splicing nearfield to farfield, does this look right?

[dlr]

Dave, I had a bit of a play with the auto-detect. it seems it always chooses the centre of the peak as time zero. It seemed however that the phase response looked more normal when I manually adjusted to have the impulse signal start to rise from time zero... I guess it doesn't really matter provided both drivers being modelled for crossover have the same method applied...

Yes, for the purpose of design, the primary requirement is that the start time marker be the same between all measurements.

dlr

 

[speaker dave]

Dave, I had a bit of a play with the auto-detect. it seems it always chooses the centre of the peak as time zero. It seemed however that the phase response looked more normal when I manually adjusted to have the impulse signal start to rise from time zero... I guess it doesn't really matter provided both drivers being modelled for crossover have the same method applied...

Tony.

I thought there were options for this. Search through the menu.

For crossover networks you shouldn't auto detect though. You have two choices: Null out depth delay and define acoustic centers taking different driver depth into acount in the model, or (better) cancel out the same amount of airpath delay for both units and define distance to the baffle rather than the driver acoustic center.

To clarify, typically the woofer has more depth than the tweeter and its acoustic center is farther back. I find it best to find the best path time for the tweeter (nearest element) and then measure all drivers with exactly that amount of delay cancelation. Then the extra phase shift for the woofer will be seen and accounted for in crossover design.

Glad you found your preflection.

David S.

 

[wintermute]

Hi David,

I will do all my measurements again, today I tried using the recent ones in my crossover sim and they were all over the place, my previous measurements of the individual drivers worked very well with excellent correlation between the crossover sim and the actual implementation (maybe I should just stick with those ), I had zero locked those, but did not know about the second try on the auto detect for setting the initial time zero, I tried that last night and could clearly see it taking two goes to get it right! Also doing it on pure loopback measurements showed that the correct place is the centre of the highest peak too!

I'm a bit mystified as to where my "preflection" is coming from! It isn't there on the nearfield measurements.... I thought it must be electrical (how can there be a reflection from a negative distance) but if it was you would think it would show in the nearfields as well....

Now a bit of a digression:

The midbass'/tweeter should be time aligned. This particular combo when mounted as I have them is supposed to be (Terry who posted earlier has done a paper on Time Alignment and used this driver combo in this configuration for his tests).... The measurement attached was taken last year at 1.5M with Holm zero locked after autodetecting on the tweeter measurement (I was using causal impulse for the initial time zero detection, which I've now read may not be the best).

This would indicate a path difference of around 22mm between the tweeter and the woofers (if I'm doing my sums correctly) Note that acording to this the tweeter is 22mm behind the woofers... (more in a minute)

I've taken speed of sound as 343M/sec and the difference between the tweeter and the woofer to be .065ms

Now the measurement was taken on axis with the tweeter at 1.5M distance (for both the tweeter and the woofers, they are in an MTM config). So there will be a difference due to the woofers being slightly off axis (centre to centre is 120mm)

Using Pythagoras I get a difference of only 4.7mm so I'm still about 17mm out... hmmm.

So either my measurements are out, my maths are out, or my speakers aren't really time aligned!

1st attachment is zoom on the impulse on the positive peak, I used the bandpass option in the holm documentation to be able to compare the two impulses on equal footing. 2nd attachment is the same measurement, zoomed out a little on the impulse, but with the tweeter measurement inverted... I'm actually not sure which way it should go... if inverted is correct then it is actually in front of the midbass drivers which is more what you would expect.. however looking at the baffles (these are semi horn loaded tweeters) you would think that they are in fact a lot closer to time aligned than 22mm! (3rd pic)... Man this post has turned into a novel!

Tony.

 

[dlr]

I will do all my measurements again, today I tried using the recent ones in my crossover sim and they were all over the place, my previous measurements of the individual drivers worked very well with excellent correlation between the crossover sim and the actual implementation (maybe I should just stick with those ), I had zero locked those, but did not know about the second try on the auto detect for setting the initial time zero, I tried that last night and could clearly see it taking two goes to get it right! Also doing it on pure loopback measurements showed that the correct place is the centre of the highest peak too!

