Buchardt S400 Speaker Review

[amirm]

I think what was pointed out which seemed interesting is that for the polar plot, yours shows a shorter distance (<1m) and theirs shows 2m, and what is assumed is that the software is showing performance at those distances.

So, is that correct, or is that distance figure irrelevant and the plot is always showing far-field performance?

It is totally irrelevant. This is just a visualization tool, not computation of soundfield. It assumes that we are operating in far field so just changes the SPL. That is the definition of "far field." Phase becomes invariant and we just have a 1/r relationship with distance.

The variations seen at low frequencies is due to output of the speaker being very low so if you reduce the projected output, you get more error. Shading also changes due to SPL, not because the soundfield generation is different.

I picked the lower "distance" for this review as to make the SPL numbers closer to spin data. That's all.

 

[MZKM]

It is totally irrelevant. This is just a visualization tool, not computation of soundfield. It assumes that we are operating in far field so just changes the SPL. That is the definition of "far field." Phase becomes invariant and we just have a 1/r relationship with distance.

The variations seen at low frequencies is due to output of the speaker being very low so if you reduce the projected output, you get more error. Shading also changes due to SPL, not because the soundfield generation is different.

I picked the lower "distance" for this review as to make the SPL numbers closer to spin data. That's all.

Yeah, I did notice that if you accounted for what SPL segments get shaded which color, they look pretty similar.

 

[amirm]

Sorry, but No. This is not right. The NFS will not just assume same result at any distance, this is the essence of the system, that when you move to different distance, you get NEW results. We can show that easily by showing you our 2m vertical contour plot and same contour plot at 66cm distance below:

No, it is right Mads. It is physics of sound propagation. It is not subject to debate.

The module that generates the directivity plot is a *visualization* tool. The sound field data is already computed in a different module and remains unchanged.

When you change the "distance" all it does is change the computational gain. This changes the false color pallet making it look like things are different due to very low resolution of these graphs. All major acoustic events are identical in both measurements. They would not be so if were in near field.

Your speaker doesn't have much output in low frequencies so that area is subject to a lot more noise and inaccuracy. So naturally when you increase the "distance" the accuracy of that data shrinks and the graph changes a bit.

It is evident, that you see a difference between the two plots – especially in the LF region. The reason for this is because at nearfield, your relative distance from visualization point to the LF drivers is shorter. The circle below simple has the speaker physically shifted upwards to the left. Also – rear-driver contribution to SPL rearside will be unevenly high, due to relative distance to front-driver is shorter, due to nearfield evaluation.

No, this implies that the measurement point was changed but that is not what we are talking about. The distance comes from telling the visualization software to simply show SPL data at one or another far field position. The software is not changing the angle as you are assuming. That would only happen with a single measurement point at two different positions.

It would be nice if it worked this way but it does not because it is projecting the data into far field where this doesn't matter.

 

[aarons915]

Just curious about the distortion discussion that was taking place. It was said that the S400 has basically the same woofer as the Revel M106, does anyone know why the distortion seems to be very well controlled for that speaker in the Soundstage measurement and not as well in this speaker? Looks like the drive level here is a bit higher, does that explain it?

https://www.soundstagenetwork.com/i...&catid=77:loudspeaker-measurements&Itemid=153

 

[GelbeMusik]

I rarely deal with ...
Sources: Zwicker, Fastl - Psychoacoustics, researchgate.net, hifi-selbstbau.de

The "hobbyists" of hifi-selbstbau.de, ... have taken a program published in DIN 45631, which can ... and have calculated the masking curves at different frequencies and sound level.
...

Would also assume … pure estimations of Zwicker, because ...
… were not simply pulled out of a hat by "hobbyists".

The limits shown are certainly not valid for ….

I already jumped out of this topic, so the answer might read more sharp harshthan it should be taken. First of all, thank You for taking the effort to look up and explain parts of Your library.

These hobbyists are hobbyists. Psychoacoustics is neither their profession nor education. So, regarding the merits of their undertaking, they tried to validate it by a public (!) survey. Only that the results are kept secret as long You (!) don't pay for it. What could I say?

