Principles: Time of Arrival Delays, vs Phase tuning ?

[john61ct]

I've been getting into the weeds on DSP for "speaker building", differences between using FIR from scratch to get linear phase crossovers vs using IIR to "correct phase" in already magnitude-working crossovers.

I've seen members state that "time alignment" is a completely different topic domain from that above sort of phase tuning.

Which makes me think they mean adjusting delays to get time of arrival from different speakers in sync at a single listening position.

Apparently MSO can automate that, across lotsa overlapping LF units anyway, and usually to "smooth the bass" across a larger multiple LP area.

In looking to create a Modularised DSP ™ system, I see some "network clock" sync protocols work so that "good enough" for same-room playback is at say 2ms, with a sliding drift up to 8ms tolerated before a perceptible "reset" to prevent random signal clicks, pops or drops.

But in phase adjustment, even hundredths of a ms are apparently significant enough to be audible, at least in the context of blending at crossover points.

Is this a fundamental difference between "latency delays" "transport / distance" delays on the one hand and phase alignment on the other?

There is also disagreements as to the importance of addressing "group delays". High SQ in transient response is a key goal for me.

I have been resisting the high expense of the conventional "single-interface approach for all my A/D conversion points", but maybe my alternative concept of using multiple cheap RPi convolvers is an unrealistic garden path?

I realise this is a big topic domain (or more than one), so links to ELI5 noob level learning resources would be greatly appreciated

My @Floyd Toole "Sound Reproduction" text has not helped much so far.

Sound Reproduction | The Acoustics and Psychoacoustics of Loudspeakers

This fourth edition of Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers, Rooms and Headphones explains the physical and perceptual

www.taylorfrancis.com

 

[Redacted]

We've been told on ASR that delays in amplifiers are "irrelevant" and AI suggests these delays are 5 to 20ms (but users can pick a number they believe in, if unable to check with a 'scope). But given some of your numbers, it seems to me that every channel (surround or XO) should use amps with the same delay or at least measure the delay and take it into account in the DSP setup. And if we are getting down to stepped speaker positioning, then clearly we are getting down to quite small delay times apparently being relevant after all.... I don't know which of all the various room correction systems deal with time delays...

I mean this as possible food for thought and not a criticism of the DSP work - I'm strictly a K.I.S.S. guy myself!

 

[john61ct]

Well this question is really on crossover tech nearly exclusively

not by other DSP functions, much less analog-only components in the chain like amps!

I may not need to use DSP at all, unless there are phase issues at the crossovers

For KISS, definitely setting DRC and MSO aside for now, let's assume no such issues below Schröder transition, lots of subs carefully and successfully placed.

IOW please address the OP questions in the context of DSP "speaker building" functions only, specifically crossover-related phase tuning

vs the much larger timing delays cause by the DSP processing itself, much greater below Schröder, and more so as you go deeper in the subs stack, say four-way simulated stereo floorstanders, plus mono units for free placing also going right down to or even past 20Hz.

I suspect the audibility of the much tinier delays involved in phase-linearizing already frequency-correct crossovers is related to changes in magnitude due to these not being FIR-linear phase crossovers, from-scratch created by DSP in the first place.

"smearing" and interference with transients performance

rather than extreme non-linearities like "echo" artifacts or explicit pops, signal dropouts...

 

[Keith_W]

The only way for you to understand is to play with some measurements. Try to time align the impulses, then watch what happens to the phase alignment. Or align the phase, and see what happens to the time alignment. Depending on your room, it may not be possible to achieve both simultaneously. You will have to choose one over the other. The next time someone posts a thread on ASR requesting help aligning subs to mains, i'll tag you and you can go download the MDAT and play with it.

As for time aligning to hundredth of a millisecond, that is absurd. 1ms = 34.3cm / about 13". That is how far your head might move while you are listening. Unless you invest in one of these:

DSP lets you time align to a fraction of a hundredths of a millisecond, but you do not need that kind of precision. It is WAY overkill. But since I am able to do it, I do it.

 

[MaxwellsEq]

We've been told on ASR that delays in amplifiers are "irrelevant" and AI suggests these delays are 5 to 20ms

A delay through an analogue audio amplifier will be in the microsecond range or less.

