• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Ascend Sierra-1 Klippel on-axis measurements

I'll stop posting. But maybe you should try to understand that is not the devine holy Klippel, but the way you measure that makes the difference. It doesn't make any difference whether you measure using Klippel or any other software, as long as you use the correct method. The Klippel robot definitely has it's advantages, but it's just another approach, it's in the end totally not doing anything different from what very smart engineers do for decades. My problem is that your little jihad is putting nice newbees who want to measure a speaker or a room in the assumption they have to buy a 100.000 dollar system to do it right. And that's pathetic.
It is doing something different. First it measures in near field but then computes and presents far field. This sharply increases SNR eliminating the need for a quiet room.

Second, it makes two passes allowing to filter out reflections in the room. This is something that requires massive anechoic chamber otherwise.

Both of these are unique features. Of note unlike methods you mention, the raw measurements are not useful until post processed.
 
I'll stop posting. But maybe you should try to understand that is not the devine holy Klippel, but the way you measure that makes the difference. It doesn't make any difference whether you measure using Klippel or any other software, as long as you use the correct method. The Klippel robot definitely has it's advantages, but it's just another approach, it's in the end totally not doing anything different from what very smart engineers do for decades. My problem is that your little jihad is putting nice newbees who want to measure a speaker or a room in the assumption they have to buy a 100.000 dollar system to do it right. And that's pathetic.
Let me see if I can follow what's going on here. Starts off with a religious conflation. Then says it makes no difference. Then you say the robot has it's advantages, which is a difference. Then you conflate fact with religion again and go right into a non-sequitur and end with a character judgement that was unwarranted.

I mean. 10/10 solid post.

I work in tech, and I work on automating things that were once manual. It's not hard to determine that consistency via automation makes a giant difference in outcomes over time.
 
It is doing something different. First it measures in near field but then computes and presents far field. This sharply increases SNR eliminating the need for a quiet room.

Second, it makes two passes allowing to filter out reflections in the room. This is something that requires massive anechoic chamber otherwise.

Both of these are unique features. Of note unlike methods you mention, the raw measurements are not useful until post processed.
Time Delay Spectronomy did in the, what was it, 80-s?, use windowing to exclude room influences. It was named Techron TEF. Mlssa was a pc-based interpretation of this, to make it a little more affordable. I'm not trying to make Klippel suck or something, that would be ridiculous, I'm only saying that when you measure a frequency response, like in the first post, it doesn't make a difference if you use mlssa, or arta, or rew or, monkey forest, or klippel de pippel. It was suggested that Klippel would do it different from other software, I said all the software is just recording mic data. As long as you have the periphery you can use all this software to measure the speaker and it won't make difference. That robot only makes things easy, costing 100k. But that is not the point. The MLSSA graph in the first post is not inferior to a klippel graph, only the way you meaure it makes a difference.
 
1699566034663.png


The differences are closer than they may appear. A few dB >3kHz on-axis may be due mic positioning, mic calibration, sample deviation or even some small change in the design. <700Hz includes smoothing and room effects, so at I can leave it at that.
 
Last edited:
Time Delay Spectronomy did in the, what was it, 80-s?, use windowing to exclude room influences.
No, it didn't "do that." Klippel NFS also uses windowing but that limits resolution below a few hundred hertz. Heisenberg principle: the smaller time window, the worse the frequency resolution. This is why many of the measurements using windowing have very smooth low frequencies (which are false) or do not show them at all. If you use gating, then you must augment it with another method for low frequencies and stich them together.

I'm only saying that when you measure a frequency response, like in the first post, it doesn't make a difference if you use mlssa, or arta, or rew or, monkey forest, or klippel de pippel.
It doesn't make a difference if you hang the speaker hundred feet up in an open field. Then yes, they are all pressure measurements. But the moment you bring that speaker indoor, all bets are off and approaches differ.

And to be clear, the individual measurements Klippel NFS makes are just the same as any other program (it uses Chirp instead of pseudo random but that is not material in this context). Klippel measurements are however raw data to solve the pde of wave equation, enabling you to get full 3-D radiation of soundfield with specialized processing.
 
