loudspeakers L/R matching are sometimes terrible !

[thewas]

Hifinews are measuring the matching frequency responce accuracy of the left and right loudspeakers.

The results with passive loudspeakers differ on average more than +-3 dB in the midrange between left and right loudspeaker.

I wonder where do you see such 6 dB fluctuations, I see less than a +-1 dB differences in the midrange on the links you posted above (always the black and red curves):

Only the 801D4 seems to have a narrow bigger deviation than that at 2,5 kHz which probably is also inaudible with most music:

And there are also many cheap passive loudspeakers with similarly good pair matching even from the same company, just exemplaric there:

B&W 603 S2 Anniversary Edition Loudspeaker Lab Report

Lab Report While the Anniversary ticket brings with it a stunning standard of finish, it's the 603's 'S2' acoustical revisions that are of greatest interest here. In practice, the forward response remains uneven [see Graph 1, below] although the 'shape' with its trough from 1.5kHz-3.5kHz and...

www.hifinews.com

KEF LS50 Meta Loudspeaker Lab Report

Lab Report To ensure the validity of our comparisons, new samples of both the LS50 and LS50 Meta were tested 'back-to-back' under identical lab conditions. So, while the LS50's pair matching is within 1.3dB and response uniformity good to ±2.9dB and ±4.0dB [grey/orange traces, Graph 1] the new...

www.hifinews.com

Monitor Audio Bronze 200 Loudspeaker Lab Report

Lab Report Competing in the 'slimline' floorstander bracket, MA's Bronze 200 employs the brand's smallest C-CAM mid/bass driver, a pair of these 140mm units enabling a narrow 166mm-wide baffle. (The 231mm-wide Bronze 100 standmount [HFN Aug '20] employs a single 200mm C-CAM unit.) The...

www.hifinews.com

By the way active loudspeakers from usually the lower cost points have similar deviations (see the measurements of Sound & Recording which give its value - unfortunately not a plot) so its not really a issue of active vs. passive but rather of manufacturing quality (especially of the drivers), allowed tolerances and measurements/selection.

 

[thewas]

JBL 4309 Loudspeaker Lab Report

Lab Report With that mid/treble horn front-and-centre there's hope of good sensitivity from this little box, and JBL's 4309 does not disappoint, achieving 87.8dB/1m/1kHz and 86.1dB from 500-8kHz, according well with the rated 87dB. Neither is there a sting in the 4309's tail as regards its...

www.hifinews.com

In this case the difference between L/R speakers are +- 5 dB in the midrange and +-4,8 dB in the treble. This is hugh differences. Clearly audible. Apparently there is the risk of bigger differences with horn loudspeakers.

??? I again see less than a dB deviation in the midrange!

 

[thewas]

This is smoothed curves . You have to read the spec tabell , scroll down.
View attachment 218502

Nope, you just are just reading them wrong(!), for example on this review of the JBL 4309 where you say its pair matching is poor:

Pair matching/Resp. error (200Hz–20kHz) 2.4dB/ ±5.0dB/±4.8dB

Only the first number is the pair matching and the other ones are the response errors!

This can be also easily be seen if you had spent few seconds to read the text on the same page:

Pair matching is a superb 0.6dB (200Hz-8kHz) but a wobble in the response of one speaker from 8-13kHz results in a poorer 2.4dB from 200Hz-20kHz.

which makes the title and thesis of the whole thread not applicable.

 

[Tangband]

I wonder where do you see such 6 dB fluctuations, I see less than a +-1 dB differences in the midrange on the links you posted above (always the black and red curves):

Only the 801D4 seems to have a narrow bigger deviation than that at 2,5 kHz which probably is also inaudible with most music:

And there are also many cheap passive loudspeakers with similarly good pair matching even from the same company, just exemplaric there:

B&W 603 S2 Anniversary Edition Loudspeaker Lab Report

Lab Report While the Anniversary ticket brings with it a stunning standard of finish, it's the 603's 'S2' acoustical revisions that are of greatest interest here. In practice, the forward response remains uneven [see Graph 1, below] although the 'shape' with its trough from 1.5kHz-3.5kHz and...

