Hi There
Been messing around with Room Eq Wizard in an attempt to Phase /Time align my three way speakers. I have found the best method is to align unwrapped phase in conjunction with polarity inversion if needed.
I have taken measurements and aligned phase at 1.5 meters and 3 meters ( listening position).
At 1.5 meters mids are delayed by 0.10ms and Tweeters at 0.15ms for phase alignment.
At 3 meters im getting mids delayed at 0.20ms and Tweeters at 0.30ms.
So these two positions represent phase alignment at two listening positions. One is aligning in the middle of the room whilst the other near the back wall and listening position.
I guess this is something that mostly concerns delay correction for large venues where I believ the engineer must somehow average out the optimal point at which time alignment occurs resulting maximum perceived fidelity over a wide area.
Any opinions on this would be welcome.
I must ad that phase alignment really makes a dramatic improvement to the sense of integration and seamlessness to the sound.
Thanks
Been messing around with Room Eq Wizard in an attempt to Phase /Time align my three way speakers. I have found the best method is to align unwrapped phase in conjunction with polarity inversion if needed.
I have taken measurements and aligned phase at 1.5 meters and 3 meters ( listening position).
At 1.5 meters mids are delayed by 0.10ms and Tweeters at 0.15ms for phase alignment.
At 3 meters im getting mids delayed at 0.20ms and Tweeters at 0.30ms.
So these two positions represent phase alignment at two listening positions. One is aligning in the middle of the room whilst the other near the back wall and listening position.
I guess this is something that mostly concerns delay correction for large venues where I believ the engineer must somehow average out the optimal point at which time alignment occurs resulting maximum perceived fidelity over a wide area.
Any opinions on this would be welcome.
I must ad that phase alignment really makes a dramatic improvement to the sense of integration and seamlessness to the sound.
Thanks
Last edited:
10ms difference is way too much, that is more than 3m!
Are you using dual channel measurement?
If you target acoustical LR crossovers then the delay should correspond to the distance difference between the radiating surfaces of your drivers.
Delay differences can vary depending on listening/measuring position if you are not in the drivers axis (and of course that is always the case with a 3 way speaker with noncoincident drivers...), but aiming for aligned delays on the mid+tweet axis at a reasonable distance should be good enough.
I have found a good way to accurately determine the time delay between drivers in a multi-way speaker system, using Audacity. Generate click tracks with an interval of say 1 sec between the two frequencies. Make sure the frequencies are centered for the respective drivers being aligned. Play the recording through the system, muting out other speaker/drivers. Record the output in Audacity, using a high sampling rate. Measure distance between the start of adjacent peaks. You need to zoom in on the time axis. Now it is a simple calculation using the sampling rate, to determine the delay. This method worked really well for me, in my 5-way system.
@praba
I f I understand well what you say, you are doing it wrong.
here is a dirac on which I applied a highcut on Left side, and a lowcut on right side.
http://i.imgur.com/f6Y4ioI.png
You can't align with the top of the impulses.
What is correct is the very start of the pulse.
For accoustic signals, especially for the low band, this method is not viable.
I f I understand well what you say, you are doing it wrong.
here is a dirac on which I applied a highcut on Left side, and a lowcut on right side.
http://i.imgur.com/f6Y4ioI.png
You can't align with the top of the impulses.
What is correct is the very start of the pulse.
For accoustic signals, especially for the low band, this method is not viable.
Start of the disturbance is perhaps what I should have said. The click track itself would contain alternating clicks at different frequencies. During each click, only one of the channels would produce output. If these clicks do not come out exactly 1 second apart, as generated in Audacity, the time alignment has to be wrong, at least at that listening position.
Just one more point -- Dirac will cause a bunch of frequencies to be produced. In contrast, the click track will produce a single frequency for each alternating beat, as given by MIDI to frequency conversion charts. That this is the case can be seen by zooming on the individual clicks.
Here is a method that I always use to discover the "time difference" at the listening/measuring position. It uses an additional interference measurement.
The basis for this measurement is this: if you have two sources separated in space the output from each will interfere. For example you may heard of "comb filtering" - that's an interference pattern from two (or more) drivers.
You make 3 measurements: driver 1, driver 2, and then a measurement with driver 1 and driver 2 both "on". The latter records the interference of the two drivers.
It turns out that the interference "pattern", that is the frequency response, contains information about the distance that the two sources are separated. I use my ACD tools and do the following:
1. Load driver 1 measurement
2. Load driver 2 measurement
3. Create the sum of driver 1 + driver 2 as the "system"
4. Import the driver 1 + driver 2 measurement and display on the same plot as "system"
5. Adjust the offset of the acoustic center for either driver 1 or 2 (this is like adjusting the delay, or distance between them) until the system and the "driver 1 + driver 2 measurement" line up.
You can get an almost perfect match, and the resolution is very, very high, e.g. 1mm or better.
To do this you need the driver 1 and driver 2 measurement to include the phase. It can be measured phase, but it's best to try and remove the extra measurement delay to reduce the amount of phase wrapping for these. The driver 1 + driver 2 measurement does not need to have phase - you are only comparing the system SPL to that SPL so phase is not needed.
This method is very powerful. For instance I could do the following:
driver 1
driver 2
driver 3
driver 1,2,3 all "on"
I can then use a fitting routine to fit the offsets for 2 out of 3 of the drivers automatically. NOTE: these are relative offsets, e.g. from each other, thus you get N-1 offsets for N drivers. One of them is used a the "reference" and is considered to be at a "zero delay" plane.
The basis for this measurement is this: if you have two sources separated in space the output from each will interfere. For example you may heard of "comb filtering" - that's an interference pattern from two (or more) drivers.
You make 3 measurements: driver 1, driver 2, and then a measurement with driver 1 and driver 2 both "on". The latter records the interference of the two drivers.
It turns out that the interference "pattern", that is the frequency response, contains information about the distance that the two sources are separated. I use my ACD tools and do the following:
1. Load driver 1 measurement
2. Load driver 2 measurement
3. Create the sum of driver 1 + driver 2 as the "system"
4. Import the driver 1 + driver 2 measurement and display on the same plot as "system"
5. Adjust the offset of the acoustic center for either driver 1 or 2 (this is like adjusting the delay, or distance between them) until the system and the "driver 1 + driver 2 measurement" line up.
You can get an almost perfect match, and the resolution is very, very high, e.g. 1mm or better.
To do this you need the driver 1 and driver 2 measurement to include the phase. It can be measured phase, but it's best to try and remove the extra measurement delay to reduce the amount of phase wrapping for these. The driver 1 + driver 2 measurement does not need to have phase - you are only comparing the system SPL to that SPL so phase is not needed.
This method is very powerful. For instance I could do the following:
driver 1
driver 2
driver 3
driver 1,2,3 all "on"
I can then use a fitting routine to fit the offsets for 2 out of 3 of the drivers automatically. NOTE: these are relative offsets, e.g. from each other, thus you get N-1 offsets for N drivers. One of them is used a the "reference" and is considered to be at a "zero delay" plane.
Last edited:
- Status
- Not open for further replies.