Bruno PutzeysIn Part One of my discussion with Bruno Putzeys, we explored some early turning points in his audio education and career, his exciting period of blue-sky experimentation and commercial-product design at Philips, and his move to Hypex in 2005, where he expanded the boundaries of class-D amplification with the development of NCore. In Part Two, the conversation returns to 2005 and the formation of Grimm Audio, then brings us up to date with the launch of Mola-Mola, the products shipping at the start of the second quarter of 2014, and the groundbreaking Mola-Mola DAC, which will follow in 2015.
Pete Roth: Looping back in time a bit -- feedback pun intended -- did your leaving for Hypex coincide with the work you began doing with Grimm Audio?
Bruno Putzeys: More or less. Grimm essentially was started because I’ve always been interested in A-to-D and D-to-A converters, back from my days of deciding on my thesis. Around 2000 or 2001, I proceeded to build a discrete A/D-converter circuit, which at that point I was hoping to sell as a design to some professional audio manufacturers. It was set up as a 1-bit front end with a decimating back end done in DSP, but I never got to the latter because the front end happened to operate at exactly the same format as DSD. In those days, we had the first wave of DSD. So I tried to sell that, but I was not really much of a sales guy, and so I failed.
In 2004, I decided the heck with it, I’ll just present it at the [Audio Engineering Society conference] and see if it catches anyone’s attention. So I did, and as I walked out of the conference I was stopped at the door by a couple of Dutchmen who said they were looking for something like my design. “We are not really a company at this point,” they said, “but want to use this for our own recordings. What do you say? Will you develop this for us?” They looked like nice enough guys, so I said, “How about just joining hands here and seeing what we can do together?” And that was the start of Grimm Audio -- not that the name of the venture was, at that point, clear. One of the guys is named Grimm, but we ultimately chose that name because we felt the sound had a good ring to it. We started the company as four equal partners, and our first product was the Grimm Audio AD1, an all-discrete, DSD analog-to-digital converter based on that circuit.
I don’t ascribe any mystical properties to 1-bit sigma-delta. It’s just a way of getting into and out of the digital domain. I’ve always felt that it is perfectly possible to get the exact same sound quality using PCM, but for the actual A/D conversion process, 1-bit sigma-delta is an excellent choice if you want to build a discrete converter. It is not that good a choice, however, if you want to make a monolithic A/D converter on a chip, but for a discrete circuit it works fine. So I had always planned to make a decimating back end for the AD1, but that never materialized.
PR: What do you mean by a decimating back end?
BP: A decimating back end is something that turns DSD or some other high-speed, low-bit format into PCM. In the meantime, I’ve really worked out what it is in a decimator and in digital filters that causes the typical PCM sound, and all of these issues can be fixed, meaning you can do PCM with all of the glorious sonic characteristics of DSD preserved. It is perfectly possible. But at some point loudspeakers caught my fancy, and while the AD1 spawned a new product in the CC1 (essentially just the clock circuit used in the AD1), the AD1 itself was left to gather a bit of dust. I think we only sold a handful of these products, although they did land in the hands of some of the best sound engineers on the whole planet -- like Jared Sacks of Channel Classics, for instance.
So I decided to look at loudspeakers, and created the Grimm LS1, which, as you can imagine, is for a completely different market from Mola-Mola. After the AD1, I did feel that I could very much outstrip the performance of available D/A converters by moving to pulse-width modulation. As you can imagine from my background with class-D amplification, I know a thing or two about pulse-width modulation. I was looking for an excuse to build a DAC or an A/D converter based on that for years, and then we [with Jan-Peter van Amerongen] started Mola-Mola, and I said, “If this isn’t the right excuse to do it, then I don’t know what is.”
PR: Now is a great time to delve into your Mola-Mola DAC. One of the things I’m hoping you can explain is the difference between a standalone digital-to-analog converter, or DAC -- that is, the separate component you feed a digital signal to and get an analog signal out of -- and the similarly named sub-component within, which in most cases is one or more DAC chips. I often feel that the two are confused, with one camp saying that whatever component uses such-and-such a chip will always sound the same, but I don’t believe that to be true.
