Measurement Length Changes Impulse Response? And a Sub Integration Question

[Ultrasonic]

Hi all .

I'm currently working on integrating a subwoofer into my traditional stereo audio system, and have just been trying to use measured impulse responses to get the time alignment and phase right. The help files states that increasing the measurement length (256k, 512k etc) would decrease measurement noise so I tried experimenting with this to see if it would give me more accurate results. What I found confused me though, since increasing the measurement length changed the shape of the impulse response as shown in the image below. The only change I made was the measurement length and an acoustic timing reference is being used. What is going on? And which is most 'right'?

(I'm running REW version 5.19 on a laptop with 8 GB of RAM and an Intel Core i5-7200U processor. The help file mentions a lack of RAM or processing power being a potential issue but I'd have thought I should be OK? I'm making the measurements using a UMIK-1 and a miniDSP 2x4 HD as the sound source.)

As well as the timing I'm also interested in the polarity of the impulse so that I can adjust my sub to match. Based on the 256k measurement I took first I'd thought it was a positive going impulse. Inverting the output to my sub, and adding in a delay, gave me the result below (main speaker in red an subwoofer in blue). Which I'm hoping looks sensible? For the longer measurement lengths above the biggest amplitude on the IR is negative rather than positive though?

Thanks for any help .

 

[John Mulcahy]

Can you post the mdat file for your measurements?

 

[Ultrasonic]

Hi John,

Thanks for the quick reply. File attached. Let's see if you can spot something dumb that I'm doing .

Thanks very much for your efforts in producing and sharing REW BTW .

 

[John Mulcahy]

Odd. Here are 4 taken on my desk just now:

Probably worth trying with the latest V5.20 beta version.

 

[Ultrasonic]

Thanks for testing this out yourself.

I've tried again with the new software version but sadly the result is the same (data attached). I selected the option to commit 2 GB of RAM during the installation process. The new version also prompted me to use a louder timing reference, which I did.

 

[John Mulcahy]

Interesting. Could try a higher start frequency for the sweeps, 20 Hz or above, in case a DC servo somewhere along the path is having an effect. Check that "Listen to this device" is not selected in the Audio properties for the UMIK, just on the off chance.

Quite an impressive 36 Hz peak you have there

 

[Ultrasonic]

Interesting. Could try a higher start frequency for the sweeps, 20 Hz or above, in case a DC servo somewhere along the path is having an effect.

I tried this earlier actually as I'd realised the speaker cone was rather flapping about down at 10 Hz (unsurprisingly in hindsight) but same result unfortunately. I've attached an example I've just measured again.

Check that "Listen to this device" is not selected in the Audio properties for the UMIK, just on the off chance.

Checked and it was not selected.

Quite an impressive 36 Hz peak you have there

Indeed . My aim is to use my miniDSP to act as a crossover between mains and sub somewhere around 80-100 Hz, and also to EQ the sub response to give a rather flatter response below this . The EQ is straightforward but getting the crossover/matching right is what I'm currently working on.

 

[John Mulcahy]

Everything turned off in the miniDSP? Maybe try connecting directly from a headphone out on the laptop to an analog in on the system.

 

[Ultrasonic]

I'll try another DAC in the morning (I'm in the UK) just to rule out any issue with the miniDSP. I believe all filtering etc was off.

I've progressed to what I think are some rather more sensible looking response curves BTW (attached).

 

[John Mulcahy]

I think I've worked out what is going on. The clocks for the input data from the UMIK and the output data via the 2x4HD are running at slightly different rates, so the longer the sweep lasts the bigger the overall drift. I was able to generate similar results, including the flip from positive to negative peak for the longer sweeps, by resampling captured data by a factor of 1.0001 (just picked a number to test the theory). Best stick with the 256k sweeps.

That does look rather improved!

 

[Ultrasonic]

Ah, great, glad you've found an explanation, and one that does mean I can at least still take useful measurements . Is there any logic to a 128k sweep being better than 256k under these circumstances?

(I did actually try my other DAC before spotting your reply, but same variation.)

 

[John Mulcahy]

May as well stick with 256k, though if you are making a lot of measurements 128k saves some time. Can also use Pink Periodic Noise and the RTA (Rectangular window, no averaging) for live response updates while you make adjustments.

 

[Ultrasonic]

May as well stick with 256k, though if you are making a lot of measurements 128k saves some time. Can also use Pink Periodic Noise and the RTA (Rectangular window, no averaging) for live response updates while you make adjustments.

I'm not worried about saving time, I was just wondering if the timing error issue meant that 128k might be more accurate.

May I ask you one other question? Having read the Minimum Phase help page I had a look at the excess group delay for the corrected data I posted last night. In line with the example in the help file, this suggests a time delay in the region of 10-15 ms over the region covered by the sub (see below for right channel example), and yet the sub vs main speaker impulse response comparison I posted at the top of this thread suggests they are reasonably well time aligned. Any thoughts/comments? Just trying to check I'm not doing something wildly wrong! Thanks.

