The Real World

The Real World

Today’s post comes to you from a 600 seat concert hall in upstate NY. The hall was originally constructed for acoustic performances, and as such is far more reverberant than we’d like for reinforced music. The seating is covered by what some of my system tech buddies jokingly call a “dash” array – that is, a ‘line array’ that’s too short to exhibit any meaningful line source behavior at any frequency. Specifically, four boxes per side, which is barely enough to cover the seating area. What’s more, they’re all driven off a single DSP channel per side, so optimization options are few.

Since this is the type of situation that we are very likely to encounter often in the real world, I thought it might be illustrative to show what “should” be done in this situations vs what can be done.

Here we see the house left array response over the seating area, with mic positions at the ONAX of each box in the array.

 Here is an alternate view of the same data with 1/12 oct smoothing, thicker traces, and the coherence trace visible for position ONAX A (top box, back rows).

The subwoofer is muted. A lot of working with an analyzer is displaying the data in a format that’s comfortable for you to parse and understand. I tend to prefer less smoothing and thin traces, but when writing or presenting I use more smoothing and thicker traces just to make the trends a little easier to see for people who may not spend all day staring at analyzers.

I was asked why I don’t generate an average traced based on these. I’m not looking for trends, I’m looking to see the difference over the space. Trace averaging would completely hide the issue that is the basis of work in this post. It’s a “video solution to an audio problem.” I don’t want to make the traces look better (otherwise I’d use Photoshop), I want to solve the problem that makes them look that way.)

The pink trace (ONAX A) is below the level of the others from 250 on up, and much lower above 6k3 or so. This, coupled with the relatively low coherence from being that far back in a reflective room, translates to poor intelligibility for the folks in the furthest seats who need it most. What are the textbook solutions?

Solution 1: More boxes. This is out, at least without a lot of budget and advanced notice: it’s a permanent install, and the rigging is a deadhang so we can’t easily bring the array in. We’d need to bring in motors and a rigger.

Solution 2: Close up the angles between the top few boxes. Besides the same logistical issues (can’t bring the array down, can’t pull pins out from a lift with the array hanging), we’re out of coverage angle. The bottom two boxes are already at maximum splay, so by closing up the top, we’ll be creating a coverage issue at the bottom. Perhaps this is preferable because front fills could be added, but they can’t stay on the deck, so for most events they won’t be added because people are lazy, and so we still haven’t solved the problem. As an added bonus, this particular model of box has particularly course options for splay angle, and the next choice is too narrow for this application.

Solution 3: DSP. Tweak the gain and HF shelf on the top box. Except we can’t, because we’re all on one channel here. Or can we?

These active boxes have a pot on the back that sets the input level to the internal amp. We can tweak that to get some gain back. There’s also a per-box option for HF compensation, and we can try our luck with that to see if we can get an HF shelf boost out of it.

As soon as we talk about level shading arrays, audio people get really nervous. The first issue is that once the array is level-shaded, some boxes are going to go into limiting before others. Whether or not this is a problem is heavily dependent on the situation. Typical shading situations involve bringing the bottom few boxes down by a few dB to reduce the extra HF down front. We try to do as much as we can with splay, but generally we still will need a few dB via gain, and this isn’t really an issue of limiting because these boxes, in being turned down, now have *more* headroom than those pointing at FOH, so there’s no problem. It’s when the higher gain boxes are further back in the house, like this example, that we need to be more careful. By setting the gain higher on the top box than for the rest, we have a situation where the top box will hit its limter first and the FOH engineer down in the middle of the house won’t hear it. If this is a high-SPL situation (rap, rock, EDM, etc) where the PA is taken to its limits every night (giggity), this is probably something we want to avoid.

However, this is a situation that prioritizes low variance and high intelligibility, and the PA is not operated anywhere even close to its limits. For the types of events staged in this hall, people would be lodging loudness complaints (it’s lots of old people, okay?) far before the PA hits its maximum output, so the dynamic range issue isn’t a concern, whereas creating a low-variance, high-intelligibility environment for the patrons is definitely a priority for the venue’s management.

The other, more common, and far more baseless objection to array shading is that any gain offset between boxes “breaks the line.” Bob McCarthy has written what I consider to be a landmark article about this which outlines a number of reasons why this concern is unfounded, but I’ll address the most commonly stated objection here: that shading down some boxes reduces the array’s headroom.

Of course this is true, but in reality the loss is so tiny that it’s negligible in all but the most extreme situations. We’re talking of course about LF because that’s where the boxes have the most overlap and thus the most summation. We have a 4 box array, which means that the LF headroom of the array is 12 dB above that of a single box. That is, 0 dB + 0 dB + 0 dB + 0 dB, which everyone knows is 12 dB. No? Here’s the same math in linear for for boxes at unity:

1 + 1 + 1 + 1 = 4

Get back to log with

 20log(4/1) = 12 dB

Now let’s say we turned one box down by 2 dB. How much LF coupling do we lose?

-2 dB is about .8 in linear (you’ve got these memorized, right?)

So we now have

0 dB + 0 dB + 0 dB + (-2 dB)

which in linear is

1 + 1 + 1 + 0.8 = 3.8

That comes out to 11.6 dB rather than 12. (20log3.8 = 11.6, or use 20log(3.8/4) to get the dB loss in relative terms as compared to the no-shading situation)

In a situation that demands low variance, a half a dB in LF headroom is a small price to pay for reducing front-back variance by 2 dB. For the same reason, I’m not concerned about goosing the top box by a dB or two.

Also for the same reason, bringing up the gain on the top box (or down on the rest, if you like) isn’t going to cause a significant effect at LF, where all the boxes are working together. It will, however, help increasingly above 1k, which is where our problem is. With one “pot notch” (whatever that is) of gain and the HF “coupling compensation” onboard DSP filter engaged, here’s what we get:

If you compare that to the first image, you can see that we actually overshot a bit: the back seats are leading the rest of the array in the HF by about 3 db above 1 kHz. Given those two options, I certainly choose the latter. The intelligibility in the rear is greatly improved over the original state. In addition, I’ll take an extra dB or two in the back simply because of the stage sound down front. What we have here is a PA that does not get louder as you get closer to it, and a high degree of confidence that whatever the FOH engineer mixes will be heard in all the seats, which those with touring experience can confirm is a relative rarity in venues of this size. Even with limited options on the table, a small change here and there goes a long way.

Leave a Reply

Your email address will not be published. Required fields are marked *