Phase-alignment based method of designing multi-way speakers

A good example is the What is the ideal directivity pattern for stereo speakers? thread - 2200+ posts and counting.

I contributed a bit in similar thread where topic of directivity of dipoles. That was also quite a discussion until I jumped in with some dipole waveguide measurements. Have to work on them more elaborate, though.

Amazingly enough, audio gear seems to follow similar circular trend patterns as those of fashion industry. At the moment trend goes towards SET amps, hornloaded and waveguided speakers typical for nineteen 50-ties and 60-ties, and I believe also for 20-ties 30-ties. At 70-ties and 80-ties it was acoustic lenses and multi-angled drivers. So I wouldn't be very surprised if Karlson's folded pipe with that cute skirt-like frontal cut pops out again sometime soon as a major must-have issue.

I too would be interested to see the results of this method - whether or not I had certain technical or editorial prejudi... err, misgivings.

PRTG - I also commend your patience and restraint.

LafeEric, thanks for your interest and friendly gesture. I was trained for patience at some point of my career, so no problem with that.. most of the time.
 
PRTG - are you going to create such a speaker? Please go on and show us your results!

Thanks Juhazi! All I needed was only one vote to go for it.

Ok, lets go with a minimalistic two-way speaker. Here is a plan.

I'll start with an easy (naturally matching) HF/LF speakers. Later as a side quest I'm going to do the same with a more difficult pair (only if it credible, clearly advantageous results are achieved with the "easy" speakers).

Now lets set target measurement values for the goal. Here is my proposal:

a) for phase-aligned method relative electrical phase shift should not exceed 7 degrees within 250Hz-12kHz range. No target for "typical" method is set here, still phase curves for drivers + filters are recorded for reference;
b) target f response to have maximum +3dB/-6dB on-axis deviation against ruler flat (in region mentioned);
c) Directivity will be measured with no target goals set. Results will be shown and analyzed after listening just to provide some food for thought;
d) Distortion values - same as c).

Measurement gear: licensed version of ARTA/LIMP + measurement box, Presonus FireStudio Mobile FireWire soundcard, Behringer ECM8000 mic, stereo gainclone "classic" amplifier (non-inverted), LCR multi-meter.

F response measurements will be performed outdoors with measurement SPL level 90dB @1kHz 3 meters away from speakers. Listening tests will be done at about the same SPL level but from 5m distance with distance between speakers being 2 meters and speakers turned at about 10 degrees inwards. SPL will vary from chosen material but will be kept within low-distortion range.

Now follows building plan for pair of test speakers.

1. First part of method (acoustical phase matching) will be done regardless of the second (filter building with and without electrical phase matching). First part involves creation of moderately wide baffle, HF speakers with waveguides, driver acoustic centers aligned vertically for first approximation).

2. Two different passive filters will be built by using different methods.

2.1 Phase-alignment method:

a) measure and study driver f response, impedance, and electrical phase curves;
b) add L-pad to match average sensitivity of drivers;
c) add resonance damping RLC notch to align the phase curves of drivers in range around HF driver's resonance towards set target phase alignment value.
d) add and tune Zobel or RLC filter for LF driver to align the phase curves of drivers at upper midrange and HF range towards set target values. Return to e) if d) makes negative impact on phase alignment made in c), re-tune both to achieve set target value;
f) re-measure f response and impedance curves of drivers;
g) choose crossover frequency at about half-octave above Fo of HF driver;
h) read driver impedance values at chosen crossover point;
i) calculate and apply 1-st order filters based on g) and h) values;
j) re-measure summed f response;
k) if response variation exceeds target value apply additional notch/shelving filter(s) to whole speaker;
l) make final summed phase adjustment by re-positioning of HF driver under pink/white noise.
m) re-measure summed f response to be sure it's still compliant against target values. If no, return to k) and tune parameters, if yes, proceed to blind tests.

2.2 "Typical" method:

a) measure and study driver f response and impedance curves;
b) add L-pad to match average sensitivity of drivers;
c) add Zobel for LF driver to achieve flat Z in HF region;
d) re-measure f response and impedance curves of drivers;
e) choose crossover frequency about 2 octaves above Fo;
f) read driver impedance values at chosen crossover point;
g) based on e) and f) calculate and apply 1-st order filters;
h) measure summed f response;
i) if response variation exceeds target values then correct LF Zobel to achieve the flattest f response. Restart at f) as we need to recalculate filter for LF to match impedance change at crossover frequency. Proceed further when target f response values are reached or changing Zobel/HF filter gives result that is closest to them;
j) Apply additional notch/shelving filter(s) to whole speaker if f response still doesn't achieve target values;
j) make final summed phase adjustment by re-positioning of HF driver under pink/white noise.
k) re-measure summed f response to be sure it's still compliant against target values. If no, return to i) and tune parameters again, if yes, proceed to blind tests.

