Hi Everyone..
I'm planning out a center channel with a DSP plate amp, and I ran across something I had not been paying attention to. Per Wikipedia, the lower filter will be in phase but a period behind the next driver up.
So, help me do the math here, imagining an ideal LR4 electroacoustic crossover with 2 ideal and identical drivers crossed at 300 Hz. Does this mean the woofer will be 3ms further away from the mid??
Is it possible or desirable to compensate for this? That is, delay the mid by 3 ms, so that the 1 period delay vanishes??
Am I overthinking it or is this actually a thing that most people do already? 😀
I'm planning out a center channel with a DSP plate amp, and I ran across something I had not been paying attention to. Per Wikipedia, the lower filter will be in phase but a period behind the next driver up.
So, help me do the math here, imagining an ideal LR4 electroacoustic crossover with 2 ideal and identical drivers crossed at 300 Hz. Does this mean the woofer will be 3ms further away from the mid??
Is it possible or desirable to compensate for this? That is, delay the mid by 3 ms, so that the 1 period delay vanishes??
Am I overthinking it or is this actually a thing that most people do already? 😀
Last edited:
If the acoustical responses are textbook like LR4 and the acoustical centers of the drivers are at the same distance (or the closer one are delayed in the DSP to acts like it would be at the same distance to rhe geometrically further driver) to the measuring/listening position, then the high-frequency driver will be 1 period ahead in time than the low-frequency driver and in phase with each other. All this caused by the normal IIR filtered LR 24 dB/octave acoustical curves.
If you want to compensate the delay of the woofer with just simply delaying the tweeter by 1 period time, then the relative phase between the drivers will be messed up.
If you want to compensate the delay of the woofer with just simply delaying the tweeter by 1 period time, then the relative phase between the drivers will be messed up.
Have a look at Linkwitz original paper for more information
https://www.linkwitzlab.com/JAES/jaes_papers76.htm
Here is an image from Douglas Self's book on Active crossovers that shows the phase relationship.
An LR4 crossover will produce a second order allpass response, that is there will be 360 of phase change through the filter with -180 degrees being at the centre frequency. The group delay is frequency dependent so a simple linear delay cannot correct for it. This is the group delay, also from Self, there is a peak below the centre frequency before it levels out to a stable value.
What you can do with a delay is to align the two drivers in time at the crossover frequency. A simple way to check that is to reverse the phase of one leg and change the delay until you get the deepest reverse null.
When the drivers are offset from each other there will be a time delay so some form of delay adjustment is needed to have the highpass and lowpass work as intended otherwise they will not sum properly.
https://www.linkwitzlab.com/JAES/jaes_papers76.htm
Here is an image from Douglas Self's book on Active crossovers that shows the phase relationship.
An LR4 crossover will produce a second order allpass response, that is there will be 360 of phase change through the filter with -180 degrees being at the centre frequency. The group delay is frequency dependent so a simple linear delay cannot correct for it. This is the group delay, also from Self, there is a peak below the centre frequency before it levels out to a stable value.
What you can do with a delay is to align the two drivers in time at the crossover frequency. A simple way to check that is to reverse the phase of one leg and change the delay until you get the deepest reverse null.
When the drivers are offset from each other there will be a time delay so some form of delay adjustment is needed to have the highpass and lowpass work as intended otherwise they will not sum properly.
I tried this once. It did not work, for all the reasons YSDR and Fluid explain. The on-axis response became ragged, and the off-axis response was even more screwed up.
My understanding is that FIR digital processing may be able to "correct" the group delay, or implement a 4th order crossover without any phase alteration. The big question is whether this would be an improvement over the more organic minimum phase behavior of an analog (or IIR) 4th order filter.
Thanks guys. I simulated this last night with XSim (it's great for practical exercises like this). I didn't look at the group delay, but looking at the impulse and step responses I saw no reason to compensate for the time, and as shown by the graph @hifijim put up, the group delay isn't simple. Attempting to compensate for that with a single value delay would leave a hot mess through the crossover area.
Hi, what has helped me, has been the realization phase and time cannot be substituted for each other.
By time I mean any constant fixed time .....or any expression of it, like the time of one period at a particular Hz, or group delay at any particular freq.
My best understanding of phase is that it has no 'real' time delay....any attempt to translate phase into a fixed delay is a mathematical apparition.
As phase has no effect on the absolute fixed time, of when all frequencies first arrive, together simultaneousy.
I see phase as a continuum relationship between frequencies angular positions that cannot be broken into an amount of fixed time at any chosen frequency.
So the concept wiki proposes, that "the lower filter will be in phase but a period behind the next driver up." is misleading at best imo.
And imo, has nothing to do with a fixed delay between woofer and tweeter.
Easiest way for me to separate phase from time, comes down to 'what's freq dependent and what's not'.
By time I mean any constant fixed time .....or any expression of it, like the time of one period at a particular Hz, or group delay at any particular freq.
My best understanding of phase is that it has no 'real' time delay....any attempt to translate phase into a fixed delay is a mathematical apparition.
As phase has no effect on the absolute fixed time, of when all frequencies first arrive, together simultaneousy.
