Can't Reproduce a Square Wave.

[cocolino]

Hallo Charles:

thanks for clarifying things!
I think I got it now. So this is a 3 driver filler driver system with the distinction that the filler driver (or maybe better mid filler driver in this case) is feed from a XO which allows the mid driver to operate over a much wider frequency range in contrary to the "classical" filler XO where the filler XO-point coincides with the lowpass respectively the highpass XO-point. Hence this looks much more than any conventional XO but with the potentially benefits to achieve good transient response.
So far so good but I still have some concerns about IM.
You are right about that this would be much better in regards to tweeter IM-distortion but IMHO the same problem about IM is now just moved to the midrange (the "filler") .
As the filler driver (preferable a fullrange as You say correctly) has likely to be a small diameter unit, the 6dB highpass together with its hump don`t prevent the poor driver from larger excursions when "seeing" frequencies near or below its highpass cutoff. This is particular bad since most of this small FR-units do not allow higher linear excursions.

I would very appreciate Your opinion about this:
Why not just use two of the electronic filler driver XO`s in a cascaded configuration as shown below. The individual blocks are the same as shown as the 2. diagram from top in Your "filler_drv.zip" here:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=5435&perpage=15&pagenumber=2
The "delay?" in the tweeter branch may be implemented as an allpass filter to equalize for the time delay introduced by the 12dB highpass (see arrow in drawing) of the midrange (but depending how far the XO-points are sperated or/and due to eventually compensating effects of the tweeters acoustical high pass response it might be left out too) .
I handdraw the theoretical amplitude responses below the block schematic.
The increased woofer FMD due to the less steep low-pass is not so much an issue IMHO but the IM thing for the midrange would be better now. Of course the circuity in series with the midrange and particular the woofer is more complex but overall this design might have some advantage though.

And yes, designations for the filter topologies is actually a good idea! Otherwise we might get lost in the discussions by one misunderstanding following the other.

 

[phase_accurate]

Hi Christoph

I agree about the FR's increased stress when only highpass-filtered with 6dB/octave compared to 2nd order that's why I recommended a FR rather than a conventional midrange which have often less LF capabilities (Xmax and fs) than many fullranges.
OTOH 6dB/octave is way better than running it fullrange, isn't it ?
My intention was not to make an ordinary pulse-response optimized three-way but rather giving relief to a FR at it's upper and lower end.

What you proposed would of course work but is way too complicated. Two cascaded 2nd order HP /1st order LP subtractive crossovers would do the same job (i.e. the same transfer functions) with much less op-amps and capacitors (e.g. four op-amps and 4 capacitors only !).

BTW despite giving more stress to the MSW I will use the 2nd order filter on the woofer and the 1st order on the MSW on my MSW project. Simply because the MSW can take quite some bass (I know somebody using an active studio monitor, used for voice only, with an MSW run fullrange and a small class-a tube amp), also Manger suggests 1st order and I am more afraid of the woofer messing up the midrange than of the MSW messing up the bass.

Regards

Charles

BTW: I don't like allpass filters that much

 

[Jared]

I had an odd thought about a square wave... can you use an array of ultrasonic transducers to create it with constructive and deconstructive interference?
or how about this:
If you had 10 transducers, say 5 turn on in an instant, then at the same time those 5 turn off the other 5 turn on, and then alternate which ones are on for the duration of the + component of the square wave. Then turn 5 on with - polarity at the same time the 5 with + polarity are turned off, to get the falling edge to the negative side, and then alternate which transducers stay on with - signal in order to maintain the flat negative wave shape.
As long as you maintain half of them on in one diretion or another at all times it would be the equivilant of a speaker cone with the DC signal on it, right?

I got this idea from hearing about a type of speaker that reproduced the audible spectrum with high frequency transducers, and I figured they must have a very very quick "slew rate" suuitable for making the rising and falling edges of the square wave.

 

[Nelson Pass]

I would very appreciate Your opinion about this:
Why not just use two of the electronic filler driver XO`s in a cascaded configuration as shown below. The individual blocks are the same as shown as the 2. diagram from top in Your "filler_drv.zip" here:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=5435&perpage=15&pagenumber=2
The "delay?" in the tweeter branch may be implemented as an allpass filter to equalize for the time delay introduced by the 12dB highpass (see arrow in drawing) of the midrange (but depending how far the XO-points are sperated or/and due to eventually compensating effects of the tweeters acoustical high pass response it might be left out too) .
I handdraw the theoretical amplitude responses below the block schematic.