Their method is kind of reverse engineering the DIN 45631. Perceptibility is defined, well, but oddly, to say the least. It doesn't go back to the original data, neither by values nor by the overall concept. As Your own presentation shows, extensive data is rare, and not that easily accessed. Only the concept is explained in the books.

Last but not least the THD is a good estimator regarding the technical predictability of a system. If it sky-rockets, anything can happen. Guitarists love strongly non-linear speakers: real chaos gives their "tone" that nasty twang. Otherwise it is not a good predictor for neutral sound, because music, voice etc consist to an overwhelming part of overtones, with which THD blends quite easily. The masking is very strong.

So, if common people easily detect an overloaded speaker, and they actually do, while ignoring the case, it must be something else. If one presents a synthetic signal with lots of overtones, common folks becomes interested. Do it with subharmonics, as intermodulation has them, people say 'meeh'. Anecdotally, not even with my bass guitar high IM was accepted: "would You please stop burping--behave!"

Do what You like, then.

 

[napilopez]

No, it is right Mads. It is physics of sound propagation. It is not subject to debate.

The module that generates the directivity plot is a *visualization* tool. The sound field data is already computed in a different module and remains unchanged.

When you change the "distance" all it does is change the computational gain. This changes the false color pallet making it look like things are different due to very low resolution of these graphs. All major acoustic events are identical in both measurements. They would not be so if were in near field.

Your speaker doesn't have much output in low frequencies so that area is subject to a lot more noise and inaccuracy. So naturally when you increase the "distance" the accuracy of that data shrinks and the graph changes a bit.

No, this implies that the measurement point was changed but that is not what we are talking about. The distance comes from telling the visualization software to simply show SPL data at one or another far field position. The software is not changing the angle as you are assuming. That would only happen with a single measurement point at two different positions.

It would be nice if it worked this way but it does not because it is projecting the data into far field where this doesn't matter.

For those of us trying to follow along without an NFS in our garage...

So what you're saying is that the way you have the NFS set up, it computes everything to the farfield? In other words, it's not trying to estimate a specific physical distance (say, 2m, the anechoic standard) but rather wherever the farfield begins for the particular speaker being measured? That's cool, but what if you do want to compute a speaker at a specific distance like 1m or 2m -- would the polar response change then?

I ask because what Mads said makes sense to me when you measure a speaker the 'old' way. The polar response does change a bit with distance because the peaker is rotated relative to the baffle and different parts of the speaker are at different distances to the mic during measurement and thus relative SPLs vary a bit -- it is one of the compromises I live with measuring at 1m. Could also be useful to know when measuring a speaker meant for nearfield use, for example.

 

[tuga]

Just curious about the distortion discussion that was taking place. It was said that the S400 has basically the same woofer as the Revel M106, does anyone know why the distortion seems to be very well controlled for that speaker in the Soundstage measurement and not as well in this speaker? Looks like the drive level here is a bit higher, does that explain it?

https://www.soundstagenetwork.com/i...&catid=77:loudspeaker-measurements&Itemid=153

Soundstage measurements are performed at 2 metres.

 

[edechamps]

So what you're saying is that the way you have the NFS set up, it computes everything to the farfield? In other words, it's not trying to estimate a specific physical distance (say, 2m, the anechoic standard) but rather wherever the farfield begins for the particular speaker being measured? That's cool, but what if you do want to compute a speaker at a specific distance like 1m or 2m -- would the polar response change then?

My understanding is that the NFS is capable of both: it can give you far field data (which I assume just means infinite distance, or something equivalent), or near field data (i.e. finite distance). It depends on what you select in the software.

From what I understand, @Mads Buchardt is claiming that @amirm is misusing the Klippel software and mistakenly generating near-field polar maps, thinking they're just far-field with some scaling applied. Amir strongly denies that's the case.

Seems like there's an easy way to end this debate and determine if the data truly is computed near-field or not: recompute the polar maps at a crazy short distance, say 5 centimeters. If @Mads Buchardt is right, the data should look very different (similar to a close-mic tweeter measurement, I'd expect). If @amirm is right, the data will look basically identical, only the absolute SPL will be much higher.

