Looks like PS Audio is upping its game with their new klippel scanning machine regarding new speaker development. Not sure this would benefit the current speaker Fr line but future speaker development must be high on their list considering the investment in this machine cost 150k. Magico also uses this to develop their new speaker lines with great success; in fact most contribute this machine as the main factor regarding the development of M9,M7 and the new S series.
Yes, I will being doing some videos of how weâre using it on future speaker development.
We have been using other klippel hardware to âlook insideâ of our woofer designs and confirm the FEA simulation I do and that the units were assembled optimally. However, that unit is based at our woofer vendor or weâd send them to the Warkwyn lab in Minnesota. That thing looks like this:
Now, we have the nearfield scanner, which is A LOT MORE EXPENSIVE but measures freefield or half space frequency response and polar response among other things. It has some very unique technology in the software and measuring specific points to very accurate simulate the anechoic far field response in a standard non-anechoic room and works all the way from below 20 Hz. So, in some ways, itâs better than an anechoic chamber and is fully automated. Pretty awesome!
Magico has one and features it prominently and I agree with them. Itâs like having an MRI machine for speakers. Itâs a pretty small club - we have serial number #90.
Yourâe like a kid with a new toy! 
How is it going to be used to improve speaker development in the future? With the development of the current Fr series how could it have helped improve the line if it was available at the time of development? Do you see PS Audio and yourself looking into the future with a vision to develop a better speaker line than the current Fr range of speakers and is this the tool needed to do that?
Hi Mikey,
To understand why this represents and improvement requires some knowledge of how we have been developing speakers. Basically, every speaker measurement has to solve the same fundamental problem: separating the speakerâs own radiated sound field from everything else the mic picks up â floor bounce, ceiling, side walls, stuff in the room.
You can either physically prevent those reflections (anechoic chamber) or mathematically reject them. The traditional gated/spliced approach is the latter, and it works, but the math it relies on is crude: take an impulse response, window it before the first reflection arrives, and FFT whatâs left.
With a typical ~10 ms gate we were able to get with the space we had available, you get roughly 100 Hz frequency resolution, and hereâs the catch â thatâs a linear 100 Hz across the whole spectrum, while everything we care about acoustically is logarithmic. At 10 kHz, 100 Hz is 1% of the frequency, finer than 1/96 octave, far more resolution than youâll ever need. At 1 kHz itâs already 10%, roughly a seventh of an octave, where narrow features start getting lost. At 200 Hz, 100 Hz resolution is half the frequency itself â youâve got half-octave bins, which is essentially no resolution at all. So a gated measurement is wildly over-resolved at the top end and under-resolved exactly where you need it most: the lower midrange and upper bass, where cabinet modes, port/PR behavior, baffle diffraction, and driver breakup all live. Narrow features can get averaged into the surrounding response in precisely the band where they matter. Below ~200â300 Hz the data is so coarse you have to abandon it entirely and splice in a ground plane measurement, which works (the floor bounce arrives essentially coincident with the direct sound and sums constructively) but stitches together two measurements taken under different boundary conditions, with a splice region thatâs always a judgment call and an implicit half-space assumption for the bottom octaves.
The NFS rejects the room differently. It scans the near field on a closed surface around the speaker and decomposes the result into outgoing and incoming wave components, a spherical wave expansion. Room reflections show up as incoming waves and are mathematically discarded; what remains is pure free-field radiation. Thereâs no time gate involved, so frequency resolution isnât set by a window length anymore, itâs set by SNR and scan density, and it can be made effectively logarithmic to match the way speakers and ears actually behave. The technique stays valid down to wavelengths comparable to the scan radius, so you get true anechoic data continuously from 20 Hz to 20 kHz in one dataset, with no splice point and no half-space assumption.
Practically, that means three things you can actually see in the data. You can resolve narrow midrange features that a 10 ms gate would have averaged out â resonances, diffraction ripples, cabinet artifacts that used to sit below the resolution floor of the old method, particularly in the 200 Hz â 2 kHz band where the linear-vs-log mismatch hurts most. You get the bass octaves at the same fidelity and from the same measurement as everything else, so thereâs no transition band where you have to mentally caveat the data. And because every angle on the sphere is reconstructed from a single automated scan rather than dozens of manually repositioned turntable measurements, the polar data is internally consistent â angle-to-angle variation is the speaker, not the operator or the splice. Plus the obvious wins: a scan that runs overnight replaces what used to be a multi-day wrestling match with a 250 lb cabinet on its side. This automation will allow me to get many more turns on the design optimization and better understand the system behavior.
We also got the in-situ compensation module, which extends the same logic to distortion. Harmonic and IM products are small enough that room contamination has historically made anechoic-quality distortion measurements impractical outside of a chamber, especially in the bass where gating doesnât help at all. The compensation module characterizes the environment and removes its contribution from the distortion data, so you can resolve harmonic structure cleanly across the full bandwidth in a normal room.
I should note that, for driver development, we have been using FEA simulation tools and also the Klippel distortion analyzer for large signal analysis and even a bit of their laser come scanning with their SCN system. We didnât own the hardware (because weâre only developing a few drive units) and were sending these out to a lab in Minnesota and using some of the Klippel QC hardware in assembly at our contract manufacturer in order to make sure that we have the parts alignment nailed in production.
That being said, the largest single thing Iâve done to improve the speaker line in the future is making it in the first place. Every time you go through a design process, you are left with a list of things that youâd like to experiment with or change the next time. I have also been working on some more âtrickle downâ product as well.
However, Iâm also working on research stuff and R&D on new drive units. The goal is to save up enough of these changes to make sure that there is a substantial enough improvement to justify a new model.
And your new data sets allow you to refine your FEA models?
At some point I hope your tech responses can be compiled into something like a sticky. You have a knack for boiling down complex topics into a few paragraphs that provide a great basic understanding.
The Klippel NFS is about anechoic frequency response and directivity measurements, primarily. This is like a microscope for speaker perceived tonality, which is a primary contributor to sound quality.
Currently, the FEA I do is on the transducer side of things, designing woofer motors etc. As i mentioned, thereâs a separate Klippel product for that (which is a module we didnât buy because our woofer build-house and a lab in the states we use has that. Klippel recently introduced an add-on/alternate model for characterizing woofer distortion, their FLSI test. Unlike their other modules, theyâre treating this one as an ongoing yearly license cost. It gives more visibility into the voice coil inductance vs excursion vs frequency of a driver and other factors that were not able to be looked at separately in-situ.
Anyway, all that to say that I am refining my FEA models but that will really only affect the woofer magnet and distortion optimization work Iâve been doing.
With the help of some AI vibe coding, I ported an old excel spreadsheet and macro/script using a freeware FEA solver to a more standalone python application and have incorporated a newer, more capable solver and beefed up the parametric modeling/drawing system to quickly âwhat-ifâ things. It now has a much more robust inductance simulation system and Iâm tacking on some flux modulation and current based distortion AC analysis bits now. Currently considering adding in support for other solvers like ElmerFEM and looking at adding more features as time allows. I could post a few screenshots if anyone is interested
Many thanks for the details. Always interested in interesting work, which this most certainly is, especially that helps us understand what we hear and why.
Anything you care to share at your convenience would be appreciated.