I think you can get the same effect using the "phase coherent" subtractive type filters as outlined in my article on same ( www.passlabs.com ). When you generate the complement to a 12 dB/oct slope, you get the same sort of bump followed by a 6 dB/oct slope. Also, if you use the same sort of filter on the mid/tweeter, I believe you can eliminate the delay circuit.

 

[Circlotron]

I haven't read the whole thread, but my idea is that there is a derivative effect at work here. If you make the speaker cone move in and out in a triangular fashion I imagine you will produce a square acoustic output; the constant velocity of the cone producing the flat top of the square wave, the abrupt change in direction at the corners of the triangle wave making the switch in polarity of the square wave.

If you actually managed to make the cone move in a square movement I think you would get a sudden positive acoustic spike that decays to zero, then a sudden negative one, just like when you feed a very low frequency square wave through an RC high pass filter

 

[phase_accurate]

Graham

In simple words: You've got it right ! The cone has to move in triangular fashion.

But you don't have to take special measures to force it to do so. It will automatically behave like that if the input signal is a squarewave - simply because it is forced to do so by the laws of physics.

Now if we could only make a driver that behaves perfectly over the whole audio frequency spectrum ......

Regards

Charles

 

[jan.didden]

sqw wave

Well, I don't know, if you move the cone in a triangle fashion, you get constant speed, which has as a derivative a zero acceleration. Viewing the square wave as the acceleration of the moved air then gives a flat top.
But is sound pressure represented by acceleration? I thought it was represented by pressure. How does that relate to the constant speed of the cone?


Jan Didden
/out on a limb here

 

[phase_accurate]

Hi Jan

Sound-pressure is proportional to cone-velocity.
Regards

Charles

 

[cocolino]

originally posted by Nelson Pass
I think you can get the same effect using the "phase coherent" subtractive type filters as outlined in my article on same ( www.passlabs.com ). When you generate the complement to a 12 dB/oct slope, you get the same sort of bump followed by a 6 dB/oct slope. Also, if you use the same sort of filter on the mid/tweeter, I believe you can eliminate the delay circuit.

Thank You Mr. Pass for the hint and the link!
I read Your article long time ago already and was aware of the possibility realizing those kind of filters by substraction.
The attached picture shows what I believe this should look like for a three-way active filter to give the same response as shown on the picture in my previous post.

As You say in the article substractive filters work perfectly in the electrical domain but that of course the overall ACOUSTIC response INCLUDING the drivers (which are filters by themselves) has to be considered.

In contrary to passive filters where it`s common sense meanwhile that driver responses have to be equalized and textbook filters to be modified to get the desired acoustic response this seem often to be forgotten for active XOvers.

What makes things difficult with substractive filters is to take account for the acoustic driver response as they don`t leave much room for modifying the highpass/lowpass slope at a Xover point.

Although the substractive way is simpler (less parts) and (seems) easier to implement, the reason I like Charles (phase_accurate) topology with the addition of a bandpass and a lowpass filter (what I was not aware before) instead of generating the lowpass by subtracting is because IMO this topology is more universal applicable in real (speaker) world situations as it would allow a phase coherent filter (-approach) simply with extended possibility to alter slopes while accounting for acoustical driver/box response.

 

[capslock]

Hi Christoph,

a passive filter has a high output impedance, hence the impedance curve of the driver should be linearized in order to make the filter calculation more straightforward. Active filters have near-zero output impedance, and hence their output frequency response is not influenced by the driver impedance. This is about the only difference.

I think it is perfectly ok to use more filters after the phase-coherent XO do make the drivers behave more ideally. A Linkwitz-transform might be used to move the resonance further away from the XO frequency, and an allpass delay can be used to compensate for differences in membrane mounting depth.

Regards,

Eric

 

[cocolino]

a passive filter has a high output impedance, hence the impedance curve of the driver should be linearized in order to make the filter calculation more straightforward. Active filters have near-zero output impedance, and hence their output frequency response is not influenced by the driver impedance. This is about the only difference.

Hi Eric,
in fact there are a couple more differences between active and passive XO but I don`t want to get off topic here as this wouldn´t adress phase coherency in XOvers but active versus passive XO in general.

Only one example that fits here: You can`t do these subtractive (this time I got it right - always wrote substractive ) filters passive.

