ProSoundWeb Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: 1 2 [3] 4 5   Go Down

Author Topic: Are linear phase filters really the holy grail?  (Read 20974 times)

Mark Wilkinson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1104
Re: Are linear phase filters really the holy grail?
« Reply #20 on: July 01, 2017, 02:24:16 PM »

Is the amount of delay that the FIR filters add because of processing time? In other words if the DSP were to be faster will is have less propagation delay or have we reach the point of no return?

Also I have heard it said that a speaker system with FIR filters is more stable. For example if a pastor with a wireless mic were to be walking around out in front of the main cluster they are likely to hit spots that will feedback in a non-FIR filtered system, so it’s not consistent over the coverage area. And the same pastor walking around under the FIR filtered system any feedback points are likely to not change as much in relation to where they walk. Is this true?

Hi Kevin, my understanding is there is some processing time that comes from CPU/FGPA/dsp utilization, but that is usually pretty minimal and relatively fixed.
The delay that's talked about is number of samples used by the FIR filter, often called taps...(sorry for using unexplained nomenclature.)
So for instance, the 1024 samples I use for live sound, represents a delay of 1024 samples /48,000 sampling freq or 21.3ms....but the most common use is to use impulse centering which means you end up with 1/2 the just calculated delay or 21.3/2=10.7ms.
You'd think increasing DSP speed would cut down delay, but what happens is when you double the processing speed (say 48k to 96k) you also double the nyquist freq, or rather the freq range the samples are working over.
So without doubling the number of taps used also, your FIR filter ends up with half the resolution you started with at 48kHz. Once you double the taps for the higher processing speed, you're back to the same delay.

The only thing I understand about stability, is that IIR filters with their infinite feedback loops, have the capability of doing just that...going unstable from feedback...
Maybe electro/acoustic feedback from the pastors mic increases the probability of IIR processing feedback ?....  just spit balling here...
Logged

Ivan Beaver

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 9534
  • Atlanta GA
Re: Are linear phase filters really the holy grail?
« Reply #21 on: July 01, 2017, 02:34:23 PM »



Also I have heard it said that a speaker system with FIR filters is more stable. For example if a pastor with a wireless mic were to be walking around out in front of the main cluster they are likely to hit spots that will feedback in a non-FIR filtered system, so it’s not consistent over the coverage area. And the same pastor walking around under the FIR filtered system any feedback points are likely to not change as much in relation to where they walk. Is this true?
That has more to do with the actual cabinets being used, how many, what sort of pattern control they have, how they are arrayed, or whether it is a single cabinet.

There are SOME things that "fancy DSP" can fix, and there are other things that they cannot fix-at least in more than 1 location at the same time.

It is one thing to fix a single cabinet on axis, but when you put that cabinet with others, getting them to work together is quite another challenge-except in a single location.

Once again, the reason a single cabinet source is the best solution.  You don't have other cabinets to mess with and how they are interacting with each other.
Logged
A complex question is easily answered by a simple-easy to understand WRONG answer!

Ivan Beaver
Danley Sound Labs

PHYSICS- NOT FADS!

Frank Koenig

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1318
  • Palo Alto, CA USA
Re: Are linear phase filters really the holy grail?
« Reply #22 on: July 01, 2017, 05:49:33 PM »

I think it has a lot to do with the phase response and how our brains process things, how we are able to focus and listen to one person even if there is extreme background noise.  The flatter the amplitude and phase response the easier it is for our brains to do this.  There are of course many other factors, but I believe a flat phase response over the critical vocal region does help.

This is all well and good. But before we get too carried away it might be instructive to do some listening tests. Real, blinded, listening tests, to reveal just how audible smoothly varying phase is.

Smoothly lagging phase is what we're talking about as the rapidly varying phase, "phase grass" as I call it, is minimum phase, and is taken out by flattening the magnitude, which can be done with a variety of causal, minimum-phase filters, FIR or IIR. What FIR filters alone allow is compensating for lagging phase, as only a non-causal filter can do. We get around the causality problem by introducing a processing delay.

I propose we tune a system to our liking and then introduce a second-order all-pass into the signal chain (the whole signal, not some pass-band that will mess with the crossovers) at a variety of frequencies and see if anyone can hear a difference. We would not be the first to try this, by the way.

To be clear, I'm not bashing the use if FIR filters. Being able to manipulate the phase in both directions (at high frequencies at least) is useful for getting smooth crossovers and getting different speakers to play together. FIR filters are easier to synthesize automatically than banks of bi-quads, even if they are just used to implement (causal) minimum-phase filters (which don't introduce any processing delay).

