ProSoundWeb Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: 1 2 [All]   Go Down

Author Topic: Emulating a black box - how would you go about this?  (Read 6612 times)

Riley Casey

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1284
Emulating a black box - how would you go about this?
« on: August 10, 2012, 10:38:43 am »

I have a set of DSP settings that have been provided by a manufacturer as blind settings.  These are loaded into the specified brand and model of DSP unit as a complete, largely locked file with no settings adjustable or visible aside from the high pass filter.  I want to create a back up to the back up DSP ( I already have a back up unit for this DSP) in a different model of DSP.  The settings will be only an approximation as while they are the same brand of processor they are different models. Does anyone have a suggestion on a systematic way to approach this endeavor? 

I have attached a screen shot of transfer function of the original DSP with green trace being a live curve of the low band of the DSP in which I am building a the back up settings. Blue is the low band, yellow the mid band and red the high band.  The green active device trace does not yet have  any EQ applied, only the high and low pass filters.  By definition this is a complete shot in the dark as to everything from filter points & topologies to EQ slopes and frequencies so if anyone has a suggestion as to a good starting point I'd love to hear it.

John Roberts {JR}

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 16249
  • Hickory, Mississippi, USA
    • Resotune
Re: Emulating a black box - how would you go about this?
« Reply #1 on: August 10, 2012, 11:52:21 am »

While low tech I would consider a simple null test, or subtracting one from the other. There may be a need to add some small delay to one or the other, but null them out with wide band noise, and you should be there...  You can listen to the null or measure it, or both.

JR

Logged
On the internet people tell you everything "they" know, not the answer to "your" question.....  http://circularscience.com/

Rasmus Rosenberg

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 55
Re: Emulating a black box - how would you go about this?
« Reply #2 on: August 10, 2012, 01:10:17 pm »

I have a set of DSP settings that have been provided by a manufacturer as blind settings.  These are loaded into the specified brand and model of DSP unit as a complete, largely locked file with no settings adjustable or visible aside from the high pass filter.  I want to create a back up to the back up DSP ( I already have a back up unit for this DSP) in a different model of DSP.  The settings will be only an approximation as while they are the same brand of processor they are different models. Does anyone have a suggestion on a systematic way to approach this endeavor? 

I have attached a screen shot of transfer function of the original DSP with green trace being a live curve of the low band of the DSP in which I am building a the back up settings. Blue is the low band, yellow the mid band and red the high band.  The green active device trace does not yet have  any EQ applied, only the high and low pass filters.  By definition this is a complete shot in the dark as to everything from filter points & topologies to EQ slopes and frequencies so if anyone has a suggestion as to a good starting point I'd love to hear it.

Hey Riley,
What is the original processor? and can't the manufacture help you? (most are really forthcoming, they will rather give the parameters or presets than have people "roll" there own) (And some of us might be able to make you a preset for you, depending on gear ... shhh...;)
If you have acess to a measurement program you can get very close by measureing and "repainting" the TF as I think your saying you will try to do. Depending on platform there will however most likely be some compromises to how things relate/transfer, so listning test is crucial too, and not for the faint hearted :) As for approach I like to start from the HF and then down And make sure you zoom in and out to make the slopes visible, watch for out of band EQ and all pass filters etc. Both the magnitude and phase trace should match and the final preset should have the same latency as the other processor so you can swap it with out worries.
Logged

Ivan Beaver

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 8752
  • Atlanta GA
Re: Emulating a black box - how would you go about this?
« Reply #3 on: August 10, 2012, 07:55:26 pm »

It looks like you didn't do a good job of setting the receive delay.  That is why the phase is so wrapped.

Since you have the transfer function of each band-also measure the delay.

Now you have something that you can 'start to copy".

It depends if there are any "special filters" that may be a bit hard to copy.  This would show up in your phase response curves.

Once you get the two curves to overlap-you should have it.

Logged
A complex question is easily answered by a simple-easy to understand WRONG answer!

Ivan Beaver
Danley Sound Labs

PHYSICS- NOT FADS!

Arthur Skudra

  • SR Forums
  • Full Member
  • *
  • Offline Offline
  • Posts: 160
Re: Emulating a black box - how would you go about this?
« Reply #4 on: August 10, 2012, 08:28:22 pm »

It looks like you didn't do a good job of setting the receive delay.  That is why the phase is so wrapped.
With the exception of the green trace, it looks like he set the receive delay correctly for each of the passbands.  Nevertheless, he does need to figure out what the time delay is for each passband so that he can replicate that as well in his settings.
Logged

Ivan Beaver

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 8752
  • Atlanta GA
Re: Emulating a black box - how would you go about this?
« Reply #5 on: August 10, 2012, 09:19:28 pm »

With the exception of the green trace, it looks like he set the receive delay correctly for each of the passbands.  Nevertheless, he does need to figure out what the time delay is for each passband so that he can replicate that as well in his settings.
I would argue that the blue and yellow traces are still a "bit early".  But they could be "close enough" as to be usable.

