KEF LS50 Meta Review (Speaker)

[Tangband]

A filter just reduces the signal level. Look at the chart, distortion is already very high from around 150Hz. A filter at 90Hz will make no difference as signal level seen by the driver at that frequency will be the same as unfiltered.

A filter slope starts higher in frequency than the target crossover , so you have a bit protection and thus less distortion even at 110 Hz .
Reading at Erins audio corner, where he has tested 3 different uni-q drivers, one can se that all of them really do well as midrange units, crossed over at 400 Hz , which should be optimal.

 

[sarumbear]

It's quite possible IMD on this is worse and more audible than THD. But I agree with you. I don't think everyone understands that THD measurements don't involve multiple frequencies, that's what I was saying.

IMD is never higher than THD at low frequencies. It becomes an issue on high frequencies where due to system bandwidth you can’t hear the harmonics. For instance IMD is an issue on DACs above 10kHz.

I can’t imagine anyone who knows what the acronym THD means not knowing that it’s about the distortion of a single frequency signal. If you are that ignorant why even bother about the distortion of a device? First learn then bother

 

[sarumbear]

A filter slope starts higher in frequency than the target crossover , so you have a bit protection and thus less distortion even at 110 Hz .

Really?

 

[Sancus]

IMD is never higher than THD at low frequencies. It becomes an issue on high frequencies where due to system bandwidth you can’t hear the harmonics. For instance IMD is an issue on DACs above 10kHz.

Citation? I have not seen any actual detailed study on the audibility of IMD in loudspeakers, just various nonstandardized attempts at measurement that result in unclear graphs, so I don't know how to make informed statements about IMD one way or another. I have seen claims that it can be a significant issue with 2-way coaxials from credible sources, but that's it, and they weren't detailed.

 

[Tangband]

Really?

It depends on the steepnes of the slopes, a 48 dB/oct HP filter will start much nearer 90 Hz , than a 12 dB/oct filter where the response will start to go down much higher up in frequency , maybe as high as 150 Hz , if the target crossover is 90 Hz .

 

[Chromatischism]

The KEFs are a lot louder so unless people realize this biases our perception, they are going to believe the KEFs sound better just because of that.

 

[abdo123]

In general i would advice people not to do any 48dB/oct crossovers it does nothing to IMD and makes the phase messy.

stick with 24dB/oct for best results.

 

[sarumbear]

…a 48 dB/oct HP filter…

 

[Tangband]

Citation? I have not seen any actual detailed study on the audibility of IMD in loudspeakers, just various nonstandardized attempts at measurement that result in unclear graphs, so I don't know how to make informed statements about IMD one way or another. I have seen claims that it can be a significant issue with 2-way coaxials from credible sources, but that's it, and they weren't detailed.

It would be really interesting if Amirm could measure IMD with loudspeakers. My guess is that a 4-way generally should be better at IMD than a 1 or 2 way loudspeaker.

 

[Sancus]

It would be really interesting if Amirm could measure IMD with loudspeakers. My guess is that a 4 way should be better at IMD than a 1 or 2 way loudspeaker.

I agree it would be interesting, but it's very complicated because a) there's next to no information on how much is audible, even less than on THD which is already very weak and primarily based on studies that use test tones which don't include masking and b) there's no standard methodology for either measurement OR visualization of it.

 

[Chromatischism]

It would be really interesting if Amirm could measure IMD with loudspeakers. My guess is that a 4 way should be better at IMD than a 1 or 2 way loudspeaker.

It should be. But are the audible benefits enough to outweigh the potential downsides?

Nearly every crossover measured here has a response dip and a directivity change. You add more of those and things could get messy. Then there's care needed in choosing the right points so you don't have people talking out of different drivers.

 

[sarumbear]

Citation? I have not seen any actual detailed study on the audibility of IMD in loudspeakers, just various nonstandardized attempts at measurement that result in unclear graphs, so I don't know how to make informed statements about IMD one way or another.

There is no standard for IMD measurements on speakers. If you know otherwise please show.

During the design of our Silver 5L we measured IMD using frequency pairs of 50-70Hz and 500-700Hz. It was order of magnitude less than THD within the SPL limit of the speaker. We gave up.

