FIR filters

[Peter Morris]

Re: FIR filters

I wont admit to knowing/understanding everything that Dave is doing even after reading his papers/patents many many times (maybe he will drop by here again), however I do know that his use of FIR filters goes far beyond that of simply flattening the phase response of a system.

15 posts in and no one has mentioned Rephase yet. If all you are looking to do is adjust the magnitude and phase response independently then play around a little with Rephase. I have had good luck importing Smaart transfer functions, creating my adjustments, and having it spit out a FIR coefficients.

Yes what David is doing goes beyond correcting phase and amplitude; he also corrects things in the time domain. The time domain is very important.

Correcting the phase and amplitude will not necessarily mean things are correct in the time domain.

A speaker can sound superb without a flat phase response. From my experience, if you then flatten the phase response it will sound more or less the same, but more real.

The price is always time. An IIR crossover will delay the low frequencies. If you then flatten the phase you are basically delaying the high frequency arrivals to match the low frequency arrivals. If you use a FIR crossover everything is delayed to start with.

FWIW I disagree with the idea that better speakers necessarily have flat responses and don’t need correction. If you model the power response of a speaker with a perfectly rigid cone etc. its response will not be flat.

Speaker designers rely on increasing directivity and cone breakup to maintain a flat response. They are masters of the art of balancing the cone profiles, materials, suspension types etc.

B&C IPAL is an example of what I’m talking about – its a large rigid cone combined with very large and long voice coil plus one of the biggest and most powerful magnet you can get / reasonably afford.

The net result is not what you would consider a good frequency response; add some DSP correction and a feedback loop, then I believe the results are stunning, more quality bass in a small box than just about anything else combined with a reasonable bandwidth.

In 2016 a speaker designer should be taking advantage of everything on offer – Mathematical modelling, DSP processing, the extreme power of modern amplifiers, and some of the new materials on offer for driver construction. I think the old idea that if you use DSP correction you initial design is wrong is an extremely flawed concept.

Having said that, it is important to understand what can, and what cannot be corrected with signal processing.

 

[Michael John]

Re: FIR filters

Yes what David is doing goes beyond correcting phase and amplitude; he also corrects things in the time domain. The time domain is very important.

Correcting the phase and amplitude will not necessarily mean things are correct in the time domain.
.

In the screenshots at the link (above) that I posted, there are two sequences of images. The 2nd sequence, without coherence, starts with a (somewhat poor) measurement of a typically 12" 2-way with a HF driver lagging the LF due to the horn depth. The loudspeaker's passive crossover has been designed to maintain a roughly flat magnitude but the LF leading the HF can clearly be seen in the time domain representation of the measurement on the "Import" tab. The "Export" tab shows the convolved result of the imported measurement with the FIR filter - which in this case is flattening the phase. Note how the impulse is improved/sharpened - I.e. the LF and HF are aligned (but at the expense of the FIR filter delay). This example shows how manipulating phase is manipulating time. The two are directly related.

In the case of time domain horn correction, the LTI behaviour of the horn is a filter and can be expressed as time domain IR or a mag/phase representation. Manipulating the mag & phase of this "filter" affects the time domain behaviour and vice versa. As an aside, I'm curious to know how well David's horn correction filter method holds for different levels, given non-linear/adiabatic air behaviour in the horn at high levels.

 

[Peter Morris]

Re: FIR filters

In the screenshots at the link (above) that I posted, there are two sequences of images. The 2nd sequence, without coherence, starts with a (somewhat poor) measurement of a typically 12" 2-way with a HF driver lagging the LF due to the horn depth. The loudspeaker's passive crossover has been designed to maintain a roughly flat magnitude but the LF leading the HF can clearly be seen in the time domain representation of the measurement on the "Import" tab. The "Export" tab shows the convolved result of the imported measurement with the FIR filter - which in this case is flattening the phase. Note how the impulse is improved/sharpened - I.e. the LF and HF are aligned (but at the expense of the FIR filter delay). This example shows how manipulating phase is manipulating time. The two are directly related.

In the case of time domain horn correction, the LTI behaviour of the horn is a filter and can be expressed as time domain IR or a mag/phase representation. Manipulating the mag & phase of this "filter" affects the time domain behaviour and vice versa. As an aside, I'm curious to know how well David's horn correction filter method holds for different levels, given non-linear/adiabatic air behaviour in the horn at high levels.

In a linear system the magnitude and phase are directly related etc., the problem is not everything we are dealing with is linear.

