How far back are we talking?
there was just this one:
http://www.diyaudio.com/forums/multi-way/159840-measurement-approach-2-way-closed-box-7.html
there was just this one:
http://www.diyaudio.com/forums/multi-way/159840-measurement-approach-2-way-closed-box-7.html
Last edited:
The problem with that discussion is that it assumes the far field response extends all the way to the frequencies where the response is in full 4Pi radiation. In general it is going to be difficult to gate the far field response with long enough window to extend that low in frequency. The method most commonly used is to obtain a far field measurement that extends to the frequency where the baffle step is still developing, typically between 200 and 300 Hz. Then a near field measurement is made. Next the 2Pi to 4Pi transition is modeled using a baffle diffraction simulation. This is then superimposed on the near field response to correct it for far field effects through the baffle step region. Then the far field and correct near field are overlaid which will show a region of overlap, again typically between 200 and 300 Hz. The near and corrected far field data can then be merged at a frequency where the overlap is best.
Hi Earl,
I'm not sure if this pdf has appeared in any DIYAUDIO threads, but it may be of interest.
http://www.fesb.hr/~mateljan/arta/AppNotes/AP4_FreeField-Rev03eng.pdf
Regards.
Peter
I'm not sure if this pdf has appeared in any DIYAUDIO threads, but it may be of interest.
http://www.fesb.hr/~mateljan/arta/AppNotes/AP4_FreeField-Rev03eng.pdf
Regards.
Peter
It was for the benefit of anyone reading the thread.
The Arta writeup and approach is better, provided that the wave length is large compared to the box dimensions.
Of course there are other "tricks" than can be played to get the low frequency response..
Cepstral editing
and Macthed Filters
Last edited:
Peter
That was it thanks. Good writup. When I went to save it I already had it. I must be lossing something.
Get a mac.
Dan
I did an AES paper on using Spectral Estimation techniques to extend the impulse response almost twenty years ago. This is the technique used in Praxis.
The only thing I see in the Praxis manual about extending low frequency response is the method suggested by Fincham, J. Audio Eng. Soc.,Vol.33,No.3, 1985March1985. The matched filter approach suggested by Benjamin, AES
Convention paper 6218, 2004, is an improvement on that technique.
I disagree that Eric's technique is better. Its a little different but in the end winds up being the same thing - extending the impulse response past the window bu using known characteristics of the system. There are lots of different ways to do this, but they will all yield the same result of they are correct.
There as subtle differences. For example Eric's method, in theory, "eliminates" the tail of the impulse associated with the HP behavior of the woofer cut off. Laurie's approach "shortens" the tail by, in effect, applying a Linkwitz transformation to shift the cut off to a higher frequency. As you know, shifting the the cut off frequency higher doesn't really shorten the impulse since the impulse decays either exponentially, or with an exponential envelope. What it does is shorten the time constant so that the impulse should decay below the noise floor before any reflections enter the picture. But, the shorter the reflection free window, the higher the required shift in cut off frequency. Eric's method does not require any decision on how to shift the cut off frequency.
In any event, important the contribution Eric made was the observation that there is no need to apply any pre-emphasis to the stimulus. It can all be done in post processing. This makes it very eazy to implement.
The contribution I made to all this when it was implemented in SoundEasy was the recognition that it is not always sufficient to use a matching filter which is only the inverse response of the system low frequency cut off. Aside from the influence of the low frequency cut off on the length of the system impulse, there is also the influence of the all pass response due to crossovers between drivers. If the system is a simple 2-way with typical 1K+ crossover frequency, then in influence of the the crossover's all pass response on the impulse has very short time constant and will decay sufficiently within the typical reflection free window. But if the system is a 3-way, with mid to woofer x-o point below 200 or 300 Hz, then the contribution of the all pass nature of the mid/woofer crossover to the system impulse may not decay sufficiently within the reflection free window. The fix is to convolve the impulse of the all pass woofer/mid impulse with that of the filter matching the system low frequency cut off. I discuss this at my web page. In particular, see the text a below Figure 4, and Figure 5 and 6.