If your software can create a model that generates phase via HBT (easily available if yours does not), so that it is not necessary to use direct measurements with measured phase, there is a solution to the issue of the relative phase between drivers. You can use a three-measurement method that does not require measured phase at all. I have an article posted at my site that was published in SpeakerBuilder ONE:2000 that shows how to do this. It's straight-forward and IMO, practically fool-proof. The only issue is that if you change something that requires a new measurement of one or both drivers, such as baffle changes, care must be taken in creating the new model(s), primarily for lowpass or the full three measurements should be made again. But it makes the start time-marker issue moot. It is a benefit of which I was unaware when I wrote the article.

I'm a bit mystified as to where my "preflection" is coming from! It isn't there on the nearfield measurements.... I thought it must be electrical (how can there be a reflection from a negative distance) but if it was you would think it would show in the nearfields as well....

Keep in mind that HOLM is making a long time measurement and is capturing data prior to the first arrival from the drivers, so there will be data that is, I believe, simply background noise and any test system noise. Time zero as you see it is relative.

So either my measurements are out, my maths are out, or my speakers aren't really time aligned!

I may have missed it, is this to be a first order system?

Dave

 

[speaker dave]

1st attachment is zoom on the impulse on the positive peak, I used the bandpass option in the holm documentation to be able to compare the two impulses on equal footing. 2nd attachment is the same measurement, zoomed out a little on the impulse, but with the tweeter measurement inverted... I'm actually not sure which way it should go... if inverted is correct then it is actually in front of the midbass drivers which is more what you would expect.. however looking at the baffles (these are semi horn loaded tweeters) you would think that they are in fact a lot closer to time aligned than 22mm! (3rd pic)... Man this post has turned into a novel!

Tony.

I'm a little confused. Your tweeter has more depth than most but I still can't imagine it being acoustically behind the woofer. Think in terms of the voice coil planes. That is approx. where the acoustic center is.

I've forgotten what the bandpass part of Holm does but won't it modify/dominate the phase response? What do the unfiltered impulses look like?

As a general design principle your objective is not to allign impulses, rather to get phase curves to line up through the crossover region. That likely will not line up the impulses. The impulses tell more about the energy arrival in the center of the respective bands. Energy arrivals at the overlapping band edges are the key.

Finally, note that inverting the impulse shifts phase 180 degrees but has no impact on arrival time.

David S.

 

[wintermute]

Thanks Dave (dlr) I will have a read of your paper!

I may have missed it, is this to be a first order system?

Depends on what you mean by first order certainly at the moment the tweeter is crossed 1st order (electrical) ie it only has a cap and some series resistance. The woofer I'm not sure what you could say electrically as it only has a 4K notch filter on it... The acoustic slope is 2nd order bessel centred at about 3Khz.

As a general design principle your objective is not to allign impulses, rather to get phase curves to line up through the crossover region. That likely will not line up the impulses. The impulses tell more about the energy arrival in the center of the respective bands. Energy arrivals at the overlapping band edges are the key.

Thanks David yes I thought that having the phase aligned was the most important. guess I was going off on a tangent (again). I hadn't actually tried lining up the impulses before but thought it might shed some light on my measurements.

The non filtered is below: in the area where the phase slope is relatively constant (unfortunately not the crossover region) the difference in phase is around 60 degrees around the 3K crossover point it is out about 100 degrees. I adjusted the tweeter to detect largest peak and then changed the woofer offset to match...

The latest measurements I did have the woofer response a bit flatter after the 2K dip, probably due to different lining and increasing the flare in the driver cutout, hence the desire to re-measure. Also have decided I'm not happy with the current sound so time to have another crack at it

2nd graph is the measurement I did at the time with the crossover, I can't remember what the crossover was at that point but I suspect it had just the cap on the tweeter and a 1k and a 4k notch on the woofer... I later got rid of the 1K notch as it seemed to be causing audible distortion However I'm thinking I need to revisit the 1K notch.

edit: one thing that has always confused me was that if the drivers really were time aligned, that I would have thought the phase should be pretty close between the two on the raw measurements.

