LinFIR – DSP Software for FIR/IIR Filter Design and Speaker Correction

[Arnwald]

In fact, there is a link between the phase curve offset and the symmetry of the impulse response.

When the phase is flattened around 0° the impulse response gets an even symmetry, while when it is flattened around ±90° the symmetry becomes odd. And when approaching ±180°, it becomes even again, but with an inverted polarity.

You can see it in the attached screenshots. The phase correction filter has an offset parameter that allows to play with the symmetry. Sometimes it can be useful to fine tune a system.

 

[Waveform Fidelity]

You can see these effects in real time with Marani AEQ. That tool provides the user a pair of cursors in the amplitude and phase windows, that allows the desired part of the frequency range to be equalized. Moving these, together with the target level, shows the corresponding gain and phase responses and the corresponding ripples in gain and phase, which is extremely useful. I had used matlab to create FIR filters, but there is no way to do this iterative “tuning” in real time.

On the topic of room equalization - I don’t do it. Not a big fan. The multi-path delay spread is large and the distance between speaker and listener change significantly in terms of wavelength at that critical band of frequencies covering for example human speech, so EQ has limited ability to compensate.,

On your last point about the audibility of phase distortion, I invite you to think instead of waveform distortion caused by the phases of all the spectral components, being shifted from the true values giving rise to a distorted SPL waveform, which is a significant form of distortion. Play a square wave through your LS and tell me what you think.

In my opinion, the LS industry, with the exception of a few, have been dismissing both phase and group delay effects on the SPL waveform because with passive cross-overs there is not much that can be done, so the claim its not audible suits the purpose well. DSP based crossovers have just emulated their analog counterparts with IIR filters. FIR filter have been around years, but designing them is specialized, that most loudspeaker users/designers don’t have, so a barrier to adoption. Now we are seeing software like yours and others that allows user that don’t code DSP and/or don’t design filters being able to use the power of FIR/IIR filters to really tackle complex crossovers and EQ.

Even KEF, have finally caught up with their LS60 in this regard.

 

[Julf]

Play a square wave through your LS and tell me what you think.

But make sure you only listen to the square wave instead of looking at it on a screen.

In my opinion, the LS industry, with the exception of a few, have been dismissing both phase and group delay effects on the SPL waveform because with passive cross-overs there is not much that can be done, so the claim its not audible suits the purpose well.

Or perhaps, just perhaps, because almost all the research has shown phase and group delay effects not to be audible despite looking spectacular on a scope?

 

[ernestcarl]

Or perhaps, just perhaps, because almost all the research has shown phase and group delay effects not to be audible despite looking spectacular on a scope?

And what about phase linearizing artifacts caused by equalizing "the room" or listening position(s) of most interest? At what point does it start to become audible in any given scenario? I am genuinely curious if all the research has already been performed because I do utilize a little bit of "room phase EQ" myself. Not by a huge extent even, but rather enough to get a much better summed IR result which also happens to "usually" be a better resulting GD/phase/magnitude/step/impulse and clarity graphs -- not "always" mind you.

I try to listen to audible artifacts whenever I can after significant enough EQ manipulation -- because there's almost always going to be audible artifacts at some point with certain test signals/tracks.

*The important part to me is whether the "poison"/cure is worth the price. With my own critical listening tests, overall sometimes it is, sometimes it isn't.

 

[Julf]

I am genuinely curious if all the research has already been performed

I doubt all research will ever be done (or even performed). A lot has.

With my own critical listening tests, overall sometimes it is, sometimes it isn't.

What does your critical listening tests involve? Sighted listening?

 

[ernestcarl]

Sighted listening?

I've done some blind testing and semi(?) blind testing if you could call it that -- posted results in other threads one being probably the "acourate" thread if I am remembering right.

Generally, more often sighted A/B switching than blind out of convenience. Almost every "expert" says to generally not use room correcting phase EQ which inevitably cause linearizing "artifacts" out of the MLP -- and I would agree -- I understand why given how easy it is to hear the side-effects of such when done to the extreme. In fact, one would be totally deaf not to hear it. Of course, it is the much smaller changes/ripples that is more difficult/dubious to assess. But it can still be done -- and at certain threshold points one can truly hear slowly worsening audible amounts of such artifacts.

 

[ernestcarl]

Ripples in the GD below can be seen in the corrective filter itself overshooting the "target" by excessively forcing a reduction in the centering time (bias to the left) to reduce latency:

SPL, Phase, GD


How big does the overshoot in any of the three graphs have to be before producing clearly unacceptable audible artifacts in the flawed room I am dubiously phase correcting for?