 

[Tangband]

Just curious about the distortion discussion that was taking place. It was said that the S400 has basically the same woofer as the Revel M106, does anyone know why the distortion seems to be very well controlled for that speaker in the Soundstage measurement and not as well in this speaker? Looks like the drive level here is a bit higher, does that explain it?

https://www.soundstagenetwork.com/i...&catid=77:loudspeaker-measurements&Itemid=153

Here are some measurements of the woofer:
http://www.audioexcite.com/?page_id=7311

 

[amirm]

My understanding is that the NFS is capable of both: it can give you far field data (which I assume just means infinite distance, or something equivalent), or near field data (i.e. finite distance). It depends on what you select in the software.

From what I understand, @Mads Buchardt is claiming that @amirm is misusing the Klippel software and mistakenly generating near-field polar maps, thinking they're just far-field with some scaling applied. Amir strongly denies that's the case.

Seems like there's an easy way to end this debate and determine if the data truly is computed near-field or not: recompute the polar maps at a crazy short distance, say 5 centimeters. If @Mads Buchardt is right, the data should look very different (similar to a close-mic tweeter measurement, I'd expect). If @amirm is right, the data will look basically identical, only the absolute SPL will be much higher.

You can't do that. The system extrapolates out but not in. As such it won't let you put in shorter distance than measurement radius.

 

[Francis Vaughan]

You can't do that. The system extrapolates out but not in. As such it won't let you put in shorter distance than measurement radius.

That seems just plain weird. Why would it stop you doing that if it was nothing but a rescaling?
More to the point - what defines the "measurement radius"? I had understood that this was the radius that the holographic field was projected to. As such it should have the angle to driver related anomalies as a function of distance welded into it. That is the point of projecting from the holographic field. One assumes that Klippell provides calculation of the field in this manner to allow for design of near-field monitors. Thus one would expect that it is an explicit parameter to the reconstruction if there is a desire to see the result in the near-field. Otherwise one assumes it is asked to reconstruct in the far-field, and any visualisation is nothing but a rescaling. Not letting you rescale nearer suggests that someone knows that this isn't entirely valid.

OTOH, this still likely doesn't matter. Since the question is the interference effect between the PR on the rear and woofer, there is no error directly on axis. cos(0) = 1. Assuming 0.3 metre depth of the box. At 45º the path length differences from on axis are - at 0.5m radius we see 57mm, 1m - 69mm, and 2m - 78mm. So at 45º maybe a 2cm error changing from from a 0.5m sphere to a 2m sphere versus re-calculating the sphere from the hologram. At 90º the error reaches 6cm. At 500Hz this will still make a small difference. Half wavelength is about 30cm, so maybe sin(1/5 radian) = 0.2, or a 20% = 0.8dB change in any interference effect. As a very very rough back of the envelope calculation. (I really have done this on the back of an envelope so errors are more than possible.)

 

[andreasmaaan]

Just curious about the distortion discussion that was taking place. It was said that the S400 has basically the same woofer as the Revel M106, does anyone know why the distortion seems to be very well controlled for that speaker in the Soundstage measurement and not as well in this speaker? Looks like the drive level here is a bit higher, does that explain it?

https://www.soundstagenetwork.com/i...&catid=77:loudspeaker-measurements&Itemid=153

Soundstage measurements are performed at 2 metres.

90dB/2m is roughly equivalent to 96dB/1m, ie the same level as Amir measured, so that can't be the explanation.

As to what the explanation is, I can only speculate that it may be:

• The M106 may use a modified version of the driver.
• At 96dB/1m, we're at/near Xmax, so 1 or 2dB changes in output (ie differences within the margins of error of the measurement systems) may lead to large changes in nonlinear distortion.
• The M106 is tuned 10-15Hz higher, so above 50Hz, at a given SPL the voice coil is displaced less.
• The NRC chamber is not accurate below around 100Hz (this is in fact obvious from their measurements).

