You are reading the older HTML site

Positive Feedback ISSUE 11
january/february 2004

 

The DSD/SACD Revolution, Part II: PF Interviews Digital Designer Ed Meitner
by Mike Pappas

 

The following interview first appeared in Issue 8, Volume 2 of the print version of Positive Feedback

Pappas: What I’m doing is trying to really flesh out the story on the DSD program for Positive Feedback magazine. And so that’s one of the reasons why we’re doing this phone interview is because I want you to tell me the story. I want to know how you got involved with it, why you think it’s a good approach.

Meitner: Originally it was through Tom Jung at DMP Records, who recommended me to David Kawakami at Sony. We got together at CES last year, in January, and talked about if I wanted to get involved and what I could bring to the table. They had an MPGA meeting coming up in New York, and for this meeting they needed some converters. At the same time at the MPGA meeting, Tom Jung did a recording of Manfredo Fest, which is a Brazilian jazz group. I made the first set of converters for that. It turned out they worked out pretty well, so we decided to continue from there. Sony needed more converter work, like eight-channel design and stuff like that. I agreed to do that, and that was pretty well the beginning of it.

Pappas: Let’s talk for a minute about your background. Did I understand that you, a long time ago, had some exposure to the DBX system back in the ‘80s?

Meitner: Yes, well, exposure. I played with it when it was available. Because back in the ‘70s, I was involved with a company called Olive Electrodynamics.

Pappas: Oh yes.

Meitner: I was the analogue designer there.

Pappas: Okay.

Meitner: We built recording studio consoles.

Pappas: In fact, I believe some of the first automated ones.

Meitner: That’s right, yes. We used the VCA, which came from Dave Blackmere. And that’s how I got to know Dave, and I think we’ve been friends ever since.

Pappas: So at that time, you had an opportunity to try the DBX system, which was kind of a bitstream also.

Meitner: Well, we used to call it "delta slope" in those days. And also at ADS we own a company called Delta Labs.

Pappas: Oh yes.

Meitner: Delta Labs did a bunch of the original design work of that scheme. And also the Shure HDS converter or HDS surround sound system used the delta slope modulator and demodulator to process the rear delay.

Pappas: Ah, very interesting. So you’ve got quite a history with this type of data conversion then.

Meitner: Well, also I built some cassette replacements for paging terminals where we stuffed that format into shift ridges to get about 15 to 20 seconds of recording time.

Pappas: Very interesting.

Meitner: And obviously, you know, that blew away cassette machines in a hurry.

Pappas: I would guess so.

Meitner: But then I played around myself a lot with it and it always came up. But the problem always was that there wasn’t a storage facility; i.e., it was too many bits to store on anything, except maybe a Umatic video machine or something. So it never really flourished until now.

Pappas: You feel one of the reasons this has got a good shot at being successful, besides the fact that you feel that — I’m assuming you feel that it sounds superior to PCM — is that storage and computer horsepower have gotten inexpensive enough to make sense.

Meitner: Yes. Also, the thing is that if you try to archive or re-archive analogue tapes and analogue masters…

Pappas: Right.

Meitner: … well, if you have it in the one bit DSD format, you can, (A) you have a pretty robust storage that way, (B) you can now convert it to any other format that may come about, PCM 96/24, whatever. So it’s a very versatile format to begin with. And don’t forget that every A to D converter that you see on the market today starts off as a DSD modulator. So then you have the DSD signal on the A to D that just goes to the PCM down sampler or decimator and gets turned into PCM, so the life of the audio in the digital world really starts off as a one-bit signal.

Pappas: And so this is just a natural extension of that.

Meitner: That’s right. Yes. So now if you have an audio chain, you have one bit coming out of the A to D and going into a storage medium, i.e., hard disk or the AIT. And your D to A converter again is relatively simple because you don’t have to do any conversion from PCM to bitstream. And most converters out today are some form of bitstream. So we just cut out the PCM part.

Pappas: So in effect, DSD is simpler because you’re not doing the kind of conversions using a decimation filter to make it into PCM. You’re just eliminating all that extra stuff.

Meitner: That’s right. Yes.

