[Tyson]

I've been running active (3 way) speakers for years.  After running different crossover slopes and points on different drivers and different speakers over time, I've come to a couple of conclusions:
1.  Phase is audible.  Steep crossovers like LR48, Butterworth 48, LR24, and Butterworth 24 all have messed up phase which eventually ends up bothering me.  Shallow slopes like Butterworth 6 or Butterworth 12 just sound better.
2.  Steep slopes sound cleaner.  Face it, if you cross a driver at 2khz, you don't want 3khz resonances poking through into audibility.  Steep slopes cut off out of band crap very effectively.
So, it's a Mexican standoff - you can have better overall sound with a shallow slop and put up with out-of-band shenanigans, or do you use a steep slope and less musical and coherent sound?
Well, I've found a way to cheat.  Here's the trick - set the crossover to 2khz, using a shallow slope, say a Butterworth 6db/octave.  The DCX also lets you use EQ to shape your Frequency response, and the cool thing is you don't have to use it in the pass band of the driver.  So, in our example, you can choose a "High Pass" filter, set it one octave higher than your crossover point, set the slope to 12db/octave, and then make it a negative EQ of -12db.  This EQ slope combines with the 6db Butterworth crossover to give you an overall 18db cutoff, but the phase characteristics are still tracking the very shallow 6db Butterworth filter.
Net result - great phase behavior AND a sharp cutoff!  Hopefully this makes sense....

[Bob in St. Louis]

Hopefully this makes sense....

Umm yea...Come on over, and show me on mine. I'll have beer.

Bob

[richidoo]

Both of those are using IIR filters, which act the same as analog filters, with same theoretical phase shifts. The error less than analog because the filters are done in math, so no parts tolerance errors, but the phase shift is still audible to sensitive listeners. 4th order crossover is 360 degree of phase shift, while 1st is only 90 spread out over a large bandwidth, so it is not perfect either, but much less noticeable. The 12dB EQ filter also has phase error of it's own, which adds to the 1st order crossover filter. It may even be worse phase response because instead of a phase coherent steep crossover around the desired frequency, you have crossovers and filters each with different phase errors that are not aligned. With LR4 the phase still shifts through 360 degrees but HP and LP filters are always in phase with each other. Why your experiment sounds better than LR4 I don't know. Luck? Did you try changing freqs a little bit to see if you found a sweet spot with your system?

You can check out linear phase crossovers, aka FIR filters. They are only possible in digital realm, and they require much more processing than IIR filters so there is usually significant latency, which is why it has not been used in low priced pro audio gear. With FIR filters  there is no phase distortion in the filter, no foggy stuffy feedback like sound, just clear open and fast, as if there was no crossover at all. EQs are done in the same way. With some implementations, the setup with a sampling microphone can remove all group delay from the entire audio band. Acourate, Audiolense, DEQX use FIR crossovers. The DEQX on RobS' accuton/raal MTM sounds awesome, best solo violin playback I have ever heard. I tried a Japanese 2way FIR crossover  plugin for foobar and the sound was much better than DCX or Rane, or other PC based digital IIR crossovers.

There was some talk of FIR plugin for miniDSP, but they decided that it's not powerful enough processor. http://www.minidsp.com/forum/3-suggestion-box/338-method-of-fir-filter-development  The Behringer is definitely all IIR filters.

[panomaniac]

LOL Bob.  That rulz! 

I'm hip to try the FIR plugins, got links?

[Tyson]

HOLM Impulse has phase measurements built in, I'll post some screen shots in the next day or so so you can see what I'm talking about....

[Tyson]

Also, I had a DEQX for a while and it was an improvement for sure, but I still tend to prefer shallow slope IIR filters, for some reason.  Not sure why, but there it is.... 

[richidoo]

I still tend to prefer shallow slope IIR filters, for some reason.  Not sure why, but there it is.... 

I love 1st filters too. Simple passives retain the life with simplicity. It's great if the drivers can handle it. Your idea is a great way to help them handle it. I didn't think about that before. You use low order slope across most of the crossover band until it is attenuated enough so that the added phase shift of the 2nd order EQ will not be so audible. Very nice idea... 

Michael, here is a link to a free but very cloogie foobar plugin. Instructions are in Japanese, but the author is very nice and helpful. I eventually made it work, but it is a basic implementation, and asymmetric crossover filters. You adjust the slope by the number of taps you specify. But it's free and gives a taste of FIR sound. http://www.aedio.co.jp/download/index_e.html

Audiolense and Acourate are PC based so they need a multichannel soundcard. Audiolense uses a non traditional crossover slope like what Tyson discovered, that starts off shallow and gets steeper and steeper until it is infinite at +/- 2 octaves. Acourate lets you create any kind of filter shape you want, traditional or not.

