How to build a flat-phase DSP preset?

Re: How to build a flat-phase DSP preset?

Bennett Prescott said:
Silas,

Now I'm tempted to make a post about building a flat phase crossover, which is not so hard with any decent DSP. Let me see if I can get someone to take a few measurements for me...
Click to expand...

Hi Bennett, Silas, et. al.

I'm curious to understand what is meant by the desire for a flat-phase speaker or crossover? If we interpret the phrase as a loudspeaker that has no phase variation over it's whole frequency range, this is nigh impossible for a multi-way loudspeaker.

Firstly, if we consider a multi-way loudspeaker with utopian crossovers that have no phase shift and where the drivers are time aligned, the loudspeaker will only exhibit flat phase in one spatial direction (or plane). At all other listening positions, there is a slight difference in time arrival which is equivalent to non-flat phase. (Even the use of linear phase brick wall FIR crossovers can't prevent this. But they can reduce off axis lobing.)

Secondly, if we look at all IIR filter based crossovers, of which Linkwitz-Riley is one, they all have a phase that varies with frequency.

Thirdly, the use of IIR filter and minimum-phase FIR filter EQ to fix irregularities in a speaker's magnitude response will result in phase shifts - some minor and some major, depending on the choice of filter. Linear phase FIR filters are the exception but they have longer delay and can have audible pre-ringing.

I suspect the interest is just to have a multi-way loudspeaker where every driver exhibits positive pressure for positive voltage, where the drivers are time aligned, where the crossovers have a low amount of phase shift variation with frequency, and where the overall phase is consistent (for a given model of cabinet) so that cabinets can be arrayed. I suggest we're already there....

On the question of the audibility of phase, I'm not aware of evidence that people can detect absolute phase variations across frequency in a typical well designed single loudspeaker. However people can detect relative phase differences between two loudspeakers or sources. This can show up as frequency response irregularities, comb filtering, imaging problems and the like. (I remember evaluating a pair of studio monitors where in the crossover region, one loudspeaker was about 100 degrees different to the other. There were stereo imaging and level problems which were confirmed with test gear. We sent them back!) There is evidence from ABX testing that some people can disguish a difference between a positive signal and an inverted signal. (Some instruments like saxophones have a very asymmetric pressure waveform.) And there is evidence that time delay between drivers is problematic, but as I mentioned, we can already time align drivers with current speaker processors.

Best,
Michael
 
Last edited:
Re: How to build a flat-phase DSP preset?

If i have to make a DSP preset for a 2 or 3 (or more) way loudspeaker i try to get the phase response as flat as possible for several reasons .
The most important 1 for me is that a loudspeaker with a "flat phase" response is easy to combine with other loudspeakers with other loudspeakers with a not so nice phase response by the use of all-pass filters .
Also a flat phase response (and a "semi" flat frequency response) somehowe makes it more easy to listen too for me anyway (i do not know why maybe it's just me) .

If i get a system like this :
slmhdone.png
I would be pretty happy

This is a 4-way system including subs measured near field and corrected with the use of all-pass filters (do not remember how many but probably 8 or 10 and the sub allignment was done via all-pass filters not delay) as described @ my blog .
 
Last edited:
Re: How to build a flat-phase DSP preset?

Michael Smithers said:
Hi Bennett, Silas, et. al.

I'm curious to understand what is meant by the desire for a flat-phase speaker or crossover? If we interpret the phrase as a loudspeaker that has no phase variation over it's whole frequency range, this is nigh impossible for a multi-way loudspeaker.

Firstly, if we consider a multi-way loudspeaker with utopian crossovers that have no phase shift and where the drivers are time aligned, the loudspeaker will only exhibit flat phase in one spatial direction (or plane). At all other listening positions, there is a slight difference in time arrival which is equivalent to non-flat phase. (Even the use of linear phase brick wall FIR crossovers can't prevent this. But they can reduce off axis lobing.)
Click to expand...
This is correct for loudspeakers that have offset drivers (like most loudspeakers).

However there are some loudspeakers that do not exhibit combfiltering as you move around the coverage pattern. Ie having the same distance between drivers at all listening positions.

With these-getting a good phase response at every seat (at the same time) is possible.

I would agree that with most loudspeakers you are correct.