I see phase as a continuum relationship between frequencies angular positions that cannot be broken into an amount of fixed time at any chosen frequency.
So the concept wiki proposes, that "the lower filter will be in phase but a period behind the next driver up." is misleading at best imo.
And imo, has nothing to do with a fixed delay between woofer and tweeter.
Easiest way for me to separate phase from time, comes down to 'what's freq dependent and what's not'.
Off topic but
Fir filters are so much better then iir I use a computer based dsp. For example I use a blesma 34b in a waveguide to 1000 hz if i 4th order xo it would still have a lot of power being sent below the xo probably would kill the tweeter sooner then later but with fir I can use a 16th order 96db/Oct cut every thing below that 1000hz off and have perfect directivity to a 8 inch driver. When measuring my speaker I had 6db more headroom on the tweeters with the 1000hz 16th then 1300hz 4th order. You can also use fir for subs when you cross them over at 80 or 100 with 4th order you can still hear the sub playing into the frequency you can locate it but with that 16th or higher it disappears. Also with higher order xo things like metal domes and cones ringing is completely a non issue. You can also do things like a active cardioid speaker my side speakers on mine play 10->600ish but 10-100 is in phase while 100+ is out of phase to give the effect of a cardioid speaker. Fir can completely change how you think and design a speaker . I also use 10x 8'' in a 3 cf box qtc over 1 do I care nope this bad boy is flat to 10hz at 105db linear phase. My point is your thinking about using fir filters while shackled to the constraints that once were real for speaker design.
weird.
When I used 24db LR active crossovers, I basically lined up the acoustical centers.
Even then, I think they were really only lined up at the crossover point (regarding delay tweet vs. woofer).
And to momo, a distortion clean sub is much harder to localize
When I used 24db LR active crossovers, I basically lined up the acoustical centers.
Even then, I think they were really only lined up at the crossover point (regarding delay tweet vs. woofer).
And to momo, a distortion clean sub is much harder to localize
"When I used 24db LR active crossovers, I basically lined up the acoustical centers."
Yeah, that's the normal with LR crossing, and if done right, then the low-freq driver is lagging 1 cycle (if 24 dB/octave are used) in the time domain, behind the high-freq driver. This lagging cannot be compensated by delaying the tweeter 1 cycle, because if we do, it messes up the phase between the drivers, which messes up the combined resulting frequency response of the drivers.
Yeah, that's the normal with LR crossing, and if done right, then the low-freq driver is lagging 1 cycle (if 24 dB/octave are used) in the time domain, behind the high-freq driver. This lagging cannot be compensated by delaying the tweeter 1 cycle, because if we do, it messes up the phase between the drivers, which messes up the combined resulting frequency response of the drivers.
Last edited:
Yes an FIR filter can be made to compensate for the all pass behaviour of the LR4 crossover. Applying a reverse/inverse all pass filter of the same type and Q will cancel the group delay and make it flat, the penalty is latency in the filter. This is easy to see the result of in an impulse or step response as it will then be much closer to an ideal version. Whether this is an audible improvement is a highly debated topic.
Or more simply linear phase versions of the filters can be used if a DSP has that capability.
FIR filters are different not better, both types have their differences and trade off's and which is the right one depends entirely on what you want to achieve.
In addition to the classical steep linear-phase FIR filters, high-performance audio ADCs and DACs nowadays usually have filter options that reduce preringing at the expense of a worse phase or magnitude response, even though the preringing of ADC decimation and DAC interpolation filters is supposed to be ultrasonic.
A linear-phase crossover filter will have preringing in the middle of the audio band. The preringing reproduced by the woofer is supposed to cancel that of the tweeter in a two-way loudspeaker, but it never exactly will.
Is preringing a concern for you or is that something only ADC and DAC designers worry about?
A linear-phase crossover filter will have preringing in the middle of the audio band. The preringing reproduced by the woofer is supposed to cancel that of the tweeter in a two-way loudspeaker, but it never exactly will.
Is preringing a concern for you or is that something only ADC and DAC designers worry about?
I have a DAC which has the option of apodizing filters, which do what you describe. I don't like the sound, so not an issue for me. I find them a little dull in the top octaves.
Nope.
Especially not when a complementary acoustic crossover is achieved.
And doubly especially not, when driver c2c spacing of 1/4 WL at crossover freq, is also achieved.
In the unit i use preringing with FIR is a non issue for complementary filters ( crossover).
It's mainly because engineers locked (to the user) the steepness of filters (if you want more than 48db/octave) to values which makes preringing not audible ( under threshold of audibility for each area in which you xover ( iow it does change with frequency implied and i don't say it isn't there, just not audible).
I'm sure a badly implemented 500db/octave could exhibit bad audible preringing but i have yet to find a possible use to this.
It's mainly because engineers locked (to the user) the steepness of filters (if you want more than 48db/octave) to values which makes preringing not audible ( under threshold of audibility for each area in which you xover ( iow it does change with frequency implied and i don't say it isn't there, just not audible).
I'm sure a badly implemented 500db/octave could exhibit bad audible preringing but i have yet to find a possible use to this.