The limitations with gating in-room measured impulse responses are directly related to the size of the room, i.e., how far away the walls/floow/ceiling are from the loudspeaker and the microphone. This has nothing to do with software.
For example, if the anechoic part of the impulse response is 3 ms long, the frequency resolution will be 1/3ms = 333 Hz. This is a typical value for measurements taken in a home environment. For better frequency resolution like 100 Hz (or even 50 Hz), one needs a 10 ms long (or even 20 ms long) anechoic impulse response, and hence one needs to do the measurements in a larger room or outdoors.
However, different software will process and display the data in slightly different ways, and it may look like some software packages produce smoother SPL curves than others. You are perfectly right that some of the Monkey Box measurements did show poor frequency resolution in the 300-1000 Hz range, and I was very transparent about this simply because I did not apply a lot of smoothing to the data with MATAA. Other software packages will do this (maybe even by default, I don't know), and the curves will look like there is better resolution (although it's not). The only real solution to the limited frequency resolution is to set up the measurement such that all walls/ceiling/floor are far away from the speaker and microphone (as I did with the Monkey Box).
I have been thinking about other ways to remove (or at least reduce) the room echoes from the measured impulse response data. I somehow have a gut feeling that it must be possible to disentangle the direct sound of the loudspeaker and the overlapping room echoes by modelling the room response somehow. First of all, the shapes of the room echoes must be similar to the shape of the direct sound emitted by the speaker, so the correlation between the first impulse (from the speaker) with the later impulses (room echoes) should allow disentangling the room echoes from the direct sound. Secondly, it might be possible to tease out the room response by measuring at different spots around the speaker. I'll have to take a closer look at how the Klippel system does this, and possibly the papers mentioned in the video from post 21 also have some clues on this.
Hello Matthias (mbrennwa),
That would truly be terrific. Are you familiar with the Klippel n.f. scanner on a shoestring thread here on diya I referred to in post 6?
Maybe Prof. Ivo Mateljan could do something similar for ARTA?
That would truly be terrific. Are you familiar with the Klippel n.f. scanner on a shoestring thread here on diya I referred to in post 6?
Maybe Prof. Ivo Mateljan could do something similar for ARTA?
I would pay upto £500 including shipping for NFS and £150 for kipple driver measurements (BL(x) etc.). A lot of what I would want measuring is very large though (big horns) so unless in the UK it would be problematic!
Probobly for me a large outdoor tower with pre-positioned microphone arc and hoist would be more useful, this is how community and light and sound used to measure large speakers polar response. If someone already has the land this would be cheaper than an NFS.
Probobly for me a large outdoor tower with pre-positioned microphone arc and hoist would be more useful, this is how community and light and sound used to measure large speakers polar response. If someone already has the land this would be cheaper than an NFS.
Last edited:
The question I don't seem to be answered, is why you need so much more resolution?
Or better asked, what extra information will you get that's not in nearfield measurements in combination with a burst decay graph, distortion graphs, impedance measurements and bafflestep simulations?
Better methods have been proposed in another topic which are actually very similar to Klippel's method.
Klippel has to do some kind of stitching method as well.
If someone is happy to do the (tedious) programming part, I am more than happy to join a group effort and think of similar ways that can be used.
In fact currently I am working on way to measure those BL(x), Kms(x) and Le(x) graphs as well.
I looked at this a long time ago, and I forgot most of it. Will have to dive in again once I find a bit of time for this. 🙄
With a 300 Hz resolution, you get a data bin centered at 300 Hz, the next one centered at 600 Hz, then at 900 Hz, etc. The ear is quite sensitive in this frequency range, so it would be good to know a bit better what the acoustics of a loudspeaker are doing in this frequency range.
The far-field SPL response.
Right, I just gave a whole bunch of other methods that will give all that information?
So the question still remains unanswered.
Well I have a neighbor who spends $27k a year to play golf. Another family dragged their rug rats down to Disney World for a week at a cost of almost $10k, and in the case of the 14 year old, literally dragged... she wanted to stay home. So $500+shipping does not seem like a big expense for a recreational hobby.
All you see with near field is the driver. So you miss anything to do with the box. For example, resonances. Or diffraction beyond the simple 2D representation of the baffle sim. Or port resonance interference. But the real problem is the wrong phase information. This is a tricky business when summing a baffle diffraction output with a near field measurement, which then has a HB Transform applied to reconstruct phase. Sometimes it's very close, sometimes it's not. And that result is only valid on axis, what about as you move off axis? I don't know of any design software that will model 3D diffraction - and correct phase! - to be summed with the driver off axis.
So what you miss, are all the other things effecting the frequency response besides the driver, and you're missing the correct phase. Have you not designed a 3-way before?
Obviously at $100k, I'm out. I thought Klippels were $20k. I have asked MISCO if they lease them, but if they are $100k new, how much would a 12 month lease be?
The DIY Klippel is actually a better possibility than a even just a few years ago. The robotic part is doable with all of the 3D printers and hobbyist CNCs having widely available stepper motors and electronics as a source to draw from. The math sounds not that impossible, @bohdan1232000 has already included cepstral editing and matched filters in Soundeasy. And gofundme makes the most important part so much easier. I know if a credible team volunteered to take this on I would be happy to donate.
The DIY Klippel is actually a better possibility than a even just a few years ago. The robotic part is doable with all of the 3D printers and hobbyist CNCs having widely available stepper motors and electronics as a source to draw from. The math sounds not that impossible, @bohdan1232000 has already included cepstral editing and matched filters in Soundeasy. And gofundme makes the most important part so much easier. I know if a credible team volunteered to take this on I would be happy to donate.
No, actually you see ALL of those things you just mentioned, ESPECIALLY resonances.
Port resonance is even one of the easiest to measure, can done and be seen in multiple ways.
