Simulated Cauer using DCX?

[AJINFLA]

This thread on DIY seems to suggest the possibility. Thought I'd throw it out to the Cauer experts . All input welcome (hint, I have a DCX and have thought of trying this approach since joining this site :T , but I'm far too lazy to be this innovative and take the lead. Thankfully someone else has :W ).

http://www.diyaudio.com/forums/showthread.php?s=&threadid=15943&perpage=10&pagenu mber=36

Image not available

cheers,

AJ

[sfogg]

(That didn't take long...)

It isn't a Cauer but it is certainly simulating an elliptic filter. My goal was to at least duplicate the response of ES600t Al K. designed. I ended up getting it sharper with about 10dB more attenuation initially and much faster roll off after the arc.

Shawn

[JonMarsh]

I've been considering getting a DCX2496 just to play with and prototype stuff, and try this for fun, too. It's just that finding any spare time to do that, well, let's just say it's an ugly situation!

I figure I ought to finish up the stuff I've already started before going off on yet ANOTHER interesting tangent.

~Jon

[mikec]

(That didn't take long...)

So, are you planning on sharing how you did it? My guess is that you daisy-chained two channels.

Edit: I see that you explained how you achieved the slopes. Thank you for sharing. Have you had a chance to listen to it?

"that response is a composite of a 48dB/octave (the rolloff after the arc) and three parametric filters."

[DIY_newbie]

So, are you planning on sharing how you did it? My guess is that you daisy-chained two channels.

My guess is BP filter (-gain) in PEQ added to LR24 or LR48 cross over..

--Chris

[sfogg]

"Edit: I see that you explained how you achieved the slopes. Thank you for sharing. Have you had a chance to listen to it?"

I haven't heard it through the Behringer yet (have to make some cables) but have been using a speaker level crossover with close to the same response for about 2 years.

As far as exactly duplicating the 600hz crossover in the Behringer you have to unlink the XOs...

Low Pass:

LP Filter = L-R 48
Frequency = 607hz

EQ 1 -15dB, 706hz, Q 10
EQ 2 -15dB, 706hz, Q 10
EQ 3 +15dB, 607hz Q 7.9

High pass:
HP Filter = L-R 48
Frequency = 581hz

EQ 1 -15dB, 510hz, Q 10
EQ 2 -15dB, 499hz, Q 10
EQ 3 +14.1dB, 594hz, Q 7.9

The high pass EQs are slightly different to smooth the blend between the high/low pass.

Shawn

[Dennis H]

As discussed in the other Cauer threads, I think Jon's curve has some advantages over Al's curve that Shawn is emulating. While the initial rolloff isn't as steep ("only" 48dB/octave), it continues much farther and the stopband ripple is suppressed much more, about 50dB down.

Also the group delay peaking (rapid phase shifts) near Fc will be much better with Jon's curve, similar to an LR4 rather than the LR8 Shawn is starting with. Quoting the man (SL) himself, "crossover filters of higher order than LR4 are probably not useful, because of an increasing peak in group delay around f0." And that sort of agrees with the subjective impressions of several Behringer users on the Mad board that something sounds a bit off with the LR8 filter compared to LR4.

So, I'd start with a 4th order Butterworth and apply notch filters as necessary to get a curve like Jon's. See the Modula threads for examples.

[sfogg]

Actually, my curve is going about 50% deeper then Al's and rolls off considerably quicker after the arc too. His is rolling off at about 6dB/octave after the arc, mine is rolling off 48db/octave.

"So, I'd start with a 4th order Butterworth and apply notch filters as necessary to get a curve like Jon's."

I'm not sure you will be able to without using a bunch of filters. If I swap my setup to 24dB/octave the arc jumps way up. The arc is literally from the difference in amplitude between the notch filters and what the crossover itself is doing to the response. If you use more notches to get more attenuation the arc doesn't come down nearly as much as the initial attenuation. It would take more notches to drop that down too and the possibility of running out of DSP power in the Behringer gets more and more likely.

For example here is the above with two more notches (DCX limited to -15dB cuts per filter)....


Image not available


Which brings the initial attenuation way down but the arc comes back up at least 30dB and the initial slope is much shallower to give a wider blend zone. I could notch the arc with more filters but this is already using 5 and isn't nearly as steep to the -10dB points.

