(MOVED) - FIR discussion spinoff

Jeff Babcock

Jeff Babcock

Senior
Jan 11, 2011
1,039
26
48
Ontario, Canada
jeffbabcock.org
(MOVED FROM Coaxial Wedge project thread)

As nice as it would be, I don't FIR processing is a reasonable target for this project. Despite having done plenty of reading regarding FIR and having played with things such as the BruteFIR convolution engine, ScopeFIR, etc, as Bennett has pointed out, being able to implement FIR is something completely different from being able to DESIGN FIR. The best I could do based on my own understanding and access to design resources would be simple brickwall filters.

I do have an interest in this and certainly think as FIR type processing starts becoming more mainstream, DIY is going to struggle to compete on a professional level without it. Look at some of the cheaper boxes coming out with FIR processing, some of them sound pretty darn good without relying on "premium" drivers. There is potential for more bang for the buck in a cab with this sort of DSP, particularly as DSP costs for this will lessen in the future.

A well explained and developed DIY process for such design work would be a significant game changer for DIY, though I don't expect to see this any time soon, as I would imagine the good folks like Dave Gunness and others aren't keen to gift the DIY community with such resources after having such significant business investments into it.
 
Last edited:
Impulse Responce correction on and off axis

Bennett Prescott said:
Simply applying FIR to a box is like equalizing to one measurement mic position, you can get it right in one place but may be making it very wrong everywhere else.
Click to expand...

It is possible to improve most loudspeaker's impulse response on and off axis with FIR correction. This is a major point of David Gunness' paper ( http://fulcrum-acoustic.com/wordpres...ssing-2005.pdf ) which I made reference to in my first post, which you may not have read. The trick is to improve the "worst" direction. To be clear, the following quote is in reference to a cone loudspeaker, which typically has little directionality at 1000 hz (for 12" speaker) and lower:

"The directionality of a loudspeaker also affects its transient response, but it is not a 2-port characteristic. Hence it cannot be corrected everywhere. However, the transient response can be modified in a way that produces greater consistency throughout the pattern. By improving the transient response in the "worst" direction, at the expense of the transient response in the "best" direction, the sound quality can be made more consistent over the breadth of the coverage pattern. This is exactly what has been implemented in the example shown. The off axis spectrogram, shown in figures 3.3.3 and 3.3.4, shows a degree of improvement similar to that observed on axis."

Attached are more measurements of a product which improves the transient response of most speaker systems both on axis and 30 degrees off axis (60 deg bandwidth) with FIR correction. Information may be located here:

http://www.musicanddesign.com/pubimages/UE-2-way.GIF
http://techtalk.parts-express.com/showthread.php?t=218580

And here are two more self-powered monitor/FOH speaker systems which have jumped on the bandwagon, to correct impulse response using FIR:

http://www.yamahaproaudio.com/products/speakers/dsr_series/
http://www.dasaudio.com/index.asp?pagina=actualidad&subpagina=2343&c=125

UE-2-way.jpg
 
DSP latency

NOTE:
What I'm about to say does not only apply to FIR calculations in a DSP, but applies to all DSP calculations such as crossovers, parametric EQ's, Graphic EQ's, time delay, limiting, and compression, etc.

GIVEN:
It is well established and accepted that DSP latency should be minimized as much as possible and any DSP developer is well aware of algorithms (FFT vs DFT for example) which affects the throughput of a DSP engine.

Bennett Prescott said:
All you need is a DSP that can do about a 2ms FIR.
Click to expand...

Please provide your source for your information so that I might learn something new here. I have never read nor heard of any such standard. If you can't provide a source then I am left to assume your information is speculation and arbitrary.

I am not sure "all you need is a DSP that can do about a 2ms FIR". Nor am I sure you can put an absolute value on the required DSP latency to perform an FIR, at least one where the industry as a whole would accept as a standard.

But, let's assume you are right and it does require "about 2ms". And because you included the modifier "about", I'll throw in plus and minus 25% error in your favor, so that leaves 2.5ms to 1.5ms.

As an example of using your low number (1.5ms), I have been studying commercial DSP engines for some time and came across this one ( http://www.symetrix.co/wp-content/uploads/2011/01/Jupiter_ds_EN1.pdf ) which is capable of FIR calculations and to their credit, they publish their idle time as 1.6ms. Which means, it takes 1.6ms for the audio signal to go from input to output with no DSP processing of any kind. According to your numbers on the low side (1.5ms), this DSP is not sufficient to be used for FIR processing, because 1.5ms (required time for FIR) is less than 1.6ms (idle time). In other words, it takes more time sitting idle than it does to perform the needed FIR processing.

This leads me to conclude your low number is off.

