PASSIVE TO ACTIVE

[Rolledmyown]

Hi folks, I'm a bit out of my depth here with some recent REW experiments - and audio in general!
Long story short, I've have a pair of kit speakers comprising Peerless 830668 woofers, Scanspeak 12MU/8731T00 mids and Classic D2905/930000 tweeters in a ported 70 L box. I made my own massive aircored chokes and crossovers back in the 90's and decided to modernise my outfit using a pair of Hypex FA123 plate amps to make active speakers.
I have spent quite a number of years testing my equipment with REW in order to optimize my listening position and sub integration but just by using SPL/phase plots. Now I have the opportunity to branch out and do some things that were not possible with my passive cross overs, specifically time align the drivers to optimize the speakers, then I was going to implement room EQ using MediaCenter as I've been doing for a while now.
Anyhow, my first attempt was to estimate the physical offsets of my 3 drivers and calculate delays to suit those measurements. I then tweaked those delay numbers and used SPL and phase plots by eye to 'find the best delay'. The optimized delays arrived at were 200 uS for the mid and 501 uS for the mid, which correspond to offsets of 69 mm and 172 mm, Those numbers sounded quite rational. Then I read some articles describing more sophisticated approaches - in particular the miniDSP article https://www.minidsp.com/applications/rew/measuring-time-delay. I chose to use my recording mike at 1 m from the midpoint between the mid and tweeter (one of the recommended setups) and then put a delay of 1 mS between the mid and tweeter in order to well-separate the impulses from the mid and the tweeter (then of course the 1 mS increment is subtracted from the measured impulse delays). From that approach I (eventually!!!) arrived at delays of 853 and 896 uS, which seemed way too big to me for the dimensions of my enclosure and the official drawings of the drivers! So today I did some experiments from my listening position with just one speaker and my sub turned off, using the delays of 853 and 896 together with several others of shorter delay but preserving the delay difference of about 43 uS between the mid and tweeter.
I checked the SPL and phase for each to see what was happening at the mid/woofer crossover (4th order LR at 250 Hz) and although there was some difference there was nothing startling showing even at delays as low as 100 and 143 uS. I then had a read of an excellent post on a Klipsch site https://community.klipsch.com/topic/182892-using-rew-to-determine-time-delays-between-drivers/. That made use of spectrograms in order to interrogate sweep data. I've attached a couple showing the plot from what I think are the unrealistically high delays (853 and 896 uS) and what I've thought to be the best of the other settings that I tested (500 and 543). I must admit, I'm a little concerned at the size and position of the peaks in the time energy plots. They make me think that maybe I should not cross over at 250 Hz?? But I'd like to hear the views from folks who are more adept than I at interpreting these plots.
I also had a gander at REW documents discussing excess group delay as a means of selecting delays, I've attached a couple of those plots below as well for the same delays as used in the spectrogram plots. The only thing that stands out to me is the size of the peak at 263 Hz, which is smaller in the 500/534 delay pair. For delays between 500/543 and 853/896 the size of that peak grows and others come and go too. What should I make of those graphs - again, the peak at 263 Hz alarms me somewhat. If it is an artefact such as a room reflection then maybe I shouldn't be placing the XO at 250, maybe closer to 200 or 300 would be better??? By the way, the woofer has a 4th order HP at 50, which should account for the other massive peak at lower frequency I suspect. I've also attached the mdat files. As I said, these were recorded from my listening position, which is about 2 m from the speakers, with no eq nor room treatments.
Any help would be gratefully received!
In any event, the 853/896 and original 200/501 delays all sound quite good to me when I've done some listening tests. Possibly a testament to my cloth ears rather than to my ability to pick the best numbers from the data!
PK

 

[sm52]

To time-match the drivers, you need to take multiple acoustically synchronized measurements. One driver is involved in one measurement, the rest are disabled. the reference driver could be the other side's speaker tweeter. Or a separate tweeter. You need to measure the tweeter, mid, woofer, sub (if there is one). 4 measurements. The microphone should be at the listening point at ear level.

