I can't really create anything like a time coherent speaker system either just by listening, or by back of the envelope calculations. Nobody can. It's too complex.
Everyone (except tweakophiles) agrees. Measurements are required. Every piece of the system has it's own time behavior, and crossovers and equalizers further modify that.
REW is really about room acoustics, and not so much speaker performance. From the standpoint of REW, speaker perfomance is a given.
To design a speaker...which is effectively what I'm doing with my combination of subwoofer and supertweeter to a "full range" eelectrostat, I needed a speaker design program.
Many years ago, when I decided it was hopeless to try to improve on the LS3/5A, I was using Liberty Audiosuite 3. That was a huge step up in my measurement capabilities, and served me well from 1999 to about 2010.
Using MLS (which I thought was a requirement--but now I see it was merely an optimization) and auto-magic gating (that really IS a requirement, unless you do outdoor or anechoic testing, or simple time recording instead of FFT convolution) LAUD3 was always able to construct plausible step response curves.
But that program HAD to run on an old fashioned ISA bus PC which could host a DSP soundcard like the Turtle Beach FIJI. When those parts started to fail, I was out of luck, for a decade.
For the past 5 years I've been occasionally trying to get my LAUD3 system working again. Without getting there yet. There's always some issue that comes up in the setup process. It wasn't much different in 1999, except the issues weren't so much hardware failures. Getting the Fiji working, and re-working after every system upgrade, I figured I had re-installed Windows 95 about one hundred times. It took re-installations because the Plug-n-Play system learned things the wrong way, and I had to start from scratch.
Anyway, I hold onto the past too much in some cases. I should have moved on. I did try to move on to the REW program, but it is unsatisfactory for the speaker time alignment. I've written about that. I needed a real "speaker designer's program" like LAUD, and for the past year or so I figured ARTA would be a good choice.
Finally, a few days ago, I got started by downloading ARTA.
For the first day, I was getting nowhere. I have yet to figure out how to do the sharp gating required to see the leading edge from my speakers. If you use the first 3 methods for deriving an impulse response, and from that a step response, you are convoluting a recording lasting several seconds. Everything beyond the first few milliseconds, maybe as few as 4 ms, is compounded by reflections. This is therefore a room/speaker analysis, and not a speaker analysis. To do the time alignment, we really need to look at the first few milliseconds somehow.
And none of the menus I've looked at yet have anything to do with gating.
But I figured out an easier and much more intuitive way to do this. Friends have mentioned the concept to me before, as much as 35 years ago. A storage oscilloscope!
Around 1985 I picked up a surplus storage oscilloscope, but it was a dud, a really poor oscilloscope with a really poor storage feature tacked on, and I never had much time to mess with it anyway with a full time job and endless attempts to find a good girlfriend relationship.
Fortunately, ARTA has a "time recorder" feature which works just like a pretty good storage oscilloscope. And with this, I can look at the actual behavior as time unfolds, not a convolution of the longer term behavior. This is another area where I think the professionals are too caught up in their own magic. Simpler is better, or at least simpler.
Even then, it seemed like I was getting nowhere. I could measure the latency of the panels and the supertweeters. But it was way different from I expected, the the supertweeter considerably delayed. Attempts to remove delay from the supertweeter by dialing back the delay setting on the Behringer DEQ seemed to do nothing at all. My first guess was that even in this supposedly time recording, with the loopback being taken from my audio interface itself, the ARTA program was doing some kind of additional automagic time shifting, just as had plagued me before when convolution was being done.
But it turned out, the problem was more in the DEQ. I noticed I was changing the delay setting for MAIN, but the AUX setting was not changing. So I started changing the delay setting for AUX, and now I was able to vary the delay. It seems that the Digital Outputs are part of the AUX circuit, not the MAIN circuit.