__________________________________________

Now, there are two things that continue to bother me in these discussions. First is the apparent belief that FIR filters are inherently "linear phase" and that they always introduce a processing delay. Neither is true. Filters have linear phase if they have a symmetrical impulse response, and hence are non-causal (and require a processing delay), be they FIR or IIR. (Homework problem: how do we get an IIR filter to have a symmetrical impulse response? Hint: it won't work for live sound :) )

Second is that compensating for smoothly lagging phase in an overall system response has a dramatic effect on sound quality. My belief is that you need a pretty carefully chosen test signal (try a 100 Hz triangle wave) for it to be audible. But I'm happy to be convinced otherwise.

On a more general subject, I think the real art in speaker tuning starts with observing what is relevant to various aspects of sound quality and what of that can be corrected electrically. Much as tuning the room sometimes requires a bulldozer, tuning the speaker often requires a table saw. Given that, what matters is taking the right measurements (spatial sampling rate), combining them in a certain way (averaging, throwing out outliers, weighting total isotropic power vs sound pressure at a point), and deciding what can be and needs to be corrected (smoothing, don't care regions). And as they say in medicine, "first, do no harm", or something like that.

Best,

--Frank

Logged
"Nature abhors a vacuum tube." -- John Pierce, Bell Labs

Peter Morris

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1467
Re: Are linear phase filters really the holy grail?
« Reply #23 on: July 02, 2017, 12:46:32 AM »

This is all well and good. But before we get too carried away it might be instructive to do some listening tests. Real, blinded, listening tests, to reveal just how audible smoothly varying phase is.

Smoothly lagging phase is what we're talking about as the rapidly varying phase, "phase grass" as I call it, is minimum phase, and is taken out by flattening the magnitude, which can be done with a variety of causal, minimum-phase filters, FIR or IIR. What FIR filters alone allow is compensating for lagging phase, as only a non-causal filter can do. We get around the causality problem by introducing a processing delay.

I propose we tune a system to our liking and then introduce a second-order all-pass into the signal chain (the whole signal, not some pass-band that will mess with the crossovers) at a variety of frequencies and see if anyone can hear a difference. We would not be the first to try this, by the way.

To be clear, I'm not bashing the use if FIR filters. Being able to manipulate the phase in both directions (at high frequencies at least) is useful for getting smooth crossovers and getting different speakers to play together. FIR filters are easier to synthesize automatically than banks of bi-quads, even if they are just used to implement (causal) minimum-phase filters (which don't introduce any processing delay).

__________________________________________

Now, there are two things that continue to bother me in these discussions. First is the apparent belief that FIR filters are inherently "linear phase" and that they always introduce a processing delay. Neither is true. Filters have linear phase if they have a symmetrical impulse response, and hence are non-causal (and require a processing delay), be they FIR or IIR. (Homework problem: how do we get an IIR filter to have a symmetrical impulse response? Hint: it won't work for live sound :) )

Second is that compensating for smoothly lagging phase in an overall system response has a dramatic effect on sound quality. My belief is that you need a pretty carefully chosen test signal (try a 100 Hz triangle wave) for it to be audible. But I'm happy to be convinced otherwise.

On a more general subject, I think the real art in speaker tuning starts with observing what is relevant to various aspects of sound quality and what of that can be corrected electrically. Much as tuning the room sometimes requires a bulldozer, tuning the speaker often requires a table saw. Given that, what matters is taking the right measurements (spatial sampling rate), combining them in a certain way (averaging, throwing out outliers, weighting total isotropic power vs sound pressure at a point), and deciding what can be and needs to be corrected (smoothing, don't care regions). And as they say in medicine, "first, do no harm", or something like that.

Best,

--Frank

Hi Frank,

Can you hear it … I have done some tests using the same speaker with “normal” IIR 24dB LR filters and then with matching linear phase FIR crossovers + some all-pass filters to fix a few small bumps making it flat from 300-400Hz up.

The conclusion I came to as I said above “Your highs, mids and lows won’t really sound any better (I hope this makes sense), but things will be clearer and more natural, more defined and separate in the mix.”  What they do is subtle, but noticeable.

I think part of the problem regarding tests to determine if we can hear phase distortion is has been using continuous tones.  Our hearing is extremely sensitive to first arrival times yet all this information is processed through a pseudo-resonant structure whose behaviour can be modelled with an auditory filter bank based on the gammatone function (as I noted above).

Without going into detail, I believe this means we can detect a lot more about the nature of random transient sounds than a continuous tone; after all this is what we were designed to do and one of the mechanisms that kept us safe from predators.
 
