Hi-def DSP

[Chris A]

1 hour ago, gnarly said:

I've seen in the Xililca XP/XD Manuals, they speak of 'Presets within the Device Only'.  I'm hoping/curious  this is along the same lines as Q-Sys Snapshots, with similar broad capabilities in what can be altered on the fly.  Do you use them much?

 

I use presets quite a bit, mostly now for supporting others on this forum--helping them dial in their setups using REW and XConsole (or the equivalent miniDSP application).  I've lost count on how many guys that I've helped thus far--it must be 50 people...all using IIR filters.

 

Once I get my setup dialed-in, I pretty much leave the Xilica preset alone until something in the setup changes.  But I used to do A-B comparisons before discovering how to remove all the phase growth from the crossover bands without the use of FIR filters.

 

Chris

[gnarly]

Awesome help to folks,  like you've been doing.

I wish i wasn't such a slow writer and could try to help more too....

(i ducked every writing class possible in undergrad and grad,.....to my current chagrin ...)

 

Anyway.... back to  to audio.....

I can not stop with A-B comparisons, on what i think you call phase growth. They keep redefining what I think i know/hear.

 

Are you simply talking phase lag, and re-coining it into 'phase growth' , or are there subtleties I'm missing with your term phase growth?

 

I keep playing with phase rather intensely.

I'll post some of my typical measurement results, and  the tuning techniques that have given best results, on the thread you linked...for critique.

Seems like they would fit better there.

I'll try to stay away from subjective results.....cause we all know well they work  😅

[Chris A]

12 hours ago, gnarly said:

Are you simply talking phase lag, and re-coining it into 'phase growth' , or are there subtleties I'm missing with your term phase growth?

The "phase growth" I talk about is the crossover-filter-induced excess phase, which always causes the lower frequency drivers to lag.  I'm not really talking about the minimum phase phase changes due to the drivers/horns themselves.  We've already discussed this with Greg B. (Edgar) on what the ear is expecting to hear, and I've settled on minimum phase--at least for the time being. 

 

When I get the computing horsepower in place to do significant amounts of low frequency phase correction (FIR filtering), I plan on spending more time on that, but it has to be within the confines of what the AVP can handle in terms of overall delay to sync the video with audio, i.e., I can't currently add, for example, another 50 ms to the video delay due to the addition of FIR filtering at low frequencies.  The AVP can't handle that.

 

12 hours ago, gnarly said:

I'll try to stay away from subjective results.....cause we all know well they work  😅

Well, let's not throw the baby out with the bathwater...

 

I did an entire thread on the subconscious effects of phase flattening and this has been one of the biggest revelations in audio that I've ever experienced. I believe that the phase growth problems that are typically seen with passive crossover fully horn-loaded loudspeakers are the source of the deal killer for so many "golden ear" audiophiles that reject horns.  The little monitors on a stick that so many people like--all of them have essentially minimal excess phase, and that's why I believe they are so popular.  Danley's Hyperion takes that effect to the next level (...reportedly...looking at his spec sheet for the Hyperion and its totally flat phase response). I think that's what the "reviewers" are hearing, including extremely low compression and modulation distortion, and full-range controlled directivity.

 

Once you get better horn profiles (full-range fully horn loaded loudspeakers, in this case), do a little in-room early reflection absorption around the loudspeakers, and then correct for the excess phase due to the crossovers and time mismatch of horns/drivers...from a subjective point of view, pure magic pours out the the loudspeakers/room once these issues are controlled.  I'm still astonished that no one has really found the same thing--or at least talked about it online. 

 

You really can't talk about this subject without bringing in the subjective aspects of the phenomenon and the magic of what you hear with your best naturally recorded acoustic music (i.e., no multitrack layering of music on top of each other and artificial mixing on a console after tracks are recorded, etc. that destroys the phase relationships that exist in real life).  How this pulled me into music that I had avoided most of my life (19th century orchestral and other acoustic recordings) is the big surprise. Once I put on a good recording of this genre (especially multichannel recordings), I can't walk away now that the excess phase and early reflections are all controlled.

