Speakers and IM distortion

[Mtbf]

It felt inappropriate to post this in Amir’s “Should we (I) get into speaker testing & measurement” thread, so I started a new thread here.

One of the sites I follow regularly on the internet is Lowbeats, see:

https://www.lowbeats.de

This is a German HiFi-Online-Magazin, as they call themselves. The quality of the posts is imo certainly above average, and they MEASURE. Not that extensively, but still infinitely (anything ÷ 0 = ∞) more than the majority of all those other audio-related sites and of many audio-related journals out there.

For whatever reason, the company Dynaudio has stayed under my radar until the past one or two years. Despite the fact that I was familiar with the name Dynaudio, that this company exists for over 40 years (!), and that Dynaudio is one of the major loudspeaker companies of the world. As I’ve encountered in the mean time, they make some seriously good stuff. Besides the enclosures they also develop and produce all drivers in-house, and from the beginning they consistently use the same cone materials for their drivers, namely MSP (Magnesium Silicon Polypropylene, a composite) for their woofers and midtoners, and silk soft domes (doped with a special coating) for their tweeters.

This is a company that does some serious R&D. They opened a new research centre in 2017, containing a 13m x 13m x 13m large loudspeaker measurement room (as was already mentioned in Amir’s speakermeasurement yes/no thread), being the largest loudspeaker measurement room in Europe. One of the first speaker line-ups that profited from this new R&D centre is the new Dynaudio Evoke series. I happened to be at an audio show late 2018, and was very pleasantly surprised by the sound quality of the smallest speaker of this line-up, the Evoke 10 (I didn’t hear the rest of the Evoke family).

Now, on the 5th of May 2019 Lowbeats reviewed the Dynaudio Evoke 50, the largest floorstander of the Evoke family, see:

https://www.lowbeats.de/test-dynaudio-evoke-50-viel-bass-viel-spass/

What I found interesting were the IM measurements of this speaker. In the review they provide the IM-spectra of the speaker at 94dB spl and 103dB spl, both at one meter.

The IM-spectrum at 94dB spl is really impressive: allmost none (with the scale used):

As they conclude: “Kein Wunder, dass die noble Standbox so fein und klar klingt.” I thought exactly the same: no wonder I liked the Evoke 10 so much at the audio show. BTW, in most if not all of the Lowbeats speaker reviews these IM-spectra are provided.

A sound pressure level of 94dB is already beyond my normal listening levels (≤80dB spl), but even at an spl of 103dB the IM-spectrum looks pretty good:

Now here comes the surprise, at least for me it did. Last week there was a rave review on Lowbeats (Editor’s choice 2019; the year isn’t even over yet..) of the brand new TAD Evolution One TX, price per pair 25.000 Euro (ahum), see:

https://www.lowbeats.de/test-standlautsprecher-tad-evolution-one-tx-editors-choice-2019/

Andrew Jones left the company, but TAD of course is still able to produce a top loudspeaker, with very good directivity and well behaved impedance and phase curves (not shown here).

But have a look at the IM-spectra of the TAD at 94dB and 103dB spl:

The TAD spectra don’t come close to the spectra of the Dynaudio Evoke 50, and the TAD’s are almost 6 times the price of the Dynaudios.

What to think of these differences?

 

[DonH56]

Price is a weak to nonexistent function of performance.

Having IMD fill in the noise floor may make the speakers sound "richer" to some people. Also note the TAD's are worse at LF where we are less sensitive to distortion.

 

[MC_RME]

Interesting. Is that the 30 sine multi-tone signal used as reference? Are the peaks supposed to be on the same level? Both speakers show nearly the same peak levels, so it seems that the signal itself is modified to be not linear.

Regarding IM: that is an often overlooked topic, but in times of LFE subwoofers not as critical anymore as it had been years ago. It is still 100% relevant for headphones, and I never understood why nobody (!) performs useful IM measurements on them. The differences are drastic and audible, especially with today's high level low frequencies that tend to modulate mids (frequency and amplitude) on single driver systems. Planars are mostly immune, dynamic drivers differ heavily.

Here is a useful paper on speaker IM, only in German, sorry:

https://www.irt.de/IRT/FuE/ak/pdf/Goossens Nichtlineare Verzerrungen TMT2006.pdf

And here an even earlier paper (1992!) showcasing all kinds of distortion measurements, from Brüel&Kjaer, including IM with notes on speakers and microphones:

https://www.bksv.com/media/doc/BO0385.pdf

 

[RayDunzl]

Interesting. Is that the 30 sine multi-tone signal used as reference? Are the peaks supposed to be on the same level? Both speakers show nearly the same peak levels, so it seems that the signal itself is modified to be not linear.