In a "perfect world" the phase should "fall" along with the top end of the repsonse of the passband-not tilt upwards-but the little tilt shown could be "workable"..
Logged
A complex question is easily answered by a simple-easy to understand WRONG answer!

Ivan Beaver
Danley Sound Labs

PHYSICS- NOT FADS!

Riley Casey

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1284
Re: Emulating a black box - how would you go about this?
« Reply #6 on: August 10, 2012, 10:44:26 pm »

Thanks for all the observations thus far particularly Ivan for calling out my flakiness in posting an uncorrected live trace.  I pulled a screen shot as soon as I thought to ask for suggestions without setting the delay first on the new processor.  In any case the high and mid bands seem to be falling into place even using the lower cost DSP unit so while the speaker maker and the DSP maker both stonewall the very idea ( quite properly ) of trying to build settings designed for a very capable DSP product into a low end unit it seems like its not out of the question for an emergency backup.  The narrow band notches in the phase trace are a bit odd though.  Not sure what to attribute those to.  Of course the brick wall is the number of filters available and in this case the low end unit ran out of filters at least one short of those required for the low pass band quite aside from not being able to match up the phase trace nearly as well as the mid & high bands.  I'm still working on that but this is the mid & high comparison to date.  Red & Yellow are the 366 and pink and orange are the 336. 

Haven't listened to anything yet though. ::)
« Last Edit: August 10, 2012, 10:46:10 pm by Riley Casey »
Logged

Tim McCulloch

  • SR Forums
  • Hero Member
  • *
  • Offline Offline
  • Posts: 19899
  • Wichita, Kansas USA
Re: Emulating a black box - how would you go about this?
« Reply #7 on: August 13, 2012, 02:14:15 pm »

Thanks for all the observations thus far particularly Ivan for calling out my flakiness in posting an uncorrected live trace.  I pulled a screen shot as soon as I thought to ask for suggestions without setting the delay first on the new processor.  In any case the high and mid bands seem to be falling into place even using the lower cost DSP unit so while the speaker maker and the DSP maker both stonewall the very idea ( quite properly ) of trying to build settings designed for a very capable DSP product into a low end unit it seems like its not out of the question for an emergency backup.  The narrow band notches in the phase trace are a bit odd though.  Not sure what to attribute those to.  Of course the brick wall is the number of filters available and in this case the low end unit ran out of filters at least one short of those required for the low pass band quite aside from not being able to match up the phase trace nearly as well as the mid & high bands.  I'm still working on that but this is the mid & high comparison to date.  Red & Yellow are the 366 and pink and orange are the 336. 

Haven't listened to anything yet though. ::)

The magnitude trace looks a lot like the VerTec v4 4889 midrange processing....  am I close?
Logged
"Practicing an art, no matter how well or badly, is a way to make your soul grow, for heaven's sake. Sing in the shower. Dance to the radio. Tell stories. Write a poem to a friend, even a lousy poem. Do it as well as you possible can. You will get an enormous reward. You will have created something."  - Kurt Vonnegut

Timo Beckman

  • Full Member
  • ***
  • Offline Offline
  • Posts: 141
  • Rational Acoustics/Isemcon/Fulcrum Acoustic NL
    • Timo Beckman Geluid
Re: Emulating a black box - how would you go about this?
« Reply #8 on: August 13, 2012, 02:40:45 pm »

The respons of the low channel has a bump after the x-over point .
It might be a eq point or it could be a NT filter or a chebichev filter . Is this a BSS Processor ?
if so it could be a 4rt order NTM filter but i would have to check that measuring my BSS Blu 80 units on the processor outputs . Could be just a eq point also .....
Logged

Riley Casey

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1284
Re: Emulating a black box - how would you go about this?
« Reply #9 on: August 14, 2012, 12:07:57 pm »