They not only should be, but they are. Don't mistake MLSSA for something silly, it was a break trough and was considered the industry standard and is still used. Audiophile Magazine or whatever it is named uses it, as a bunch of other magazines and I know a leading speaker manufacturer that uses my ols MLSSA board for R&D. They had some finished systems measured by a Klippel system but they immediately saw that didn't work out right. That Klippel robot is very clever but I would like to see what it can do that absolutely cannot be done with a large room and software like ARTA, REW, Monkey Forest or ..... MLSSA :).

Bert is correct, MLSSA is not and has never been some hobbyist device, it was indeed the industry standard for loudspeaker measurement for decades. Harman used it for the majority of their measurements. You can see this here:


All of the older discontinued products that you see on that above page were taken with MLSSA.

Stereophile (John Atkinkson) still uses it.

For a maximum length sequence analyzer (hence the name, MLSSA - Maximum Length Sequence System Analyzer) it is still the standard and far superior to anything else that uses this technology exclusively. Here is info on what a maximum length sequence is: https://en.wikipedia.org/wiki/Maximum_length_sequence

MLSSA is hardware based and this is the reason it hasn't evolved to produce more graphically friendly plots. The MLSSA board requires a full size ISA slot. I had many conversations with Doug Rife back in the day, he had given thought to modernizing the board so it can be used in a PCI slot. However, as I recall - he did quite well and decided to retire instead. MLSSA cards aren't inexpensive, back then - the base board was about $3500, then you can add SPO (speaker parameter option) for taking Thiele/Small parameters as well as other very useful options. We also added what is known as the RCAI box (Remote Controlled Analog Interface) to both of our complete MLSSA systems, which automated many functions and eliminated the hassle of continually connecting / disconnecting various cabling required for different testing procedures. We also integrated our main MLSSA system with a turntable (MLSSA can control various turntables) such that off-axis measurements (and spinorama) can be fully automated. This was 25+ years ago :)

I often search ebay to find old 386 based PC's that can take a full size ISA slot so that we have backup computers available, as well as having old copies of windows 95/98 around, as MLSSA software is DOS based but it runs well in a dos window on these old operating systems. Loads of fun and great nostalgia for me using DOS.

The problem with MLSSA, or really for any measurement system except Klippel NFS, is - of course - how to get an accurate anechoic bass response without having an anechoic chamber. For MLSSA (and nearly all other measurement systems), the process is to use nearfield or groundplane measurements and then "splicing" that measurement into the more accurate response where reflections can be gated out of the time domain. Accuracy of this technique was perfectly acceptable back in the day. MLSSA automates that splicing but the results are never that great and this can be seen in every Stereophile measurement where there is always a large bump in the bass response. One can manually adjust the the frequency of the splice, which can help reduce that bump, but then other inaccuracies appear above that bump. Harman used the MLSSA in their anechoic chamber, so their bass response measurements were far more accurate.

Comparing our original published Sierra-1 measurements (taken about 20 years ago with MLSSA) against a Sierra-1 we built up just to measure on the NFS is an interesting study. The measurements are actually closer than I expected them to be, again - we are comparing a speaker assembled measured and designed 20 years ago to one we just recently built up. Low frequency response between the two measurements are quite similar, while the NFS is showing -4dB at about 50Hz and MLSSA was showing -2dB at the same frequency. MLSSA measurement is not showing that midbass droop however, and that goes back to what I mentioned above regarding the splicing. If we spliced at a lower frequency, that upper midrange response would show more droop, but then we would have a large bass "bump" - basically, it comes down to choosing the inaccuracy. Sensitivity appears to be nearly identical, and we can clearly see that slight rise in the response starting at ~600Hz, and the dip at 3kHz which is caused by cabinet diffraction. Both NFS and MLSSA are showing this dip to be about 3dB. NFS is showing a slight HF rolloff at about 17kHz while MLSSA does not show this. This could simply be tweeter tolerances or even microphone differences. Technically speaking, the mic we have with our MLSSA is the better mic compared to what Klippel supplies with the NFS. We are using a very expensive reference standard microphone with MLSSA (ACO Pacific 7012/4012) But again, we are comparing measurements taken 20 years ago to measurements taken maybe 6-7 weeks ago on a speaker assembled 2 decades later than the original.