In our case (live sound) the critical thing we need to do is hear the human voice or voices within the mix as clear as possible …

http://www.audioholics.com/room-acoustics/human-hearing-phase-distortion-audibility-part-2

In this respect, dare I say, Dr Floyd Toole got it wrong to some extent … I’m with Lipshitz who said “Using pre-recorded music (male & female singing) fed through a 2nd-order all-pass network, with a Q of .5, audible effects were noted with a 95% confidence level. Using a variety of unpitched sounds recorded anechoically, phase effects were again audible.”

The other thing to note is most of these tests were done with 2nd order networks, while we typically use 4th order filters in our speakers, I believe this make things much more noticeable. I don’t think a smooth slowly varying phase is that noticeable.

A speaker designer can also do a lot more tricks with FIR processing than just flatting the phase. You can use extremely high crossover slopes to minimise radiation errors or limit effect of out band problems that your transducers may have, and you have much greater ability to finely correct amplitude response issues.
Logged

Peter Morris

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1467
Re: Are linear phase filters really the holy grail?
« Reply #24 on: July 02, 2017, 01:26:19 AM »

Is the amount of delay that the FIR filters add because of processing time? In other words if the DSP were to be faster will is have less propagation delay or have we reach the point of no return?

Also I have heard it said that a speaker system with FIR filters is more stable. For example if a pastor with a wireless mic were to be walking around out in front of the main cluster they are likely to hit spots that will feedback in a non-FIR filtered system, so it’s not consistent over the coverage area. And the same pastor walking around under the FIR filtered system any feedback points are likely to not change as much in relation to where they walk. Is this true?

If you what to use an FIR filter to implement a linear phase crossover it will take time. The time taken is proportional to the crossover frequency; the lower it is the longer you need.  If you had a perfect DSP with zero processing latency it would still take time because of the way the filters are implemented.

If you use an IIR or analogue crossover, the low frequencies are delayed. You can use an FIR filter to flatten the phase so that the HF delay matches the sub delay … and like the FIR crossover the time you end up delaying things is proportional to the crossover frequency, the lower the crossover frequency the more you have to delay the HF to match the initial delay caused by the crossover.

As a generalization, the speakers I have used with FIR filters do behave better in terms of feedback.  I think this is due in part to being able use the FIR filters to correct for amplitude and time domain issues better than ever before.  They can also be used to minimize radiation errors around the crossover frequencies.

http://dolby.invisionzone.com/index.php?app=core&module=attach&section=attach&attach_id=8 see pages 8 to 12
« Last Edit: July 02, 2017, 01:30:13 AM by Peter Morris »
Logged

Frank Koenig

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1318
  • Palo Alto, CA USA
Re: Are linear phase filters really the holy grail?
« Reply #25 on: July 02, 2017, 07:03:53 PM »

Hi Peter,

Responses inserted below. (Hope all the quotes work.)

Quote
Can you hear it …

In my limited (unscientific) experiments, for recorded music, no, not so far. Interestingly, on certain continuous test signals, such as the 100 Hz triangle wave I alluded to, yes. But just because I can't hear it doesn't mean that someone else can't. And I might hear it under different circumstances. There are many things to confound such tests, such as, I'm guessing, a small-room reverberant environment with ample early reflections.

Quote
Without going into detail, I believe this means we can detect a lot more about the nature of random transient sounds than a continuous tone; after all this is what we were designed to do and one of the mechanisms that kept us safe from predators.

Indeed. I remember hearing a talk in which research came up suggesting that there is an inverse relationship between distinct pitch perception (critical bandwidth) and time discrimination in various species. As I recall domestic cats have only about 5 critical bands but have good enough localization (time discrimination) to be able to pounce on a mouse in total darkness guided only by its sounds.

Quote
In our case (live sound) the critical thing we need to do is hear the human voice or voices within the mix as clear as possible …

Yes, agree. Get speech intelligibility right and most of the rest will follow. OK, not the boom-boom from the subs :)

Quote
The other thing to note is most of these tests were done with 2nd order networks, while we typically use 4th order filters in our speakers, I believe this make things much more noticeable. I don’t think a smooth slowly varying phase is that noticeable.

Yes, and speakers themselves, even if you ignore all the wiggles, are >> 4th order networks. So there is much to mess with.

Quote
A speaker designer can also do a lot more tricks with FIR processing than just flatting the phase. You can use extremely high crossover slopes to minimize radiation errors or limit effect of out band problems that your transducers may have, and you have much greater ability to finely correct amplitude response issues.