Pappas: So what are the challenges of designing converters for this format?

Meitner: Really, the way you drive the A to D converter, you have to be careful with your analogue circuitry there. In all this digital audio, analogue is the weak link. So it requires a good deal of attention to the classic analogue problems and analogue issues that have to be dealt with. And you know, other than that, it’s really no big magic.

Pappas: When you say the analogue is the weak part of the chain, can you explain that a little further for me?

Meitner: Well, you have an analogue input that goes into the A to D.

Pappas: Right.

Meitner: Somehow it needs to be processed. It has to have probably some gain or some gain adjustment. It has to be balanced in, single-ended in. And then the actual ports of the A to D need to be driven with a fairly low impedance. There is some reactive impedance from the A to D input, which means often op amps generally are not happy there. So you need analogue circuitry that can live and do the best to drive the A to D inputs. Once you have that done, of course, then you have the power supply issue and stuff like that.

On the D to A side you have your digital part; in other words, the bitstream part that comes into the D to A and has to put out analogue. So the same issues are there. You have high frequency in the presence of op amps or other analogue circuitry. So you just have to be very careful. And I think my specialty is that interface part, how to deliver audio to the A to D converter, and how to take it from the D to A converter and present it to the outside world again. And then there are issues of jitter and clock distribution, relatively sensitive issues that over the years I’ve learned to deal with and make them as best as we can today.

Pappas: Now one of the other issues with DSD is that you’ve got to do fifth-order noise shaping filters to shift the noise that is a result of the conversion process out of the audio band.

Meitner: Yes.

Pappas: Can you tell me what kind of challenge that presents from a design standpoint?

Meitner: It’s what I said before. There is high-frequency noise present in the system. And the analogue parts have to deal with that, which means it has to be low distortion, low noise, and very high speed analogue circuitry that can live without generating any intermod products in the presence of the kind of noise the DSD signal presents.

Pappas: The other thing is you’ve got a pretty wide input band, somewhere between DC and 100 kilocycles, which has also got to present some challenges.

Meitner: Oh Yes. Yes. It again comes back to the same thing. The tradeoffs are low noise, low distortion, wide bandwidth and the ability to drive, for instance, reactive loads, like the A to D input for instance. So those are usually tradeoffs that are diametrically opposed.

Pappas: The wider the bandwidth, generally the more noise you end up with.

Meitner: That’s right. Yes.

Pappas: And so when you try to squeeze 120 dB of dynamic range and still maintain a DC to 100 kHz signal, that’s got to be pretty tricky.

Meitner: Mm-hm.

Pappas: Does that translate into "expensive"?

Meitner: Not necessarily.

Pappas: So you feel that you can build these products for a competitive price compared to a PCM converter?

Meitner: Oh, absolutely. Because in a PCM converter you have much of the same problems, except you don’t see the high frequency problems directly. But you might still see them as intermod, and if you don’t see them, you hear them as bad sounds. And, you know, what was very helpful for me is that for years I worked with a company called Amber Electro Design and we built distortion analyzers.

Pappas: Very good distortion analyzers.

Meitner: Yes, so again I did all the analogue work on that. So, low distortion, low noise, and wide bandwidth is not a strange thing.

Pappas: The world is working on two different standards here. One end of the spectrum is pushing 96/24 PCM which basically is nothing more than hot-rodded 44.1…

Meitner: Right.

Pappas: … and then you guys are coming in from a completely different angle saying, "Forget PCM. There’s a better solution here."

Meitner: Uh-huh.

Pappas: I understand why I think it’s a better format. Tell me why you guys think it’s a better format.

Meitner: To convert audio into PCM is a very alien thing, whereas if you look at the convert audio into one-bit format, it’s a very natural thing. In any form of conversion, you will lose something. You have to choose the format where you lose the least, which means the format that’s the friendliest to audio, which is definitely DSD over PCM.

Just look at one problem with PCM. Imagine what happens at your zero crossing. You have all those bits flipping. You have, you know, noise shock in the system coming off the power supply if all of a sudden 23 bits change from all zeros to all ones. You have that at every zero crossing. And you need really good error correction, because if a sign bit gets screwed up in the process, all of a sudden instead of your signal being positive, it thinks it’s negative.

