Understanding the state of the DSP market

[levimax]

Re the distortion in the low end, you have to pick what you deem acceptable, but yeah my main argument was that I'm sure we can simplify things by just looking at distortion and frequency response. Re "delay" that you mention are you saying that all frequencies are delayed by 100ms in some of these subs in terms of based on when they first receive the electrical signal from your DAC?

Group Delay is frequency dependent and is caused by DSP filters (In the case of a small DSP sub much of it by the "high pass protection filter) not by the DAC and typically it rises rapidly toward lower frequencies. See below measured group delay for my SVS SB 3000 (red) and the group delay for my "main" speakers (green) which are large sealed with no DSP. The "common wisdom/ rule of thumb" is that "one cycle" of group delay is acceptable but that breaks down at LF as the delay become quite long for one cycle at 20 Hz. The other "common wisdom/ rule of thumb" is less than 20 ms at LF which makes sense to me. Many subs, usually larger well designed ones, do manage <20 ms @ 20 Hz but the many of the heavily DSP'd small subs have much higher group delay. The problem with group delay is not just audibility but also it makes integrations more difficult and I have seen it even cause cancellation issues.

 

[ernestcarl]

things can be simplified by saying lets just make sure the distortion and frequency response is ok.

I have no qualms with that sentiment, but levimax has repeatedly mentioned already integration was not satisfactory due to the non-defeatable DSP — SVS’s own choice. You at least have a bit more choice with Rythmik subs, for example.

The concern shared is valid so there is really no reason to dumb down the conversation to whatever you might think are the main LF requirements.

 

[Robbo99999]

Group Delay is frequency dependent and is caused by DSP filters (In the case of a small DSP sub much of it by the "high pass protection filter) not by the DAC and typically it rises rapidly toward lower frequencies. See below measured group delay for my SVS SB 3000 (red) and the group delay for my "main" speakers (green) which are large sealed with no DSP. The "common wisdom/ rule of thumb" is that "one cycle" of group delay is acceptable but that breaks down at LF as the delay become quite long for one cycle at 20 Hz. The other "common wisdom/ rule of thumb" is less than 20 ms at LF which makes sense to me. Many subs, usually larger well designed ones, do manage <20 ms @ 20 Hz but the many of the heavily DSP'd small subs have much higher group delay. The problem with group delay is not just audibility but also it makes integrations more difficult and I have seen it even cause cancellation issues.

View attachment 451384

So in that example a 20Hz sine tone would be delayed by about 85ms vs the time it received the signal? And conversely at 90Hz a sine tone would play immediately vs the time it received the signal? If so I suppose I can imagine in music where it's obviously composed of different frequencies being played at the same time then I could almost imagine that if some elements are delayed then it would place that frequency band in a slightly incorrect place in time vs what else is going on in the music so I could imagine some "smearing". Thinking of it in lines of distortion then distortion is other frequencies being activated by an original tone, the harmonic distortion like 2nd order, 3rd order, etc where the frequencies are multiples of the original - well in the case of this group delay then it's almost like zero order distortion or 1st order distortion (I'm making up terms here), yes it would be 1st order distortion because if it's delayed no sound is being produced when it is being expected to be produced at that frequency (I'm making up terms). No but I'm just trying to imagine the effect on the sound. The lower bass notes are less audible though and more felt, so almost if anything then delay of that feel rather than sound is what would happen. Perhaps if it were ever possible that delay was shown at the higher more readily audible frequencies then maybe that would be more detectable. I don't know, I'm just trying to imagine it.

EDIT: maybe this group delay would be most audible in recordings of single instruments that activate that range. Maybe in really low bass drums, because a strike on a drum would offer different decaying frequency patterns over time and if the lower frequency band is out of time with decaying sound of the drum after the strike then maybe it could make it sound different and I was gonna say mask some of the other sounds that it's out of time with, but because of the low low frequencies involved that are more felt then less likely to mask those more audible frequencies that are higher up.

EDIT#2: in the example of the drum strike above it would actually make the length of the drum strike shorter by whatever the group delay was at the main frequency if you assume that the lowest frequencies of a reverberating drum strike occur at time zero of the strike. 100ms is a tenth of a second. How long is a drum strike for drums that can activate these low frequencies? And if the lowest frequencies occur at the end of a reverberating drum strike then it would be the opposite effect and the drum strike would be lengthened.