 

[Waveform Fidelity]

In fact, there is a link between the phase curve offset and the symmetry of the impulse response.

When the phase is flattened around 0° the impulse response gets an even symmetry, while when it is flattened around ±90° the symmetry becomes odd. And when approaching ±180°, it becomes even again, but with an inverted polarity.

You can see it in the attached screenshots. The phase correction filter has an offset parameter that allows to play with the symmetry. Sometimes it can be useful to fine tune a system.

Interesting. Equalizing the phase to zero does matter as you mentioned -its not just zero group delay, but its not widely discussed, so I thought the following reference might be helpful if others come across this post. Often its called phase intercept distortion. Is that what your offset control is doing?

https://electroagenda.com/en/frequency-constant-phase-shift-and-distortion/

 

[Waveform Fidelity]

But make sure you only listen to the square wave instead of looking at it on a screen.

Or perhaps, just perhaps, because almost all the research has shown phase and group delay effects not to be audible despite looking spectacular on a scope?

Hi Julf et al

I have an open mind about this, but we differ in what the research in the past has actually concluded. I suggest looking back through AES papers that will show that many experiments and evaluations have been conducted to check the audibility of group delay, phase distortion etc. Large amounts of group delay are clearly audible (the reason why higher order IIR Biquads are not used in crossovers), some more so in lower frequency bands and some depending on the material used and the user/evaluator. So I think its a stretch to claim “almost all” research…

I have converted my KEF R900 speakers to create flat SPL and zero acoustic phase and I have a switch on the rear panel to do A/B testing. In my tests, I did not find that switching to flat SPL zero acoustic phase from KEF’s OEM response an immediate and obvious difference, except in the bass region (only Eq above 110Hz) where drums sound different in their attack transients. I’d like to give concrete examples in the mid-range but I have not fully convinced myself yet if and what the differences are.

The other points worth noting:

A. How many loudspeakers that have flat SPL and zero phase have been built? let alone subjectively evaluated by the masses? So how can anyone arrive at the conclusion that phase-induced waveform distortion doesn’t matter?

B. In the past, its plausible to me that other forms of distortion, non-linear drivers/crossovers, break-up resonances, etc could have been masking the waveform distortion effects of group delay. However, as driver, enclosure, crossover components, and other distortions have reduced (eg KEF R3 MAT driver ~1/4% THD level), other artifacts could become more audible.

I’d also be interested in hearing from members that have actually listened, built or measured flat SPL/zero phase shift loudspeakers and what the subjective evaluation revealed.

LINFIR software is now giving people to option to build and equalize speakers with FIR/IRR filters - its tiime to do new experiments and projects.

In my opinion, the jury is still out on this one. I’d be delighted to be proven wrong.

 

[Arnwald]

You can see these effects in real time with Marani AEQ. That tool provides the user a pair of cursors in the amplitude and phase windows, that allows the desired part of the frequency range to be equalized. Moving these, together with the target level, shows the corresponding gain and phase responses and the corresponding ripples in gain and phase, which is extremely useful. I had used matlab to create FIR filters, but there is no way to do this iterative “tuning” in real time.

On the topic of room equalization - I don’t do it. Not a big fan. The multi-path delay spread is large and the distance between speaker and listener change significantly in terms of wavelength at that critical band of frequencies covering for example human speech, so EQ has limited ability to compensate.,

On your last point about the audibility of phase distortion, I invite you to think instead of waveform distortion caused by the phases of all the spectral components, being shifted from the true values giving rise to a distorted SPL waveform, which is a significant form of distortion. Play a square wave through your LS and tell me what you think.

In my opinion, the LS industry, with the exception of a few, have been dismissing both phase and group delay effects on the SPL waveform because with passive cross-overs there is not much that can be done, so the claim its not audible suits the purpose well. DSP based crossovers have just emulated their analog counterparts with IIR filters. FIR filter have been around years, but designing them is specialized, that most loudspeaker users/designers don’t have, so a barrier to adoption. Now we are seeing software like yours and others that allows user that don’t code DSP and/or don’t design filters being able to use the power of FIR/IIR filters to really tackle complex crossovers and EQ.

Even KEF, have finally caught up with their LS60 in this regard.