Pappas: That wouldn’t be very good.

Meitner: No. And the other thing is archiving our recordings, which to me is a very important part, because the old audio tapes are falling apart, and somebody needs to do something with them. If you convert them to PCM now, you’re nailed into that format. You can’t ever get out of it. Whereby, if you do it in the one-bit domain, in the DSD domain, you can convert it, like I said before, to any other format again.

Pappas: Through the use of…

Meitner: … through the use of a decimator and down sampler. So we can take the DSD and put it back into 44.1 or 96/24 or whatever comes about.

Pappas: And not lose as much.

Meitner: That’s right.

Pappas: Because the guys who were doing 96/24 and trying to convert it to 44.1 are going to end up tossing something.

Meitner: Well, yes, that’s a nightmare.

Pappas: Well, not if you believe their propaganda.

Meitner: Well, it’s a nightmare. I don’t believe the propaganda.

Pappas: They’re talking about it like it’s no big deal. And for those of us who know that 96/24 does not equally divide into 44.1, know that that’s not going to —

Meitner: Yes, but also don’t forget, those are the same people that have said, "Digital is perfect forever."

Pappas: You guys have got to be taking a pretty big risk on this thing. Let’s…

Meitner: I don’t think so.

Pappas: Let’s move to the more practical side. The world currently seems to be moving towards just taking PCM and extending it. There’s a lot of hardware out there, not a lot of it’s particularly good, but there’s hardware. Tell me why I shouldn’t think you’re in a "come-from-behind" mode here.

Meitner: What do you mean by "come from behind"?

Pappas: Well, 96/24 in the pro audio business, there are a myriad of vendors providing at least digital audio work stations that run on 96/24.

Meitner: Yes. Those are also the naysayers because they’ve got all this invested interest.

Pappas: I believe that DSD is a much better way of doing this and I pretty much feel that us going to 96/24 is like a kid trying to stuff a bigger engine in a Chevette.

Meitner: That’s right.

Pappas: You know, you still have all the limitations of the drive train and chassis, it maybe it goes a little faster.

Meitner: See, I always have to come back to the same thing, it’s that your A to D converter starts off its life — I should say the audio through an A to D converter starts off its life — as a one-bit, or similar to a one-bit signal, and then you muck it up.

Pappas: And as you refer to that process, PCM is alien.

Meitner: Yes, it has really nothing to do with audio. You have minimal resolution at zero crossing, whereby with DSD you have maximum resolution and on and on.

Pappas: So in other words, DSD is pretty much the —

Meitner: It’s the least conversion of a conversion.

Pappas: And with PCM, you were mentioning that you know your biggest problems are when you get around the least significant bit, which is at zero crossing; with DSD, that’s where you have your maximum resolution.

Meitner: That’s right.

Pappas: That’s very interesting.

Meitner: You have, in fact, a situation that is very akin to what we hear. We mostly hear velocity changes. Now velocity changes are at maximum at zero crossing of the sine wave. So this is where you have to be so careful. And if you look all through the high-end audio industry, it’s class A, class A, class A. And you know the old solid state zero crossing distortion amplifiers and stuff like that never worked.

Pappas: Because the ear is more sensitive at that point.

Meitner: Yes, "maximum velocity" means "maximum intensity," which means the point where we hear the most. So now look at a PCM signal at zero crossing, and all you’ve got at that moment where it crosses zero is you have zero-bit resolution. The only resolution you’ve got is dither.

Say you have a 16-bit system where one bit is dithered. You now have your noise floor quantized at one bit, and that’s why, when you go to higher resolution tracks or higher resolution PCM converters, people think they are better — and they might be — but the really fundamental reason in my mind why they are better is because you quantize your noise with more bits.

Pappas: And that exists through the zero crossing issue.

Meitner: Well, anywhere near zero crossing. Because we know that as your signal level goes down, with every 6dB decrease you lose one bit.

Pappas: Right.