But I agree that the decoupled driver impedance is of the main benefit with active.

I think it is perfectly ok to use more filters after the phase-coherent XO do make the drivers behave more ideally. A Linkwitz-transform might be used to move the resonance further away from the XO frequency, and an allpass delay can be used to compensate for differences in membrane mounting depth.

IMO it`s not only "perfectly okay" to use other filters or equalizer circuits in phase-coherent XO but more often than not they are REQUIRED to get a decent OVERALL acoustic response.
That`s what I was finally talking about in my previous post .
Linkwitz-transform and allpass delays circuits might be one option to get there but these cannot cure everything.
The Linkwitz-transform circuit for example is suitable only to correct errors introduced due to driver high-pass filter behavier.
But what about acoustic driver low-pass effect due to restricted high frequency response? One possible solution could be to take account for this by including the acoustic driver low-pass response into the electronic part of the filter. However, as I said already, this is almost impossible with subtractive filters as they don`t leave room for modification of a lowpass (or highpass, whatever one use to derive the other) without at the same time affecting the corresponding highpass (lowpass respectively) filter slope at given "fc" .
Again, the somewhat more "complicated" (none-subtractive) topology according the basic schematic in my last but one post could help here (which in turn is based on Charles works in his thread: Active Subtractive XOs ).

After all one should never forget: all these filters work fine in the electronic domain only. To maintain their excellent phase coherent characteristic in the acoustic domain with real world drivers also, the individual drivers inherent (electro-)acoustic behaviour HAS to be taken into account to get decent results overall.

 

[JAZZY JESS]

A high power true minimum phase loudspeaker system is possible

I was the Chief Technician at Dunlavy Audio Labs for several years all the way through it's demise. for every speaker that went out the door I calibrated to +/- 1dB which could only be accomplish by using first order cross overs. needless to say, they suffer from SPL and where blowing all kind of tweeters and drivers like crazy. Andrew Rigby got me started on my own research and development to fix the problem, at least a little with out sacrificing what John Dunlavy was only intrested in, that was; perfect measurements on paper only. when Andrew Rigby would bring it to Johns attention his reply is we can not do anything about it, if they want SPL over quality tell them to buy JBL's speakers. through out the years I have discovered a High Power True Minimum Phase Loudspeaker system with a flat excess group delay over the entire audio range, which is an extreme accurate reproducer with all the perfect measurements on paper and capable of reproducing high SPL with out damaging the low voltage transducer drivers. Thats Right, it reproduces square waves and plays at very very high levels. I did a lot of CES with Dunlavy and listened and measured a lots of speakers, none of them even come close to what I have. I offered my design to JBL in northridge but to no avail. I can not belive it, if pepole could only hear what an accurate reproducer sounds like at real high levels. So yes it can be done; if you know what your doing, only the Military that I know of has high power true minimum phase and use it as a weapon to disperse crowds without violence called LRAD, but not with overhung voice coils.

 

[chris661]

You can't hear square waves anyway. Your eardrum has mass and therefore inertia. Trying to move it infinitely fast will never happen. This follows with speakers. You can move them fast, but now try to stop them.

I discussed a topic similar to this (reproducing a drum) with others, and it was decided that it will never happen. The mic diaphragm has inertia, so does the speaker cone, so does your ear drum. All of these mean the steepness of the leading edge will never be as steep as the original. Furthermore, if we tried to reproduce a bass drum (like in a brass band), you'll need a 31" speaker with PR and the BL equivalent to a bloke hitting it with a weighted stick.

Sorry to be the pessimist, but I don't think it can be done.

Chris

 

[phase_accurate]

I'ts not about reproducing a real squarewave with infinitely steep slopes and a flat "top and bottom" - nobody listends to them either. It is about having flat group delay AND flat amplitude response over the audio range. If both these targets are met, a reasonable squarewave reproduction is achieved.
This can't be done with the ubiquitous LR crossovers alone. It needs some more tinkering (although not even that much - it is a question of attitude IMO) than just use what is published everywhere.
But yes, it can be done. And yes, it is definitely easier to do it actively.

Regards

Charles

 

[catapult]

I was the Chief Technician at Dunlavy Audio Labs for several years all the way through it's demise. for every speaker that went out the door I calibrated to +/- 1dB which could only be accomplish by using first order cross overs. needless to say, they suffer from SPL and where blowing all kind of tweeters and drivers like crazy.

