Help understanding group delay

[Chris A]

Here goes.

1) I'd first think about moving the measurement microphone to be 1m in front of the front baffle of your loudspeakers, then place a good amount of temporary absorptive material over the floor between the microphone and the loudspeaker under measurement, as well as temporarily covering anything that's acoustically reflective between the loudspeakers and on any nearby sidewalls within 4-5 feet of the loudspeakers. I'd center the microphone on the tweeter's axis. Group delay and excess group delay:

2) I also recommend adding both absorption and diffusion to your room. You have a local rise in RT values above 1700 Hz and your average RT values are in the 350-450 ms range, which is a little high for a smaller listening room. You have acoustic reflectors at 8.2', 19', and 21.4' (path length) from the right loudspeaker that significantly affect the phase and group delay measurements. In fact, below 2kHz, the group delay curve is noisy (no areas where the GD curve is flat and smooth, even using psychoacoustic smoothing). Try the closer microphone measurement position on-axis with absorption on the floor and covering anything in the nearfield of the loudspeakers (especially between the loudspeakers). I don't know your room dimensions, but it looks like it could be in the 2K-3K cubic foot category:

3) It looks like your subs may not be very well time aligned to your left/right loudspeakers, but the measurements are noisy enough to mask exactly what the sub lag time(s) might be. You might think about pairing your left or right loudspeaker with each sub (1 and then 2), taking measurements with just one loudspeaker and one sub (the sub closest to the loudspeaker). The spectrogram view is fairly noisy and doesn't attenuate very quickly, even after 100 ms...

Chris

 

[Arindal]

But you have to ask yourself whether you are improving your listening experience, or just the measurement?

Fortunately, audible group delay issues usually translate to a very limited number of apparent sound quality flaws, like ´slow/delayed deep bass transients´ or ´hollow, kick-heavy beats´ with the fundamental impulse coming late, localization stability issues or (rare) perceived proximity issues. If you have tight, bone-dry lower bass with synthetic (EDM) recordings, stable localization and phantom sources not appearing too close or too distant to the listener, I would stop worrying about GD, no matter what measurements show.

Interestingly, I found the popular step response graph to be the least reliable source for judging group delay.

My take on it is: yes, I know the thresholds are unknown.

If I recall it correctly, there were some thresholds formulated by Zwicker/Zollner/Fastl in one of their standard books, in the region of 1 to 2ms per octave for midrange and higher frequencies. For bass, the thresholds were said to be higher, like 5 to 10ms.

We should note, though, that these scientific thresholds originate from experiments with no interaural differences, as there were executed using headphones. In the real world of loudspeakers and rooms, group delay issues can, depending on the listening position and slightly varying listening distance, easily result in very tiny interaural differences, for which our brain is way way more sensitive (one of main mechanisms of localization).

 

[Mort]

Does DSP processing on a subwoofer increase group delay?

 

[Chris A]

Interestingly, I found the popular step response graph to be the least reliable source for judging group delay.

I agree. I found that the step response plot is basically a qualitative view of loudspeaker/room acoustic performance since there isn't really anything directly usable from the plot itself to make direct adjustments to DSP settings. Step response is what you use to show that "everything is working right". It's also a one-look view of any issues you've got with:

• crossover all-pass phase growth (due mostly to using IIR-type higher order crossover filters),
• individual acoustic driver minimum phase growth (typically more severe for direct radiating drivers whose acoustic centers move around a lot more vs. frequency), and
• room reflections--if the acoustic measurement is done in-room, showing the effects of primary (slap) echos and insufficient levels of in-room absorption.

You then have to look at other plot types (excess group delay, spectrogram view, phase plot, impulse response, etc.) to know what to adjust.

Fortunately, audible group delay issues usually translate to a very limited number of apparent sound quality flaws, like ´slow/delayed deep bass transients´ or ´hollow, kick-heavy beats´ with the fundamental impulse coming late, localization stability issues or (rare) perceived proximity issues.

