Audyssey Labs' MultEQ - page 2
"What does it do if you're measuring dipoles?" I ask. Tom answers, "The system is waveform dependant and since dipoles are not dipolar at 80Hz MultEQ will measure their distance to the audiophile seat the same as it would a monopole." Chris adds that "It looks at the first arrival. The waveform doesn't flip with reflections. You still have a positive peak."
The MultEQ system now sets delays for the prime seat. Why? Other technologies within the MultEQ package require a baseline map of the speaker systems' position with respect to the audiophile seat before the other imbedded technologies can be implemented. And the FIR filters which are utilized can easily be designed to be "linear phase". Such filters, by definition, delay the input signal but don't distort its phase.
The Denon AVR-5805 allows for three subwoofers and each of three optimum crossover frequencies to be determined independently. In finding the best low-pass/high-pass filter set Tom explains they used "a priori knowledge" of what the best splices might be. "For instance, the THX spec calls for a 4th order low-pass filter and a second order high-pass. This is not asymmetrical as one might think if you take into consideration that an acoustic suspension satellite system capable of getting down to 80Hz will have its own second order slope." Tom gives an example. "We say, 'If we get this set of measurements then they must be coming from this speaker'. Does that make sense?" When I say "no" Tom goes on to explain "It is determining what a speaker must be doing versus what the room is doing. Based on the timing (first arrival) information there are very few possibilities from which to choose. So Audyssey fixes both the point and the slope of the sub-sat combo. THX works but you have to use all the ingredient parts and you have to space them in a particular way. That's all you could do at the time. But my gosh, it's been ten years"
Chris then talks about an adjunct application to MultEQ intended for HTIB systems. It's called PrevEQ. "The main problem with home theater-in-a-box systems is the huge hole between the sub and sat. With a HTIB system we would have the luxury of having the speaker systems in advance so we could pre-characterize the speakers and boost the subwoofer to make it go up higher in frequency (to match with the satellites)." Tom and Chris then both clarify that "the filter placed on the content side is 120Hz for Dolby and 80Hz for DTS which is a problem for movie theaters also. Audyssey can be set up in the Denon for the three subs as filter pairs. Each to match with a different satellite set. MultEQ is in fact the only existing system to figure out the distance to the sub for timing information."
"The approach to solving this problem in the past has been based on parametric EQ which is an extension of what was done with analog equalizers just, done digitally. The first problem is that you never have enough bands, typically 10, using an IIR (infinite impulse response) filter. IIR filters allow you to do things in the frequency domain but it does unknown things to the time domain. In many cases it manifests itself in ringing or smearing."
"Our approach is based on FIR filters which in the past have been computationally intensive but this is not an issue any more because the DSP power has increased so dramatically. FIR filters allow us to correct the time domain and frequency domain at the same time. 'Well, you might say, FIR filters don't give you enough resolution if you want to keep them relatively short.' And that's true. This is the reason we implemented Dynamic Frequency Allocation (another of the imbedded technologies) which gives non-linear spacing. So instead of having only 80Hz or so resolution we can get down, at low frequencies (where it matters), to under 5Hz of resolution. It's on a Bark Scale but the resolution starts below 5Hz at the lowest frequencies and goes up to a few tens of Hertz at 20KHz." (The Bark Scale ranges from 1 to 24 barks, corresponding to the first 24 critical bands of hearing. For computing all-pass transformations, it is preferable to optimize the all-pass fit to the inverse of the map, i.e. Barks vs. Hz, so that the mapping error will be measured in Barks versus Hz.)
The conversation now turned to the bottom line technology within MultEQ. The ability to have every seat be a good seat. Again Tom provided his historical perspective from tuning theaters in the early eighties. "While real-time analysis is 'time-blind' (so you have to know something about the time domain before you use it) nevertheless, if you clean it up, it has some advantages over the FFT-based analyzers. The THX R2 (from the eighties) was readily able to do spatial averaging and temporal averaging and we realized if we made an extension of it using a laptop with an add-on spectrum analyzer peripheral that we could send signals across dynamically from the analyzer and do a lot of mathematics to it and therefore clean up the signal."
Chris takes over, "So part 1 was, we knew if you EQ for the single sweet spot then every other position would suffer from much poorer frequency response. (And that was one of the reasons for the bad name 1/3 rd octave equalizers were given.-Tom) Initially Denon and every other potential customer thought 'let's have two modes'. One for a sole listener and one for when you have several listeners in a room. Well, it turns out if you EQ a whole room the audiophile seat gets better. If you take more of the problems of the room into account you're fixing a bigger area than just the audiophile seat so there's no need for two modes."
Chris continues, "The approach other people have taken is to throw DSP at it. There are room correction units on the market that do just that. They can do 8000-tap FIRs and you need 3 DSPs per channel. But if you want to be in a consumer product you have to make some computing decisions. So that was the thinking that went into Audyssey's Dynamic Frequency Allocation.
I then asked "Does it give the same response at each listening location? How is it possible, for instance, if you have a standard D'Apollito-style center channel which is known to have a lobe which points mostly toward the audiophile seat." Chris responds, "By measuring the response at different locations we use a fuzzy-logic based clustering approach which, after computation, makes the sound at the audiophile seat better. The average assigns equal importance to each seat, an importance of 1. Now by applying a weighting factor automatically we use an approach based on pattern recognition. It doesn't have anything to do with what we know about acoustics," Chris stresses. "This is the leap of faith. It is the first application of fuzzy logic that I know of in audio."
"If we were to treat the time domain version of these responses and say which of the criteria are closer to each other as far as pattern similarity, then I find for instance that seats 1, 3 and 5 in the room are "clustered" as far as similarity, seat 2 is by itself and seats 2 and 4 are similarly grouped together."
I interject and ask if the sound the system is reading is mainly direct sound and first order reflections and the answer was "No". "The response that we're taking is quite long. It's 8000 samples over 200 milliseconds. If you look at the time response, it has a pattern. But if the seats have similar problems, they will fall into similar clusters as set up by our pattern recognition method. Where it gets fuzzy is that a particular seat can belong to more than one cluster. In other words, what it says is that based on our theory that seat #2 has 80% of the characteristics of seat #3 but 20% of the characteristics of seat #1. So there are no hard boundaries."
"So now we have six responses which we've clustered into 3 groups. From each response we elect a representative of the cluster. It's not any one (exactly within the cluster), it's one that represents each one in the cluster in the optimal way. That's called a cluster centroid. So now, of the 3 clusters you have, you have 3 representative responses. So you do it again until you finally end up with the "President response" which represents the constituent responses in the optimal way. So the final representative response is the one we take and invert. When we invert we are inverting proportionally and non-linearly."
Tom now breaks off and talks about the early two-channel $12K SigTech box which had been brought to Skywalker Sound when Tom was still working at The Ranch. The SigTech supposedly fixed the first 50 milliseconds of time. Tom had tried an A-B with and without the SigTech on the Skywalker sound stage which he had previously calibrated using the THX R2. The SigTech was inaudible in double-blind testing. This prompts my question " So if a room is properly treated, can the system sound better?" And the answer from Tom, "Y ou can't fix a first reflection " (electronically).