Meitner: That’s why when you do digital recordings, you nail this thing to zero dB as much as you can. So let’s look at a typical mike pre-amp that might have, say, an equivalent input noise of -131 or -130 dB, right?

Pappas: Right.

Meitner: So then we put, say 40 dB of gain, which is a typical gain, I guess, and we end up now with a noise at -90. So now we have a 16-bit converter. That converter is going to be now dithered from that noise with one or two bits. Right?

Pappas: Right.

Meitner: So the resolution of that noise at the bottom is one or two bits. Now, we take a 20-bit converter. All of a sudden we have the same situation, the same noise floor still, except that that noise floor is now quantized with five or six bits, and so on and so on. So I call that "the bits that are dancing."

Pappas: Dancing bits.

Meitner: Which is what gets you through the parts where you need resolution and you just don’t have the bits to do it.

Pappas: And DSD gets around these problems.

Meitner: Yes. And that’s what I call a friendlier conversion or a lesser conversion or minimal conversion, I should almost say.

Pappas: Do the same issues about jitter apply to DSD as they would for PCM?

Meitner: Of course.

Pappas: Okay, there are no disadvantages or advantages in terms of jitter performance?

Meitner: No. Jitter is still an important issue. The way I see it, the problem with jitter is that very few people — almost none out there — have an analyzer for it. I think we were about the only ones who, for the past five or six years — from the moment when Stereophile adopted our jitter analyzer — actually measure this stuff. And we measure it in such a way that you don’t only know how much jitter you’ve got, but you also know what your frequency components are in that jitter.

Pappas: That’s the LIM device?

Meitner: Yes.

Pappas: Can you give us a little background on it? Because I think it’s important.

Meitner: Basically, it is a face demodulator that you can look at any sample frequency within a digital system from 2FS to 768FS. In a 20 kHz bandwidth, it gives you your jitter components and your frequency components in the jitter, which is more important than just knowing what the actual value of the jitter is. Because if it’s just noise, it’s not so bad, but if you have components in there that are alien to audio, well, that’s a different story. What are these "components"? Well, typically, it’s artifacts coming from a CD drive, a deck; you have components in there that are jitter components that come from the focus adjustment, tracking adjustment, motor speed adjustment, and all sorts of other staff that have nothing to do with audio. So this device shows you what those components are.

Pappas: Very interesting.

Meitner: You know, when Stereophile adopted that, we agreed and promised that any of our competitors that would want that equipment, we would make it available at a very reasonable cost, which we did.

Pappas: And the primary advantage of this is not just knowing the number, but knowing what the spectra of the jitter components is?

Meitner: Precisely.

Pappas: And that’s a lot more —

Meitner: Well, then you can find it.

Pappas: Otherwise you’re just hunting around.

Meitner: That’s right.

Pappas: So, let’s talk about the future. It would appear that there are some vendors who are getting on board with DSD in terms of being able to provide things like digital audio work stations and things like that. Is working in the DSD realm easier or tougher for things like building digital mixing desks —

Meitner: It’s really no different.

Pappas: Okay.

Meitner: It’s just re-learned. That’s all. At the Sony demo they had a mixing console that was purely one bit, filters and all.

Pappas: Everybody’s brain is so accustomed to having PCM represent an absolute value, and there are certainly a fair number of naysayers out there who say that you can’t do it.

Meitner: Yes, well, there always is. The thing is, there is an obsession with absolutes, totally forgetting that human sensory inputs are not so much sensitive to absolutes as they are to deltas, and our hearing is really no different. So, you know, this is also one of those philosophical things where PCM is this absolute machine, and DSD is this relative machine. And it’s just better for us as humans. And I firmly believe that general health would be better if PCM would not exist.

Pappas: And why, particularly, is that?

Meitner: Because there is a subliminal irritation about PCM that may just affect the psyches of people in a bad way, and certainly distracts from the pleasure of listening to music. And if listening to music was considered as relaxation and was supposed to be a way to relieve stress, then PCM, like CD playback, certainly doesn’t do it as well as some of the old analogue stuff did.

Pappas: So maybe that’s one of the reason the music industry sales have been down.