With this change to my method, I was able to adjust the delays for panel and tweeter (on left channel) to match within about 0.02 ms, by measuring one at a time and adjusting. There are some tiny undulations prior to the supertweeter step up signal which I'm assuming are digital artifacts. Other than that there is finally zero ambiguity. Finally. I feel confident enough in this approach that I can now fearlessly pivot to using a different sampling rate in the supertweeter path, 96kHz vs 48kHz, and restore my system to virtual 96kHz operation (albeit only in the supertweeter path...but that's where it needs the 96kHz after all), and re-set the alignment by doing this again...it only takes a few minutes. For that matter, I can now easily "roll" the DAC used in the midrange path (the panels), and likewise readjust the alignment in a few minutes.
But moving on to adjusting the subwoofer time alignment, at first I seemed to be back in the same old boat. Changing the delay adjustments in the subwoofer path seemed, at first, to make no difference, and this time I started out adjusting the "AUX" path. Only after I did a number of measurements, first moving the mike closer as a sanity check (moving mike closer DID have the expected effect), and flipping the MAIN/AUX switch a few times, did it seem finally to be working again.
Perhaps the answer isn't so much adjusting the main vs aux delay, but flipping the switch so it records the change??? I notice also that the AUX path, which includes the digital output, is selected to "be the same" as the MAIN path in an earlier I/O menu page. However, that page doesn't show where the adjusted delay is applied.
For now, it still seems that the answer is to adjust the AUX delay, but possibly flipping back and forth between AUX and MAIN to be sure it "sticks."
But even with that little technical glitch fixed, the Subwoofer time response leaves me with a conundrum.
The subwoofer response begins with an 7.5 ms negative cycle before the positive cycle begins. This is large enough for me to conclude it is NOT a digital artifact.
I've also done endless earlier adjustments of the subwoofer polarity, using a variety of methods, including a polarity checking app. As far as I know, the subwoofer is In Polarity, it just responds with this little leading negative cycle.
So now the question is, do I line up the midrange response, which appears not to have such a leading negative cycle (or, if it does, it's so small it isn't clear on the default scale) with the part where the subwoofer response begins, at the beginning of the leading negative cycle?
Or, do I line up the midrange response with subwoofer curve where it's back to baseline and now moving positive? Or, do I line up the midrange response with the subwoofer curve where it's at the bottom of the leading negative cycle and is starting to move up, even though it hasn't reached the zero axis yet? Or something else entirely?
I suppose it would help to understand what this leading negative cycle is all about. I suspect now, if I studied the midrange time response I'd see a similar thing, only very much smaller. If that turns out to be true, it could be this leading negative portion is related to the high pass response of all speaker drivers. (I suppose ionic and fan-based speakers could be an exception, but those are hardly worth mentioning due to other impracticabilities.)
Linkwitz talks about this inevitable high pass response, and where it is most important to deal with is in the subwoofer, because it creates a phase error in the crossover with the midrange that can be quite large.
The phase lead is 180 degrees at the natural (or unnatural) low frequency cutoff of the subwoofer, but in his example reduces to 23 degrees at the 100 Hz crossover (same crossover frequency as I am using).
That's because the phase lead falls as the frequency goes higher relative to the cutoff frequency.
But meanwhile, the lowpass being applied to the crossover for the subwoofer has phase lag, which is supposed to be 180 degrees at the crossover frequency of an LR4 crossover.
At that same crossover freuqncy, the LR4 highpass has 180 degrees of phase lag. Thus both highpass and lowpass have gotten to the same point, but in different directions. As we move either way from the crossover frequency, they remain in phase, as the leading goes up or down and the lagging does the exact reverse.
Linkwitz proposes either using an allpass filter (and that is what he does) or a time delay system (applicable to DSP type crossovers) to correct the phase shift caused by the woofer's high pass function.
But now it occurs to me, why isn't he concerned about the midrange high pass function? That highpass function is closer to the crossover point, and it seems to me would contribute MORE relative phase error at the crossover frequency.
In fact, the ideal for ordinary crossovers (without going to FIR linear phase types) is precisely this: subwoofer and midrange having exactly the same inherent highpass function. One way to fix the problem is to alter the subwoofer's highpass function to match the midrange highpass function.
In my case, the subwoofer has rolloff very close to 20Hz, whereas the electrostats roll off around 40Hz--somewhat extended by their own resonance and in my case the room resonance.