Tony.

 

[DBMandrake]

I'm a bit unhappy that my Audigy II ZS is not playing well with Windows 7, it never gave me any troubles. though the built in card gives a perfectly flat response in holm if I do a measurement with a loopback cable so it probably is environmental..

I've got my audigy looking like it will work reliably now, so think I will have to do all the measurements again using it, as I've never had weird stuff like this before with it. I think my reservations about the on board sound card may have been justified...

Tony.

I took the measurement at a lot higher spl level, the other thing was the sampling rate was only 44.1K instead of 96K and previous measurements were done on the audigy sound card.

Hi Tony,

I've been lurking on this forum for a while but your posts about the Audigy 2 ZS problems have given me a good excuse for a first post, as I also use one and think I can answer your questions regarding measurement problems.

The Audigy 2 ZS is a really great card for it's price and capable of extremely good measurements but there are a few tricks required to get it working properly for speaker measurement, and with the wrong driver or software settings the results can be terrible. Some of these you might have already discovered, so forgive me if I go over old ground.

The most important thing is that the card only has two native sample rates - 48Khz 16bit, or 96Khz 24 bit. It actually uses two different sets of ADC/DAC for the two sample rates as well, with an analogue audio switch to route inputs/outputs to one or the other. (More on that later)

If you use any other sample rate in your software, such as 44.1Khz, the card will do onboard resampling between the two rates, and it does a VERY poor job of this. It applies a low pass filter that starts rolling off above 10Khz, so you won't get a flat response to 20Khz, but worse still, despite the low pass filtering there will be a severe amount of aliasing above 10Khz - if you run a sine sweep and measure it in a real time analyser app you'll see (and even hear) the aliasing products clearly.

Depending on the sample rate you choose you can even see very obvious ripple in the treble response on the order of a dB or more.

If you set your app to 48Khz sample rate the resampling is avoided and the response will be dead flat to just above 20Khz with no aliasing artefacts.

The next issue is the 96Khz 24 bit mode - this is implemented as a completely separate ADC/DAC to the 48Khz mode, and it bypasses the EMU DSP processor, allowing for direct un-processed input/output only.

What this means is if you are using 96Khz 24 bit mode in your software and you apply any DSP effects, the 96Khz ADC/DAC will not be used, rather the signal will be resampled to 48Khz 16 bit and routed via the EMU DSP processor to the 48Khz 16 bit ADC/DAC, resulting in both early 10Khz rolloff and severe aliasing as described above. DSP effects include CMSS 3D, EAX audio effects, and even bass and treble controls.

If you set the bass or treble control even a smidgen above or below centre, 96Khz mode is effectively disabled and results will be horrific.

Also, in 96Khz mode only one app can open the audio device - if you have any other app open that has the audio input or output device open (such as iTunes open but not playing) it will also force the card back to 48Khz 16bit resampled mode once again.

The most sure-fire way of avoiding this problem (and apparently the only way on Windows Vista or 7 - I use XP) is to set another sound card - such as your onboard sound as the default sound device in windows, and manually select the audigy in the device settings of your speaker measurement app - thus ensuring that only that one app attempts to access the sound card.

Another limitation of 96Khz mode is that only the Line input can record at 96Khz, the Mic input can only do 48Khz, and if the software is in 96Khz mode you'll again get early roll off and aliasing attempting to use the Mic input.

It's worth it to use 96Khz mode though - I use 96Khz 24 bit almost exclusively with ARTA, as you can achieve an analogue noise floor of 108dB and a flat response out to just beyond 45Khz.