BP: I’d like to dispense with the two camps first. To be awfully frank, I think that if you can’t judge whether a product is good without taking the lid off, it’s probably not going to help to take the lid off and see what it does inside. What I mean is that you ought to be able to characterize the quality of a product like a DAC by a) listening to it, and b) measuring it, without having any idea of how it is constructed inside. It gets right up my nose when audiophiles decide whether they even want to listen to a product based on something they know that’s inside the box. That is really irrelevant from an end-user point of view. A DAC takes a digitally encoded electrical signal on one end, and outputs an analog signal on the other end. How well it does that job should suffice to work out whether you like the product or not. Building a product that is good is the designer’s responsibility, and anything goes, so long as the performance in the end is good. I very often see the reverse happening, with designers specifying up front what kind of technology they will want to use -- or, worse, that they think is sellable -- and what kind of circuit they’ll press into service. Then, having taken that as their overarching philosophy, they live with whatever result comes out. That is quite wrongheaded. As our mutual friend Philip O’Hanlon [of distributor On a Higher Note] says, “If you can do it with a stick and a ball of twine, then hey, by all means!” But to go to your question: Given a particular DAC chip, its inherent performance places a limit on how good the end product can get, although you can mollycoddle some of the weirder artifacts out, which some people are doing, with good success.
Bruno and Pete Roth
PR: Yet your forthcoming Mola-Mola DAC will not use an off-the-shelf chipset, isn’t that right?
BP: Yes, that is correct. But not for some deep philosophical reason. For instance, for the Grimm LS1 speaker, I was perfectly happy to use a D/A chip that was suited to the job, one that sounds good and measures well. On top of that, it’s remarkably affordable too. So again, I don’t really care what tools you use, so long as the result is there. The only reason why I took to discrete in the case of the Mola-Mola DAC is because I wanted to get out of that rat race. Like, one day everyone’s choice was a certain TI or Burr-Brown chip, then afterward it was suddenly Cirrus Logic, and then Analog Devices, and then TI again. Now it’s Sabre. Whenever some company announces a new chip, everyone has to scurry to get a new product out in order not to be seen as outdated. If I don’t want to spend my days building applications for the latest chip du jour, I think I ought to build a circuit that simply outperforms the current best by such a large margin that you can extrapolate from historic data that ICs (DAC chips, I mean) won’t reach this level of performance within, say, 10 or 15 years. So at least for 10 or 15 years, I can sleep soundly, knowing that the DAC I did way back in 2014 is still the go-to product. That’s really the idea, because I do feel the current crop of good chips is exceedingly useful, and can be turned into really nice end-user products.
Even DAC chips that are currently considered outdated are astonishingly good if you just apply them correctly. You should go and listen to this product [the Manley Labs Heart Monitor Controller] that Dave Collins has made. Dave Collins is a recording engineer and EveAnna Manley’s [of Manley Labs] husband. What he’s done is to take a classic chip [the Analog Devices AD1955 DAC] and just worked out some of its nasties, in particular the DC problem that I mentioned, and he did some funky tricks with adding and subtracting noise on either end of the signal chain, and is getting brilliant low-level performance. He decided not to try his hand at a discrete DAC, but just to work the problems at hand, and has very gratifying results.
There are so many variables when you buy a chip. Usually, people will just feed the chip a digital signal, and let it do all the standard processing it has built into the chip, and just put it out. In that case, what will determine the sound quality, apart from the chip, will be things like the quality of the power supply for the chip, the quality of the clock, and the quality of the necessary output filter and buffering circuitry that need to be added -- or the I/V converter, in the case of current-output DAC chips. I think, especially when you work with current-output DACs, there are several ways of deviating from what the chip manufacturer proposes as recommended application circuits, and some of these ways are better, although some are worse.
PR: From the reports on your blog about your Mola-Mola DAC -- which, you say, offers performance that chips won’t get to for 10 to 15 years -- there are a lot of interesting and seemingly unique applications for it. You hinted at some elements, namely the pulse-width modulation (PWM) scheme, years ago in a white paper on the Hypex website. Please give us a brief rundown of what you’re doing in this DAC.
BP: Actually, I think every part has something unique about it. The seed for the whole project was a pulse-width modulation scheme that I did in 2004. I had done it then just for fun, because a lot of people in those days were doing open-loop class-D and were spending huge amounts of research effort in designing a low-distortion digital PW modulator, as though that would bring them closer to a good amp. Then one of my colleagues who was walking around the lab at Philips came up to me and said, “Philips is having this internal DSP conference, and we want to represent our lab in Leuven with as many papers as we can. Can you work something out and present it?” So I was thinking I should probably just make a better PW modulator, just for giggles. I sat down behind the simulator, and something like a half hour later I had [a PW modulator] that was completely distortion free. It outdid everything that all of the guys had been writing papers for the previous 20 years did, so I figured it would do for an internal presentation. Then I presented it to the AES as well.
