While you may not yet have heard of Bruno Putzeys, it’s likely you will in the coming years. Already, his name is very well established in Audio Engineering Society circles. Based in Belgium, Putzeys graduated magna cum laude from the National Technical School for Radio and Film, in Brussels, and began his career at Philips’s Applied Technologies Lab, where he worked for the better part of a decade. He is responsible for the development of the Universal class-D (UcD) amplifier modules now made by Hypex for industry manufacturers and DIY hobbyists, as well as the new, state-of-the-art NCore class-D amplifier modules featured in cutting-edge audio equipment such as Bel Canto Design’s newly launched Black. Putzeys is also one of several equal partners in Grimm Audio, a company that serves the professional recording industry and makes several well-regarded products designed by Putzeys, including the AD1 DSD analog-to-digital converter and the LS1 studio monitor.
In the second quarter of 2014, Putzeys’s first consumer products will begin shipping via his first consumer venture, Mola-Mola. The line will begin with a monoblock amplifier based on his NCore technology, and a highly configurable analog preamplifier that will include modular slots for the installation of an internal phono stage and, later, an onboard DAC and possibly even bypassable tone controls. Mola-Mola’s third product will be a standalone DAC, projected to ship in early 2015, followed by an integrated amplifier (according to Putzeys, a high priority) and perhaps even a standalone phono stage. Based on my experiences of listening to Mola-Mola prototypes, I think the industry is in for a treat -- the DAC has the potential to rewrite the rules for what’s possible in high-performance digital. It was my pleasure, at the 2014 Consumer Electronics Show, to talk for 90 minutes with Bruno Putzeys as he stood on the verge of much wider recognition, to learn more about his past, his accomplishments and philosophies, and his hopes for Mola-Mola and the industry at large.
Pete Roth: Thanks again for carving out some of your precious CES time for me this morning. When did the audio bug first bite you, and when did you decide to pursue it formally in your studies of electrical engineering?
Bruno Putzeys: The first moment I got seriously interested in audio was when I was about 16. Before that, I was just your typical ’80s computer whiz kid. However, my father was heavily into audio and I sort of followed what he was doing, but I never took part -- until the particular moment when a friend of his came over with some very fancy new amplifier. We listened, but weren’t that impressed. Then some other guy walked in with a really old little tube amp with two EL84s from a defunct brand called Armstrong, and we plugged that in. It just blew this fancy new transistor amp so far out of the water straight away, I thought, “This is curious -- this shouldn’t be allowed. What’s going on?” It was just that moment of curiosity, to have my expectations about something poked, that piqued my interest.
Audio never really appeared as a potential career choice, though, until I was well into my electronics training. Before that, I had been looking at everything. Either I wanted to study computer sciences, or at some other point I thought I may actually go into industrial design or advertising, or what have you. Really flip-flopping all the time. It was only once I started a bachelor’s course [in electrical engineering] that I thought I might specialize in audio, and I pretty much made up my mind, at that point, that for my thesis work I would either try to do something in class-D or work on a discrete D-to-A converter -- it would be one of those two. And I think I’ve stayed the course ever since.
Bruno Putzeys and Pete Roth
PR: Which of them did you choose for your thesis?
BP: It was class-D amplifiers. In those days I was still heavily into the hand-waving side of audiophilia, apart from the technical side. I was suspicious of negative feedback. So that particular class-D amplifier was a power stage optimized for really low open-loop distortion. Effectively, I pulled off the very first proper power D-to-A converter. Like the TACT Millennium Power DAC, but before the TacT Millennium. Never did anything with it commercially, but it did land me a job at Philips.
PR: So that’s how you got from the National Technical School for Radio and Film, in Brussels, to the Philips Applied Technologies Lab, in Leuven, Belgium?
