Does DSP latency matter?

[Steve81]

You are close in your speed calls (actual speed at STP is 1087.42 fps). Your calls for delay are fine. And yes even 1millisecond (hence a foot) can be noticed and can make a noticeable difference. That is why in large speaker system boxes the highs can be delayed just 1.345 milliseconds. It makes a difference in the wave front! And you can see the arrivals on a graph and they are frequency dependent.

Gordon

Hi Gordon,

No question that 1ms can make a difference in the highs, but we're not talking about the highs.

 

[Gordonj]

Hi Gordon,

No question that 1ms can make a difference in the highs, but we're not talking about the highs.

Indeed! However, That same 1 millisecond can make a difference between a muddy LF and a clear and punch LF, and it is very noticeable!

Actually it is amazing when a sub system comes to life by being dialed in correctly. And truly 1 millisecond can make a huge difference! I do it and experience it all the time!

Gordon

 

[Steve81]

And can you see a problem with those calcs? Gordon

Please clarify. Mr. Ricci's measurements are pretty straightforward. That PB12-NSD is just one example. Adding ~1ms (for 1 foot) or ~8ms (for 8 feet) may make a difference, but compared to the group delay already present in these systems, it doesn't amount to a whole heck of a lot. Where it can present a problem as I see it is if that extra delay causes the total GD+DSP delay to exceed the one to one and a half cycle threshold for audibility. Clearly at 1kHz or 20kHz, we're not talking about a lot of time. But at 80Hz, you've got a few milliseconds to work with (12.5 to 18.75). Clearly an extra 8ms can be a problem if you've got a sub with high group delay. But if your sub has reasonably low group delay, I don't see the problem.

 

[Gordonj]

Please clarify. Mr. Ricci's measurements are pretty straightforward. That PB12-NSD is just one example. Adding ~1ms (for 1 foot) or ~8ms (for 8 feet) may make a difference, but compared to the group delay already present in these systems, it doesn't amount to a whole heck of a lot. Where it can present a problem as I see it is if that extra delay causes the total GD+DSP delay to exceed the one to one and a half cycle threshold for audibility. Clearly at 1kHz or 20kHz, we're not talking about a lot of time. But at 80Hz, you've got a few milliseconds to work with (12.5 to 18.75). Clearly an extra 8ms can be a problem if you've got a sub with high group delay. But if your sub has reasonably low group delay, I don't see the problem.

Hey Steve please define GD or Group Delay. Do you mean Latency?

Gordon

 

[Steve81]

Indeed! However, That same 1 millisecond can make a difference between a muddy LF and a clear and punch LF, and it is very noticeable!

I just have trouble seeing it I guess.

Here's one example of a fine subwoofer, the Velodyne DD18+.
http://www.data-bass.com/images/measurements/67/C dd18+ servo gain 1 gd.jpg

As you can see, even its group delay is well beyond 1ms, hitting 20ms at 20Hz. This is exceedingly low by most standards, but certainly not 1ms. You can also note that little spike around 60Hz which is also about 15ms.

 

[Steve81]

Hey Steve please define GD or Group Delay. Do you mean Latency?

Gordon

The best way to visualize it is to look at the spectrogram compared with the group delay graph.

http://www.data-bass.com/images/measurements/71/C pb12nsd group delay.jpg

http://www.data-bass.com/images/measurements/71/D pb12nsd spectrogram.jpg

See how at 80Hz, the PB12 is hitting the peak right around 0ms? Now look at 20Hz; the PB12 isn't hitting peak until about 100ms. You can also see that it doesn't even start firing at 20Hz until about 100ms after an 80Hz wave would begin.

In short, yes, I believe you can basically think of it as frequency dependent latency. It is also heavily related to frequency response. The reason the PB12's delay is so high is because of the shaped FR and steep filter to cut out ultra low frequencies (<20Hz). Porting also obviously has an impact.


Edit
Also to add courtesy of Ilkka.