The room will cause that, even if the speakers don't, with a flat signal from the amplifier.

 

[Krunok]

The room will cause that, even if the speakers don't, with a flat signal from the amplifier.

Sure, but average peak level on 94dB graph is below 80dB and none of the peaks actually reaches 94dB. The same goes for 103dB graph - average peak level is below 90dB and none of the peaks reaches 103dB.

I would be very interested to hear @Floyd Toole 's comment on how listeners reacted to various levels of THD/IMD in their blid tests as this is not covered by spinorama graphs. Are there THD/IMD measurements for Revel speakers available from Harman?

 

[RayDunzl]

Sure, but average peak level on 94dB graph is below 80dB and none of the peaks actually reaches 94dB. The same goes for 103dB graph - average peak level is below 90dB and none of the peaks reaches 103dB.

The individual lower intensity tones add to give the combined SPL.

For the thirty tones at 80dB average:

Ah... That worked out nicely.

 

[Ilkless]

What perplexes me is how Dynaudio clearly has the R&D capability to build excellent drivers, and the manufacturing chops to build nice speakers in Denmark at a relatively affordable price, yet insists on using first-order electrical crossovers against all empirical evidence about the importance of polar response.

 

[Krunok]

The individual lower intensity tones add to give the combined SPL.

For the thirty tones at 80dB average:

View attachment 29447

Ah... That worked out nicely.

Sure Ray, but what I wanted to say is that graph looks like 75dB average, not 80. But ok, the point is that kudos to Germans for making such measurements - I can't even remember when was the last time I saw one. And Dynaudio results certainly look really well.

 

[Krunok]

Interesting. Is that the 30 sine multi-tone signal used as reference? Are the peaks supposed to be on the same level? Both speakers show nearly the same peak levels, so it seems that the signal itself is modified to be not linear.

I believe signal was linear but peaks are not of the same height because of speaker's non linear FR.

 

[Floyd Toole]

Sure, but average peak level on 94dB graph is below 80dB and none of the peaks actually reaches 94dB. The same goes for 103dB graph - average peak level is below 90dB and none of the peaks reaches 103dB.

I would be very interested to hear @Floyd Toole 's comment on how listeners reacted to various levels of THD/IMD in their blid tests as this is not covered by spinorama graphs. Are there THD/IMD measurements for Revel speakers available from Harman?

Non-linear distortion measurements are routine in the design of transducers and systems, but they are mainly useful in a relative not absolute sense, meaning that if a measurement gets smaller, that is good, but whatever the measurement is does not reflect how it might sound in music, in a room. It is very rare for distortion to be a factor in listening test results, but it has happened, most recently IMD in a concentric mid/tweet driver. Here is something I wrote on the topic a while back.

Non-linear distortions originate in a non-linear input/output relationship which changes the waveform by adding spectral components to the audio signal. Rather than quantifying the non-linearity itself, we probe the system with simple signals; single or multiple pure tones being the most common. Then we measure what comes out and, knowing what went in, a percentage distortion can be calculated. Problem is that because of simultaneous masking all of the distortion components are not heard. So, the measured percentage distortion, which includes all components, is wrong. The correlation with objectionable sounds when listening to music is poor. We need evaluations that incorporate some properties of the human hearing system's inherent masking in order to make a serious start.

The existing metrics, like harmonic or intermodulation distortion are useful to design engineers, but the only truly meaningful number is zero.

In order to meaningfully interpret distortion measurements above zero, the metrics must involve elements of psychoacoustics: masking being the dominant one. A few have been tried but I know of none that have been tested sufficiently to achieve widespread acceptance. But it is a topic that I have not closely followed in recent years - I'm now retired. There is a discussion, with references in my book. The good news is that over my nearly 50 years of evaluating loudspeakers it has been extremely rare for non-linear distortion to be an audible factor in sound quality evaluations of conventional "hifi" products. Recently popular small wireless and "smart" loudspeakers face real challenges in that respect.

Measuring non-linear distortion in a room makes a difficult situation even worse, because amplitudes of distortion products are affected by reflected sounds which are highly dependent on the venue, the loudspeaker location and the mic location. Background noise is another challenge; such measurements need to be done in a very quiet anechoic chamber or equivalent. Being able to make a measurement does not make it a useful measurement.