Its actually for 88s but good eye Tim.  Timo these are BSS processors but when I was in the early stages of testing I tried an NTM filter in that spot and did not get a good match.  I was actually able to get a better match by building the components of an NTM filter with an L-R 4th order and then dropping in a separate notch.  The project, now that I've been outed, is an attempt  to build a close approximation Vertec 4888 v4 filters into a BSS FDS336 using transfer function measurements of the factory stock settings from an FDS366.  I decided to build from the top down since it became clear early on that I was going to run out of filter DSP resources before I got everything I needed.  If I'm going to have to run through some outboard EQ to dial it in further I'd rather it not be in the voice range. Not too surprising of course, at least part of the reason the 366 costs almost three times the price of a 336 but I had the processor on the shelf and wanted a spare unit for those crash and burn moments.  This high lights my somewhat tenuous grip on FFT measurement fundamentals though.  The  FDS366 in the stock settings has what appears to be a 3.5ms delay inserted into the low pass output.  Now since this is the "device under test" as seen by the measurement system why is the phase display so unstable on both the 366 and the 336 when the delay is set to the same 3.54 ms ?  I posted a screen cap of the magnitude and phase angle of both units as well as the arrival time selection in the delay finder ( the white vertical line on the first peak ).  The best looking phase display I could get is the second transfer function screen shot with the arrival delay offset as shown in the second delay finder screen shot.   

The magnitude trace looks a lot like the VerTec v4 4889 midrange processing....  am I close?

Quote
Timo Beckman wrote : It might be a eq point or it could be a NT filter or a chebichev filter . Is this a BSS Processor ?

Timo Beckman

  • Full Member
  • ***
  • Offline Offline
  • Posts: 141
  • Rational Acoustics/Isemcon/Fulcrum Acoustic NL
    • Timo Beckman Geluid
Re: Emulating a black box - how would you go about this?
« Reply #10 on: August 14, 2012, 02:42:35 pm »

I've had simular problems with SIM3 while working on phase correcting a line array system build in the netherlands . Working with SIM gives you the possibility of looking at the processor phase and frequency response without having to repatch things .
It took me a while to figure out to get the real response of the processor within SIM because if i synchronize it the processor also gets synchronized (find delay auto small or what ever it's called within the software you use) .
Because i'm working a lot with all-pass filters i wanted to see what the processor was doing without all kinds of delay settings messing up the view of what these filters do .
So because i had to start from scratch i took a full-range trace without filters what so ever of each driver (at very low level in order not to burn up the drivers and to find the best points for the x-overs) .
Within SIM you have to program branches where you tell the machine measure this processor channel/speaker with this mic (out of six in my case) .
The last time i programed each branch and then dubbeled them so i could allways go back to the point where i only measured the latency of the processor itself and not a trace that's influenced by group delay caused by different filters . This allows me to get a view of what filters do if i do not synchronize after that (on the branch i double at the start of the measurement).

To do this with the BSS units and your software you might want to take a measurement of a output with no delay's or filters aplied so you only measure the latency of the processor it self and store the delay value (probably about 1.5 to 2ms).
After that take a processed output and measure it at the time you got for the processor latency .

http://timobeckmangeluid.wordpress.com/2012/05/03/alignment-the-x-over-point-using-2nd-order-all-pass-filters/   
I think this is the link to the first part of that measurement (not sure and i have no time to check that at this moment)

Hope this helps.....

By the way i already tought it was a notch but it has been a while since i worked with NTM filters when i ways practicing with reading the phase display and training myself to see delay time values without having to think about it (not a chance in my case:-). Couldn't remember how the response was on them but it looked a bit like the NTM .

« Last Edit: August 14, 2012, 02:48:46 pm by Timo Beckman »
Logged

Paul Tucci

  • SR Forums
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 371
Re: Emulating a black box - how would you go about this?
« Reply #11 on: August 14, 2012, 09:13:40 pm »

Riley,

May I question your thinking?
Why does the phase response of the second of the two screenshots look better to you?
What criteria detrmine that?

My view of this is that if the bandwidth of interest is the low end, then the low end phase response is where to judge the quality.
Repeated measurements of mic cables or analog gear in bypass mode have taught me that a flat phase trace indicates a proper offset time. It's not quite so flat when real gear is in play. If the crossover output is only passing low end that bandpass should have a relatively flat (with that familiar smiley face) response. The first screenshot shows that. If what you're not liking is the muck in the high end, remember, it's outside the bandwidth of interest. Pay it no mind. Don't know if Foo has a coherence blanking feature that would NOT DISPLAY a phase or magnitude trace if it were deemed unrelated, but if it does I'd use it. That would force your thinking to look for what is important and that is a relatively flat phase trace.