For info. the delay I'm using is close to what would be expected based on how much further the sub is away from the MLP than the main speakers. DSP use will add to this of course.

 

[John Mulcahy]

It is very difficult to align a sub using the impulse response. From the group delay and the wavelet spectrogram I'd say the sub is quite delayed relative to the mains, though it is hard to get an accurate picture given the various room reflections. jtalden uses a phase tracking approach to align subs and mains, see his posts in this thread and this one for examples.

 

[Ultrasonic]

It is very difficult to align a sub using the impulse response. From the group delay and the wavelet spectrogram I'd say the sub is quite delayed relative to the mains, though it is hard to get an accurate picture given the various room reflections. jtalden uses a phase tracking approach to align subs and mains, see his posts in this thread and this one for examples.

Thanks for the links . I'll give them a read.

 

[jtalden]

Hi Ultrasonic,
Regarding SW timing, I looked at the Post-3 mdat file and noted that the SW and main was measured full range rather than with the chosen XO active. I suggest that the proper delay timing is impacted by the XO chosen and thus it is best to measure the SW, L, and R all with 20-10k Hz sweeps with the XO active. It's

good to leave the PEQs that were later applied to smooth the response as active also. A smoother SPL response helps to smooth the phase response as well. I can answer any questions and help with more direction as needed.​

 

[Ultrasonic]

Hi Ultrasonic,
Regarding SW timing, I looked at the Post-3 mdat file and noted that the SW and main was measured full range rather than with the chosen XO active. I suggest that the proper delay timing is impacted by the XO chosen and thus it is best to measure the SW, L, and R all with 20-10k Hz sweeps with the XO active. It's

good to leave the PEQs that were later applied to smooth the response as active also. A smoother SPL response helps to smooth the phase response as well. I can answer any questions and help with more direction as needed.​

Thanks for taking the time to comment. I'd spotted from your other posts that you'd normally recommend measurements with the crossovers applied and was going to take some measurements like this later. My only slight concern is if doing this and then focusing on alignment over the crossover region could lead to a significant timing disparity either side of the crossover?

This may come down to the choice of crossover actually, for which I have a choice within the miniDSP software. Above I went with 48 dB/octave Linkwitz-Riley crossovers (low-pass on the sub and high-pass on the mains). This sounded sensible from the little I've read but I wonder if you'd suggest anything else? The options are various Butterworth and Linkwitz-Riley cross-overs, and a Bessel crossover.

I'm still playing about with what EQ to apply, in terms of what balance of flat amplitude response vs how much the associated filters introduce their own ringing. Ultimately I know I'm going to have to listen to a few options to choose between them, so all of this is clearly going to be an iterative process.

I'd underestimated the complexity of the time alignment though, in part because the miniDSP website suggests it's simple to do based on impulse responses. One thing I did try actually was to band-limit the output of my main speaker to roughly match that of the sub, and to then compare their impulse responses (red vs blue in figure below). This made me think I had things about right by delaying the mains by 3.6 ms...

 

[Ultrasonic]

OK, I've had a go at trying the phase tracking method (data attached), with LR 48 dB/octave crossovers at 100 Hz applied.

If I've understood correctly I think a 4.6 ms delay on the main speaker and inverting the subwoofer signal looks like the best fit? Zero delay is actually pretty close to the non-inverted sub signal but that isn't sensible given the physical separation, so ignoring this then 9.5 ms matches this best, but I think not over quite as wide a range as the inverted/4.6 ms combination?

 

[jtalden]

My preferred target for delay timing is to provide; close phase tracking of the direct sound throughout the XO range. For an 80Hz LR-24 that would be at least a 40-160Hz range. When that is done we are assured that there is a smooth handoff between the 2 drivers. The direct sound central lobe will thus remain pointed straight at the LP throughout the entire operating range. This approach does not consider the impact of the room that often creates disparate phase issues between the drivers in the XO range. As such, the SPL results may not be better than other timing alignments, although it often is marginally better.

Background:
As you noted there are several considerations in choosing the XO settings. A review of the data is needed to provide any comments on what may be advantageous in a particular situation.
Considerations/approaches may be:
> Raise or lower the XO frequency to try to avoid a room issues.
> Raise or lower the XO frequency to better control the shape of the rolloff slopes.
> Raise or lower the XO frequency to better distribute power to the drivers more suited to balance the SPL capacity of the drivers.
> Change the XO point and filter slopes as needed on each driver to best match the acoustic SPL target response of each driver.
> Adjust positions of the drivers or listening position to improve the relative room influences on the XO.
> Change room treatments to address a problem.