My standard example of human tolerance for non-linear distortion is the LP. In the early years LPs were the source of program material for loudspeaker listening tests, so I made it my business to examine the performance of the medium. I published several articles in Canadian audio magazines on the topic, and it got to the point of making a test recording that contained music and test signals. There were easily audible differences between the master tape and what came off the LP - those inherent problems are still with us. The levels of measured non-linear distortions of all conceivable kinds were shocking - whole percentages being common, and, during incipient or real mistracking, off the charts. Masking is a powerful factor in what we hear. Needless to say, digital audio was a great improvement. But for some, obviously, the placebo effect is profound.

 

[Krunok]

Non-linear distortion measurements are routine in the design of transducers and systems, but they are mainly useful in a relative not absolute sense, meaning that if a measurement gets smaller, that is good, but whatever the measurement is does not reflect how it might sound in music, in a room. It is very rare for distortion to be a factor in listening test results, but it has happened, most recently IMD in a concentric mid/tweet driver. Here is something I wrote on the topic a while back.

Non-linear distortions originate in a non-linear input/output relationship which changes the waveform by adding spectral components to the audio signal. Rather than quantifying the non-linearity itself, we probe the system with simple signals; single or multiple pure tones being the most common. Then we measure what comes out and, knowing what went in, a percentage distortion can be calculated. Problem is that because of simultaneous masking all of the distortion components are not heard. So, the measured percentage distortion, which includes all components, is wrong. The correlation with objectionable sounds when listening to music is poor. We need evaluations that incorporate some properties of the human hearing system's inherent masking in order to make a serious start.

The existing metrics, like harmonic or intermodulation distortion are useful to design engineers, but the only truly meaningful number is zero.

In order to meaningfully interpret distortion measurements above zero, the metrics must involve elements of psychoacoustics: masking being the dominant one. A few have been tried but I know of none that have been tested sufficiently to achieve widespread acceptance. But it is a topic that I have not closely followed in recent years - I'm now retired. There is a discussion, with references in my book. The good news is that over my nearly 50 years of evaluating loudspeakers it has been extremely rare for non-linear distortion to be an audible factor in sound quality evaluations of conventional "hifi" products. Recently popular small wireless and "smart" loudspeakers face real challenges in that respect.

Measuring non-linear distortion in a room makes a difficult situation even worse, because amplitudes of distortion products are affected by reflected sounds which are highly dependent on the venue, the loudspeaker location and the mic location. Background noise is another challenge; such measurements need to be done in a very quiet anechoic chamber or equivalent. Being able to make a measurement does not make it a useful measurement.

My standard example of human tolerance for non-linear distortion is the LP. In the early years LPs were the source of program material for loudspeaker listening tests, so I made it my business to examine the performance of the medium. I published several articles in Canadian audio magazines on the topic, and it got to the point of making a test recording that contained music and test signals. There were easily audible differences between the master tape and what came off the LP - those inherent problems are still with us. The levels of measured non-linear distortions of all conceivable kinds was shocking - whole percentages being common, and, during incipient or real mistracking, off the charts. Masking is a powerful factor in what we hear. Needless to say, digital audio was a great improvement. But for some, obviously, the placebo effect is profound.

Thank you for taking time to answer!

The graph below is a THD+N measurement of Performa3 F206 speaker taken by SoundStage! magazine.

As it can be seen it was taken at 90dB SPL and it shows that north of 200Hz THD+N is lower than 45dB, so app <0.5%. I normally don't listen music at higher SPLs than 90dB and ambient noise in my room is typically 25-30dB, depending on the time of the day, so probably there would be significant portion of distortion componenets that would not oly be masked by music but also burried in the noise floor. Taking that into account would you say that distortion shown here for this aprticular speaker is sufficiently good and that making it better than this can actually go unnoticed by listeners in blind test?

 

[Floyd Toole]

Thank you for taking time to answer!

The graph below is a THD+N measurement of Performa3 F206 speaker taken by SoundStage! magazine.

View attachment 29482

As it can be seen it was taken at 90dB SPL and it shows that north of 200Hz THD+N is lower than 45dB, so app <0.5%. I normally don't listen music at higher SPLs than 90dB and ambient noise in my room is typically 25-30dB, depending on the time of the day, so probably there would be significant portion of distortion componenets that would not oly be masked by music but also burried in the noise floor. Taking that into account would you say that distortion shown here for this aprticular speaker is sufficiently good and that making it better than this can actually go unnoticed by listeners in blind test?