Edited to fix a typo

 

[alex-z]

You can have phase alignment with poor time alignment, a full cycle is 360 degrees of phase rotation.

For example, if you had 2 sources playing a 100Hz signal but the second source arrived 10ms late it would have good phase alignment so you would expect a good frequency response but unnecessary group delay.

With time aligned sources the phase should match over the entire bandwidth. In practice you will see deviations that increase with frequency due to room reflections and speaker radiation pattern.

Sources can refer to discrete speakers/subs or multiple drivers within a single speaker/sub.

MSO does not optimize for time alignment. It tries to balance your frequency response and reduce spatial variation, aka the Harman approach to managing room modes.

https://www.harman.com/documents/multsubs_0.pdf

Time alignment of subwoofers can increase spatial variation in residential rooms, I would only recommend it for systems where the listening position is fixed.

You can run multiple DSP units from a single source, if you are doing it over a network make sure it uses a protocol like Dante that is built for audio. 2ms of drift would be significant, Dante for example mutes hardware outputs over .25ms of jitter/drift.

If you want good transient response you should address room acoustics first. Even advanced MIMO DSP like Dirac ART cannot compensate for uncontrolled reflections above 150Hz, absorption and diffusion is needed for balanced ratio of early vs late energy arrival.

 

[john61ct]

Thank you!

The only way for you to understand is to play with some measurements.

I will continue to try to understand more on this prior to spending big money (for me) on the multichannel enabling interface hardware.

If I can successfully build a compute with Firewire working, then my 48 ports of cheap I/O will be a solid hands-on playground, maybe even run by a single RPi5 which would be ideal

As good value for PC I/O as Echo Audiofire 12 ?

Total interface noob here, please ELI5 I don't need to record or mix music in the creative sense, just looking to get multi-channel analog in and out of HTPC type hardware. Primary use case is DIY convolving, loudness contour, bass management, room correction PEQ. The Echo Audiofire 12 unit...

www.audiosciencereview.com

I suppose I could meantime play-learn using an RPi5 + two 8-port HATs one ADCs the other DACs.

> Try to time align the impulses, then watch what happens to the phase alignment. Or align the phase, and see what happens to the time alignment. Depending on your room, it may not be possible to achieve both simultaneously. You will have to choose one over the other.

But that goes away if I have enough processing power to do FIR linear phase crossover filters from scratch, correct? Not worrying about total system latency...


> you can go download the MDAT and play with it

I'm grateful, but flattered you think so, I have no PC working at this point just parts. I think my kids only have big gaming desktops, all the laptops are Chromebooks...

> As for time aligning to hundredth of a millisecond, that is absurd. 1ms = 34.3cm / about 13"

Exactly, yet when creating crossover filters in the time domain, even within co-located multiple boxen stacks with less time differences than that, all I'm reading talks about comb filtering, stereo imaging / soundstage getting shot, the need to get the phase alignment as close as possible ??

That exact cognitive dissonance is the very crux of this thread!

 

[-Matt-]

Time delays in an amplifier are tiny (microseconds as suggested above) and crucially they should be always the same.

Likewise, if you have a single multi-channel interface you can be confident that any delay between channels that you specify will be respected (probably to sub ms level) because they share a common clock.

The main problem with what the OP suggests (combining multiple separate dacs) is that you can't be sure that there won't be randomly changing delay between the separate dacs (each having its own separate clock).

These delays could potentially be in the ms range (depending on the dac driver) and importantly they could randomly change each time the system restarts, or could drift out of sync during playback.

Syncronisation across multiple dacs, especially of different brands is never going to be guaranteed.

 

[john61ct]

You can have phase alignment with poor time alignment, a full cycle is 360 degrees of phase rotation.

For example, if you had 2 sources playing a 100Hz signal but the second source arrived 10ms late it would have good phase alignment so you would expect a good frequency response but unnecessary group delay.

But if that "transport delay" was say 4ms it would not be a problem right? (assuming room resonances already handled)

...

> MSO does not optimize for time alignment. It tries to balance your frequency response and reduce spatial variation, aka the Harman approach to managing room modes.

So the "Maximize SPL using only delays and all-pass filters" is a different technology?

See I thought "All-Pass" were ONLY time / phase domain, and how are those different from the "discrete delay" blocks available in MSO ?

