Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Doug Fowler]

something similar to what Clair is doing and what VUE is doing in terms of applying array processing

AFMG FIRmaker

http://firmaker.afmg.eu/

[Lee Buckalew]

AFMG FIRmaker

http://firmaker.afmg.eu/

Exactly

Lee

[Mark Wilkinson]

Hi Lee,

I just took a quick look at the plots posted in the link you provided and though coverage looked relatively even, it appears that (according to the scale provided), hf is significantly down at about 14khz, some 20db and nosedives above that, pretty much in line with what Ivan was talking about earlier.  Whilst I have had a subjectively good experience hearing MLA, it still looks like there are many of the issues that all systems suffer from due to interaction from multiple drivers in multiple boxes, though I would concede these would be likely much worse without the processing power MLA offers.

I just can't see how DSP processing can do fine optimization for multiple drivers at 14Khz and higher....given current processing speeds.

At 14khz frequency, one cycle takes 1/14,000 second, or about 0.07ms.  So 0.07ms for 360 degrees.
A 98khz dsp takes 1/98,000 second for one sample, or about 0.01ms.  So 0.01ms is the smallest time adjustment the dsp can make.
Doesn't that mean that the smallest adjustment that a 98khz dsp can make at 14khz frequency,
is 0.01/0.07 * 360 or 51 degrees ?

That seems pretty coarse, and obviously gets worse as freq climbs.
48khz processing looks like it would almost be worthless for aligning multiple VHF drivers.

But I guess the error between any two adjacent VHF drivers, each individually driven, would be +/- one half the minimum adjustment... so maybe summation isn't too bad...at 14khz and 98k processing...
It's just hard to see how this slippage fits together for more than a couple drivers..

[Lee Buckalew]

I just can't see how DSP processing can do fine optimization for multiple drivers at 14Khz and higher....given current processing speeds.

At 14khz frequency, one cycle takes 1/14,000 second, or about 0.07ms.  So 0.07ms for 360 degrees.
A 98khz dsp takes 1/98,000 second for one sample, or about 0.01ms.  So 0.01ms is the smallest time adjustment the dsp can make.
Doesn't that mean that the smallest adjustment that a 98khz dsp can make at 14khz frequency,
is 0.01/0.07 * 360 or 51 degrees ?

That seems pretty coarse, and obviously gets worse as freq climbs.
48khz processing looks like it would almost be worthless for aligning multiple VHF drivers.

But I guess the error between any two adjacent VHF drivers, each individually driven, would be +/- one half the minimum adjustment... so maybe summation isn't too bad...at 14khz and 98k processing...
It's just hard to see how this slippage fits together for more than a couple drivers..

Phase correction is not about time delay as you are thinking of it, it is about phase correction via the proper filter selection and design.  With FIR processing delay is created to allow for time to process the required coefficients. 

Here is some info from the QSYS information page on creating custom FIR in QSYS.  A FIR coefficient includes; magnitude (20Hz - 20kHz, -20dB to 20dB), magnitude/phase (same magnitude adjustment but adds phase from -180 degrees to 180 degrees) , impulse (amplitude vs. time from 0.00ms to 21.3ms)

The larger problem for FIR filters is LF control, not HF control.  The lower the frequency being affected by an FIR filter the longer the processing takes so you get greater processing delay.  FIR is not always the correct filter choice depending on what you are trying to accomplish.

Lee

[Tom Danley]

Phase correction is not about time delay as you are thinking of it, it is about phase correction via the proper filter selection and design.  With FIR processing delay is created to allow for time to process the required coefficients. 

Here is some info from the QSYS information page on creating custom FIR in QSYS.  A FIR coefficient includes; magnitude (20Hz - 20kHz, -20dB to 20dB), magnitude/phase (same magnitude adjustment but adds phase from -180 degrees to 180 degrees) , impulse (amplitude vs. time from 0.00ms to 21.3ms)

The larger problem for FIR filters is LF control, not HF control.  The lower the frequency being affected by an FIR filter the longer the processing takes so you get greater processing delay.  FIR is not always the correct filter choice depending on what you are trying to accomplish.

Lee

"The larger problem for FIR filters is LF control, not HF control.  The lower the frequency being affected by an FIR filter the longer the processing takes so you get greater processing delay."

This part is correct but the reason is that since there is no such thing as inverse delay (read ahead processing), to correct the phase at low frequencies, everything else must be delayed back to the lowest frequency / phase one is correcting. The processing latency is added to this FIR imposed correction delay for a total processing delay
Tom

[Mark Wilkinson]

Phase correction is not about time delay as you are thinking of it, it is about phase correction via the proper filter selection and design.  With FIR processing delay is created to allow for time to process the required coefficients. 

Here is some info from the QSYS information page on creating custom FIR in QSYS.  A FIR coefficient includes; magnitude (20Hz - 20kHz, -20dB to 20dB), magnitude/phase (same magnitude adjustment but adds phase from -180 degrees to 180 degrees) , impulse (amplitude vs. time from 0.00ms to 21.3ms)

The larger problem for FIR filters is LF control, not HF control.  The lower the frequency being affected by an FIR filter the longer the processing takes so you get greater processing delay.  FIR is not always the correct filter choice depending on what you are trying to accomplish.

Lee

Thanks Lee,

I've been doing a lot of FIR experimentation, with a lot of taps available...6000+ per channel, for a 4 way system...
Yes, as you say.....I get the the issues down low, and think they definitely extend into lower mid as well...

What has surprised me, and I what i don't see mentioned much on the web, are the HF/VHF issues I've experienced.
Maybe the issues don't exist and i am just in the state of self-learning that always has the huge learning holes.....good chance...

but as i try to tune HF to VHF that crosses at about 6300, I find that this has been more problematic than expected.