 

[OCA]

Group Delay is frequency dependent and is caused by DSP filters (In the case of a small DSP sub much of it by the "high pass protection filter) not by the DAC and typically it rises rapidly toward lower frequencies. See below measured group delay for my SVS SB 3000 (red) and the group delay for my "main" speakers (green) which are large sealed with no DSP. The "common wisdom/ rule of thumb" is that "one cycle" of group delay is acceptable but that breaks down at LF as the delay become quite long for one cycle at 20 Hz. The other "common wisdom/ rule of thumb" is less than 20 ms at LF which makes sense to me. Many subs, usually larger well designed ones, do manage <20 ms @ 20 Hz but the many of the heavily DSP'd small subs have much higher group delay. The problem with group delay is not just audibility but also it makes integrations more difficult and I have seen it even cause cancellation issues.

View attachment 451384

A good part of it is caused by port induced phase shifts, GD is derivative of phase. Sealed subs have less and flatter GD. Another interesting feature of GD is that, it's harmless when it's flat no matter how large, GD with a slope is the real problem. My 2 cents!

 

[levimax]

So in that example a 20Hz sine tone would be delayed by about 85ms vs the time it received the signal? And conversely at 90Hz a sine tone would play immediately vs the time it received the signal? If so I suppose I can imagine in music where it's obviously composed of different frequencies being played at the same time then I could almost imagine that if some elements are delayed then it would place that frequency band in a slightly incorrect place in time vs what else is going on in the music so I could imagine some "smearing". Thinking of it in lines of distortion then distortion is other frequencies being activated by an original tone, the harmonic distortion like 2nd order, 3rd order, etc where the frequencies are multiples of the original - well in the case of this group delay then it's almost like zero order distortion or 1st order distortion (I'm making up terms here), yes it would be 1st order distortion because if it's delayed no sound is being produced when it is being expected to be produced at that frequency (I'm making up terms). No but I'm just trying to imagine the effect on the sound. The lower bass notes are less audible though and more felt, so almost if anything then delay of that feel rather than sound is what would happen. Perhaps if it were ever possible that delay was shown at the higher more readily audible frequencies then maybe that would be more detectable. I don't know, I'm just trying to imagine it.

EDIT: maybe this group delay would be most audible in recordings of single instruments that activate that range. Maybe in really low bass drums, because a strike on a drum would offer different decaying frequency patterns over time and if the lower frequency band is out of time with decaying sound of the drum after the strike then maybe it could make it sound different and I was gonna say mask some of the other sounds that it's out of time with, but because of the low low frequencies involved that are more felt then less likely to mask those more audible frequencies that are higher up.

EDIT#2: in the example of the drum strike above it would actually make the length of the drum strike shorter by whatever the group delay was at the main frequency if you assume that the lowest frequencies of a reverberating drum strike occur at time zero of the strike. 100ms is a tenth of a second. How long is a drum strike?

Group delay audibility, especially at lower frequencies, is not settled science unfortunately. The most common concern with Group Delay is with something like a "drum hit" where ideally you would want all the frequencies that are part of the drum hit to play at the same timing as original so you get "impact" of the original. You also bring up some interesting points about how group delay would interact with distortion which I have never considered. The safest and most reasonable conclusion is that minimizing distortion and group delay is a reasonable way to improve sound quality as well as making integration easier. The levels of distortion and group delay we are dealing with for many subs is quite high and is not at all like chasing SINAD from 100 dB to 120 dB.

 

[Fidji]

So in that example a 20Hz sine tone would be delayed by about 85ms vs the time it received the signal? And conversely at 90Hz a sine tone would play immediately vs the time it received the signal? If so I suppose I can imagine in music where it's obviously composed of different frequencies being played at the same time then I could almost imagine that if some elements are delayed then it would place that frequency band in a slightly incorrect place in time vs what else is going on in the music so I could imagine some "smearing". Thinking of it in lines of distortion then distortion is other frequencies being activated by an original tone, the harmonic distortion like 2nd order, 3rd order, etc where the frequencies are multiples of the original - well in the case of this group delay then it's almost like zero order distortion or 1st order distortion (I'm making up terms here), yes it would be 1st order distortion because if it's delayed no sound is being produced when it is being expected to be produced at that frequency (I'm making up terms). No but I'm just trying to imagine the effect on the sound. The lower bass notes are less audible though and more felt, so almost if anything then delay of that feel rather than sound is what would happen. Perhaps if it were ever possible that delay was shown at the higher more readily audible frequencies then maybe that would be more detectable. I don't know, I'm just trying to imagine it.