As the old saying goes "align to where? Different positions will provide a different alignment".
 
Re: How to build a flat-phase DSP preset?

Michael Smithers said:
Hi Bennett, Silas, et. al.

I'm curious to understand what is meant by the desire for a flat-phase speaker or crossover? If we interpret the phrase as a loudspeaker that has no phase variation over it's whole frequency range, this is nigh impossible for a multi-way loudspeaker.

Firstly, if we consider a multi-way loudspeaker with utopian crossovers that have no phase shift and where the drivers are time aligned, the loudspeaker will only exhibit flat phase in one spatial direction (or plane). At all other listening positions, there is a slight difference in time arrival which is equivalent to non-flat phase. (Even the use of linear phase brick wall FIR crossovers can't prevent this. But they can reduce off axis lobing.)

Secondly, if we look at all IIR filter based crossovers, of which Linkwitz-Riley is one, they all have a phase that varies with frequency.

Thirdly, the use of IIR filter and minimum-phase FIR filter EQ to fix irregularities in a speaker's magnitude response will result in phase shifts - some minor and some major, depending on the choice of filter. Linear phase FIR filters are the exception but they have longer delay and can have audible pre-ringing.

I suspect the interest is just to have a multi-way loudspeaker where every driver exhibits positive pressure for positive voltage, where the drivers are time aligned, where the crossovers have a low amount of phase shift variation with frequency, and where the overall phase is consistent (for a given model of cabinet) so that cabinets can be arrayed. I suggest we're already there....

On the question of the audibility of phase, I'm not aware of evidence that people can detect absolute phase variations across frequency in a typical well designed single loudspeaker. However people can detect relative phase differences between two loudspeakers or sources. This can show up as frequency response irregularities, comb filtering, imaging problems and the like. (I remember evaluating a pair of studio monitors where in the crossover region, one loudspeaker was about 100 degrees different to the other. There were stereo imaging and level problems which were confirmed with test gear. We sent them back!) There is evidence from ABX testing that some people can disguish a difference between a positive signal and an inverted signal. (Some instruments like saxophones have a very asymmetric pressure waveform.) And there is evidence that time delay between drivers is problematic, but as I mentioned, we can already time align drivers with current speaker processors.

Best,
Michael
Click to expand...

Michael,

You make many valid points about limitations on implementing linear phase designs. There is much to write on the various aspects of this topic. The link below is an excellent overview document to address many of your questions:
http://www.grimmaudio.com/whitepapers/speakers.pdf
 
Re: How to build a flat-phase DSP preset?

Phil Graham said:
Michael,

You make many valid points about limitations on implementing linear phase designs. There is much to write on the various aspects of this topic. The link below is an excellent overview document to address many of your questions:
http://www.grimmaudio.com/whitepapers/speakers.pdf
Click to expand...

Thanks very much Phil. The author describes the use of a non-causal filter to correct for the Linkwitz-Riley crossover phase shift. Something I'd completely missed. This will add some delay though, and the accuracy of the filter is proportional to the delay. (The "non-causal" filter has to be truncated at some point to realisable.)

Best,
Michael
 
Re: How to build a flat-phase DSP preset?

Ivan Beaver said:
This is correct for loudspeakers that have offset drivers (like most loudspeakers).

However there are some loudspeakers that do not exhibit combfiltering as you move around the coverage pattern. Ie having the same distance between drivers at all listening positions.

With these-getting a good phase response at every seat (at the same time) is possible.

I would agree that with most loudspeakers you are correct.

As the old saying goes "align to where? Different positions will provide a different alignment".
Click to expand...

Thanks Ivan. Yep just after I hit post, I figured synergy horns might come up. Looking at the SH50 specs, the phase is excellent and has minimal change - only about 360 degrees across the whole frequency range. My question to the OP was, what is meant by the desire for "flat phase." What I see from everyone's posts is a desire for minimal phase change. With the synergy horns, I am really curious how the passive crossover phase change is dealt with. I guess that's somehow related to the acoustic properties of the relative offsets of the drivers on the horn. :)

Best,
Michael
 
Re: How to build a flat-phase DSP preset?

Michael Smithers said:
Hi Bennett, Silas, et. al.