 

[Rolledmyown]

Thanks SM52, I did my close field measurements following the miniDSP recommendations using one speaker as the reference and in the test speaker just the tweeter and mid then just the mid and woofer, so two measurements for 3 drivers. You are saying to measure each individual driver - which method do you then use to calculate the delay from the measurements collected?
PK

 

[Rolledmyown]

Sorry, I should also have asked - should I deactivate the digital crossovers on each driver? For the sub I think the only thing I can do is to run at the max XO value as I don't think I can disable the LP.
PK

 

[Keith_W]

Here are a few points about driver time alignment.

1. The acoustic centre of the driver depends on the driver, where the XO point is (because of frequency dependent group delay), any phase rotation introduced by the XO or the driver, and lastly its physical offset on the baffle. Contrary to what you might expect, the physical offset plays a very small role, especially for low freq drivers. The ONLY way to determine the acoustic centre is by acoustic measurement.

2. The mic should be placed far enough from the speaker so that parallax error is not a concern. 1m mic placement is fine if it's a small speaker, but it may be a problem if we you are measuring a large speaker with drivers placed far apart.

3. If the goal is to achieve sufficient time alignment so that the result is inaudible, the psychoacoustic audibility thresholds are surprisingly large. For this reason, IMO it is more important to phase align so that there is no cancellation at the XO point. However, I do understand the satisfaction of achieving a perfect looking impulse and step response. My DSP allows me to do it, so I strive for it even though I know it's rather pointless. My only defence is that I like pretty graphs and that it costs me nothing.

This is how I prefer to do my time alignment:

1. Make sure the crossover and all driver corrections (if any) are in place prior to time alignment.
2. Place the mic a sufficient distance away from the speaker to avoid parallax error.
3. Delay the tweeter by a known quantity. Let's say 30ms.
4. Sweep the driver under test (DUT) together with the tweeter.
5. Examine the impulse response. You will see two peaks. The peak on the left (arriving early) will be the DUT. The peak on the right is the tweeter.
6. If it's a linear-phase crossover, measure the distance between the two peaks and subtract the delay you introduced, in this case 30ms.
7. If it's a minimum-phase XO, measure the distance between the START of the two impulses and subtract the delay you introduced.

 

[Rolledmyown]

Thanks for that Keith!
The parallax error did occur to me, and that is why I started to go down in delay as I expected that the path differences between the mid and woofer would be shorter than assumed. And my speakers are quite tall 3-ways. But it seems like the other assumption that I was making - that the physical offset of the driver is the only important factor to consider - is not the major concern! OK, so I've lived and learned!!
So in your method do you only have one speaker in operation - the one containing the DUT and the reference tweeter?
As I understand it, a 4th order LR XO is not linear phase so I should work on the commencement of the impulses.
With the Hypex I'm not restricted to that particular XO, I can dial in Butterworth or Bessel types and concatenate them to increase the order. I gather that I can also use FIR with my particular version of the plate amps, but that seems to be way beyond my capability!
I am just listening to music in stereo, not attempting to integrate video into my machinations so, up to a point, latency from DSP on the plates won't be an issue. However, I am using a pre-amp as my volume control and downstream of that are my speakers and sub. I gather the Hypex plate amps are pretty quick with analogue input (well under a mS) and my sub has a latency of about 6 mS. I guess I don't want to slow down the plate amps so much that they are slower than the sub.
Thanks for the advice folks!
regards
PK

 

[sm52]

I did my close field measurements

Then you should make your listening position in this place. Drivers aligned at the 1m point will not be aligned at the 2.5m point, much less the 4m point. The microphone, to achieve your goal, must be in the center of the listening position.

You are saying to measure each individual driver

Yes. Take 3 measurements with acoustic synchronization. If you are measuring the left tweeter, midrange and woofer, then specify the right channel as the reference channel. The microphone and drivers should not move until all measurements are completed. Then post your measurements. It is best to turn off all DSP and crossovers.

 

[Keith_W]

So in your method do you only have one speaker in operation - the one containing the DUT and the reference tweeter?

Correct, I align all the drivers on the left speaker to the left tweeter. That makes sense to me

As I understand it, a 4th order LR XO is not linear phase so I should work on the commencement of the impulses.