Is not about distortion and harmonics, but about fundamental frequency of a note and harmonics, that make the timbre of the instrument or voice. E.g. this is frequency spectrum of open E bass note. You want them all to arrive at the same time.

Actually Group Delay is one of the more important things to get right if you want really high performant system, especially below 100Hz. Usually sealed speakers behave much better in that respect and ported speakers trade-off their bass extension for worse Group Delay.

To get FR right is trivial.

 

[levimax]

A good part of it is caused by port induced phase shifts, GD is derivative of phase. Sealed subs have less and flatter GD. Another interesting feature of GD is that, it's harmless when it's flat no matter how large, GD with a slope is the real problem. My 2 cents!

Ports do add GD but in the case of the SVS SB3000 they are sealed subs so much of the GD has to be caused by DSP filters with a high order "protective high pass" to protect a boosted LF driver being the most likely. There are also plenty of well designed ported subs with less than 20 ms of GD. To me the problems come in when you "push" things to make specs look better (high SPL at 20 Hz) at the expense of distortion and GD. Those are my 2 cents

 

[OCA]

one of the more important things to get right if you want really high performant system, especially below 100Hz

There's very little one can do to correct GD below 100Hz. You'd need to use time-inverted allpass phase filters, which introduce pre-ringing. An accepted inaudibility threshold for time inversion is around 5 ms, allowing corrections only down to about 200 Hz. This restricts GD correction primarily to phase shift issues at woofer/midrange and midrange/tweeter crossovers.

 

[OCA]

Ports do add GD but in the case of the SVS SB3000 they are sealed subs so much of the GD has to be caused by DSP filters with a high order "protective high pass" to protect a boosted LF driver being the most likely. There are also plenty of well designed ported subs with less than 20 ms of GD. To me the problems come in when you "push" things to make specs look better (high SPL at 20 Hz) at the expense of distortion and GD. Those are my 2 cents

I hear you. Probably the amp is not powerful enough to sustain 20Hz in a sealed box at high SPLs.

 

[Fidji]

There's very little one can do to correct GD below 100Hz. You'd need to use time-inverted allpass phase filters, which introduce pre-ringing. An accepted inaudibility threshold for time inversion is around 5 ms, allowing corrections only down to about 200 Hz. This restricts GD correction primarily to phase shift issues at woofer/midrange and midrange/tweeter crossovers.

Same speakers, different DRC [Optimizer + Waveforming vs MSO].
E.g. Trinnov is very much focused on correcting Impulse and GD, with balancign pre-ringing artifacts.

This is why I prefer sealed speakers overall, includign subs. They are inherently better in GD and require less correction. Now I have all my speakers sealed and it is quite a differnce in how easily you can integrate subs to mains/satellites.

 

[kyuu]

Same speakers, different DRC [Optimizer + Waveforming vs MSO].
E.g. Trinnov is very much focused on correcting Impulse and GD, with balancign pre-ringing artifacts.

View attachment 451456

This is why I prefer sealed speakers overall, includign subs. They are inherently better in GD and require less correction. Now I have all my speakers sealed and it is quite a differnce in how easily you can integrate subs to mains/satellites.

Seems like the speakers being sealed is really the only necessity for ease of integration. A ported sub only has phase/GD issues around the port's tuning frequency, which should be well below where it works be interacting with the other speakers. Or am I overlooking something?

 

[dbaudio]

Any ideas on AI-based DSP FIR filter design?

 

[mitchco]

Yes, working on it now.

 

[levimax]

Seems like the speakers being sealed is really the only necessity for ease of integration. A ported sub only has phase/GD issues around the port's tuning frequency, which should be well below where it works be interacting with the other speakers. Or am I overlooking something?

Due to DSP used to push FR response lower on my main sealed active speakers it has more GD at higher frequencies than my large ported sub. It is not terrible but it makes integration harder than it could be. The more I get into it sub integration the more I realize how difficult it is to do correctly.

 

[dualazmak]

I found that really nice and invaluable discussions have been going-on starting last Sunday on DSP integration of subwoofers with main-SP's woofers mainly focusing on group delay (GD) of subwoofers.