Again, in my slightly rebellious way, I put in two papers at the same session about class-D amplifiers. The first paper was one that figured in the whole feedback story, and was called “All Amplifiers Are Analog, But Some Are More Analog than Others.” It explained why it was such a fundamentally stupid idea -- and a waste of one’s time -- to try to make an ideal PWM digitally, and then just blindly try to drive a power stage with it. It really made that argument very strongly. The second paper basically said, for those cranks out there who still want to make the best digital PWM, here’s how to do it. By then, of course, I did realize that while digitally generated PWM is not good for an amplifier, it is great for a DAC. I wasn’t the first in this regard. Peter Craven had been hired by B&W, in the early ’90s I think, to build a better DAC. That thing never saw the light of day, but he did invent a pulse-width modulator which had pretty good performance, but which was fiendishly complicated and was later snatched up by Cirrus Logic to be used in their class-D line (open loop: as you can imagine, it didn’t work out as a commercial success, but was spun off to one company after another until it spun out of existence). Nevertheless, you see that this crossover between PWM for DACs and PWM for class-D is a historical one.
So I felt that I really wanted to harness the power of this PWM scheme in a DAC, because the theoretically available performance is staggering. I lived for a very long time, as I’ve said, just looking for the excuse, and am now pressing it into service in this DAC. I’ve combined it with another scheme, which is not altogether a novelty, as it’s been done before (especially in integrated circuits), where you run the modulator output through a shift register and add all the taps together. Well, they don’t usually take a PWM signal, but this trick combined with PWM is really the perfect combination. Consider a digital PWM signal as a string of ones followed by a string of naughts. For instance, if you have a large positive signal, you have a long string of ones followed by a very short string of naughts; and if you have a large negative signal, you have a short string of ones followed by a long string of naughts.
Bruno with a prototype DAC board
In my DAC, the total number of every such cycle is 32 samples. So the PWM signal cycle repeats at the rate of 3.125MHz, and in that interval you have 32 samples, so the bit-clock rate is 100MHz. So there’s this 32-stage shift register that clocks in this data at a corresponding rate of 100MHz, and each of those stages sees the full PWM signal. And you add the 32 outputs together with 32 resistors. Each of those stages alone would, in principle, suffice to convert that signal into analog, but because you’re making a running average of a complete cycle in there (for each bit of the cycle you shift out, there is an equivalent bit of a new cycle that you shift in), the big square-wave part of the PWM signal is being averaged out, and all you are actually left with is just the audio content and a much-reduced amount of out-of-band noise. That is quite an interesting thing. In 2004 or so I presented a paper at the AES (available on the Grimm site) explaining how the choice of conversion format influences the jitter sensitivity of the converter; in this case, filtering out 3.125MHz (and its multiples) sharply reduces the sensitivity of the DAC to high-frequency jitter. The result is that the sensitivity of the signal/noise ratio to clock noise is greatly reduced. Effectively, the only noise I get from this DAC is just resistor noise, nothing else -- no digital noise whatsoever.
On the input side (immediately before this PWM generator, which is running at 3.125MHz), any signal that comes in, be it DSD or PCM, first gets converted to an intermediate format, 32 bits at 3.125MHz. It’s a form of asynchronous sample-rate conversion. People often frown at asynchronous sample-rate conversion, because they don’t like the sound of it. Again, this was one of those things where "first approximation theory” should have said that it can’t be a problem, that it should actually be better because of the jitter removal. But then you start looking to see if theory had more details up its sleeve, and yes, it did. You really have to think long and hard about the algorithm that tracks the ratio between incoming sampling frequency and the sampling frequency that you are trying to transform it to. I think I’ve pulled some interesting tricks there, with the bandwidth of that circuit constantly reducing as long as the circuit is in lock, so that after ten minutes you are into the microhertz region. After only a few seconds, whatever the input jitter might be, the purity of the output signal is already better than that of the extremely low-jitter clock oscillator that’s on the DAC. At this point it’s as good as synchronous. It does mean that I can’t use an ASRC chip from one of the usual suspects because they aren’t good enough, they’re all sensitive to jitter, so I have to “roll my own” in software.
So there are these three really important things going on technologically: the asynchronous resampling to 3.125MHz, followed by the 100MHz pulse-width modulation (it was quite an interesting task to port the algorithm, which was originally a hardware description, into DSP code), and the shift-register-based converter. Then, the fourth bit is an otherwise normal output filter, but where several amplifiers are placed in one loop so that every stage of the output filter (which is a third-order) will actually help reduce distortion from the remaining two.
PR: Bear with me while I try to wrap my brain around all that. What results are you getting at the moment, and what little bugs remain before you have a commercial product by this time next year?