A 4th order LR can be either linphase or minphase. If you are using linphase, then the 90deg phase rotation per order does not exist. All symmetrical filters of whatever order will sum perfectly. So the only reason to choose a linphase LR4 is the slope.

With the Hypex I'm not restricted to that particular XO, I can dial in Butterworth or Bessel types and concatenate them to increase the order. I gather that I can also use FIR with my particular version of the plate amps, but that seems to be way beyond my capability!

The FIR taps on your plate amps is likely to be limited. This means they can not create low frequency XO's with steep slopes. They are more useful higher up in the frequency range.

 

[Rolledmyown]

OK, I've tried a couple of experiments and attached some of the results.
First, I placed my mike at the listening position then used L+R sweeps and slight adjustments in the mike position slightly until I no longer saw two impulses. So I think I've got my speakers equidistant from the mike. So far, so good.
After that, though I've got really confusing results, which I've attached. I've included results using substantial delays with and without crossovers. I've also include a couple or results where the delay was changed from 20 to 30 mS. I can't really spot any consistent differences between results, but that easily could be that I'm not treating the data correctly. For example, it took me a while to even present impulse data properly - eventually I noticed that I had to be using % as the Y axis units, so I may have other settings incorrect. I'm pretty sure the delay values are being properly configured on my plate amps, as I keep uploading the settings every time I make a change. There is a limit of 32000 uS for any delay so 20000 uS at least and even 30000 should be OK I hope.

 

[sm52]

Fine. And not good. None of these measurements can be used to move towards your goal. Do not click 'estimated IR delay'. Where are the measurements for each driver separately? Measurements must have a full frequency range, from 0 Hz to 24 kHz. If you don't want to give the tweeter full range (which is safe when using a sweeptonе), the range should be as wide as possible, such as 100Hz - 24kHz. Centering the microphone between the right and left sides is not necessary for your task. The microphone should be approximately where your ears are when you listen.

 

[Rolledmyown]

Ah, OK, 100 - 22k should be fine. I have a big cap hardwired to protect the cap, but it is arranged to roll off at about 30 - 40 Hz, so hopefully commencing the sweep at 100 will suffice, I could always short the cap if you think it still will have an impact.
I have to say that these answers to my questions have opened my eyes about drivers, especially the information that the physical location of the driver isn't its most important criterion. I guess that drivers are producing a spherical wave but the cone distorts the radius of curvature of the sphere and the "source" of the sphere is a virtual one. I also suppose the source position is frequency-dependent??
Also (if I can ever do these tests correctly!) time alignment seems to be the least audible tweak that I can carry out and I would be well advised to move on to and concentrate on ensuring phase matching. So far I have been attempting that by trying to minimize excursions in the phase trace of the phase and SPL plots. I did that by adjusting marginally the driver delays at the XO points. Is that the correct strategy or is there another approach that I should get familiar with?

 

[AJ Soundfield]

I'm no expert but do you have basic gated FR measurements at 1m on your design axis, whatever you decided that was?

 

[sm52]

time alignment seems to be the least audible tweak that I can carry out and I would be well advised to move on to and concentrate on ensuring phase matching

If you do the first one first, and then the second one, then after that you don’t have to invent anything, but just listen to the music.
If timing the drivers isn't of much value to you, you might want to connect a passive crossover to the drivers and do some EQ work. That's it.
Why is driver timing needed? The tweeter always processes the impulse faster than the mid, and the mid is faster than the woofer. With flat front panel drivers, you never hear the original sound. You hear the tweeter first, then the mids, then the woofer. The original sound is whole. All drivers must work so that a whole sound reaches your ear. This is what Vandersteen speakers are designed for. They are already 50 years old and they are still for sale.

 

[Rolledmyown]

I'm no expert but do you have basic gated FR measurements at 1m on your design axis, whatever you decided that was?

I thought I did, but I'll check! Good point too.

 

[AJ Soundfield]

My only defence is that I like pretty graphs and that it costs me nothing.

Well it also gets all the oohs and ahhs from 99.99% of online audio forum members, similar to those smashed flat in room FRs. Very funny stuff.