In the past, I did some experimentations with phase distorted square waves generated with a Python script, and I must say I didn't hear any difference. But to be fair, this is a steady state signal, so maybe differences would be more audible with transients. But here again, I couldn't hear any difference when comparing linear phase crossovers to traditional minimum phase LRs.

Not a scientific protocol, I agree, but I wanted to believe and ended up disappointed

That being said, I'm a huge fan of brickwall (or sinc) crossovers because they are so much easier to manage, and they clean up the directivity pattern by limiting off-axis interferences. This probably has the most audible impact, as the indirect sound becomes much more balanced. On the subject, I find the definition of the standard spinorama somewhat too lenient, given that most rooms are wider than they are tall, and floors and ceilings are hard to treat.

Interesting. Equalizing the phase to zero does matter as you mentioned -its not just zero group delay, but its not widely discussed, so I thought the following reference might be helpful if others come across this post. Often its called phase intercept distortion. Is that what your offset control is doing?

https://electroagenda.com/en/frequency-constant-phase-shift-and-distortion/

Thanks for bringing this article to my knowledge. Yes, that's exactly it.

Audio signals being purely real, the filter must be Hermitian: the negative frequency part of its spectrum is the complex conjugate of the positive part, so that the result of the inverse Fourier Transform is a real signal. So the phase offset is a constant phase added to all positive frequencies and subtracted to all negative frequencies.

 

[Julf]

I have converted my KEF R900 speakers to create flat SPL and zero acoustic phase and I have a switch on the rear panel to do A/B testing.

So sighted tests?

A. How many loudspeakers that have flat SPL and zero phase have been built? let alone subjectively evaluated by the masses? So how can anyone arrive at the conclusion that phase-induced waveform distortion doesn’t matter?

There was a period in the late 1970s and early 1980s when "phase coherence" was all the rage, and we got stuff like the B&W "pregnant penguin" DM6 and many others. The fad died because the benefits were not audible.

I’d also be interested in hearing from members that have actually listened, built or measured flat SPL/zero phase shift loudspeakers and what the subjective evaluation revealed.

There are hundreds of audiophile forums full of subjective "evaluations". Yes, this forum seems to be constantly sliding that way, but can we please respect the "Science" part of the forum name?

 

[Waveform Fidelity]

So sighted tests?

There was a period in the late 1970s and early 1980s when "phase coherence" was all the rage, and we got stuff like the B&W "pregnant penguin" DM6 and many others. The fad died because the benefits were not audible.

There are hundreds of audiophile forums full of subjective "evaluations". Yes, this forum seems to be constantly sliding that way, but can we please respect the "Science" part of the forum name?

Yes agree with your more science comment. I have been an R&D Engineer almost all of my working career, so the Science and Engineering of both the transducer and the hearing system are the most important to me, and I look forward to more scientific posts on this forum, based upon science and real data vs lots of opinion.

On the group delay topic, one might reasonably argue if waveform distortion is not audible, then whats the motive for seeking vanishing low THD/IMD? Or other distortions for that matter?

I use FIR filters to get the SPL correct and use steep crossovers, the equalization of phase comes free, so why not do it? I don’t see any downsides to it, and like I mentioned at bass frequencies it is audible.

In the end music is there to be enjoyed. I had some email conversation with Angelo Farina years ago and he basically said equalize your sound system contour to your preference. Floyd Toole basically says the same thing - provide tone controls for user preference.

Anyway thank your comments and observations.

 

[Julf]

On the group delay topic, one might reasonably argue if waveform distortion is not audible, then whats the motive for seeking vanishing low THD/IMD? Or other distortions for that matter?

I think we all agree that THD is not very audible unless excessive. IMD is a different story.

I use FIR filters to get the SPL correct and use steep crossovers, the equalization of phase comes free, so why not do it? I don’t see any downsides to it, and like I mentioned at bass frequencies it is audible.

Is it?

In the end music is there to be enjoyed. I had some email conversation with Angelo Farina years ago and he basically said equalize your sound system contour to your preference. Floyd Toole basically says the same thing - provide tone controls for user preference.

Sure - but that is a different discussion. My point is always "would you really want to add ketchup and salt (because that is what you like) to every dish in a michelin-star restaurant"?

 

[Waveform Fidelity]

I think we all agree that THD is not very audible unless excessive. IMD is a different story.

Is it?

Sure - but that is a different discussion. My point is always "would you really want to add ketchup and salt (because that is what you like) to every dish in a michelin-star restaurant"?