 

Chris

[gnarly]

Ok, very cool....maybe this is the right thread to continue on with phase some.

I very much share your views and subjective comments about phase audibility. It's a very narrow tightrope to tune to, but when right, for me it's like washing a windshield (to use a visual analogy).

 

Phase audibility is what has driven me to find the best processor i can, to be able to both implement linear-phase and then easily A-B it against traditional IIR.

Q-sys was the first affordable platform (used) i found that allows instant A-B, FIR vs IIR.

 

In case folks are interested in the Q-sys Cores' FIR capability, I've found a Core110f can run 8 channels of 4k taps per channel (@48kHz), on stereo 4-ways.

Different tap allocations per band are of course possible, but I haven't been able to put more than 6k on the sub FIRs. Due to that, i recently acquired an older Core500i, that allows 16k taps per channel on at least 16 channels.  16k taps is the current Q-sys maximum, raised from 8k a year of so ago...

 

Of course like you say, such latencies are a no-go for home theater without video delays. 

Do such long video delays exist....?  I have no clue as I'm audio only, (and use computer logitech speakers for video. lol)

 

Anyway, I feel I've had super success with phase linearization, both audibly and with measurements. 

I've used it on a number of different type 4-way builds, always on a driver by driver tuning approach, where each driver gets it's own FIR file and amplification channel.

 

Here's a spectrograph example from a flat-phase MEH build, that uses a CD straight to 10"s.  (I used the spectrograph settings you showed in the linked phase flattening thread.)

It's steeply high passed at 100Hz. 

 

Here is a dual 18"s push-push slot loaded sub's transfer function, that I'm using with the MEH.

It was taken outdoors at 4m. (I don't bother with indoor sub measurements.)

Green is raw, blue processed.

It was with 6k taps.  Looking forward to dragging it back onto the driveway and retuning it with Core500i.

I think the processed phase trace is already pretty good, but 16k taps will raise frequency resolution to a little better than 6Hz (from the 6k's resolution of 16Hz), to maybe gain some bass definition....who knows 🙂

 

 

 

 

 

 

[Edgar]

48 minutes ago, gnarly said:

Phase audibility is what has driven me to find the best processor i can, to be able to both implement linear-phase and then easily A-B it against traditional IIR.

 

My big concern about linear phase is pre-ring. An unavoidable consequence of linear-phase filtering is that the time-domain pre-ring and post-ring are mirror-images of each other. So whatever ringing occurs after the main pulse, also occurs before the main pulse. It may be arguable as to whether this is audible at high frequencies, but at low frequencies it is likely to be, due to the long durations involved.

 

The compromise that I proposed to @Chris A was minimum-phase, or at least group delay that is monotonically decreasing with increasing frequency. Classic highpass filters, such as Butterworth or Bessel, exhibit this characteristic. Chebyshev and Elliptic filters typically do not. Chris' counter-proposal is elimination, or at least minimization, of excess phase (right half-plane zeroes), and I consider that to be an attractive alternative explanation.

 

Quote

Of course like you say, such latencies are a no-go for home theater without video delays. 

Do such long video delays exist....? 

 

I have been looking for a high definition video delay, but have not found anything that costs less than about a year's salary. 1920x1080x3byte video needs ~6MB per frame, and each frame is ~17 milliseconds at 60 Hz. 4K 60Hz video quadruples that storage requirement.

[gnarly]

34 minutes ago, Edgar said:

 

My big concern about linear phase is pre-ring. An unavoidable consequence of linear-phase filtering is that the time-domain pre-ring and post-ring are mirror-images of each other. So whatever ringing occurs after the main pulse, also occurs before the main pulse. It may be arguable as to whether this is audible at high frequencies, but at low frequencies it is likely to be, due to the long durations involved.

 

Yep, been my concern too. 