Other things you need to do is make sure the bass and treble controls are set exactly to neutral (50% in the Creative Volume control app) and all DSP effects like EAX, CMSS 3D, Karaoke, Bass enhancement are turned off.

Set Line in and Analog mix on the Creative Volume app to slightly below 0dB to avoid clipping. (the line in can only go up to -1dB without some extra distortion) Mute playback for CD/Aux/TAD/PC devices, otherwise their analogue noise will be recorded when recording line in via "Analogue mix", raising the noise floor.

In the normal windows sound control panel go into Options->Properties->Recording, Analogue mix, Advanced, and make sure Record without monitoring is ticked, otherwise some of the line input will be fed to the output during measurement upsetting results. Unfortunately this only disables pass-through during recording, there doesn't seem to be any way to disable pass-through all the time, so if you have a high gain loop you may get feedback when not taking a measurement. Beware!

Always use 48Khz 16 bit or 96Khz 16/24 bit in your application. (Either 16 or 24 bit is ok for 96Khz - 24bit mode will give about 12dB more S/N ratio)

Finally, I'd recommend always doing a quick analogue loop-back measurement before each measurement session by linking line in and line out jacks on the card and sweeping the FR with your measurement app to verify everything is in order - especially if using 96Khz mode. You'll instantly spot a configuration problem causing the card to drop back from 96Khz to 48Khz for example.

It's far too easy to have some other app change some of your sound card settings such as bass or treble and ruin your carefully configured setup. (Or someone else used the computer etc...)

Hope that's helpful

 

[DBMandrake]

Always use 48Khz 16 bit or 96Khz 16/24 bit in your application. (Either 16 or 24 bit is ok for 96Khz - 24bit mode will give about 12dB more S/N ratio)

Finally, I'd recommend always doing a quick analogue loop-back measurement before each measurement session by linking line in and line out jacks on the card and sweeping the FR with your measurement app to verify everything is in order - especially if using 96Khz mode. You'll instantly spot a configuration problem causing the card to drop back from 96Khz to 48Khz for example.

Here's a couple of FR sweeps that show just how wrong things can go...

The first screen shot is at 96Khz 24 bit - the red line is the correct response flat to nearly 45Khz obtained with ARTA as the only sound using application open, the yellow response is the same measurement taken with the same settings in ARTA simply with iTunes launched in the background, not even playing. The roll off beginning below 10Khz is obvious and shows the card has reverted to 48Khz resampling mode with an excessive low pass filter.

The second screen shot shows the difference between 48Khz at 16 bit, (the yellow line) and 48Khz at 24 bit - in theory this shouldn't cause a problem, but there is some weird kind of resampling going on here too causing ripple in the treble. Moral - don't accidentally use 24 bit together with 48Khz mode. (Easy to forget to change both together in ARTA)

The third screen shot shows the difference between 48Khz 16 bit (red) and 44.1Khz 16 bit. (yellow) Fluctuations are visible across the spectrum but especially above 1Khz. (I probably should have zoomed the vertical axis a bit more) What's not shown by a straight FR measurement is that there is a large amount of aliasing going on in the treble. Definitely to be avoided.

Note that in all cases there are also differences in the relative analogue signal levels in the different sample modes, I haven't introduced a level change for clarity of the graphs - so taking the same measurement in different sample rates/bit depths and expecting the levels to be identical is an exercise in futility - either stick with the same sample and bit rate all the time or check the level calibration when switching modes...

 

[wintermute]

VERY Helpful!! Thank you! Now I know why I had SO much trouble getting decent results (in XP I had set everything as per the RMAA how to, windows 7 is completely different)!

I haven't yet taken any actual speaker measurements since I thought I'd got the Audigy working, Those (loop back) that were looking ok were the 96/24 ones, but I still had unexplained bad measurements with those settings... now you have explained I can guess that what made them work was that at the time I had the default sound set to the onboard, more good luck than good management!

Also I will use 48/16 when doing my low freq measurements, I'd used 44.1 had no idea about the re sampling! Will also do the loop backs as suggested as they really do show when something is screwed up!