I wanted very high slope for my setup which is why I did them this way. I'm getting around 210dB/octave that is transitioning to 48dB/octave compared against 48db/octave transitioning to less then that.

" And that sort of agrees with the subjective impressions of several Behringer users on the Mad board that something sounds a bit off with the LR8 filter compared to LR4."

And in an old center channel I bi-amped I prefered 8th order LR crossovers to 4th. The whole question of audibility of phase *by itself* (not when it creates FR changes from driver interaction) is still up in the air from what I have seen.

But... if I can find the time I can certainly try it both ways and switch back and forth at the touch of a button.

Shawn

[Dennis H]

Maybe move your notches higher in frequency and maybe lower the Q to suppress the ripple? It's a bit hard to picture what's going on with your linear frequency scale. Jon's goal is to emulate the shoulder and slope of an LR8 for the first octave. That means the notches would need to be more than an octave above Fc to fit that shape, assuming they dip well below -48dB. About the processing power, an 8th order uses more than a 4th order so that gives you some more filters to play with.

Edit: so why do all that rather than a real LR8? It's the old phase thing again. I don't personally know how audible it is but there is a nasty looking peak in the group delay of an LR8 when you plot it out. Jon's filter looks about like an LR4 which is pretty benign.

[sfogg]

"Maybe move your notches higher in frequency and maybe lower the Q to suppress the ripple? "

Then my slope isn't as sharp.

"Jon's goal is to emulate the shoulder and slope of an LR8 for the first octave."

OK, but mine starts off much sharper then that (down 35dB in about 1/6 of an octave compared against 8dB down), then ends up at about -48dB one octave out and continues to roll off eight order after that point. I have a *much* sharper initial slope and more ultimate attenuation too. Win win in my book.

"About the processing power, an 8th order uses more than a 4th order so that gives you some more filters to play with.

Edit: so why do all that rather than a real LR8? It's the old phase thing again."

But throwing more parametrics at it is going to screw with the phase too. And *if* that is only audible because of FR changes from driver interaction then a *steep* slope limits driver interaction and therefor doesn't let the phase changes screw with FR in the blend or contains them to a very narrow area.

Shawn

[Dennis H]

The peak in the group delay happens in the passband. The phase anomolies of the notch filter happen in the stopband. But I'm just talking a bunch of theory, no practical listening experience. So, carry on the good work....

[Dennis H]

I finally figured out how to get the demo of LspCAD to run. I've got a lot to learn about how to use it but I did manage to whip up a filter somewhat similar to Jon's. Target is -6dB at 1000Hz, -48dB at 2000Hz, stopband ripple more than 50dB down. Settings in the DCX are 24dB Butterworth XO at 1000Hz and two notch filters at 2210Hz, each with Q=3.5 and gain=-15dB. Once I figure out how to do an optimization run, I can probably get the curve closer to LR8 but, by eyeball, it's not bad now.

[JonMarsh]

Looks pretty good, Dennis! :T

[Dennis H]

More fun and games with LspCAD. This thing is addictive.

Comparison of Shawn's lowpass and my version of Jon's lowpass. Both have a nominal Fc of 607 Hz.

Shawn's magnitude:



Jon's magnitude:



Shawn's group delay:



Jon's group delay:



It's pretty clear that Shawn is paying a penalty in group delay spiking for his sharp shoulder and steep rolloff. Which one sounds better is left as an exercise for the reader. It probably depends on the particular drivers.

[sfogg]

Now you just need to get the Behringer (or something like it) to give them all a whirl!

"It's pretty clear that Shawn is paying a penalty in group delay spiking for his sharp shoulder and steep rolloff. "

Never claimed I wasn't. I'm still not convinced that is an issue by itself. Looking at the graph keep in mind the rolloff. That second dip is occuring where the driver is already down at least 20dB and still dropping fast. That is going to be swamped by what the upper driver is doing.

OTOH I'm gaining less driver interaction which will mean less comb filtering, less off axis FR changes..etc...etc.

Which one sounds better?

Neither of us can answer that. They both have their pros and cons. Like everything else there are tradeoffs involved.

