Here’s a plan view of a sound system with two mains, a sub, and a measurement mic at a fixed location. If I asked you to align this system, would you prefer to start by muting the left side to align the right side to the sub or would you like to play both left and right together with the sub?
My guess is that you’d probably start by muting the left side to align the right main to the center sub, especially if you’re in a hurry and indoors where your measurement is prone to even more contamination than from just another loudspeaker. If so, you’d be in good company based on a few polls I ran.
What if I show you a prediction at 80 Hz with both left and right mains playing that reveals that the mic is in a power valley?
You’d definitely want to mute that other side to do your alignment now because you know it’s going to contaminate the measurement and make your phase trace go crazy.
What if I rotate the room 90º and ask you to do the same alignment?
Would you change your strategy?
Nothing has changed so hopefully you would do exactly the same as before and mute one side to do your alignment.
What if I cut off half of the room and enable the bottom boundary?
Let’s make it look more like a section view.
Everything is in the same place. Nothing has really changed for your alignment.
You might say, “Sure, I’d like to mute the other side, but that’s impossible.”
Maybe this dissonance is reflected in these polls. Or maybe I just didn’t phrase the question clearly.
Why is that?
Most of us have made our peace with the first scenario in plan view. You don’t necessarily need to align with both sides on because the left side will arrive with a slightly lower level and the right side will be slightly more dominant. So you mute the other loudspeaker and not think much of it.
If you rotate the room 90º the story doesn’t change except that house left in plan view has become the mirror image—living on the other side of the floor—in section view and you don’t have the luxury of turning it off. That’s the conflict. Your phase response becomes compromised and you never had the opportunity to prove to yourself what is was like in the absence of the mirror image.
Measured indoors, the phase trace can be hard to interpret or even so jagged that you think “Why bother?” or resort to a gratuitous amount of smoothing which is not the same as muting the mirror image.
This has made field tests of SubAligner in small rooms a real problem. I have done four tests in Canada, Spain, Greece, and Minnesota and they have all gone the same way. All of the measurement and listening tests are good until the mic goes to head-height. In the direct field it’s clear, but as soon as the room gets involved it falls apart.
This is a pretty big problem since I’m trying to sell a product that is supposed to create an audible result. If you can’t hear an improvement, why would you buy it?
SubAligner is a simple web app that uses a method of comparing relative distance offsets in main+sub alignments to pre-alignment values in its database. It short: it saves time on sub alignment.
I didn’t invent the method. I learned it from Merlijn van Veen. He got it from the L’Acoustics Preset Guide (Time alignment with geometric distance) and d&b Line Array Design along with his own field experience. The process is also described in the 2011 AES e-brief Time Alignment of Subwoofers in Large PA Systems.
From what I can tell the method is solid, but (and this a big BUT) you won’t always be able to hear or measure the results. It requires a leap of faith.
Fight the room
A floor bounce will cause a negative shift in the average phase measured at the microphone. Delay may be introduced to correct this local misalignment with unknown effects on the rest of the audience.
Here are two matching sources. With no boundaries enabled they create a perfect line of summation along the perpendicular bisector or median plane between the sources.
If the floor is enabled the central power alley seems to be pushed up.
It is tempting to add delay to steer it down. Here the summation pattern has been outlined.
The entire shape is pivoted around the main speaker back into alignment so that the ground sub is properly delayed.
That should do it, right?
Damn it. I’m not sure if that’s better. There may be more of the audience in the blue color (lower level) now. It got better in one spot, but the rest of the audience has suffered.
Measuring LF indoors is scary
One challenge to alignment decisions is the excessive ripple encountered when measuring indoors, especially in smaller spaces.
Here’s a prediction of a subwoofer measurement using the previous example with 4 boundaries enabled. The more reflections, the more ripple, the harder the story is to read.
In this case the envelope of the measurement is still visible. You can see the trend and and indication of the original uncontaminated trace.
Let’s look at field data from a real room featuring two more walls. Here are measurements for a main+sub alignment that I attempted at a small church north of Minneapolis.
Unless the audience is close enough where they can practically touch the subwoofer, most audience members will suffer from poor direct-to-reverberant ratios. Most audio professionals, myself included, would throw up their hands and say, “Forget about it.”
Average out the floor bounce
If measuring at a single location indoors is highly prone to error, how can better data be obtained?