Have had the feedback problem got caught thinking that the record without monitoring should make it safe, nearly deafened me!


Welcome to the forum!! A very helpful first post


Tony.

 

[DBMandrake]

Glad to be of help.

I've used the Audigy 2 ZS to measure speakers since it first came out around 2003, even back then I had noted anomalies in the response with some sample rates and realised that 44.1Khz caused aliasing artefacts, so I settled on 48Khz 16 bit for all measurements.

At the time I couldn't seem to get 96Kbit 24bit to work, (using SpectraLAB at the time) results were always much worse than 48Khz, (as per the first screenshot in my previous post) so I had decided that either the 96Kbit mode was marketing lies, or was only for playback! (They added the 96Khz 24bit DAC to play back DVD-Audio discs after all, so I thought maybe there wasn't a 96Khz ADC on the board)

It was only very recently when trying out ARTA that I had a Eureka moment and stumbled onto getting 96Kbit 24 bit working properly, so I spent a few hours exhaustively trying all combinations of sample rate, bit rate, mixer settings and so on, with analogue loopback measurements to measure FR, noise floor, distortion, aliasing etc, so I could once and for all find out what was going on and be able to achieve consistent measurement results.

In hindsight I think my failure to get 96Khz to work in the past was due both to using the microphone input for most measurements (as the mic input can only work at 48Khz) and also not realising that even a slight error in the bass/treble controls, or a background app with the audio device open would disable the 96Khz ADC/DAC.

I now connect the mic (an ECM8000) to the line input via a balanced to unbalanced stepup transformer so I can use the 96Khz mode (also so I can use Dual Channel mode - seeing as the Mic input is Mono) - the gain is about 20dB down on the Mic input despite the step up ratio of the transformer, but for most measurements it's still more than adequate and the noise floor is a lot lower on the Line input than the Mic input anyway, so your effective SNR is about the same.

I'm chuffed now that I've got accurate repeatable 96Khz 24bit measurements at the sound card level - that's one less thing to worry about. (My measurement environment is a disaster though, and I think only moving house will help there...)

 

[speaker dave]

The non filtered is below: in the area where the phase slope is relatively constant (unfortunately not the crossover region) the difference in phase is around 60 degrees around the 3K crossover point it is out about 100 degrees. I adjusted the tweeter to detect largest peak and then changed the woofer offset to match...

2nd graph is the measurement I did at the time with the crossover, I can't remember what the crossover was at that point but I suspect it had just the cap on the tweeter and a 1k and a 4k notch on the woofer... I later got rid of the 1K notch as it seemed to be causing audible distortion However I'm thinking I need to revisit the 1K notch.

edit: one thing that has always confused me was that if the drivers really were time aligned, that I would have thought the phase should be pretty close between the two on the raw measurements.

Tony.

I think your units are fairly well time alligned in that they track with a constant phase shift of about 60 degrees apart. Slope of the phase curves are the same and that means that delay time is the same. The constant phase shift between the two is not indicative of delay. This does not mean that you will have an easy time of getting linear phase blending.

You had asked about the addition of time alligned units (perhaps as judged by the impulses). The way to think of it is that a multiway system is always the sum of bandpass elements or elements with roll offs at both ends. Frequently you will see articles where people add theoretical sections or electrical filters with no relative delay and infer what ideal crossovers are. Usually these simulations have little practical relevance because most real systems have interunit delay, but in your case the interunit delays are minimal and you will be close to the idealized filter summations.