Subwoofer Tests Explained - Home Theater Forum and Systems - HomeTheaterShack.com

Group delay

Mathematically group delay is the negative derivative of the phase response. Phase response was measured using the ETF-5 program which calculates it from the measured impulse response. Since frequency response and phase response correlate with each other, also FR and GD correlate with each other. Meaning flat or gently sloping FR results in low GD. That's why the final in-room frequency response should be as flat as possible. There are no studies defining the audibility of GD at low frequencies, but the threshold is suggested to be in the range of 1-1.5 cycles. Therefore I have included the 1 cycle curve on the graphs.

Ideally GD should be as low as possible, or at least gradually/gently rising.

 

[Gordonj]

I just have trouble seeing it I guess.

Here's one example of a fine subwoofer, the Velodyne DD18+.
http://www.data-bass.com/images/measurements/67/C dd18+ servo gain 1 gd.jpg

As you can see, even its group delay is well beyond 1ms, hitting 20ms at 20Hz. This is exceedingly low by most standards, but certainly not 1ms. You can also note that little spike around 60Hz which is also about 15ms.

I nderstand your use of GD, there are two defintions in play here. GD of the components that you are using in the DSP and GD of the speaker componts.

What I am referring to is time delay of the audio spinal to the sub system. That delay would be in addition to any GD that would be present.

Gordon

 

[Steve81]

I nderstand your use of GD, there are two defintions in play here. GD of the components that you are using in the DSP and GD of the speaker componts.

What I am referring to is time delay of the audio spinal to the sub system. That delay would be in addition to any GD that would be present.

Gordon

I understand, but what I'm getting at is that if the two combined don't exceed the audible threshold of 1 to 1.5 cycles, I'm not sure that there will be a problem. Its possible an extra 1ms or 8ms across the band could do it, but it really just depends on the sub and where you're planning to cross it. Take the SVS PB13U for example.

http://www.data-bass.com/images/measurements/55/C pb13 15hz group delay.jpg

Adding 1ms from 80Hz to 20Hz really isn't going to make a difference. Even 8ms probably wouldn't make a difference. Adding 8ms up toward 100Hz, well that's a different story.

Edit
I guess its also probably worth noting that if you've got relatively bad GD ala the PB12-NSD, adding another 1-8ms on top of a 110ms delay probably isn't the end of the world either.

 

[TLS Guy]

The problem is we only have accurate audibility thresholds for group delay above 500 Hz.

Frequency Threshold
500Hz 3.2 ms
1kHz 2 ms
2kHz 1 ms
4kHz 1.5 ms
8kHz 2 ms

After Blauert and Laws.

Ported enclosures, especially band pass ones, add huge group delay. This is almost certainly audible and probably accounts for the highly negative reactions to band pass enclosures, especially the higher order ones.

Unfortunately group delays sum in the last octave, because of the cumulative effects of DC blocking caps in circuits in microphones all the way to the final power amps. The ported enclosures with their twin resonant peaks separated in time add massive insult to injury in the last octave.

There is reasonably good circumstantial evidence that group delays greater than 50 msec are deleterious in the last octave.

This area needs further research and attention.

 

[Gordonj]

The problem is we only have accurate audibility thresholds for group delay above 500 Hz.

Frequency Threshold
500Hz 3.2 ms
1kHz 2 ms
2kHz 1 ms
4kHz 1.5 ms
8kHz 2 ms

After Blauert and Laws.

Ported enclosures, especially band pass ones, add huge group delay. This is almost certainly audible and probably accounts for the highly negative reactions to band pass enclosures, especially the higher order ones.

Unfortunately group delays sum in the last octave, because of the cumulative effects of DC blocking caps in circuits in microphones all the way to the final power amps. The ported enclosures with their twin resonant peaks separated in time add massive insult to injury in the last octave.

There is reasonably good circumstantial evidence that group delays greater than 50 msec are deleterious in the last octave.

This area needs further research and attention.

Thank you! I was going to the office to pull that paper from the ASA archives and you beat me to it - DANG IT!