So, actually, despite the vast difference in appearance of the time output signal (because...the subwoofer goes far lower but not as high) the actual phase lead difference at the crossover frequency is fairly small.
I'm sure most of my friends would believe in lining up the first semblance of output, whether it's up or down, phase lead or lag. One of my friends would always look at speakers claiming to be time aligned, and if the voice coils weren't lining up, he'd say it is a lie. I'd always say there were probably other factors, but I never knew what those other factors were, until the one day, for just a few moments, I myself got to talk to Linkwitz at a trade show (Santa Monica, 1987 IIRC, or maybe it was San Francisco in 1997). He told me more in that one minute than anyone else has in a lifetime. I asked why I couldn't have phase linear crossovers and he told me I could, with FIR, but that had other issues (as is aptly described on a webpage now). And about the voice coil lining up not being indentical with time alignment, that it was due to the high pass function of the drivers (as is aptly described on a webpage now).
So it would seem a first order approximation would be to line up the signals where they begin, regardless of direction.
AND then, to possibly adjust further, to correct the relative phase lead at the crossover frequency.
From what I know already, I can sort of predict what that will be like. At the crossover frequency, ignoring the effect of the crossover itself, the panels will have MORE lag than the subwoofer because they are closer to their cutoff than the subwoofer is to it's.
(The disconnect between this observation and Linkwitz' focus on correcting the subwoofer lag, confuses me.)
If the subwoofer is lagging by 23 degrees, the panels might be lagging by 46 degrees.
Lagging more, means the panels should be corrected with less delay than would otherwise be required.
So after lining up the starting points (which will, btw, require adding much more delay to the panels, for some reason the subs simply have far more delay than I expected, possibly because internal digital processing) I will then possibly have to reduce that additional delay, perhaps back to more or less where it was before (!?!)
What is 23 degrees at 100 Hz? 360 degrees is about 10ms, so my back of the envelope calculation says about .64 ms.
A workable process might be to do the first alignment based on first driver movement, then decrease panel delay to maximize output at the crossover frequency, up to about 2 ms, expecting the maximum to be around .64 ms.
Everyone (except tweakophiles) agrees. Measurements are required. Every piece of the system has it's own time behavior, and crossovers and equalizers further modify that.
REW is really about room acoustics, and not so much speaker performance. From the standpoint of REW, speaker perfomance is a given.
To design a speaker...which is effectively what I'm doing with my combination of subwoofer and supertweeter to a "full range" eelectrostat, I needed a speaker design program.
Many years ago, when I decided it was hopeless to try to improve on the LS3/5A, I was using Liberty Audiosuite 3. That was a huge step up in my measurement capabilities, and served me well from 1999 to about 2010.
Using MLS (which I thought was a requirement--but now I see it was merely an optimization) and auto-magic gating (that really IS a requirement, unless you do outdoor or anechoic testing, or simple time recording instead of FFT convolution) LAUD3 was always able to construct plausible step response curves.
But that program HAD to run on an old fashioned ISA bus PC which could host a DSP soundcard like the Turtle Beach FIJI. When those parts started to fail, I was out of luck, for a decade.
For the past 5 years I've been occasionally trying to get my LAUD3 system working again. Without getting there yet. There's always some issue that comes up in the setup process. It wasn't much different in 1999, except the issues weren't so much hardware failures. Getting the Fiji working, and re-working after every system upgrade, I figured I had re-installed Windows 95 about one hundred times. It took re-installations because the Plug-n-Play system learned things the wrong way, and I had to start from scratch.
Anyway, I hold onto the past too much in some cases. I should have moved on. I did try to move on to the REW program, but it is unsatisfactory for the speaker time alignment. I've written about that. I needed a real "speaker designer's program" like LAUD, and for the past year or so I figured ARTA would be a good choice.
Finally, a few days ago, I got started by downloading ARTA.
For the first day, I was getting nowhere. I have yet to figure out how to do the sharp gating required to see the leading edge from my speakers. If you use the first 3 methods for deriving an impulse response, and from that a step response, you are convoluting a recording lasting several seconds. Everything beyond the first few milliseconds, maybe as few as 4 ms, is compounded by reflections. This is therefore a room/speaker analysis, and not a speaker analysis. To do the time alignment, we really need to look at the first few milliseconds somehow.
