If you place importance on the spec of your recording system's analogue-to-digital converters, then you shouldn't overlook the quality of your D-As either. We check out a surprisingly affordable high-quality example from the States...
The ultimate sonic quality of any digital audio system is defined by the quality and resolution of the analogue-to-digital conversion used — any loss of signal resolution at this early point in the signal chain cannot be recovered later. This simple truth is widely understood by the majority of serious amateurs and professionals who have embraced digital audio recording, and most allocate a significant part of their equipment budgets to purchasing the best A-D converters they can afford.
However, the other end of the same argument is that you won't be able to hear the limitations of a sub-optimal A-D if the digital-to-analogue converter is incapable of resolving the detail encoded within the digital signal. In short, you require a superb D-A if you wish to assess and appreciate the quality of the best A-Ds! Sadly, as we all know, the very best converters — and I'm thinking of the likes of DCS and Prism — always cost a great deal of money, putting them outside the reach of most of us. Even the next best offerings from the likes of Apogee, RME, Audio Design and so on, usually cost well over £1000 in the UK.
An American company has a product that has forced me to re-evaluate these ideas, though. Benchmark Media Systems are based in the state of New York and manufacture a range of professional digital and analogue audio products, many incorporating rather intriguing technology. The DAC1 is one such device — a stereo, 24-bit, 192kHz D-A converter that costs less than £1000 in the UK, but performs at a level that eclipses more familiar devices costing two or even three times as much.
The fundamental reason for such a high standard of performance is the DAC1's almost perfect rejection of interface and conversion clock jitter, which causes unstable stereo images and varying anharmonic distortion in D-As, colouring and clouding low-level audio detail (see the box later for more on this). The DAC1 avoids this with a novel approach which eschews the use of phase-locked loops (PLLs), the conventional means of dealing with jitter, as these can themselves cause problems.
The DAC1 is a compact self-contained stereo D-A, housed in a sturdy half-rack-width, 1U case that extends a modest 8.5 inches (216mm) behind the rack ears. The substantial sculpted aluminium front panel contains standard rack ears on both sides, so one side can be bolted into a rack frame in the usual way, and the other side is attached using an optional 'coupler' either to another DAC1 or a bespoke blanking panel to fill the void to the other side of the rack (a matching stereo A-D is currently under development, and would make an ideal rackmount partner for the DAC1).
The black painted front panel carries an engraved Benchmark logo, a red DAC1 name tag, and white function labels. A recessed toggle switch selects one of three digital inputs (co-axial, XLR or optical) and a trio of LEDs indicate the presence of power (an essential blue LED), a clock error, and non-PCM data — ie. a DTS- or AC3-encoded data stream (both red). The input selector switch can be disabled via an internal jumper link and one specific source selected permanently. There's also a pair of full-sized headphone sockets (wired in parallel) and a large knurled and detented volume control. The latter always controls the volume of the headphone outputs, but can be switched to control the analogue line output level if required — making this D-A converter an ideal front end for active monitors.
Moving around to the rear panel (shown overleaf), the analogue outputs are provided on a pair of isolated unbalanced phono sockets as well as a pair of electronically balanced XLRs. Another three-position toggle switch configures the analogue outputs to operate at a fixed 'calibrated' level, or to be varied according to the front-panel level control. Alternatively, the outputs may be muted if the switch is placed in its unmarked central position.
The XLR outputs are aligned at the factory to provide +24dBu for a 0dBFS input (+4dBu at -20dBFS), although this can be adjusted in three ways. Most obviously, switching the front-panel level control into service allows the output levels to be continuously varied. The factory-calibrated level can also be adjusted via a pair of multi-turn trimmers on the rear panel, allowing a range of +5 to -15dB from the factory setting. Furthermore, if the DAC1 is to be used to drive sensitive inputs (or active monitors), internal links can be used to provide a further 10, 20, or 30dB of attenuation.
The unbalanced outputs are designed to provide a level of -10dBV when the XLRs are aligned for +4dBu (with a -20dBFS input), and are affected by any trimmer adjustment when switched to 'calibrated' mode, but not by the XLR attenuator link settings.