Now let's consider your high side number of 2.5ms. What would be the penalty of a DSP calculating an FIR which takes 2.5ms or longer? To keep the math simple and ignoring any atmospheric interference, let's assume the speed of sound in air is one foot per millisecond. Then the penalty of a 2.5ms FIR calculation is simple. It would have a similar affect as moving your coax wedge 2.5 feet farther from your ears, and the SPL would not change. That is it. Seems like a benign penalty to me. Even if the calculations took twice or three times longer than 2.5ms, it would still appear to be benign.

Pictures are worth a thousand words, so I have attached a simple diagram to show the affect of no latency, 2ms latency, and 4ms latency.

To quote David Gunness' paper ( http://fulcrum-acoustic.com/wordpre...ponse-with-digital-signal-processing-2005.pdf ):

"All loudspeaker systems are subject to significant latency, because of the relatively slow propagation speed of sound in air. Therefore, a small amount of added latency is usually inconsequential."


DSP_latency.JPG
 
Re: DSP latency

Kimo,

The FIR is on top of the base latency of the processor, which is due to consequences of the AD and DA and other such largely unavoidable issues. I am well aware of the effects of latency on loudspeaker systems, and feel in general it should be minimized in any live reinforcement situation. As Ales Dravinec once memorably said, I want the audio to happen now, not later. Using longer FIRs to attempt to solve LF problems that can be easily mitigated with IIR processing is just sloppy engineering, IMHO.

I don't want to put words into Dave Gunness's mouth, but I believe he told me he uses a fixed 4ms time window on his FIRs, and only introduces 1.5ms of latency. A shorter FIR is possible depending on how tolerant you are of artifacts and how low a frequency you want to go to.
 
Re: DSP latency

Bennett Prescott said:
Using longer FIRs to attempt to solve LF problems that can be easily mitigated with IIR processing is just sloppy engineering, IMHO.
Click to expand...

From this comment, I'm not sure you do understand the benefits of impulse response correction using FIR, which are not limited to mid and high frequencies, it is also beneficial for low frequencies. If you had read and understood Dave Gunness's paper which I have referred to three times now, we would not be having these conversations. It's all there.

Here are all of Dave's papers at his business... Feel free to read away and then we can argue about data, not subjective opinions, subjective numbers, and subjective measurements, and subjective quotes.

http://fulcrum-acoustic.com/technologies/whitepapers

How do you determine it is "sloppy engineering". Is this one more example of over stating the conversation?

Bennett Prescott said:
I don't want to put words into Dave Gunness's mouth, but....
Click to expand...

If you don't want to put words in Dave Gunness' mouth, then don't. But for some reason, you can't stop yourself.
 
Re: DSP latency

Kimo,

The application of FIR is subjective. If you feel that you have a low frequency phenomena that is best addressed by FIR and think the time penalty is inconsequential, then go ahead and design your box that way.

I have read the majority of Dave's papers. As I am a lousy math student perhaps as you suppose I did not understand them, but I am under the impression that I have a decent understanding of the potential benefits and application.
 
Re: DSP latency

Please note I have separated this discussion from the Coax wedge thread as it is starting to drift from the project focus somewhat. At this present time, as nice as it might be, I don't believe FIR processing is a reasonable goal for the coax wedge project. Feel free to continue the FIR discussion if desired.
 
Re: DSP latency

Bennett Prescott said:
I have read the majority of Dave's papers. As I am a lousy math student perhaps as you suppose I did not understand them
Click to expand...

I once again refer you to Dave's paper. It only contains a single formula for the impedance of the phase plug termination of a compression driver. This formula may be ignored with no side affect of misunderstanding the main points. Outside of that formula, there is no other math, just lots of good data and good pictures. This is exactly why I thought it was a great paper for all to read.

http://fulcrum-acoustic.com/wordpre...ponse-with-digital-signal-processing-2005.pdf

http://fulcrum-acoustic.com/wordpre...ponse-with-digital-signal-processing-2005.pdf

Too bad you didn't read this paper as mentioned in my first post, especially after all the banter which could have been avoided. I believe you would have understood it.
 
Re: DSP latency

I have been arguing for for clarity in DSP control interfaces (Q definitions, etc). One interesting suggestion that came out of a discussion over at the LAB forum about FIR and IIR filters in DSP, was providing the facility to directly enter filter coefficients into DSP platforms to bypass the "Babelicious" control interfaces.

While computing these filter coefficients may be a little beyond the typical DIY crossover designer, and awkward to key in raw numbers, some lower machine level interface could be imagined that would allow for more accurate and powerful control of DSP platforms.