Let’s measure at eight positions across the audience and average them together. Here’s the example from the beginning of the article with the addition of a standing height audience.
Here’s all 8 measurement of the main combined with its mirror image, plus the average.
Here’s what that average looks like compared to a solo on-axis measurement without the mirror image (ie. anechoic).
The mirror image’s delayed arrival time varies throughout the audience such that it averages out. The jaggedness is inversely proportional to the relative level offset between the main speaker and its mirror image.
If you are wondering why there seems to be more phase difference around 100Hz, it’s because those measurements made closer to the speakers have greater magnitude and a greater influence on the average. This is generally helpful since we’d like to average out the nulls of the comb filters. Here is an unweighted average for comparison.
Ian Robertson from GerrAudio in Canada graciously agreed to test this with me.
The setup included eight Lina and a single 900-LFC.
After consulting the Sub Align calculator we could see that the phase span from front to back was very small (28º) and position eight was chosen for the alignment with SubAligner. (at this point I should probably apologize for titling my app SubAligner so similar to Merlijn’s calculator Sub Aligner or even Waves Sub Align plugin, but the URL was available so here we are)
Here’s are the measurements at position eight with the crossover region highlighted.
Here are the solo measurements of main and sub at each microphone position along with their respective averages.
Here are the averages alone.
I have to be honest that I wasn’t sure this would work, so I was really happy to see the alignment clearly visible, but also the similarilty with SubAligner’s predictions.
The first time we tried it, it didn’t work. My mistake was in calculating the coupling zone and alignment position in Sub Align based on a ground-plane height instead of head height, which has a much smaller phase span.
Here’s is the method Ian and I used to test SubAligner.
I should first note that if you’re measuring outside away from boundaries you might not need to do any of this. These steps will be necessary indoors and anywhere where signal-to-noise ratio is low.
In order to find a global average free of local anomalies measurements must be balanced in level and phase. In level, in order to not tip the average towards one measurement with greater amplitude (eg. positions closer to the source). In phase, in order to not tip the phase trace too far in one direction or another.
Use the Sub Align calculator to find an alignment position in the center of the coupling zone. To do this, If the phase span is less than 240º, divide the phase span by two. If it is more than 240º, use 120º as the phase offset.
- Example 1 – Alignment phase span 226º: Enter a max phase offset of 210/2 or 105º.
- Example 2 – Alignment phase span 436º: Enter a max phase offset of 120º.
Use at least eight positions distributed equally around the XOVR position. This may seem counter intuitive at first if the XOVR position is not near the center of the audience, but if there are an unequal number of measurements on one side then the average will favor that side and will appear to be mis-aligned. Of course, this is all much more critical with very large phase spans. If the phase span is 10º then the distribution is less critical.
Results indicate that a subset of eight measurements locations spread over the target audience area represents a rational solution to characterize the loudspeaker system response.Optimum Measurement Locations for Large-Scale Loudspeaker System Tuning Based on First-Order Reflections Analysis
Most designs will have an asymmetrical relationship between the main, sub, and audience, and therefore require an asymmetrical placement of microphones.
Start at the closest position in order to avoid overloading the mic preamp. Use this position as the magnitude reference. At each new position reset the mic preamp to match this reference and reset the delay locator to the main. This will equalize magnitude and phase offset introduced by distance.
Back to the original question
When faced with an unknown number of mirror image reflections in a room, what do you do?
Do you rely on what you know to be true, or speculate on what might be true?
What you know to be true:
- Phase offset caused by distance offset is predictable and therefore alignable. Geometry tells us exactly where to align and what parts of the audience will be in the coupling zone and what areas will be in the cancellation zone.
- The pre-alignment values you created for this array were validated at the warehouse. You know they work.
What we can only speculate about:
- Attributes of the speaker’s mirror image living on the other side of the floor and every other wall.
- The validity of LF measurement data under poor D/R conditions.
- Where the null caused by floor bounce will end up at once the room is full of humans.
Back to Milaca. If we took that same system outside and measured on a calm day under near anechoic conditions I’m 95% positive we would get the data and see alignment. Inside where reflections dominate, suddenly the analyzer is showing me a different story. Unfortunately, it’s not a video game where you can turn off the walls. (the irony of Phase Invaders is not lost on me here)
Too much noise will render any measurement system ineffective. If it’s not working, try a different tool.
What has been your experience with trying to get actionable data indoors. Let me know!