I think we're still debating different issues
As I pointed out earlier, the thread started out on an acoustic crossover and the inversion of the connection in a 2nd order passive crossover. This is a different matter than adding an allpass filter into a signal that is meant only to add a phase shift at some point as one done in the Linkwitz reference.
Yes and I'm not taking issue with this aspect. The allpass is extended on the acoustic side to include odd-order Butterworth and all Linkwitz-Riley crossovers.
Reactance does not by nature have to change the shape. Keep in mind that we're not talking about it's use as an allpass (at least the thread starter didn't) inserted into a signal. We're talking about a combination of a highpass and lowpass acoustic response and the sum of the two. It's the sum that concerns us. A first order Butterworth and a few other select transient-perfect crossovers can indeed be constructed passively that maintains the signal ideally. The only caveat is that if other than a 2nd order type is used, the eventually transition of the driver response to a 2nd or higher slope will occur. This can be designed to be far into the stop-band, so that it's impact is minimized.
Certainly there's a slew of upstream components in the chain. However, if it were not important as you imply, then we likely wouldn't be so sensitive to the differences between different crossover designs. The best that one can hope for and the goal of many of us is to reproduce what is in the recording as faithfully as possible. That is all that's within our control.
Yes. What we are debating, though, is not a signal and itself, it's the same signal with both a highpass and lowpass applied to the two contributors. The BW1 elicits the same result in a sense. At Fc, each leg is down 3db from the design target reference. Each leg is rotated 45 degrees, one in one direction, the other in the reverse direction. This leads to phase quadrature that when summed yields a result of +3db over each section's signal. In other words, flat summed response results since both legs are down 3db at Fc.
Coincidentally, at all other points in the crossover region, the phase remains in phase quadrature as each rotates through its transition region with the vectoral sum resulting in 0db relative result, i.e. a flat summed response.
Dave
As I pointed out earlier, the thread started out on an acoustic crossover and the inversion of the connection in a 2nd order passive crossover. This is a different matter than adding an allpass filter into a signal that is meant only to add a phase shift at some point as one done in the Linkwitz reference.
hailteflon said:
Yes and I'm not taking issue with this aspect. The allpass is extended on the acoustic side to include odd-order Butterworth and all Linkwitz-Riley crossovers.
Reactance does not by nature have to change the shape. Keep in mind that we're not talking about it's use as an allpass (at least the thread starter didn't) inserted into a signal. We're talking about a combination of a highpass and lowpass acoustic response and the sum of the two. It's the sum that concerns us. A first order Butterworth and a few other select transient-perfect crossovers can indeed be constructed passively that maintains the signal ideally. The only caveat is that if other than a 2nd order type is used, the eventually transition of the driver response to a 2nd or higher slope will occur. This can be designed to be far into the stop-band, so that it's impact is minimized.
Certainly there's a slew of upstream components in the chain. However, if it were not important as you imply, then we likely wouldn't be so sensitive to the differences between different crossover designs. The best that one can hope for and the goal of many of us is to reproduce what is in the recording as faithfully as possible. That is all that's within our control.
Yes. What we are debating, though, is not a signal and itself, it's the same signal with both a highpass and lowpass applied to the two contributors. The BW1 elicits the same result in a sense. At Fc, each leg is down 3db from the design target reference. Each leg is rotated 45 degrees, one in one direction, the other in the reverse direction. This leads to phase quadrature that when summed yields a result of +3db over each section's signal. In other words, flat summed response results since both legs are down 3db at Fc.
Coincidentally, at all other points in the crossover region, the phase remains in phase quadrature as each rotates through its transition region with the vectoral sum resulting in 0db relative result, i.e. a flat summed response.
Dave
soongsc said:
A perusal of back issues of JAES will knock down that assumption pretty quickly. Lipshitz in particular has done a superb job and I'd highly recommend his papers on the subject.
How do you separate out the huge variable of polar pattern? Different crossover orders and sharpness are quite different in this respect.
The graphs show ideal responses as an illustration
You would be well off to go to the link and review it, it will provide more than I can in a thread. To quote one point in it:
The theory of filters holds whether it's in the electrical or acoustical domain. Since our goal is the output from drivers, our only concern is the acoustic response of some driver and a properly designed crossover, passive, active or DSP.
Measurements ultimately have to be made. There are issues and limitations with them, but they can be very accurate if done properly taking into account any known deficiencies, such as low frequency limitations. Fortunately, with the PCs and measurement software widely available and relatively cheap (some can be had free), it's far easier than it once was.
This is because we only care about the acoustic response. The rest is the means to the end. Certainly the electro-mechanical side enters into it, but we simply use the known properties relative to a known signal as applied to the electrical input of the driver and measure the acoustic response.
