The design of active crossovers- Douglas Self wants your opinions

[Ouroboros]

You might have a carefully designed 3-way active 4th order LR crossover, which keeps the drive signals to all of the speaker units in phase at all frequencies. If the acoustic centres of the drive units in a speaker are not all at the same distance from the listener, then the waveforms radiated by the units will no longer be in phase at the listening position, and will therefore alter the FR by phase addition/cancellation at the crossover points. Of course this will always be the case as you move your head up and down, but it pays to get as close as possible to the ideal, either by having a stepped baffle or by adding the required time delays to acheive the same result.
For those of us who are convinced by 'transient perfect' crossovers (Doug Self is not), then correct time alignment is also required, otherwise even with the best drive units, your carefully aligned crossover circuitry will not produce the desired acoustic response at the listening position.

 

[soongsc]

Since time alignment can only be achieved at a single point, it seems reasonable to question whether "transient perfect" crossovers are better or not. But different design imperfections can mask audibility of others. So if the effect of more dominant imperfections can be reduced one by one, we will find that others will become audible. The tough thing is that when a few things have the same magnitude of effect, you kind of question whether you are doing the right thing or not. But once you get through each of those, you will find a new level of performance.

 

[Bob Richards]

I have a few questions/concerns in regard to Chapter 10: Time Domain Filters. Mr. Self recommends the introduction of time delays into the crossover outputs to correct for the physical alignment of drive units.

Siegfried Linkwitz also makes use of all pass, time delay filters in his analog crossover designs (ASP), and has gone so far as to claim that "Multiple sections may delay the tweeter output and compensate for the driver being mounted forward of the midrange. Active crossover circuits that do not include phase correction circuitry are only marginally useable."

However, it's well known in science that the human ear is very insensitive to phase. Given the complexities of normal listening environments, different sound frequencies will reach your ear through different paths. Per the "The Scientist and Engineer's Guide to Digital Signal Processing," from a physics standpoint, the phase of an audio signal becomes randomized as it propagates through a complex environment.

As expressed from Lenar Audio: "Small shifts caused by vertical crossover alignments are difficult to hear and measure. A trained ear can hear subtle differences of alignment shifts, during real time switching comparison only but not when stationary. That is, the ear is unable to pick which speaker is in alignment, forward or back"

Given Mr. Self's well rationed, objectivist, engineering, and science based approach to audio design in his previous works, he gives little attention to the theory behind the supposed necessity to introduce time alignment compensation. In addition, Mr. Self makes several references to Wikipedia through the text. I find this lacking for a serious engineering reference. At a minimum I would like to see original source references.

Can someone knowledgeable please give a science based answer as to why delay compensation does or does not make the slightest bit of difference for normal listening environments? I'm NOT interested in an audiophile, golden ears based subjective answer...... because I've tried altering the delay in my digital crossovers and I can't tell the least bit of difference in my living room.

Another design factor to consider when implementing analog time delay compensation / equalization: a slight change in listening position will cause a variation in frequency response. Also, any object in the room will alter the standing wave pattern, thus changing the frequency response at any particular location. Time delay compensation must account for listening distance and height, and also room temperature. All these factors are variable across listening environments and people, which makes me question why time delay is useful for any real world domestic situation.

I have to generally agree with the above. I don't listen in an anechoic chamber. It's my opinion that the weakest link in most stereo systems is not the speaker, but the way the speaker interacts with the listening room acoustics. The room WILL, in most cases, make a real mess out of the frequency response and phase response by the time the energy reaches the listener. Comb filter generation off every reflective surface... At higher frequencies the comb filter effects largely fill each other in since the cancellations are closer together and so abundant. At the low frequencies below about 300HZ (depends on the room size) there are so few cancellations due to very few effective reflection paths (due to the wavelengths involved) that you get a particularly uneven frequency response at the listing position which causes excessive boominess... This is why vertical line arrays have better sounding, less boomy bass. What exactly is left of the phase response in the real world? If you do get the perfect phase response somehow, what if you move your head four inches?

 

[Ouroboros]

Obviously, if the drive units are all aligned in a vertical line, then side-to-side head movements have only a small effect on the time alignment. Vertical movements have a large effect at high frequencies, but not to the catastrophic effects that they do with line arrays. Have you ever heard the enormous change in treble between sitting down and standing up with the true ribbon tweeter large Magneplanar panel speakers?

 

[Davey]

What exactly is left of the phase response in the real world? If you do get the perfect phase response somehow, what if you move your head four inches?