Again clearing up my confusion on these terms is EXACTLY what I'm asking for help with here.

? @andyc56

...

> Time alignment of subwoofers can increase spatial variation in residential rooms, I would only recommend it for systems where the listening position is fixed.

Well that is the case 95% of the time for me.

I am willing to swap things around for dance parties, background listening and occasional film watching in groups (with a delay in the video for lip sync). That covers all my use cases I believe...

> You can run multiple DSP units from a single source, if you are doing it over a network make sure it uses a protocol like Dante

Too pricey afaict, if needed I hope to find other solutions.

> If you want good transient response

I do, very much.

> (then) you should address room acoustics first.

Nope, simplynot possible. this system (or the more portable subset) is mobile, deployed off grid in various widely different spaces large and small, and often used outdoors.

So that's why, for this thread I've stated "ignore DRC", there will be little to no "physical room" treatment and in practice if I do start looking at DRC it will be the last finishing touch when setting up for a session or maybe a few days deployment at a time.


> Dirac ART

Not anticipating ever having budget for that sort of solution. Gsonic Reference sure? miniDSP 2x4 is expensive for me, and of course only a learning tool toward 16+ channels...

 

[john61ct]

Yes, and these issues are exactly why I'm here!

if you have a single multi-channel interface
...
Synchronisation across multiple dacs, especially of different brands is never going to be guaranteed.

Well I'm hoping there are ways, and not pricey like Dante.

...

I'm thinking this idea of using SlimProto may not work, although it's cheap enough to test

LMS + SqueezeDSP for Modularised DSP ™ - perfect network clock sync ?

LMS = Lyrion Music Server, formerly Logitech Media Server, assume here a noisy powerful server out in the garage, also running the HTC, NAS etc. SqueezeDSP is a convolution plug-in that runs on LMS, and squeezelite is the protocol stack used for comms between LMS and the renderer ("player")...

www.audiosciencereview.com

...

So yes, my Plan B is one Big Cahuna master clock interface, hence

ADAT interface hosts, high SQ but cheaper than RME Digiface USB?

I am looking for an interface "hub" based on multiple ADAT, clean and transparent. I'm not recording / mixing, just for recreational home DSP - crossovers timing delays / phase alignment tuning Happy to buy secondhand to save money, vintage discontinued no problem. 24-bit, 44.1 / 48kHz is...

www.audiosciencereview.com

...

But meantime I hope to find explanatory texts that help me understand the difference between "transport delays - not phase misalignments - are OK if kept below say 6ms"

And "even slight phase issues below 1ms can reduce system SQ"

before I'm in a position to test, hands-on

 

[Redacted]

A delay through an analogue audio amplifier will be in the microsecond range or less.

I believe you if you say so. I got this off AI - so it may well be nonsense, just gleaned from 800 "blokes from down the pub" or "listed sources" as the pic shows!

I would measure it just for fun, but I only have a cheap 'scope, and not too bothered.

I was just curious about the various threads talking about time delay, phasing and room correction - it's not something I'd bother about, as my modest kit sounds good enough if I sit anywhere in the room, but the best "stereo effect" or image if you like, is in a 2-3 sq. metre area.
.
.

.

 

[john61ct]

Please note also I am not talking about using multiple "DAC units" as that is commonly used here.

My Echo Audiofire12s AS CONVERTERS have 12x ADCs 1/4" ports for input + 12x DACs 1/4" ports for output. The "interface" part being firewire make it likely I can't go that way.

Hence my asking above about "ADAT converters" just A/D conversion boxen, to use via a "hub interface" like the RME Digiface.

Those are my "DACs" and a total hardware cost of > $20/analog port is "expensive" from my POV, only want to go that way if there truly is no path forward for the multiple-RPi Modularised DSP ™ system concept.

 

[john61ct]

> I believe you if you say so. I got this off AI - so it may well be nonsense

...

I appreciate your transparency, but please abstain from posting "facts" gleaned from AI sources here.

Very unreliable, and in any case such delay sources (analog only components) are OT in this thread as I stated.

 

[Keith_W]

I'm grateful, but flattered you think so, I have no PC working at this point just parts. I think my kids only have big gaming desktops, all the laptops are Chromebooks...