... also in a multi way system a resonance issues that is out of band e.g an octave below the crossover point, will have little impact on the magnitude and phase response but will be clearly heard by the listener. A corrective filter on the broad band input signal will not be able to solve this problem, it must be dealt with by the processing specifically assigned to that driver.

To see that's happening in the time domain you need to look at plots like this ... (DIY Mid Hi)

 

[Rob Timmerman]

Re: FIR filters

Yes what David is doing goes beyond correcting phase and amplitude; he also corrects things in the time domain. The time domain is very important.

Correcting the phase and amplitude will not necessarily mean things are correct in the time domain.

A speaker can sound superb without a flat phase response. From my experience, if you then flatten the phase response it will sound more or less the same, but more real.

The price is always time. An IIR crossover will delay the low frequencies. If you then flatten the phase you are basically delaying the high frequency arrivals to match the low frequency arrivals. If you use a FIR crossover everything is delayed to start with.

FWIW I disagree with the idea that better speakers necessarily have flat responses and don’t need correction. If you model the power response of a speaker with a perfectly rigid cone etc. its response will not be flat.

Speaker designers rely on increasing directivity and cone breakup to maintain a flat response. They are masters of the art of balancing the cone profiles, materials, suspension types etc.

B&C IPAL is an example of what I’m talking about – its a large rigid cone combined with very large and long voice coil plus one of the biggest and most powerful magnet you can get / reasonably afford.

The net result is not what you would consider a good frequency response; add some DSP correction and a feedback loop, then I believe the results are stunning, more quality bass in a small box than just about anything else combined with a reasonable bandwidth.

In 2016 a speaker designer should be taking advantage of everything on offer – Mathematical modelling, DSP processing, the extreme power of modern amplifiers, and some of the new materials on offer for driver construction. I think the old idea that if you use DSP correction you initial design is wrong is an extremely flawed concept.

Having said that, it is important to understand what can, and what cannot be corrected with signal processing.

Quoted for truth. Fulcrum has designed drivers that perform *worse* without DSP correction so that the corrected performance can improve. B+C has done the same with the IPAL. And I'll bet that many other manufacturers have done the same.

The key here is recognizing what can and cannot be corrected with DSP, and what can and cannot be corrected purely in the mechanical domain.

 

[Frank Koenig]

Re: FIR filters

Peter. I agree completely that speaker designers should use all available tools and that the days of passive crossovers with miles of copper, banks of resistors, and nonlinear elements (light bulbs) for protection should be over. To the extent that speaker-level crossovers are used to reduce amplifier channel count, they should be as simple and lossless as possible, their deficiencies compensated for by upstream DSP.

I have a few comments on math and nomenclature. In audio the terms "linearity" and "distortion" are thrown about in ways that dull their meaning. When we talk about linear systems we should mean linear in the conventional sense that the function F(x) is linear if F(Ax + By) = AF(x) + BF and F(0) = 0. This allows for all manner of frequency or time responses, including non-minimum phase responses where the phase is not uniquely determined by the magnitude, but not for the types of non-linearities where new frequencies are produced, such as harmonic distortion. Traditionally, many linear response anomalies are called distortions (phase distortion, etc.) and sometimes, incorrectly, nonlinear, and this contributes to the confusion.

Further, if we limit ourselves to system models that are time invariant and linear in the above sense, then time and frequency domain representations are equivalent given sufficient detail and duration in each. The "complete" frequency response does indeed tell us everything about the time behavior (the converse is true, too). The smoothed and band-limited magnitude plot in the speaker brochure does not.

As with any subject, it's hard to nail down a logical framework and I don't claim the above ramblings to be complete, just an attempt at increasing the discussion's SNR.

Best,

--Frank

 

[Peter Morris]

Re: FIR filters

Peter. I agree completely that speaker designers should use all available tools and that the days of passive crossovers with miles of copper, banks of resistors, and nonlinear elements (light bulbs) for protection should be over. To the extent that speaker-level crossovers are used to reduce amplifier channel count, they should be as simple and lossless as possible, their deficiencies compensated for by upstream DSP.

I have a few comments on math and nomenclature. In audio the terms "linearity" and "distortion" are thrown about in ways that dull their meaning. When we talk about linear systems we should mean linear in the conventional sense that the function F(x) is linear if F(Ax + By) = AF(x) + BF and F(0) = 0. This allows for all manner of frequency or time responses, including non-minimum phase responses where the phase is not uniquely determined by the magnitude, but not for the types of non-linearities where new frequencies are produced, such as harmonic distortion. Traditionally, many linear response anomalies are called distortions (phase distortion, etc.) and sometimes, incorrectly, nonlinear, and this contributes to the confusion.

