time aligning speaker systems

I just spent two or so weeks in the wilderness trying to time align my horn system. Just like with info on starting VR tubes, this stuff is not described well in one single place on the internet. So now that I have pieced together a successful (enough) method to time align, I thought I pass it along.

First, let’s define an example speaker system; 3-way, woof–mid–tweet, 200Hz and 2kHz x-over frequencies:

The first thing I had to learn from various sources on the net is that there is no such thing as all frequencies arriving at the same time (at your ear or measuring mic), even if you have achieved the mother of all time alignments. It is much better to think of the response of every driver to arrive as trains of different length. This image, from this page of Yuichi’s audio lab, shows the different trains:

The highest frequencies that come out of a driver (say, up to 20kHz for the tweeter, 200Hz for the woofer) dominate the front of their arriving train and define how sharp this front looks. The lowest frequencies that come out of a driver (say, down to 20Hz for the woofer, 200Hz for the mid driver) dominate the back of their train and define how long each is. Both of these are shaped by the x-over before the driver.

It is easy to define time alignment as when the front of all the trains of all drivers arrive at the same time. The problem is that the huge differences in sharpness when you look at the fronts of the impulse response trains make them incomparable.

The other thing I had to pick up from the net was to concentrate on the x-over region when aligning two drivers. This started to make a lot of sense.

It can only confuse us to compare the arrival of the 800Hz component of the mid driver with that of the 80Hz of the woofer. But the 150, 200 and 300Hz that are coming out of both the woof and mid are comparable, if we avoid getting confused by the rest.

So the trick for time aligning the woofer with the mid driver is to limit the bandwidth of both for a moment to the x-over range, the area around 200Hz that I hatched in the drawing above. This is achieved by leaving the low-pass of the woofer and and the high-pass of the mid driver as set up for 200Hz, and set the high-pass of the woofer and the low-pass of the mid also to 200Hz, with slopes to match the other diver’s response. Now (roughly) the same highest and lowest frequencies are coming out of both drivers, which makes their impulse trains much more comparable.

This is of course much easier to do with a digital x-over. Setting high-and low-pass at the same frequency pushes down the max output of each driver quite a bit. I compensated by setting my post-x-over volume attenuators 10dB higher.

Now we can run a sweep or a pulse through either driver, and compare the resulting impulse response (using the step response gives significantly different results, I am more happy with the impulse results). You will still see that the pulse from the woofer is still a fraction more stretched out that the one of the mid; it is in their nature. I time aligned both drivers by centring the first major peak of either on another. It was easy to read out the times of the apex of both peaks, take the difference and put it in the digital x-over as a delay of the driver that was ‘early,’ both with five microseconds of precision (you could also physically move the driver enclosure to delay).

The whole thing is then repeated for the mid and tweet, but now for the 2kHz x-over range (also hatched in the drawing above). Delay times simply add up. If you have to delay the mid for the tweeter, then you’ll have to add this delay time to the woofer too.

It should be clear now from the last drawing that there is no such thing as time aligning the woofer with the tweeter in our example system; they do not overlap in range. The tweeter is aligned with the mid and the mid with the woofer, and that is how the response arrives at the listener.

Having heard at least twelve wrong ways to measure and time align drivers in the last two weeks, I can testify how awful it sounds to be 0.1 or more milliseconds off. But when they are all well aligned (enough), the speakers snap together like magic.

(I am still curious if theoretical or practical improvements can be made to this method; if you can contribute, leave a comment)

new: ready to try it out? Check out my speaker time alignment cookbook.

update: more experimentation, and another stretch of wilderness, shows best results through centring the very first peak in the impulse trains. For instance: I have—effectively, i.e. electrical + acoustical combined—fourth order x-over slopes in my system. The initial impulse response of the (correctly phased) drivers looks like this: small peak up, large peak down, small peak up. It is that very first small peak (instead of the major one, the large one down) that the measuring and centring has to be done.

ps: curious about which horns I was aligning? Stay tuned. I am still trying to work out one kink, but will publish them, with kink if needed.