Dave
hailteflon said:
You would be well off to go to the link and review it, it will provide more than I can in a thread. To quote one point in it:
The theory of filters holds whether it's in the electrical or acoustical domain. Since our goal is the output from drivers, our only concern is the acoustic response of some driver and a properly designed crossover, passive, active or DSP.
Measurements ultimately have to be made. There are issues and limitations with them, but they can be very accurate if done properly taking into account any known deficiencies, such as low frequency limitations. Fortunately, with the PCs and measurement software widely available and relatively cheap (some can be had free), it's far easier than it once was.
This is because we only care about the acoustic response. The rest is the means to the end. Certainly the electro-mechanical side enters into it, but we simply use the known properties relative to a known signal as applied to the electrical input of the driver and measure the acoustic response.
Dave
Dave, this is a problem with forums. The earlier posts are viewed in later days by those who are not fully aware of the original meaning of the early posts.
My first post was in relation to the original post where a question was asked about phase response. The poster wanted to know if the whole issue was a lot of baloney.
My post about broadband all-pass filters was to indicate that phase response is not all that important. These broadband filters gave me a useful understanding about shifting phase at various frequencies.
You seem to be a bad case of reading too many spec sheets. Circuitry is seldom what it is supposed to be. It usually does things that it shouldn't.
Your assumption that you are dealing with objective factual filter data is somewhat ficticious. I still don't know where you get your purely acoustical info.
It is the nature of commercial audio to get people going in a certain direction with specifications such as to keep them from seeing the realities involved. This appears to be a severe case of it.
The problem with reactance in general is that there are curves. In this situation you have overlapping curves.
On top of this a power amp does not have total control over the reactive load. High slew rate and high current capacity help a lot, but there is still overshoot and undershoot. There is an inherent propagation delay from input to output such that the feedback loop is giving the inverting input information that is a fraction of a cycle late.
I'm turning this thread loose, but I will close with this. One should look deeper into such issues as this before jumping to conclusions.
My first post was in relation to the original post where a question was asked about phase response. The poster wanted to know if the whole issue was a lot of baloney.
My post about broadband all-pass filters was to indicate that phase response is not all that important. These broadband filters gave me a useful understanding about shifting phase at various frequencies.
You seem to be a bad case of reading too many spec sheets. Circuitry is seldom what it is supposed to be. It usually does things that it shouldn't.
Your assumption that you are dealing with objective factual filter data is somewhat ficticious. I still don't know where you get your purely acoustical info.
It is the nature of commercial audio to get people going in a certain direction with specifications such as to keep them from seeing the realities involved. This appears to be a severe case of it.
The problem with reactance in general is that there are curves. In this situation you have overlapping curves.
On top of this a power amp does not have total control over the reactive load. High slew rate and high current capacity help a lot, but there is still overshoot and undershoot. There is an inherent propagation delay from input to output such that the feedback loop is giving the inverting input information that is a fraction of a cycle late.
I'm turning this thread loose, but I will close with this. One should look deeper into such issues as this before jumping to conclusions.
I brought up this phase distortion thing because the pulse response graph explained how signal is affected when we invert tweeter polarity in the case of LR2 xover (acoustic LR2).
This is how the original filter config is designed. And most of us have been using this kind of All-Pass-Config xover for so far so long.
Below is spice sim for the electrical reponse of perfect LR4 xover. No driver alignment, driver acoustic, or amplifier overshoot is involved. Dave is right, amplifier feedback has nothing to do with this phase distortion.
This is how the original filter config is designed. And most of us have been using this kind of All-Pass-Config xover for so far so long.
Below is spice sim for the electrical reponse of perfect LR4 xover. No driver alignment, driver acoustic, or amplifier overshoot is involved. Dave is right, amplifier feedback has nothing to do with this phase distortion.
Attachments
Dave, I apologize for the criticism. Perhaps I should read more spec sheets.
I looked at what I wrote after reading through later posts and I must admit it is quite a mess. I was speaking in generalizations pertaining to ideal vs. the real world in audio. Years ago I was oriented primarily by specifications when considering audio equipment. People less "scientific" would say that one should match an amp with a speaker if it sounds good. In later years I began to realize that it is common for an amp and speaker to be a mismatch. It surely has to do with how the amp deals with the particular complex load of a given speaker system.
A wikipedia article on the LR crossover says the following about 2nd order.