So I'll leave the plan open for a week open for discussions and suggestions. Then I'll finalize it and proceed to the actual building.

Meanwhile I'll prepare similar plan on blind testing methodology.

Lets have fun! :)
 
...Amazingly enough, audio gear seems to follow similar circular trend patterns as those of fashion industry.

That is because people seem to forget that the wheel has already been invented, refined, refined again, to the point where it is no longer recognized as a wheel. Thus, people come around and say, "Look what I discovered" and the process repeats itself. It's called short attention span and not learning from history.

Good luck with your endeavor, but it's all been done before, with varying degrees of success.
 
Hi,

The described "typical" way of designing a speaker is simply wrong,
its not the way its done at all. Neither is the "new" way of doing it.
Two wrongs don't make a right.

rgds, sreten.


So, it's a wrong, but not a new way of doing it? In other words, it is a persistant, but erroneous custom amongst loudspeaker designers? So, at least you are not blaming PRTG for introducing a new fallacy.

I am happy with everybody who is willing to invest time and effort in understanding how to design loudspeakers. I would like to commend again PRTG for his effort to put his thoughts on paper. He also invites critique -quo- That's about it! Questions, corrections and additions are welcome -unquo-. That's appropriately humble and almost scientific, to invite peer review.

I am disappointed with the tune adopted by some who consider themselves a step higher on the learning curve. I am convinced that loudspeakers designed by the method proposed by PRTG will come out all right. He may be overdoing it a bit with the design goal to have identical group delay of all drivers at all frequencies, but it will do no harm either.

Vinigar is nice to spice up a salad, but excreting it in public doesn't look good on anybody.

vac
 
Hi,

Quite frankly you are insulting those who have all been there and done
it before. Your rather large posts are are condescending / insulting to
those who have worked it all out before you, and much more, and
your perception of two separate approaches is just stunningly dull.

Personally I just don't get why would somebody be insulted if someone finds something already found and brings it up from debris of forgotten once again. If indeed somebody had described such method before and found it useful/useless, I'd be more than happy to drop the burden of building and testing and comparing. So please refer to the particular articles or threads so we can get into details and sort it out and eventually have happy ending for everybody :)

Regarding dullness - if you have found any errors already, please point them out. Remember that method two taken from reference (with minor improvement as I will use measured impedance values for x-over calculation not some predefined constants) in the beginning of the first post I personally find flawed and it is there for comparison's sake only.
 
....and almost scientific, to invite peer review.

vac

Hi,

Scientific peer review can be very vicious and can be very dismissive.

What is the crux of this thread when you sort the wheat from the chaff ?

Its being implied there is something sour in my attitude, perhaps there
is, that is because the presentation, full of assumptions and rules of
thumb, doesn't lend it self to simple useful comment.

AFAICT some alleged "typical" methodology of ignoring the tweeter Fs
impedance peak and choosing a x/o point of 4 x tweeter Fs is to be
replaced by a "new" methodogy of compensating the tweeter Fs peak
and using a x/o point of 1.4 x tweeter Fs as a better methodology.

It might be a better way of doing it if the "typical" methodology is
true, but it simply isn't, speakers aren't designed that way. Baffle
step is accounted for in the x/o, zobels are only used if needed,
mid and treble Fs impedance peaks are compensated if needed,
and x/o points are not chosen simply based on driver Fs's.

The holes in the assumptions are there to make the conclusion seem
reasonable, and that is simply not how it actually works scientifically.

rgds, sreten.
 
AFAICT some alleged "typical" methodology of ignoring the tweeter Fs
impedance peak and choosing a x/o point of 4 x tweeter Fs is to be
replaced by a "new" methodogy of compensating the tweeter Fs peak
and using a x/o point of 1.4 x tweeter Fs as a better methodology.

It might be a better way of doing it if the "typical" methodology is
true, but it simply isn't, speakers aren't designed that way. Baffle
step is accounted for in the x/o, zobels are only used if needed,
mid and treble Fs impedance peaks are compensated if needed,
and x/o points are not chosen simply based on driver Fs's.

Thanks Sreten, I found your comment very useful. "Typical" methodology was actually hard to describe as there are a lot of general recommendations floating around that are correct only to some degree. I'd re-define that "typical" approach states that flat F response is the only things that matters for good sounding speaker.

- BS is omitted intentionally because wide-enough baffle is used to get room help as already stated in introductory part;
- In my version of "typical" method Zobel is used along with its tuning against target frequency response flatness, so it also contains possibility of omitting it completely;
- Mid and treble Fs impedance peaks are compensated if needed - we need some criteria of why would we do it if flat F response is the only that matters;
- Sure, filter's x/o points cannot be chosen simply based on driver Fs's, but see my comment about general recommendations. But I see your point. So to have less degree of "apples to oranges" I'm going to change "Typical" so that x-over point will be chosen at the same F as for phase-aligned method.