[Davey]

If it were only that easy.  You're trying to cheat....but it's not really cheating. 

If you look at the result you'll see it's exactly the same as would be created with an analog equivalent.  You're still not attenuating out of band the way a full 18db/octave filter would.
Remember that a shelving filter still has phase-shift within its band related to the amplitude.  Once you stop shelving yes, the phase stops shifting, but it's still phase shifted.

It might be better if you analyze/test using your DCX just electrically and leave the speakers out of it.  You'll see that there is no free lunch.

Cheers,

Dave.

[Tyson]

Davey, I'd agree, but measurements tell a different story.  I'll post soon....

[richidoo]

That HOLM Impulse app looks really cool. Interested to see how you use it and what you learned.

[Tyson]

OK, here's the measurements.
First - Just the midrange driver Frequency Response and phase, with no EQ and no crossover slopes applied.  This is our baseline, we can probably make things worse, but unlikely to make things better:




Here is the previous, sharp filters I was using - Linkwitz-Riley 48db/octave (8th order) slopes, set at 80hz and 2kh.  Original midrange only measures are in Red still, the LR measures are in Blue:




And lastly, here is the measure of the shallow slope Butterworth 6db/octave (1st order) slopes I am using now, along with the "cheat" of using out of band EQ to keep the actual rolloff rather steeper than 1st order.  Original midrange only measures are in Red still, the B6 measures are in Orange:



As you can see, phase is much, much better with this approach, and the Frequency Response I was able to dial in is quite good also.  Sound quality is much higher with this approach, IMO.

[panomaniac]

Rich thanks for the links, I'll check them out.   

Tyson.  Thanks for the graphs and phase plots it's interesting stuff.  Can you post your DCX settings for this?  I'd like to have a look.  Or post the DCX control file if you use that software.  I want to look at what is going on, too.  Will measure with HOLM straight out of the DCX.

EDIT:  Re-read your post, I see what you are doing.  Think I can do the same thing here.  It's just adding the -12dB/octave shelving filter 1 octave about the crossover.

[Tyson]

panomaniac,
yep, that's it.

[Davey]

I guess I'm not sure what your point is.  What you have there is essentially a first-order crossover.....with the associated phase shift, in comparison to a higher-order crossover with its associated phase shift.

You did some EQ'ing outside of the pass band of the particular drivers which looks to have created a nicer first-order acoustic target......but it's still a first-order crossover.  You're not "cheating" on anything.  Your post title is a misnomer.

I think what this experiment does show is the difficulty of implementing first-order acoustic crossovers.  The usage of the DCX makes it easier than it would be in the analog world.....but it's still tricky. 

Cheers,

Dave.

[Tyson]

What I'm saying is that the out-of-band EQ cascades with the 1st order filter to give you a de-facto 18db/octave rolloff at the crossover point, rather than the 6db/octave roll off you'd get using a 1st order filter without the EQ. 

[panomaniac]

What I'm saying is that the out-of-band EQ cascades with the 1st order filter to give you a de-facto 18db/octave rolloff

Was curious so I measured this on the DCX.  Actually the roll-of is very close to Linkwitz-Riley 2nd order, then flattens out to 1st order about 25dB down.  See the graph below.  The blue line is 3rd order, the green line LR 2nd and the red line 1st order with the 12dB shelf. Phase is taken as relative to no filter or all pass (not shown).



I might use the LR 2nd instead, but the more shallow slope of the 1st+EQ down low might match well with some drivers acoustic roll off.  They can sometimes roll off more steeply than they start off as they get out of bandwidth.  This curve might counteract that change.

[Tyson]

panomamiac - right on!  You can also cascade multiple EQ's.  For example, you could put in another -12db eq at 4khz and it would keep the rolloff from flattening out for a bit longer.
I've used 2nd order filters too, of course, and even though they do measure pretty closely, I still prefer the sound of the 1st order slopes w/the out of band EQ.  Again, not sure why, but there it is.

[Tyson]

Also, they don't measure quite the same either - the blue line is the Butterworth 12db/octave slope, and the orange line is the Butterworth 6db/octave slope with cascaded EQ out of band.

[panomaniac]

They should sound different, they are different!   
Try your 1st order +EQ compared to LR 2nd at the same frequency.  Use the "compare" button on the DCX.  They should be very similar sounding.  I think that if you added just a tiny touch of delay on the 1st + EQ they would be near identical.

[JohnR]

This is all very interesting...! Does the DCX support Bessel filters?