In general, targeting an acoustic LR-24 or LR-48 is a common and well defended approach for a hobbyist. To that end, the 2 XO filters may be chosen independently. This just comes down to the time we want to spend experimenting on these variables. With careful setup, the measurements can be improved. It is more difficult to say if this will have a practical impact beyond that. Given reasonable delay timing, a smooth SPL through the bass range range is the primary consideration. We may be able to hear differences depending on how we get there, but it would probably be difficult for us to say which setup is better.

If your not a DIY audio hobbyist, it's common to choose the recommended 80 Hz XO setting with LR-24 filters and not worry about all these possibilities. It is a good mainstream approach that is likely to sound just as good as a tweaked design.

It's also not necessary to set delay timing based on direct sound phase tracking for acceptable results. Just aligning the initial rise of the 2 impulses puts us close to the correct timing. This is similar to measuring the distances to the drivers and setting the delay accordingly. Beyond that, we could check for a deep null at the XO when one of the 2 drivers polarity is reversed. From there, the timing could be adjusted a little as needed to move that null nearer to the XO frequency. [Here is where the room influences often becomes a problem.]

The phase alignment method I prefer is more complicated to do, but is the only one that provides evidence that the direct sound arrives correctly timed at all frequencies.

 

[Ultrasonic]

We may have cross-posted above?

Quick reply as I need to get to bed...

 

[Ultrasonic]

Somehow posted above before I'd finished typing!

My current plan is to apply PEQ to the sub but not my main speakers and as such my reasoning for the crossover being at 100 Hz rather than 80 Hz was to try to give me a flatter bass response to a higher frequency.

Does the lobing argument apply to the separate subwoofer and main speakers that I have? I'd naively thought this was more an issue for multiple drivers on a single loudspeaker?

I'll have another read through your post tomorrow...

Thanks for your help .

 

[Ultrasonic]

Oh, and as for whether I'm a 'DIY audio hobbiest' I guess it depends on your definition. I have no intention of building my own speakers as my cabinet finishing skills would not be up to it! I'm a physicist by training with an interest in acoustics, and as such I'm keen to do what I can to try to optimise the setup the the equipment I currently have. I come from a stereo audiophile background where most would rather look down their noses at the use of a subwoofer but am interested in exploring their potential for music. I do use my system for TV and movies too though.

My current sub I bought pretty cheaply second hand but if it works out I have it in mind to buy something potentially rather better, like a PSA S1510.

 

[jtalden]

Your method worked out okay.
You chose 2 + 7 as a good choice based on this data.
I chose 1 + 9 as better for aligning the direct sound.
Both are good alignment.
2 + 7 delays SW vs the main an additional 4ms vs my choice. This is not a particular problem in this low frequency range.

The SPL support is similar (very good) for both these settings and you may prefer the one to the other for sound quality.

The phase tracks well for both. Your choice tracks slightly better given the impact of the room.
My choice is slightly better for the direct sound tracking. Applying a frequency dependent window of 8 (and moving down to 4 cycles) we can move to lessen the impact of the room on the phase and better see the phase tracking of the direct sound.

The L main may be different from this R main in the way it interact with the room so I suggest it be measured also.

 

[Ultrasonic]

Thanks for looking at my data @jtalden .

As you probably realised I hadn't explored using frequency dependent windows above. My initial aim was very much to prioritise time alignment of the direct signal though, which was why I was trying to use the impulse responses to get the initial arrival times for the sub and mains to be the same.

I've just had a look at using frequency dependent windows to try to see what you are but I'm not quite there. When I first go to the IR Window tab and click through the different data files it picks out a different 'Window Ref Time' for each one based on the location of the impulse response peak, with the window left and right lengths being either side of this. Is this what I want, or do I perhaps want to apply windows all with a fixed relationship to the timing reference signal used for the measurements? My hunch is the latter...

From the help file I can't tell quite what the 'Apply Windows To All' and 'Apply Windows to All, Keep Ref Time' buttons will actually do in this regard (@John Mulcahy ?)

Can I just check that you realised all of the delays that I applied were to the main speaker, not the subwoofer?

Oh, and yes my speaker layout is unfortunately rather asymmetric and the amplitude spectrum is rather different for each, and so I expect the room dominated phases will be too. That's another reason to prioritise the direct signal I guess. Not sure if I'll get chance to do any more measurements before the weekend right now but I'll see.

I will just say that I've been listening to some modern acoustic jazz (Go Go Penguin's Humdrum Star album) whilst looking at this, and with my 3.6 ms delay things like drum impacts are at least not sounding obviously 'wrong'. The music is sounding rather good in fact .

 

[John Mulcahy]

From the help file I can't tell quite what the 'Apply Windows To All' and 'Apply Windows to All, Keep Ref Time' buttons will actually do in this regard?

The first applies the windows and reference time of the selected measurement to all the other measurements, the second leaves the reference times for the other measurements as they are. Don't change the reference times.