BP: From the performance perspective I am getting an output that looks perfectly analog. There are no inharmonic components in the output, no noise modulation, and the signal-to-noise ratio currently sits at 130dB and will improve slightly still. Whatever distortion is there is really low, so that when you drive a full-scale signal flat out, I get a third harmonic at around -130dB. (But that is roughly around the analyzer limit of the Audio Precision AP2722, so I actually cannot know if it is --130dB or -140dB -- I’ll have to find a creative way to eventually prove that it is lower than -130dB.) Second harmonic, same problem. I still have a fourth and fifth poking up to about -140dB, and I’d like to drop that to -150dB. Not that I think it actually matters for sound -- this is already uncharted territory. Even with the uncertainty about the second and third harmonics, we’re already 10 to 20dB better than anything else. Nevertheless, I think I just want to keep chugging along until it is not just off the charts, but, as I said, really, really future-proof.
PR: Am I correct in understanding that the Mola-Mola DAC will first be implemented in a standalone product?
BP: You are correct. We listened to the prototype, and it outperformed something which is actually very well engineered, and which sells for nearly a six-figure price. The American sales guys basically said we’d have to be crazy to sell this as a $5000 plug-in for the preamp. We’ll do the standalone product first. There is still a lot of work yet to be done. We need to sort out a few EMI issues on the analog side and, hopefully, reduce distortion even further. On the digital front, we’ve got a sample-rate switching gremlin that intermittently hangs up the DAC, which we obviously can’t have in the finished product, so there is some software work to be done there. The USB interface still needs to be done. Now that we’ve decided to make it an external DAC, I’ll have to make a front board for it. I’m still trying to work out whether I actually need a display on the front -- I’ll probably need one -- so there is some hardware design and software design to be done for that. All the hardware design and the back-plate connectors will need to be finished. I’ll likely bang the prototype boards into a box so that we can demo with it starting at Munich, hopefully, but do not expect anything on the market before next year.
Mola-Mola Kaluga amplifier
PR: The casework problems for the Mola-Mola monoblocks and preamplifier have now been worked out, I understand.
BP: Yes, they have. Getting it sorted effectively delayed the delivery of the products to March or April, although the extra time will allow us to refine the tablet interface for the preamp. Apart from sheer audio performance, the tablet interface is the preamp’s most powerful aspect, given its programmability and flexibility. Then, of course, if you want to sell the circuit on the merits of those features, you have to get the tablet interface working really smoothly. We’ll get that done by shipping time.
PR: So at the initial launch in the second quarter of 2014, you’ll have the Makua preamp and the Kaluga monoblocks?
BP: Yep, you’ve got the names right. I shouldn’t actually worry too much about these names -- I was planning on calling them Preamp and Power Amp, until our Japanese distributor said we need to have a type number or a name. So I proposed denki unagi (electric eel) in Japanese characters, and he said no, Japanese customers want Western print on their boxes. Then I just went looking for other names of fish that I like. Kaluga is a type of sturgeon, and Makua is actually Hawaiian for Mola-Mola.
PR: Will the phono-stage module be available at launch?
BP: Oh yes, it will be. Most of the boards have already been manufactured, and they are just waiting for cases.
PR: What other products do you contemplate -- perhaps a standalone version of the phono stage, or an integrated amplifier?
BP: A standalone phono stage is something I’ve been asked about, again because we had it pitched against something that sold for ten grand, and the difference was hardly subtle. It was obvious that some people would like this product as a standalone phono stage, so that will come. Of course, now that the DAC has been upgraded to a standalone product, I still need something for an internal DAC, because I don’t want to let go of the idea of having a really nice, self-contained system. Three boxes -- a preamp and two monoblock power amplifiers -- and you are ready to go. An integrated is very high on the list. I still need to work out whether I can actually cram the existing preamp, as it is, and two power stages, into one box. That would be perfect -- if I can get that fantastic quality with just lower-power NCore amps into one box, that’s what I’m hoping I can do, but it’s a mechanical issue. If it doesn’t fit, then I might need to skimp on the phono stage and build an RIAA-only phono stage that just fits on the channel board along with the rest of the electronics, to make space for the power stage and the bigger power supply.
Mola-Mola Makua preamplifier
PR: We are now in January 2014, one quarter away from the first shipment, and there is a development path fairly well defined for Mola-Mola. From your present perspective, what has been the big difference between working at Hypex, working at Grimm (where your customers are other manufacturers and the professional marketplace), and Mola-Mola? You have a different type of customer, and a distributor and dealer network to build. What has been most exciting and most frustrating?