I've come to believe complementary linear-phase xovers cancel each sides preringing, and are completely safe to use. 

So even for the low frequencies, like for a sub-to-main xover at 100Hz, i feel comfortable with a linear-phase xover.

 

Where i don't have comfort with linearizing-phase, are on the spectrum ends, 20Hz and 20KHz.  Because there is no complementary offset.

So I've been using IIR for sub high-pass and 20kHz+ low-pass (which for me has helped clean up any tiny digital impulse oscillations)

 

I also don't feel comfortable linearizing a driver's phase from its natural frequency response rolloff.

There, I like to flatten the driver's out of band response with minimum phase EQs, hopefully to the point that the added electrical xover order becomes the driver's measured acoustic order.

That's a big part of the reason I like steep linear-phase xovers....it makes the out-of-band flattening much easier (and protects against lower-end over excursion from the min-phase flattening).

Steep also has helped me minimize the width of lobing region, making it easier to measure smoother polars thru xover region.

And i think that thereby also minimizes the freq range where complementary linear-phase xovers might not fully cancel pre-ringing acoustically.

 

Any comments / concerns welcomed....🙂

34 minutes ago, Edgar said:

 

 

 

I have been looking for a high definition video delay, but have not found anything that costs less than about a year's salary. 1920x1080x3byte video needs ~6MB per frame, and each frame is ~17 milliseconds at 60 Hz. 4K 60Hz video quadruples that storage requirement.

Gotcha. 

Like said, I'm clueless about video.

[gnarly]

On 6/26/2021 at 11:20 AM, gnarly said:

 

 

 

I also don't feel comfortable linearizing a driver's phase from its natural frequency response rolloff.

There, I like to flatten the driver's out of band response with minimum phase EQs, hopefully to the point that the added electrical xover order becomes the driver's measured acoustic order.

That's a big part of the reason I like steep linear-phase xovers....it makes the out-of-band flattening much easier (and protects against lower-end over excursion from the min-phase flattening).

 

Hi all, apologies for quoting my own post, but i wanted to build on it and try to get some feedback.

 

I want to compare the technique of out-of-band fattening combined with complementary 'named crossovers', to building 'un-named' xovers using combinations of parametric EQ's, shelving EQs, perhaps along with high or low pass filters, etc.

Because i believe the two techniques give identical "total filter" results, for a given acoustic order/slope target.

By "total filter" results, I mean a transfer function of the whole electrical filter package of each approach.

The out-of-band flattening when combined with a named xover, becomes the actual xover...and as as such, the combination formed also can't be named.

That combo should look just like the "un-named filter approach", again assuming they both produce the same desired acoustic target.

 

Imo, the really cool things about this equivalency are:

that it lets us use the named xovers in our processors,

and i find it is much easier to EQ response up to a horizontal flat line, than it is to try to nudge one curve on top another (which is what we have to do when using combinations of parametric EQ's, shelving EQs, etc.) The vertical distances between curves can be quite deceiving. But flat against flat is easy to judge.

 

Anyway, this technique works so well for me, i keep trying to share it.... (credit for it goes to POS of rePhase)

 

 

 

[Edgar]

6 minutes ago, gnarly said:

 

I want to compare the technique of out-of-band fattening combined with complementary 'named crossovers', to building 'un-named' xovers using combinations of parametric EQ's, shelving EQs, perhaps along with high or low pass filters, etc.

Because i believe the two techniques give identical "total filter" results, for a given acoustic order/slope target.

By "total filter" results, I mean a transfer function of the whole electrical filter package of each approach.

 

 

There's a lot of hand-waving in your description, @gnarly, but there doesn't have to be. There are various ways to determine approximate transfer functions of drivers from mag/phase measurements. Once you have their transfer functions, you can mathematically change them to whatever you want (within reason) by means of inverse filtering (for minimum-phase systems), pole-zero cancellation (e.g., Linkwitz Transform), and other means. No need to use Kentucky Windage to try to get the transfer function "just right" -- the math will do it for you.