To further express what I have been saying, in tuning a sound system's LF a minor adjustment of 1ms can put the entire LF in time with or behind or in front of the mains. Obviously 1ms by itself does not effect the LF the same (from 0ms to 1ms) however going from 10ms to 11ms (or a similar change) can be night and day. Therefore, one 1ms can be heard and noticed, in that circumstance.

Gordon

 

[Steve81]

There is reasonably good circumstantial evidence that group delays greater than 50 msec are deleterious in the last octave.

This area needs further research and attention.

Probably worth noting that 50ms at 20Hz is exactly 1 cycle, which is the threshold (1-1.5 cycles) cited by both Ilkka and Ricci. And yes, as the quote I cited from Ilkka notes, the area could use some study.

 

[Steve81]

Obviously 1ms by itself does not effect the LF the same (from 0ms to 1ms) however going from 10ms to 11ms (or a similar change) can be night and day. Therefore, one 1ms can be heard and noticed, in that circumstance.

What do you think I've been trying to convey when I've been saying

what I'm getting at is that if the two combined don't exceed the audible threshold of 1 to 1.5 cycles, I'm not sure that there will be a problem. Its possible an extra 1ms or 8ms across the band could do it, but it really just depends on the sub and where you're planning to cross it.

Is it possible that 1ms will be the difference between audio bliss and audio heck in a handbasket? Sure. Is it probable? Not IMO; certainly it cannot be argued that 8ms would make the issue of audibility a lot more probable. Nonetheless, the answer of whether it is audible or not depends on the subs delay as well.

 

[Steve81]

Also regarding one of my prior posts on group delay and Audyssey. Won't help for the OP, but for those concerned about GD in subs in general.

 

[Gordonj]

What do you think I've been trying to convey when I've been saying



Is it possible that 1ms will be the difference between audio bliss and audio heck in a handbasket? Sure. Is it probable? Not IMO; certainly it cannot be argued that 8ms would make the issue of audibility a lot more probable. Nonetheless, the answer of whether it is audible or not depends on the subs delay as well.

Steve with all due respect, you have been saying that you can not make out 1ms of time regarding LF....

And anytime you want to come out to a tuning of a system so that i can demonstrate the results of proper time alignment, and the resulting alignment of the LF with one additional millisecond, I would be more then happy to demonstrate. After all it seems as if you are close, relatively speaking (at least within 1ms ).

Gordon

 

[Steve81]

No, I've said that several times that if the total delay doesnt exceed the audible threshold, then it won't matter. All I've really said is that an additional 1ms is unlikely to cause a breech of that threshold, although it theoretically is possible. Obviously adding 8ms can be a greater problem since the threshold at 80hz is about 12.5ms.

 

[Ed Mullen]

DSP latency and group delay are completely different things.

The AS-EQ1 has 7.5 ms of DSP latency. That means the entire signal is uniformly delayed (i.e., the delay is not frequency dependent) before it leaves the subwoofer. This is why the AVR will add approximately 8 feet to the actual distance of the subwoofer when an AS-EQ1 is in the system.

Group delay is the derivative of delta phase with respect to frequency. Anything which creates a change in phase response over a portion of the frequency response will cause a change in group delay. A vented subwoofer has a large phase change near tuning, and this is evidenced by an increase in GD. If a high pass filter is also added to protect the woofer below tuning, the HPF will also obviously create a phase change, which will increase GD even more.

As discussed previously, GD audibility is frequency dependent, so it's best to think of GD in terms of cycles rather than simply looking at the absolute value in ms. 50 ms of GD at 20 Hz (i.e., 1 cycle) is not really a big deal, but 50 ms of GD at 1 kHz (i.e. 50 cycles) would be very problematic/audible.

 

[Steve81]

Hi Ed,

I understand that DSP latency and group delay are different, but unless I'm misunderstanding what I'm looking at in Ricci's spectrograms, group delay is literally a measure of the (frequency dependent) delay of a subwoofer's output. In the OPs case, he won't be able to correct for that 8 feet, so the question is how big of an impact would that have. It seems like it would be akin to adding 8ms across the entirety of the GD curve of a subwoofer, which can obviously have negative consequences in some (probably a lot) of cases. Please correct me if that impression is mistaken.