I'm not sure which or all (if any) of the above may apply.

EDIT: and ofc, the passive radiator will be contributing some distortion.

 

[Juhazi]

Distortion looks different plotted as spl vs %.

Loudspeaker drivers are not linear with distortion, at some point % will jump and D2, D3, D4, D5 etc. live their own life. All this comes from a typical dynamic loudpeaker driver's mechanic properties - construction details (excursion limit, behaviour at overexcursion, cone stiffness and resonances, spider stiffness, surrond modulus etc.)

When a driver is put on the surface of the box, things get even more difficult because of internal standing waves, reflex tuning, port resonances etc.

https://audioxpress.com/article/Measurement-and-Perception-of-Regular-Loudspeaker-Distortion
Conclusions
Linear and nonlinear distortion is unavoidable in current electroacoustical transducers using a moving coil assembly driving diaphragms, cones, and other radiators. The regular distortion is deterministic and can be predicted by using linear and nonlinear models and identified loudspeaker parameters in an early design stage. Finding acceptable limits for those regular distortions is an important part in defining the target performance at the beginning of loudspeaker development.

Subjective evaluation is required to assess the audibility and the impact on perceived sound quality. Some distortions which are audible might still be acceptable or even desirable in some applications. Systematic listening tests, nonlinear auralization, and objective assessment based on a perceptual model are useful tools to assess regular distortion.

 

[tuga]

Distortion looks different plotted as spl vs %.

Loudspeaker drivers are not linear with distortion, at some point % will jump and D2, D3, D4, D5 etc. live their own life. All this comes from a typical dynamic loudpeaker driver's mechanic properties - construction details (excursion limit, behaviour at overexcursion, cone stiffness and resonances, spider stiffness, surrond modulus etc.)

When a driver is put on the surface of the box, things get even more difficult because of internal standing waves, reflex tuning, port resonances etc.

https://audioxpress.com/article/Measurement-and-Perception-of-Regular-Loudspeaker-Distortion
Conclusions
Linear and nonlinear distortion is unavoidable in current electroacoustical transducers using a moving coil assembly driving diaphragms, cones, and other radiators. The regular distortion is deterministic and can be predicted by using linear and nonlinear models and identified loudspeaker parameters in an early design stage. Finding acceptable limits for those regular distortions is an important part in defining the target performance at the beginning of loudspeaker development.

Subjective evaluation is required to assess the audibility and the impact on perceived sound quality. Some distortions which are audible might still be acceptable or even desirable in some applications. Systematic listening tests, nonlinear auralization, and objective assessment based on a perceptual model are useful tools to assess regular distortion.

I think that at this point it is becoming obvious that Amir's measurements should be complemented with a serious listening assessment, preferably long-ish term, adequately positioned in an acoustically-neutral room, with proper methodology, by trained listener(s).

 

[GelbeMusik]

You can't do that. The system extrapolates out but not in. As such it won't let you put in shorter distance than measurement radius.

Because You cannot conclude from a given soundfiled where it comes from, but You clearly predict into what it will develop? Huygens principle at work, I guess. Green's function won't do.

But, why shouldn't the apparatus ignore some 'reference axis' while measuring? I would expect that to be abitrary, or, better to say, one might choose it afterwards to ones liking. The relative amplitudes would be referred to the chosen one and done.

 

[MZKM]

The NRC chamber is not accurate below around 100Hz (this is in fact obvious from their measurements).

Yep, anything below ~150Hz should be disregarded for the frequency response. Now, does this also make their distortion measurements below ~150Hz invalid, I do not know.

 

[tuga]

Just found this on the Hi-Fi World website and was reminded of the S400 blip.

Common response errors are shown by dashed lines.
2 - under damped bass peak gives lively, 'obvious' bass.
3 - over damped bass gives 'tight' bass and suits near wall placement.
4 - a small up / down blip suggests a strong internal reflection and resultant colouration.
5 - a dip around 3kHz is due to driver mismatch in the crossover and softens the sound.
6 - raised output from the tweeter, above 3kHz, results in a bright sound with strong detail.
7 - treble peak at 15kHz produces sharp treble.
8 - falling output from the tweeter, above 3kHz, results in warm sound.