[Dennis H]

I'm not sure I see the advantage of the ultra-steep initial rolloff. Both filters hit -50dB at about the same frequency, about an octave above Fc, so I doubt one filter has much more audible overlap than the other. The only situation where it would seem to be an advantage is where you are pushing the limits of the woofer's capabilities and need to cut it off hard before a resonance. But, even then, Jon's highpass will give less driver excursion on the tweeter because it is 6dB down at Fc. So, you could cross a bit lower with Jon's if the woofer is a problem.

[jdybnis]

Dennis,

Just for comparison can you post the group delay plots for an LR4 and LR8. Thanks.

[sfogg]

"I'm not sure I see the advantage of the ultra-steep initial rolloff. Both filters hit -50dB at about the same frequency, about an octave above Fc, so I doubt one filter has much more audible overlap than the other."

What happens at -50dB is a lot less relevant then what happens at much higher level. Look at where each filter is starting to kick in (just below 300hz compared to 600hz) and compare that against the -3, -6, -10dB point or -20dB point.

One takes a lot longer to get there then the other. The high pass will have to be similar too (for smoothest blend) so that will make the difference larger yet.

"But, even then, Jon's highpass will give less driver excursion on the tweeter because it is 6dB down at Fc. So, you could cross a bit lower with Jon's if the woofer is a problem."

That is a something of a stretch and a bit of a toss up. Just by going off the low pass his hits -6dB at FC but within 50hz of that I'll already be down around -15dB, John's needs 200ish hz to get that low. If you are trying to push it to avoid a resonance a sharper slope is going to let you do that. If for example you need to keep that driver below -20dB to have an audible resonance you can run the low pass higher with the steeper slope and get away with it. With the shallower slope you would have to cross over earlier.

Shawn

[Dennis H]

Shawn, Feyz did some tweeter excursion graphs over on the Mad board. For an LR4, the highest excursion is at Fc and drops down above and below there. For crossovers with steeper slopes and/or higher Q, the max excursion is actually greater than the LR4 because of the sharper shoulder.

[Dennis H]

Lr4:



lr8:

[AJINFLA]

Audible differences aside, can you see what a fantastic tool this $200 box can be? What would be the parts cost for any of these XO's done passive :E ? Not to mention that later on, after extended listening, you decide to tweek........ :W

cheers,

AJ

[jdybnis]

Thanks.

[cjd]

Audible differences aside, can you see what a fantastic tool this $200 box can be? What would be the parts cost for any of these XO's done passive :E ? Not to mention that later on, after extended listening, you decide to tweek........ :W

cheers,

AJ

Can you do a 3-way with a single one of these magic boxes and run these elliptic filters? It might break even if you'd need two, but would be a savings if one could do it.

On the other hand, the "problems" I am currently dealing with simply would not exist.

LR4 is about the maximum most people find works to their ears: many bits of data that some pay more or less attention to provide clues as to why this might be the case. A C-E filter emulating an LR8 seems to treat the ear like a plain LR4 though. . .

C

[sfogg]

"Can you do a 3-way with a single one of these magic boxes and run these elliptic filters?"

Maybe, they have six channels of outputs but depending upon how you set them up you may or may not have enough DSP resources to work with for three way with elliptics.

"many bits of data that some pay more or less attention to provide clues as to why this might be the case. A C-E filter emulating an LR8 seems to treat the ear like a plain LR4 though. . ."

There is a numerous reasons why that could be the case. It may be the one that is discussed here (though the LR8 listed here is below Blauert's suggested level of audibility) or it could be other things. Such as widely different dispersion patterns between two drivers at the crossover point. The steeper the slope there the more apparent that is going to be depending upon the room. An elliptic acting like a LR8 would really be no different then an LR8 in that situation.

Or a driver that is down somewhat at the crossover point. More overlap from the other driver would help to mask that.

Or that the interference between the two drivers helps cut down on the off axis FR somewhat to reduce floor/ceiling bounce.

Drivers that have a different sound to them might stand out more with sharper slopes compared to a more gradual transition between them.

Or simply that a person may like the sound of comb filtering between the drivers and may not like having that reduced.

There are a lot of variables. It isn't as easy to only change one to see how it audibly effects things while playing back music.