We still use MLSSA for many functions, it is incredibly fast - I can take an accurate quasi-anechoic measurement with the MLSSA in seconds, and MLSSA still offers many functions that no other measurement software offers. It has extremely robust mathematical functionality, complex macros can be written to fully automate procedures and it is, by far, the best option for production line QC testing. We have even worked out a unique process such that I can use the NFS to fully measure a speaker, compare that true anechoic result to MLSSA measurements of the same speaker, and then using MLSSA's robust math functions to essentially remove the environment, such that MLSSA will then produce true anechoic measurements. This works quite well provided that the environment where we are using MLSSA does not change at all. We do this with our production line testing of all LX, and we will eventually expand this procedure to our other models as well such that instead of customers receiving a production line frequency response measurement that is limited to the range of about 300Hz to 20kHz, it will be accurate down to 20Hz. Problem with this is that my workers are always moving pallets around and if there is anything within a few meters of our testing area - it completely throws the results off so we have to move things around.

Combining Klippel NFS and MLSSA is a very powerful combination and fully expands on the functionality of both devices.

I should mention that Klippel NFS actually uses the exact same technology as MLSSA for frequencies above about 1-2kHz, below these frequencies is where NFS really stands out by switching to wave expansion, but wave expansion fully relies on being able to accurately take a measurement at one distance, and then another at a slightly further distance which in order to do so with the needed accuracy, the system must be able to precisely move the microphone.

We had the Sierra-1 measured at Canada's National Research Council 16 years ago. We had those results, which are more accurate than our measurements due to being taken in an anechoic chamber, linked on our website. There is no longer any reason to link to those measurements because NFS is even more accurate than the NRC measurements, for which everything under about 100Hz needs to be ignored as output from a rear port is not properly captured and the chamber limitation itself is about 80Hz as I recall, maybe even higher. I don't think there are any chambers that are tuned to 20Hz, I don't believe it is physically possible. But if anyone is interested, here they are:


and for anyone interested in learning more about MLSSA:


and


Hope the community here can find this info useful :)
 
View attachment 324999

The differences are closer than they may appear. A few dB >3kHz on-axis may be due mic positioning, mic calibration, sample deviation or even some small change in the design. <700Hz includes smoothing and room effects, so at I can leave it at that.

Yep, they are indeed closer than they appear due to differences in vertical scaling between NFS software and MLSSA software. Biggest difference is that 200Hz-500Hz low Q droop. Glad you noticed and posted this, but if I may ask, how did you capture both response measurements and compare them like this? This is incredibly useful to me.
 
The key difference between the Klippel NFS and MLSSA (and pretty much all other loudspeaker measurement software/systems, e.g. REW) is how they come up with the "quasi-anechoic" measurements.

Klippel uses complex mathematical modeling (holography) and measurements all around the loudspeaker to generate its "quasi-anechoic" data, with no loss in frequency resolution and basically unlimited spatial resolution. Others have to use time gating which comes with the well known limitation of low frequency resolution and the measurement result only apply at the measurement point (or in the direction of the measurement point with distance corrections).

The test signals, maximum length sequences for MLSSA and exponential sine sweeps for Klippel NFS, are basically equivalent for transfer function (complex frequency response) measurements. The benefit of exponential sine sweeps is that it also gives the harmonic distortions (for free).
 
Hope the community here can find this info useful :)
Super informative and interesting response! This is the sort of thing that makes ASR a really cool community, we can argue about something and then the actual company in question comes along and clears things up and educates the rabble. :)
 
Yep, they are indeed closer than they appear due to differences in vertical scaling between NFS software and MLSSA software. Biggest difference is that 200Hz-500Hz low Q droop. Glad you noticed and posted this, but if I may ask, how did you capture both response measurements and compare them like this? This is incredibly useful to me.
The holy of holies of data extraction from images: https://automeris.io/WebPlotDigitizer/tutorial.html

Nice lecture by the creator here:
 
Super informative and interesting response! This is the sort of thing that makes ASR a really cool community, we can argue about something and then the actual company in question comes along and clears things up and educates the rabble. :)

My sincere pleasure, I didn't sense any arguing or negativity in this thread. This is really an interesting study, very few speakers have had the longevity of our Sierra-1 - which was fully designed and optimized using what was the absolute state-of-the-art equipment for that period of time ~20 yrs ago, versus how it measures now using the current state-of-the-art equipment. As I previously stated, they are much closer than I expected. The bass response measurements we took were far more accurate than those taken in the chamber at the NRC, while the midbass and midrange response was more accurate compared to our measurements. Klippel NFS solves this issue by producing more accurate results from 20Hz to 20kHz.