THD and IMD are just two different ways of measuring a non-linearity. Both are convenient test signals, but a poor representation of a music signal, which has AM/FM/PM fluctuating envelope and non-stationary stats. That’s possibly one reason that systems that measure well with thge simple signals, don’t sound as good as one might expect from the measurements. In my opinion, its yet another example of how R&D and innovation in measurement side of the LS business has remained unchanged for decades.

 

[Julf]

THD and IMD are just two different ways of measuring a non-linearity. Both are convenient test signals, but a poor representation of a music signal, which has AM/FM/PM fluctuating envelope and non-stationary stats. That’s possibly one reason that systems that measure well with thge simple signals, don’t sound as good as one might expect from the measurements. In my opinion, its yet another example of how R&D and innovation in measurement side of the LS business has remained unchanged for decades.

Isn't this the old "but music is different from test signals" argument that people who don't understand Fourier series keep repeating?

 

[theREALdotnet]

Isn't this the old "but music is different from test signals" argument that people who don't understand Fourier series keep repeating?

Or perhaps they do understand the Fourier transform, i.e. the representation of any signal by the superposition of a set of harmonic (sinusoidal) oscillations. And maybe they also understand that the approach of studying the signal transmission properties of a system by looking at its response to single, steady-state oscillations and inferring its response to any signal is only valid for linear, time-invariant systems.

 

[Waveform Fidelity]

Isn't this the old "but music is different from test signals" argument that people who don't understand Fourier series keep repeating?

Well, if you are referring to me. I know very well about Fourier, his transforms and the limitations of looking at things in the frequency domain. My experience comes from 30+ years in communications and signal prcoessing engineering in R&D.

The point I was trying to make is that test sine wave is constant amplitude and 100% deterministic, 2 tone IM is not much different but has a small PAPR as the ratio is normally 4:1 in audio test. These signals are great, and still great for static distortion, FR etc tests and are great for comparing different electro-acoustic systems.

Music has a large PAPR, its stats are non-stationary, it contains AM/FM and PM all warying as a function of time. Is a loudspeaker drive a LTI system? Clearly not the time varying mean level biases the cone at different locations in the coil/magnet gap where B is not linear with cone displacement. My only point is that the industry can do a much better job on devising a new type of test signal that more closely resembles a music waveform and measures this dynamic distortion. Short FFT (or Gabor transforms) are a way to look at the time varying aspects, but not the complete answer.

Why not use an actual sample of music and compare the input and output time waveforms for example?

 

[Salt]

...
Why not use an actual sample of music and compare the input and output time waveforms for example?

So give us an example by, let's say, Deltawave by @pkane, and we will see the difference ... not?

 

[Waveform Fidelity]

Didn't know that software existed - thanks for pointing this out. I’ll find time to experiment

 

[Arnwald]

LinFIR 1.3.0 is out.

The headline change is on the Auto EQ side: the algorithm has been reworked to better respect the configured gain limits and boost cap, which should make the results more predictable. A new auto-generate toggle (licensed feature) lets the filters update in real time as you tweak parameters, instead of requiring a manual generate step each time.

The Adaptive Window also gets a meaningful upgrade: the internal wavelet transform moves from CDF 5/3 to CDF 9/7, which roughly doubles the stopband attenuation (~40 dB vs ~18 dB). In practice this eliminates the frequency ripples that could let high-frequency echoes bleed through the window boundary.

Two new IIR filter types are available: asymmetric low shelf and asymmetric high shelf, each with independent zero and pole parameters. These are particularly handy for shaping a baffle step with an asymmetric transition, or for implementing a Linkwitz transform to equalise a sealed enclosure.

The Hypex FusionAmp export window has been improved: the patch/create mode is now a proper dropdown rather than a toggle, the channel mapping terminology has been clarified, alignment delays are validated against the 19.2 ms hardware limit before export, and first- and odd-order filters now export as the correct HFD filter types.

On the graph side, a pre-filter overlay is now available for each driver (licensed feature), showing the unfiltered response when a filter configuration window is open. The crossover smoothing behaviour has also been refined: reduction is now proportional to filter slope, with no effect below LR2 and full reduction at LR6 and above.

This release also brings a general interface refresh: boolean options have been replaced by iOS-style toggle switches throughout the application, and several UI labels and panel layouts have been clarified.

A handful of bugs have been fixed, including a crash with order-0 Bessel filters (which doesn't make sense but could appear when changing the order through keyboard arrow keys), a phase alignment issue on off-axis measurements, and a sonogram that would persist after measurements were cleared.