Beyond the Ariel

[soongsc]

Hello Jean-Michel,

I don't know what to say because I've done a few designs, all of which are series XO, and I try to focus on overal system performance. During one design session, I tried a design that looked very good on analysis and measurements which had two drivers in opposite polarity, but when I listened to music, some instruments sounded right and some sounded like they were a bit muffled. Since these instruments had very distinct frequency ranges. I swapped the absolute polarity on the system, then I heard the instruments that originally sounded right then sounded a bit muffled, and vice versa; so I went to modify the design such that both drivers were connected with the same polarity, and the result was much better. After that experience, I just make it a rule to always design with drivers connected in the same polarity.

Originally Posted by Jmmlc

Hello Soongsc

Better to not mix 3 different problems.

1) The trivial problem of the wrong polarity due to a negligent people.

2) The trivial problem problem of the absolute polarity

3) The question: is it possible to reach a quasi linear phase over the widest audible frequencies interval without using FIR and DSP.

I don't think this is a place to adress the problem 1) and 2). Better to concentrate on 3)

John has correctly written that using conventional (causal) crossovers only the 6dB/octave can reach the optimal solution. ( note that people using 6dB/octave crossovers don't pay enough attention on a possible polarity reversal of one of the polarity of the loudspeakers. We have to remind that for the 6dB/octave crossover if the 2 drivers are in phase opposition, the frequency response curve is still flat, with no hole. But if one loudspeaker has a polarity inversion then the phase is no more linear as it is with the 2 loudspeakers having the same polarity)

I don't want to speak about digital linear phase crossovers here. They are theorically perfect but few people report hearing the preondulation of the filters used (don't forget that our hearing is causal)

Several people (John, Francis Brooke, Samuel Harschn myself,...) are trying to overcome the weakness of "classical" crossovers ( weakness of the phase linearity of the LR crossover is an example. Bad dephased operation of the 2 loudspeakers when using the classical Butterworth 3rd order is another...). They brought solutions that have each of them advantages and their weakness but which are a true ameliorations of existing crossovers.

Best regards from Paris, France

Jean-Michel Le Cléac'h


Best regards from

[gedlee]

Originally Posted by soongsc

Hello Jean-Michel,
After that experience, I just make it a rule to always design with drivers connected in the same polarity.

Thats a bit of a rash conclusion based on one casual test since your subjective results could be caused by a myriad of factors other than the crossover.

[mige0]

Originally Posted by john k...

I'm actually working with someone on such a projects right now. Here are some preliminary results using my NaO Mini, panels as a test mule. The panels response is 100 Hz LR4 high pass, LR4 2k Hz mid/tweeter x-o. The system amplitude remains unaltered after the DSP preprocessing.

[mige0]

Originally Posted by john k...
... the best approach is to design a system to have good polar response ...

Originally Posted by gedlee

Therein lies the real challenge, not the DSP to flatten the response along some axis.

Actually - there is a strong point to achieve both goals – no need to stop half ways....

We want it all and we want (and even got) it now - felice me that is!



John's min phase concept is the key!

Michael

[john k...]

John has correctly written that using conventional (causal) crossovers only the 6dB/octave can reach the optimal solution.

I don't think that is what I wrote and it is certainly not what I meant. There are numerous causal crossover filters which can achieve true linear phase aside from the 1st order pair. I won't list them because I believe most here are aware of them. Many of them can be achieved passively as well.

My interest in building such a system has waned because DSP provides a much more reasonable solution as it can not only address transient distortion due to the crossover, but also that associated with low frequency cut off of the system. And while I have been direct in my discussion of the approaches defined by Jean-Michel I am not in total disagreement that above a certain frequency time distortion becomes more difficult for listeners to identify. We would all agree that a system is grossly misaligned if we hear individual pulses form midrange and tweeter. This is purely a alignment issue and not necessarily related to transient "perfectness". I.e. a time aligned system with one driver inverted will be superior to a severely misaligned system with correct polarity.

[Jmmlc]

Hello John,

No, in my method the first arrival time are not perfectly aligned as you erroneously say.

If you use my spreadsheet (the link of which has been given previously), you'll see that there is still a minor group delay difference between the bass and the HF.

See attached file: the comparison between a classical 3rd order Butterworth 2 ways with Fc 1kHz and the equivalent Le Cléac'h croossover.

The in "coincidence" curve is very near of the response curve indicating a very good in phase operation between the waves emitted by the 2 loudspeakers. (an horizontal "in coincidence" will indicate a perfect in phase operation of the 2 loudspeakers).

But look at the variation of the group delay (expressed in equivalent distance travelled at the speed sound) between the classical Butterworth and the Le Cleach: 72mm of varition for frequency <4000Hz for the Butterworth, 17mm of variation for the Le Cl&#233;ac'h, indicating for that one an increased phase linearity.

Above 4kHz the phase of the tweeter is for sure progressively tending toward 180 degrees but this is inaudible IMHO.

