John Bau: Interstellar Overdrive Page 2
Bau: Right. Basically, how we got to the TC-50 was by throwing out all preconceptions, and just trying, number one, to find a transfer function for a crossover network that actually summed to linear amplitude and phase. That was the biggest thing that needed to be solved. I had already come to realize, with the help of the FFT and later the Time Delay Spectrometry (TDS) measurement technique system—which I found to be superior, more revealing—that the possible clue was in the phase information and the timing relationships which were totally mucked up on almost every system out there. Since there was no system that I could go and buy to see if this were really true, I needed to build one to see if it really did result in the ability to, at least in some measure, recreate the sense of space and localization in a more coherent manner. It was a very involved process because I was having to set up a measurement system that was accurate and predictable, and this was before the TDS system could measure phase. That box hadn't showed up yet. In fact, there was no box that you could buy to give you accurate phase information—something that took me several months to discover.
Atkinson: Do you think that part of the traditional engineer's attitude that phase doesn't matter is because there was no easy way of measuring it? Because it's more convenient to push it under the carpet?
Bau: Sure. And any time you look at something and you see that there are just a number of roadblocks if you go in that direction, things to overcome that are either going to be very costly or relatively impossible, then you turn your attention to a way to go around it!
This project was as much for my own edification as anything, though I did hope that it would result in something. I had obtained a fairly large loan at the bank and had a time deadline on it. And when I got to the end of that deadline, I still hadn't found the key in terms of the crossover function. I could now accurately measure phase, but it took me four or five months to get a measurement system rigged where I could actually eliminate all of the nonlinearities of the measurement system with the help of software. For example, there were timing uncertainties with the FFT analyzer. It would get the trigger from the signal generator but there was a variance as to when it was going to start collecting data. The software was sitting there looking for the trigger, so it depended where the microprocessor was in the software loop when it was actually going to send the hardware trigger and start collecting the data.
Atkinson: And then if you just wanted to average the data, you couldn't because the waveform would be in a different part of the time window for each sample...
Bau: Sure, it's jumping all around all the time. Thankfully, the company whose FFT box I had selected was intrigued that I had actually found this problem—they knew about it but, again, had swept it under the rug. You had access to the raw data and you could compute the phase, but they didn't display it for you. So they never thought it would ever get caught. But I was very fortunate, because the guy who had designed the box was still with the company. He got right on it, sent me out new ROMs, and I had already evolved the software to verify that it was OK. Finally, this part works! But then it's something else and I now have to model all the nonlinearities of the microphone and of the preamp and all of the stuff so that they do not affect what I see on the screen.
Atkinson: Did it come as a surprise to find that the tools you wanted to measure with were, in effect, bent? That the rules had unequal gradations?
Bau: It was very frustrating. When you get down to having to define the problems of the very tools that you need to rely on, it really throws you back. But it did create a very positive attitude because it made me want to trust nothing unless I had actually verified it myself, which is really the definition of Scientific Method. However, it meant that I had to forget about sound for months.
When I did get back to sound, I had still not found the key I knew I needed. To me, the first-order crossover systems were unsatisfactory for a two-way speaker system. Even for three-way speaker systems.
Atkinson: Because of the wide overlap between the drive-units?
Bau: Because of the huge overlap region and because of what I think has been described in the literature as the lobing effect. With a first-order crossover system, you actually have more energy in the crossover region bouncing off the floor and then reaching your ears than is coming to you directly from the speaker. When I was playing around with first-order crossovers, we would do experiments where we would put huge piles of cotton on the floor in front of the speaker—when we'd take it away, it was like day and night. Now I don't think there's any psychological way of getting over that. I don't see that it's possible.
And you would be getting a lot of distortion from the bass-mid driver because of the very gentle rolloff into the high frequencies. There were a lot of things that were not desirable.
Anyway, I still hadn't gotten this key, and it was getting down to a couple of weeks before I had to go to the bank and tell them why I couldn't start paying them back. Out of frustration I decided to go to California to spend a few days at the AES Convention there. I was sitting very dejectedly, listening to all these boring papers, when somebody got up and started talking about his experiments with Bessel networks. The more I listened, the more I saw that he had not been able to take it through to completion because he didn't have the tools. He was working with fairly crude hardware and needed a computer with modeling capability and iterative optimization, which we have, to bring it to fruition. He could see the potential but couldn't take it further.