BP: Actually, the thesis was already sponsored by Philips. Well, as you might guess, as I got my degree, Philips said, “Well, we’ve got to keep this guy.” That is not to say that I did any class-D work for at least the next year and a half. When I started with Philips, in 1995, I got dumped into the rat race of designing these shitty black mini-stereos that used to be en vogue in those days -- just making iteration after iteration of these standard Sanyo hybrid-based amplifiers. That was hugely frustrating, as I really wanted to do something with class-D. So all the time I was walking around, talking to management and anyone who would listen, saying, “The future is in class-D!” Believe it or not, for the longest time, they didn’t get it.
I found one guy, though, who had a tiny bit of budget from the television business to let me work on class-D for one month. I said, “OK, this is my make-or-break moment.” In that one month I made a 25W amplifier module that cost about $2 and immediately outperformed something Philips Semiconductor, in those days, had been working on for five years or so. Then we scrambled together a tiny budget for the next phase, and the next phase, and the next phase. That went on for almost two years, at which point the local lab at Leuven, where I was based, realized that it was time to think strategically. Thereafter the whole thing took flight, and I really got the liberty to try out different ideas and different technologies, pitch them to various clients. I must say, I had a great time. I was a bit of a rebel, but very well protected by the management, and ended up doing the most fantastic projects. The one I’m probably best known for, UcD [Universal class-D], was done in 2001.
PR: So, having started in 1995, and after working through 1996 on those little crap boxes, by 2001 you’d been working on class-D for about five years, culminating in UcD?
BP: Yes, exactly. But not just UcD, as there were some amazing things I did in those days. For example, in 1999 I built, again, a zero-feedback class-D amplifier that, literally, directly processed DSD with very good audio performance, especially considering the fact that there was no error correction onboard whatsoever. It still is the best-performing, non-feedback class-D amplifier ever made. Funny, enough, it did not sound as good as the better UcD models I made. So that amplifier, as historic as it is, never made it into a product and is gathering dust in my garage. But just for the technical tour de force, I’m immensely proud of it.
PR: It’s interesting that you’ve now talked twice about zero-feedback class-D amplifiers. I’ve read -- without fully understanding more than a fraction of -- your article on negative feedback, published in Linear Audio magazine. What I took away from the paper is that you either use zero negative feedback and make the best circuit you possibly can, or you otherwise use a whole lot of feedback and take full advantage of the theoretical maximum. Also, I think you made the point that the feedback itself is just one tool in the toolbox, and that to maximize the overall result, you have to start with a fundamentally sound circuit design. Did I get it right?
BP: It’s a correct statement, but that was not the emphasis I wanted to get across. The article, just so people can find it, is called “The F-word.” The subtitle was added later on: “or, why there is no such thing as too much feedback.” Maybe I should step back and explain how it was I got into that direction, which was around the time when I started designing UcD. As I’ve said, the management was quite happy to have me go wild on very out-there, blue-sky schemes. But they did ask me to design something sweet and simple that could see them through while I was doing the more crazy stuff, something they could drop into whatever application that they had for linear amplifiers.
I then remembered that, a couple years before, I had a vision in my head of a power stage that was switching, and of the way that the output filter would swing in response to the switching and then couple back into some other circuitry. The output filter was actually a part of the bit that makes the amplifier oscillate and modulate. And I also had begun to form the suspicion that the sound of feedback has nothing really to do with the fact that there is feedback, but with the fact that, usually, the amount of feedback you get (called the loop gain, technically) typically decreases as you increase in frequency. So an amplifier with a massive amount of feedback at bass frequencies might actually have quite little at high frequencies, because that’s where it gets difficult. What you get is high frequencies that are much more distorted than the low frequencies. Well, maybe that explains this very brittle, thin, choked kind of sound that is typically associated with negative feedback. Maybe the real cause isn’t there being too much feedback at low frequencies, but too little at high frequencies.
So, as I was starting work on UcD, I thought to have a secondary goal, a private one, of trying to build a circuit which has a substantial amount of feedback that is constant all across the audioband, just to see how that hypothesis pans out once I listen. If it hadn’t worked, if it had turned out that it sounded so-so, then it would still be OK, as they would have something to use in those shitty black boxes. But if it did turn out right, then I would have made headway into high-quality class-D, and more importantly, I would have understood more about technical performance and sonics.
