Time Alignment using ARTA (description and pictures)

Time Alignment showing microphone at listening position

ARTA running on laptop and Focusrite Scarlett

For many years up until now I had not been satisfied with my methods of measuring time alignment.   Measurement is the hard part.  Actually adjusting the time alignment is trivial, I simply dial in time delay values into the separate Behringer DEQ 2496 units that digitally EQ the signals for my SVS PB13 Ultra subwoofer (the bass below 100 Hz), the Acoustat 2+2's (the midrange from 100Hz to 17kHz), and the super tweeters (two small dome tweeters with response to 40kHz mounted on a wood box).  Now that I am also using miniDSP OpenDRC-DI units to perform the actual crossover functions for each of these 3 ways, I could dial the delays required into the MiniDSP's instead, but I find it easer to make adjustments with the Behringer DEQ units because they don't require a computer to be attached.

Stack of MiniDSPs on top of stack of Behringer DEQ's

Previously I had used the measurement program that is part of my Tact Room Correction System 2.0 Preamp (I don't use the Room Correction itself, but I have often used the RCS measurements).  That is strange and hard to use.  Last year I started using Room EQ Wizard (REW), which is OK for doing loudspeaker+room measurements, but didn't seem helpful specifically for the time alignment.  Just last week I downloaded and registered the ARTA program used by many loudspeaker designers.  For only $100 it is a far better program IMO than REW and after a week of fiddling around with it, I came up with a very intuitive, repeatable, and I believe accurate methodology for time alignment.  The method has two parts:

1) Align the leading edge of the acoustic output of each driver.  I measure each driver separately with the Signal Time Record feature of ARTA.  For midrange and super tweeter, I use a pulse with width "1".  I expand the vertical and horizontal scales of the time record display, and put the cursor at the exact position where the signal begins, as differentiated from room noise.  This is easy to see with the scales expanded, though less easy for the subwoofer output.  I adjust the delay times in my Behringer DEQ's so that the signal beginning times for each driver are the same, within about 0.02 ms for the panels and 0.06ms for the subwoofer (which is hard to see as clearly).  When measuring the subwoofer, I change the pulse width to 1000, otherwise it doesn't show up at all.

Leading edge of Acoustat signal

The ARTA Signal Time Record is an unprocessed display, similar in principle to a storage oscilloscope.  It can be used with a simple up and down pulse, which results in the measurement shown above.

The Impulse and Step displays are derived from longer term signals using FFT, and therefore show the room response as well as the speaker response, and not just from the very first instant.  For the first phase of time alignment, we need to examine ONLY the very first instant of the signal.

2) For the subwoofer and electrostatic panel alignment, I further fine tune the delay adjustments to either maximize the output at the 100 Hz crossover frequency, or minimize the output with the polarity of the panels reversed.  I think the minimization method is the better one, and IIRC was specifically recommended by Linkwitz himself.

When the polarity of the panels is reversed, the subwoofer and the panels are "cancelling" each other, because normally the LR4 crossover has them "in phase" at every frequency, especially the crossover.  I was pleased to achieve deep null using the cancellation method.

Deep Null at 100 Hz Crossofer Frequency (Green Line)

To measure the output at 100 Hz, I run an Impulse measurement with ARTA, and then run the Frequency Response and Distortion analysis.  This gives a remarkably stable frequency response graph, so stable that averaging his hardly needed, but I usually averaged 3 runs (a setting in the Impulse Response dialog) to be sure the measurements were not contaminated by ambient noise.  Two runs in a row typically show the same value at 100.3 Hz with 0.1dB precision.

*****

I had never been able to do the leading edge alignment before with any feeling of confidence.  No other tool has given me as clear a display of the signal vs time as ARTA.  You may be able to find pictures of my previous attempts farther back in this blog.  In every previous case, a lot of guesswork and interpretation was involved.  Not so with the Signal Time Record feature.

Although the ultimate alignment is done with the frequency output at the crossover frequency, the leading edge alignment is still important, because it is a good place to start.  When starting the alignment on a system like mine, one doesn't really know the latency of processors and DACs in line.  It all has to be included in the "starting point" which can then be further optimized at the crossover frequency.

Sometimes I run the signal time record feature several times if the transient looks affected by nearby noise.  I didn't at first realize how to zoom and scroll the ARTA window, so my earliest pictures are not as revealing.

Here is the final determination of the leading edge of the Acoustat panel output.  I put the marker where the signal begins abruptly downwards first.  (I do not know why it goes downwards first, by all other measures all my speakers are unambiguously "in polarity."  I've spent much time verifying that by various technical means, including and hand-made asymmetric signal and a smart phone app.)

Leading edge of Acoustat signal

Here is the final determination of the leading edge of the supertweeter output (with the miniDSP OpenDRC-DI running at 24/96 using minisharc 4x10 plugin).  I line up the marker with the leading large edge, ignoring smaller pre-ringing.