Let me summarize the major factors affecting the cause of the GD and DSP correction/tuning thereof as follows;

1. Most upstream DSP software and its functionalities i.e. selection of XO Fq, XO filter type and slopes, phase/polarity, relative gains, EQ(s) if applied, etc.,

2. Ported or sealed in subwoofer design,

3. Active (directly driven by dedicated amplifier) and/or passive LCR low-pass (high-cut) network, or DSP-type low-pass filter (AD for DSP then DA), in subwoofer,

4. Ported or sealed in main SP design for woofers,

5. Active (directly driven by dedicated amplifier), or passive LCR network with high-pass (and of course low-pass), for the woofers.

Edit:
6. Individual room acoustic environments.

The selections and combinations of these factors greatly vary in our individual audio setups, and therefore the optimal (compromise) DSP configurations/corrections and tunings would also very much vary/dependent on each of our setups.


I essentially agree with the following nice comments;

The problem is pushing LF down with a boost requires a strong High Pass "protection" filter which is what causes the crazy high Group Delay on a sub like this. Audibility is not proven but 100 ms is bordering on crazy and almost has to be audible (10 ms is limit for lip sync).

Group Delay is frequency dependent and is caused by DSP filters (In the case of a small DSP sub much of it by the "high pass protection filter) not by the DAC and typically it rises rapidly toward lower frequencies.

Sealed subs have less and flatter GD. Another interesting feature of GD is that, it's harmless when it's flat no matter how large, GD with a slope is the real problem.

There are also plenty of well designed ported subs with less than 20 ms of GD. To me the problems come in when you "push" things to make specs look better (high SPL at 20 Hz) at the expense of distortion and GD.

Having noticed all the above factors and comments, I assume it would be worthwhile and "of-reference-for-you" sharing (again) my setup and tunings in my PC-DSP-based multichannel multi-SP-driver multi-amplifier fully active stereo audio rig (ref. here #931 for the details).

a. I use simple and reliable DSP "EKIO" (IIR filters in cascade 2nd order direct form II biquad in 64 bit floating point) in upstream digital domain in PC.

b. My L&R subwoofers are large-rigid-heavy (48 kg) YAMAHA YST-SW1000 (ref. here) with 30-cm cone driver in ported design and in YST-Helmholtz resonance mechanism, and also have built-in passive -24 dB/Oct LP variable filter as well as with its built-in dedicated powerful amplifier; even with these rather old-fashion configurations, I found (ref. here) that YST-SW1000 is well within the above @levimax's comment of "There are also plenty of well designed ported subs with less than 20 ms of GD." I found the GD of YST-SW1000 in 20 Hz - 50 Hz zone was around 16 msec against woofers (and midranges, tweeters super-tweeters) (ref. here #493 and #494).

c. My main SP's woofer, 30-cm YAMAHA JA-3058 is in heavy(39 kg)-rigid sealed cabinet (ref. here), and now dedicatedly and directly driven (passive LCR has been eliminated) by powerful and excellent-damping-factor amplifier YAMAHA A-S3000.

Prior to fully deciding the DSP integration for my ported subwoofer YST-SW1000 and sealed woofer JA-3058, I thought that it should be critically important to know/observe the transient behaviors (responses to rectangular tone-burst input/excitation) of YST-SW1000 and JA-3058 around the possible low-Fq XO zone, as I shared in my posts #495, #497, #503, #507 on my project thread.

I fortunately (and unexpectedly) found that the woofer JA-3058 in the heavy-rigid-sealed cabinet driven directly by dedicated amplifier (with no passive CLR network) still have excellent transient behavior with little distortion even in 50 Hz - 65 Hz zone.

On the other hand, the subwoofer YST-SW1000 showed unexpectedly excellent transient behavior in 32 Hz zone, but it showed some "distortions" in 50 - 65 Hz zone (I may better to say "muddiness" here) by a little bit prolonged after-shock of the subwoofer cone and/or slow low-gain lower-Fq Helmholtz resonance of the inner air of the so-designed ported cabinet.

The 3D color Gain-Fq-Time spectrum given by ADOBE Audition 3.0.1 was very much effective in semi-objectively measure/observe the tightness and cleanliness of the sound energy distribution in 3D Gain-Fq-Time space. These spectral observations were much helpful for me to decide the XO FQ and the filter slopes thereof for XO between YST-SW1000 and JA-3058.

Spoiler

Ref. #495 , #503

Ref. #507

The whole of above-described measurements and observations have led me to my compromise optimal DSP and other configurations, not only the 16.0 msec GD setting for woofers to synchronize with subwoofers but also selection of XO Fq and filter slopes, in my own audio rig as shown in detail in the diagram under the below spoiler cover.