How does that sound?
 
Regarding recent cross-accusations my call is to keep the focus on topic. I don't see need to argue on why who wrote whom what way ago. We are underreacting and overreacting on lots of things everyday and for very personal reasons, so i'ts barely enough to think about our own attitude which gets easily missed when we start bothering about other's (like I'm doing right now). If somebody will be interested in outcome of this thread lets make it shorter to read, so we'll be saving a lot of time (most valuable asset for a life). Thanks! :)
 
Regarding recent cross-accusations my call is to keep the focus on topic. I don't see need to argue on why who wrote whom what way ago. We are underreacting and overreacting on lots of things everyday and for very personal reasons, so i'ts barely enough to think about our own attitude which gets easily missed when we start bothering about other's (like I'm doing right now). If somebody will be interested in outcome of this thread lets make it shorter to read, so we'll be saving a lot of time (most valuable asset for a life). Thanks! :)
So sorry PRTG. I don't mean to appear argumentative, but I am genuinely curious just what sreten was on about. It must have been VERY important - it certainly sounded like you were hammering a stake into his soul - and in the interests of avoiding his cross hairs in the future, if that is possible, I thought I would make some inquiries.

I'm sure the normal bedlam will quickly resume after sreten clears this all up. :)
 
PRTG,

I would think you tend to illustrate things more complicated than they are.

Basically you have the raw in-situ driver's IR's, impulse responses, (that can be displayed as magnitude + phase response vs frequency), preferably with frequency dependant windowing applied, combined/merged with close field measurements for the LF stuff. These IR's also must contain time of flight differences (use HolmpImpulse to get there without fiddling).

Then you simply must find the correcting filter function (and then implement it -- which is the section you put emphasis on) to acheive your choosen total and complete driver response, which not only does contain the XO but also the XO of adjacent sections (or the final LF rolloff). These correcting filter functions may contain additional low/high passes and/or allpass sections (things get tricky when the tweeter is ahead of the woofer, though). See this post for an (analytical) example of time-of-flight compensation.

The core question is now if 0deg phase-matched responses (of final acoustical response) of all driver sections throughout their whole operating (down to -30dB or so) is acoustically to be preferred or not. I personally do prefer it (and I think this is generally agreed on), for technical reasons like lobing and distortion, except for D'Appolito type of arrangements were 0deg tracking phases doesn't work well (you need constant 90deg phase offset there).


For example I recently designed (the electrical part of) an active sub+sat system which was planned to work along the THX-guideline of a "textbook" LR4 XO at 80Hz. The sats were ported and highpassed (8th order total), so resulting IR was obviously only able to partially follow the LR4 reponse. But of course the sub has its own LF-rolloff (6th order, at ~35Hz) and I managed to tailor the additional phase from the 8th order sat alignment to complement the additional phase from the woofer's LF-rolloff for a perfect 0deg phase tracking (+-5deg down to -30dB points) with almost equally perfect flat summing. The subwoofer's HF roll-off, when viewed in isolation, is perfect LR4, which required the woofer passband being EQ'd flat up to 500Hz and then the textbook LR4 applied (which had to be variable from 50...120Hz as a design requirement)

Note that raw reponses were measured only once, all the rest was done with simulators (LTspice mostly, and some LSPcad), then built. I happened to land spot on when finally measuring the thing.
 
Last edited:
KSTR, thanks for suggestions, I see your point. While these things are relatively easy to do with I'm aiming for minimalistic as possible passive filtering here for usage with single SET amp (quite a a niche, still many people enjoy it). Thanks for pointing to HolmImpulse and time-of-flight measurements. I was going to analyse mostly with group delay available in ARTA.
 
...I'm aiming for minimalistic as possible passive filtering...

PRTG: I noticed this before, and now that you mention it again I thought I would comment. You stated that you want to use "first order" filtersin your first post. Please keep in mind that first order filters, especially passive ones, are not very good for loudspeaker crossovers because they do not sufficiently attenuate the signal to the drivers. If you do not attenuate the signal to a woofer enough at higher frequencies the breakup region will not be sufficiently suppressed. If you do not sufficiently suppress the signal to the tweeter at low frequencies enough, the cone will still be driven to high excursions (for a tweeter). Both of these will increase distortion, and for the tweeter, more power than necessary will be sent through the voice coil, which will cause it to heat up more than needed. In general, first order filters lead to a rash of bad thing, and this is why they are not typically used in high performance speakers. Occasionally you see it, but the designer must really know how to address (and measure) the problems that can occur in order to make sure that they are controlled and minimized.