Further, if we limit ourselves to system models that are time invariant and linear in the above sense, then time and frequency domain representations are equivalent given sufficient detail and duration in each. The "complete" frequency response does indeed tell us everything about the time behavior (the converse is true, too). The smoothed and band-limited magnitude plot in the speaker brochure does not.

As with any subject, it's hard to nail down a logical framework and I don't claim the above ramblings to be complete, just an attempt at increasing the discussion's SNR.

Best,

--Frank

Thanks Frank … I was using the term too loosely, perhaps it would have been better to say linear / minimum phase.

FWIW the classic example of linear / non-minimum phase in this discussion would be an all-pass filter.

The problem with the “complete frequency response" is that it is typically the sum of 2 or 3 separate sources (Hi Mid Low). The location of the sources are generally different as is their directivity. This makes things complicated, they will not mathematically sum nicely. Accordingly each individual source will need to behave perfectly in the time domain in order to achieve the best result … and ... then there is the reverberant energy we hear.

This is one of the things I love about Danley’s approach with his synergy horn / multi driver point source approach - just one point and one (virtual ) speaker to deal with.

 

[John Roberts]

Re: FIR filters

This is one of the things I love about Danley’s approach with his synergy horn / multi driver point source approach - just one point and one (virtual ) speaker to deal with.

Yes it's elegant we just need to figure out how to use a FIR filter in it... needs to stay modern.

JR

 

[Peter Morris]

Re: FIR filters

Yes it's elegant we just need to figure out how to use a FIR filter in it... needs to stay modern.

JR

Ahhhh yes ... there are a couple of things I would like to do with Tom's design - reduce the weight, throw away the passive filters (more weight reduction) and "attack" it with some FIR crossovers & filters :twisted:~:-D~:grin: ... and if possible make it more saleable.

 

[Tim McCulloch]

Re: FIR filters

Yes it's elegant we just need to figure out how to use a FIR filter in it... needs to stay modern.

JR

From birch to fir - there's your filter. I'll take 20% of gross sales, thankyouverymuch.

 

[TJ Cornish]

Re: FIR filters

Yes it's elegant we just need to figure out how to use a FIR filter in it... needs to stay modern.

JR

I know you're kidding, but I think you're right. I did a shootout between a Danley product and an EAW product that were similar in design. We spent several hours of tuning time and we were never able to get the Danley product to sound as good as the EAW. I don't know how much of this was the magic GF/FIR work EAW did, but starting with a manufacturer-specified tuning surely makes a difference. I think Danley is moving at least in the direction of supplying DSPs and presets. Maybe they will eventually do FIR-level processing.

 

[John Roberts]

Re: FIR filters

I know you're kidding, but I think you're right. I did a shootout between a Danley product and an EAW product that were similar in design. We spent several hours of tuning time and we were never able to get the Danley product to sound as good as the EAW. I don't know how much of this was the magic GF/FIR work EAW did, but starting with a manufacturer-specified tuning surely makes a difference. I think Danley is moving at least in the direction of supplying DSPs and presets. Maybe they will eventually do FIR-level processing.

Yes, I was kidding. I'm not smart enough to second guess TD's designs.

DSP are all but free inside modern high tech power amps, so it makes sense to use them if there is something that needs DSPing.

I would need to know more about your direct comparison, but I am not the answer man for Danley. Ivan will be better equipped to respond to such a comparison thoughtfully.

JR

 

[Ivan Beaver]

Re: FIR filters

.

I would need to know more about your direct comparison, but I am not the answer man for Danley. Ivan will be better equipped to respond to such a comparison thoughtfully.

JR

Without any more information, it is like saying I compared a Ford to a Chevy and the X (you choose) was better. In what way? what models? etc.

Some people like a smooth ride-others like to "feel the road", some like gas mileage, others like acceleration.

With loudspeakers, some people like accuracy, others like a particular colored sound.

When you have accurate loudspeakers, some material sounds better on less accurate speakers-since things are covered up.

So as usual-it depends.

 

[TJ Cornish]

Re: FIR filters

Without any more information, it is like saying I compared a Ford to a Chevy and the X (you choose) was better. In what way? what models? etc.

Some people like a smooth ride-others like to "feel the road", some like gas mileage, others like acceleration.

With loudspeakers, some people like accuracy, others like a particular colored sound.

When you have accurate loudspeakers, some material sounds better on less accurate speakers-since things are covered up.

So as usual-it depends.