With bandpasses with no excess delay you will always have phase near 0 in the middle of a bandpass, with rising (+) phase for low frequencies below the corner and falling (minus) phase at and above the high frequency corner. The phase curve of the woofer and the phase curve of the tweeter may look similar (but shifted in frequency) meaning they might match at the middle of the bands but they will deviate from each other at the crossover frequency. (Woofer phase will be drooping down while the tweeter phase will be rising up away from their band centers). What you must do to get the system to blend is to have enough droop on one and upswing on the other for them to come around into phase again. For this you play with two variables: corner rolloff slope and driver polarity. For the tweeter adding a crossover or increasing the order of a crossover will swing the phase up. Likewise, adding a crossover or increasing the crossover order of the woofer will swing its phase down. The right network orders will get the phases to overlap. Alternatively you can get them to track 180 degrees apart and flip the polarity of either unit to achieve phase overlap.

That your units are about 60 degrees apart and that the mid phase is above means that a minimal crossover slope can get then to pull together. If you need more than first order for power handling reasons then you probably need to go higher, say 2nd and 2nd and flip the tweeter polarity.

The arcane art of crossover design.

David S.

 

[wintermute]

Glad to be of help.

I now connect the mic (an ECM8000) to the line input via a balanced to unbalanced stepup transformer so I can use the 96Khz mode (also so I can use Dual Channel mode - seeing as the Mic input is Mono) - the gain is about 20dB down on the Mic input despite the step up ratio of the transformer, but for most measurements it's still more than adequate and the noise floor is a lot lower on the Line input than the Mic input anyway, so your effective SNR is about the same.

I initially used the mic input with a cheap computer mic with Speaker Workshop just to see what I could get... even that proved useful as I discovered I had my mids wired with the incorrect polarity. I then decided to build the Eric Walin mic preamp which has been excellent I'd recommend it! and I think Vikash still has some PCB's. I'm using a linkwitz modded panasonic WM60AY mic mounted in the end of an approx 1M long copper tube which is the same diameter as the mic capsule.

I'm chuffed now that I've got accurate repeatable 96Khz 24bit measurements at the sound card level - that's one less thing to worry about. (My measurement environment is a disaster though, and I think only moving house will help there...)

You never know, I tried moving around, things, changing the position of the mic and speakers till I found something that worked reasonably well (not being parallel to any walls seems to help) I just tried to find the position that gave me the longest time before the first reflection, if you have a low ceiling or very small room though then it could be a losing battle....

Recently moving outside onto the balcony seems to be even better.... the best part with that is that I can angle so that reflections off nearby walls get bounced out at an angle to open space so I only have to worry about floor and ceiling reflections... I've been thinking lately that if I made a boom I could place the speaker out near the edge with nothing above nothing to the sides, and about 3M to the ground below (via some bushes which should break up reflections anyway)... That may make for some decent measurements down to quite low frequencies !

PS. I didn't see the second post till just now. the effect of having iTunes open is appalling!!

Tony.

 

[wintermute]

David, Thank you for an excellent explanation!! I have struggled with understanding phase since I started, that helps more than anything I've read in any books or anywhere online!

The arcane art of crossover design.

The crossover was always the part that I feared the most! I've got pretty good results (measurement wise) so far considering it is my first attempt at crossover design, but in the end something just doesn't sound quite right... Only on some music though, most notably (and annoyingly) on Pink Floyd! I reckon the problem is the hump between 1K and 2K but when I flattened that out before I thought the speakers sounded distorted... I will persist (I'm nothing if not tenacious).. I also have to consider it might be the room.. the response at the listening position is pretty awful (though I lost those measurements the other night when holm crashed, the upside was that It is now actually calibrated and it won't be difficult to re-measue) I guess at some point I should take them somewhere outside and have a listen!

Tony.

 

[wintermute]

I have an article posted at my site that was published in SpeakerBuilder ONE:2000 that shows how to do this. It's straight-forward and IMO, practically fool-proof. The only issue is that if you change something that requires a new measurement of one or both drivers, such as baffle changes, care must be taken in creating the new model(s), primarily for lowpass or the full three measurements should be made again. But it makes the start time-marker issue moot. It is a benefit of which I was unaware when I wrote the article.