And none of the menus I've looked at yet have anything to do with gating.
But I figured out an easier and much more intuitive way to do this. Friends have mentioned the concept to me before, as much as 35 years ago. A storage oscilloscope!
Around 1985 I picked up a surplus storage oscilloscope, but it was a dud, a really poor oscilloscope with a really poor storage feature tacked on, and I never had much time to mess with it anyway with a full time job and endless attempts to find a good girlfriend relationship.
Fortunately, ARTA has a "time recorder" feature which works just like a pretty good storage oscilloscope. And with this, I can look at the actual behavior as time unfolds, not a convolution of the longer term behavior. This is another area where I think the professionals are too caught up in their own magic. Simpler is better, or at least simpler.
Even then, it seemed like I was getting nowhere. I could measure the latency of the panels and the supertweeters. But it was way different from I expected, the the supertweeter considerably delayed. Attempts to remove delay from the supertweeter by dialing back the delay setting on the Behringer DEQ seemed to do nothing at all. My first guess was that even in this supposedly time recording, with the loopback being taken from my audio interface itself, the ARTA program was doing some kind of additional automagic time shifting, just as had plagued me before when convolution was being done.
But it turned out, the problem was more in the DEQ. I noticed I was changing the delay setting for MAIN, but the AUX setting was not changing. So I started changing the delay setting for AUX, and now I was able to vary the delay. It seems that the Digital Outputs are part of the AUX circuit, not the MAIN circuit.
With this change to my method, I was able to adjust the delays for panel and tweeter (on left channel) to match within about 0.02 ms, by measuring one at a time and adjusting. There are some tiny undulations prior to the supertweeter step up signal which I'm assuming are digital artifacts. Other than that there is finally zero ambiguity. Finally. I feel confident enough in this approach that I can now fearlessly pivot to using a different sampling rate in the supertweeter path, 96kHz vs 48kHz, and restore my system to virtual 96kHz operation (albeit only in the supertweeter path...but that's where it needs the 96kHz after all), and re-set the alignment by doing this again...it only takes a few minutes. For that matter, I can now easily "roll" the DAC used in the midrange path (the panels), and likewise readjust the alignment in a few minutes.
But moving on to adjusting the subwoofer time alignment, at first I seemed to be back in the same old boat. Changing the delay adjustments in the subwoofer path seemed, at first, to make no difference, and this time I started out adjusting the "AUX" path. Only after I did a number of measurements, first moving the mike closer as a sanity check (moving mike closer DID have the expected effect), and flipping the MAIN/AUX switch a few times, did it seem finally to be working again.
Perhaps the answer isn't so much adjusting the main vs aux delay, but flipping the switch so it records the change??? I notice also that the AUX path, which includes the digital output, is selected to "be the same" as the MAIN path in an earlier I/O menu page. However, that page doesn't show where the adjusted delay is applied.
For now, it still seems that the answer is to adjust the AUX delay, but possibly flipping back and forth between AUX and MAIN to be sure it "sticks."
But even with that little technical glitch fixed, the Subwoofer time response leaves me with a conundrum.
The subwoofer response begins with an 7.5 ms negative cycle before the positive cycle begins. This is large enough for me to conclude it is NOT a digital artifact.
I've also done endless earlier adjustments of the subwoofer polarity, using a variety of methods, including a polarity checking app. As far as I know, the subwoofer is In Polarity, it just responds with this little leading negative cycle.
So now the question is, do I line up the midrange response, which appears not to have such a leading negative cycle (or, if it does, it's so small it isn't clear on the default scale) with the part where the subwoofer response begins, at the beginning of the leading negative cycle?
Or, do I line up the midrange response with subwoofer curve where it's back to baseline and now moving positive? Or, do I line up the midrange response with the subwoofer curve where it's at the bottom of the leading negative cycle and is starting to move up, even though it hasn't reached the zero axis yet? Or something else entirely?