This is where I diverge a bit...

From my own observations and reading into Bohdan Raczynski's March 2013 paper entitled, "Some Attributes of Linear-Phase Loudspeakers" (from Siegfried Linkwitz's site), some effects that I've personally noticed after reducing all-pass phase growth in-room and significantly smoothing the group delay response, along with controlling early reflections within 4-5 feet around the loudspeakers (across the audible band):

1. Tighter and more prevalent perception of bass
2. Elimination of perceived harshness
3. Wider and deeper subjective sound stage
4. Increased subjective perception of realism or musical presence
5. Increased subjective soundstage resolution--being able to hear musical details that were not heard before.

So note that I'm not just talking about reducing subwoofer group delay distortion (as most have been talking about in this thread) but also the reduction of at least three types of full-range distortion.

Chris

 

[john61ct]

I thought UMIK-2 is required for measuring phase / timing issues?

Does that not include group delay?

How else can I check that my (non-DSP) active crossovers at 24dB slopes are OK?

If I need to use lower slopes, aren't they more susceptible ?

 

[levimax]

The audibility of group delay is debatable and as mentioned there is not a scientific answer yet. Where it can cause audible issues is with sub integration, either between subs and mains or between multiple different subs. If the group delay profile of the speakers is different in the frequncies where they overlap you will get cancellations even if time is aligned correctly at the crossover point. If possible you want to cross over where both speakers have a flat group delay curve.

 

[Chris A]

I thought UMIK-2 is required for measuring phase / timing issues?

I don't own one of those. Presently, I don't see the use for one.

Does that not include group delay?

Group delay data is present with all REW measurements made with swept sine upsweeps.

How else can I check that my (non-DSP) active crossovers at 24dB slopes are OK?

Take a single loudspeaker measurement at 1m, centered on the tweeter using plenty of temporary absorption on the floor and on nearby acoustic reflectors. Then you'll see the all-pass phase growth and group delay response.

If I need to use lower slopes, aren't they more susceptible ?

More susceptible to what? Passive lower order IIR-type crossover filters are arguably more stable in transfer function response with temperature transients than higher order crossover filters.

Does DSP processing on a subwoofer increase group delay?

Only if you don't get the (typically higher frequency) channel time delays and subwoofer placement(s) in-room correctly set.

That's what REW measurements enable: they verify what you're actually achieving acoustically in-room.

Chris

 

[Flaesh]

Does DSP processing on a subwoofer increase group delay?

DSP processing should reduce group delay. But this isn't always the case, I gave an example yesterday.

 

[john61ct]

I don't own one of those. Presently, I don't see the use for one.

So what mic do you use for measuring timing issues?

My understanding is UMIK-1 cannot ?

 

[john61ct]

More susceptible to what? Passive lower order IIR-type crossover filters are arguably more stable in transfer function response with temperature transients than higher order crossover filters.

I am not planning on getting into passive design, and DSP only if I need to.

So, within the active electronics and bass management context, is it not the case that 24dB LR provides better (chance of) phase alignment than "gentler" slopes?

Separate question - if a DIY sub has group delay issues, can simple placement changes - just moving it closer to the LP or the main pair further away - solve them?

 

[Keith_W]

I thought UMIK-2 is required for measuring phase / timing issues?

Does that not include group delay?

How else can I check that my (non-DSP) active crossovers at 24dB slopes are OK?

If I need to use lower slopes, aren't they more susceptible ?

All right, this discussion is starting to get a bit advanced with @Chris A 's recent observations (which I happen to agree with, BTW). It is starting to stray into linear-phase vs. minphase DSP and whether there is any advantage in linearising the phase / group delay of the speaker.