I'm curious to understand what is meant by the desire for a flat-phase speaker or crossover? If we interpret the phrase as a loudspeaker that has no phase variation over it's whole frequency range, this is nigh impossible for a multi-way loudspeaker.

Firstly, if we consider a multi-way loudspeaker with utopian crossovers that have no phase shift and where the drivers are time aligned, the loudspeaker will only exhibit flat phase in one spatial direction (or plane). At all other listening positions, there is a slight difference in time arrival which is equivalent to non-flat phase. (Even the use of linear phase brick wall FIR crossovers can't prevent this. But they can reduce off axis lobing.)

Secondly, if we look at all IIR filter based crossovers, of which Linkwitz-Riley is one, they all have a phase that varies with frequency.

Thirdly, the use of IIR filter and minimum-phase FIR filter EQ to fix irregularities in a speaker's magnitude response will result in phase shifts - some minor and some major, depending on the choice of filter. Linear phase FIR filters are the exception but they have longer delay and can have audible pre-ringing.

I suspect the interest is just to have a multi-way loudspeaker where every driver exhibits positive pressure for positive voltage, where the drivers are time aligned, where the crossovers have a low amount of phase shift variation with frequency, and where the overall phase is consistent (for a given model of cabinet) so that cabinets can be arrayed. I suggest we're already there....

On the question of the audibility of phase, I'm not aware of evidence that people can detect absolute phase variations across frequency in a typical well designed single loudspeaker. However people can detect relative phase differences between two loudspeakers or sources. This can show up as frequency response irregularities, comb filtering, imaging problems and the like. (I remember evaluating a pair of studio monitors where in the crossover region, one loudspeaker was about 100 degrees different to the other. There were stereo imaging and level problems which were confirmed with test gear. We sent them back!) There is evidence from ABX testing that some people can disguish a difference between a positive signal and an inverted signal. (Some instruments like saxophones have a very asymmetric pressure waveform.) And there is evidence that time delay between drivers is problematic, but as I mentioned, we can already time align drivers with current speaker processors.
Click to expand...

Michael,

I find speakers with minimal phase shift sound better, and are certainly easier to align with other speakers. Perhaps I am fooling my bronze ears, but knocking a few phase rotations out of a system sure makes it sound better to me. Of course, we're talking gross discrepancies, not 30° here and there. Certainly there are compromises to be made, but reasonably good designs can be made reasonably phase flat for most of their coverage pattern. Certainly a 12" woofer and a silk dome tweeter crossed at 3kHz are going to exhibit lots of warping even slightly off axis, no matter how flawless they may be on axis.

What I am referring to in my post is not utopian. It can be implemented with standard crossover filters and a little delay, with the help of a loudspeaker design that allows for significant overlap between passbands. This is not so unusual, there is always some overlap or else crossovers wouldn't work, so another half an octave between friends on either side of the "crossover point" is only a big deal in pretty marginal designs. The result has some magnitude ripple and is nothing like textbook, but it does maintain the impulse response pretty well and I think sounds significantly better. If you're really clever (and I sometimes am by accident) you can even use the overlap to increase power response and reduce pattern dissimilarities between passbands.

I'm not sure that the interest is to have a multi-way system where every driver exhibits positive pressure for positive voltage. I don't believe this is a requirement for a low phase shift system. Perhaps I am wrong.
 
Re: How to build a flat-phase DSP preset?

Perhaps some one can shed some light on pre-ringing. I'm aware there are various methods of minimizing rining but never bothered to investigate them in detail. I was stuck with what my Lake Contours could do…. so there wasn’t much point investigating further. (They offered brick wall and 24 / 48 dB FIR linear phase options as well as the standard IIR filters.)

Most of the stuff I can find, is about filters used in DAC’s

http://www.wolfsonmicro.com/documents/uploads/misc/en/Ultra_High_Performance_DAC_whitepaper.pdf

I may be wrong, but I also understand that pre-ringing in adjacent pass bands is equal and opposite, so it should be possible to cancel it acoustically.

These guys (below) seem to suggest you can … (Q9)

http://webcache.googleusercontent.com/search?q=cache:gvTYh2VKX88J:http://www.deqx.com/techquest.php+l....266.1280.0j1j5.6.0...0.0.Ce8no6dW7B4&ct=clnk

The plots I posted above used FIR filters, although everything hung together on and off axis very well, the polar response was not so good, and it had a total latency of about 17.7ms . The "production version" of the settings used IIR filters for the sub crossover and 24 & 48 dB linear phase filters for the mid and HF crossover. The FIR time was only 2.5ms.

