 1000s of Hz, as I mentioned

And I never said that stepped sine was inferior to log sweeps for groundplane measurements. I think that was SIY discussing that, in the context of in-room measurements (not groundplane), where stepped sine measurements fail to remove the contribution of the room.