JR
 
Re: DSP latency

Kimo,

I am not going to continue arguing with you on this topic. I am reasonably certain, having consulted someone who understands this topic at a base level better than I, that I do, in fact, understand the mechanisms at work. I am not entirely sure of the basis of our argument in the first place, perhaps some turn of phrase has convinced you that I am attacking you or that we are otherwise at odds. If you feel you have a solid enough grasp of the principles to develop your own FIR processed box then please do, it would be a great advance for the DIY community.
 
Re: (MOVED) - FIR discussion spinoff

Just doing some digging re more advanced DIY processing....

I did not realize that MiniDSP has recently released the ability to support up to 68 custom IIR biquads on their inexpensive balanced minidsp box, and are supposedly working on FIR support in a software update (according to a post by one their Dev Team).

http://www.minidsp.com/applications/advanced-biquad-programming

http://www.minidsp.com/onlinestore/detail/7-minidsp-kits/flypage/65-minidsp-balanced-kit?sef=hcfp

With the cost of the minidsp balanced unit at just $120 and $10 for the "advanced" software add on which enables the custom biquads, this seems like an interesting potential entry point for DIY, particularly if FIR comes on board.

Given this, it could be possible to build a custom processor either as an external box (add as many minidsp's as you want (each is 2x4) into your own box), or attached to a plate amp for powered speakers at a really low cost.

As an example if applied to the coaxial wedge project, you could build a processor capable of providing 4 biamped monitor mixes for just $260 plus the cost of your blank rack mount box, connectors, psu, and cabling. Easily well under $400. Hmmmmmmmmmmmm...... if their FIR update ever comes out then there's a potential stepping stone to get DIY FIR gaining some interest which may lead to better DIY resources for design.
 
Last edited:
Re: (MOVED) - FIR discussion spinoff

Jeff Babcock said:
Just doing some digging re more advanced DIY processing....

I did not realize that MiniDSP has recently released the ability to support up to 68 custom IIR biquads on their inexpensive balanced minidsp box, and are supposedly working on FIR support in a software update (according to a post by one their Dev Team).

http://www.minidsp.com/applications/advanced-biquad-programming

http://www.minidsp.com/onlinestore/detail/7-minidsp-kits/flypage/65-minidsp-balanced-kit?sef=hcfp

With the cost of the minidsp balanced unit at just $120 and $10 for the "advanced" software add on which enables the custom biquads, this seems like an interesting potential entry point for DIY, particularly if FIR comes on board.

Given this, it could be possible to build a custom processor either as an external box (add as many minidsp's as you want (each is 2x4) into your own box), or attached to a plate amp for powered speakers at a really low cost.

As an example if applied to the coaxial wedge project, you could build a processor capable of providing 4 biamped monitor mixes for just $260 plus the cost of your blank rack mount box, connectors, psu, and cabling. Easily well under $400. Hmmmmmmmmmmmm...... if their FIR update ever comes out then there's a potential stepping stone to get DIY FIR gaining some interest which may lead to better DIY resources for design.
Click to expand...


FYI, Minidsp has done futher investigation and have concluded that their hardware platform does not have enough DSP horsepower for implementing FIR to an extent that would be useful. So the custom biquads are as far as it goes with those units. The Symetrix Jupiter which starts at around 1K seems to be the lowest cost FIR capable entry point. http://www.symetrix.co/wp-content/t.../img/screenshots/sound_reinforcement_3_j4.png
 
Re: (MOVED) - FIR discussion spinoff

Jeff Babcock said:
The Symetrix Jupiter which starts at around 1K seems to be the lowest cost FIR capable entry point. http://www.symetrix.co/wp-content/t.../img/screenshots/sound_reinforcement_3_j4.png
Click to expand...

Used BSS Soundwebs pop up for $500-$600, and if the power supply capacitors are inspected/changed preemptively (the most common failure mode), I know of many that survive life on tours and they are almost unmatched for their feature/dollar ratio (including FIR). Newer options offer more user friendly connectivity (Ie ethernet vs serial, cobranet, etc) but if you are looking for the minimum cost entry point that might be it. Some would say the Soundweb isn't robust enough, but at the price-point they are selling for now it may be a worthwhile consideration at the budget end. The model you want is the 9088iis and while the 9088ii could be considered at the right price (it's really the same but older), the original 9088's are really getting on in age and are best avoided, though my little venue has two of the original 9088's humming away without issue. Also try to get ones that have mic input cards (which can do both mic/line).

If I was considering getting into the Soundwebs from scratch I'd likely buy two - setup each one with one mic card and one line card (or two mic cards each if you can find them - just make sure the machines match either way) and you have a pretty good audio swiss army knife for briefcase gigs and a backup for your monitor DSP. For example we used one last week as 6 channels of EQ for monitors and FOH and 2 channels of compression for a corporate band show that required a tiny FOH footprint but whose soundman wasn't familiar with the 01v96 or similar. If you are used to digital mixers the Soundweb can be a real godsend if you stuck behind a 1604 or something.

