Doing it by numbers.
Over the past couple of months, I have looked at why sound systems often go so wrong and how, just by listening, you can often tell what the problem is. Optimizing the performance of a sound system can be a repetitive (iterative) and time-consuming task as one goes around a space tweaking equalizers, individual cabinets or arrays.
It never ceases to amaze me how adjusting one thing seems to affect something else. Equally, there are times when, after an hour of expert tweaking, you stand back, take a dispassionate listen and decide to start all over again!
Several years ago, an audio DSP manufacturer contacted me and asked if they could join me on my next commissioning session to see what I did, because they felt that they could automate the whole EQ process. Great, I thought; if that would save me some time, it could be the best invention next to sliced toast. I forget exactly which venue that the nominated boffin accompanied me to, but I clearly recall that, after about two hours into setting up the system, he came over to say that he was off home because, having watched the process, there was no way he could replicate what I was doing—even using his most sophisticated DSP! I was not altogether surprised, but the idea of automating the process has always been in the back of my mind.
However, the first step would be to undertake a methodical analysis of the whole process. The introduction of line array loudspeakers has simplified this analysis considerably because the process becomes much more definable and manageable. Clearly, however, a computer would be needed not only to control the DSP but the entire process. This, in turn, would mean setting and defining performance parameters and permissible variations.
A characteristic of computers is that they are very good at repetitive (or iterative) tasks. They don’t “need to know” what they are doing…but give them a target to achieve and a way at getting there, and it doesn’t matter if they try out thousands of wrong solutions; they will get there eventually. (Just as theoretically, if you were to equip an infinite number of monkeys with typewriters, they would eventually type out the complete works of Shakespeare, although there are going to be some spectacular failures and near misses on the way: “To be or not to be, that is the gnorganshonoboplatz”?).
The rapid take off of digitally steered line arrays has brought some of the best minds in the business together, to both acoustically conceive of and to steer the devices. This has required loudspeaker designers, DSP engineers/programmers and those versed in abstract mathematics, rather than dBs, to work closely together. For those of you not versed with early array aiming and optimizing, it was (and still is, in some cases) rather akin to shooting in the dark. You know that you have something you want to aim at but, until you press “go” for the first time, you have little idea of exactly what is going to come out of the array or where it is going to go.
It seemed to me that this was an obvious case for a computer to assist with. Given a target area to cover and some information about the array, even a fairly simple program can improve the productivity of the array aimer dramatically. Most digitally steerable array products have an associated aiming program, but those from the likes of Duran Audio (recently “Harmanized”) and Tannoy, for example, take this to another level by allowing targets to be set or areas to be specified where sound radiation should be minimized, such as the rear wall of a space or the stage or ceiling, etc.
Martin Audio’s MLA software for controlling its digitally based music line array took this to a new level and has been successfully employed not only to optimize the performance of the array, but also to steer sound away from reflective surfaces in venues or away from noise-sensitive areas when employed outside at festivals and concerts, etc. The system works by drawing an outline of the venue, inserting an array and then determining the desired level variation across the audience and where you don’t want the sound to go.
A series of receptors (test points) is then created throughout the seating and in the areas where you want the sound to be attenuated or restricted. Figure 1 shows an example of the process, illustrating a longitudinal profile of an auditorium with a simple line array (shown by white outline above the stage). The green line shows the seating areas, whereas the red points indicate the room boundaries where sound has to be attenuated; the blue area shows the stage, where it is desired to strongly reduce the sound from the array.
The top left view shows a sample point in the seating (indicated by the white asterisk), while the top right window shows the corresponding response and sound level. The middle pane shows a sample point toward the upper part of the rear wall, and the sound level plot opposite shows the sound level at that point (red curve). The seating response is shown by the yellow dotted curve for comparison purposes.
From the curves, it can be seen that the sound striking the rear wall at this point has been attenuated by about 15dB from approximately 500Hz to 600Hz upwards, though at bass and lower mid frequencies, the attenuation is lower at around 5dB. (Greater attenuation could probably have been achieved if the “importance factor” allocated to this surface been increased). This was the case for the stage, where greater attenuation was targeted, with the result being shown in the bottom pane. Here, it can be seen that more than 15dB of attenuation has been achieved from 250Hz upwards, with a respectable 10dB being achieved at 200Hz. However, below 100Hz, there is a slight gain in level on the stage, though that should be relatively straightforward to deal with in most applications. (Maybe MLA should stand for Miracle Level Adjustment?)
Apart from optimizing the sound level distribution within the venue, the Martin Audio software also enables the frequency response of the system to be optimized by manipulating the EQ and signal level applied to each element of the array.
Figure 2 illustrates this process in operation. The top plot shows the raw frequency response of the system, prior to optimization, viewed (predicted) using multiple points within the seating area. The center pane shows the EQ optimization process when about halfway through. I stopped the process at about 1500Hz, so the left side of the plot is showing the optimized balance of response and levels across the seating, while the right side is showing the higher frequencies that have yet to be optimized. The bottom pane shows the completed process. The improvement, not only in the frequency of the response but also in the reduction of spatial variations, can be seen clearly.
To achieve this improvement would take a skilled operator using multiple measurement microphones located around the venue (with each connected to an analyzer with equally multiple measurement channels) several hours to achieve.
But this is the beauty of using computers to do it for you. They keep on trying until they achieve the target responses and variations set (assuming that these are physically achievable…an “anti blue smoke feature” should always be incorporated into the program).
Next month I will throw some more numbers into the optimization pot and see what other approaches are doing and achieving.