Kevin Voecks: Loudspeakers, Crossovers, & Rooms Page 2
Voecks: The conceptual process is the part I like—by far—the most, because it's the really creative part, and comes pretty easily. You usually say that you need a speaker in a certain price range—that's usually the direction that you start in—then you start to think about what kind of things could be done for that price. For instance—at $2000 for the CIII, what can we do to make it demonstrably better and worth twice the price from our model below it, the Type E? You would think about what things you could optimize—more dynamic range, lower extension in the bass, refined midrange, particularly the off-axis midrange response and power response, so that pretty much defines that it has got to be a three-way system with a woofer bigger than 8" so that you can get some low end and some power-handling capability. When you're at that point, you start looking at potential drivers and designing woofers.
Norton: You're not particularly tied to any one concept of woofer enclosure—I notice some of your designs are sealed and some are ported.
Voecks: I definitely don't believe that one is inherently superior. With our small speakers a bass-reflex alignment wouldn't make sense. Given the amount of money you're going to spend on the woofer magnet, the size of the cabinet, and all of that, it becomes very clear when you do the simulations that it needs to be a sealed enclosure.
Norton: Most of your larger designs do tend to be ported, though, except for the A. I'd like to pursue your concept of crossover design. There's always been a difference of opinion between designers as to slow-slope vs fast-slope. You tend to go more for faster rolloffs on your slopes, generally.
Voecks: That's true. Actually, if you wouldn't mind, I'll go back a step and discuss something that I think is really important about the way we look at the whole design process. That is, we look at it in an objective, methodical sort of way. For instance, let's say some manufacturer comes out with a new loudspeaker that has a new x factor, and that they have done everything they can to optimize that, often at the expense of other characteristics—believing it [the x factor] is important. Maybe the speaker sounds good, and gets a good review, and the reviewer concludes, somewhat logically, "It's the x factor in the speaker that has made it sound good, so that must be an important characteristic." Other high-end manufacturers, usually pretty small companies, look at that and say, "Wow, these guys have come up with this new thing that makes a difference." And they may start designing toward that end, again failing to optimize other parameters that are not in vogue at the moment.
At Snell, we intend to completely avoid that situation. We look at everything, whether it be parts quality—for crossover components, say, iron-core vs air-core inductors, mylar vs electrolytic capacitors—or whether it's something like a high-order filter vs a low-order filter, and try and determine what is optimum without having assumptions beforehand, which is really the way to go. In terms of the high-order filters, back in my Mirage days, we used first-order filters on the original line, believing that improving phase response must improve the sound. I had done some crude experiments (where unfortunately I had had to alter more than one parameter at a time) that seemed to indicate that, yes indeed, this made an improvement. Of course, when you hear a difference you usually believe that it's an improvement, when you want it to be.
Then Stanley Lipshitz, at the University of Waterloo, wrote a paper saying that he had developed a device that would allow you to alter phase without altering amplitude—usually in minimum-phase speakers the two are linked together. In this paper he said—to paraphrase generally—that in a room with music used as a program source, phase really wasn't very audible. When it was, it was basically in the midbass—and then it was hard to say which was "better." That sounded like heresy to me, so Stanley was kind enough to let us use the test equipment that he had developed. We reproduced the test, using Type As and also using electrostatic speakers and state-of-the-art analog equipment for that time [pre-CD], and came to the very same conclusions. We decided that giving away all of the other potential advantages of high-order crossovers in order to chase after the proper, flat phase response had been a mistake. At that point we went to a multi-slope speaker, where it started out being first-order, then, as it got an octave away from the crossover into the stop band, the rate of rolloff would increase. That gives you really demonstrable improvements; the power handling goes up, the distortion goes down, and the interference between the drivers is reduced because the range where two of them are reproducing the same frequency range is minimized.
Norton: You don't have the lobing problems (footnote 3), even though they're in the vertical plane in most loudspeakers, if designed right.
Voecks: That's right. Those were differences that were definitely real. It also allowed us to get the amplitude response much flatter. Peter Snell had clearly designed for ultra-flat frequency response; his speakers were always the flattest that were out there. It seems that as more research is done, more and more of his ideas are proven to be true.
Norton: When you start a design, you probably pick what you feel to be representative drivers that you're interested in working with. Sometimes you use manufacturers' standard drivers, sometimes you have certain modifications made. I know you've said that on your new tweeter you did quite a lot of work with Vifa to get what you wanted. How does that process normally work?
Voecks: You have a choice of designing and building your own drivers or buying somebody else's. We believe that it makes more sense for us to be buying somebody else's. First of all, because that's their area of expertise and they can do it better than I think we could do it. But also, importantly, if you build your own drivers you can't build them to such a QC standard that they're all perfect—it can't be done. One turn of wire on a tweeter voice-coil, for instance, will make a significant difference. So if you decide to build your own, you have to decide what to do with the rejects. Some manufacturers use the same driver throughout a line, and put the rejects in the less expensive models. Other manufacturers sell those rejects as their OEM drivers for hobbyists. I don't feel either of those is really a very acceptable solution. What we would rather do is get someone to build drivers as we like them, and then have that driver manufacturer make them available to everybody so that our rejects can go to people who are less particular.