I finally got the time to sit down and read your article Dave, I must say ingenious in it's simplicity I had gone part of the way in that I did my measurements for midbass' and tweeter at the same position for use in my crossover model, but the idea of measuring both drivers playing together with no crossover components and comparing the summing of the raw with the actual measured combined response is very clever.. no other components to potentially be introducing errors!

I figured that by making my measurements at the same position I didn't have to put any offset info into speaker workshop (or Jeff Bagbies PCD) because it was already built into the measurements). I will have to try this now!! I'm pretty certain speaker workshop allows you to put in the driver offsets... and I might be able to remove the last remaining inaccuracies in my model vs actual result!

I also had a bit of a browse of your site, will be returning for more! The section on using felt has made me decide I have to put that order in for felt before I do any more serious measurements.. Time to redo the grills (the plan is to integrate the felt with the grills) and do the measurements WITH the grills in place rather than without.

Tony.

 

[dlr]

I finally got the time to sit down and read your article Dave, I must say ingenious in it's simplicity I had gone part of the way in that I did my measurements for midbass' and tweeter at the same position for use in my crossover model, but the idea of measuring both drivers playing together with no crossover components and comparing the summing of the raw with the actual measured combined response is very clever.. no other components to potentially be introducing errors!

If you use software such as the PCD, there's another benefit to this. Once you've created the models with HBT minimum-phase and adjusted the z-offset as described, you can then easily find the precise angle of the lobe for any crossover you create. The key here is that the phase for each let will be for the driver/crossover combination. The rolloff in the driver passband will have some impact as will any non-ideal matching to the target curve.

Essentially, you try to match a target for each driver individually, then you can "rotate" the system up or down in the PCD. This will let you see where the summed response is at a peak. If this angle is low, you probably will be fine just trying to smooth the on-axis response, for example by "relaxing" the woofer crossover to get a better phase response in the crossover area for better summed response. If it's significantly far from the on-axis response, smoothing on-axis will introduce a power response anomaly in and around the crossover area.

I figured that by making my measurements at the same position I didn't have to put any offset info into speaker workshop (or Jeff Bagbies PCD) because it was already built into the measurements). I will have to try this now!! I'm pretty certain speaker workshop allows you to put in the driver offsets... and I might be able to remove the last remaining inaccuracies in my model vs actual result!

If the measurements are accurate, then this is correct, no offset is needed. However, if the measurement is off a bit, say due to start time-marker issues, then it will help. Keep in mind that if you generate phase, you should tail each driver. These tails will affect the calculation of the HBT phase. In the end, once you have a model with tails for a driver, don't change them as this will alter the calculated phase response.

I also had a bit of a browse of your site, will be returning for more! The section on using felt has made me decide I have to put that order in for felt before I do any more serious measurements.. Time to redo the grills (the plan is to integrate the felt with the grills) and do the measurements WITH the grills in place rather than without.

I don't make anything without felt. Even my current dipole has some in empirically determined strategic spots. Using it with a grill frame is an excellent idea, you'll be surprised at the improvement.

One key reason I like the use of felt is the improvement of the polar response.

dlr

 

[speaker dave]

If the measurements are accurate, then this is correct, no offset is needed. However, if the measurement is off a bit, say due to start time-marker issues, then it will help. Keep in mind that if you generate phase, you should tail each driver. These tails will affect the calculation of the HBT phase. In the end, once you have a model with tails for a driver, don't change them as this will alter the calculated phase response.

dlr

I know that people have success with both approaches but I have always preferred to measure phase rather than Hilbert Transform and worry about tail correction and such. I make fewer mistakes with that approach. With packages like Holm you can have proper phase measurements easily. Fix your airpath delay and don't worry about it.

David S.

 

[dlr]

I know that people have success with both approaches but I have always preferred to measure phase rather than Hilbert Transform and worry about tail correction and such. I make fewer mistakes with that approach. With packages like Holm you can have proper phase measurements easily. Fix your airpath delay and don't worry about it.

David S.