www.hifinews.com

KEF LS50 Meta Loudspeaker Lab Report

Lab Report To ensure the validity of our comparisons, new samples of both the LS50 and LS50 Meta were tested 'back-to-back' under identical lab conditions. So, while the LS50's pair matching is within 1.3dB and response uniformity good to ±2.9dB and ±4.0dB [grey/orange traces, Graph 1] the new...

www.hifinews.com

Monitor Audio Bronze 200 Loudspeaker Lab Report

Lab Report Competing in the 'slimline' floorstander bracket, MA's Bronze 200 employs the brand's smallest C-CAM mid/bass driver, a pair of these 140mm units enabling a narrow 166mm-wide baffle. (The 231mm-wide Bronze 100 standmount [HFN Aug '20] employs a single 200mm C-CAM unit.) The...

www.hifinews.com

By the way active loudspeakers from usually the lower cost points have similar deviations (see the measurements of Sound & Recording which give its value - unfortunately not a plot) so its not really a issue of active vs. passive but rather of manufacturing quality (especially of the drivers), allowed tolerances and measurements/selection.

My comments and faults are now corrected and other examples are made after your suggestions . There were other, much worse faults on other loudspeakers .

You are right in the way you say that this is also a problem with active loudspeakers if not done right .

But - a dsp crossover has almost 0 % deviation between different units and a massproduced passive crossover sometimes has +- 10 % in its coils .

The drivers are as they are , in both active and passive loudspeakers, so there is a equal selection problem here .

 

[thewas]

But - a dsp crossover har 0 % deviation between different units

I wouldn't say it 0% because of the electronics but very low.

and a massproduced passive crossover often has +- 10 % in its coils .

Could you please show some examples? Personally I doubt it is larger than +- 5% which might sound large but its effect in the end is usually lower than most would expect, just as an example below I just changed the main/series inductor of the woofer of a typical 2-way loudspeaker from 4,7 mH to 5,2 mH (so + 10%!) and the deviation is less than a dB (dotted line):

The parallel inductors on the mid and tweeters have even lower sensitivity/deviation on value tolerances.

Also it would be expedient to edit your initial post and title as it is based on an misinterpretation of you.
(Which can be also seen that you thought that only the Blade One Meta has a low pair deviation while what makes its stand out is the lower deviation from the ideal neutral response).

 

[Tangband]

I wouldn't say it 0% because of the electronics but very low.

Could you please show some examples? Personally I doubt it is larger than +- 5% which might sound large but its effect in the end is usually lower than most would expect, just as an example below I just changed the main/series inductor of the woofer of a typical 2-way loudspeaker from 4,7 mH to 5,2 mH (so + 10%!) and the deviation is less than a dB (dotted line):

View attachment 218505

The parallel inductors on the mid and tweeters have even lower sensitivity/deviation on value tolerances.

Also it would be expedient to edit your initial post and title as it is based on an misinterpretation of you.
(Which can be also seen that you thought that only the Blade One Meta has a low pair deviation while what makes its stand out is the lower deviation from the ideal neutral response).

Some corrections are now done !

10% deviation in coils, in a 24 dB/octave passive crossover can result in 20% errors because there are two of them .

 

[thewas]

My comments and faults are now corrected after your suggestions .

Not enough though, as you now write:

The results with loudspeakers often differ more than +-2 dB in the midrange between left and right loudspeaker.

while for the 2 "poor" examples you posted the deviations are 2.2dB, 2.4dB which is equivalent rather to +-1 dB and usually those deviations are not in the midrange where they are lower. Also the title/generalisation that this terrible is not really the case then.

 

[thewas]

10% deviation in coils, in a 24 dB/octave passive crossover can result in 20% errors because there are two of them .

It doesn't work like that as the secondary coil value doesn't have such a strong influence on the pass band like the first one.

 

[Mtbf]

I remember that in the past B&W speakers were matched based on their individual FR. On the other hand, this kind of variation from sample to sample should be taken into account in the differences in measurements between different speakers.

 

[sergeauckland]

Only the 801D4 seems to have a narrow bigger deviation than that at 2,5 kHz which probably is also inaudible with most music:

Depends what one means by 'inaudible. Pair-matching errors aren't so much in the frequency response, but in the image sharpness. A perfectly matched pair will give a very sharp central image on a mono two-channel signal. As the pair-matching errors increase, the width of the central image widens, and becomes diffuse, and will drift with frequency as the errors give different L-R amplitudes as frequency varies.