I suppose it would help to understand what this leading negative cycle is all about. I suspect now, if I studied the midrange time response I'd see a similar thing, only very much smaller. If that turns out to be true, it could be this leading negative portion is related to the high pass response of all speaker drivers. (I suppose ionic and fan-based speakers could be an exception, but those are hardly worth mentioning due to other impracticabilities.)
Linkwitz talks about this inevitable high pass response, and where it is most important to deal with is in the subwoofer, because it creates a phase error in the crossover with the midrange that can be quite large.
The phase lead is 180 degrees at the natural (or unnatural) low frequency cutoff of the subwoofer, but in his example reduces to 23 degrees at the 100 Hz crossover (same crossover frequency as I am using).
That's because the phase lead falls as the frequency goes higher relative to the cutoff frequency.
But meanwhile, the lowpass being applied to the crossover for the subwoofer has phase lag, which is supposed to be 180 degrees at the crossover frequency of an LR4 crossover.
At that same crossover freuqncy, the LR4 highpass has 180 degrees of phase lag. Thus both highpass and lowpass have gotten to the same point, but in different directions. As we move either way from the crossover frequency, they remain in phase, as the leading goes up or down and the lagging does the exact reverse.
Linkwitz proposes either using an allpass filter (and that is what he does) or a time delay system (applicable to DSP type crossovers) to correct the phase shift caused by the woofer's high pass function.
But now it occurs to me, why isn't he concerned about the midrange high pass function? That highpass function is closer to the crossover point, and it seems to me would contribute MORE relative phase error at the crossover frequency.
In fact, the ideal for ordinary crossovers (without going to FIR linear phase types) is precisely this: subwoofer and midrange having exactly the same inherent highpass function. One way to fix the problem is to alter the subwoofer's highpass function to match the midrange highpass function.
In my case, the subwoofer has rolloff very close to 20Hz, whereas the electrostats roll off around 40Hz--somewhat extended by their own resonance and in my case the room resonance.
So, actually, despite the vast difference in appearance of the time output signal (because...the subwoofer goes far lower but not as high) the actual phase lead difference at the crossover frequency is fairly small.
I'm sure most of my friends would believe in lining up the first semblance of output, whether it's up or down, phase lead or lag. One of my friends would always look at speakers claiming to be time aligned, and if the voice coils weren't lining up, he'd say it is a lie. I'd always say there were probably other factors, but I never knew what those other factors were, until the one day, for just a few moments, I myself got to talk to Linkwitz at a trade show (Santa Monica, 1987 IIRC, or maybe it was San Francisco in 1997). He told me more in that one minute than anyone else has in a lifetime. I asked why I couldn't have phase linear crossovers and he told me I could, with FIR, but that had other issues (as is aptly described on a webpage now). And about the voice coil lining up not being indentical with time alignment, that it was due to the high pass function of the drivers (as is aptly described on a webpage now).
So it would seem a first order approximation would be to line up the signals where they begin, regardless of direction.
AND then, to possibly adjust further, to correct the relative phase lead at the crossover frequency.
From what I know already, I can sort of predict what that will be like. At the crossover frequency, ignoring the effect of the crossover itself, the panels will have MORE lag than the subwoofer because they are closer to their cutoff than the subwoofer is to it's.
(The disconnect between this observation and Linkwitz' focus on correcting the subwoofer lag, confuses me.)
If the subwoofer is lagging by 23 degrees, the panels might be lagging by 46 degrees.
Lagging more, means the panels should be corrected with less delay than would otherwise be required.
So after lining up the starting points (which will, btw, require adding much more delay to the panels, for some reason the subs simply have far more delay than I expected, possibly because internal digital processing) I will then possibly have to reduce that additional delay, perhaps back to more or less where it was before (!?!)
What is 23 degrees at 100 Hz? 360 degrees is about 10ms, so my back of the envelope calculation says about .64 ms.
A workable process might be to do the first alignment based on first driver movement, then decrease panel delay to maximize output at the crossover frequency, up to about 2 ms, expecting the maximum to be around .64 ms.
No comments:
Post a Comment