There is no time delay in the subtractive XO circuit you posted. What you have is some group delay (rate of change of phase), meaning that the *envelope* of a wavelet (some sine cycles windowed with a raised cosine, for example) is delayed (the peak of the output envelope appearing later than at the input), but the actual *waveform* isn't. Adding the waveforms yields a new waveform with no group delay, by design (1-x + x = 1).
Attachments
Well, of course, you guys are right. BUT to me, a crossover had just as much potential delay as a single low pass filter. That is where the surprise came from. I had PRESUMED that a 6dB/octave xover had excess phase shift, because you could measure the 45 degree shift at -3dB rolloff. What was surprising to me was that this phase shift WITH its attendant phase delay, could be easily canceled out. It was even more interesting with the TP filter.
Last edited:
Yes this can be done. JC's circuit is 1st order + 2nd order assymetric slopes.
2nd slopes are easy to synthesize this way (see below, different -- non-subtractive -- implementations are possible as well), and I was successful doing 3rd order HP and 2nd order LP (or vice versa). BUT, it doesn't work well in reality, phase difference in the XO region is 120deg or more, killing most of the individual driver signals in the sum, except for distortion compoments. Ranges have high overlap, lobing is horrible. Much better, in my book, is to design for zero phase difference but allowing the overrall allpass which can be easily corrected by inverse filtering the source signal (convolution with the time-inverted impulse of the allpass), that was my point in post 44444.
Hawksford has a XO paper where he presented some higher order XO which have an allpass function of lower order than it would normally have, using the same ideas and having the same drawbacks.
Attachments
You can build up a lumped approximation that within limits approaches what looks like a pure delay. I think some early Apogee filters had a strong sin(x)/x like component in their impulse response with the peak well after the input.
I'm not following you. 45 deg at -3dB is not excess, it is exactly what is expected for a first order filter. No delay, just phase shift. Have I misunderstood you?john curl said:
Yes, but that is in essence a lumped approximation to a transmission line so it provides an approximation to a true delay. I suppose it could be argued that the true delay occurs when you have an infinite number of infinitesimally small filter sections.scott wurcer said:
A chap not too far away has those NHTs, with Krell integrated amp and Meridian CD player. Had a listen one evening, system had the usual "hifi" sound, but I very much liked the concept of the speakers - relatively easy to make very physically stable, plenty of room for tweaking ...
Well, I was close, but no cigar. G. R. Koonce's article was in _Speaker Builder_ 2/95. His state-variable implemented crossover gives the same results as Erik Baekgaard's filler driver: all pass and constant voltage and second order high pass. It's a superior "transient perfect" tunable crossover. (I mis-remembered in the earlier post: the Baekgaard crossover is B2 rather than LR2).
All good fortune,
Chris
Now we get back to a point i made almost a year ago when i discovered this Site and decided to put my toe in it (or something)---> The MLSSA "waterfall" plot showing the longer decay at the corner freqs of a LP filter (or HP).
So what happens with this topology cross-over when subjected to same time decay test -- if GD is constant as a non filter - is the time decay also constant at all freqs thru the cross-over ?
Thx-RNMarsh
So what happens with this topology cross-over when subjected to same time decay test -- if GD is constant as a non filter - is the time decay also constant at all freqs thru the cross-over ?
Thx-RNMarsh
Last edited:
Of course real world crossovers are made imperfect by non-coincident drivers, and real drivers' fundamental resonances, keeping geometric solutions like D'Appolito's and driver fundamental resonance shifting (nowdays called a Linkwitz Transform) still in play.
The cool thing to me about Koonce's paper is the idea to sum LP and bandpass, and make the LP driver do both. It's well worth digging out if y'all can. The original Baekgaard is in AES Loudspeakers Anthology (I think).
All good fortune,
Chris
Edit: John Curl is reading my mind! I knew it!
The cool thing to me about Koonce's paper is the idea to sum LP and bandpass, and make the LP driver do both. It's well worth digging out if y'all can. The original Baekgaard is in AES Loudspeakers Anthology (I think).
All good fortune,
Chris
Edit: John Curl is reading my mind! I knew it!
Just to state what I think that means: It means that all of the Fourier components (i.e. at all of the odd harmonic frequencies) still had the correct amplitudes and relative (to each other) phase angles. So the phase angles either had not changed, or, they all had changed by the same constant. So the phase angle change did not vary with frequency. Does that seem correct?
- Status
- Not open for further replies.