If one wants very sharp stereo imaging, as I do, then pair-matching is more important than actual frequency response, which mainly affects timbre/coloration.

S.

 

[Tangband]

Depends what one means by 'inaudible. Pair-matching errors aren't so much in the frequency response, but in the image sharpness. A perfectly matched pair will give a very sharp central image on a mono two-channel signal. As the pair-matching errors increase, the width of the central image widens, and becomes diffuse, and will drift with frequency as the errors give different L-R amplitudes as frequency varies.

If one wants very sharp stereo imaging, as I do, then pair-matching is more important than actual frequency response, which mainly affects timbre/coloration.

S.

One example I have edited in the first thread, is the JBL hdi 3800. The deviation between L and R speaker are here 4,6 dB !

JBL HDI-3800 Loudspeaker Lab Report

Lab Report The breakup mode of the Teonex compression driver, or its interaction with the horn throat, is clearly illustrated just outside of the audioband [pink shaded area, Graph 1] at 22kHz, rising some +15dB above the mean midband output of the HDI-3800. The CSD waterfall [Graph 2] suggests...

www.hifinews.com

And another :

Paradigm Founder 120H Loudspeaker Lab Report

Lab Report Paradigm's very generous 92dB sensitivity rating for the Founder 120H is on the money, these flagship speakers delivering a thundering 93dB (2.83V/1kHz/1m). However, sensitivity does fall to 91.1dB across 500Hz-8kHz thanks to a loss in axial output around the 2.4kHz crossover between...

www.hifinews.com

5,7 dB

 

[Tangband]

Thewas: after reading the reviews in the correct way, the deviances between certain ( other ) loudspeakers where even worse than I wrote in the beginning !
Thanks for correcting me though , I have edited the first part of the thread with other examples.

Two of the worst differences between L and R was from two very expensive loudspeakers. 5,7 dB in one case and I wonder what excuse they have to have such big tolerance between different units ?

 

[Tangband]

Depends what one means by 'inaudible. Pair-matching errors aren't so much in the frequency response, but in the image sharpness. A perfectly matched pair will give a very sharp central image on a mono two-channel signal. As the pair-matching errors increase, the width of the central image widens, and becomes diffuse, and will drift with frequency as the errors give different L-R amplitudes as frequency varies.

If one wants very sharp stereo imaging, as I do, then pair-matching is more important than actual frequency response, which mainly affects timbre/coloration.

S.

Totally agree . I would certainly not except those variations, not for this money.

 

[MAB]

Crossover components are often built from components with +- 10% tolerance... capacitors, inductors and resistors. Each tolerance can have an equivalent magnitude of impact depending on the application.
+-10% is the maximum tolerance (3-sigma of the component distribution), it a statistical value so keep that in mind.

10% deviation in coils, in a 24 dB/octave passive crossover can result in 20% errors because there are two of them .

Fortunately, statistics don't work like this. The probability of two 3-sigma events (two random -10% pieces ending up in a single unit) is less than ten in a million.

The bad component tolerance in many passive crossovers with +-10% differences in coils is showing up in measurements.

Why are you assuming it is the inductors (if it is even the passive components)? The capacitors in the crossover have the same tolerance band, with the same impact to overall response.
So, you should really model your hypothesis. It is so easy. And it will be illuminating and we can compare to the results from Hifinews and perhaps come to some conclusions.

So I modeled 10% component variations in a first order low-pass and first order high-pass crossover (single inductor or capacitor in series with the driver).

The mismatch grows to 1dB at the 6dB down point. So, a 10% mismatched component can have no larger than a 1dB response deviation, and that deviation would be below the cutoff frequency for that element. That gives information as to where the actual deviations from ideal

Putting this all together, here are 8 Ohm woofer and tweeter in a first-order alignment with (relatively) flat response at the crossover point:

Here is the same crossover with one element 10% mismatched (capacitor in this case but inductor mismatch would be the same magnitude deviation from ideal):

And, to add the very important comment that @thewas made:

The parallel inductors on the mid and tweeters have even lower sensitivity/deviation on value tolerances.

Here is a 10% component variation in a second order crossover (the control was flat, the deviation on the mismatched crossover components are about 0.25dB!!!):

I didn't model resistors, but a resistive element will have similar impact per percentage point of tolerance except that it will typically affect the driver response in the entire audible band (your mileage will vary depending on the crossover and driver specifics).