And it actually worked. I built prototype after prototype, until the practical circuit behaved as theory predicted, so I wouldn’t have any hidden variables affecting the sound. Only at that point [did I go and listen to it] -- six iterations before listening; talk about patience -- and that thing immediately killed the open-loop DSD amplifier, plus practically any other high-end amplifier I had access to at that time. Certainly there will have been several others that were better, but it really punched way above its weight. That set me thinking that feedback itself is not a sonic evil, but it’s how you do it. As time went on, I realized that feedback is not just one tool in the toolbox, it is the most important tool in the box. Feedback is what keeps jet fighters flying, it’s what keeps nuclear power plants from melting down, it is everywhere. It is one of the most important engineering subjects in electrical-engineering courses. Indeed, I came to realize that feedback had unjustly been given a bad rap in audio, and that you could actually use it to your advantage in terms of subjective sonic result.
Of course, the next question is how to explain that when so many people disagree with that point of view. You can’t just go around saying, “Hey, I’ve made a negative-feedback amplifier that sounds great, so you are all wrong.” You have to accept that, for those people who say they tried feedback and it didn’t sound good, they had real experiences -- they didn’t make it up or start a religion. People have really, honestly heard what they have heard, and what they heard didn’t sound good. So I had to reverse-engineer all these experiments they had been doing and work out exactly what caused that particular subjective sonic experience. The “F-word” article is, in its first part, just a rundown of feedback structures and an attempt to get terminology straight. An interesting observation identified in that first section of the article is that local feedback with a bit of global feedback is, actually, identical to full global feedback -- mathematical fact. The second part looks back at the history to see what, in different scenarios, was responsible for feedback sounding bad in those particular cases. One of those you hinted at is that, if you take a simple amplifier which has acceptable distortion (just a second harmonic is what I use as an example) and you start applying feedback, harmonics will appear that were not there originally. Higher-order harmonics, even and odd, turn up out of the blue. So if you apply a little bit of feedback, the second harmonic that you wanted to reduce drops by a little, but out of the blue you get this whole smattering of high harmonics. It is quite understandable that this doesn’t sound good. That observation has been made and published by various people over the years, but the most important conclusion was never drawn: If you keep increasing feedback, if you turn the feedback knob up and up and up, you quickly hit a point where those distortion products all start coming down again and the signal does start getting cleaner. And if you get to very large amounts of feedback, the result is just supersmooth. So that is why I say that it is normal for an experimenter to experience that if you take a good-sounding zero-feedback amplifier and add 6dB of feedback, the result sounds worse. They heard that right. But had they been in a position to add 60dB, well then, suddenly they would have been confronted with a sound that is little short of magical.
PR: So you were working on UcD, exploring different ideas, identifying hypotheses and subjecting them to scientific experiment, and in doing so you came to the conclusion that feedback is perhaps the most important tool. Maybe we can take a moment to talk about the differences you’ve seen between textbook theory and practical applications. Where is there a disconnect? The one most often cited, at least on the audio forums, relates to cables.
BP: “Textbook theory” is very often just a shortcut. When people say something like “In theory, it should happen like this . . . ,” what they actually mean to say is, “In the very first approximation, on a basic level, this is how it should go.” That’s oversimplification, not theory. Real theory isn’t so simple. It is like you say: in theory, cables shouldn’t make any difference. Well, hang on. Does that imply that you’ve actually looked at all of the established textbook physics that explains exactly what happens within a cable? I don’t mean “new physics,” like microdiodes or what have you, because I do think that’s a load of crock -- but, really, all the things you know happen when you, for instance, intersperse two conductors with a dielectric between them. How will that behave, for instance, when you actually put it up in a listening room and subject it to the vibrations that are caused by the speakers -- the triboelectric effect? Or just ordinary electromagnetic noise pickup from nearby mains cables? All these things are entirely known by physics and fully understood by theory. But the people who say that “in theory” it shouldn’t matter, they just look at one small corner in one particular textbook, where it doesn’t mention all these other things. Usually, where theory and practice deviate, it just means that your theory hasn’t gotten into enough theoretical detail. So far, I have not yet bumped into anything in terms of audible differences that I, or anyone with me, could hear that did not at some point connect with established theory and known physics -- by which I mean ordinary street-level physics, none of your fancy quantum stuff. You really do not need to invent laws of physics from a parallel universe to explain things. And you don’t have to excuse yourself to say that theory does not connect with practice. If you look close enough, you will find [the connection]. If practice and theory seem to deviate, you better have a sharp look at your theory.
