Home Theater Forum and Systems banner

1 - 17 of 17 Posts

·
Premium Member
Joined
·
74 Posts
Discussion Starter #1 (Edited)
Many years ago I bought an "Imp" kit from Liberty Instruments. This was more if a speaker measurement tool than room mesasurement. One of its best features was being able to measure gated (quasi-anechoic) impulse responses using MLS or Maximum Length Sequences. Instead of using an audible click to record an impulse, it used MLS sequences which sound like a burst of pink or white noise. The program would then mathematically auto-correlate these recorded bursts with the original MLS signal to extract an impulse response from the first repetition of the MLS sequence. Here's a link that explains the theory. Fortunately this first MLS repetition occurred early enough to still allow evaluation of a gated response that eliminated all room reflections from being evaluated (assuming the speaker under test and microphone were setup in the middle of the room about 1M apart). The downside was the lower end of the final frequency response was limited by the size of the gated interval, or how long between the direct sound to the first reflected sound. In a normal size room with 8' ceilings, the lowest measureable frequency was around 300-400 Hz.

I've been searching REW documentation for a way to measure quasi-anechoic responses in REW. I realize a swept-frequency sine wave can be used to generate an impulse response, but the stimulus takes too long to complete for gating room reflections. Also the audible click impulse method doesn't contain enough energy to provide an accurate measurement, which is why Liberty Instrument went with MLS.

Does anyone know how to do an accurate quasi-anechoic measurement with REW?

Thanks,
Darrell
 

·
Premium Member
Joined
·
74 Posts
Discussion Starter #3 (Edited)
The signal and math are different, the theory and limitations are the same. With REW, gated measurements are accurate and effective and good down to the 300 to 500 Hz or so limitation in-room.
Are you talking about using gated swept sine waves to create the quasi-anechoic impulse response? How do you fit the sweep pattern inside the gated interval?

EDIT: My goal is to measure the quasi-anechoic response of an individual speaker placed in the middle of the room, and not to measure the speaker+ROOM response.
 

·
Premium Member
Joined
·
6,711 Posts
You are misunderstanding how MLS and swept sine work. The repetition intervals are long for both (typically 128k or 256k samples) but that has no bearing at all on making quasi-anechoic measurements. The measurement system extracts the impulse response, then that impulse response is gated to remove the reflections and provide a quasi-anechoic response. MLS and swept sine are the same in that respect. The length or type of the stimulus doesn't make any difference.
 

·
Premium Member
Joined
·
74 Posts
Discussion Starter #6 (Edited)
Thanks guys for clearing that up. It still seems counter-intuitive to me that you can gather anechoic data from a 1-2 sec test signal during all of the room reflections. I was assuming the MLS rep interval was very short (<5 msec) and would finish before room reflections occurred. But if the MLS pattern lasts as long as the swept sine stimulus, then I guess it works somehow.

EDIT: Just to clarify, the article I linked in my first post says the IMP software evaluates the second repetition of the MLS pattern and uses the first rep as a "warm up". I couldn't find in that article where it says how long a single rep lasts, or how many reps are in the test burst, but I do remember the test burst lasting 1-2 sec.
 

·
Premium Member
Joined
·
74 Posts
Discussion Starter #8 (Edited)
Open the IR Window (gate) and set the left and right windows to block the first room reflection as shown below.
I understand how to do the gating once the impulse response has been extracted. My concern is how the initial impulse peak in your example is not already contaminated with room reflections since the swept sine data used to derive this impulse response lasts much longer than the first reflection.

EDIT: Let me further explain my concern with an explicit example. Say as the swept frequency of 1000 Hz is first reaching the microphone, won't remnants of frequencies just below 1000 Hz (from earlier in the swept stimulus) still be bouncing around the room? Won't these spurious (late) frequencies mess up the Fourier Transform that extracts the impulse response?
 

·
Premium Member
Joined
·
1,562 Posts
Sorry, I misunderstood your concern.
I don't know the detailed concepts or the math. It was enough for me that it is a well established method. I have also tried all 3 signals in HolmImpulse (log sweep, linear sweep and MLS) and obtained the same IR. The MLS signal used in Holm is much longer than the sweep so it does not really fit the <5 ms signal you referenced. My understanding is that the log sweep provides a much better S/N ratio making it more popular for many situations.