And, if you get that one in a million speaker that has two elements mismatched by 10%, the errors from ideal roughly twice the figures I calculated here and still tiny.

Lastly, look carefully at the graphs you referenced and ask if a mismatched passive element in a crossover could possibly explain the deviations you are observing. They just can't. I can speculate why the Hifinews measurements show left/right mismatches but it is not realistic to think that Hifinews got shipped a statistical sample of speakers that have passive component mismatches in the crossovers large enough to generate these measurements.

To be honest, I have read about speaker manufacturers hand-matching components, blah blah blah. I hear this matching talk from the zero-negative feedback crew as well. It is cheap talk that fuels paranoia and causes non-physical reasoning like this.

If you have further questions, you should model the actual response of a driver to various component values. You will quickly stop worrying about this phenomena.

 

[Tangband]

Thanks MAB for the simulations.

Are you suggesting that the 5,7 dB deviance in the Paradigm founder 120h loudspeaker might be a measurements fault at Hifinews ? It might be of course.

Is the +- 0,45 dB difference in Kef blade also a faulty measurement ? Or is it a sign of that they have matched the crossovers and drivers to small tolerances ? There is a youtube video where they show how Kef do this on the blade speaker.

From my perspective : a passive crossover with more than 1 dB difference between different speakers is not good enough. Not even close. Because you also have the driveunit deviances, which you cant correct easily with a passive crossover.

I have played around a lot with active dsp crossovers during my work with my own active DIY loudspeaker. I wouldnt tolerate more differences than 0,5 dB between left and right loudspeaker between 200-12000 Hz - that include crossover AND driver missmatch, all of this can be corrected with a dsp crossover with some peq and a lot of nearfield measurements.

From my own experience, 1 dB difference in the midrange is audible as a slightly worse stereo perception.

So why are there still people who says that much larger deviances between L/R speakers doesnt matter ?

As a semiproffessional recording amateur , doing recordings with two microphones in good acoustic rooms, I use two Line audio OM1 microphones with +-1 dB tolerance between 20 - 20000 Hz . Why would I tolerate bigger differences between my L/R listening speakers in the midrange, in my listening setup ?

 

[MAB]

Thanks MAB for the simulations.

Are you suggesting that the 5,7 dB deviance in the Paradigm loudspeaker might be a measurements fault at Hifinews ?

No, that would be jumping to conclusions but it is possible. I am suggesting that it cannot be explained with normal (+-10%) inductor or capacitor mismatch in the crossover, even extraordinary mismatch. It could be the drivers are mismatched, but that would also be speculating. And any time I see measurements showing mismatches in the crossover region, the first thing I think of is the measurement was the variation source. Without being there for the measurement, or having Hifinews explain it in detail, I cannot tell and would be jumping to conclusions, which is what I am trying to prevent! I will say that with REW, I can get dips and peaks in midrange-tweeter interaction similar to your plots. I will also say, room-speaker interactions have a bigger impact on L/R matching than any passive component has.

From my perspective : a passive crossover with more than 1 dB difference between different speakers is not good enough. Not even close. Because you also have the driveunit deviances, which you cant correct easily with a passive crossover.

I tried to show that 1dB response variation is not crossover component matching, even for a pair of speakers with maximum mismatched components. You are holding on tight to this hypothesis!

So why are there still people who says that much larger deviances doesnt matter ? Because it clearly does, in most cases.

No, I actually showed that even components with a full 10% mismatch in component values will have less than 0.5dB variation for a first-order crossover, and less than 0.25dB for a second-order crossover. I don't want to debate if you can hear 1dB variation in midrange, but 0.25dB to 0.5dB mismatch isn't audible and far below room interaction impact. And my simulations are for worst case matching, 99.7% of shipped pairs will have better matching than what I calculated.

 

[thewas]

One example I have edited in the first thread, is the JBL hdi 3800. The deviation between L and R speaker are here 4,6 dB !