Of course, one must examine full spinorama data to get a true understanding of overall performance, and even that data was better than I expected it to be, but there is most certainly room for improvement.

The big question is, what is next? How will our current NFS measurements compare to the next breakthrough 20 years from now?
 
The differences are closer than they may appear.

Speaking for myself personally, this was the illustration I was hoping to have discussed. And which succeeded admirably.

General side note: I don't think anyone in the thread ever intimated that MLSSA was a hobbyist tool. Just less advanced tech as techniques have progressed.
 
Last edited:
It's not inaccurate at certain frequencies, the data it was being fed was.

That's technically true, but a distinction without a difference. Unless I'm mistaken the Klippel NFS post-processing advanced math algorithms can remedy those issues with the data where the MLSSA cannot - that's kind of the point. Someone can correct me if I'm wrong.
 
The differences are closer than they may appear. A few dB >3kHz on-axis may be due mic positioning, mic calibration, sample deviation or even some small change in the design. <700Hz includes smoothing and room effects, so at I can leave it at that.

Great job on the graph, but I guess it's a matter of opinion what "close" is. There's deviations of 3-4dB and maybe even more depending on how you align those graphs. To me that's more than enough to be termed "inaccurate". A great speaker is one that's +/- 3dB of flat, and if your measurements have deviations that high you basically can't build one of those, like, at all.

I agree it's quite unfair to judge the Sierra-1 by modern standards, but I think this shows that things have improved a lot over those 20 years. And yes, off-axis is definitely important as well but if you're not getting on-axis correct you can't possibly get the off-axis either.
 
A great speaker is one that's +/- 3dB of flat

That would be extremely desirable, absolutely, but I don’t quite agree that should be the only focus for speakers. Many Meyer and Danley speakers wouldn’t be great at all by this strict definition. Have you seen actual measurements in a large venue system? A concert arena or outdoors sports center?
 
A very interesting comparison of "gated" versus NFS measurement - it brings out the nerd in me.

First, both FRs were "scanned" and 1/10 oct smoothing was applied to minimize/smear possible scanning errors:
1699614037809.png 1699614050457.png

The measurements were then "calibrated" to 1kHz, as this is normally the reference frequency for calibrations:
1699614078980.png
Here you can already see that there are basically three frequency ranges with different matches.
This becomes even clearer when we look at the differences in sound pressure level normalized to the NFS measurement:
1699614321032.png

Roughly speaking, in the 500-5000Hz range, the measurements match those of FR measurements with a calibrated mic and with good consistency of the DUT.

Below 500Hz, several problems could come together. Firstly, the gate causes extreme smoothing of the FR.
In a normal room, the gate is usually 3.5ms. So after 3.5ms the first reflections arrive and therefore everything that arrives later than 3.5ms is "zeroed out" in the measured impulse response.

This means that at 280Hz there is just one oscillation within the gate and at frequencies below 280Hz there is not even a whole oscillation available for the processing of the FR. So if the FR changes in this frequency range, this is no longer showed correctly in the gated measurement.
This results that the near-field measurement being added "incorrectly" to the gated measurement.

In addition, there may be errors in the generation of the near-field measurement, such as baffle step correction for woofer and BR port, alignment of SPL for near-field woofer and near-field BR port measurement via the radiating surface and possible phase error if the BR port is not on the same "plane" as the woofer - for example if the BR port is on the rear the BR sound is slightly delayed (wich could for example screw up interaction of BR port resonances with FR of the woofer).

Above 5kHz, the deviation of up to 3.5dB is not small. This could be caused mainly by manufacturing tolerances of the tweeter and the choice of reference point for the measurement.
Since MLSSA is still used by Ascend today, if there were a deviation in the measurement equipment, it would be noticeable in all current measurements compared to the NFS measurements.
 
A very interesting comparison of "gated" versus NFS measurement - it brings out the nerd in me.

First, both FRs were "scanned" and 1/10 oct smoothing was applied to minimize/smear possible scanning errors:
View attachment 325149 View attachment 325150

The measurements were then "calibrated" to 1kHz, as this is normally the reference frequency for calibrations:
View attachment 325151
Here you can already see that there are basically three frequency ranges with different matches.
This becomes even clearer when we look at the differences in sound pressure level normalized to the NFS measurement:
View attachment 325153
Roughly speaking, in the 500-5000Hz range, the measurements match those of FR measurements with a calibrated mic and with good consistency of the DUT.