I will say this: it is NOT my intention to confuse beginners by talking about cutting edge DSP or the very latest in psychoacoustic research which hasn't made it to more mainstream publications yet. There are threads on ASR where you can get that fix, but a beginner who is struggling to understand GD is not the place for it, IMO. This is why I present a very "conventional" view whilst privately holding different beliefs. Having said that, I think Chris' contributions are valuable because you need a hint that there is more to it than what I said in my previous post. Something you can delve further into once you understand the basics.

Take a look at the eBook in my signature. In it, I make a very strong recommendation, which is this: DON'T look at the phase graph, or group delay. Any problems with phase will manifest in the amplitude response. So for example, two speakers may individually measure flat, but if you play them together, you see a dip ... then there is something funny going on. NOW you can look at phase / group delay / polarity / etc. to obtain the diagnosis. If the frequency response looks perfect, then there is no need to examine phase/group delay at all. Think of these graphs as a diagnostic tool to help you understand what is going on in the frequency response.

Yes, it is possible to straighten out these curves with DSP. And NO you should not attempt to do it unless you really understand what you are doing, and where to look for side effects. One of my cardinal rules of DSP is this: do not introduce an audible side effect by trying to fix something which is not audible or minimally audible.

To answer your question: a 24dB slope is a fourth order slope. Each order adds 90 deg of phase rotation, so fourth order rotates the phase 360deg at the XO point. You can check if they are OK by summing the LPF and HPF. Don't forget to convolve the driver measurement into the HPF/LPF (or measure the speaker with the XO's in place) before summation. If it sums to a flat line, it's all good. Lower order slopes, odd order slopes, and asymmetric slopes greatly increase the risk of improper summation. Vance Dickason wrote an entire chapter in his book about crossover summation. I can't summarize that here, so all I will tell you to do is sum HPF/LPF and look at the result.

 

[Keith_W]

Separate question - if a DIY sub has group delay issues, can simple placement changes - just moving it closer to the LP or the main pair further away - solve them?

You posted while I was typing my previous reply. The short answer is "no". Subwoofer latency consists of:

- any electrical delays applied to the subwoofer, e.g. with DSP, passive XO's, etc.
- the inherent group delay of the subwoofer (can be up to 100ms or even more depending on what frequency we are looking at)
- the "time of flight" delay, which depends on how far the sub is located from the listener.

Think about how far sound travels in 100ms. The speed of sound is 343m/s, so in 100ms, sound travels 34.3m. Even if you are sitting on the subwoofer you can not compensate for the group delay of the sub by physical positioning. The only way is with DSP.

And the bigger question, which I have emphasized over and over, is whether you even need to. The psychoacoustic detection threshold is very long! This is why I told you to stop worrying about it!

 

[john61ct]

Excellent pointers to further my learning, but much of that goes whoosh over my head.

At this stage, as stated I am abstaining from any DSP, so could you maybe rephrase these in that context?

> summing the LPF and HPF

> convolve the driver measurement

I do of course plan to have the xovers working when measuring with REW, if that's all it means

To answer your question: a 24dB slope is a fourth order slope. Each order adds 90 deg of phase rotation, so fourth order rotates the phase 360deg at the XO point. You can check if they are OK by summing the LPF and HPF. Don't forget to convolve the driver measurement into the HPF/LPF (or measure the speaker with the XO's in place) before summation.

 

[john61ct]

I am not a worrier, even about life and death issues my F&F think I **should** "worry about"

Even if I do nothing to "fix" these issues while I am using REW, I do want to measure phase / timing issues anyway if nothing else to investigate cause & effect in speaker design and using different active crossovers, and to further my learning.

 

[john61ct]

I googled the verb "convolve" and infer that is only relevant to using DSP signal processing.

 

[Mort]

After a good nights sleep I realize that I'm not as free as I thought. Per amir's comment about not worrying about GD - if it's not worth worrying about it then why does it get so much attention?

Getting any sleep now?

 

[Keith_W]

At this stage, as stated I am abstaining from any DSP, so could you maybe rephrase these in that context?