"Second order Linwkwitz-Riley crossovers (LR2) have a 12 dB/octave (40 dB/decade) slope. They can be realized by cascading two one-pole filters, or using a Sallen Key filter topology with a Q0 value of 0.5. There's a 180° phase difference between the lowpass and highpass output of the filter, which can be corrected by inverting one signal. In loudspeakers this is usually done by reversing the polarity of one driver if the crossover is passive. For active crossovers inversion is usually done using a unity gain inverting op-amp."
This issue is probably what concerned the original poster.
National ApNote AN346 has an interesting 12db active crossover on the 5th page. Goggle "AN346 LM833" and there will be a link to an adobe file for AN346. This active crossover seems to be different in that it does not have this phase reversal problem.
The following is from AN346. Note zero phase shift from 20 to 20KHz.
"The low-pass and high-pass constant voltage crossover
outputs are plotted in Figure 10. The square-wave response
(not shown) of the summed outputs is simply an inverted
square-wave, and the phase shift (also not shown) is essentially
0§ to beyond 20 kHz."
I will add here that the LM833 is unique in that it has zero input-output lag from 20-20KHz at a gain of up to 40. I have tested this. Perhaps this filter would not give this phase accuracy with an opamp that had input-output lag.
I did finally ascertain that the acoustic qualities that everyone is concerned with here are actually taken from sound pressure measurements in a listening room. At least that is what they did on this particular website. I thought you may have been assuming that the ideal output of the crossover was what was happening in the room.
Mark
I looked at what I wrote after reading through later posts and I must admit it is quite a mess. I was speaking in generalizations pertaining to ideal vs. the real world in audio. Years ago I was oriented primarily by specifications when considering audio equipment. People less "scientific" would say that one should match an amp with a speaker if it sounds good. In later years I began to realize that it is common for an amp and speaker to be a mismatch. It surely has to do with how the amp deals with the particular complex load of a given speaker system.
A wikipedia article on the LR crossover says the following about 2nd order.
"Second order Linwkwitz-Riley crossovers (LR2) have a 12 dB/octave (40 dB/decade) slope. They can be realized by cascading two one-pole filters, or using a Sallen Key filter topology with a Q0 value of 0.5. There's a 180° phase difference between the lowpass and highpass output of the filter, which can be corrected by inverting one signal. In loudspeakers this is usually done by reversing the polarity of one driver if the crossover is passive. For active crossovers inversion is usually done using a unity gain inverting op-amp."
This issue is probably what concerned the original poster.
National ApNote AN346 has an interesting 12db active crossover on the 5th page. Goggle "AN346 LM833" and there will be a link to an adobe file for AN346. This active crossover seems to be different in that it does not have this phase reversal problem.
The following is from AN346. Note zero phase shift from 20 to 20KHz.
"The low-pass and high-pass constant voltage crossover
outputs are plotted in Figure 10. The square-wave response
(not shown) of the summed outputs is simply an inverted
square-wave, and the phase shift (also not shown) is essentially
0§ to beyond 20 kHz."
I will add here that the LM833 is unique in that it has zero input-output lag from 20-20KHz at a gain of up to 40. I have tested this. Perhaps this filter would not give this phase accuracy with an opamp that had input-output lag.
I did finally ascertain that the acoustic qualities that everyone is concerned with here are actually taken from sound pressure measurements in a listening room. At least that is what they did on this particular website. I thought you may have been assuming that the ideal output of the crossover was what was happening in the room.
Mark
The problem with the filter of aforementioned appnote is the same as with any other one: It is working perfect by itself (in this special case it is even transient-perfect or phase accurate) but the topology doesn't take the driver responses into consideration.
The filter in the app note is just one of three ways (it is called state-variable filter BTW) to build such a 2nd/2nd order symmetrical & transient-perfect crossover. The other two possibilities are 1.) the subtractive approach and 2.) two fully seperate filter topologies.
Case 2) has the advantage that the driver responses can be taken into consideration more easily than with the other ones.
Regards
Charles
P.S. I am currently doing some drawings and explanations for a transient-perfect 2nd/1st order crossover that is suitable for fullrange and woofer topologies and which takes driver responses into consideration.
The filter in the app note is just one of three ways (it is called state-variable filter BTW) to build such a 2nd/2nd order symmetrical & transient-perfect crossover. The other two possibilities are 1.) the subtractive approach and 2.) two fully seperate filter topologies.
Case 2) has the advantage that the driver responses can be taken into consideration more easily than with the other ones.
Regards
Charles
P.S. I am currently doing some drawings and explanations for a transient-perfect 2nd/1st order crossover that is suitable for fullrange and woofer topologies and which takes driver responses into consideration.
soongsc said:
?
Well, my points in this thread has been:
1. When listening to a signal that has been affected ONLY by the phase shift from a LR2 filter, the difference is very very small. For crossover frequencies higher than 500 Hz I would say that they pose no problem at all.