Your assertion then is that speaker system designers should not consider or worry about phase response and/or time delay?

Yes, the "real world" is the real world, but everyone's listening room is different...but system designers need to begin with some starting point (any point) and design toward an objective. This means you select a design axis, a measurement distance, a target amplitude response, etc, etc. A target phase response may be one of those objectives.

I'm surprised "greenmo1" can't hear any audible difference when adjusting relative driver position with a digital crossover. In my experience this is easily audible and measurable with typical (multi-way) speakers and testing schemes at typical measuring distances...say 1 meter or so. I understand he may be referring to listening position audibility with music. Yes, more difficult to audibly determine then, but that's not the point, is it?

Cheers,

Dave.

 

[CharlieLaub]

However, it's well known in science that the human ear is very insensitive to phase. Given the complexities of normal listening environments, different sound frequencies will reach your ear through different paths. Per the "The Scientist and Engineer's Guide to Digital Signal Processing," from a physics standpoint, the phase of an audio signal becomes randomized as it propagates through a complex environment.

I think that you are taking a marginally relevant point and running with it. Mostly the direct sound (which predominates) propagates through air, which is not a "complex environment".

The important issue in loudspeakers related to phase is phase "alignment". This is most important when you have more than one driver radiating a particular frequency or frequencies, e.g. within the crossover region(s). Around the crossover point, phase differences exceeding 90 degrees start to cause cancellation and the result is a dip or dips in the frequency response. If this happens at the listening position, then it is audible and you have poorly reproduced audio reproduction. That's about it, in a nutshell.

Mr Self does not address the phase response of the drivers themselves in his book, or even mention this important fact, and this is (in my opinion) a glaring omission in the crossover designs that he presents (he admits he is not a loudspeaker designer). His circuits are technically excellent as far as I am concerned. But he seems to ignore the fact that, unless you are several octaves away from the edge of the driver's passband, the phase response (of the driver alone) is changing with frequency. Mr. Self treats drivers as phase invariant, which is rarely if ever true. Although Mr. Self does use delay to correct for the acoustic center offsets of the drivers, this is actually of secondary concern.

What is really the crux of the matter is phase tracking within the crossover region. A good loudspeaker designer strives to keep the phase responses as close as possible across the entire crossover region. Each driver has a different frequency and (therefore) phase response and these must be accounted for in the crossover design. Typically this requires measurements of the driver response before designing the final crossover. Given that information, there are several tricks (in addition to phase shaping circuits) that can be used to get good phase tracking, even if the time alignment is poor (as you mention, the ear is not sensitive to that). Good phase tracking will allow the on and off-axis responses to be as smooth as possible throughout the crossover region, and that should be goal #1 for the designer since the direct and near direct sound is what dominates at the listening position.

-Charlie

 

[whgeiger]

Analog Bliss

Here we are ignoring the fact that most signal sources available today, reside in the digital domain. Processing them there to produce ideal drive signals for each of a diverse set of loudspeaker drivers would appear to be an intrinsically superior approach. In this setting, the need for cost effective power amplifier designs becomes apparent, as the requirement to drive a muscular woofer is materially different than that required to drive gossamer tweeter. The inherent flexibility in this approach makes it a good choice for the DIY venue as well as for those professionally engaged in loudspeaker development.

WHG

 

[forr]

Each driver has a different frequency and (therefore) phase response and these must be accounted for in the crossover design.

For the high-pass behaviour of a closed driver, this can to be done by a transform which can increase its apparent cut-off frequency. There is currently an interest for tranforms using circuits simpler than the Linkwitz one, see Linear Audio #2 for examples by Marcel van de Gevel.

 

[CharlieLaub]

Each driver has a different frequency and (therefore) phase response and these must be accounted for in the crossover design.

For the high-pass behaviour of a closed driver, this can to be done by a transform which can increase its apparent cut-off frequency. There is currently an interest for tranforms using circuits simpler than the Linkwitz one, see Linear Audio #2 for examples by Marcel van de Gevel.

I have a designs for both a general MFB-based (Friend's circuit) biquadratic filter (like the LT) that uses a single op amp, as well as a 6-op-amp version. The single op amp version allows for a mix of boost and cut, all boost, or all cut, in the response of the circuit and I have done sensitivity studies using Monte Carlo and it looks very usable. I wasn't aware of the Linear Audio article and I will have to track that down.