Nice job!

I was recently thinking about Dunlavy speakers and came up with my own "improved" filter. Basically, you use the Duelund concept of stacking acoustic (not electric) transfer functions, either highpass or lowpass, all at the same frequencies, on all the drivers. If you do that with 1st order filters, rather than the 2nd order Duelund preferred, the system retains its "transient perfect" behavior but the ultimate roll-off of most of the drivers is steeper than 1st order so there is much less strain on the drivers.

HTGuide Forum - Duelund meets Dunlavy (aka Duelund meets transient perfect)

 

[john k...]

One problem that doesn't often get discussed is that after you are all done with the crossovers you are still left with the significant GD associated with the high pass nature of the system's woofer.

 

[badman]

One problem that doesn't often get discussed is that after you are all done with the crossovers you are still left with the significant GD associated with the high pass nature of the system's woofer.

Darned good reason to stick with low Fs drivers where possible, and crossovers far from Fs for midranges and tweeters.

 

[KSTR]

One problem that doesn't often get discussed is that after you are all done with the crossovers you are still left with the significant GD associated with the high pass nature of the system's woofer.

Good point. And there is only one**) way to really get around that (unless this roll-off is very low in frequency -- single digit Hz numbers preferably -- and very low order, too), and that is phase correction with a convolver/FIR-filter, to obtain a linear-phase roll-off.

**) Well, with a true hell of cascaded peaking (not optimized for Butterworth GD) allpass analog filters one might be able to get this managed (partly) in the analog domain, too (and thanks to Charles for pointing us to this dsign idea/method in a german forum)

In the graph we see a square wave burst at 1.33 times the pole frequency of a LR4-type high pass (two Butterworth 2nd order cascaded).

Upper trace: mininum-phase (causal) filter like some bass-reflex speaker would sport. Not very akin to a square wave, indeed.

Lower trace: same LR4 amplitude response filter, this time linear phase. Looks pretty much like a square wave, doesn't it? Of course we see the intrinsic, unavoidable pre-ringing, but it doesn't show negative acoustic effects. Rather the contrary is what happens with such a filter, the bass remains "tight" with the rest of the music even for frequencies in the vicinity of the high-pass corner and below (below it just gets softer in volume but is not lagging behind).

- Klaus

 

[KSTR]

Through out the years I have discovered a High Power True Minimum Phase Loudspeaker system with a flat excess group delay over the entire audio range, which is an extreme accurate reproducer with all the perfect measurements on paper and capable of reproducing high SPL with out damaging the low voltage transducer drivers.

Would you mind to share some breadcrumbs about the pricipal design ideas behind that? That is, unless you lips are sealed (for understandable reasons).

- Klaus

 

[catapult]

One problem that doesn't often get discussed is that after you are all done with the crossovers you are still left with the significant GD associated with the high pass nature of the system's woofer.

Yup. Even if you get the crossovers perfect, the top of the square wave won't be flat. If the woofer rolls off 2nd order at 20Hz and the tweeter rolls off 2nd order at 25K, here's the best you can expect a 1K square wave to look.

Good point. And there is only one**) way to really get around that (unless this roll-off is very low in frequency -- single digit Hz numbers preferably -- and very low order, too), and that is phase correction with a convolver/FIR-filter, to obtain a linear-phase roll-off.

**) Well, with a true hell of cascaded peaking (not optimized for Butterworth GD) allpass analog filters one might be able to get this managed (partly) in the analog domain, too (and thanks to Charles for pointing us to this dsign idea/method in a german forum)

In the graph we see a square wave burst at 1.33 times the pole frequency of a LR4-type high pass (two Butterworth 2nd order cascaded).

Upper trace: mininum-phase (causal) filter like some bass-reflex speaker would sport. Not very akin to a square wave, indeed.

Lower trace: same LR4 amplitude response filter, this time linear phase. Looks pretty much like a square wave, doesn't it? Of course we see the intrinsic, unavoidable pre-ringing, but it doesn't show negative acoustic effects. Rather the contrary is what happens with such a filter, the bass remains "tight" with the rest of the music even for frequencies in the vicinity of the high-pass corner and below (below it just gets softer in volume but is not lagging behind).

- Klaus

Your linear-phase FIR example will fix the phase distortion of the crossover but it doesn't do anything to fix the non-flat-topped square wave. The only way to fix that is to lower the frequency response as close to DC as possible.