2. I don't hold it impossible, though, that one can detect the difference under very special conditions (crossover frequency ~100 Hz, appropriate signal, possibility for direct comparison). In fact I have detected it myself with reasonable statistical significance and blind conditions.
3. The above does not mean that the "sound of LR2 filters" is identical to that of a phase linear first order filter. The filter characteristics affect so many things, like loobing, stopand attenuation etc, that there is bound to be audible difference between filters. But these are not due to the allpass phase that the filter introduces.
4. Impulse response graphs suck as means to illustrate reproduction quality.
The above points are the result of listening, much more so than theoretical reasoning.
Please include information about the deficiencies/idiosyncrasies of the drivers. When you say driver responses are you referring only to frequency response? I am wondering if there are phase issues peculiar to a particular driver, possibly due to the mechanical impedance of the suspension.
I will add here that I have seldom heard an audio system that sounded good. It seems that phase response would be a vital issue. The question is, how altered is the phase response of the original signal? By nature an RC filter shifts phase. Every adjustment of the signal in the studio, especially RIAA and NAB equalizaton, shifts phase, doesn't it?
The one audio system that really taught me a lesson was in a little audio store in 1983. After listening to all his convential equipment I wasn't impressed. Then we went to the next room. Unbelievable sound, flawless, capable taking one off of the planet for a moment. It was a MAGNETIC AMPLIFIER. Whatever it is that makes them different is very important. This rig was very powerful. He showed it to me sitting on the floor outside the room. A little bigger than a MAC 240. Huge transformers. He cranked it way up, and there was nothing but pure powerful music.
In post 33 soongsc has four graphs from the Linkwitzlab website.
Where is the phase distortion. Is it a polarity issue?
Seems like the output pulses would be shifted to the right if they were phase distorted.
Is the distortion in these graphs due solely to the RC constants in these active filters? Or is some of it due to slewing of the opamp?
The first three are probably the result of the time constants of the filter network.
The fourth one looks like amplifier instability. This graph is only on the Linkwitzlab site. Is there a pulse to the right (in graph 4 on the Linkwitzlab site) that has already occurred, but isn't shown on the graph?
This is a crash course in filter theory. I had no idea that the waveform would be changed this much. Yes, this is a square wave pulse, the worst of them all.
I'm hoping someone will show a filter that doesn't do this. Surely it takes a toll on fidelity.
Where is the phase distortion. Is it a polarity issue?
Seems like the output pulses would be shifted to the right if they were phase distorted.
Is the distortion in these graphs due solely to the RC constants in these active filters? Or is some of it due to slewing of the opamp?
The first three are probably the result of the time constants of the filter network.
The fourth one looks like amplifier instability. This graph is only on the Linkwitzlab site. Is there a pulse to the right (in graph 4 on the Linkwitzlab site) that has already occurred, but isn't shown on the graph?
This is a crash course in filter theory. I had no idea that the waveform would be changed this much. Yes, this is a square wave pulse, the worst of them all.
I'm hoping someone will show a filter that doesn't do this. Surely it takes a toll on fidelity.
As a coarse approximation drivers are minumum-phase devices. An important property of those is that amplitude-response and phase-response have an inevitable mathematical interrelationship. Though there are effects where this isn't true anymore. But the LF rollof can in fact be modelled quite closely as a highpass whereas the HF rolloff can be modelled as a lowpass.
Another interesting fact is that if one is equalising a response irregularity (that is minimum-phase) with its (also minimum-phase) mirror image one also equalises the phase-response. This has implications on speaker equalising as well as the NAB and RIAA equalisation: If done well the result would be a flat phase-response within a reasonable frequency range.
Now the ususal speaker/crossover combinations are not minimum-phase, they are allpasses. That means even if the amplitude-response is flat there is frequency dependant phase-shift.
The crossover topology that I mentioned considers the drivers being bandpass filters and it showed to be quite accurate. I don't want to claim that it is the most accurate crossover possible since one has to make truncations and approximations - but when I tried it with my Manger/Audiotechnology combination I had to listen endlessly to records that I though I couldn't listen to anymore (because they began to be boring or even annoying after listening to them too much before) !
Regards
Charles
This thread is the biggest mess I saw in weeks, 
Hi HailTeflon,
seems that you're confused :
"Where is the phase distortion. Is it a polarity issue?"
"Is the distortion in these graphs due solely to the RC constants in these active filters? Or is some of it due to slewing of the opamp?"
"The fourth one looks like amplifier instability. This graph is only on the Linkwitzlab site."