 

[Mads Buchardt]

No, it is right Mads. It is physics of sound propagation. It is not subject to debate.

The module that generates the directivity plot is a *visualization* tool. The sound field data is already computed in a different module and remains unchanged.

When you change the "distance" all it does is change the computational gain. This changes the false color pallet making it look like things are different due to very low resolution of these graphs. All major acoustic events are identical in both measurements. They would not be so if were in near field.

Sorry, but I need to correct your understanding of the visualization tool here. All data of the visualization tool are shown as they are at the actual shown distance. The only exceptions are:

• Sensitivity data, which is referenced from 10m SPL curve to 1m distance. Below is the headline of sensitivity plot shown, where you can see how it is noted when data is referenced to new radius:

• CEA2034 curve, which is 2m SPL curve referenced to 1m distance.

This is the entire point of the use of NFS system, that we as designers can optimize a speaker to work well at all distances without having to redo the measurements. We had a call with Klippel (C.B., name left out, but I think you know him) yesterday, to sort this misunderstanding out – and they confirmed to us, that our understanding of the system is correct.

You will see this clearly, if you show our S400 contour plot at 10m distance instead of 66cm distance. You will see clearly that the patterns are not the same – and that true far-field data looks like the one we present in our documentation.

The reason for us to highlight this is that since you choose to show objective measurements in your reviews, which we fully support, it is important that the measurements are shown with the right backgrounds. Our recommendation remains – always show far field data for speakers >2m distance. These near-field data will be misinterpreted by users unfamiliar with how the NFS data works, and will show near-field artefacts, that are not to be considered correct.

Your speaker doesn't have much output in low frequencies so that area is subject to a lot more noise and inaccuracy. So naturally when you increase the "distance" the accuracy of that data shrinks and the graph changes a bit.

Wrong again. Fitting error for your measurement will show you at which frequencies the speaker spherical model is valid. The speaker spherical model will be valid at any distance you calculate based on that measurement. You are correct that SNR at low frequencies may cause it is difficult for the system to calculate a proper speaker model. But in our case of S400 measurements, we have valid data all the way down to 15Hz(which indicates speaker has proper output there), with correctly selected ‘tweeter point‘, sweep level and calculation points. See below curve.

No, this implies that the measurement point was changed but that is not what we are talking about. The distance comes from telling the visualization software to simply show SPL data at one or another far field position. The software is not changing the angle as you are assuming. That would only happen with a single measurement point at two different positions.

It would be nice if it worked this way but it does not because it is projecting the data into far field where this doesn't matter.

Sorry to play it hard here, but please firstly understand the NFS system, before commenting any further on this. This is exactly why this is such a valuable tool for engineers. We can see the exact radiation patterns for the speaker at any distance and angle, but you must understand the system firstly before interpreting the data. Actually what we do in development is to measure each driver of the speaker one-by-one, and then bring them into an advanced simulation tool, where we can analyze them in detail.

Just to state again, im not the engineer here, this is the information I get from our engineering team

 

[QMuse]

Hmm.. The way I see it this kind of difference in opinions would have to be resolved, if necessary with Klippel as arbitrator.

 

[amirm]

Wrong again. Fitting error for your measurement will show you at which frequencies the speaker spherical model is valid. The speaker spherical model will be valid at any distance you calculate based on that measurement. You are correct that SNR at low frequencies may cause it is difficult for the system to calculate a proper speaker model. But in our case of S400 measurements, we have valid data all the way down to 15Hz(which indicates speaker has proper output there), with correctly selected ‘tweeter point‘, sweep level and calculation points. See below curve.

The "curve below" undermines your case. Noise spikes exceed the speaker output starting at 20 Hz. This impacts both the sound field prediction *and* measurements used to confirm the same. To try to rely on accuracy of a single when signal is below the noise makes no sense. Clearly this impacts the visualization of that field in that region and any comments made regarding what it looks like is quite suspect.