PR: There are parallels here to what we were talking about regarding measurements -- that when folks dismiss measurements and the ability to measure subjectively observable aural distinctions, they are simply not taking a full battery of appropriate, available measurements. When I spoke with Paul Barton [of PSB Speakers], he made a point about it being critical what you measure, how you measure, and then applying it against what we know about the ear/brain interface and how sound is perceived in real space. He informed me that he can largely tell how something is going to sound based on his interpretation of the full set of measurements, and that, similar to how Beethoven could write symphonies after he lost his hearing, Barton could continue to design speakers were he to lose his hearing, due to the exhaustive sets of measurements he takes and his 42 years of correlating those measurements with the resulting perceived sound.
BP: I agree. In fact, I very often have to invent new measurements on the fly when I suspect there might be something going on that doesn’t show up clearly on standard measurements. To give one example, you could take a DAC and do something very classical, like sweep the level of a sinusoidal signal from full scale to nothing, and then look to see how distortion changes with signal level. You might find some minuscule squiggles at lower levels and shrug them off as measurement errors, like, “OK, that is just the machine not correctly measuring noise.” But I got suspicious at some point and said, “Hang on, let me try to find explicitly whether something happens in the noise floor with the signal modulation, but then I have to do so without a signal present. How do you do that?” Well, you sweep a DC input to a DAC. You feed it a constant code, some small value, and measure the noise. Increase that code and repeat. Suddenly you’ll find that some of these D-to-A converters will do these frightening things, like the noise floor suddenly shooting up or an audible whistle actually just walking through the audioband as you sweep, going from supersonic down to zero and then back up. You have to be creative when you measure, not just do the standardized battery.
It’s a bit like real science. When you have a scientific hypothesis, what you do in order to make it stick is try everything you can to push it over, and only if it doesn’t topple, then it might be correct. Same thing in audio -- if your hypothesis is that this particular D-to-A converter is a really good converter, then you’ll test it, and you’ll measure, and you’re actively going to look for bad news. If, after not finding any bad news despite how hard you’ve tried, then you can start to say this is probably good. That simple example shows that, very often, the standard measurements might simply cover something up that is very measurable and very glaring, but just happens to be under the radar of regular test methods. You might then have this DAC which listeners feel has a “shine,” that puts an unnatural shine everywhere, and then suddenly you find this noise modulation when digging deeper, and you realize that might be the answer for why.
It is my experience -- confirmed by every new thing I do -- that when you get into really high measured performance, really low distortion, superlow noise, then the ultimate subjective sound quality starts improving and continues to improve in step with the measurements. At some point you will find that a product that measures absolutely perfectly under an extensive battery of tests will sound a lot better than a product with more typical high-end audio performance that has been tuned by ear for years. The upshot is that measurements do matter. The way you should translate this into a development process is not to listen and tweak your circuits, but rather to measure and adjust, measure and adjust, and then listen. If, at that point, something sounds off or is not quite working, that tells you something about what and how you measure. You calibrate by ear your set of measurements and the methods by which you measure, but you optimize your circuit by measurement. That is much more logical. You should take science to the absolute limit and crosscheck your scientific, technical procedures with what you are hearing, to make sure you’re not forgetting anything. The purely technical road in the end will yield a circuit that really sounds better than what you can get by mere philosophy and tuning parts. Start shooting for fantastically low distortion, fantastically low noise.
PR: Tell us a little bit about your transition from Philips and your work with UcD there, to Hypex.