 

[Ed Mullen]

Hi Ed,

I understand that DSP latency and group delay are different, but unless I'm misunderstanding what I'm looking at in Ricci's spectrograms, group delay is literally a measure of the (frequency dependent) delay of a subwoofer's output. In the OPs case, he won't be able to correct for that 8 feet, so the question is how big of an impact would that have. It seems like it would be akin to adding 8ms across the entirety of the GD curve of a subwoofer, which can obviously have negative consequences in some (probably a lot) of cases. Please correct me if that impression is mistaken.

Hi Steve -

DSP latency and actual acoustic distance can be compensated for by the AVR subwoofer distance setting.

There is no way to compensate/correct for the group delay profile of a given subwoofer - it is an inherent function of the electro/acoustic design of the subwoofer. Nor would I be concerned about this.

A variable tune subwoofer like the SVS PB13U might help you understand this. Take a look at the GD profile for this subwoofer in all three tunes (20 Hz reflex, 16 Hz reflex, and Sealed).

SVS PB13-Ultra Subwoofer Measurements and Analysis &mdash; Reviews and News from Audioholics

While the three GD curves obviously differ at the deepest bass frequencies (which is perfectly normal and expected), notice how they are all virtually identical overlays > 28 Hz.

Were you to set-up a PB13U and an AS-EQ1 in a system with an auto-set-up AVR, it would set virtually the same overall subwoofer distance each time if you ran it separately for all three tunes. This is because the GD profiles for all three tunes are virtually identical >28 Hz.

While the audibility of GD in subwoofers seems to be an evergreen topic, it is IMO overrated. While I actually advised Josh to add the 1.0 and 1.5 cycles to his GD charts in order to provide at least some frame of reference, those values were referenced from GD audibility studies conducted at much higher frequencies. To the best of my knowledge, there has yet to be a definitive study with a statistically high confidence level which clearly defines the audibility of GD at very deep (i.e., <30 Hz) bass frequencies. No one listening to a PB13U in the 20 Hz mode would subjectively think 'gee the 18-22 Hz bandwidth is obviously lagging way behind the mid/upper bass regions' - we just don't subjectively perceive extremely deep bass that way. Josh basically makes a statement to this effect in all his subwoofer reviews, and has similar thoughts on the whole GD audibility issue as it relates to subwoofers.

Things like a flat frequency response, low distortion, high dynamic output capability, excellent bandwidth uniformity, and a lack of audible overdrive artifacts have far more influence on the subjective listening experience than does the GD profile of the subwoofer.

 

[Steve81]

Hi Steve -

DSP latency and actual acoustic distance can be compensated for by the AVR subwoofer distance setting.

I know, but the OP is looking to use it in a simple 2 channel system where he can't make that adjustment, or at least that's the impression I got. Aside from the suggestions already given, I'm not sure what else he can do, but that's what got this going, ie how much it matters.

There is no way to compensate/correct for the group delay profile of a given subwoofer

According to Chris K. at Audyssey, his product can. Don't know whether he misunderstood the question, but that's what he claimed when asked; in any case though, I'd agree, I'm not worried about it for my usage.

 

[Ed Mullen]

I know, but the OP is looking to use it in a simple 2 channel system where he can't make that adjustment, or at least that's the impression I got. Aside from the suggestions already given, I'm not sure what else he can do, but that's what got this going, ie how much it matters.

We include a delay function (up to 10 ms) on the line level outputs of our Sledge 800/1000 platforms in order to compensate for the lack of distance controls on most 2-channel pre/pros.

Aside from that, the only other option to time-align the mains and subwoofer is to physically move the subs closer to the LP than the mains.

While a discussion of minimum phase is way beyond the OP topic, the below article does delve into how a subwoofer interacts in the room, and how the use of EQ can be used to modify the in-room FR and in-room phase response. Naturally the in-room phase response of the subwoofer differs from the quasi-anechoic phase response of the subwoofer - and the two GD plots will of course be different too.

Minimum Phase