 

[amirm]

It would be really interesting if Amirm could measure IMD with loudspeakers.

I have tried. If you are looking for IMD as a single number, I have found it to not make any sense. Your first problem is to decide on which pair of frequencies and what amplitude ratio. I tried tons of combinations and none correlated to anything useful as I swept the level. A coaxial that was supposed to have higher IMD didn't and so on.

Note that THD information just falls out of frequency response measurements. A special "log chirp" is used where harmonic distortion just shows up as shadows. With harmonic distortion, we get the non-linearity at all frequencies at once. There is no such thing for IMD. You have to pick two frequencies per above and pray that they are the right ones, deal with noise floor issues, etc.

 

[Sancus]

There is no standard for IMD measurements on speakers. If you know otherwise please show.

During the design of our Silver 5L we measured IMD using frequency pairs of 50-70Hz and 500-700Hz. It was order of magnitude less than THD within the SPL limit of the speaker.

Well that's fine, but it doesn't prove anything about speakers in general, nor does it answer my question because you can measure the amount but you can't say how audible it is or is not.

In general I don't think THD/IMD are very useful. A number of studies(Perception & Thresholds of Nonlinear Distortion using Complex Signals, Geddes' work, etc) have found that the numbers they produce have little correlation with perceptual preference. There are suggested alternative distortion measurements but nothing that has seen significant adoption as far as I know.

The state of distortion measurement is honestly very poor. It should be possible to produce graphs where the information corresponds, in a clear and understandable manner, with audibility thresholds so that one can see, in one step, where a speaker is exceeding those thresholds and how bad it would be. However we cannot do that. So the whole situation is a big mess as far as I can see.

 

[Sancus]

I have tried. If you are looking for IMD as a single number, I have found it to not make any sense.

Have you looked at alternative proposed distortion metrics eg the DSMetric evaluated in the above study or any others? There seem to be a few, with claimed much better correlation with preference. But I'm not knowledgeable enough to evaluate whether any of them are worthwhile.

 

[sarumbear]

you can measure the amount but you can't say how audible it is or is not.

That’s because everyone will have a different perception. You can’t place a human within a test system.

 

[amirm]

Have you looked at alternative proposed distortion metrics eg the DSMetric evaluated in the above study or any others? There seem to be a few, with claimed much better correlation with preference. But I'm not knowledgeable enough to evaluate whether any of them are worthwhile.

I have read a few but not that paper. I just glanced at it and while there is excellent introductory information, I am not a fan of creating a single number metric like that, especially since there is no tool provided to readily compute it. The criticism there is also about THD. That is the wrong focus. I like to look at the harmonic spectrum and then using psychoacoustics, determine level of audibility in the full spectrum. Louis Fielder has done excellent work in this area with respect to audibility of low frequency distortion. I hope to put this in place at some point but finding the time to do the computation has been hard.

 

[NTK]

At one point Amir's speaker test suite included multi-tone tests. Unfortunately, the sound annoyed Mrs a great deal and Amir had to stop running this test. Here are examples from the Salk WoW 1 review.

 

[sarumbear]

I have read a few but not that paper. I just glanced at it and while there is excellent introductory information, I am not a fan of creating a single number metric like that, especially since there is no tool provided to readily compute it.

They say: “sound pressure can be a metric in the context of human sound perception or temperature can be a metric for human perception of heat. The audio industry has the need for a proper metric relating to the human sound perception of nonlinear distortion.”

They compare a one dimensional and static SPL and temperature measurement with a two dimensional (frequency and level) and dynamic audio signal. At best they are very naive or their understanding of human perception of sound is lacking.

 

[amirm]

At one point Amir's speaker test suite included multi-tone tests. Unfortunately, the sound annoyed Mrs a great deal and Amir had to stop running this test. Here are examples from the Salk WoW 1 review.

It wasn't just the annoyance but the fact that I could not interpret them in any useful way. The multitone test gets heavily impacted by the frequency response differences and noise floor. Combined, they make it next to impossible to draw any conclusion from them. But yes, it is not for lack of trying.