Spoiler

You would please note that, in my setup, the DSP LP filter for subwoofer YST-SW1000 is 50 Hz (-24 dB/Oct), and furthermore the built-in passive LP in YST-SW1000 is set at 55 Hz (-24 dB/Oct). DSP HP filter for woofer JA-3058 is set at 55 Hz (-12 dB/Oct). These mean YTS-SW1000 and JA-3058 sing well together (or very smooth transition) in 40 Hz - 60 Hz zone.

Spoiler

Edit: on May 18
Let me emphasize again that it should be always very important and indispensable not only objective and semi-objective "measurements and tunings" but also the utilization of persistent/consistent and suitable "audio reference/sampler music playlist" (in my case please refer here), throughout our audio tunings as well as room acoustic treatments for our final intensive subjective assessments.

 

[DWPress]

All of my speakers and subs are sealed as well but I have a SB1000 (also sealed) in the mix too. My big passive 15" sub is in the front wall left corner underneath the main speakers and the SB1000 is diagonal in the room rear right corner which is closer to LP and delayed 2.8ms - both subs run in mono and are currently XO at 80Hz.

LR averages at LP show a pretty smooth response but the right channel impulse is corrupted by the proximity to the DSP'd to death built in stuff on the SVS.

Below are vector averaged LR measurements of my speakers with subs integrated



I keep hopping between this thread and the Time Alignment with Wavelets thread that sprung from this one learning a lot along the way and a few new methods. I've got to try measuring again soon and take the SVS out of the chain see if I can get better decay and GD. If @levimax is correct I might be building another passive sub soon.

 

[ernestcarl]

There's very little one can do to correct GD below 100Hz. You'd need to use time-inverted allpass phase filters, which introduce pre-ringing. An accepted inaudibility threshold for time inversion is around 5 ms, allowing corrections only down to about 200 Hz. This restricts GD correction primarily to phase shift issues at woofer/midrange and midrange/tweeter crossovers.

Where again is this 5ms threshold limit specified? From headphone studies?

I know that movement of the mic/LP even a few cm or inches often shows up unwanted precausal artifacts in the measurements esp. when cutting GD peaks in the bass, and yet, even obvious visibility in the measurements doesn't always necessarily translate to audible pre-ringing.

It's definitely better to not have to force a phase inversion... but sometimes it can also work out fine if the excess phase is not so huge.

 

[OCA]

Where again is this 5ms threshold limit specified? From headphone studies?

I know that movement of the mic/LP even a few cm or inches often shows up unwanted precausal artifacts in the measurements esp. when cutting GD peaks in the bass, and yet, even obvious visibility in the measurements doesn't always necessarily translate to audible pre-ringing.

It's definitely better to not have to force a phase inversion... but sometimes it can also work out fine if the excess phase is not so huge.

You're right, it's hard to find this kind of information since those who know often don't share it, and I don't remember the exact source, but I recall seeing that number in a forum discussion years ago. I later tested and confirmed it myself with many inverted phase filters (which are essentially "compensate" mode all-pass filters in rePhase, or can be generated in REW with gated inversion) in my own system. More recently, I also discovered that Dirac uses 4.5 ms, and even Audyssey left a 4.5 ms buffer seemingly intended for inverted all-pass filters, though it appears they decided against implementing it in production.

You can easily test this yourself (just not with headphones, of course). Pre-ringing, or pre-echo in most cases, is quite audible with the right track.

 

[OCA]

Same speakers, different DRC [Optimizer + Waveforming vs MSO].
E.g. Trinnov is very much focused on correcting Impulse and GD, with balancign pre-ringing artifacts.

View attachment 451456

This is why I prefer sealed speakers overall, includign subs. They are inherently better in GD and require less correction. Now I have all my speakers sealed and it is quite a differnce in how easily you can integrate subs to mains/satellites.

They are not playing it fair though since they are exclusively using MIMO capabilities of the receivers. To you and me, only a graphic equalizer (not even parametric) is available and even that has been obliterated by removing any equalizer for subwoofers from the equation.

 

[OCA]

Yes, working on it now.

Hi Mitch. What are you training it on exactly? If you need measurement data, I have about a thousand (possibly more) Atmos measurements lying around from a wide range of D&M systems, sourced through A1 Evo and would be happy to share.