The transport part of the player is even easier to nail, this thinking goes, because all it needs to do is extract the data accurately, something any box-store CD player can do. Jitter? No need to worry about that, or anything happening in the time domain, as long as the data are transferred to a decent DAC via an asynchronous isochronous interface and reclocked inside the converter. Reclocking salves all digital wounds, or so this thinking goes.
What’s especially reassuring to like-minded audiophiles is that all this can be verified with a simple set of measurements that almost anyone can do; all you need is some affordable software and a $150 USB computer interfaceor, at most, an Audio Precision analyzer, which isn’t cheap but costs half as much as Michael Fremer’s reference phono preamplifier.
Such an approach allows the manufacture of players and DACs that can be sold for perhaps $1000, or even several hundred less than that, assuming it’s manufactured in a low-wage country. Manufacture it in the US or Europe and, even if it’s built to an exceptionally high standard, the price can remain quite low.
A top-quality digital source, then, is a commodity, like gasoline, a dozen eggs, or flash drives. It’s pointless to spend more, or so the thinking goes. Or perhaps not.
Digital is analog
The subject of this reviewthe CH Precision D1.5is hardly a commodity. Fundamentally, it’s a transport, built to a very high standard and equipped to read and output data from CDs, SACDs, and MQA CDs. But it’s modular. It accepts add-in cards that turn it into a CD/SACD/MQA-CD player.
Equipped as a transportwith, of course, a digital output cardthe D1.5 costs a formidable $41,000. Equipped as a player, with two mono DAC boards added in, the price rises to $46,000.
As I prepared to write this review, I spoke by Zoom with CH Precision’s two principals: Florian Cossy, the “C” in CH Precision and also in “CEO,” and Thierry Heeb, the “H” in CH Precision and a senior researcher at the University of Applied Sciences and Arts of Southern Switzerland specializing in audio/video DSP. Cossy and Heeb are both engineers, Cossy on the analog side and Heebobviouslydigital.
During our chat, both admitted the possibility, even the likelihood, that other, quite different approaches could be equally valid. Still, they have their own vision, their particular approach. Their job, as they see it, is to execute that vision to the best of their collaborative abilities. So, what is that vision?
“I would say that one very important point in digital products apart from the pure software part is that it’s actually analog design,” Heeb said at the start of our interview. If there was a broad theme, that was it. “Even if the signals or the electrical signals are supposed to be digital, basically just two levels, a zero and a one, as soon as you get into an electronic board, they are actually analog signals, current or voltage flowing through components. That is especially true, for instance, for clock signals. If you just consider clock signals as being a shift between two values between zero and one, you don’t really get what clock is. The most important point in clocking is in the time domain, with finite resolution. Basically, it boils down to an analog signal again.”
Digital has two faces. On the one hand, it’s symbolic; that’s the “digital” part. Ones and zeros can be stored, read, and processed almost error-free (almostbecause there can be computational errors due to the imprecision of digital data). But when you introduce time into the pictureas you must in audioand anytime you require signals to be transmitted from place to place, it matters how those ones and zeros line up. Analog concepts like noise and distortion become important, not just for the analog part of the process but in the digital conversion itself. Those imperfections may be too small to cause inaccurate calculations, but that’s not the point. The problem comes at transitions: Precisely when does zero become one, and vice versa?
This is hardly a new idea. The concept of jitter has been around as long as digital audio itself. (Longer, actually.) But isn’t it a solved problem? Perhaps not. In an “on background” interview some months ago, a different well-known designer told me that even very small amounts of jitter can affect ultimate performancemuch smaller than, eg, the results of the Miller-Dunn J-Test used by Stereophile. Indeed, many digital designers I’ve spoken to seem laser-focused (sorry) on timing accuracy.
Is that all it is thenjust jitter? That’s a big part of it, but no, not all. Also important is the timing precision of the digital conversion itself; Cossy and Heeb call it time-smearing. (So do Bob Stuart and the MQA folks.) More on this below.
The D1.5
I’ll be auditioning the D1.5 as a player, but, as I said near the beginning of the article, it is fundamentally a transport, so let’s focus first on the transport part.
Transports are relatively simple things: They spin silver discs and read the data on them. Even inexpensive transports can read data just fine; in the absence of defects, damaged discs, and intense vibrations, reading errors are rare. But if you’re a perfectionist and you’re building a transport, you acknowledge the analog nature of digital and aim to produce a datastream that’s as pure as it can be, as perfect as possible when it arrives at that last little bit of wire before the D/A conversion. To do that, you have to account for vibrations and electronic noise.
A CD rotates at midrange frequencies, in the 200500rpm range. (The spinning frequency varies depending on which part of the disc is being read.) SACDs spin faster than that. Any eccentricity in the disc will cause vibrations at those frequencies. Especially inside an electric or magnetic field, vibrations can translate directly to electrical noise.
If you want to solve this problem, you start by leaving off components, such as ceramic capacitors, that translate vibrations directly into electrical noise. (As Heeb mentioned in our interview, a ceramic capacitor is essentially a microphone.) The other thing you must do is keep vibrations away from wires and other circuit elements.