Below 500Hz, several problems could come together. Firstly, the gate causes extreme smoothing of the FR.
In a normal room, the gate is usually 3.5ms. So after 3.5ms the first reflections arrive and therefore everything that arrives later than 3.5ms is "zeroed out" in the measured impulse response.

This means that at 280Hz there is just one oscillation within the gate and at frequencies below 280Hz there is not even a whole oscillation available for the processing of the FR. So if the FR changes in this frequency range, this is no longer showed correctly in the gated measurement.
This results that the near-field measurement being added "incorrectly" to the gated measurement.

In addition, there may be errors in the generation of the near-field measurement, such as baffle step correction for woofer and BR port, alignment of SPL for near-field woofer and near-field BR port measurement via the radiating surface and possible phase error if the BR port is not on the same "plane" as the woofer - for example if the BR port is on the rear the BR sound is slightly delayed (wich could for example screw up interaction of BR port resonances with FR of the woofer).

Above 5kHz, the deviation of up to 3.5dB is not small. This could be caused mainly by manufacturing tolerances of the tweeter and the choice of reference point for the measurement.
Since MLSSA is still used by Ascend today, if there were a deviation in the measurement equipment, it would be noticeable in all current measurements compared to the NFS measurements.
Did you measure this yourself? Or do you exactly know how these measurements are done? I don't understand the gating part. You shouldn't consider frequencies that tall outside the applied time window. MLSSA and ARTA have a time-bandwidth requirement
A very interesting comparison of "gated" versus NFS measurement - it brings out the nerd in me.

First, both FRs were "scanned" and 1/10 oct smoothing was applied to minimize/smear possible scanning errors:
View attachment 325149 View attachment 325150

The measurements were then "calibrated" to 1kHz, as this is normally the reference frequency for calibrations:
View attachment 325151
Here you can already see that there are basically three frequency ranges with different matches.
This becomes even clearer when we look at the differences in sound pressure level normalized to the NFS measurement:
View attachment 325153
Roughly speaking, in the 500-5000Hz range, the measurements match those of FR measurements with a calibrated mic and with good consistency of the DUT.

Below 500Hz, several problems could come together. Firstly, the gate causes extreme smoothing of the FR.
In a normal room, the gate is usually 3.5ms. So after 3.5ms the first reflections arrive and therefore everything that arrives later than 3.5ms is "zeroed out" in the measured impulse response.

This means that at 280Hz there is just one oscillation within the gate and at frequencies below 280Hz there is not even a whole oscillation available for the processing of the FR. So if the FR changes in this frequency range, this is no longer showed correctly in the gated measurement.
This results that the near-field measurement being added "incorrectly" to the gated measurement.

In addition, there may be errors in the generation of the near-field measurement, such as baffle step correction for woofer and BR port, alignment of SPL for near-field woofer and near-field BR port measurement via the radiating surface and possible phase error if the BR port is not on the same "plane" as the woofer - for example if the BR port is on the rear the BR sound is slightly delayed (wich could for example screw up interaction of BR port resonances with FR of the woofer).

Above 5kHz, the deviation of up to 3.5dB is not small. This could be caused mainly by manufacturing tolerances of the tweeter and the choice of reference point for the measurement.
Since MLSSA is still used by Ascend today, if there were a deviation in the measurement equipment, it would be noticeable in all current measurements compared to the NFS measurements.
Did you measue this? Or do you know exactly how it is measured? I also don't understand the gating part. You shouldn't consider frequencies outside the applied time window. MLSSA and ARTA have a time-bandwidth requirement for that. Or am I missing stuff?
 
Did you measure this yourself? Or do you exactly know how these measurements are done? I don't understand the gating part. You shouldn't consider frequencies that tall outside the applied time window. MLSSA and ARTA have a time-bandwidth requirement

Did you measue this? Or do you know exactly how it is measured? I also don't understand the gating part. You shouldn't consider frequencies outside the applied time window. MLSSA and ARTA have a time-bandwidth requirement for that. Or am I missing stuff?
Oh f*ck, I'm not even allowed to post :D:D.
 
Back
Top Bottom