I would think your priority should be to get a working PC. You will need it to design your filters. And you can also use it as a stand-in for your Rpi5. What good is getting a DSP platform ready if you don't have a PC to design filters for it?

Also, you don't need FIR to implement delays. It can be done with IIR very easily.

Anyway, one good way to learn how all these things work is to create simulations. You can use REW to create simulations, but it's a bit clunky at doing that. I learnt a lot by playing with Acourate.

 

[Waveform Fidelity]

I've been getting into the weeds on DSP for "speaker building", differences between using FIR from scratch to get linear phase crossovers vs using IIR to "correct phase" in already magnitude-working crossovers.

I've seen members state that "time alignment" is a completely different topic domain from that above sort of phase tuning.

Which makes me think they mean adjusting delays to get time of arrival from different speakers in sync at a single listening position.

Apparently MSO can automate that, across lotsa overlapping LF units anyway, and usually to "smooth the bass" across a larger multiple LP area.

In looking to create a Modularised DSP ™ system, I see some "network clock" sync protocols work so that "good enough" for same-room playback is at say 2ms, with a sliding drift up to 8ms tolerated before a perceptible "reset" to prevent random signal clicks, pops or drops.

But in phase adjustment, even hundredths of a ms are apparently significant enough to be audible, at least in the context of blending at crossover points.

Is this a fundamental difference between "latency delays" "transport / distance" delays on the one hand and phase alignment on the other?

There is also disagreements as to the importance of addressing "group delays". High SQ in transient response is a key goal for me.

I have been resisting the high expense of the conventional "single-interface approach for all my A/D conversion points", but maybe my alternative concept of using multiple cheap RPi convolvers is an unrealistic garden path?

I realise this is a big topic domain (or more than one), so links to ELI5 noob level learning resources would be greatly appreciated

My @Floyd Toole "Sound Reproduction" text has not helped much so far.

Sound Reproduction | The Acoustics and Psychoacoustics of Loudspeakers

This fourth edition of Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers, Rooms and Headphones explains the physical and perceptual

www.taylorfrancis.com

My recommendation to undertsand everything, try to make a clear distinction between time delay and phase - they are 2 different topics. Think of phase as a property of the source (the loudspeaker) and time delay a property of the propagation channel (the air path between speaker and ear)

 

[MaxwellsEq]

I believe you if you say so. I got this off AI - so it may well be nonsense, just gleaned from 800 "blokes from down the pub" or "listed sources" as the pic shows!

I would measure it just for fun, but I only have a cheap 'scope, and not too bothered.

I was just curious about the various threads talking about time delay, phasing and room correction - it's not something I'd bother about, as my modest kit sounds good enough if I sit anywhere in the room, but the best "stereo effect" or image if you like, is in a 2-3 sq. metre area.
.
.

.View attachment 535884

I used to build very large high bandwidth networks for media companies. We had a rule of thumb that light took ONE millisecond to traverse 200km of fiber. Copper is quicker!

If you take a line-level preamplifier, the signal probably traverses at most 3 metres of copper trace and wire, so if it was a "passive preamplifier" the delay would be a few nanoseconds! Resistors, capacitors, op-amps, inductors, transistors will all will only add sub-microsecond delay.

 

[kemmler3D]

We've been told on ASR that delays in amplifiers are "irrelevant" and AI suggests these delays are 5 to 20ms

Your AI answer is conflating DSP and amp. This is why you shouldn't rely on AI for this stuff, it only sounds like it knows anything. Amps that don't do DSP should have negligible delay, like far less than 1ms.

the difference between "transport delays - not phase misalignments - are OK if kept below say 6ms"

Transport delay here probably means a gross delay of the whole signal.

And "even slight phase issues below 1ms can reduce system SQ"

This is probably referring to things going out of phase, especially individual drivers - phase when expressed as delay is frequency dependent. If that doesn't immediately make a lightbulb go off (it wouldnt' for me if I hadn't studied this in college a little) spend a few minutes messing with this calculator: https://sengpielaudio.com/calculator-timedelayphase.htm

180 degrees out of phase means full cancellation. But at 20hz, 180 degrees means 25ms of delay. At 2000hz, it's 0.25ms. The length of time is determined by the wavelength. At 20hz it takes a whole 25ms for a soundwave to go by, at 2000hz, it's much shorter and so the wave takes much less time to go by.