Leading edge of Supertweeter signal

Here is determination of the leading edge of the subwoofer signal, after I finally figured out to expand the horizontal and vertical scales first (previously, it was looking for the first pixel).  Prior to exampding the scale, it was virtually impossible to see where it started closer than 1ms or so, because the initial start is so low in level.  There is still some ambiguity because of noise, and sometimes I will try another run to see more clearly.

Leading edge of Subwoofer Response

The subwoofer start time does not actually optimize its phase at the crossover frequency.  That is best measured at the crossover frequency by putting one way (the midrange) out of polarity.  The supertweeters are deactivated.  At the crossover frequency, if the levels are set properly, there should be a null in the response.  I was pleasantly surprised that there was a deep null.  And then that null is made deeper by moving around in 0.02ms increments until the optimal delay value is determined.  First I tested positive and negative 0.02ms increments away from the leading edge alignment, to determine which direction to go in.  Then larger increments in delay time are tried, backing up when the notch starts getting less deep.  Ultimately, the optimal delay adjustment causes a very deep null, differing significantly from 0.02ms forward and backward delay adjustments around it.

In this picture the green line is the system response, and the lines below are % distortion (which is very high where the fundamental cancellation is occuring, an artifact of the measurement being done).  I set the marker as close as possible to the crossover frequency (it is set to 100.3 Hz).  Then I can read the amplitude very precisely (from one run to the next, it is often exactly the same to 0.1dB, except at the deep null where it goes lowest, it actually varies a lot, but always lower than the surrounding points).

You may note the notch in the response actually appears to be centered below 100Hz.  This is because of room acoustics and rear wall reflection.   I measure the level as close as possible to 100 Hz, ignoring any deeper notch below 100 Hz.

The high end response in the graph above rolls off because supertweeter is disconnected.


Here are the delay adjustments dialed in on June 19, before converting the supertweeter miniDSP to 96kHz operation.

Here are the ultimate delay adjustments dialed in, after the miniDSP for the supertweeter was switched to 96/24 operation on the afternoon of June 20.  Note that the super tweeters require 1.3 ms more delay than before because the 96/24 path has lost 1.3 ms latency for some reason, compared to when it was running at 48/24.  The change in latency might be from the miniDSP itself, or in the following Behringer EQ, or in the DAC, or all of these combined.

The final delay adjustments on June 20

I was unhappy to see the following left channel system response when all was done.  It was not very flat in the bass, with 100 Hz itself still in a notch (after correcting the Acoustat polarity), with a huge rise over 15dB from there throughout the deep bass below 100 Hz.  This was not my intention (I intend to have "flat bass", or what I call "electrostatic bass" even though realized with a dynamic subwoofer).

After adjusting the magnitudes of my pre-existing parametric EQ's, but NOT their frequencies (which were adjusted by hand tuning an oscillator, and are therefore "real" resonances and not digital artifacts) I was able to make the bass much flatter.  I still couldn't totally flatten the deepest bass without creating undesired holes there.  So I left it like this:

This graph is also showing the effect of the super tweeter in the rise abov 15kHz.  The super tweeter is optimized through other means, mainly using the microphones of my smartphone (which use algorithms to simulate a perfect spherical omnidirectional response).  The rise in on-axis response shown here above 15kHz is not representative of the average room response, or even at ear positions a few inches to the sides, because of the high degree of beaming from the supertweeter at those frequencies.  The on-axis level must be exaggerated at the exact center (where the microphone is) to create the equivalent auditory sensation as a live performance.  This is the opposite of the situation which calls for the Gundry (aka Linkwitz) Dip, which is another useful alteration of flat on-axis response,  because small rooms reflect unnatural amounts of 2-6kHz directly into the ear from side reflections.  Also, I have not as yet incorporated my microphone calibration into ARTA, and my microphone has a slight HF resonance contributing to the rise shown here.

Whenever I've lowered the supertweeter level to make it appear flatter with some kinds of measurements (and not so much my smartphone held at the listing position, which shows almost flat response at 20kHz) it sounds wrong to me, dull and more irritating actually.  The supertweeter takes away any sense of "strain," making things more rather than less listenable, stridency is reduced when the super tweeter is active.  But when the super tweeter is too low, the strain comes back, the magic is gone, and I'm just listening to 30 year old electrostats not a live band.

Here are the new PEQ adjustments for the left channel bass:

Notably I reduced the notch filter at 106Hz down to 0.5dB, to help neutralize the dip around 100 Hz.  I didn't zero out the notch at 106 Hz so it could be increased again if it later seems this decision was not altogether for the best.  I also increased the depth of the notches at 45.3 and 71 Hz and the broad cut around 28.3 Hz to help flatten the deepest bass.


The right channel looked better, especially in the bass, so I didn't make any EQ changes there: