Phase problems combining LP and HP xo in Speaker Workshop

[soongsc]

Okay. I played around with SW. Looking at the tweeter, the phase would probably look something like the attached picture. Amazing it doesn't use the original MLS data, probably using a different algorithym to take out the delay.

The network capability is probably better suited for generating response with the XO included instead of using COMBINE. since the combine was probably designed for combining near field and far field data, I do not thing the phase information is combined. But I havn't trie it though.

The phase of the bass looks mixed up. If you can get a different in distance between the speaker to mic versus speaker first reflection to mic of about 1.2M, then the you should be able to get better phase measurements for the woofer.

 

[tcpip]

Since your mic was stationary, any phase or time delay change you make to the tweeter measurement should be duplicated on the woofer measurement. The mic is essentially your ear, and you will be listening from a fixed position in relation to all the drivers.

Okay, got it. Will remember this.

Well, you could just increase the time between your start and stop markers. Some of the reflections will get into the measurment, and the graph may not look as pretty, but it will give more resolution of the data. There are compromises that have to be made when you don't have an ideal measurment chamber, and you are encountering them now....sorry.

I'll see what happens when I try this. However, I'm already getting enough data points for getting a fairly detailed curve at the xo region, I thought. Don't a lot of DIYers work with 1metre measurements with floor bounce? They'll all get my kind of data points, I'd think.

As Zaph wrote, Roman Bednarek can probably help more at Madisound forum or here: http://www.rjbaudio.com/

Thanks for this pointer. I'll check it out.

Okay. I played around with SW. Looking at the tweeter, the phase would probably look something like the attached picture.

Thanks for taking the trouble.

One question: how do you determine how much phase delay to add which gives you the "right" result? Is it purely visually?

Secondly, does this correcting of phase delay affect xo design? I was under the impression that it doesn't.

The network capability is probably better suited for generating response with the XO included instead of using COMBINE. since the combine was probably designed for combining near field and far field data, I do not thing the phase information is combined. But I havn't trie it though.

As I'd mentioned I was getting exactly identical results by using the COMBINE operation and the physical aggregation of xo components. Also, the near and farfield combination is done using SPLICE, IIRC, not using COMBINE, in SW.

The phase of the bass looks mixed up. If you can get a different in distance between the speaker to mic versus speaker first reflection to mic of about 1.2M, then the you should be able to get better phase measurements for the woofer.

So basically, this means I should move the mic back further? Let me do another set of measurements (tonight) and see.

Thanks for the trouble, all of you guys.

 

[lndm]

No, I'm doing LR4, so the phase difference should be 360 degrees. See the goal curves on my xo graph. Both the goal curves will show 24dB/oct slopes.

The drivers are not coherent to begin with. You gets what you get. (for example you may be dealing with 270 or 450 degrees versus 360). The same in the end.

 

[soongsc]

...

So basically, this means I should move the mic back further? Let me do another set of measurements (tonight) and see.

Thanks for the trouble, all of you guys.

Move the speaker and mic away from wall or any reflecting surfaces. (speaker to closest reflecting point to mic total distance)-(speaker to mic distance) > 1.2M Whereever you put the mic, it should be the same location for both tweeter and woofer measurement.

The only way to determine the where the combing effects from driver are is using analysis and simulation. This will tell you at what frequencies it will occuc, but not necessarily the correct magnitude. Still usefull however.

 

[soongsc]

Thanks for the inputs, John.


...

One question: how do you determine how much phase delay to add which gives you the "right" result? Is it purely visually?

Secondly, does this correcting of phase delay affect xo design? I was under the impression that it doesn't.

...

1. The proper way to do it is to do a hilbert bode transform of the amplitude and compare with the measured. The results will not be exactly the same, but the closest is the right one. Since SW does not have hilbert bode transform capability, you need to kind of guess a little. But the "remove access delay" function seems to come pretty close if you do it two or three time (have to close it between each time).

2. If SW implements everthing correctly, the amplitude and phase will be used to determine the final amplitude. It's like if you have two sine waves equal aimplitude but 90 degrees phase difference between them, then the summed wave form will be different from that if there were no phase difference.

 

[tcpip]

The drivers are not coherent to begin with. You gets what you get. (for example you may be dealing with 270 or 450 degrees versus 360). The same in the end.

Sorry I'm a late entrant to this club, so can you please explain what makes drivers coherent or incoherent? My reading is limited to v5 of Vance Dickason and a few thousand posts on these forums; I can't seem to recall any discussion about looking for coherency among drivers to get them to behave themselves in an xo.

 

[tcpip]

Move the speaker and mic away from wall or any reflecting surfaces. (speaker to closest reflecting point to mic total distance)-(speaker to mic distance) > 1.2M Whereever you put the mic, it should be the same location for both tweeter and woofer measurement.

Done that. Another friend too said I should take a second set of readings at 2m distance, just to see whether things improve. His theory was that if my observations were due to combing from the MTM multi-driver radiations, then maybe moving the mic position further away would get reduced combing.

Well, the datafile is here with the new data. There is a folder in it called "2metres" which contains the SPL curves taken at 2 metres distance for just the drivers of the left channel. I tried making xo's out of them, and combining the xos' outputs, and I found exactly the same sort of jagged SPL curves near the xo point there. My datafile this time contains the xo networks I tried using. In the "2metres" folder, there will be one network called "L-hp-2m" and another called "L-lp-2m". These are the xo networks.

If you are curious, you'll also see the earlier 1metre readings and their xo's in the "left" folder (for the left channel). Here, the SPL curves are objects whose names end with "-v3" (they were the third round of readings). You'll also see the xo networks I designed around them.

 

[tcpip]

2. If SW implements everthing correctly, the amplitude and phase will be used to determine the final amplitude. It's like if you have two sine waves equal aimplitude but 90 degrees phase difference between them, then the summed wave form will be different from that if there were no phase difference.

My question was a little bit different.

I know that phase and amplitude both must be taken into account when one does an xo. However, I was under the impression that the optimisation function of xo design software usually optimises both the phase and the amplitude, by tuning the L and C components, so that the phase of the high-pass and the low-pass are correct at the xo point, and there is no phase mismatch.

Therefore, I was under the impression that SW would do the job of phase coherency for me, by tweaking the values of the xo components. I was under the impression that I would not need to play with the phase delay or other things myself.

Am I wrong? Does one routinely need to tweak the phase after taking acoustic readings, before one builds the xo?

 

[lndm]

can you please explain what makes drivers coherent or incoherent?

Ok, there are two areas of concern: the impedance of the drivers and the acoustic output. Each of these two has an amount (magnitude) and phase. If these four attributes on each your woofer and tweeter were flat, then a passive crossover would work perfectly every time. 'Tain't so.

Your goal is probably to produce an acoustic rolloff for each of your drivers that resembles a theoretical crossover. Someone mentioned superimposing a target slope and trying to match it. You also want to have the acoustic phase plot of the two drivers close to each other (within a small range) over an octave or two around the crossover.

You might produce the perfect acoustic (magnitude/SPL) for the two drivers and finf the phase of the two is not close to each other across the frequencies of interest. This will give nulls etc, and generally less than perfect results. If you have done it well, there should be a null when you reverse the tweeter.

Achieving SPL and phase targets both at once can be a challenge. You may feel that one particular crossover frequency is just impractical for your two drivers. The 'un-flatness' of the impedance will hinder your efforts and may produce unexpected results, including apparently illogical peaks. Persistence pays off.

You also want to avoid a 'kinky' combined system phase plot, but one thing at a time.

In the end, your ideal crossover components may look nothing like the text book values (or topology). A novice may look at them and think they are poor values, but they are not.

 

[soongsc]

My question was a little bit different.

I know that phase and amplitude both must be taken into account when one does an xo. However, I was under the impression that the optimisation function of xo design software usually optimises both the phase and the amplitude, by tuning the L and C components, so that the phase of the high-pass and the low-pass are correct at the xo point, and there is no phase mismatch.

Therefore, I was under the impression that SW would do the job of phase coherency for me, by tweaking the values of the xo components. I was under the impression that I would not need to play with the phase delay or other things myself.

Am I wrong? Does one routinely need to tweak the phase after taking acoustic readings, before one builds the xo?

I've played around with SW somewhat, and I can say for sure that the "network" capabilities are not really that accurate, but it's a good ball park estimate. I'm not sure how SW uses phase data in the "network". But it does use the FF or On-axis data for the simulation is what I recall. I've always been in the habit of trying to get the right data every step of the way.

If you have your markers set correctly so that the measured phase is right, then you do not have to do the "remove access delay" if the distance between the mic and driver acoustic center never changes. Otherwise, yes you need to tweek it with each measurement. At a sample rate of 48K, a difference of 4mm would probably result in phase shift.

Right now I'm using Sound Easy, which actually tells you how far the acoustic center is from the mic, which is very convenient.

 

[lndm]

BTW tcpip, I appreciate you have experience here and if what I have said, you already know, feel free to ignore it as required .

It's that I have found optimisers to only give me what I want under better conditions, and I have chosen to forego them and do it by hand. YMMV

 

[tcpip]

BTW tcpip, I appreciate you have experience here and if what I have said, you already know, feel free to ignore it as required .

It is as you have suspected --- I already knew most of what you've written in that earlier post. However, the part that I fail to understand is the failure of the optimiser to even out some of these problems. When I tell the optimiser that I want a target curve, that curve has both a magnitude plot and a phase plot. Both need to match, or else it's simply not matching the target. For instance, when one specifies an LR4 lowpass at a certain Fc, it means that the phase too is specified, not just the magnitude. And the optimiser tweaks the values of the L and C components to get things right for both phase and magnitude. Why is the optimiser doing such a good job with the magnitude in my case, and failing with the phase?

Incidentally, I've tried alternate Fc. I get the same jagged curve each time on combining. I can't believe it's some little problem at just 2KHz... it must be something fundamentally wrong I'm doing.

It's that I have found optimisers to only give me what I want under better conditions, and I have chosen to forego them and do it by hand. YMMV

I'd love to hand-tweak, if I only knew what to tweak. I've hand-tweaked some of the things. For instance, I've added a series resistor with the tweeter, and this gives me a handle to control the slope of the tweeter's SPL in the region above the Fc. Similarly, if I could control the phase in the octave above and below the Fc, I'd tweak it too. But where to start? Just perturbing all the component values at random does not seem like a good enough idea... that'd just duplicate the job of the optimiser, without its speed.

 

[lndm]

I can't say I'm sure, but I feel fairly certain that you cant just dial up a slope and a given phase plot together. There is a limit to the ways you can achieve a target slope, and to a point, it seems you have to accept the phase as it falls.

This is why I mentioned possibly finding one particular frequency unsuitable to cross at. True or not, it seems that way sometimes.

Just perturbing all the component values at random does not seem like a good enough idea...

Au contraire It works for me. As for as your cause and effect observations about the tweeter resistor. I was very surprised at some of the effects of various components.

that'd just duplicate the job of the optimiser, without its speed.

Except that you are watching both the slope and the phase. Actually, in my experience the rolloff slope of either driver is less important than the combined response, and the individual phase plots.

BTW, there is the delay ladder, which can introduce a constant delay, FWIW. It is here under delay lattice calculator http://www.hal-pc.org/~bwhitejr/

 

[jbateman]

It is as you have suspected --- I already knew most of what you've written in that earlier post. However, the part that I fail to understand is the failure of the optimiser to even out some of these problems. When I tell the optimiser that I want a target curve, that curve has both a magnitude plot and a phase plot. Both need to match, or else it's simply not matching the target. For instance, when one specifies an LR4 lowpass at a certain Fc, it means that the phase too is specified, not just the magnitude. And the optimiser tweaks the values of the L and C components to get things right for both phase and magnitude. Why is the optimiser doing such a good job with the magnitude in my case, and failing with the phase?

Incidentally, I've tried alternate Fc. I get the same jagged curve each time on combining. I can't believe it's some little problem at just 2KHz... it must be something fundamentally wrong I'm doing.


I'd love to hand-tweak, if I only knew what to tweak. I've hand-tweaked some of the things. For instance, I've added a series resistor with the tweeter, and this gives me a handle to control the slope of the tweeter's SPL in the region above the Fc. Similarly, if I could control the phase in the octave above and below the Fc, I'd tweak it too. But where to start? Just perturbing all the component values at random does not seem like a good enough idea... that'd just duplicate the job of the optimiser, without its speed.

When you tell SW to optimize a filter so that it will mimic your LR4 GOAL, all it does is adjust the component values to try to acheive the same AMPLITUDE response as your goal. It doesn't create any particular phase value.

I think you are misunderstanding SW abilities somewhat. When you use the optimizer to get HP and LP values, this is just a starting point. And SW is only good at optimizing for INDIVIDUAL DRIVERS...it will not optimize for a COMBINED woofer and tweeter circuit...it's just not that powerful.

So your original component values are just a place to begin. You will have to examine the amplitude and phase, and experiment with different crossover frequencies, maybe even completely different circuit topologies until you find some combination that gives acceptable results.

This is why it was suggested to contact Roman Bednarek, as he has a lot of experience with various speaker designs, and may be able to recommend some circuits that will get you close to your goals.

 

[tcpip]

When you tell SW to optimize a filter so that it will mimic your LR4 GOAL, all it does is adjust the component values to try to acheive the same AMPLITUDE response as your goal. It doesn't create any particular phase value.

This is news to me -- thanks. I was under the impression that anything which fine-tunes the amplitude will also affect the phase, but then I guess the implicit effect on the phase may not be enough to actually align it with the goal's phase.

So your original component values are just a place to begin. You will have to examine the amplitude and phase, and experiment with different crossover frequencies, maybe even completely different circuit topologies until you find some combination that gives acceptable results.

I have tried LR4 at various Fc. I'll now try LR2, and may even try asymetrical slopes... let's see.

This is why it was suggested to contact Roman Bednarek ...

I've written to him. He hasn't yet replied. I'm keeping fingers crossed.

 

[tcpip]

This is why I mentioned possibly finding one particular frequency unsuitable to cross at. True or not, it seems that way sometimes.

I'm going to spend tonight trying out various Fc, then different slopes, then maybe some asymmetrical slopes. Wish me luck.

 

[lndm]

Using a first order low pass and a second order high pass is not an uncommon end result in my experience. Looking at commercial designs, I used to think it was just a way of saving money on the second capacitor , then I started finding it worked for my designs.

You might try a '1.5' order low pass (capacitor in series with a resistance).

Good luck!

 

[soongsc]

This is news to me -- thanks. I was under the impression that anything which fine-tunes the amplitude will also affect the phase, but then I guess the implicit effect on the phase may not be enough to actually align it with the goal's phase.

....

All simulation and optimization is only as good as the models can be. If everything were as simple as clicking the mouse, it takes lots of fun out of the process.

 

[tcpip]

Problem Solved!!!!

As long as his hardware does not have a shifting latency problem, which might make the starting point for the 2 measurements occur at different times, it should be OK.
It would be helpful to have ALL of the data, however, if others are going to be evaluating it..

I'm glad to report that my problem has been solved, partly due to this hint in your earlier post, which got me thinking last night.

I am now able to get neat summing of the LP and HP curves, without a single dip or peak anywhere which was not there in the original LP/HP curves. So, I don't see any big phase misalignment problem any more.

The problem was my sound card. Its latency jumps randomly. I can take five different readings and they'll have five different delays between 0msec and the start of the MLS spike. I have always known this, but I've never known that this would cause phase misalignment between the drivers. (You have to excuse my ignorance... this is my first design project.)

With my card, it's not even feasible to "get the delay right" using the "Pulse Response" operation and then proceed to the FR measurement. You simply have no way of knowing whether the latency during the FR measurement was the same as the latency during the Pulse Response reading. So I gave up doing FR measurements altogether. I just did Pulse Measurement and applied FFT to it to get my SPL curve.

To get the tweeter and woofer to phase-align, I took pulse readings of tweeter and then did multiple pulse readings of the woofers till the MLS spikes in both aligned perfectly, by visual comparison. Then I applied FFTs on both, plugged the SPL curves into the xo's, summed the xo's, and got a well-behaved sum.

Luckily for me, the latency jumps forward and back in largish, discrete steps. It never seems to jump to completely random points... it always jumps back and forth to specific points, which are almost 1msec distant from each other. So the spike will either be at 2msec or at 3msec, but not at 3.1msec. Therefore, a visual alignment of two pulse measurements seems to work very easily. The two spikes either align very well, or they very obviously don't align.

So, this problem appears to be licked.

The last niggling doubt or unease I have is that the reverse-polarity summing is not giving me a deep notch... the notch I'm getting is barely 4-5dB deep. However, this is far less serious than the situation that I was in earlier.

Any comments about why my reverse notch is so shallow? My guess is that the phase data of my woofers is "smudged" because the two woofers are not in exact phase alignment with each other. Don't ask my why this is so, but this is the best guess I can hazard. Maybe it has something to do with the fact that the two woofers are in series.

Thanks for the inputs and patience, all of you. I'd be nowhere in speaker design without all of your help. Even my friend who guides me in every step of speaker design is someone I "met on the forums."

 

[lndm]

One possibility is that using discreet measurements will prevent you seeing the deep notch. It may be there though. Look at the phase plots and if there is 180 degrees difference at the null point, there really will be a deep notch even if it doesn't show.

I remember reading earlier of someone saying you should remove the excess delay. This will make the phase plot more straight instead of wrapping around, but it is still the same. You must remove the same time (milliseconds) from each the tweeter and woofer or else your plots mean nought.

The main issue is relative phase. Do to one driver what you do to the other. So, you can measure with a poor microphone, use a poor sound card, delay both channels etc, and you will still be able to see whether the two drivers are in phase with each other at any given frequency.

The microphone should be in the same spot for each measurement. I like to do each driver in quick succession to make sure I don't bump anything.

I like to use pink noise to do my phase measurements with the mike at 2'. Some may disagree. I can watch the FFT spectra for anomalies and re-start the test if desired. I can set the averaging up and let the curve settle. This can give a more robust noise input than a MLS signal.

I think that room contributions are less critical when measuring relative phase than, say, SPL measurements. YMMV.