BP: Philips is a great incubator of new technologies. They used to have a lab called “Nat.Lab,” just south of Eindhoven, and they had places like Leuven, where all sorts of exciting things were happening. But they are not at all good at turning a profit from all the fantastic technology that they have. So the Netherlands is littered with ex-Philips people who do really exciting audio things that they carried away from their work at Philips. They usually stay about ten years at Philips, and then they become disenchanted by how they never actually see their work come to fruition. The same thing happened with me. During my time there, Philips never actually used UcD in their own products. We only sold some licenses to a handful of firms. Yamaha had a license. The Löwe television company had a license. And then it sort of petered out.
There was also this tiny two-person company in Groningen, Netherlands, called Hypex, that had bought a license. The owner of Hypex, Jan-Peter van Amerongen -- he’s a pretty technically savvy guy himself, although he tends to downplay that -- I sent him some schematics and an example board layout, we talked on the phone and exchanged some e-mails. Then, two months later -- bang -- he had a perfectly working UcD amplifier which he then proceeded to commercialize. Shortly after that, when I finally decided it was time to widen my horizons, Jan-Peter said, “Come work for us.” That was 2005.
PR: At Hypex, obviously, you continued development and implementation of the UcD modules for OEM clients -- but tell me about the much newer NCore class-D devices. Is NCore a further extension of what you’ve been doing with UcD, or is it an entirely different class-D scheme?
BP: I think it would be fair to say that it builds on from UcD. Certainly the fundamental math is the same. The really crucial part of NCore was to figure out how to improve the loop gain even further from what we had -- UcD had substantial loop gain up to 20kHz of 35dB or something -- and I wanted to go beyond that. As I said, there is no such thing as too much feedback, so I was looking for a way to add 20 more dB. That’s actually a very hard problem to crack, because once you start doing that, you have to remember that a class-D amplifier has a limited bandwidth. A reasonable switching frequency for a class-D amplifier is just under 500kHz or so. If you go much above that, you run into efficiency and headroom problems. That, in turn, implies that you have no more than some 200kHz of bandwidth to play with -- actually, less than that. And if you want to cram 50 or 60dB of audioband loop gain into that bandwidth, you have to think completely differently from the way that linear amplifiers are usually designed.
Linear amplifiers typically have what we call single-pole compensation; some of them have two-pole compensation, but nothing much beyond that. UcD has four-pole compensation, and NCore has five. Once you reach into the four- and five-pole compensation, you have this problem that the amplifier can be operating in perfect stability until you clip it, and then it will suddenly start oscillating at a frequency that will immediately damage the amplifier and the loudspeaker -- so you want to avoid it like the plague. And not just that, but you actually want to return the amplifier, once it comes out of clip, to its normal operating regime so quickly that you don’t hear any glitches. The whole NCore patent revolves around the practical solution to that stability problem, the way that it actually catches the feedback loop at the moment that it’s thinking of going unstable, and then lets it go when it is safe to do so.
Apart from that, of course, I did learn some more tricks as concerns driving the output stage. If you have 50dB of feedback, and you are aiming for -100dB of distortion, you’ll still need to manage to get -50dB of open-loop distortion. In that respect, you are right in your previous question: You have to start out with something that’s good, because there is always a trade-off. If you’ve got yourself 50 or 60dB of negative feedback, but if you can also get 10dB improvement in open-loop distortion, why not do so? So the actual power circuit has also changed between UcD and NCore, and obviously, then, the later UcD modules actually use an improved power stage that was borrowed from NCore. The core of NCore is the feedback circuit, but the actual product contains some more improvements that are now trickling back into the UcD range.
PR: How do switching power supplies, aka switch-mode power supplies, and linear power supplies fit into all of this? One of the arguments against switching power supplies and class-D amplification centers on the high clock speeds, the high switching frequencies, and the impact that they have on either the power supply being modulated, or the line noise that affects the system’s electrical infrastructure. Perhaps a brief tutorial . . . ?