-Charlie

 

[Bob Richards]

OK, I'll admit it, I did include a time delay section in my 3 way active crossover, so right at the crossover frequency (1.4kHZ), the signals being put out by the midrange driver and the tweeter simultaneously are such that the lobe of maximum energy points straight forward, toward the listener, in theory anyway. But my listening room is such that there are so many reflection paths, and my speakers are open baffle di-poles, so at the listener position, the number of reflections is so high that I don't pretend that I've fixed anything.

What I tend to believe is that unless you are in an anechoic chamber, more room reflections are actually better. The cancellations from any given comb filter mechanism are more likely to get filled in by the presence of other comb filter mechanisms, due to other reflections with different delays. Each reflection path will give you an effectively random additional phase, besides the frequency response anomolies it creates.

In the lower midrange and bass where there are usually only a few effective reflection paths in a typical living room, due to the size of the wavelengths involved, the comb filter cancellations don't usually get filled in very well by the few (if any) additional reflection paths. Having many reflections is arguably a step away from "fidelity", but in the real world it appears to be a tradeoff with frequency response at the listener position, and I think FR is more audible and important than phase or even a certain amount of what could be called room ringing, from all the additional reflections, as long as it's relatively consistent over frequency. The way I see it, the real game is how do you work with the room?

 

[greenm01]

Time-Delays

Your assertion then is that speaker system designers should not consider or worry about phase response and/or time delay?.

Yes, In a nutshell.

I'm surprised "greenmo1" can't hear any audible difference when adjusting relative driver position with a digital crossover. In my experience this is easily audible and measurable with typical (multi-way) speakers and testing schemes at typical measuring distances...say 1 meter or so..

Really? I've tried, but can't tell a bit of difference between a 328 microsecond time-alignment delay and a crossover with zero delay in my listening space. I challange you to detect an audible difference in a blind subjective test, with music. I seriously believe that time aligned speakers are pure maketing hype. A delay is trivial to implement digitally, although in an analog signal processor (ASP) the added circuitry and complexity do not justify the cost.

Floyd Toole, in his book "Sound Reporoduction, Loudspeekers in Rooms" addresses this issue in section 18.6.1:

"Delays are non-minimum-phase phenomena. In the crossover regions, where multiple transducers are radiating, the outputs can combine in many different ways depending on the orientation of the microphone or listener to the loudspeaker..... The result is that if one chooses to design a loudspeaker system that has linear phase, there will be only a very limited range of positions in space over which it will apply. This constraint can be accommodated for the direct sound from a loudspeaker, but even a single reflection destroys the relationship."


He also makes the important assertions (direcly quoted):

1. Because of reflections in the recording environment there is little possibility of phase integrity in the recorded signal​
   ​
   ​
   ​
2. There are challenges in designing loudspeakers that can deliver a signal with phase integrity over a large angular range
   
3. There is no hope of it reaching a listener ia normally reflective room
   

"All is not lost, though, because two ears and a brain seem not to care."

Even more to the point:

"There is quite general agreement that with music reproduced through loudspeakers in normally reflective rooms, phase shift is substantially or completely inaudible."

I understand he may be referring to listening position audibility with music. Yes, more difficult to audibly determine then, but that's not the point, is it.

But, isn't that exactly the point? I listen to music, not log sine sweeps or pink noise.

Differences in phase are rarely corrected in the recording environment, and would be extremely difficult to perfect regardless. When I go to the Symphony the instruments are situated in different positions across the soundstage, and have wide varation in z, y, z coordinates from my seating position.... But, the sound is still absolutely fantastic without any sort of time alightment, which would be virtually impossible to correct for in any regard. And if one used advacned DSP processing to get an exact phase match between individual instruments, you wouldn't be able to tell a difference regardless..... No time alignment necessary in the real world.

 

[Davey]

Well, there are a variety of opinions on this subject. And it's been hashed to death on various forums for years.

There are multiple (well respected) speaker system designers and many audiophiles that feel time alignment and/or phase coherency are necessary design elements that (if not included) diminish the system performance.

If you don't believe it's necessary for your setup/designs, then don't worry about it.

Cheers,

Dave.

 

[Bob Richards]

I love what Greenm01 had to say two comments above this one.

To be fair though, It may be arguable that in the lower midrange (90 - 1kHZ) where we detect image location more by timing comparisons than by amplitude comparisons, a consistency of phase response between the speakers over frequency may be worth worrying about.