I've experimented with high-slope crossovers a little. There no doubt are situations where they work well but for the time being I'm getting along pretty well with 4th or lower order crossover filters as I'm working mostly with Danley boxes or coaxes where radiation errors (polar weirdness around the crossover) are less of a problem. (Not counting the order of any sub-band FIR filters which have an order equal to the number of taps - 1.)

This all started when I noted that I'd worked on FIR tunings for some Danley products, and their philosophy seems very much to be low slope, wide overlap when it comes to crossovers. And they know a lot more about their speakers than I know about any speakers.

Best,

--Frank
Logged
"Nature abhors a vacuum tube." -- John Pierce, Bell Labs

Mark Wilkinson

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1104
Re: Are linear phase filters really the holy grail?
« Reply #26 on: July 02, 2017, 09:24:27 PM »

Hi guys,

Can you hear it? 

My take, where i feel pretty secure.....
On well recorded music......When I use enough FIR to flatten phase all the way through the sub to main summation region (down to about 60Hz)...YES, I can hear it, and everyone I've asked to listen critically says they can too. 
It's simply clearer with a very real wow factor mono.  And stereo imaging gets much, much more solid. Wow again. 
Most of my critical listening/measuring has been using the PM60s, and by enough FIR, I mean 64ms  impulse centered, because that's what my minidsp units deliver. 
I can tell by sims, and then measure as well, that even with these 6144 taps, there is still a little slippage mag and phase, vs hoped for correction, once below 100Hz. 
I do not know if flat phase is really better than mildly sloping phase (ala all pass A/B comparisons, or for another comparison, vs meyer signature phase traces)...I kinda doubt it. 
I do know phase matching throughout x-over summation matters, if nothing more than for magnitude summation....heck, we all know that. 
IMO, linear flat phase just makes it easy and guarantees summation, even when we move x-over freq up or down.
I also know, that in trying to learn how to pare down delay, or chop taps, for live sound...the wow factor starts disappearing...dishearteningly so.
I was bumming last weekend at the reduction in SQ, as I lowered tap count/ changed processing, on the PM90s to get set up for a live show..so I gotta learn more here....

Which begs......Peter, I would love to know the tricks you've learned at phase alignment, while keeping delay minimized. 
I get IIR for x-over down low.  But I don't see how you get get any kind of brickwall filtering at any freq below up way high (say 6300 -:). 
And I need to get a better grip on how to use all pass filters....
It amazes me what you get done with a couple of ms......pls help !  :)
Logged

Peter Morris

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1467
Re: Are linear phase filters really the holy grail?
« Reply #27 on: July 03, 2017, 12:47:15 AM »

Hi guys,

Can you hear it? 

My take, where i feel pretty secure.....
On well recorded music......When I use enough FIR to flatten phase all the way through the sub to main summation region (down to about 60Hz)...YES, I can hear it, and everyone I've asked to listen critically says they can too. 
It's simply clearer with a very real wow factor mono.  And stereo imaging gets much, much more solid. Wow again. 
Most of my critical listening/measuring has been using the PM60s, and by enough FIR, I mean 64ms  impulse centered, because that's what my minidsp units deliver. 
I can tell by sims, and then measure as well, that even with these 6144 taps, there is still a little slippage mag and phase, vs hoped for correction, once below 100Hz. 
I do not know if flat phase is really better than mildly sloping phase (ala all pass A/B comparisons, or for another comparison, vs meyer signature phase traces)...I kinda doubt it. 
I do know phase matching throughout x-over summation matters, if nothing more than for magnitude summation....heck, we all know that. 
IMO, linear flat phase just makes it easy and guarantees summation, even when we move x-over freq up or down.
I also know, that in trying to learn how to pare down delay, or chop taps, for live sound...the wow factor starts disappearing...dishearteningly so.
I was bumming last weekend at the reduction in SQ, as I lowered tap count/ changed processing, on the PM90s to get set up for a live show..so I gotta learn more here....

Which begs......Peter, I would love to know the tricks you've learned at phase alignment, while keeping delay minimized. 
I get IIR for x-over down low.  But I don't see how you get get any kind of brickwall filtering at any freq below up way high (say 6300 -:). 
And I need to get a better grip on how to use all pass filters....
It amazes me what you get done with a couple of ms......pls help !  :)

Hi Mark,

With the Lake set to a FIR processing time of 2.5ms you can have a 24 or 48 dB/oct LR shaped linear phase crossover down to 500Hz. At 6K you can have a brick wall – ish 93 dB/oct.  At 3.5ms I can achieve 80dB/oct at 1.5KHz.