BP: Switch-mode power supplies and switching amplifiers are, in fact, very closely linked. In both, the power devices ideally operate either fully on or fully off, and they alternate very rapidly between those states. By and large, when it comes to unwanted behavior of the circuit (you were referring to the noise, with the more generic term being EMI -- electromagnetic interference), most of these things happen in that space between turning a device off and turning another one on. This happens really rapidly, and whenever you have rapid changes in voltage and current, every bit of wire will turn into a transmitting antenna. You really have to think extremely hard about how fast you really want to switch -- can I reduce the problem at the source? -- and then look at the conduit by which the noise gets out: the wiring and the printed circuit board. Everything that is left after that -- even after you’ve done your best to make the switching power stage as quiet as you can as far as radio-frequency interference [RFI] is concerned, and you’ve done your best to make sure there is no internal wiring that actually radiates this stuff -- in the end you might still need to add some further filtering to the incoming and outgoing lines. And you really need to be extremely careful with these things, because a lot of audio equipment is astonishingly sensitive, sonically, to any of this high-frequency noise circulating around power cables, speaker cables, and interconnects.
As a reference, I keep a CD player at home which sounds quite good, but which is fairly typical in its sensitivity to this noise, so I can do listening tests on power amplifiers and switching-mode power supplies to judge whether the power supplies are quiet enough not to disturb the connected equipment. As a rule of thumb, for something that is directly connected to a source, like a preamplifier, the power supply ought to have a measured EMI performance that is at least 20dB, and preferably better than 30dB, quieter than the legal FCC requirements. Simply designing a power supply that complies with FCC class-B is not good enough for good sound quality -- it actually has to drop off the scale of the measuring instrument entirely. Then you can be quite sure that whatever source you are going to connect, it is not going to be bothered by this RFI.
I should add that in the pro world they’ve learned to work around this. Designers of pro-audio gear nowadays make a habit of constructing the balanced interface in accordance with a set of rules called AES48, and those actually immunize equipment to an extreme degree. So if you have a complete audio system that is constructed as per AES48, then you could have some things sitting in the signal chain -- like a television or computer that produces staggering amounts of EMI -- and still get good sound. Mola-Mola products follow AES48, but you can’t be sure of any of the other kit people have in their rack, so I’ve got to make sure my gear plays ball with the rest. Many of these compatibility issues -- where people say this preamp sounds good with that CD player -- some of these mysterious interactions actually happen through the power wiring, and sometimes even through direct coupling from a power cable into a speaker cable. This, then, does to an extent explain why people put such an inordinate effort into speaker cables, and cables in general. All I can say is that once you realize you are looking at these crosstalk issues as one of the reasons for a sonic difference, you might want to do some more targeted experimenting, because I believe most cable manufacturers just construct things and listen to them without really understanding what’s behind them.
PR: In a recent Mola-Mola blog entry about your DAC, you wrote about a modification to the power supply from a linear supply to a switch-mode power supply, presumably to limit the power consumption of the DAC. You wrote that you initially heard both benefits and drawbacks, but then, after applying a cap-multiplier circuit, the result was nothing but positives. What was going on?
BP: EMI is, of course, one very big and underestimated part of the sound of a power supply. But obviously, just plainly, the quality of the DC itself matters just as much. The quality of the DC of that experimental switch-mode power supply was quite good, but initially not as quiet, not as good as the linear power supply we were trying to replace. A C-multiplier circuit is essentially a pass transistor, sort of like a linear regulator, that uses a capacitor as its reference voltage so that it just hangs, like, one diode drop underneath the input, and it filters out the ripple. I didn’t need the rail to be strictly regulated, but I needed it to be fairly quiet. The added circuit simply returned the sound quality to that of the linear supply. Before that, however, I had huge fun sorting out the EMI. As I said before, it was quite good to begin with, but I really had to work hard to drop it down to this -30dB level, below FCC legal requirements, before it started to sound the way that it should. What it shows, however, is that while most people only remember to measure the DC output and forget about the EMI, clearly I did the reverse.
PR: This is probably a good place to pause. In Part Two, we can delve into your work at Grimm Audio, the launch of Mola-Mola, and your exciting new DAC.
. . . Peter Roth