It’s relatively easy to produce a “flat” frequency and phase response at one point, the hard bit is getting it all to work on and off axis as well maintaining good pattern control through the crossover region. As Bennett, mention you can overlap bands to control directivity … here is one method …

http://www.four-audio.com/data/AES32/html/img15.html
 
Last edited:
Re: How to build a flat-phase DSP preset?

Bennett Prescott said:
Michael,

I find speakers with minimal phase shift sound better, and are certainly easier to align with other speakers. Perhaps I am fooling my bronze ears, but knocking a few phase rotations out of a system sure makes it sound better to me. Of course, we're talking gross discrepancies, not 30° here and there. Certainly there are compromises to be made, but reasonably good designs can be made reasonably phase flat for most of their coverage pattern. Certainly a 12" woofer and a silk dome tweeter crossed at 3kHz are going to exhibit lots of warping even slightly off axis, no matter how flawless they may be on axis.

What I am referring to in my post is not utopian. It can be implemented with standard crossover filters and a little delay, with the help of a loudspeaker design that allows for significant overlap between passbands. This is not so unusual, there is always some overlap or else crossovers wouldn't work, so another half an octave between friends on either side of the "crossover point" is only a big deal in pretty marginal designs. The result has some magnitude ripple and is nothing like textbook, but it does maintain the impulse response pretty well and I think sounds significantly better. If you're really clever (and I sometimes am by accident) you can even use the overlap to increase power response and reduce pattern dissimilarities between passbands.

I'm not sure that the interest is to have a multi-way system where every driver exhibits positive pressure for positive voltage. I don't believe this is a requirement for a low phase shift system. Perhaps I am wrong.
Click to expand...


Thanks very much. You're definitely not wrong. :) This and Timo's blog are very, very helpful. (I completely retract the "nigh impossible.")
 
Last edited:
Re: How to build a flat-phase DSP preset?

Michael Smithers said:
Thanks very much. You're definitely not wrong. :)

This and Timo's blog entry showing crossover and phase adjustment for a 12" 2-way are very helpful.

I completely retract the "nigh impossible."
Click to expand...

Michael,

While you are investigating this topic, you should read Vanderkooy and Lipshitz's paper "A Family of Linear-Phase Crossover Networks of High Slope Derived By Time Delay" in J. AudioEng. Soc.,Vol. 31, No. 1/2 1983 (and also in the loudspeaker anthologies).
 
Re: How to build a flat-phase DSP preset?

Michael Smithers said:
With the synergy horns, I am really curious how the passive crossover phase change is dealt with. I guess that's somehow related to the acoustic properties of the relative offsets of the drivers on the horn. :)

Best,
Michael
Click to expand...

Here is a link that talks about the basic concept-in simple terms.

http://www.danleysoundlabs.com/danley/wp-content/uploads/2012/01/The-Tapped-Horn.pdf

It all STARTS with having the drivers in the proper position to each other-not only to keep the expansion rate correct-but also the physical distance between them for the particular passbands.

This makes the crossover design a bit easier.

Since the drivers are always the same distance from each other (in different listening positions in the coverage area) the phase and freq response does not have the errors in it that designs that have offset drivers do.

The crossovers in the Danley cabinets are not trival (some more complicated than others). For example in the Sh50-there are basically 4 (kinda) "crossovers" on the high freq driver alone. The highest one is around 17Khz.

There are some 'unusual" things going on-that at first do not make sense-but aid in the overall response. When I first saw it-I told Tom he had messed up on the crossover design. Why would anybody put a 12dB/oct LOWPASS filter on the high freq driver? Well the reason is to ADD more high freq via the crossover. He told me to remove it and see what happened-and being the doubting Thomas that I am-I did. And the HF response went DOWN! ie didn't extend as high. Another one of those learning moments for me.

I won't go any further than that-but we use the same process on a number of our cabinets to extend the top end.

The more "flat" the phase response is-the more the cabinet will sound like a single driver. If you can't see from the phase response where the crossover points are-then that is good indication of a good alignment.
 
Top Bottom