We've conversed about this before. I compared the SH60 to the EAW QX566i. Both boxes have the same pattern and physical size. Driver complement is different. We hung them side by side in the same room at the same trim height, pointing at 3 experienced audio engineers. The SH60 was powered by a Crown ITHD 12K, the EAW was processed by a UX8800 and powered by an amp that I don't recall - it may have been another ITHD12K, with the amp set to be flat and the correct focused preset for the EAW box in question.

We tuned and listened for 3 hours playing a large variety of music - gentle female-led material, up-tempo country, some Sting, Counting Crows, Mr. Mister, John Mayer, James Taylor, etc. In some tracks the difference was minimal; in others the EAW outdistanced the Danley box fairly significantly. We tried a number of objective (Smaart-measured) and subjective (trying to pick out what we thought the EAW was doing better and attempting to replicate on the SH60) measurements. There were three experienced sound engineers in the room doing the testing.

I was rooting for the Danley because I like the concept, I like the TH-118s I own, and the box was cheaper than the EAW product when processing was considered, but in the end the decision was unanimous - the EAW box was more pleasant to listen to with everything we threw at it.

Ivan, you're famously quoted as saying "It depends". We actively worked to remove as many "depends" factors. In this situation with the tools provided doing a head-to-head shootout, with the [significant] skill of the engineers present and significant time spent, we were unable to make the Danley perform as well as the EAW did nearly out of the box.

Could another engineer have done a better job? Very possibly, but that's exactly my point - in this one time and place, the Danley box was the loser at least partly because EAW had done the heavy lifting for us, making a better result easy to attain.

You may think I'm a Danley basher, but I'm not, and I've made a substantial investment in Danley equipment. My objection is the incredulity that your team sometimes gives the impression of that other methods and solutions might possibly be better than a Danley solution - i.e. if we didn't pick Danley we must be "listening with our eyes", or wanting some bizarre tonal curve that's beneath a Danley system. In this situation, I assure you that neither of those situations were the case. There were no suits to please, no egos to stroke, no corporate kickbacks to sway our objectivity; just 3 engineers passionate about audio trying to pick the system that sounded best in our space.

 

[Jack Arnott]

Re: FIR filters

just 3 engineers passionate about audio trying to pick the system that sounded best in our space.

My guess is the mid driver is the difference. This is the same type of driver used in Peter's box. Some of us like the sound of the mid diaphragm. Others, like Evan and Art do not. They are quoted as saying that a cone will always sound better than a compression driver, and/or perform better than a compression driver. To me there is a big difference between the two, so I think that "it depends" fits. To me the compression driver sounds more precise, and the cones sound warmer. It is interesting to me to play sine waves between the two. Like 600hz. There is an audible difference, the compression driver sounds thinner. I guess if you like "phat", the cones are the way to go, but I certainly like the definition better. I have a customer who, given the option between standard, and with a shorting ring for lower distortion, chooses standard. He thinks the shorting ring sounds thinner, and he is right. Art comments on how thin the drivers sound without lows, and that is to me how it should be. There are low mids, lows, and subs for body. I like the definition and transient response of the compression drivers. I'm sure the compression drivers take more processing, but as JR notes, they are practically giving that away now.

 

[Jack Arnott]

Re: FIR filters

Back to FIR, I read that FIR comes in both analog and digital. This was a bit of a surprise to me. What would be an example of an analog FIR filter? Also, what do FIR filters look like? As in, on the screen, adjusting them, not in the bits as laid out by Michael. When going from analog to digital, there was not much learning curve adjusting to PEQ, or crossovers, but I have no idea what the world of FIR looks like.

 

[Art Welter]

Re: FIR filters

Originally Posted by TJ Cornish
just 3 engineers passionate about audio trying to pick the system that sounded best in our space.

My guess is the mid driver is the difference. This is the same type of driver used in Peter's box. Some of us like the sound of the mid diaphragm. Others, like Evan and Art do not. They are quoted as saying that a cone will always sound better than a compression driver, and/or perform better than a compression driver. To me there is a big difference between the two, so I think that "it depends" fits. To me the compression driver sounds more precise, and the cones sound warmer. It is interesting to me to play sine waves between the two. Like 600hz. There is an audible difference, the compression driver sounds thinner. I guess if you like "phat", the cones are the way to go, but I certainly like the definition better. I have a customer who, given the option between standard, and with a shorting ring for lower distortion, chooses standard. He thinks the shorting ring sounds thinner, and he is right. Art comments on how thin the drivers sound without lows, and that is to me how it should be. There are low mids, lows, and subs for body. I like the definition and transient response of the compression drivers. I'm sure the compression drivers take more processing, but as JR notes, they are practically giving that away now.