The digital input options comprise a TOSlink port for optical S/PDIF signals, an XLR for AES-EBU signals, and a BNC connector. The last of these will accept either the unbalanced form of AES (AES3-id) or, using a supplied BNC-phono adaptor, a standard S/PDIF signal. In both cases, the interface provides the required 75Ω impedance, although an internal link can be removed to provide a high-impedance 'bridge' mode, which is useful if a signal needs to be daisy-chained between several devices.
The DAC1 incorporates a mains power supply (no tacky wall-warts here!) and is fitted with an IEC mains socket with integral voltage selector and fuse holder for 115 or 230V operation. There is no mains power switch on either the front or rear panels — the DAC1 is intended to operate continuously to maintain a constant internal temperature for optimal results. For more on just how much the operating temperature can affect the unit's performance, see the box overleaf.
Jitter is an inherent part of any digital interface. Whenever data with an embedded clock signal is passed from one device to another, the cable will act to blur the clocking edges and produce a degree of uncertainty over the precise position in time of each clock edge. Provided the data is being passed between digital processors this is of no relevance whatsoever (assuming the jitter is not so excessive that the interface breaks down completely!). However, where the digital data is converted to analogue (or vice versa), the clock timing is absolutely critical to the performance of the sampling and reconstruction processes. Timing errors translate into anharmonic sidebands in the signal — not entirely dissimilar to intermodulation distortion — and these dramatically reduce the resolution of low-level detail and compromise the stability and clarity of the stereo image.
Whereas most A-Ds operate from an internal crystal, and thus potentially have very low conversion jitter, D-A converters have to derive their clock from the incoming data, which will inevitably suffer from some degree of jitter. The jitter on a 24-bit signal arriving at a simple D-A with no (or poor) jitter reduction can result in the converter performing at 16-bit resolution, or even less!
The usual way to remove jitter is to use a circuit called a phase-locked loop (PLL) which detects the incoming clock and essentially averages out any timing variations to produce a more stable reference clock. However, there is no such thing as a free lunch, and PLLs suffer from various technical trade-offs. For example, a narrow-bandwidth PLL which will give fairly good jitter reduction will take a very long time to synchronise to a new input, and varispeed sources may not be handled well at all. Also, PLLs don't provide equal jitter rejection at all frequencies, so some jitter components almost always get through. There are of course ways to improve jitter performance — multi-stage PLLs are a common approach, but the problems mentioned are very hard to overcome completely.
Benchmark's approach is to use a very high-quality sample-rate converter to isolate the incoming jittery clock completely, and instead refer the D-A conversion clock to a stable local crystal reference — a technology they refer to as 'UltraLock.' The same technique can be used in reverse to derive a totally stable A-D conversion even when the output is referenced to an external jittery clock — and Benchmark use exactly this approach in their A-D converters.
Internally, the unit is built to high standards using mainly surfacemount components on a single PCB covering the majority of the case floor. The linear power supply employs a relatively large torroidal transformer and large reservoir capacitors, with the analogue audio running on ±18V rails and separate regulation for the 5V and 3.3V digital rails. All of the analogue signal handling is done with industry-standard NE5532 dual op-amps, of which there are three per channel.
The headphone amplifier employs Benchmark's HPA2 design (which retails for $150 as a stand-alone unit). This is based around another NE5532 op-amp per channel augmented with a BUF634 high-speed buffer chip to provide higher current handling capability and fast slew rates. Since the two headphone sockets are wired in parallel, it is important to use identical models of headphone if both outlets are used at the same time (or at least models with the same rated impedance) to avoid significantly different volumes.
The digital signal path is quite unusual — but this is the reason for this D-A's remarkable jitter-suppression qualities. The input source selection is made with an AKM AK4114 transceiver chip which not only selects the input, but also decodes the various AES and S/PDIF data formats, applying automatic digital de-emphasis when required, and controlling the front-panel warnings for non-PCM data and incoming clock problems. The input clock is recovered from the selected digital input signal using a very wide-bandwidth PLL that allows it to lock to very poor digital signal sources with large amounts of jitter. This allows the DAC1 to recover good audio even when connected to poor-quality sources at the ends of long cables.