[gnarly]

On 6/21/2021 at 5:18 PM, DirtyErnie said:

Can anyone please point out a DSP that runs at more than 48KHZ?

Ideal is 24 bit, 96KHz or greater.

 

Thanks!

 

Hey, here's one that might meet your requirements....i forgot about it in my zeal for Open Architecture stuff.

Linea-Research ASC-48. (Or also available rebadged as a Danley SC-48)  https://linea-research.co.uk/asc48/

4 in-8 out. analog/AES3. 96kHz.  High spec / super clean device. (I have the Danley version.)

 

One of its really cool features is what it calls LIR xovers, which are essentialy 24dB/oct linear-phase Linkwitz-Riley's.

What makes them cool, is that the LIR xover frequencies can be dialed in on the fly, just like any usual IIR xover.....

AND the processor automatically adjusts internal delays to keep phase alignments spot on

I haven't seen that anywhere else....

[gnarly]

1 hour ago, Edgar said:

 

There's a lot of hand-waving in your description, @gnarly, but there doesn't have to be. There are various ways to determine approximate transfer functions of drivers from mag/phase measurements. Once you have their transfer functions, you can mathematically change them to whatever you want (within reason) by means of inverse filtering (for minimum-phase systems), pole-zero cancellation (e.g., Linkwitz Transform), and other means. No need to use Kentucky Windage to try to get the transfer function "just right" -- the math will do it for you.

 

Sure, there is definitely alot of hand waving.  !!!   Here's some more !!! 😄

For me, the fact that the transform can be done mathematically in a number of different ways is almost entirely  immaterial to choosing a processor,

and figuring out exactly how I'm going to use that processor to implement (what the math says can be done).

The math alone isn't going to do any processing is it ? 

 

 

 

 

 

[Edgar]

9 minutes ago, gnarly said:

 

For me, the fact that the transform can be done mathematically in a number of different ways is almost entirely  immaterial to choosing a processor,

and figuring out exactly how I'm going to use that processor to implement (what the math says can be done).

The match alone isn't going to do any processing is it ? 

 

 

If you are unable to specify the filter coefficients in your processors' user interface, then you are lost.

 

Other than that, what matters is floating-point vs. fixed-point (not many fixed-point processors around any more), and processing power.

[gnarly]

50 minutes ago, Edgar said:

 

If you are unable to specify the filter coefficients in your processors' user interface, then you are lost.

 

If you mean being unable to specify bi-quads and such for IIR work , then I'll prefer to stay lost. 😂

If you mean unable to specify FIR coefficients for a FIR processor, i of course obviously agree.

 

 

 

50 minutes ago, Edgar said:

Other than that, what matters is floating-point vs. fixed-point (not many fixed-point processors around any more), and processing power.

 

I take 24-32 bit floating point, 48-96kHz, with sufficient processing, as a given.

 

So for me, I/O channel counts, I/O input and output types and their type connectors, are the first screening factors.

Then comes whether I want FIR, and how much.

Then comes the breadth and depth and counts of IIR filter types, and orders. 

Precision of delays, gains, limiters if needed...etc. etc

Iow, What can it implement quickly and precisely.

Ease of use is an ever present factor....  whether setup/controlling from software, or from the unit's front panel.

 

 

 

[Supersteff]

Hi Mark

 

On 6/26/2021 at 5:20 PM, gnarly said:

I also don't feel comfortable linearizing a driver's phase from its natural frequency response rolloff.

There, I like to flatten the driver's out of band response with minimum phase EQs, hopefully to the point that the added electrical xover order becomes the driver's measured acoustic order.

 

I have been wondering , how far out of band do you have to flatten? If we consider a two-way MEH with a crossover-frequency of say 500 Hz between woofer and CD (I can´t remember how low you cross in your synergy´s?), up to what frequency do you flatten the woofers? Also how far down for the CD? Is it one octave or is half an octave enough?  I wonder about the 1/4 wave cancelation-notch, how do you get around that, or do you need much overlap for that reason? just curious.