I believe if you ask one by one, some people here can answer your questions, it seems you're very confused,my suggestion is, it would be much less confusing if you take some time to read what Linkwitz is saying, and assume that he don't have any broken amp / opamp to begin with(which I believe is the case).
Hartono
Hi HailTeflon,
seems that you're confused :
"Where is the phase distortion. Is it a polarity issue?"
"Is the distortion in these graphs due solely to the RC constants in these active filters? Or is some of it due to slewing of the opamp?"
"The fourth one looks like amplifier instability. This graph is only on the Linkwitzlab site."
I believe if you ask one by one, some people here can answer your questions, it seems you're very confused,my suggestion is, it would be much less confusing if you take some time to read what Linkwitz is saying, and assume that he don't have any broken amp / opamp to begin with(which I believe is the case).
Hartono
you could use those simulation softs :
- active filter
- phase audibility
You can listen to the signal while you modify various parameters and find for yourself what is audible or not.
Hi Nelson
I agree, I have found that one benefits from time preservation even higher up.
The problem in “hifi” is however with acoustically small direct radiators, in the Heyser view of acoustic phase, lag behind the input signal, up to about –90 degrees, over much of there bandwidth. This prevents them from preserving the waveshape of a complex broadband signal (like a square wave) even without a crossover.
An electrostatic driver, a resistively governed system like a Manger or critical horn can preserve the waveshape over some broad band but each normally has other issues.
The direct radiator is a mass controlled, acceleration device.
Its radiator velocity has to fall in order to compensate for its increasing radiation efficiency.
The slope of that curve is compensated by the Rdc / moving mass as C filter, internal to the driver, while the amplitude is compensated, the radiation resistance curve does not have the normal phase shift, it is largely a frequency dependent R. As a result, the phase shift associated with the R/C filter, remains as a residual, while the amplitudes have canceled.
I love hifi, it is my hobby and always has been and why I read these forums etc.
I design speakers for a living and have several solutions to the problems I see here.
If the hifi market were different, if our company were going in a different direction, I would try come up with one of these for the home.
Anyway, while not at all for the home, this commercial speaker solution (SH-50) might be interesting technically as it’s phase response shows no crossover.
This particular speaker also radiates as a single source, has near constant directivity, can reproduce a square wave, anywhere where in front of it, from fair to near perfect, anywhere between about 220Hz to 2.6KHz.
Explanation starts about half way down.
http://www.danleysoundlabs.com/pdf/danley_tapped.pdf
Best,
Tom Danley
I agree, I have found that one benefits from time preservation even higher up.
The problem in “hifi” is however with acoustically small direct radiators, in the Heyser view of acoustic phase, lag behind the input signal, up to about –90 degrees, over much of there bandwidth. This prevents them from preserving the waveshape of a complex broadband signal (like a square wave) even without a crossover.
An electrostatic driver, a resistively governed system like a Manger or critical horn can preserve the waveshape over some broad band but each normally has other issues.
The direct radiator is a mass controlled, acceleration device.
Its radiator velocity has to fall in order to compensate for its increasing radiation efficiency.
The slope of that curve is compensated by the Rdc / moving mass as C filter, internal to the driver, while the amplitude is compensated, the radiation resistance curve does not have the normal phase shift, it is largely a frequency dependent R. As a result, the phase shift associated with the R/C filter, remains as a residual, while the amplitudes have canceled.
I love hifi, it is my hobby and always has been and why I read these forums etc.
I design speakers for a living and have several solutions to the problems I see here.
If the hifi market were different, if our company were going in a different direction, I would try come up with one of these for the home.
Anyway, while not at all for the home, this commercial speaker solution (SH-50) might be interesting technically as it’s phase response shows no crossover.
This particular speaker also radiates as a single source, has near constant directivity, can reproduce a square wave, anywhere where in front of it, from fair to near perfect, anywhere between about 220Hz to 2.6KHz.
Explanation starts about half way down.
http://www.danleysoundlabs.com/pdf/danley_tapped.pdf
Best,
Tom Danley
Tom
Have you ever done any experiments to see if there is greater perceived loudness with a transient accurate system over the non-transient ones ?
How much improvement in clarity is there with transient-perfect systems in live-sound reinforcement situations ?
If size doesn't matter that much your solution is quite elegant. Unfortunately not everyone can set up a system like this for home use.
Regards
Charles
Have you ever done any experiments to see if there is greater perceived loudness with a transient accurate system over the non-transient ones ?
How much improvement in clarity is there with transient-perfect systems in live-sound reinforcement situations ?
If size doesn't matter that much your solution is quite elegant. Unfortunately not everyone can set up a system like this for home use.
Regards
Charles