> summing the LPF and HPF

LPF = Low-pass filter, HPF = High-pass filter.

When you play (say) a woofer and a midrange together, the result is a summation of the output of each driver. In this case, it acoustically sums "in the air". However, you could do the same thing with a program like REW. Measure the woofer and the midrange, then sum them using REW's "Trace Arithmetic A+B" function. The simulation is so exact that it will look exactly the same as an acoustic measurement of both drivers taken from the same position.

> convolve the driver measurement

If you place a crossover (XO) in series with a driver, the result is a multiplication of the XO phase/slope together with the driver response. You can simulate this in REW with "Trace Arithmetic A*B". But strictly speaking, it's not multiplication - it is convolution. I don't wish to confuse you by citing the convolution theorem here, but you can think of it as fancy multiplication.

If you want to simulate the effect of a Linkwitz-Riley 4th order (LR4) on a woofer and midrange, these are the steps:

1. Measure the woofer and midrange mounted on the baffle with no crossover under quasi-anechoic conditions. This isn't easy. See the link in my signature and read that book for some pointers how.
2. Create a LR4 HPF and LR4 LPF with the appropriate corner frequency in REW.
3. Convolve the LPF with the woofer, and HPF with the midrange with "A*B"
4. Sum the LPF-woofer and HPF-midrange result with "A+B". Examine the summation for any issues.

 

[klettermann]

<<<<<<<<<<<<<SNIP>>>>>>>>>>>>>

3) It looks like your subs may not be very well time aligned to your left/right loudspeakers, but the measurements are noisy enough to mask exactly what the sub lag time(s) might be. You might think about pairing your left or right loudspeaker with each sub (1 and then 2), taking measurements with just one loudspeaker and one sub (the sub closest to the loudspeaker). The spectrogram view is fairly noisy and doesn't attenuate very quickly, even after 100 ms...

View attachment 510507

Chris

This thread will take me months, if no years, to grasp. In the menatime - more questions. I actually put fair amount of effort into time alignment, though god only knows if I did it right. Any, my spectrogram of the whole system looks like the below. Not perfectly perfect but not so terrible either from what I can gather. How do I gert from that to the version you posted? We're obvioously using different settings. Mine was just the REW default. Can you clarify? Cheers,

 

[john61ct]

Thanks for that, I will for now just be measuring to optimize the stuff that I physically have, which includes the crossovers.

I guess as I learn more I may better understand the benefit of simulating things I don't.

LPF = Low-pass filter, HPF = High-pass filter.

When you play (say) a woofer and a midrange together, the result is a summation of the output of each driver. In this case, it acoustically sums "in the air". However, you could do the same thing with a program like REW. Measure the woofer and the midrange, then sum them using REW's "Trace Arithmetic A+B" function. The simulation is so exact that it will look exactly the same as an acoustic measurement of both drivers taken from the same position.



If you place a crossover (XO) in series with a driver, the result is a multiplication of the XO phase/slope together with the driver response. You can simulate this in REW with "Trace Arithmetic A*B". But strictly speaking, it's not multiplication - it is convolution. I don't wish to confuse you by citing the convolution theorem here, but you can think of it as fancy multiplication.

If you want to simulate the effect of a Linkwitz-Riley 4th order (LR4) on a woofer and midrange, these are the steps:

1. Measure the woofer and midrange mounted on the baffle with no crossover under quasi-anechoic conditions. This isn't easy. See the link in my signature and read that book for some pointers how.
2. Create a LR4 HPF and LR4 LPF with the appropriate corner frequency in REW.
3. Convolve the LPF with the woofer, and HPF with the midrange with "A*B"
4. Sum the LPF-woofer and HPF-midrange result with "A+B". Examine the summation for any issues.

 

[klettermann]

Getting any sleep now?

Consider the unfortunate insomniac, agnostic, dyslexic man. He stays up all night wondering if there really is a Dog. And so it goes with this.....