Shawn

[Dennis H]

If the drivers are comb filtering at any normal seated or standing listening position, they are too far apart for the given Fc. Even if you cut them off with brick-wall filters, they are still too far apart and an instrument will appear to shift position as the sound moves from one driver to the other. That's one good reason to go for low XO points. With a sufficiently low XO point, the best reason to go ultra steep with the filters is to suppress the cone breakup modes which are a real problem with stiff cones. It's nice to get that stuff more than 40dB down.

[sfogg]

"If the drivers are comb filtering at any normal seated or standing listening position, they are too far apart for the given Fc. "

You are thinking of normal direct radiators.... I'm using horns.

The mouth of the horn has an effective 'size' of 22"x28"ish and the mid is roughly 24"x13". There is no additonal spacing between them but depending upon how you want to figure it the path distance between them they could still comb. Pre-Behringer they aren't time aligned drivers either because of different lengths of the horns. The Behringer will let me time align the drivers.

"they are still too far apart and an instrument will appear to shift position as the sound moves from one driver to the other."

Certainly have never noticed that and that was one of the things I was listening for when I first moved to elliptic crossovers. From the listening position it is a few degrees of change in the vertical. Localization in the vertical plane doesn't work nearly as well for humans as it does horizontal. The Haas effect probably plays a roll there too. The sound gets localized to whichever driver you hear first so overtones from the other driver are heard as coming from the driver you heard first.

Maybe if you were trying to listen for this on a slow sweep tone it might be noticeable but in my setup it isn't an issue on music.

But really, this is no different then any vertically arranged speaker with multiple drivers. The actual height of the sound source varies by frequency.

Shawn

[Feyz]

Fwiw

Here is the page to high pass xover vs cone excursion comparisons I had done at one time:

http://members.fortunecity.com/pirimoglu//images/HPExcursionComps.html

Note, the cauer elliptic plot in there has an early roll off between LR8 and BW7, so does its excursion plot. If it had emulated LR8, its excursion would have looked like LR8.

[ergo]

Jon, could you please post or send me some samples of the LspCad simulated transfer functions of typical Cauer type filter from some of you projects. Even better if a full LspCad project (ver 6 one ).

What I would like to do is to try to make and optimize as precise as possible transfer function with LspCad built in DCX blocks and then enter the result to real DCX. Would be very interesting to know how close I get to your result and also if the Behringer DCX can handle such filter from processing power point of view.

Regards,
Ergo

ergo.esken AT mail.ee

[sfogg]

Ergo,

" and also if the Behringer DCX can handle such filter from processing power point of view."

Sure, I'm doing it.

How much processing power you have will depend upon if you are running two or three way. If you do two way and turn off the crossovers on the two unused outputs (unlink all crossovers) that gives you more DSP available. From there you would just be limited by the number of parametrics the DCX gives you per output and if you are using EQ for more then just the elliptic crossovers.

When you are balancing resources in the DCX you need to have both enough DSP available as well as available parametric filters per output.

Depending upon how the DCX is configured it will give differing numbers of parametrics per output. If you play around with their software you can see the different numbers/combinations of parametrics per output when you change its default configuration from stereo three way to mono and so forth. That is just the starting point in the software as you can then reconfigure it by changing the routing of input/outputs to however you need.

Because I'm crossing three way L/C/Rs I was using two DCXs both configured as stereo three way operation. I was also using some of the parametrics for actual response shaping and had the DSP resources pretty much maxed out with 1% DSP remaining. As such I was crossing to the tweeter non-elliptic. I ended up going with a third DCX and now dedicate one per speaker to have enough resources for all the above and to go elliptic to the tweeter. The tweeter was limited to only two parametrics on it so I only used one notch for the low/high pass crossover on the mid/tweeter so it isn't as sharp as the woofer/mid but is still pretty high slope.

Oddly, it was better to configure the DCXs as stereo three ways instead of mono when I used one per speaker. With it set to mono it only allowed a couple of parametrics on all the outputs and I wanted more then that on the bass/mids. You can re-route the input/outputs as needed to make it work and turn off the crossovers on the other channels though to get a combination that will likely work. Each of my units has two inputs and three outputs unused with a fair amount of DSP power left in them. I'll likely use that for other aspects of my system at some point too.

Shawn