The following message was posted by a forum member, Brad.
Measuring the impulse response of a loudspeaker/room system is a key to success with Acourate. The measurement is the basis for the correction filters that you will listen to. How to do this is a topic of much debate.
Should you first linearize the speakers with a filter generated from a near field measurement?
Should you measure in one position where you sit or measure multiple positions in the listening area?
How do you set the Frequency Dependent Windows (FDW) for analysis?
I have contemplated this topic for many years and generated countless correction filters with many different software and hardware tools.
Some measurement objectives:
I am mainly interested in making my listening position sound good. I do not believe a solution for the entire listening space is possible with good results. Maybe you can get the frequency response pretty good if you are lucky. The big issue is time of arrival from the speakers. It is absolutely critical that left, right, and center arrival times are simultaneous to your seating position. The sound stage absolutely depends on it. I am very sensitive to this.
I generate a set of filters for my seat and one guest seat. I can easily switch between them to provide another the “best seat in the house” experience.
I remove the listening chairs from the room during measurement. They do more harm than good to the in-seat measurements in my experience. I want to correct for the speakers and the room, not the seats. They are going to move and change dramatically with people in them anyway. You are totally accustomed to listening while in a seat – to people talking for example. They do not need to be processed by a FIR filter to correct for the seat.
Now, where to measure? I usually put the mic in the listening position and that’s it, however, something in me says it’s a bad idea to base my entire effort on this one point in space. There must be a better way to get more robust and confident data and still keep the focus on the listening position of interest.
I found what looks to be a good solution in the attached document. Please look it over.
The technique involves measuring multiple positions along a straight line through the listening position projected from the loudspeaker under test. Then align the measurements in time, add them together, and normalize. You get a more robust measurement that is now more loudspeaker, less room, and less ambient noise. This should also allow for more filter resolution with increased FDW settings. I would consider using ten measurement positions for each of my front speakers over a span of 50 cm or so.
I’m planning to skip linearizing the speakers separately – at least for now. Measuring up close brings a new set of issues and concerns. Blending near and far field filters with serial convolution adds more variables with no assurance that the end result will be improved.
I completed some “beamforming” measurements.
I measured 10 positions at 4″ (100mm) spacing along a line from the speaker through the listening position equally ahead and behind the “sweet spot” over a span of 36″ (900mm). I used an Earthworks M30 omnidirectional measurement microphone. I repeated the measurements for bass, midrange, and treble drivers before moving the mic. There are no passive crossover components.
I added the first five measurements into a new curve, then the second five measurements into a new curve. I added the two results together and then applied -20 dB gain. This works because Acourate automatically time aligns the impulse measurements. Easy.
I believe this measurement method reduces background noise, adds confidence, improves accuracy, while focusing results on the listening position of interest. The results look very encouraging to me. I will send the data files to Uli so he can comment on using this measurement method for filter generation.
In the following posts I will include amplitude plots for my left bass, midrange, and treble drivers which are all horn loaded (Klipschorn bass and PSE-144 coaxial horn on top).
thanks for the pictures and for the pulses
Yes, the average response clearly shows the removal of the “noise” in the frequency response added to the direct signal by reflections. This noise is characterized by the sharp dips in the steady-state frequency responses.
I have made a quick test: summing up all 10 tweeter responses and dividing by 10 (gain factor 0.1). Then I have calculated FDW of 05LT.dbl and the sum with 10/10. You will see that the results are pretty close. The FDW of 05LT show a bit more influence of reflections. So anyway the FDW prooves to work.
Of course you need to test for the final correction by listening Please tell.