Tuning Recap: Mixing Subwoofer Models, FIR in Action, and A Question of Tilt
One nice thing about hanging the same rig in the same room regularly is that I get to explore and experiment, and try to improve things each time. Here are some of the things we worked on this time around:
Mains
Mains are 6 boxes per side zoned as three composites (2 @ 3°, 2 @ 5°, 2 @ 10°).
The last few times in this room, I had been slowed by the need to move the mic so many times. This time around I used three mics at OnaxA, B and C, plus a live average, which allowed us to bang out the entire tuning in minutes. An important first step with multiple mics is to verify them before you start, otherwise you could be chasing level or spectral variance that doesn’t actually exist.
Here are all three with their capsules coincident and their gains adjusted at the I/O to level-match. This is two iSEMcon EMX-7150 and a single Behringer ECM8000, with no cal files. Close enough to dance.
My level variance goal was 6 dB front to back, which we accomplished largely with splay angles. There’s a hi-shelf cut on the bottom composite to keep the HF in line, which you’ll see shortly.
The hump around 200 Hz at OnaxC is the relatively short array going into lobing. That can’t be EQ’d away, and it’s the reason for the average (gold) pulling up around there. Averages never tell the whole story, and EQing solely based on that average would leave a hole at 200 Hz for most of the room. Note also the phase trace hanging around 180°. This boxes use internal FIR to linearize the phase above 250 Hz, and for design reasons that results in a polarity inversion.
Although the gentle LF tilt is inoffensive, we decided to tune to a flatter target curve for two reasons:
-
the board recording is far more accurate without the need for further EQing later on. The artist had requested a board mix.
-
In this room I often find myself cutting back lo and lo-mid energy from multiple inputs. Flattening out the PA tilt should alleviate this.
A -11 dB lo-shelf at 700 Hz to untilt the response (Left), plus the hi-shelf cut required to get the bottom composite to fall in line (right):
Here we see the result of the untilt, showing A,B,C plus the live average (black) and the original average (gold) for comparison. Notice how the inverted system EQ trace, also called 1/EQ, (mint green) lies over the top of the response we seek to offset.
Subs
Although not a sub-heavy show, the production company had a pair of single 18 subs from a different manufacturer that had just come back from repair so we threw them on the truck with the plan to test them if time allowed.
People often ask about combining subs from multiple manufacturers. A dose of caution is advised here but this time I got lucky.
With differing on-box LPF settings, the responses matched very well both in magnitude and phase. Since they are both direct-radiating single-18 boxes from related manufacturers, this is a better outcome than should be expected with “any two subs.” I’ve also seen wildly differing phase responses and even some lurking polarity reversals so combining unlike subs is best approached with measurement data. Otherwise you have a “box of chocolates” situation.
Front Fills
In the previous post about this room I showed the fundamental phase incompatibility between the mains and the boxes used as frontfills. This time I came prepared with a copy of FIR Creator EX.
The black trace shows the target curve (the averaged response of the mains throughout the main coverage area). The purple shows the response of the mains at the Main-FF crossover location. The mains are relatively consistent off-axis, and they’re wide enough that in this room there’s not really a hole in the coverage down front. But we decided to go ahead with the fills anyway – in case the stage volume turned out to be an issue, we wouldn’t lose the vocals to stage wash. Their response at XO shows a fundamental phase incompatibility.
We are in time around 3 kHz (flat trace), early above, and late below. Delay can shift the small “in time” region in frequency but can’t change the overall contour of the unit’s phase response.
Luckily, the FF arrival at XO is 13 ms early, so we have some time to play with. A 480 tap FIR filter (10 ms long, 5 ms filter delay) means I can take the fill delay in my DSP down to 8 ms and linearize the phase of the box above 1 kHz or so while still arriving on time.
I could have loaded the target curve into the software to use as a design target, but due to the mains having a flat phase response in the region of interest, we can speed things up by simply aiming for a flat phase response and then inverting the polarity of the fill. This does in fact throw things off at LF but it’s a fill and it’s going to get high-passed so we don’t care.
You could use FIR Creator’s Auto-Phase function but we’re asking a lot of correction here from a relatively short / low-resolution filter (100 Hz) which results in significant error in both phase and magnitude in the final response. So I try to get as close as possible to the target phase response using a series of cascaded minimum and maximum phase all-pass filters, and then let the auto-phase algo put on the finishing touches.
Think about the idea of using EQ to equalize deviations in a system’s magnitude response. We use an EQ curve that is equal and opposite to the deviations we wish to correct. In simple terms, the goal here is the same. If we design a filter that has the inverse phase response of the loudspeaker, they should “sum to flat” in the same way.
The top right pane shows the inverted phase response of the front fill as measured (light red) and the green trace shows the combined response of seven cascaded all-pass filters (two maximum phase and five minimum phase) that comes close to matching the target. The bottom pane shows the magnitude response in blue, along with the phase response as measured (light red) and after the APF stages (dark red).
Finally we have everything within about a +/- 90° range of target, and we can allow the auto-phase algorithm to close the gaps.
Note the bottom pane now shows the corrected response to be flat above 500 Hz. That’s the goal.
The transfer function of the filter and DSP itself:
Notice the small amount of magnitude ripple around 500 Hz. This would have been much greater if we hadn’t started building the APF by hand.
Let’s load that FIR into the DSP, invert the polarity, take off 5 ms of delay, and remeasure the FF:
Yeah, baby: the filtered FF (brown) now matches the mains response (purple) very closely above 500 Hz, and sums wonderfully with no weird issues (green).
For comparison, here’s the high-end of the main/ff summation from the last time I worked in this room (without FIR), in blue and offset for clarity.
That, my friends, is a comb filter in action. The good news is it didn’t sound as goofy as it looked, but it was definitely audible. The FIR approach, on the other hand, sounded extremely uniform throughout the crossover region.
Important note: this is, for me, a minority case. Massaging a main/fill crossover with FIR is very low on the priority list when it comes to optimization. It’s rare that I have enough extra time and resources to chase something like this, and although it’s definitely an improvement for the relatively small percentage of the audience in those affected seats, there are generally much bigger fish to fry on a show day. The primary goals of level and spectral variance for the main audience areas are much higher up on the list, and in most venues that’s where the challenges lie.
