Careful with that.I tried with minimum phase measurements and it works fine
Be aware that "minimum phase" measurements which calculate the minimum phase response only from the amplitude response rather than actually measuring true phase cannot capture any relative time delay differences between the two drivers due to misalignment of the acoustic centres.
This means that unless the drivers do have their acoustic centres exactly aligned with respect to the design axis and measurement microphone, you won't get valid phase information to allow you to get the phase tracking through the crossover region correct.
For example in my case where the tweeter is ahead of the woofer and needs the delay of the all pass filter the minimum phase responses would make it seem as if everything was ok and no all pass filter was required, when in fact that's not the case and the phase is out by nearly 180 degrees at the crossover frequency due to the different time of flight between the drivers.
To get valid phase data for the simulation you also need the microphone height equidistant between tweeter and woofer (or on the design axis of you choosing, which some people make the tweeter axis) in the exact same place for both driver measurements, and either take the measurement in dual channel mode (where the 2nd channel is a loop back connection from the driver terminals) or use a sound card that has a very reliable, consistent processing delay from one measurement to another.
Some sound cards are just not usable in single channel mode for phase measurement because there are slight random differences in processing delay each time the sound device is opened that will cause random errors in the phase measurement. Even if the sound card seems good for single channel measurements you can never quite be sure whether phase errors are being introduced.
The only truly reliable way to get the phase measurement right is to use dual channel measurements where one input is the microphone and the other is a direct electrical sample from the driver terminals used as a reference for the cross correlator. With this method it doesn't matter if the processing delay of the sound card changes from one measurement to another, it will be cancelled out. This allows for very accurate repeatable phase measurements. Dual channel mode also cancels out any frequency response errors due to the test amplifier or speaker cable resistance which is handy, provided the pickup point is directly at the speaker terminals.
It's a bit more work to set up dual channel measurement and capture true phase, and requires a sound card that can input line level on one channel and microphone level on the other channel simultaneously (or an external mic pre-amp) but IMO you really do need to do this to get valid results for the summed response in a simulator.
Last edited:
Yes, of course, right up until round-off error starts to wreak things, but this is implementation specific. I am not positive which is less sensitive to round-off, but I would not be surprised if FIR was.
FYI, the whole theory of these kinds of filters goes back to the 50's in statistical quality control. Box and Jenkins researched the limits of what control systems could do and basically developed the whole theory of "spectral methods." These methods show what can and cannot be done by orders and types of filters.
FIR is an all zero filter, no poles, and IIR is both poles and zero's always in conjugate pairs (poles) - hence minimum, phase. There are filters that are all poles, called Auto-Regressive, which are quite handy in stock market predictions, but can't be implemented with electric components. They are sometimes used in SONAR however.
Completely agree.
With a speaker that is designed properly from the beginning, heroic crossover efforts are not needed. There is still engineering and craft in getting it just right, but brute force is not required.Better sound and/or lower cost (monetary or electricity). Except for hiss, as I said.
Last edited:
"Better sound" is debatable and "lower cost" is certainly not the case when you factor in the need for two amps instead of one.
Pano, your very simple question is very thought provoking.
I repeated a few times that my passive makes no hiss, but the same is not true for an active, but let's leave that aside but ignore that for a moment.
If there is no "care" in crafting, one could spend a few cents on a electrolytic cap for a tweeter - oh wait, many speakers have exactly that. But it is not sounding good of course. Active wins on sound quality easily even when used with minimal amount of work.
Then one would phase align, fix a response peak here, a dip there, and get a pretty complex crossover, which may require e.g. a 12awg Litz air core to pull off, by that time it is pricewise firmly in audiophile territory. An active can easily offer the same level of complexity at a lower cost, even after accounting for extra channels of amplification.
There must be a break-even point, I think, where the cost of active vs cost of passive are the same. What I am essentially claiming is that the active will win on sound at the same cost, because active will always allow you to tweak the sound to your liking with a single up-front cost. (You can always put more care into active without increasing cost, but the same is not true for passive.) No one would realistically implement Linkwitz Transform passively, for example, unless you are talking about line-level passive.
I repeated a few times that my passive makes no hiss, but the same is not true for an active, but let's leave that aside but ignore that for a moment.
If there is no "care" in crafting, one could spend a few cents on a electrolytic cap for a tweeter - oh wait, many speakers have exactly that. But it is not sounding good of course. Active wins on sound quality easily even when used with minimal amount of work.
Then one would phase align, fix a response peak here, a dip there, and get a pretty complex crossover, which may require e.g. a 12awg Litz air core to pull off, by that time it is pricewise firmly in audiophile territory. An active can easily offer the same level of complexity at a lower cost, even after accounting for extra channels of amplification.
There must be a break-even point, I think, where the cost of active vs cost of passive are the same. What I am essentially claiming is that the active will win on sound at the same cost, because active will always allow you to tweak the sound to your liking with a single up-front cost. (You can always put more care into active without increasing cost, but the same is not true for passive.) No one would realistically implement Linkwitz Transform passively, for example, unless you are talking about line-level passive.
Great post.
The only way i trust single channel phase measurements is one driver at a time.
With loopback, or acoustic timing reference, as found in REW, I can compare measured delays, and set processing delays to then look at multi driver traces,......
but still I'd much rather be looking at dual channel traces.
And even with dual channel traces, the measurement software is always trying to build a continous trace...which means provide continuity when maybe there really isn't all that much continuity to tie together...
The best test IME, is to capture phase traces separately, and have a way to overlay them to compare, incorporating time delay.
But then again, you can skip all those worrisome efforts, by just going FIR and taking phase to zero on drivers and x-overs
Cyberstudio - I simply do not agree with your "lower cost" claim unless one is wasting tons of money on "audiophool" caps and inductors. A sensible/competent designer can always beat an active system on total cost for comparable performance. Back when DSP modules cost 1000's we did an identical system active and passive and the differences were negligible. Hence, back then it was no contest cost wise. Today it is quite competitive and I have implemented many active systems, but they still suffer a cost burden due to the second amp.
My speakers are now active, as I have said, and I will be replacing them with passive for the greater convenience and lower cost.
My speakers are now active, as I have said, and I will be replacing them with passive for the greater convenience and lower cost.
Thanks.
The stuff about poles is over my head.
One thing I do get...the engineering theory isn't even close to new.
The only thing that seems new is more affordable, faster DSP processing...
This has not been my experience with Holm and my USB card. I cannot detect any problems with the individual driver delays and phase as long as the time base is synced and working properly (which is most of the time.) If there were problems then this would show up in a comparison between the simulated and actual implementations, which is small enough so as to not be an issue with me. I've seen it happen, sure, but most times the two are lock-step together.
I took DBMandrake's post to be saying single channel doesn't work across multi-driver phase measurements...
Like you, I have no problem with single channel, and single driver...
....but multi-driver is a different animal IME.
I don't see the sound card issues he mentioned, just the 'minimum phase calculated from measured magnitude' issue.
Like you, I have no problem with single channel, and single driver...
....but multi-driver is a different animal IME.
I don't see the sound card issues he mentioned, just the 'minimum phase calculated from measured magnitude' issue.
Last edited:
Most of the time, is the critical point. Why take the chance of it being correct "most" of the time ?
I'm also referring to taking separate measurements of woofer and tweeter one at a time with identical measurement conditions including identical microphone position.
Like you I used to use signal channel measurement mode back when I had a sound card that did not lend itself to dual channel measurements. (built in mic pre-amp but only mono recording on the mic input)
And that particular sound card which was a good quality one did seem to give fairly consistent phase/timing information from one measurement to another.
But I have used other sound cards where this is absolutely not the case and there is a significant fraction of a millisecond difference in processing delays from one sound card device open to another (between measurements) rendering relative phase between multiple driver measurements untrustworthy unless they are repeated multiple times and cross checked against each other to look for any outliers.
Knowing a little bit about how the sound card drivers in windows work, I simply would not trust single channel mode for professional level time critical measurements.
A sound card that can support dual channel mode is not hard to come by these days. I now have a Behringer UMC204HD which has completely independent left and right inputs - I can switch the left input to mic pre-amp mode directly plugging in a phantom powered XLR microphone, and set the right input to line level, connecting my isolation/step down audio transformer which then connects directly to the speaker terminals for the 2nd channel sample. Both have independent gain controls and attenuator pad buttons to allow the two inputs to be adjusted to roughly the same level as seen in the measurement software.
As well as eliminating all possible sources of timing error and giving 100% consistent/reliable relative time/phase measurements, it eliminates frequency response errors due to the test amplifier and speaker cable resistance, eliminates amplitude errors between measurements for example due to the amplifier gain changing slightly, (bumped volume control, imperfect speaker cable connections) and eliminates most forms of frequency response error of the sound card itself. (except for a differential frequency response error between left and right input channel)
So why wouldn't you want to use it if you can ? I tend to use it even for measurements where I'm not worried about absolute phase/time due to the other advantages like elimination of amplifier gain and frequency response errors. The only time I use single channel mode is in RTA mode in ARTA where the dual channel RTA mode is significantly slower. But I only use RTA mode for casual use when I want to very quickly check something in real time, not for measurements that might be used in a crossover design.
Yes you might get away with single channel phase critical measurements if you're careful, have a known good well tested and carefully configured sound card and follow a careful measurement process but I couldn't in good conscience recommend it as a general method that is universally applicable when I know from first hand experience the ways it can go wrong.
You don't want to spend weeks designing and building a crossover based on faulty phase measurements and have to scrap it and start over...
Last edited:
Simon, I was measuring with HolmImpulse with time zero locked, however I did notice that there was some drift I could not account for when I did a second tweeter measurement.
I decided to extract minimum phase and work out my actual offset and use those instead (using Jeff Bagbies method of measuring each driver in turn and then both playing together, extract minimum phase and then use those measurements in sim with both drives playing till the sim curve overlays almost perfectly with the measured curve) as a sanity check.
It turned out I needed -3.5mm on the tweeter. Note that these drivers are very close to time aligned, so I guess the difference was me not being quite symetrical on the tweeter vertical axis (it is an MTM).
Below is the difference I got between my first tweeter measurement and the last sanity check one. around 0.06ms offset. I'm assuming that this was a latency change in the sound card. I found that my midbass measrements seemed to be agreeing with my tweeter measurements from a time perspective, and sims using the measured phase (without offset) and the minimum phase (with offset) were basically the same.
I used to use speaker workshop for measurements (which does do the two channel thing, but found it too unreliable and switched to holm impulse. After reading the vituixcad measurements doc, I might have to give ARTA a try
Note that one of these measurements was high passed at 200Hz and the other was from 20Hz.
Tony.
I decided to extract minimum phase and work out my actual offset and use those instead (using Jeff Bagbies method of measuring each driver in turn and then both playing together, extract minimum phase and then use those measurements in sim with both drives playing till the sim curve overlays almost perfectly with the measured curve) as a sanity check.
It turned out I needed -3.5mm on the tweeter. Note that these drivers are very close to time aligned, so I guess the difference was me not being quite symetrical on the tweeter vertical axis (it is an MTM).
Below is the difference I got between my first tweeter measurement and the last sanity check one. around 0.06ms offset. I'm assuming that this was a latency change in the sound card. I found that my midbass measrements seemed to be agreeing with my tweeter measurements from a time perspective, and sims using the measured phase (without offset) and the minimum phase (with offset) were basically the same.
I used to use speaker workshop for measurements (which does do the two channel thing, but found it too unreliable and switched to holm impulse. After reading the vituixcad measurements doc, I might have to give ARTA a try
Note that one of these measurements was high passed at 200Hz and the other was from 20Hz.
Tony.
Attachments
Aren't you concerned that diffracted energy will be incorporated in the minimum phase assumption?
Attached is a comparison between the two sims and actual measurement.
Black is vituixcad
Blue is speaker workshop
Green is actual measurement at 1.8M taken directly after the individual midbass and tweeter measurements.
As can be seen there is something not right in the low pass section. I suspect some out of spec components. Both speaker workshop and vituixcad are out up to about 2Khz (and pretty much agree). vituixcad is very good above that, speaker workshop has a bit of an issue between about 4Khz and 5Khz.
Tony.
Black is vituixcad
Blue is speaker workshop
Green is actual measurement at 1.8M taken directly after the individual midbass and tweeter measurements.
As can be seen there is something not right in the low pass section. I suspect some out of spec components. Both speaker workshop and vituixcad are out up to about 2Khz (and pretty much agree). vituixcad is very good above that, speaker workshop has a bit of an issue between about 4Khz and 5Khz.
Tony.