Another point that I believe is true and usually overlooked is the phase response of any speaker driver when you get within an octave (or so) of where it's amplitude response starts rolling off, before you've even looked at what the crossover will add to that. Tone bursts showed me some wicked nasties on relatively good drivers, when we got within about an octave of where they rolled off. Yet many speakers take a driver right up to it's amplitude rolloff, and even use that as part of the crossover, and I've heard cases where it actually sounded pretty good...

 

[Bob Richards]

The best argument I know of for having phase correction in speaker electronics is where you have two drivers putting out the same signal at the same time, but physically displaced (the tweeter being a few inches above the midrange driver, for example). Right at the crossover frequency the two acoustic signals will add at the listeners ears, but likely with some phase shift, which is effectively differential delay at that frequency. It seems best if the projection of maximum acoustic amplitude is straight forward out in front of the speaker right at the crossover frequency (the two signals should add in phase when measured from the listeners ear location, X, Y and Z axis), rather than shooting it's maximum amplitude addition up toward the ceiling or toward the floor due to a phase/timing differential. Straight forward would likely stimulate room reflections less, relative to the direct acoustic signal. In some rooms, some listeners might think it sounds better when there is the higher room reflection energy relative to direct. Room acoustics are often a "wildcard". If the real goal is enjoyment, it's hard to say if this correction would improve anything, or if it's even audible. I use all 4th order LR active crossovers to minimize the range of frequencies where this could be an issue. With a one pole crossover, you might hear a difference, but whether or not it makes it better or worse may be a crapshoot.

 

[spladski]

I have a speaker design sitting somewhere in storage for reference. With a pair of speakers for stereo, the central image sits 4ft above the top of the speakers. The sound is not distorted. So tell me which aspect of speaker performance is doing this. This is not a sole incidence. Others have reported similar.

 

[pinkmouse]

Most likely psychological. The human ear does not differentiate height information very well, we allow conditioning to do that for us - if we hear a birdsong, we look up, as experience shows most birds are up in trees or flying, but if we hear a car horn, we look around, as prior knowledge tells us that cars don't fly or burrow underground.

 

[Bob Richards]

I've heard that a small peak in the frequency response around 8kHZ makes a sound seem to come from above the speaker. The exact frequency probably varies with each individual. Apparently the way the ear canal/pinea affects the frequency response depending on the angle, is something we are sensitive to. The 8kHZ energy may need to be compared to energy at other frequencies for the perception to work (a learned response?). I first heard about this back in the 1980's at Dolby Labs where I worked at the time. I also read it in a paper a few months ago. One of the latest Dolby surround sound synthesizers (ProLogic IIz I think) has now created the option of having "height" speakers, that are positioned above the front left and right speakers. I haven't heard this, but I'd guess that they are using this phenomenon there.

 

[forr]

I have a speaker design sitting somewhere in storage for reference. With a pair of speakers for stereo, the central image sits 4ft above the top of the speakers. The sound is not distorted. So tell me which aspect of speaker performance is doing this. This is not a sole incidence. Others have reported similar.

Most likely psychological. The human ear does not differentiate height information very well, we allow conditioning to do that for us - if we hear a birdsong, we look up, as experience shows most birds are up in trees or flying, but if we hear a car horn, we look around, as prior knowledge tells us that cars don't fly or burrow underground.

There was in Electronics World a paper by Edeko about this effect.
It seems that not much can be done to counteract this effect.
Vertical emitting areas (ribbons) could help.

 

[CharlieLaub]

I have a speaker design sitting somewhere in storage for reference. With a pair of speakers for stereo, the central image sits 4ft above the top of the speakers. The sound is not distorted. So tell me which aspect of speaker performance is doing this. This is not a sole incidence. Others have reported similar.

I would be interested to know if the same effect happens if you set up the speakers in a different listening environment, e.g. differently shaped room and/or with a different amount of adsorbing material. My guess is that the room is reflecting some energy back to your listening position that generates a comb filtering type effect and this generates an ambience effect, giving the illusion that the soundfield is high.

Room response can be very, very influential on the sound of a loudspeaker!

Try it and report back please!

-Charlie

 

[greenm01]

Room response can be very, very influential on the sound of a loudspeaker!

Try it and report back please!

I agree 110%. You could also try listening to the speakers, in the same room and setup, with digital room correction (DRC).

I've added DRC-FIR to my system, via. MiniDSP's OpenDRC box, and the result is quite impressive, and subjecively audible.