This is why using multiple DACs (and especially multiple clocks) in a single speaker system is problematic. For two drivers in one speaker, as you can see 0.25ms delay can be a problem. Separate DACs tend to drift much more than this over the course of a few minutes. This is not obvious fron DAC measurements, but apparently differences of several miliseconds can accumulate over time between two decent DACs.

If you are running your woofer and tweeter on separate clocks it's going to sound insane. Even two clocks for two separate loudspeakers is likely to cause problems.

Is this a fundamental difference between "latency delays" "transport / distance" delays on the one hand and phase alignment on the other?

So bottom line, yes, big difference. Phase is frequency dependent, the delay will vary. "latency / distance" delays are a fixed delay for the whole signal, not frequency dependent.

So if you rotate phase 90 degrees for the whole signal, you get 12.5ms of delay at 20hz and 0.125ms of delay at 2000hz.

If you delay the whole signal by 5ms, you get 3600 degrees of delay at 2000hz (goes around 10 times) and 36 degrees of delay at 20hz.

Having looked at DIYing speakers with a focus on DSP crossovers a while back, I concluded that either you concoct something with Raspberry Pis to save money, or you "buy once, cry once" and go with Hypex fusion, IMO miniDSP is expensive enough that I'd rather just go for the Hypex.

It turns out DIYing speakers is not as cheap as you'd think.

 

[Waveform Fidelity]

Your AI answer is conflating DSP and amp. This is why you shouldn't rely on AI for this stuff, it only sounds like it knows anything. Amps that don't do DSP should have negligible delay, like far less than 1ms.


Transport delay here probably means a gross delay of the whole signal.


This is probably referring to things going out of phase, especially individual drivers - phase when expressed as delay is frequency dependent. If that doesn't immediately make a lightbulb go off (it wouldnt' for me if I hadn't studied this in college a little) spend a few minutes messing with this calculator: https://sengpielaudio.com/calculator-timedelayphase.htm

180 degrees out of phase means full cancellation. But at 20hz, 180 degrees means 25ms of delay. At 2000hz, it's 0.25ms. The length of time is determined by the wavelength. At 20hz it takes a whole 25ms for a soundwave to go by, at 2000hz, it's much shorter and so the wave takes much less time to go by.

This is why using multiple DACs (and especially multiple clocks) in a single speaker system is problematic. For two drivers in one speaker, as you can see 0.25ms delay can be a problem. Separate DACs tend to drift much more than this over the course of a few minutes. This is not obvious fron DAC measurements, but apparently differences of several miliseconds can accumulate over time between two decent DACs.

If you are running your woofer and tweeter on separate clocks it's going to sound insane. Even two clocks for two separate loudspeakers is likely to cause problems.


So bottom line, yes, big difference. Phase is frequency dependent, the delay will vary. "latency / distance" delays are a fixed delay for the whole signal, not frequency dependent.

So if you rotate phase 90 degrees for the whole signal, you get 12.5ms of delay at 20hz and 0.125ms of delay at 2000hz.

If you delay the whole signal by 5ms, you get 3600 degrees of delay at 2000hz (goes around 10 times) and 36 degrees of delay at 20hz.

Having looked at DIYing speakers with a focus on DSP crossovers a while back, I concluded that either you concoct something with Raspberry Pis to save money, or you "buy once, cry once" and go with Hypex fusion, IMO miniDSP is expensive enough that I'd rather just go for the Hypex.

It turns out DIYing speakers is not as cheap as you'd think.

On DAC drift its typically <5ppm, so no idea where you conclude that the time drift between 2 DAC >0.25 msec over a few minutes. Rotating the phase of a signal doesn’t introduce any delay at all. A signal leaving the amplifier with 60 or 90 degree phase, leaves at the same time, but its starting phase is different.

 

[wwenze]

I don't know which of all the various room correction systems deal with time delays...

Look under "distance" setting in your AVR

 

[john61ct]

I hope light bulbs start going off for me eventually.

Please keep on track folks, room correction and AV gear is off topic for sure

Rotating the phase of a signal doesn’t introduce any delay at all

Wow, light bulb blown