The Lake does not let you manipulate the phase separately nor enter your own tap coefficients.  To flatten the phase you need to use all-pass filters … it’s tricky but you can make it work.

If you can ensure your drivers are behaving well out side of their operating band there is often no need or advantage using brick wall filters.  On the boxes you mentioned, the PM60 & 90 I used 24dB linear phase crossover filters on the mid-range but as steep as I could get on the VHF because of the way the VHF driver has been designed and behaves.
Logged

Merlijn van Veen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 311
    • www.merlijnvanveen.nl
Are linear phase filters really the holy grail?
« Reply #28 on: July 03, 2017, 02:02:06 AM »

It's a damn shame though, because for playback only, I use 6144 taps (64ms delay), and get very nice flat phase through sub range, which does tighten up the sound of bass.
It kinda make the bass seem less, until you realize it's just cleaner. And then you get to crank it a bit more, with better dynamics, and an overall clearer sound.

This is something my students and I conclude unanimous in class time  and again.

According to Coda Audio, it's also a reoccurring "complaint" with respect to their sensor controlled subwoofers featuring less phase shift due to lack of an electronic HPF for speaker protection.

I'm starting to suspect that the audibility of phase distortion is somehow tied to harmonic distortion. Distorted sound is typically perceived as being louder over clean sound.

A lot of research regarding phase distortion has been done using square waves or transient signals in general. Phase distortion regarding the former, manifests itself as spectral or timbral change, altering the sequence of the harmonics (at least in my mind).

In general there's good consensus that less phase distortion sounds clearer. All loudspeakers ultimately distort (often excursion related), audibly or inaudible. Regardless, this is evidently the moment where sine waves become square waves. Systems with more phase distortion (relatively speaking) alter the sequence of harmonics which is perceived as spectral or timbral change making them sound less clear.

I surmise auditory masking (including temporal) plays an important part here, as well as the perception of loudness over time (integration).



Verzonden vanaf mijn iPad met Tapatalk
« Last Edit: July 03, 2017, 02:04:18 AM by Merlijn van Veen »
Logged

Peter Morris

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1467
Re: Are linear phase filters really the holy grail?
« Reply #29 on: July 03, 2017, 05:20:40 AM »

This is something my students and I conclude unanimous in class time  and again.

According to Coda Audio, it's also a reoccurring "complaint" with respect to their sensor controlled subwoofers featuring less phase shift due to lack of an electronic HPF for speaker protection.

I'm starting to suspect that the audibility of phase distortion is somehow tied to harmonic distortion. Distorted sound is typically perceived as being louder over clean sound.

A lot of research regarding phase distortion has been done using square waves or transient signals in general. Phase distortion regarding the former, manifests itself as spectral or timbral change, altering the sequence of the harmonics (at least in my mind).

In general there's good consensus that less phase distortion sounds clearer. All loudspeakers ultimately distort (often excursion related), audibly or inaudible. Regardless, this is evidently the moment where sine waves become square waves. Systems with more phase distortion (relatively speaking) alter the sequence of harmonics which is perceived as spectral or timbral change making them sound less clear.

I surmise auditory masking (including temporal) plays an important part here, as well as the perception of loudness over time (integration).



Verzonden vanaf mijn iPad met Tapatalk

Hi Merlijn,

I have had a similar experience with one of my designs where I have pushed the flat phase response to what I consider a practical limit.  Linear phase crossovers on all pass bands and no HPF on the sub (you just have to be careful selecting your kick drum mic so it does not drive the subs too low). The sub is a ported design but this and its natural roll off are the only things that impact the phase response. I have managed to contain the total latency to about 15ms so the system is still usable for FOH applications. Despite the sub being a double 21” people are amazed at how tight and punchy the low frequencies are.

(must check out Coda's sub)

I total agree about changing the sequence of harmonics. If we consider several people signing together; if we can keep the timing of the harmonics that make out each voice linear then I believe our hearing / brain is able to determine much more easily which harmonics are associated with which fundamental – this results in the sound becoming clearer and each voice is separated in the mix … and the music becomes more enjoyable because we don’t have to concentrate so hard.
« Last Edit: July 03, 2017, 05:28:19 AM by Peter Morris »
Logged

ProSoundWeb Community

Re: Are linear phase filters really the holy grail?
« Reply #29 on: July 03, 2017, 05:20:40 AM »


Pages: 1 2 [3] 4 5   Go Up
 



Site Hosted By Ashdown Technologies, Inc.

Page created in 0.033 seconds with 22 queries.