Impedance graph, nearfield graph of the port, burst decay of this measurements.
You will probably even able to see this in the nearfield of the woofer.
Second, those port resonance are very predictable from what is expected from theory and simulations.
But even better is to already design it in such a way that they will never be an issue to begin with.
To see the overal phase response you don't need so much frequency accuracy.
Unless you are trying to correct tiny little peaks and dips, which is the wrong way of doing anyway.
Very important next, is it's very debatable what is actually audible here.
Any major diffraction issues will show only up from say roughly 3-4kHz and above, where we have plenty of relative resolution left to see what's going on.
Simulations aren't very accurate here, although that mostly starts from 6-8kHz upwards.
To be very practical here, in the majority of cases, form goes above function here.
Also there isn't an awful lot one can do about it either.
I have build and consulted plenty of 3-way systems.
Have you ever learned to check peoples credentials before responding a certain way?
Kind of silly and disrespectful comment towards someone who has this a professional day-job for the last 15 years or so.
Probably have seen an awful lot more than you as well.
But apparently these days hobbyists don't seem to care anymore whatever professional experience and background certain people have here (I am talking also about other people fyi).
Or even take the time to read ANY literature AT ALL, at the same time even acknowledging that their own knowledge and math is not on par.
Yet seem to have an "opinion" ready at the go about almost any subject they very clearly have no clue about.
Very sad and pretty disrespectful in my eyes.
Yeah, I bumped into a few occasions that a Klippel system would have been very handy. (read: convenient)
But in the end the costs are not worth the benefits in general.
Renting an anechoic room is also not impossible and that expensive either.
For 100k you can rent an awful lot of days in such rooms.
I would totally agree with you if you were talking about line-arrays.
Because those are absolutely a pain to measure well.
But unfortunately they also don't fit on a Klippel system either.
Last edited:
I am sympathetic to your position. A high quality gated set of measurements, vertical and horizontal, along with a good near field response of each driver and vent, is generally good enough. With a monopole speaker (i.e. a woofer in a box), we are making some assumptions about low frequency response, but those assumptions are generally pretty safe assumptions. Especially when the merged response is cross-checked with an outdoor ground-plane measurement.
But I see a great deal of value in having a reference standard measurement. It would serve as a calibration of my gated+NF+merge process. I personally would like to have a confirmation that I am getting the 20 - 1k region right.
For people attempting to build a cardioid radiation speaker, one which maintains a directivity index down to 100 Hz, Klippel is really important. Without an accurate low frequency measurement, you are almost flying blind.
Let's put it this way... if you could get a Klippel NFS set of measurements for $10, would you do it? If the answer is yes, then we agree the data is useful. The only disagreement is over the cost.
j.
I have seen examples where a few elements are stacked and measured (to model adjacent element diffraction effects) and then a model of a single element is created to import into array modeling software.
Martijn from Dutch & Dutch is one of the many people that works with just very standard techniques.
When measured with a Klippel system there were never any real major differences or surprises.
But yes, as I said before, cardiodes are a bit trickier.
Although there seem to be some great extrapolation techniques combined with stitching that seem to work a lot better.
What really helps is that the "difficult" part is within a frequency range that is extremely predictable.
The system here just follows a higher order control system response. Nothing fancy at all.
Obviously it's all costs, if a whole system was $1k, yeah why not.
But people seem to be totally brainwashed by it, and what is much more important is forgetting to understand the fundamental physics at play.
You still don't know a thing if you don't understand what you're measuring.
An higher resolution is never gonna help with that.
Therefor you can get the confirmation yourself already, by checking what you're measuring is in line with what is being expected. For any closed or BR system, that is absolutely no rocket science anymore.
You just have to work systematically, something most people are not doing at all.
Yes, that is one way.
That only works when there are modular modules available.
Some companies have like one big active array.
Which seem to be very popular these days in smaller to mid size venues or buildings like churches etc.
It wasn't the amount which seems reasonable for a hobby interest but what was considered to be of such high value. You are not alone it would seem in placing high value where I and one or two others can see only a modest value to DIYers. There are also one or two odd claims about accuracy/inaccuracy which may have some relevance. Interesting though when people don't see things as expected.
I think a Zoom with b_force and andy19191 would be more productive than talking past each other for pages and pages. You guys available today?
You can f.i. do raw data 0-90 degrees horizontal in a rented anechoic chamber equipped with an electrically controlled turntable and Klippel modules or some other system (the one we use has an absorption coefficient >0,99 down to 70Hz and happens to be use Klippel ) - use this data for VCad work - validate the design the same way - and THEN do the full spin. No use letting that robot work for hours for a full Spinorama on a design besed only to find out it was based on flawed data imo.
Member
Joined 2003
I assume the reason the system costs $100K isn't because that's the parts cost, it's because Klippel is basically amortizing the cost of it's intellectual property. IE, there's probably around 100-ish companies in the world that need a Klippel and they have 26 employees, so you're paying for that.
$100K x 100 Klippels divided by 26 employees = $384,615 per employee.
At that rate, Klippel is barely breaking even. To put that in perspective, companies like Google or Apple generate about $500K - $1M per employee, per year!
I'd estimate that a reasonable price to measure a speaker using a Klippel is around $5000-$10000 per speaker.
If one plans to resell the Klippel, you might be able to get that down to about $2500-$5000 per speaker.
$100K x 100 Klippels divided by 26 employees = $384,615 per employee.
At that rate, Klippel is barely breaking even. To put that in perspective, companies like Google or Apple generate about $500K - $1M per employee, per year!
I'd estimate that a reasonable price to measure a speaker using a Klippel is around $5000-$10000 per speaker.
If one plans to resell the Klippel, you might be able to get that down to about $2500-$5000 per speaker.