Meitner: Possible. Aside from the fact that, right now, it doesn’t seem to be the same scene as I remember from the ‘60s and ‘70s. This is a hard thing to say, but I hear from a lot of people that, with an LP, you used to sit down, close your eyes, and sort of float away with the music, relax and unstress. And with CD, it’s just not the same thing anymore. So even though you might not hear the problems glaring at you immediately, I’m sure they wear.

Pappas: Do you think the other thing might be the fact that when you ran an LP, it was generally about 22 minutes, and then it was time to get up and change it?

Meitner: Well, that could be too. Now with the CD you have the remote control. You can change it at all times. But I find a lot of people don’t even get to 20 minutes.

Pappas: They’re bailing out long before then.

Meitner: Yes. And you know the funny thing is, on top of it, you know all the converter people who started off in the multi-bit scene, have all changed to single bit. Phillips with their bitstream; and look at the vendors of DAC chips and A to D chips and it’s all gone to single bit. You would be hard-pressed today to still find multi-bit converters because of the added problems of them not being enabled to do zero crossing and stuff like that properly.

Pappas: This problem is going to get worse with 24-bit. Because trying to get linearity at those low levels out of converters is a real problem for PCM guys.

Meitner: I think the 24-bit is a crock of lies. I don’t know how many marketing bits have snuck into that.

Pappas: So what do you think they really actually deliver, in terms of linear bits?

Meitner: Oh, in the digital domain, I’m sure they deliver the 24-bits. But when it comes to analogue, I’ve never seen an analogue circuit that can do that.

Pappas: So it’s marketing bits, huh?

Meitner: Well, yes. But again, you know, the nice thing about it is that at least now your noise quantization has more bits.

Pappas: Which improves your performance at zero crossing.

Meitner: Yes, yes. And also that fact that usually those converters are of delta sigma modulation type design.

Pappas: Which is DSD is once you take out the decimation filters.

Meitner: That’s right. Yes. And you know, with audio traditionally, less has always been better.

Pappas: I don’t know, unless you’re an English console manufacturer.

Meitner: Usually less is better. If you can make a shorter feedback path, a shorter this, a lesser that, it usually works better sonically.

Pappas: Has the output of DSD got some sort of a standard format?

Meitner: It’s S/DIF. Except that it’s three wires so you have your sync, which is the clock, and then you have left DSD and right DSD.

Pappas: Very interesting.

Meitner: So of course you have none of that squirrelly S/PDIF where you have to extract a clock from the data and the whole data stream, which is another jitter problem.

Pappas: And you also don’t have to extrapolate left and right out of it, either.

Meitner: No.

Pappas: That’s very interesting.

Meitner: It’s discrete, you know, clock left and right.

Pappas: That makes a big difference in trying to get the reconstituted clock out of it.

Meitner: All this audiophile work with their separates? Like a CD drive and then a converter? Well, the weak link really is the S/PDIF. If that would have gone to even a clock and left/right on one wire, that would have been a lot better, instead of this one wire where everything is on one wire and then you have all these problems.

Pappas: Very interesting. So in other words, it might actually be better to buy a one-piece unit —

Meitner: Oh Yes. Oh Yes.

Pappas: Because you don’t have to make the conversion to S/PDIF…

Meitner: That’s right.

Pappas: … and then back again. You can keep it all …

Meitner: I think every time there is a conversion, nobody could be as arrogant as to assume that there isn’t going to be some loss.

Pappas: And so effectively, you’re making two conversions by going to a two-piece system.

Meitner: That’s right. No free lunch.

Pappas: Not even a reduced-price lunch.

Meitner: No.

Pappas: Well, if this format is going to be accepted vendors are going to have to come forward with all kinds of hardware.

Meitner: Actually less hardware.

Pappas: But you’re going to need a multi-track machine.

Meitner: Yes.

Pappas: You’re going to need some sort of a way to do this into a two-track environment. You’re going to need some way to put it in a computer and manipulate it. When I say "lots of hardware," that’s what I mean.

Meitner: Okay. But don’t forget that Sony and Phillips are pretty powerful engines.

Pappas: Oh, absolutely.

