Low frequency in room measurement

[john k...]

I have added a page to my web site addressing recovery of low frequency quasi-anechoic low frequency response from in room measurements using what has been referred to as the matched filter approach. I've tested it and if done correctly it works quite well. It will be included in the next release of SoundEasy. Here is a link to my web page.

[john k...]

Oh, also, if you have any questions or comments post them here.

[applehorn]

Hi John,

The local AES section recently had a lecture from John Vanderkooy on this very subject.

http://www.aes.org/sections/uk/meetings/index.html Scroll down to March 10th 2008

The concept was first proposed by Laurie Fincham (KEF) about 25yrs ago, but ignored by all. Another recent proponent is Juha Backman.

It would seem that the critical elements are, the correct model for the LSpkr, and the selection of the right type of window. 1/2 Hann seems to be the best. But John V showed a window that he had designed, which might be better.

Iain.

[john k...]

Originally posted by applehorn
Hi John,

The local AES section recently had a lecture from John Vanderkooy on this very subject.

http://www.aes.org/sections/uk/meetings/index.html Scroll down to March 10th 2008

The concept was first proposed by Laurie Fincham (KEF) about 25yrs ago, but ignored by all. Another recent proponent is Juha Backman.

It would seem that the critical elements are, the correct model for the LSpkr, and the selection of the right type of window. 1/2 Hann seems to be the best. But John V showed a window that he had designed, which might be better.

Iain.

Yes, Finchman (1985) at KEF did something similar then Benjamin (2004) took it to the step that is in SE and discussed on my web page. SE includes a variety of windows including the 1/2 Hanning and the 10% cosine Benjamin proposed. I haven't seen Backman's work not Vanderkooy's but the component missing in Benjamin's work was the effect of a crossover at lower frequency as in a 3-way. As I showed at my site, if you don't include that all pass part of the response in the matching filter the results are junk.

In my mind it's all just impulse editing because fundamentally you need to know what the low frequency response of the system is ideally supposed to be to apply the method. For a sealed box that is pretty easy to obtain through T/S parameters. Vented boxes may be a little more problematic.

[djarchow]

John,

You beat me to the punch. How does a vent fit into the process?

Regards,

Dennis

[AndrewJ]

I was at KEF when this method was developed. Fincham's original concept was to extend the response to zero, but of course using analog techniques this was not feasable. We could not do all the processing in the computer because of limits in the accuracy of encoding of the measured impulse response.
Extending the response to anything other than zero frequency ended up lengthening the impulse.
That was when the idea of converting the response to a 2nd order HPF at 200Hz was proposed. We knew that the impulse response of such a filter would decay well within our truncation period requirements, with room to spare for errors.
The truncation window was a half cosine, of somewhere around 10% of the total window, but the exact length varied with the specific conditions.
We had many conversations with John Vanderkooy at that time about the validity of the technique.
We had Peter Baxandall build us an analog filter box that we could dial up any combination of Q and Fo that the speaker had, and it would apply a step frequency response to convert to a true 2nd order 200Hz HPF.
Because of S/N requirements, this was inserted after the microphone.
The technique worked very well and was used in all of our measurements where we needed to measure the low frequencies accurately.
In fact the circuitry developed for this became the basis for the first KEF Kube!
To determine what the approximation to the Q and Fo of the speaker was, we would either measure the nearfield response or measure the impedance, fit an equivalent circuit to this and from that calculate the response.
In the case of a 4th order system (a reflex speaker), the filter requirements were too onerous for this to work well. The simple solution was to block the vent to convert to closed box, measure this response, then measure the blocked and unblocked impedance, fit both impedances to the equivalent circuit model, then calculate the difference frequency response between blocked and unblocked and apply this as a correction the measured blocked vent farfield response.
Sounds very involved, but in practice it was easy, plus it was routine to measure all theses impedances and fit to the equivalent circuit anyway as part of our development procedures.
Of course since those early days (late 1980's) I have been able to implement this without recourse to analog filters, and routinely do this within MLSSA or LEAP.
Of course, the secret to good results is not quite straightforward, as there are many errors that can creep in, but the basic technique does work.
As for the effect of the xover, well since we were developing the speakers, we were making all the measurements on the drivers directly, no xovers. The Fourier analysers we used were able to do any and all the xover synthesis and optimization that is nowadays available for example in Soundeasy or LEAP, so we did not need to measure with the filter in circuit.
Clearly, the effect of the filter is to extend the impulse response, so it needs to be taken account of in the matching filter response.

Regards

Andrew

[soongsc]

Sounds like a wish come true.
I do hope your are using "window" insteady fo "widow" though.
Is there a minimum room size using this method?

Quite a while back, I recommended a "moving rate of change limiter" type of impulse editing applied out side the window. I think this would probably solve the ported enclosure design problem.

[john k...]

Dennis, George: a vented box is be not a problem if you can define the correct matching filter. Benjamin's paper used a vented system as the example. SE V16 will have some canned filters where the user specifies cut off frequency, Q and slope, or the user can define the matching filter any way he likes using the CAD screen. I have also suggested to Bohdan that it would be nice to allow the user to use the result of a box simulation directly. In effect, you can do a box simulation and end up using that as the matching filter as the code is now, with a little work.

Andrew,

I read Fincham'she paper some time ago, as well at Benjamin's. Bohdan (the SE developer) and I discussed Fincham's work back when cepstral editing was introduced in SE. For sealed box it is basically nothing more than applying a Linkwitz transformation or other pole shifting filter to the system. As you note, extending the response to zero would be ideal, but impractical. Aside for the requirement of infinite gain at DC there is also the problem of excess driver excursion at low frequency, well before the DC point is reached. That is what is nice about the Benjamin approach. It requires no preprocessing of the stimulus. Any method of obtaining the system impulse response can be used. The inverse of the matching filter is then applied to the measured, un-windowed impulse. This removes the long time contribution of the impulse arising from the low frequency cut off (and any woofer/mid crossover if included) from the measured impulse. This impulse can then be windowed to removed room effects (since they are the only remaining contribution to the long time response) zero padded, and convolved with the matching filter to obtain the quasi-anechoic response of the system. My contribution, at least with regards to SE, was recognizing the need to include the all pass response of, for example, the woofer/midrange crossover if it has a contribution to the long time impulse. Perhaps Backman or Vanderkooy discuss that, but I haven't seen their work.

Noise is a consideration, but no more so that with any quasi-anechoic measurement. There must not be any spurious noise sources in impulse. The other thing here is that you can play with the matching filter and see how the response changes without needing to obtain a new impulse response. It is pretty obvious when the matching filter isn't correct.

[soongsc]

I think we are all very fortunate to have people like John and Bohdan enhancing tools to help us make good design work easier with less resources. These are not easy tasks, and I enjoy seeing items come off my wish list every year. It's also fun to dream for new features too.

[applehorn]

John K,

During the lecture, John Vanderkooy talked about the effects of cabinet edge diffraction, and how this can contaminate the measurement. Also he talked about the need for long impulse measurement times, because of settling time of woofers. The data presented was very convincing. Do you have any comments on these issues with regard to the subject.

Thanks.

Iain.