About the very low frequency why one would care if his tweeter is not perfectly in phase at -40dB level or so, with the bass loudspeaker...!

It is far most important IMHO to consider the interval of frequency wher the 2 loudspeakers have a significant SPL and in that range, with the Le Cleac'h crossover the 2 loudspeakers operate with very low phase difference.

(BTW, I don't set up the crossover using square waves, I used pulse response and phase curves, group delay curves and CSD... square waves are given in the document because during discussion in French forum, many people asked me to give them in order to compare with published square waves ... )

Many people use my crossover in a 3 ways systems and my spreadsheet is devoted to 3 ways systems...

Best regards from Paris, France

Jean-Michel Le Cl&#233;ac'h

Originally Posted by john k...

Let me summarize what I see. You have a crossover with ... It is not phase coherent or transient actuate. For the case you are discussing the crossover frequency (in the power point presentation) is 1 K Hz. The HP and LP in the filters you mention are not in phase at all frequency below 4k Hz.

The part of the spectrum you claim to be in error, above 4k is also misleading because the presentation looks at a 125 Hz square wave for which the majority of harmonic content is below the crossover point. Move the crossover point lower, as would be required for a 3-way, or the square wave frequency higher, and the distortion become clearer.

The crossover can not be transient accurate because one driver must be connected with inverted polarity. Again, this, aside from the fact that the attack of any transient is in the wrong direction, introduces a NONLINEAR, frequency dependent delay of the form 1/f. Note, 1/f behavior is not linear, though it is stated as such in the power point presentation.


I accept that your crossover approach may be useful for improving the characteristics of 2-ways systems but it is not appropriate for 3-ways.
In any event, there is another problem with using offset to time align. Since the acoustic centers are purposely misaligned and do not lie in, or close to in the same plane, moving off axis in the horizontal direction causes loss of the alignment.

[john k...]

Originally Posted by gedlee

Therein lies the real challenge, not the DSP to flatten the response along some axis.

I don't want to use DSP to correct amplitude, just system phase response. If you look at the plost I posted a while back the system amplitude remained constant. Only phase was manipulated with DSP.

[gedlee]

Originally Posted by john k...

I don't want to use DSP to correct amplitude, just system phase response. If you look at the plost I posted a while back the system amplitude remained constant. Only phase was manipulated with DSP.

The idea being to make the impulse response as compact as possible, correct? (Hence good "transient response" and square wave behavior). If this is true then I entirely agree that one needs to have the system impulse response as compact as possible and that this impulse response should remain so over as large a frontal angle as possible. This has always been my goal. I have tried DSP before and found that its advantages were outweighed by its additional cost. But I don't bother with complex analysis on paper since in the end, with a waveguide, the waveguide is going to dominate the ability to achieve the end result and including it in an analysis, while very attractive, is analytically not reasonable.

I do have to admit to be attracted by the idea of using DSP at the very end just to clean up any phase and amplitude anomally that the passive design was not getting, but it would all come down to "bang for the buck". My experince with DSP has been that its not worth the cost (except in things like receivers and radios etc where it can do a hole lot of stuff at the same time.) Worth a try however.

[soongsc]

Originally Posted by gedlee

Thats a bit of a rash conclusion based on one casual test since your subjective results could be caused by a myriad of factors other than the crossover.

Well, if I were doing a PHD thesis, I probably would spend more time to do a whole bunch of studies. Since only the XO changed, and the resulting effect was logical, I did not see much reason to dig deeper. What other factors could cause that?

[Jmmlc]

Hello Songsc,

Polarity of a loudspeaker fed through a crossover (e.g. an "high-pass" crossover) is quite meaningless in itself, it only becomes important when the summation of 2 filtered loudspeakers is considered.

Let's consider a single crossover : a single high pass (+ a perfect loudspeaker) for example.

You have to remember that using conventional polynomial crossovers having an order larger than 2, a high-pass crossover will turn the phase within it's bandwith more than 180&#176;. (Said in other form: we have frequencies at the output of the crossover for which the phase is + or -180&#176; different of the phase at other frequencies).

Inverting the polarity will just add 180 degrees to the phase at every frequency, but the variation (rotation) of phase through the bandwith of the high pass crossover will still be there. So, considered in itself the polarity is quite meaningless.

When used in a 2 ways system (or more) the polarity inversion has to be performed sometimes on one loudspeaker in order that the wave it emits (at a given frequency inside the interval of frequency around the common cut-off of the LP and the HP) will be in phase with the wave emitted at the same frequency by the complementary loudspeaker (+ crossover). The perfect example of that is the 2nd order Linkwitz-Riley which needs a phase inversion of one of the 2 loudspeakers in order to have them operating in phase at every frequency around the common cut-off (f-6dB in that case).

Best regards from Paris, France

Jean-Michel Le Cl&#233;ac'h

Originally Posted by soongsc

Hello Jean-Michel,
so I went to modify the design such that both drivers were connected with the same polarity, and the result was much better. After that experience, I just make it a rule to always design with drivers connected in the same polarity.