I know that sampling freq determines the smallest time alignments possible.
Past time alignment I am left with trying to fine tune and compensate phase alignment via FIR guesswork.  I say guess work, because I've found that predicted flat phase via FIR generation,  and then measured flat phase, seems questionable at VHF.
I'm even wondering how good  our dual FFt's are at measuring VHF given their 48k sampling...

Maybe, this is where the engineers are doing what I can't yet see how to do...adjust minute phase at VHF via the FIR files they generate (or IIR embedded in FIR).  Either way the same.   
Dunno ...

But I do know enough to get that FIR correction has difficulty  down low (due processing time)...and I currently believe it has difficulty up high too (due to processing speed).
 
So it makes me wonder about claims i see...
Having said that,  I totally think the MLA processing tech is awesome.
Hell, I want to buy a big "multi-driver horn-loaded point-source" and try my hand with active FIR... to see if it can be tuned even further 

[Tom Danley]

Hi Mark
If you’re having problems at high frequencies you might be running into something else.

The magic of FIR filters is that it separates adjustment of the magnitude from the phase where with a circuitry based conventional equalizer any change in the magnitude also changes phase following the ”minimum phase” relationship. 

While transducers are mostly minimum phase what happens as the sound radiates may not be.

The issue is that while you can correct many things about a loudspeaker, the tough part up high becomes basing that correction on the “right thing”. 
By that I mean you can correct the output which is tied to the voice coil but not spatial issues forward of that.

The problem up high is often the device does not radiate or have a pattern like a simple expanding sphere and could be shaped more like a sea urchin and so, one must take a number of measurements across the front and average them to get to the correctable portion of the radiation and NOT the spatially variant part.     
If you don’t do that, you might be chasing your tail correcting spatial issues related to the radiation shape and not the transducer.   
6300Hz is high enough where you might be dealing with the interference patterns radiation shape with two drivers operating at the same frequency. They need to be within about a half inch apart edge to edge to radiate as a single new source and at an inch or more are independent sources.

As for the "multi-driver horn-loaded point-source", what I have done is used a physical offset instead of signal time delay and use adapted shape crossovers to do most of the crossover phase correcting but one can go farther with FIR filters if you have an active system.
A low budget but good way to play is using mini-dsp and Re-phase

https://www.minidsp.com/

https://www.minidsp.com/applications/advanced-tools/rephase-fir-tool

Best
Tom Danley

[Lee Buckalew]

Thanks Lee,

I've been doing a lot of FIR experimentation, with a lot of taps available...6000+ per channel, for a 4 way system...
Yes, as you say.....I get the the issues down low, and think they definitely extend into lower mid as well...

What has surprised me, and I what i don't see mentioned much on the web, are the HF/VHF issues I've experienced.
Maybe the issues don't exist and i am just in the state of self-learning that always has the huge learning holes.....good chance...

but as i try to tune HF to VHF that crosses at about 6300, I find that this has been more problematic than expected.

I know that sampling freq determines the smallest time alignments possible.
Past time alignment I am left with trying to fine tune and compensate phase alignment via FIR guesswork.  I say guess work, because I've found that predicted flat phase via FIR generation,  and then measured flat phase, seems questionable at VHF.
I'm even wondering how good  our dual FFt's are at measuring VHF given their 48k sampling...

Maybe, this is where the engineers are doing what I can't yet see how to do...adjust minute phase at VHF via the FIR files they generate (or IIR embedded in FIR).  Either way the same.   
Dunno ...

But I do know enough to get that FIR correction has difficulty  down low (due processing time)...and I currently believe it has difficulty up high too (due to processing speed).
 
So it makes me wonder about claims i see...
Having said that,  I totally think the MLA processing tech is awesome.
Hell, I want to buy a big "multi-driver horn-loaded point-source" and try my hand with active FIR... to see if it can be tuned even further 

A couple of things here.

Read and reread Tom's comments.

Think about what you said about trying to create a flat phase speaker output.  This is not what cellular drive does. 
A flat phase output at the speaker cabinets face can never create a phase coherent coverage at the audience plane when it is combined with any other cabinets.  Cellular Drive is creating a phase and frequency coherent coverage across the audience plane, not at the speaker.

Lee

[Roland Clarke]

Ultimately, isn't the problem trying to achieve  reasonably flat phase and frequency response at every listener location in the hall or on the field?  Taking into the equation the number of possible changes available to a system be it MLA, ANNYA, KV2, Danley, L'acoustics, Meyer, D&B, et al, there wouldn't be enough permutations to provide close to ideal.

Perhaps we need to start looking more at finding practical solutions?  It's well known that our hearing adapts, go to a concert with too much top end on the system and after about 10 mins it sounds about right.  Similarly, engineers tweaking eq only to find the eq is by-passed.  Perhaps more work in exploiting the weaknesses in our perception to gain where we are most sensitive? 

[Merlijn van Veen]

Taking into the equation the number of possible changes available to a system be it MLA, ANNYA, KV2, Danley, L'acoustics, Meyer, D&B, et al, there wouldn't be enough permutations to provide close to ideal.

"An exact and unique solution does not exist since in general there are more equations than unknowns" - Dr. Evert Start

AFAIK Dr. Evert Start of Duran-Audio aka Axys in the Netherlands, now owned by Harman / JBL Pro, was the first to come up with a solution to this challenge.

Duran was seriously pioneering with beam-steering going back to 1994. Their column loudspeakers with Digital Directivity Synthesis can be found all over the globe in airport and train terminals, churches and cathedrals and HOW's (close to 300 in Mekka alone).

Click here to read the full story.


Verzonden vanaf mijn iPad met Tapatalk

[ProSoundWeb Community]