I think this is an incredibly important point about Snell—we all know that tiny amplitude response differences are very, very critical; a tenth of a dB in level change over a third of an octave or more is something that would be demonstrably audible in blind listening tests. So if you think about most drivers being manufactured to a tolerance of, say, ±4dB, typically, it's ridiculous. So we first screen the drivers to make them fit within a particular window. Then we adjust the crossover components so that a specific driver's characteristics, when used with this individually tuned crossover system, will result in amplitude response within ±0.5dB of the original reference speaker that made. In practice we approach more like ±0.25dB, but what we're guaranteeing is 0.5dB.
There are a couple of rare cases where manufacturers make matched pairs of speakers—usually computer selection of drivers that are paired—but that is actually an admission that every pair of their speakers sounds different—I can't see that a customer is getting a very good deal if they have no idea how their speakers compare to what they're supposed to sound like. However, we can say that this speaker sounds like the pair that the Stereophile reviewer just heard and like the pair that we played at the show and like the pair in the dealer's showroom. That's a central point.
Norton: You have a fairly high reject rate on your drivers, then?
Voecks: It depends on the driver, because we can make quite a bit of difference by altering the crossover components. We actually adjust the turns in the inductors, tweak capacitor values and adjust resistors inside the crossover (footnote 4).
Actually, the way it works in driver selection, typically, is—let's say in a three-way speaker—I use a simulation program where we can go through what used to take months of trial and error or designing drivers and testing them in various enclosures, all on the computer. Low-frequency response is what we have the best handle on, in terms of being able to model it really accurately. So I can come up with a set of specs for drivers, and can give that to various people that we would like to, potentially, buy drivers from and say, "can you build this?" or "do you have something that's like this already?" Oftentimes, with woofers, there's something that's stock that works very well, and we would then go in that direction. Otherwise they build it for us. Woofers—particularly in a three-way system, where a woofer is only handling wavelengths much greater than its size—don't vary much from one to another. If you're talking about using a woofer in a two-way system, then the high-frequency characteristics of the woofer vary from one to another in a given production run and are very critical. The next step would be the midrange and tweeter.
Typically what I do is get it down to 10 or 20 driver finalists in doing the testing here—quasi-anechoic response using a maximum length sequence analyzer—then go to the National Research Council, which has, literally, the most advanced loudspeaker design facilities in the world. I do the final measurements there, which will include the cabinet effects, and I can thoroughly analyze it and decide which ones are the finalists.
Norton: I believe Dr. Floyd Toole was involved with the development of many of these advanced measurement techniques at the Canadian NRC.
Voecks: He's been working for about 12 years on determining the correlation between measurements and sound quality in speakers. He doesn't have any particular axe to grind because he's not affiliated with any commercial venture. We have independently verified most of his results as really being accurate...it's very powerful to be able to make a set of measurements that you can say for sure that, if the speaker does poorly in those measurements, it's not a good speaker. Unfortunately, we're not to the point where if it does well in those measurements, we can definitely say it's a good speaker. While double-blind listening tests usually minimize or eliminate the differences between pieces of electronics, with speakers they make those differences seem larger.
They have a facility at the NRC—it's the IEC prototype, standard listening room. When you think you're done with a speaker design, a technician will take that prototype into the room and four other speakers. You don't know what the four others are, at all. The speakers are hidden from view. They're randomly attached to five numbers. The volume levels are all equalized by measurements of pink noise. You sit and listen at any length. You don't have to make rapid switches, you can listen to whole cuts per speaker—any way you like, any way that makes it easier for you. You write up your responses: "Speaker #1 sounded bass-heavy." That sort of thing. It's interesting that, as I said, it seems to make the differences even greater than they are, and you usually end up tearing the speakers apart terribly.
In the case of the CII, for instance, its measurements looked good. We went into the listening room, and I ravaged the speaker—then went back to the drawing board. We wait until our speaker comes out demonstrably better—to myself and to a listening panel—than the competitors' speakers that are also behind the screen before we come out with the speaker. So we really know that, objectively, without having any egos involved, that our speaker is the best at the price point (footnote 5).
Footnote 3: Additive and subtractive interference in the axial response when two or more drivers are reproducing the same frequency.
Footnote 4: I once inspected the crossover of an older Snell Type A/II—I owned a pair for several years—after disassembling it according to Peter Snell's directions to locate a rattle inside (the grille is nailed in place on the Type A). I observed the use of multiple parallel capacitors to make up a required value and at least one tapped resistor, as I recall.
Footnote 5: Technically, of course, this refers to the listener's preferences under the conditions of the test and among those speakers with which it was compared. I have no reason to doubt that the tests at the NRC are among the most thoroughly researched in the world (if not the most thoroughly researched). But, like everything else in life, not all will agree with the NRC's findings. For those interested in reading about Dr. Toole's work in loudspeaker assessment, a bound edition reprinting his earlier papers is available from the National Research Council, Division of Physics, Ottawa, Ontario K1A 0R6, Canada.