 

Regards

 

Steffen

[gnarly]

20 hours ago, Supersteff said:

 

I have been wondering , how far out of band do you have to flatten? If we consider a two-way MEH with a crossover-frequency of say 500 Hz between woofer and CD (I can´t remember how low you cross in your synergy´s?), up to what frequency do you flatten the woofers? Also how far down for the CD? Is it one octave or is half an octave enough?  I wonder about the 1/4 wave cancelation-notch, how do you get around that, or do you need much overlap for that reason? just curious.

 

Hi Steffen,

 

How far we want flatten to is set by the driver's acoustic target order we want to achieve, and how deep we want to match the target.

 

For example, say the acoustic target is LR 24 dB/oct.  And say the goal is hold-to-target down to -30dB.  

Here's a 500Hz LR4 lowpass.  As you can see, flattening to -30dB would need to go out to about 1200Hz.

 

 

Here's the acoustic target of 500Hz with a much steeper LR96 dB/oct, and goal is the same hold-to-target to -30dB.  Flattening to -30dB needs to reach  only about 620Hz.

 

 

Ok...how far can we actually flatten.   The raw measurements speak. 

Here's the raw woofer/mid section of syn7, a pair of 10 inchers that reach up to the CD. (with round ports in horn centers)

 

 

 

I think you can see that flattening response up to about 700Hz is pretty easy.

 

That out-of-band minimum phase frequency flattening to 700Hz, also takes care of the section's out-of-band rising phase (to 700Hz)

The in-band minimum phase frequency flattening (which almost everyone does), takes care of in-band phase flattening.

So when a steep linear phase xover is applied on top of the all the minimum phase flattening, the result is flat phase across the driver's bandwidth  down to -30dB.

 

When the same flattening process is applied to the CD's low end, and uses a complementary linear phase high pass  (500Hz LR 96 dB/oct), the result is very flat frequency and phase response across thru xover region (and across the board).

 

 

 

Hopefully,  our woofer/mid ports are located where the 1/4 wave notch frequency is well above the -30dB down frequency, which makes the notch mostly immaterial after the steep lowpass.

 

When I try a shallower acoustic target for the woofer/mids like the LR4 first pictured above, it ends up taking alot of parametric work,along with shelving filters to flatten out to -30dB (or about 1.2kHz).  I always end up saying screw this, this is nuts......cause steep works 😁

 

 

I've used this technique on everything from MEH's, to co-axials,  to PA traps, to line arrays, and all of them with subs (crossed to with same technique). 

Iow, anywhere there's a xover ..Lol

 

Always a cost in audio though, huh ......Price is FIR latency, and need/expense/complexity of multi-channel processing / amplification.

 

 

[Supersteff]

Hi Mark

 

That makes sense, and you seem to have succes with keeping pre-ringing-effects low/"inaudible" with 

 

15 minutes ago, gnarly said:

the goal is hold-to-target down to -30dB

 

Would that in a way imply, that your tuning technique kind of fixes/moves pre-ringing-effects to sub -30 dB, when the HP- and LP-slopes start to differ from the ideal curves?

 

I don´t know how pre-ringing sounds, but i guess its like muddy waters or something. So the lover the "hold-to-target"-requirement, the "clearer" the sound (waters)? Maybe this is obvious, I just try to picture it for my self!?

 

Thanks for your explanation and pictures.

[gnarly]

1 hour ago, Supersteff said:

Hi Mark

 

That makes sense, and you seem to have succes with keeping pre-ringing-effects low/"inaudible" with 

 

 

Great, glad that made sense.

 

Quote

 

Would that in a way imply, that your tuning technique kind of fixes/moves pre-ringing-effects to sub -30 dB, when the HP- and LP-slopes start to differ from the ideal curves?