Distortion at very low frequencies is especially hard to translate into audibility. The measurements shown were made in a 4-pi anechoic chamber. The chamber has been calibrated at the low frequencies of interest, and not knowing the frequency content of the THD it may or may not be accurate in an absolute sense.

You will be listening in a reflective room, with nearby acoustically supporting floor and walls - i.e. the sound level will go up at low frequencies and the percentage distortion will drop. I think you are quite safe with these speakers - I have a pair in my office and they are very rewarding. If you ever decide to bass manage and add sub(s) this distortion issue is transferred to the subs and the F206 can play louder.

 

[Krunok]

If you ever decide to bass manage and add sub(s) this distortion issue is transferred to the subs and the F206 can play louder.

Thank you once again!

In that scenario, what would you recommend as XO point with the subs assuming they are linear in the 20Hz-180Hz range?

 

[Floyd Toole]

Thank you once again!

In that scenario, what would you recommend as XO point with the subs assuming they are linear in the 20Hz-180Hz range?

The default crossover frequency in most (all?) bass management schemes is 80 Hz. It is well chosen for a reason: it opens up more possibilities for subwoofer placement. As Chapter 8 in the 3rd edition of my book elaborates, multiple subwoofers offer huge advantages in attenuating room resonances, enabling multiple listeners to enjoy similarly good bass, and such systems offer substantial efficiency gains. Win, win, win.

 

[Krunok]

The default crossover frequency in most (all?) bass management schemes is 80 Hz. It is well chosen for a reason: it opens up more possibilities for subwoofer placement. As Chapter 8 in the 3rd edition of my book elaborates, multiple subwoofers offer huge advantages in attenuating room resonances, enabling multiple listeners to enjoy similarly good bass, and such systems offer substantial efficiency gains. Win, win, win.

Thank you! I managed to cover first 100 pages so far, almost to the end of chapter 7, but holidays start in 10 days so I'll have more time for reading.

 

[amirm]

One issue with speaker THD measurements is the distortions of the measurement microphone itself. There is a catch-22 in that what we see is the sum of the distortion of the microphone+speaker. I have seen little attempt to separate these factors.

I have been researching distortion metrics of measurement mics and I have yet to find any useful data. By default, dynamic range of a measurement microphone is specified between 1 and 3%!

In some limited testing I have done with my budget headphone measurement mic, its distortions are dominant relative to the headphone.

I have done tons of critical listening tests using headphones (and some with speakers) and I am able to detect very small impairments with them. This runs counter to high reported distortions of headphones/speakers. What is there clearly is not masking secondary distortion elsewhere.

 

[DonH56]

Have you tried measuring with your Earthworks?

 

[Krunok]

I have been researching distortion metrics of measurement mics and I have yet to find any useful data. By default, dynamic range of a measurement microphone is specified between 1 and 3%!

What exactly do you mean when you say "dynamic range"? Are you saying that THD is in the 1-3% range?

In some limited testing I have done with my budget headphone measurement mic, its distortions are dominant relative to the headphone.

Is there a practice that manufacturers specify distortion of the measurement microphones, at least for top models?

 

[Blumlein 88]

One issue with speaker THD measurements is the distortions of the measurement microphone itself. There is a catch-22 in that what we see is the sum of the distortion of the microphone+speaker. I have seen little attempt to separate these factors.

I have been researching distortion metrics of measurement mics and I have yet to find any useful data. By default, dynamic range of a measurement microphone is specified between 1 and 3%!

In some limited testing I have done with my budget headphone measurement mic, its distortions are dominant relative to the headphone.

I have done tons of critical listening tests using headphones (and some with speakers) and I am able to detect very small impairments with them. This runs counter to high reported distortions of headphones/speakers. What is there clearly is not masking secondary distortion elsewhere.

Okay simple example. You are listening over headphones with 2% 3rd harmonic distortion. You detect impairments of microphones. Wouldn't you detect a microphone distortion of .1 % 3rd harmonic distortion as you would hear 2.1% vs 2.0%? Wouldn't this be a case where high headphone distortion wouldn't mask?

 

[Blumlein 88]

What exactly do you mean when you say "dynamic range"? Are you saying that THD is in the 1-3% range?



Is there a practice that manufacturers specify distortion of the measurement microphones, at least for top models?

Well here is on example:

Total Harmonic Distortion (THD), 2500 ohms load <0.08% for a 120 dB equivalent SPL input.

Does that mean less than .08% for 120 db SPL, or does that mean the circuitry only distorts at that level for a signal one would get at 120 db SPL (the equivalent part), but not necessarily that the output of a real signal is this clean?