On the plot I don't see such a deviation though so probably it is on the ultrasonic tweeter resonance

and in the text of the review it writes "while the pair matching remains tight at 1.3dB"

And another :

Paradigm Founder 120H Loudspeaker Lab Report

Lab Report Paradigm's very generous 92dB sensitivity rating for the Founder 120H is on the money, these flagship speakers delivering a thundering 93dB (2.83V/1kHz/1m). However, sensitivity does fall to 91.1dB across 500Hz-8kHz thanks to a loss in axial output around the 2.4kHz crossover between...

www.hifinews.com

5,7 dB

Thats really a big and wide one, which looks to me a like a defective unit which shouldn't have passed the control.

 

[thewas]

Thewas: after reading the reviews in the correct way, the deviances between certain ( other ) loudspeakers where even worse than I wrote in the beginning !

Not really, you had written initially +-3 dB and this poor one example which as I write above looks like a defective unit is not worse with its 5,7 dB (< +- 3dB).

Two of the worst differences between L and R was from two very expensive loudspeakers. 5,7 dB in one case and I wonder what excuse they have to have such big tolerance between different units ?

As said such shouldn't have passed QC, but it is not impossible that the damage on this one happened also during shipping.

 

[thewas]

Depends what one means by 'inaudible. Pair-matching errors aren't so much in the frequency response, but in the image sharpness. A perfectly matched pair will give a very sharp central image on a mono two-channel signal. As the pair-matching errors increase, the width of the central image widens, and becomes diffuse, and will drift with frequency as the errors give different L-R amplitudes as frequency varies.

If one wants very sharp stereo imaging, as I do, then pair-matching is more important than actual frequency response, which mainly affects timbre/coloration.

S.

While what you say is correct, I strongly doubt that such a narrow deviation as the shown one will really significantly worsen the imaging sharpness, also in my experience for sharp imaging of much higher importance are the secondary sound sources and also very symmetric room acoustics. When measuring the FR at the LP in normal rooms usually deviations of more than 1 dB appear in the mid/highs which are unfortunately much more detrimental.

 

[thewas]

Crossover components are often built from components with +- 10% tolerance... capacitors, inductors and resistors. Each tolerance can have an equivalent magnitude of impact depending on the application.
+-10% is the maximum tolerance (3-sigma of the component distribution), it a statistical value so keep that in mind.

Fortunately, statistics don't work like this. The probability of two 3-sigma events (two random -10% pieces ending up in a single unit) is less than ten in a million.


Why are you assuming it is the inductors (if it is even the passive components)? The capacitors in the crossover have the same tolerance band, with the same impact to overall response.
So, you should really model your hypothesis. It is so easy. And it will be illuminating and we can compare to the results from Hifinews and perhaps come to some conclusions.

So I modeled 10% component variations in a first order low-pass and first order high-pass crossover (single inductor or capacitor in series with the driver).
View attachment 218543

View attachment 218561

The mismatch grows to 1dB at the 6dB down point. So, a 10% mismatched component can have no larger than a 1dB response deviation, and that deviation would be below the cutoff frequency for that element. That gives information as to where the actual deviations from ideal

Putting this all together, here are 8 Ohm woofer and tweeter in a first-order alignment with (relatively) flat response at the crossover point:
View attachment 218583

Here is the same crossover with one element 10% mismatched (capacitor in this case but inductor mismatch would be the same magnitude deviation from ideal):
View attachment 218584

And, to add the very important comment that @thewas made:

Here is a 10% component variation in a second order crossover (the control was flat, the deviation on the mismatched crossover components are about 0.25dB!!!):
View attachment 218586

I didn't model resistors, but a resistive element will have similar impact per percentage point of tolerance except that it will typically affect the driver response in the entire audible band (your mileage will vary depending on the crossover and driver specifics).

And, if you get that one in a million speaker that has two elements mismatched by 10%, the errors from ideal roughly twice the figures I calculated here and still tiny.

Lastly, look carefully at the graphs you referenced and ask if a mismatched passive element in a crossover could possibly explain the deviations you are observing. They just can't. I can speculate why the Hifinews measurements show left/right mismatches but it is not realistic to think that Hifinews got shipped a statistical sample of speakers that have passive component mismatches in the crossovers large enough to generate these measurements.

To be honest, I have read about speaker manufacturers hand-matching components, blah blah blah. I hear this matching talk from the zero-negative feedback crew as well. It is cheap talk that fuels paranoia and causes non-physical reasoning like this.

If you have further questions, you should model the actual response of a driver to various component values. You will quickly stop worrying about this phenomena.

Such quality analysis posts are what I love in ASR, thank you!