Jack,

You are "quoting" me and Evan out of context. IIRC, Evan encountered BMS co-axial compression drivers in a line array that uses a horns a fraction of the size used by you, Peter, and EAW, yet was crossed lower than any of you cross yours, with predictable results- it did not sound as good as a properly designed system would.

I have never stated that a cone will "always" sound better than a compression driver, and/or perform better than a compression driver. I have stated that cone mids can have far lower distortion at high output levels than compression drivers due to their much greater displacement potential, the opposite of what you imply, a compression driver may sound warmer, or more "phat" than a cone driver due to increased harmonic distortion. Given the same acoustical cut off, the diaphragm material makes no difference, any driver will sound "thin" with no lows. Turning off the lows does make HF distortion (and SPL) more readily apparent.

FIR cannot change pattern control (other than in multiple driver "beam steering") or distortion, but can smooth phase and frequency response. If the frequency and polar response of the SH-60 is similar to the SH-50 (the DSL spec sheet for the SH-60 does not even have an on axis response!), the roughness in both compared to the QX566i should be quite noticeable in an A/B test. The QX566i has four 12" LF drivers compared to the SH-60's two, giving it a 6 dB advantage, and a more sensitive mid/high section, also working to it's advantage in terms of headroom/distortion.

My first introduction to FIR was 2007, comparing a client's Mackie HD1521 to a DSL SH-100. The HD1521's flater frequency and phase response simply sounded better to the client and me, and the flatter response also allowed more gain before feedback. Although you can't replace basic good design with processing, that comparison made it quite apparent that the better processing FIR affords can make what seemed to be quite average components in a typical arrangement sound superb. Did not make a sale, but learned something that day. Actually, it was after that that I realized "Gunness Focusing" used FIR.

That said, it's been over a year now since purchasing a BSS Blu 100 capable of FIR, but I have still not got around to learning how to implement the filters on my system, maybe after I finish my center fill project.
The raw response of the center fills have almost flat phase response from 250, I'll be forced to implement FIR on the mains to align the two…

Art

 

[Peter Morris]

Re: FIR filters

One of the things David and EAW do with FIR filters is fix things in the time domain, it's not just about the phase and amplitude response. They combine the best driver design, box design and signal processing within the required design limits. Those fancy filters do actual work!

 

[Michael John]

Re: FIR filters

I agree, they work.

Remember phase and time are related. Take, say a 90x60 horn, take a bunch of IR measurements within the coverage pattern, then time-align and average the measured IR's. In theory the averaged IR will represent the direction-invariant characteristics of the horn. These characteristics include the internal resonances/reflections or time domain effects of the horn. Now take the averaged IR and generate a filter which flattens the HF phase (and optionally magnitude) from say 3kHz and up. Put the FIR filter inline in a processor and remeasure. The result is an improved or sharper time domain response and the FIR filter need only be a few milliseconds long. I've done this with the FIR design tool for just a few measurement points, and it works. (I haven't tried this for the whole coverage pattern of a horn, nor at many levels.)

As I understand it, the real tricks/challenges are in getting accurate, useful measurements at the right drive levels, and in the time alignment and averaging. One could also model the IR of the early part of the horn. From what I've read and heard, Dave has developed methods for doing both to "fix" the time domain response.

There's something in your email.

 

[Peter Morris]

Re: FIR filters

I agree, they work.

Remember phase and time are related. Take, say a 90x60 horn, take a bunch of IR measurements within the coverage pattern, then time-align and average the measured IR's. In theory the averaged IR will represent the direction-invariant characteristics of the horn. These characteristics include the internal resonances/reflections or time domain effects of the horn. Now take this IR and generate a filter which flattens the HF phase (and optionally magnitude) from say 3kHz and up. Put the FIR filter inline in a processor and remeasure. The result is an improved or sharper time domain response and the FIR filter need only be a few milliseconds long. I've done this with the FIR design tool for just a few measurement points, and it works. (I haven't tried this for the whole coverage pattern of a horn, nor at many levels.)

As I understand it, the real tricks/challenges are in getting accurate, useful measurements at the right drive levels, and in the time alignment and averaging. One could also model the IR of the early part of the horn. From what I've read and heard, Dave has developed methods for doing both to "fix" the time domain response.

There's something in your email.

... I'm not in Oz at the momenton - on a boat in New Zealand, so I will be short. This is exactly what I have done with the DIY

 

[Michael John]

Re: FIR filters

... on a boat in New Zealand...

Nice. I'm jealous!