Meitner: They are really working — did you go to that AES show?

Pappas: No,

Meitner: Because you would have been really surprised how much gear there was there already.

Pappas: Really?

Meitner: Yes. And the issues with the dual layer disc which is backwards compatible. You know that never happened before in this industry.

Pappas: Yes, you had to throw out everything that you used before.

Meitner: Exactly.

Pappas: And that was I think one of the reasons why the CD was so successful was that retailers could dump their LP stocking requirements and they put more product in a given amount of floor space than with LPs and they didn’t have as big of a return problem.

Meitner: Right.

Pappas: And one of the things that you will be very hard-pressed to do is get retailers to double-stock inventory.

Meitner: Uh-huh.

Pappas: You’re just not going to see guys like Tower Records want to give up any floor space for another format. So I think the dual layer thing to me was a brilliant piece of work, because it gets you around the hole. The retailer doesn’t know any difference. The consumer doesn’t know any difference. It just depends on what machine he plugs it into.

Meitner: That’s right, Yes.

Pappas: And that to me is going to be the key for making a format successful is it’s got to be transparent to the customer.

Meitner: Yes.

Pappas: Because frankly, you know, most customers think the perfect sound forever is okay right at this point. Also most customers are listening on boom boxes and ghetto blasters.

Meitner: Right.

Pappas: But they’re the guys who buy the majority of the music and drive the market.

Meitner: So it’s not a bad thing. You know, it’s not some evil big company that wants to railroad the new format. In fact, they very seriously thought about it and came up with this dual-layered thing where nobody really loses anything. It’s a win-win situation the way I see it.

Pappas: Well, and it looks like they’ve done a good job at least getting people like you and the folks at Sonic Solutions involved in it so that people know it’s not a one-stop-shop.

Meitner: Right.

Pappas: Because the industry does not like that. They want to go to multiple vendors and pick and choose who they’re buying stuff from and where they’re getting it from. And a one-stop-shop is not a good thing for this business.

Meitner: I think this whole development is politically very correct. It’s the people that are interested in better sound will get the benefit from it. 96/24 can run in parallel. 96/24 can even be the top layer.

Pappas: Oh really?

Meitner: Well, why not?

Pappas: And if you shoot it all in DSD then you can convert to 96/24.

Meitner: Yes, you’ve got an open format that you can use for any sort of conversion. So I really don’t understand why people have a pickle up their rear and slam it.

Pappas: Well, perhaps because they didn’t think of it.

Meitner: Well, maybe, yes. I don’t really want to dwell on the naysayers; I know what their thoughts are. For sound quality, here’s one simple test. I’m doing some transfers of vinyl LPs onto DSD. And, you know, in DSD this is a conversion with a minimal amount of damage to the original sound. If you consider playing back vinyl and liking all the good things about it, now we can have it in DSD format. We could possibly take some of the clicks and pops out of it and still have the general good flavor preserved. The same holds true for analogue tapes and any kind of conversion. So it’s really a very nice thing.

Pappas: Well, the analogue tape situation is becoming real critical, because all that stuff is really starting to fall apart.

Meitner: That’s right.

Pappas: And the ‘60s stuff actually isn’t quite as bad as the ‘70s, when everybody switched to the high-output tapes; those puppies are really falling apart.

Meitner: Yes.

Pappas: And if you make a conversion into 96/24, then that’s it: you’re stuck. You’re not going anywhere else.

Meitner: You should also get the Tom Petty disc that MoFi did. Compare that to the one that was done with the DSD system and then down sampled with the Sony Direct, SBM direct versus the original. It is a world of difference.

Pappas: Really?

Meitner: Yes, you definitely want to get that. Because that will show you that when it’s down converted properly, how much of the good flavor is preserved.

Pappas: I like your term for it: "good flavor." To my way of thinking, the DSD is a significant change in thinking in terms of how digital works. We have been doing PCM, effectively, since the late ‘70s. And I think DSD presents a real shift in thinking. I have to assume that you guys are in that same kind of vein thinking that way.

Meitner: It’s the undigital digital.