The charts that I posted above were measured with the mic at my 13ft LP position. It is a little more noisy than when the mic is at 1m so there is some very minor leakage/contamination in the math? It is amazing to me that it holds up so well even at that distance.

I scanned the following paper some time ago, but much of it is beyond me. Maybe there is an answer there for you?

http://melaudia.net/zdoc/comparisonMesure.PDF
 

·
Premium Member
Joined
·
6,711 Posts
The impulse response, however derived, is the response of the whole system, including the room and its reflections. Windowing eliminates the reflections by limiting the region analysed to that containing the direct sound, before reflections arrive - the reflections have to travel further so they arrive later. That's why you move your speaker into the middle of the room - the further it is away from all surfaces the longer til the first reflection arrives and so the more of the direct response you have available to analyse. The wider that window, the lower the frequency response extends. The bit which seems to be tripping you up on the impulse response analysis is confusing the frequency domain measurements (MLS or swept sine) used to derive the impulse response with the time domain result.
 

·
Premium Member
Joined
·
74 Posts
Discussion Starter #11
The impulse response, however derived, is the response of the whole system, including the room and its reflections. Windowing eliminates the reflections by limiting the region analysed to that containing the direct sound, before reflections arrive - the reflections have to travel further so they arrive later. That's why you move your speaker into the middle of the room - the further it is away from all surfaces the longer til the first reflection arrives and so the more of the direct response you have available to analyse. The wider that window, the lower the frequency response extends. The bit which seems to be tripping you up on the impulse response analysis is confusing the frequency domain measurements (MLS or swept sine) used to derive the impulse response with the time domain result.
Thanks John, the frequency vs time domain explanation rings a bell now. I'm kind of ashamed to admit it, but I should have remember that from my EE classes, but that was over 30 years ago.
 

·
Banned
Joined
·
4,838 Posts
The way I think about it, right or wrong, is that, with gating, REW is "looking at and analyzing the influence of" only the first "x" milliseconds of the signal immediately following any given frequency, ignoring the room's influence in that way. At low frequencies, there is not enough of a cycle of information within the x window time to allow accurate ayalysis. At high frequencies, there might be several cycles of information within the x window time, and analysis can be very accurate.
 

·
Premium Member
Joined
·
74 Posts
Discussion Starter #13
The way I think about it, right or wrong, is that, with gating, REW is "looking at and analyzing the influence of" only the first "x" milliseconds of the signal immediately following any given frequency, ignoring the room's influence in that way. At low frequencies, there is not enough of a cycle of information within the x window time to allow accurate ayalysis. At high frequencies, there might be several cycles of information within the x window time, and analysis can be very accurate.
So in other words, the signal being picked up by the mic is passed thru a narrow swept frequency filter that is 'x' msec behind the swept frequency of the stimulus, correct? Since the value of 'x' msec must be dependent on the distance from speaker to mic, how is this value determined? Is this 'x' msec the same or different than the window used to gate room reflections?
 

·
Banned
Joined
·
4,838 Posts
John to the rescue. Thank you, John, as always.

Although I do not know the math, and John obviiously does, my simplistic way of thinking of the analysis was not intended to imply that the actual analysis is simple or anythinig like it. If you really want to understand the analysis in detail, start with a masters degree in engineering. Otherwise, you are just going to have to trust guys like John.
 

·
Premium Member
Joined
·
74 Posts
Discussion Starter #16
Thanks for the response guys. I have one more question concerning impulse windowing. Since the lower end of the derived frequency response is limited by the duration of the window, would it be possible to edit the impulse data file and zero out all data points past the window to allow lower frequency measurements?
 

·
Premium Member
Joined
·
6,711 Posts
No, zero padding doesn't add any information, the low frequency content just isn't there - that goes back to AudiocRaver's comment, if the frequency you are interested in doesn't fit within the window you are out of luck. The typical way to deal with that is to make a very near field measurement for the low frequency information (within an inch or so of the LF driver) and splice that to the windowed measurement with the mid and high frequency content (using the merge function in the All SPL graph's trace arithmetic controls).
 
1 - 17 of 17 Posts
Top