“In the D1.5, we completely redesigned the optical unit,” Cossy said in the interview. In the D1, which used a high-quality drive from another manufacturer, the optical units were “mounted on very thin steel plate and inverted rigid dampers. There is a resonant frequency of this block which is between 300 and 800Hz.
It is exactly where we don’t want it to be”right in the midrange and right in the range of the spinning CD.
So, in moving from the D1 to the D1.5, they redesigned the optical unit. The optical pickup and motor are now mounted on a brass “sled.” The unit as a whole now “weighs 1.5kg instead of a few hundred grams.” Resonant frequency is inversely proportional to mass, so that brings the resonant frequency way down. “We’ve been able to lower the resonant frequency to 25Hz,” Cossy said.
Alpha GEL dampers (footnote 1) isolate the mechanism from the chassis and the chassis from the mechanism; these dampers are “fine-tuned to filter vibration down to AC mains frequencies,” Cossy said. The chassis itself is reinforced with a rigid support frame made with more than 4lb of machined aluminum.
One reason some peopleincluding meare fond of turntables is that they’re simple, mechanical devices. I like CD transports for the same reason: They employ principles of design that are easy to understand. As a transport, the D1.5 checks all the boxes.
Built in to the D1.5after the transport but before the add-in DACs installed in the “card cage”is a DSP unit that serves two main purposes. It decodes MQA CDs and upsamples CD data.
As a player, the D1.5 is capable of full MQA-CD decoding, which is followed by upsampling to DXD. Via its proprietary CH Link connection, the D1.5 can output “MQA core” data stored on disc, “unfolded” to 88.2kHz, equivalent to what Roon or Tidal output with MQA-encoded files if you don’t have an MQA-capable DAC. MQA says this is better than CD but not as good as all-out, fully decoded MQA.
Conversion
Cossy: “The DAC chips we are using are the WM8742.” It’s a sigma-delta chip produced by Wolfson/Cirrus Logic. (Cirrus bought Wolfson in 2014.) It became clear to me during this interview that CH Precision uses these DAC chips only for their core conversion function, ignoring the chip’s peripheral features that are performing other critical calculations in software. “These chips accept eight times base frequency, so it’s DXD rate. What we do prior to conversion is to transform everything into DXD, so it can be DSD to DXD inside the DSP, it can be 44.1kHz from a regular CD to DXD, or it can be MQA CD to DXD.” For MQA data, the MQA “black box” interpolation filter is used.
Early digital focused largely on the frequency domain. As a result, mistakes were made. The “Red Book” standard for CDs settled on a sampling frequency of 44.1kHz because that was the minimum rate needed to cover the full audible range (you must sample at twice the bandwidth in order to allow “perfect” recovery of the original time series, which gets us up to a sampling frequency of 40kHz) plus a narrow transition band to allow for bandwidth-limiting (footnote 2).
But the folks who defined the “Red Book” spec didn’t allow enough room for optimal filtersjust sharp, ones. Sharp and fast in the frequency domain equal broad and slow in the time domain. At CD resolution, you can get near-perfect frequency response or good time-domain performance, but you can’t have both.
“Another point, which is very specific to CH, is that we don’t believe that much in the necessity of achieving pure 20kHz, 0dB passband,” Heeb said in our interview. “That almost never happens in nature. As soon as you go back from a sound source, you will have a high-frequency rolloff in any case.”
The D1.5’s converter “has been designed on a set of principles that we recognize at CH as being important for proper digital audio reproduction or audio reproduction in general. One of the key points is to limit the time smearing of those filtersthat is, limit basically the time dispersion that a sample would bring once it is passed through the filter.”
Even ignoring jitter, and beyond the narrow transition band mentioned above, the traditional approach to digital conversion has some time slop in it due to the fact that the sampling “kernels” usedthe mathematical functions used to divide audio into samples in ADCs and then to reassemble it into a continuous whole in DACsare longer in the time domain than they need to be. They’re too slow.
To address this issue, the CH Precision interpolation filter utilizes splines, an algorithm that carries the acronym PEtER. A spline is a certain kind of mathematical function, a smooth, piecewise polynomial.
“It’s compact support,” Heeb said. “This is exactly what I was talking about before when I told you we want to reduce the time smearing. Time smearing is basically if you put a single pulse through the system, if you have a filter with a very long impulse response, that single sample will extend over a large number of samples. We prefer to use splines, which have a much more compact support, which makes it so that when the sample goes in, what comes out has, in our case, [no more than] 100µs of pre-ringing and post-ringing.” 100µs is the target because it’s a level of timing precision where errors are thought to be audible. It’s a conservative figure; I’ve seen estimates in the literature as low as 56µs.
Footnote 1: Alpha GEL is a trademark of the Taica corporation of Japan.
Footnote 2: Here’s the first problem for those who believe that bits-is-bits: The perfection Shannon’s theorem promises can be realized only for information that is strictly band-limited, to half the sample rate. Which, for real music, means that you must low-passfilter the signal before you can convert it. You can argue that nothing matters if you can’t directly hear it, but if your standard is to perfectly recreate what was captured on the recordingwell, there goes “perfect.”