BP: Well, from that perspective, obviously the DAC is the most exciting part of the project. (I had the amplifiers already done, to a large extent, at Hypex, of course.) The most fun part has been designing the really sweet, simple, discrete op-amp circuit, which performs phenomenally. When you look at the circuit, you say, “How on earth can such a simple circuit ever do -150dB distortion!” I love it when something is really elegant. The biggest frustration has been the casework -- just not being able to explain to people what the word perfect means. Personally, getting to know the high-end market, especially the US high-end market, has been an interesting cultural experience. It is so different from pro audio. It is actually rather different from the European high-end market too. You meet entirely different people than what you ordinarily do in pro, or when you do audio products for industrial customers. I think the human aspect has been the wildest part of the ride for me.
PR: Is that a good thing, a bad thing, or just a wild thing?
BP: It’s neither good nor bad, it is just a new experience. You know, as your prototypical Aspie geek, I’m still not sure whether I’ve properly figured out my own country. Suddenly landing here in this market, where people really do think differently and have different cultural preferences, even different humor, it’s like, instead of going to a different country, you’re going to a different planet. Even within the US, between [T.H.E. Show] Newport Beach and [the Rocky Mountain Audio Fest, in] Denver, you get a completely different crowd. I haven’t done the New York show yet, so it’ll be interesting to experience that too.
PR: What’s interesting to me is that at this moment, as many writers complain about the “death spiral” of stratospheric pricing at the top of the high end, a group of young people are just coming in from the low end of iPods, lossy downloads, and fashion-oriented headphones, and are becoming aware that there is more than supercompressed, lossy MP3s. There is growing demand at the entry level for higher-quality sound, whether from headphones or computer speakers or simply preconfigured small audio systems. If the right products are shown to these consumers in the right environment, they may be willing to push their playback envelope rather far. In 5, 10, or 15 years, these consumers will be the high-end marketplace -- or else there may not be one.
BP: The current audiophile market is a graying one in Europe as well. If we don’t break open this new market, the whole industry is going to pot, I’m afraid. The thing which makes it so fun for the traditional high-end clientele -- let’s say the “tweaky bit” -- is actually off-putting for the younger crowd. They do not want to see half of their carpet space taken up by cables. They don’t particularly go in for multi-box products. They just want something that works and sounds good. There is a huge opportunity here, and it is misunderstood both by the traditional high end and by the mass-market-oriented consumer companies that are supposed to cater to precisely that younger market. On the one hand, the mass-market units simply do not offer the same kind of sound quality, because you really want the same experience that you have, say, in our [Mola-Mola/Luxman/Vivid] room upstairs. If you could get that same experience, but you could just plop down two speakers and point at something on your iPhone, that’s what they want. You don’t want the Sonoses of the world, because they’re just not good enough. But in the high end, even those who pay lip service to this market by having some iPod interface or USB interface make products that are still too tweaky.
PR: Perhaps, after you get through the release of your Mola-Mola integrated amp a few years down the road, you can revisit your first year and a half at Philips -- but instead of making crappy little shit boxes, you could make a product in that category that is as good as you could make it. What could you make for $1000 retail?
BP: That, actually, would be really great fun. I find it at least equally gratifying to do the best for a fixed price and target as to do the best possible. But in order to do that, you do need to have the manufacturing wherewithal and the sales organization and marketing organization to put a high-volume product into the market in one go. You need to have something like [what Andrew Jones has for the $100 speaker he designed for] Pioneer -- the marketing people who have that vision for sound quality, plus such an understanding of the market that they can securely commit to a large quantity. That is really something that I would love to see happen in the future -- probably not Mola-Mola branded, but certainly it is an interesting challenge to really try and pare down until you are just left with the essence of what’s important for quality sound.
I believe, personally, that I’m uniquely placed, because I’ve taken it as a personal quest to separate the sizzle from the steak and not get bogged down in certain preconceptions, certain illusions regarding what type of components you need, what type of circuits. If you do not believe in feedback, for instance, you cannot make a low-cost, high-quality amplifier, whereas someone who understands negative feedback can. Much the same with converters. If you know exactly what to look out for in a chip, you can find chips that are really suited for use in a $1000 product that, for the price, will offer fantastic performance, simply incredible value. Speakers ditto: You wouldn’t believe the performance of some $10 drive-units, but if you’re hung up on exotic cone materials, you just won’t discover those.
PR: Thanks so much, Bruno. Perhaps an enlightened executive out there with the proper vision of sound quality, with a sales network in hand and the marketing muscle to commit to the necessary volume, will give you free rein to champion (in what little free time you have) the development of the highest-quality-possible $1000 playback system. I’d love to see that happen. In the meantime, I look forward to hearing that your first Mola-Mola products are shipping, and to seeing you again at T.H.E. Show Newport Beach at the end of May. It has been a pleasure.
. . . Peter Roth
peter@soundstagenetwork.com