Pappas: I’ll just briefly tell you that in the ‘80s I worked at Otari Corporation. And we did a big shoot out at Wally Heder’s place in San Francisco with all the PCM machines and DBX systems back in, boy, it had to be like ‘81 or ‘82.

Meitner: Was that with the evil PCM F1?

Pappas: Oh, well, we listened to the evil PCM F1 and we also had Mitsubishi 850 machines and the JVC system. We had the DBX system. At the time, the DBX system was the best-sounding digital system that was out there. We did the same kind of thing. We hooked up all the digital systems and we listened to direct, off-the-console versus the processed digital system feed. And at the time Otari attempted to license that technology from DBX, but DBX underwent a couple of ownership changes and everything fell apart.

Meitner: Yes, I think it was — who was it that bought them? Garrard or something like that?

Pappas: Oh, my God! I don’t know. They went through like about four owners in about three years.

Meitner: Yes, right.

Pappas: And the folks at Otari didn’t like that kind of stuff. And the deal fell apart. But we were all ready to cut a deal to build multi-channel digital machines and all that stuff.

Meitner: That would have been nice.

Pappas: Well, at the time that product was certainly light years better than any of the PCM systems.

Meitner: Well, you know, there was amongst us people that knew about that this big outcry — why CD had to be the way it was designed. You know, it could have been better. But I guess the straight-laced engineering guys went at it and said, "Well, we’re going to do it this way."

Pappas: Well, and I also think that they made a couple of decisions based on the technology that was available right then and there, without a lot of thinking about what would happen in the future.

Meitner: Yes.

Pappas: And for a lot of reasons, PCM is relatively easy to manipulate on low horsepower computer systems. And you know, if you get into the way-back machine and go to that time, computers that had 640K RAM cost eight grand.

Meitner: Oh yes, I remember that. As a matter of fact, at Olive, we used to build our own set of computer mechanisms, and you know, there were lots of chips. Lots and lots of them. And they weren’t very reliable either.

Pappas: Well, and I remember when 40 megabyte hard disk drives were huge and expensive. And also not very reliable. "But other than that, Mrs. Lincoln, they were fine." So I think what those guys did was say, "Okay, well this is how we’ll make it and it will be easy to manipulate and we’ll just live with it."

Meitner: Because, between you and I, I venture to say that even at 16 FS, delta slope would have been better than PCM.

Pappas: Oh, I would vote along those lines also.

Meitner: You know, nobody really cared about a little bit of background noise that you have in most systems; they are never played loud enough to hear it. The problem with record noise wasn’t so much the noise, it was all the soda biscuits in it.

Pappas: Yes, the transient "snap-crackle-and-pop" got a little annoying.

Meitner: That’s right, yes. But just straight even noise, you know, 80 dB weighted or something wouldn’t have been such a big problem.

Pappas: No. I think there was a fair amount of input from the marketing department. And the marketing guy said it’s gotta be, you know —

Meitner: "Perfect."

Pappas: Yes, and it’s gotta be silent.

Meitner: Yes.

Pappas: And I think that’s kind of why we ended up doing 44.1 PCM. And frankly, when you think about how short a duration it took for CD to overtake vinyl, if you looked at the RIAA numbers, that conversion and the consumer acceptance of that format was very rapid, all things considered. And quite remarkable. And, you know, I think that was all part of the marketing hype: "look at these specifications and don’t confuse us with listening to it, just, the numbers look great!"

Meitner: With 44.1, the filter, the so-called oversampling filter or the interpolator, is a phase linear device, right? So you have a ringing before the transient. And you have a ringing after. Now the ringing before the transient, our hearing system is anticipating and we hear that as a noise modulation, ever so slightly, but grating. I remember whenever we played with CD here, we always went back to a one FS system with an analogue filter, because that analogue filter rings after the transient. So where you look at a step and it’s going up and then you have some ringing, okay? And then it doesn’t ring on the way down. In other words it doesn’t know the future of the signal and that’s another very alien part of it. For five or six years, we’ve made a converter where we have oversampling, we have digital filtering, but we do not have the ringing. And that has made a huge difference; people like Tom Jung at DMP will attest to that. It is the closest thing to DSD.