I generally use direct measurements as well. There are times, however, when HBT phase-generated measurements are useful and can be, in fact, a necessity. If one wants to examine (or optimize for) locations other than the point at which the measurements were taken, the relative path length changes due to position require that minimum-phase (HBT) phase be used. This is, for example, the case in CALSOD and other software that can display multiple points simultaneously. It's also needed in software such as the PCD that allows one to examine locations by rotating the mic position using spinners. Direct measurements will not work in these instances.

Of course these do not account for diffraction changes (some do provide driver diameter entry), but it does allow one to investigate the impact of the crossover with driver separation.

Dave

 

[speaker dave]

I'm not familiar with Calsod but I used to do a lot of XOpt and then transitioned over to LEAP crossover shop. I don't know why either approach would require Hilbert transform derived phase. Obviously if your Hilbert transform is accurate (and your driver is minimum phase, generally the case) then the phase curves will be the same.

One slight distinction is that if you fix the air path delay (say for the shallower tweeter) then you will measure phase as if the source were on the center point of the woofer but on the plane of the baffle. The Hilbert transform gives you phase relative to the acoustic center depth which you must derive by inspection or via your itterative procedure.

Even with multiple listening axis calculations the geometries will be so close as to give the same curves.

David S.

 

[dlr]

I'm not familiar with Calsod but I used to do a lot of XOpt and then transitioned over to LEAP crossover shop. I don't know why either approach would require Hilbert transform derived phase. Obviously if your Hilbert transform is accurate (and your driver is minimum phase, generally the case) then the phase curves will be the same.

They will be the same with the exception that one will include the excess-phase between the acoustic center and the mic. That excess-phase will be different between drivers in most cases where measurements are based on the tweeter axis.

One slight distinction is that if you fix the air path delay (say for the shallower tweeter) then you will measure phase as if the source were on the center point of the woofer but on the plane of the baffle. The Hilbert transform gives you phase relative to the acoustic center depth which you must derive by inspection or via your itterative procedure.

Yes, this is the issue. The only points in space for which the non-HBT setup provides accurate results will be on an arc with radius equal to the mic distance and the plane it defines orthogonal to the line that passes through the acoustic center of both drivers. In essence, for a typical 2-way with top-mounted tweeter that arc will have a downward tilt. This arc is the only area where the excess-phase of both drivers will remain constant.

Even with multiple listening axis calculations the geometries will be so close as to give the same curves.

I've not made any detailed study of the variability, but much will depend on the driver depth(s), the vertical spacing and the crossover Fc as well as the design point. Two-ways with large midwoofers and subsequently larger separation leave more room for error. The difference may be insignificant. Or not.

I suggest doing this because 1) it's really not that difficult and 2)I suspect it's likely that with the option in software such as the PCD it will be very tempting to check out various locations. Why not remove as much possibility for error as one can?

Dave

 

[wintermute]

thanks guys, I was going to post the sim I did this morning with the old measurements... the phase was matching exactly between tweeter and midbass' from about 1000 to 3000 hz was pretty stoked about that.. but speakerworkshop is dying when trying to open the file.. typical!!

One problem I have with PCD is that it doesn't support adding the series resistance for coils in the "additional" crossover components part. Since the bass section of my crossover is mostly (this latest sim did have an added series coil) affected by Notch filters, the PCD doesn't give anywhere near accurate results due to it assuming that the inductors are perfect Speaker workshop does allow me to put in the series resistance of the inductors and it gives a much better correlation... PCD allows me to find a starting point for the notches, but it needs to be tweaked in SW to get something that approaches reality...

I think I missed something earlier, will have to read up on Hilbert Transforms.

One more small question. Is it sensible to move the crossover frequency either apart or overlapping somewhat if it results in the phase matching better? It is something I was playing with in PCD for the active crossover that the nearfield measurements were required for.... ie rather than having the slopes for each driver being at say -6db at 300Hz have them at say 280Hz and 320Hz (either appart or overlapping depending on which way improves the phase match)... Obviously the question stands for either active or passive

Tony.