Although wide-bandwidth PLLs are generally bad for conversion clock recovery (because they won't remove the jitter), in this case the clock recovered by the AK4114 is only used to feed the input side of a sample-rate converter. The recovered input clock is not related to the final conversion clock at all, and to make sure it has no influence at all, the circuit board layout has been carefully designed to keep the jittery input clock well shielded and away from the crystal-referenced output clock circuits.
The bitstream is then passed to an Analog Devices AD1896 stereo asynchronous sample-rate converter (SRC) with a 142dB dynamic range, which is used to oversample the incoming digital signal, accepting any input rate from 32 to 192kHz. This particular SRC chip was chosen for its sonic transparency along with its ability to reject jitter from the input signal — in fact Benchmark claim the jitter rejection is 'nearly perfect.' This is because the chip's input PLL has a bandwidth of only 3Hz, giving jitter attenuation of 100dB at 1kHz, and over 200dB at 100kHz.
Conceptually, the SRC chip oversamples the input signal by a ratio of 1,048,576:1 — so a 48kHz input signal is effectively oversampled to a new sample rate of 50.3GHz. It is then downsampled again to a fixed-output sample rate derived from a stable crystal clock reference at the rather peculiar sample rate of 110.6kHz. As well as allowing the D-A conversion clock to be completely isolated from the input signal's jittery clock, this approach also has the advantage of providing excellent alias rejection.
It is this asynchronous sample-rate conversion that provides the isolation between the fixed, stable, crystal-controlled and jitter-free internal D-A clock, and the inherently jittery incoming data — a technology Benchmark refer to as 'Ultralock'. The fixed clock reference is derived from a 28.322MHz crystal mounted right next to the SRC and D-A chips, operating at 256 times the required rate of 110.6kHz.
The output from the SRC is passed to an Analog Devices AD1853 multibit sigma-delta stereo D-A converter chip. This is Analog's first D-A to support all sample rates from 32 to 192kHz, and it has been designed specifically for high-end DVD-A players, mixing consoles and digital effects processors. The D-A converter is driven directly from the crystal oscillator, and its clock signal is delivered using a shielded controlled-impedance transmission line arrangement on the PCB. Again, lots of careful attention has been paid to the PCB design to isolate this reference clock from the input clocks, the audio signals, the power supply, and internal RF fields. The circuit board in current DAC1 units is a 'Revision C' design, highlighting just how critical the board layout is to the overall performance.
The choice of a 110.6kHz sample rate for the D-A conversion may seem odd, but there are some very specific reasons for it. Firstly, this places the DAC's reconstruction filter transition band above the bandwidth of a 96kHz digital signal. Secondly, the AD1853 chip actually provides its best performance specifications when operating between 100 and 120kHz sample rates, providing a 20dB improvement in the performance of its digital reconstruction filter to remove aliases. Thirdly, all current D-A converters (including the AD1853) operate relatively poorly at quad sample rates (176.4 and 192kHz) because to accommodate such high rates they have to use reduced internal oversampling ratios for the digital filters. Often this ratio is as low as 2, resulting in much poorer reconstruction filter characteristics, and thus poor alias rejection.
The use of the 110kHz D-A sample rate means, of course, that 176.4 and 192kHz sources have to be downsampled, reducing the audio bandwidth from roughly 96kHz to about 55kHz... but Benchmark claim that the improved filter performance more than compensates for the reduced ultrasonic bandwidth. For the more practical standard and double sample rate sources, there is no downside at all, and the approach provides a completely jitter-free D-A conversion.
The aspect that initially troubled me with this unusual approach to D-A conversion was the fact that the input is always passing through an SRC using complex non-integer ratios. The received wisdom is that non-integer SRC processes produce potentially audible artefacts, but Benchmark claim that the design of the AD1896 SRC chip is such that it does not require (or benefit from) integer ratios between input and output, and its performance exceeds that of the D-A converter, so it is not a quality bottleneck. With this in mind, I put it to a listening test.
After I had finished the review of the Benchmark DAC1, I had an opportunity to compare two identical units side by side. Both were fed with identical optical S/PDIF signals from a commercial signal splitter, and both were feeding identical active monitors, set up side-by-side. The only difference between the two DACs was that one had been running for about 90 minutes while the other had spent the previous two hours in a cold car travelling across the country before being plugged in.