Pappas: And that converter is the — ?

Meitner: The IDAT. Basically what we’ve done is we’ve made a detector that detects when the FIR filter gets into trouble. So we switch between FIR and linear interpolation.

Pappas: That’s a neat trick.

Meitner: So we don’t get the typical distortion characteristics that some of our competitors get with spline fitting and French curve fitting and stuff like that. We still have flat 20 kHz response on a sine wave. But if you look what an interpolator like an eight times oversampled interpolator will do, it will give you a flat frequency response. Which means that it deals with the sine-x-over-x error. It pushes it out eight times. So now the 3.2 or 3.4 dB down at 20 kHz is now 0.04 dB down at 20 kHz, okay?

Pappas: Okay.

Meitner: But it only knows to do that on a sine wave. Imagine a transient coming along.

Pappas: It won’t be able to figure it out.

Meitner: It can’t do it. It will now give you a possibility of a 3.4 dB error at 20 kHz.

Pappas: On a transient.

Meitner: On a transient.

Pappas: Wow. I never realized that.

Meitner: So you have this transient coming along and those transients are amplitude modulated, but they are positioned within the sampling system and that is the killer.

Pappas: I never even thought of it.

Meitner: So it is much better to live with the 3.4 dB error on the transient but keep the damned transient stable, and the linear interpolator will never lie.

Pappas: That’s very interesting.

Meitner: We’ve implemented it with the first product that immediately got class-A rating at Stereophile. And from then on, that was a $15,000 showpiece that we made, and then we learned how to make it a lot cheaper and more elegant and dropped it down to $2,000. So we only sold six of those expensive ones.

Pappas: You sold a lot more of the two grand ones.

Meitner: Yes. But that has worked so well. And as I think about it, to some degree the 96/24 will still have the problem that we’re discussing if they use brick wall filters and stuff like that. Not to the same degree, but they’ll still have it.

Pappas: Well, they’ve just moved the problem up a little further.

Meitner: Yes.

Pappas: It’s still a problem.

Meitner: Yes. And any kind of phase equalize filter — I don’t know if you remember on all those digital tape machines, the input had to have an aliasing filter.

Pappas: Right.

Meitner: Well, there were some that were the non-phase equalized ones, just the elliptical filter. And then Apogee came out with their phase-equalized filter and the sound started to sound like CD. You cannot introduce the future into an audio signal.

Pappas: I never looked at it that way. That’s very interesting.

Meitner: So any time for us when we develop this project, the sanity check was the one FS converter.

Pappas: Very interesting.

Meitner: And I don’t know why people don’t latch onto that, but it is so evident. You put an amplitude modulation in with something, and all this theory that says that some of this alias are pushed out beyond 20 kHz don’t mean dick, because you still have a ripple-down effect that happens within the pass band.

Pappas: You have to.

Meitner: Yes. And that’s why I said it’s a subliminal thing because you don’t hear very well. Amplitude changes at 10, 15 kHz, you don’t hear very much. But nobody really knows what this does when it carries on over long periods of time. The most vicious instrument is a trumpet.

Pappas: Tell me about it!

Meitner: So you look at the signature of a trumpet and you have these fine needle points of the curve which have a very fast turnaround. Well, you know, that FIR filter is continuously in ringing mode.

Pappas: And you’ve got lots of harmonic components.

Meitner: That’s right.

Pappas: And they’re very high amplitude.

Meitner: The FIR is not a filter. It’s a window. And that window has to have a finite opening and it is a question of how many samples are in that window.

Pappas: At high frequencies, not enough.

Meitner: Well, still too many that give it what is called a Gibbs phenomenon, which is the ringing. And once this is taken out, I have people in the industry that absolutely fell in love with the result. In fact, we have one customer who asked a very interesting question: how many pieces of five year old digital equipment are out there that you would want to own?

Pappas: None.

Meitner: Right. And you know something, like with all this DSD work, you know, there will still always be the need to convert PCM into single bits. And that’s where probably those kinds of methods will shine.

 

POSITIVE FEEDBACK ONLINE © 2004 - HOME

BACK TO TOP