 

I don't think differing from ideal curves matters too much to potential pre-ringing.  As long as the difference is relatively minor, and not near the central xover frequency.  

With complementary lin-phase xovers , if mag and phase look good thru xover region, pre-ringing is simply not worth thinking about. 

-30dB is my standard because I've found it easy enough to do...probably overkill in terms of ring.

 

What I think matters more to potential pre-ringing, is how complementary is the acoustic coupling both on and off-axis. 

The idea that complimentary linear-phase xovers cancel pre-ringing, where the driver on the low pass side negates  the driver on the high pass side, depends on acoustic symmetry.

So good old geometric lobing rears it's head again.

 

Which again gets back to why I like steep xovers....as they minimize the region of potential lobing. 

 

My take on pre-ringing is it's mostly the same type of audio overconcern/overthinking, as with so many other audio things buried deep under the noise floor.  Don't mean to offend anyone, but i believe in ranking marginal impacts, maximizing SNR, which sometimes means living with a little more noise i can't hear to gain a whole lot more signal i can hear.

 

That said, i do think many of the experiments where folks have tried to linearize phase globally on top of existing passive or active designs, have had the capability of bringing pre-ringing up to the surface.. 

The success of a global phase correction overlay depends on how well the speaker had already achieved acoustically complementary xovers, and also level in-band mag and phase response for each of its drivers.   And again, for both on and off-axis.

Otherwise the global correction will be valid only to the mic location.....and complementary ringing cancellation could be sucking wind everywhere else.

 

 

[gnarly]

5 hours ago, Supersteff said:

 

I don´t know how pre-ringing sounds, but i guess its like muddy waters or something. So the lover the "hold-to-target"-requirement, the "clearer" the sound (waters)? Maybe this is obvious, I just try to picture it for my self!?

 

 

Hi again Steffen, 

i think you've asked the real question....and i kinda ducked it my previous post.

 

"What does pre-ringing sound like?"

I truly wish somebody would nail it down ! 😉

 

The only time i've ever thought i heard pre-ringing, was when i was doing some super long FIR files  (64k taps) using j-river convolution, with the same long FIR file size on every output channel.  Kinda sounded like a pre-echo maybe, when there was silence before transients.  But that experimentation was so short lived, without any real confirmation, it's almost stupid of me to mention it.

 

The thing with FIR in my mind, is that it is a VERY powerful tool.....which requires VERY judicious use.

FIR questions me, "I don't give a shite what you want to fix.  Do you really know what to fix and why?"

Pre-ringing only happens when we can't answer FIR's question,  i think.....

 

Being a lover of "hold-to-target down to -30dB" may be as much BS as pre-ringing 😁

[Chris A]

 

[gnarly]

Yes,  I've seen and heard many of those studio type presentations regarding pre-ringing.

They all have a  simple common trait, imo..... they all use a very high Q electrical filter supposedly linearized (the idea of which is nonsense)  ..... carefully constructed to prove their point.

 

Fabfilter is one of the worst....pure marketing on a fish hook. Lol 

 

Bogus amogus big time! 🙂 imho.

 

 

 

 

[Chris A]

It really needs to be a lossless audio file - not AAC, which is what you get with YouTube--involuntarily.  To hear the difference in YouTube AAC, the pre-ringing has to be emphasized, as it is above.

 

At least you can hear what you're listening for.  The issue is, how much pre-ringing is audible on your setup/room and how sensitive are you to non-crisp impulses.  In my listening to the above YouTube video, it was really dramatic, but then it's based on an overemphasized (big bass PEQ) to hear it, and the base case (no extra processing except AAC that YouTube applies) sounded okay - but not spectacular.

 

A little more discussion of the effects of AAC and MP3 processing on impulse response here...(and, no, I'm sure the linked thread is correct): https://www.audiosciencereview.com/forum/index.php?threads/the-effects-of-lossy-encoding-on-phase-and-impulse-response.9701/

 

Chris