The difference in sound quality between the two units was stunning. The unit which had been running a while sounded smooth, detailed and worth every penny while by comparison, the cold unit sounded thin, harsh, granular, and lacking in depth and resolution. The difference was so pronounced that we all became concerned that it might actually be broken in some way! However, over the space of about 20 or 30 minutes, the cold unit warmed up and the sound evolved to provide all the fine qualities and attributes already described, finally matching that of the first model.
The sonic benefits of allowing electronic equipment to 'warm up' are well known, but I have never experienced such a dramatic example as this before. This issue doesn't arise during testing for SOS, because I always plug review units in and have them running long before I start to make any aural assessments, but I thought it worthwhile bringing this to your attention as an important quality when you are auditioning equipment.
I rigged up the DAC1 alongside my trusty Apogee PSX100, each feeding through a simple changeover switch to my reference Bryston 4B amplifier and PMC IB1 three-way monitors. Using a Meridian CD player as the source and a Neutrik Minilyser as the measurement tool, I calibrated the output level of the DAC1 to match that of the Apogee, and then sat back to listen. I started by familiarising myself with the Apogee, playing a selection of commercial CDs as well as some of my own recordings. Then I switched to the DAC1 and played through the same selection of material... and I was seriously impressed.
The PSX100 is, as converters go, a pretty good mid-range device. It's not in the same league as the high-end converters, but it resolves low-level detail and imaging information pretty well, and is certainly a lot better than most of the built-in converters in recorders, desks and soundcards. However, it was immediately apparent that the DAC1 was superior — most notably in a small but recognisable improvement in resolution, especially through the mid range. It provides a fantastically transparent conversion, completely free from any hint of digital 'graininess,' with a smooth, detailed high end and a really solid, full bass. The stereo image is truly three-dimensional, with not only impressive width but realistic depth cues too, and the dynamic and transient response is phenomenal!
Full of enthusiasm, I then hooked up the DAC1 to the output of my Yamaha DM1000 mixer, and remixed some eight-track 24/96 concert material from a Genex hard disk recorder. Compared with the both the Yamaha's internal converters and the Apogee, the DAC1 had a noticeable advantage, allowing minor EQ changes and subtle mix level tweaks to be heard just a little more clearly. I felt I could hear further into the mix — especially through the critical mid range.
I also tried the headphone output, and this proved equally impressive. Driving a pair of low-impedance Sony MDR7509s, the DAC1 had masses of level and drive capability, producing crisp transients with no hint of compression or clipping, even when pushing out ridiculous sound pressure levels.
Finally, I tried running the DAC1 as a monitoring preamp, using the front-panel volume control to adjust the listening level of a pair of powered monitors (a review pair of PMC TB2As). The relatively high output level of the DAC1 (+24dBu as standard) proved way too much for comfortable listening with the volume anything more than about a quarter of the way around. Not only was it difficult to adjust listening levels accurately, but this also revealed the only flaw I was able to find in the DAC1: some audible mistracking between channels, causing slight but detectable image shifts.
However, after removing the lid of the unit and changing some internal jumpers to attenuate the output by 20dB, I was able to operate the volume control over a more practical range, giving greater precision and much better tracking between channels. A quick meter check suggested that when using the volume knob around the 12 to 3 o'clock positions, channel matching is better than ±0.25dB, but it deteriorated with errors over ±1dB when nearing the bottom end of rotation. That is not unusual, and not really a criticism either, and the provision of output attenuators negates the issue in a very practical way.
So, my overall impression of the DAC1 is extremely good. Its performance is quite extraordinary for the UK retail price, and it competes favourably against units costing considerably more. It is a very neat, well-designed package with an excellent feature set, including an internal mains PSU, usefully clean and powerful headphone monitoring, and the provision to configure the analogue outputs for fixed or variable levels. It can accommodate the roughest input signals with ease, extracting whatever sonic beauty is hidden within.
If you are looking to improve your monitoring, or to hear what your A-D converters are actually providing, you really must audition the DAC1. I must warn you, though, that this will be hazardous to your wallet — because you'll have to buy it once you hear it. I did!