Attachments
Last edited:
+10 for the 2-channel UMC204HD....incredible value, $59 B&H
I use the 4-channel 404, to make multi-mic measurements, $100.
I also use mine as my main "Hifi" music source with Volumio (on a Cubox-i, not a Raspberry Pi, as USB audio is fairly broken on the Pi) when not being used for measurements, using the fixed level RCA outputs on the rear, and it works and sounds great there too, handling all hi-res formats natively without resampling.
Great purchase, unbeatable value for money for what it does and I "reviewed" it in another thread a while ago. The only real limitation it has is that you can't use the front panel balanced TRS line inputs in unbalanced mode without introducing a small but measurable amount of 2nd harmonic distortion due to the action of the servo balancers.
The solution to that is either use a balanced input signal (for crossover measurements I use an audio transformer for isolation so that automatically converts it to balanced) or "abuse" the TRS insert inputs on the rear if I really do want to directly connect to a line level unbalanced signal, such as using an impedance measuring jig.
The insert inputs are natively unbalanced and also completely bypass the pre-amps used for the front inputs. Using these inputs no 2nd harmonic distortion is generated when using an earth referenced unbalanced line level signal. (A disadvantage though is you can't use the level controls on the front, as they are part of the pre-amps)
Great purchase, unbeatable value for money for what it does and I "reviewed" it in another thread a while ago. The only real limitation it has is that you can't use the front panel balanced TRS line inputs in unbalanced mode without introducing a small but measurable amount of 2nd harmonic distortion due to the action of the servo balancers.
The solution to that is either use a balanced input signal (for crossover measurements I use an audio transformer for isolation so that automatically converts it to balanced) or "abuse" the TRS insert inputs on the rear if I really do want to directly connect to a line level unbalanced signal, such as using an impedance measuring jig.
The insert inputs are natively unbalanced and also completely bypass the pre-amps used for the front inputs. Using these inputs no 2nd harmonic distortion is generated when using an earth referenced unbalanced line level signal. (A disadvantage though is you can't use the level controls on the front, as they are part of the pre-amps)
Last edited:
When you have a balanced system design and excellent cabinet design, there is not much work left for the crossover to do and there is no need to build a complex, expensive crossover. So, what you said is more a testimony of your design prowess than anything.
One can for example passively cross a tweeter at 2k, or use a FIR brick wall to cross the same at 1.5k and get a subjectively better overall sound. (Distortion will increase but dispersion will improve.)
On the other hand there has been an industry-wide trend to use DSP to cut cost. Cost can be many things, like space, electricity, development time, money. It is more costly to build a speaker with high power handling but it is cheaper to use DSP to implement a limiter so that the woofer is never driven to over-excursion - the consumer likes it because he/she wants a sound that feels loud as opposed to actually loud but overwhelming. Ditto for bass enhancement algorithms because consumers do not like big speakers.
So when I connect the dots together, I now see a bigger picture: begin with a good design, and the crossover designs itself, at a reasonable cost, with no advantage whatsoever going active (your design philosophy, I think). However some people like to push things beyond their limits, and then use DSP to stop them from self-destruction.
I do not have decades of experience like you do, and when I wrote that post it was my big day building for the first time my first passive design. I had zero confidence in passive at that point, and designated the stock miniDSP setting of LR 24dB/oct as the benchmark to beat. Subsequently I finished the tweeter section and I am happy to report that my passive defeated the stock active setting decisively. Imaging is one of the aspects I rank very highly because it reveals a system's flaws very well. Hindsight being perfect it is now no surprise that a careful passive will beat a careless active soundly.
That was an enormous educational experience. As a beginner I read about all the people how switching to active biamp improves imaging by leaps and you will hear details you have never heard before with the same drivers. I played with miniDSP and apparently that was true - the steeper slope always wins. The reasons are many: a high C2C leads to a complex lobing pattern whose bandwidth can be limited by steeper slopes. Cone break-up/tweeter resonance issues that can be limited by steeper slopes. The take-away from this is a balanced system design with proper cabinet implementation did not have these problems to begin with and do not benefit from more expensive passive crossovers or active crossovers.
When a passive crossover is capable of "disappearing", mission is accomplished and throwing more money/space/time at it won't improve it any further. This cannot be accomplished by considering the crossover in isolation. It has to start with good system design.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.