PT
Logged

JoeHenson

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
Re: Emulating a black box - how would you go about this?
« Reply #12 on: August 15, 2012, 02:38:30 am »

Riley, as soon as I saw this, I knew it was the 88 box because of the crazy boost on the high end...but hey, if the 2" drivers can handle it, more power to them.  So I know that Sound Image made tunings for their Lake Processors for the V4 presets...maybe reach out to them and see if they'll give you any insight or files???
Logged

Rasmus Rosenberg

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 55
Re: Emulating a black box - how would you go about this?
« Reply #13 on: August 15, 2012, 04:50:33 pm »

Its actually for 88s but good eye Tim.  Timo these are BSS processors but when I was in the early stages of testing I tried an NTM filter in that spot and did not get a good match.  I was actually able to get a better match by building the components of an NTM filter with an L-R 4th order and then dropping in a separate notch.  The project, now that I've been outed, is an attempt  to build a close approximation Vertec 4888 v4 filters into a BSS FDS336 using transfer function measurements of the factory stock settings from an FDS366.  I decided to build from the top down since it became clear early on that I was going to run out of filter DSP resources before I got everything I needed.  If I'm going to have to run through some outboard EQ to dial it in further I'd rather it not be in the voice range. Not too surprising of course, at least part of the reason the 366 costs almost three times the price of a 336 but I had the processor on the shelf and wanted a spare unit for those crash and burn moments.  This high lights my somewhat tenuous grip on FFT measurement fundamentals though.  The  FDS366 in the stock settings has what appears to be a 3.5ms delay inserted into the low pass output.  Now since this is the "device under test" as seen by the measurement system why is the phase display so unstable on both the 366 and the 336 when the delay is set to the same 3.54 ms ?  I posted a screen cap of the magnitude and phase angle of both units as well as the arrival time selection in the delay finder ( the white vertical line on the first peak ).  The best looking phase display I could get is the second transfer function screen shot with the arrival delay offset as shown in the second delay finder screen shot.

Thats a tight squeeze... filterwise.

If possible, and in the light of compromise, I would make that LPF steeper and at higher frequency and maybe a tat lower in gain. I think it would be better to focus on the 250hz-1kzh than below 125. That low bump filter is IMO better use at around 300hz or 600hz, if possible.
Just an opinion, not facts.... yet... :)

Logged

Riley Casey

  • Hero Member
  • *****
  • Offline Offline
  • Posts: 1284
Re: Emulating a black box - how would you go about this?
« Reply #14 on: August 16, 2012, 11:08:15 am »

I welcome questions about my thinking  ::)

Yes Foo does in fact have trace 'fade' function to indicate low coherence traces which would have worked fine here but which I never turn on because I tend to default to the procedures needed for acoustic measurements where poor coherence is usually a small bandwidth phenomena born of a the occasional reflection.  I didn't think to turn it on when measuring electronics.  Nonetheless I'm glad that the error was there for you to comment on as it was a failure to connect the dots on my part.  I'm aware that response outside of the passband can be a clue to behavior within the passband and I assumed that phase response would be a valuable component of that without taking it a step further and realizing that a couple of octaves would probably be the limit for that sort of thing. 

This screenshot is an expanded version of the previous 366 low pass curve showing more detail and what appears to be a second NTM type filter notch. Based on this I'm even further away from a complete emulation than I originally thought with the available DSP and I need to rethink my procedures for measuring DSP and other electronics.  A window of 24db is my norm for acoustic measurement.  Clearly 80db is better for this measurement.

 
Riley,

May I question your thinking?
Why does the phase response of the second of the two screenshots look better to you?
What criteria detrmine that?

My view of this is that if the bandwidth of interest is the low end, then the low end phase response is where to judge the quality.
Repeated measurements of mic cables or analog gear in bypass mode have taught me that a flat phase trace indicates a proper offset time. It's not quite so flat when real gear is in play. If the crossover output is only passing low end that bandpass should have a relatively flat (with that familiar smiley face) response. The first screenshot shows that. If what you're not liking is the muck in the high end, remember, it's outside the bandwidth of interest. Pay it no mind. Don't know if Foo has a coherence blanking feature that would NOT DISPLAY a phase or magnitude trace if it were deemed unrelated, but if it does I'd use it. That would force your thinking to look for what is important and that is a relatively flat phase trace.

PT

Paul Tucci

  • SR Forums
  • Sr. Member
  • *
  • Offline Offline
  • Posts: 371
Re: Emulating a black box - how would you go about this?
« Reply #15 on: August 17, 2012, 03:02:29 pm »

Riley,
Happy to help. I was mostly referring to the extremely out of bandwidth stuff with the nonsensical phase response. The nearby out of bandwidth range you mentioned adds another important element to the conversation and revealed some very useful info to try to replicate. At 20+dB down you would think those filters wouldn't be effective, but I trust the design guys are smarter than I.
PT
Logged
Pages: 1 2 [All]   Go Up
 


Page created in 0.069 seconds with 22 queries.