John
The techniques that you mention require a prior knowledge of the HP function of the DUT or at least a guess. If you use Spectral Modeling (LPC, ARMA, etc.) approaches, then you don't need this info. The available data gives you an estimate of the impulse from the lowest HP filter and this is used to extrapolate the impulse response past the window. No assumptions need be made about the HP filter. In the end, I believe that a combination of techniques would work best. For example, we know that there has to be a HP filter, we just don't know its frequency or Q. A Prony technique could be used to estimate these two parameters based on available data and then use this to extend the impulse out to infinity or whatever window you want.
As I said, in the end they all have to give the same answer and they are all affected by noise and errors in different ways. How robust each one is to these errors has never been examined that I know of.
The techniques that you mention require a prior knowledge of the HP function of the DUT or at least a guess. If you use Spectral Modeling (LPC, ARMA, etc.) approaches, then you don't need this info. The available data gives you an estimate of the impulse from the lowest HP filter and this is used to extrapolate the impulse response past the window. No assumptions need be made about the HP filter. In the end, I believe that a combination of techniques would work best. For example, we know that there has to be a HP filter, we just don't know its frequency or Q. A Prony technique could be used to estimate these two parameters based on available data and then use this to extend the impulse out to infinity or whatever window you want.
As I said, in the end they all have to give the same answer and they are all affected by noise and errors in different ways. How robust each one is to these errors has never been examined that I know of.
No disagreement, but that is the nice thing about having everything done in post processing. It is easy to change the matching filer and look at the effect on the response. And I can't say too much about gaining experience by actually applying the technique rather than just looking at the theory. The low frequency behavior is pretty much dictated by box design (or T/S parameters) so in practice there isn't as much guessing as you seem to imply. As implemented in SoundEasy it works very well, assuming some proficiency in making measurements to start with.
When I first started discussing the matching filter idea with the developer of SoundEasy (SE) I was not very positive. My previous experience had been in trying to implement Laurie's approach and I wasn't very impressed with what I was getting. But when I started using the beta version of the implementation of Eric's approach in SE I quickly reversed my position. There can be no question that in any of these approaches we are really throwing away the long time behavior of the impulse and replacing it with what we believe is the correct behavior. The result will depend on how realistic that "correct" behavior is.
Still, form my point of view these approaches are little more than convenience features. Personally I still prefer the far field/baffle corrected near field merging approach if I can not make a single anechoic measurement. If it is possible to get 1/2 way through the 2Pi to 4Pi transition with the far field measurement then most decent baffle simulation codes will do a good job of correcting the near field result to the far field distance and it's pretty easy to merge these response in SE.
Nearfield
Hi,
Just to add this link which might make it clearer to some.
Audio Engineering Society UK ‘Can we make quasi-anechoic measurements in normal rooms?’
Iain
Hi,
Just to add this link which might make it clearer to some.
Audio Engineering Society UK ‘Can we make quasi-anechoic measurements in normal rooms?’
Iain
Nice writup, but seems wrong in step 2. Shouldn't the filter have a much LOWER cutoff that the loudspeaker rather than the much HIGHER as stated? Much higher would be a problem in reconstruction since the data that we want is now gone.
The only filter that is going to make the frequency response flat to DC is the inverse of the speakers low frequency roll off. That is exactly Eric Benjamin's Mathced Filter approach.
Thats correct, but any target response that is below the true speakers roll-off will also work. Seems to me that it would be better to use a target other than flat to DC because DC target will yield a filter with a double pole at zero Hz. If the target is a HP filter then this is just a simple bandpass and no poles at zero.
But remember that this is being done as post processing to an impulse that has finite length. Thus the poles at DC never come into the picture.
John
Thats not true. A minimum phase filter that "flattens the response to DC" would always have two poles at zero Hz because the acoustic radiation always has two zero's at DC. Now if you are saying that because of errors in the data, truncation, noise, etc. the response is never zero at DC, then that is true, but that is because of the errors not because of the post processing.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.