You might want to keep that in mind. If you are planning on using ARTA, use STEPS to measure distortion. Do some modeling of the tweeter excursion to see how power you can apply before reaching Xmax. Using a tweeter with a resonance frequency closer to where you will cross it over will help, because the tweeter's own rolloff will decrease excursion. On the other hand, this region of the tweeter's response typically has higher harmonic distortion... and the tweeter's phase response will still be changing in its passband, making the phase matching/tracking harder to achieve... also, flattening the impedance peak at resonance does not change the phase behavior of the driver output, so you will still have that to contend with... there are only difficult tradeoffs to choose from I am afraid.

Speaking of phase, in your first post you mention aligning the tweeter and woofer using a waveguide on the tweeter to "align acoustic phase". Unfortunately this does not align the phase, only minimizes the acoustic delay part of the phase response. Remember, phase = minimum_phase + excess_phase, the excess phase coming from the delay of the signal that results when the distance from the drivers' acoustic centers to the listening position is different for each driver. So, when excess phase is zero, there is still the minimum phase response to consider, which is solely a function of the frequency response of the drivers. Since the woofer and tweeter have different frequency responses, their minimum phase responses MUST be different. This is why, in general, the relative phase cannot be the same everywhere, unless you use some DSP tricks.

-Charlie
 
Thanks, CharlieLaub, I knew someone would touch this sooner or later.

Using a tweeter with a resonance frequency closer to where you will cross it over will help, because the tweeter's own rolloff will decrease excursion. On the other hand, this region of the tweeter's response typically has higher harmonic distortion... and the tweeter's phase response will still be changing in its passband, making the phase matching/tracking harder to achieve... also, flattening the impedance peak at resonance does not change the phase behavior of the driver output, so you will still have that to contend with... there are only difficult tradeoffs to choose from I am afraid.

This isn't quite so. My observations are that when you flatten driver's resonance peak electrically (or mechanically) excursion is minimized so is THD in this region as THD comes from resonant character and not the opposite way. Same can be said about impedance. Electrical correction of impedance peak is matching LRC values to electrical equivalent circuit of driver's mechanical properties to counter-measure its resonance. When it is done f response roll-off gets less steep as Q changes its value, impedance is flattened, THD is lowered and phase curve gets straightened. Actually that's the first and most important point of the method. I believe directivity smoothness must also be improved as my observations are that HF driver's resonance also mean extremely wide dispersion at resonance therefore crippling directivity smoothness. We'll see.

Regarding "high performance speakers" I believe you actually mean capable of playing "loud and clean". First, we don't have defined criteria here (still you're welcome to suggest), second the goal isn't set for such, rather for pleasant listening at moderate levels. With target speaker sensitivity at least ~93dB/2.83V and cleverly designed baffle I don't see much of the trouble here as 1-2W power will make it "quite loud" for in-room listening with minimum of excursion and audible distortion.

For any conditions requiring more power crossing actively with high filter orders and damping tweeter resonance (for reasons explained above) would give the best result. Many would disagree for second part to be reasonable at all, so its yet to be proved unless "this was all already done before" (still waiting for some referrals to related articles). Still active filtering/DSP-based correction/multi-amping is approach that is far from being minimalistic, simple and easy to build, SET amp friendly.

Thanks for suggestions anyway! :)
 
Last edited:
This isn't quite so. My observations are that when you flatten driver's resonance peak electrically (or mechanically) excursion is minimized so is THD in this region as THD comes from resonance-like character not the opposite way. Electrical correction of impedance peak generally gives counter-measuring equivalent circuit of driver's mechanical properties. When it is done f response roll-off gets less steep as Q changes its value, THD is lowered (and output as well), and phase gets straightened. Actually that's the whole point of the method. Plain physics.:)
I don't have much experience with this, though I have been in electronics for years.

It seems to me it might be possible by carefully manipulating the Q of the resonance compensation network to minimize power losses by balancing it's effect. Do you know if this has been tried?
 
I don't have much experience with this, though I have been in electronics for years.

It seems to me it might be possible by carefully manipulating the Q of the resonance compensation network to minimize power losses by balancing it's effect. Do you know if this has been tried?

Have a look, I added a bit to original reply to CharlieLaub, still you were very quick :)

I see that power loss is in place only in R in parallel LRC filter. R value is somewhat 2-4 times Ro for result close to total damping. I'd say energy loss would be about 25-50% in this particular region (a very rough estimate). Also part of energy gets shifted to lower frequencies due changes in slope steepness. So we'll be expecting dip in combined output here if LF driver isn't very capable or if we have tweeter waveguide diameter too small to compensate for it. Still it will be "clean" dip without artifacts that usually accompany resonances.

Regarding balancing it is surely possible to compensate for a lesser degree as a compromise, depends on whether you have to do it at all.