Just my rambling thoughts on the matter....
 
Re: (MOVED) - FIR discussion spinoff

Hi Kimo,

Just to add to Bennett's points, I believe the reason FIR works as it does in methods like Gunness focusing is that the effects being tamed are spatially localised, fairly predictable and fairly consistent. Effects such as certain horn resonances or horn shape characteristics. At mid and lower frequencies, other effects come into play including the cabinet's interaction with adjacent cabinets and the room and weather effects. These effects are not consistent, and tuning FIR impulse response correction filters in these ranges could lead to worse performance.

One example. Years ago I designed a simple concept system to do active reverberation in small rooms. (Not the first by any means.) The speaker to mic equalization was done using ~8000 tap FIR filters (@ 48 kHz). Naturally, the system operated close to the edge of feedback. Every hour or two I had to re-measure the impulse responses due to temperature changes in the room. People moving through the room also affected the system stability.

Best,
Michael
 
Last edited:
Re: (MOVED) - FIR discussion spinoff

Just some more comments about FIR filters.....

Often when I see FIR filtering mentioned in the context of live sound, people frequently mean Linear Phase FIR filters. These filters have a symmetric impulse response and a fixed group delay (across frequency) which is half the impulse response length. FIR filters can be designed to have other phase characteristics, including minimum phase, with little to no latency.

Regardless of the FIR filter design, the difficulty is implementing the filtering with low latency. (This implementation latency is a separate issue to the latency/delay in the impulse response itself.) There are techniques whereby an impulse response can be chopped up into smaller pieces and the smaller pieces processed in a parallel fashion. Lake/Dolby has some IP in this area. The implementation latency is then reduced to the size of the smaller piece. The convolution result is identical to brute force convolution of the long impulse response. The overall latency is the sum of the latency of the filter design, the smallest block size of the filtering and finally the A/D and D/A converters. In the interest of full disclosure, I work for Dolby.

Someone earlier mentioned linear phase brick wall FIR filtering. I've successfully used these in biamping some (previously passive) EAW VR21 cabinets to essentially eliminate off-axis lobing in the crossover region. Below is a link to a photo of one system I designed using this. Here the VR21 is using linear phase brick wall FIR crossover with a group delay of 5ms. The "sub" is delayed 5ms to match but operated up to about 300Hz to help steer the low mid frequencies. It has a 24dB/oct LR low pass filter and the VR21 has an all-pass filter to match the phase change of the low pass on the "sub." The result was great - directed sound with no significant lobing issues below the cabinet and no feedback with mics 1.5m below.

http://www.flickr.com/photos/michael_smithers/6310460687/

Best,
Michael
 
Last edited:
Re: (MOVED) - FIR discussion spinoff

Jeff Babcock said:
FYI, Minidsp has done futher investigation and have concluded that their hardware platform does not have enough DSP horsepower for implementing FIR to an extent that would be useful. So the custom biquads are as far as it goes with those units. The Symetrix Jupiter which starts at around 1K seems to be the lowest cost FIR capable entry point. http://www.symetrix.co/wp-content/t.../img/screenshots/sound_reinforcement_3_j4.png
Click to expand...

Just by way of update, MiniDSP recently announced a new DSP platform called OpenDRC which now supports FIR. A 2 in 2 out box is set to be around $300 and there will be both analog and digital versions. I suspect this will eventually scale to larger channel count units, as their previous platform could do up to 8x8.

This would certainly be the lowest cost FIR-capable DSP device that I have seen available to the public. It will be interesting to see if manufacturers such as Fulcrum Acoustic create presets for these as a lower cost DSP alternative for their products.

This may also spark increased DIY interest in FIR, though the DIY resources for creating FIR presets have a long way to go.
 
Re: (MOVED) - FIR discussion spinoff

But have they designed a tool that lets Joe Schmoes like me actually build FIRs? There are plenty of DSPs that you can stick a several millisecond FIR into, but making one (and doing it smart) in the first place is beyond most mere mortals.
 
Re: (MOVED) - FIR discussion spinoff

Bennett Prescott said:
But have they designed a tool that lets Joe Schmoes like me actually build FIRs? There are plenty of DSPs that you can stick a several millisecond FIR into, but making one (and doing it smart) in the first place is beyond most mere mortals.
Click to expand...

Indeed, noted when I said that it has a long way to go.... there are some tools such as a free program called DRC, but are largely targeted at room correction, not loudspeaker preset design. Doing a brickwall style FIR is about the limit for most folks heads to wrap around. There are some commercial tools such as ScopeFIR, among several others, but again, this is not trivial for the average user to just jump into, the resources just aren't there.
 
Top Bottom