There is a vast array of digital converters on the market, but relatively few offer 96kHz capabilities, and fewer still provide built‑in sample‑rate conversion. Hugh Robjohns checks out the latest offering from Digital Audio Denmark.
Digital audio converter boxes are rarely exciting things to look at! Typically, they are 1U rackmount affairs with a modest handful of front‑panel buttons and a couple of LED bar‑graph meters. It's not as if a converter is much of a hands‑on tool either — once set up it is unusual to have to make any adjustments, so the machine simply sits in the rack doing the job it was designed for, installed and forgotten! However boring converters may seem, though, their role in the overall quality of recording can not be underestimated — the A‑D conversion defines the maximum quality of the chain and the D‑A conversion determines the resolution of the window through which you monitor your recording.
The new ADDA 2402 converter from Digital Audio Denmark is no more exciting to look at than any other combined A‑D and D‑A unit — it is just another 1U rackmount box with a handful of simple buttons and the obligatory LED meter. In fact, it simply goes about the job of converting between the analogue and digital worlds with the minimum of user intervention.
However, this particular unit is a little more interesting than some of its rivals. For a start, as well as a pair of stereo A‑D and D‑A converters (both supporting all sample rates up to 96kHz at 24‑bit resolution), the unit also incorporates sample‑rate conversion between digital input and output. Another feature is an extremely accurate word‑clock generator, with the ability to synchronise to an external clock signal if required. The unit's anti‑alias filtering is also to a tighter specification than normal. Much is made of this last feature, which is claimed to minimise the risk of intermodulation distortion (see box). Finally, and perhaps most importantly, it is rather more affordable than some of its peers and will undoubtedly steal sales from the Apogee PSX100, which is deemed to be its main competition.
The Guided Tour
Starting at the rear of the 2402, analogue inputs and outputs are catered for with electronically balanced XLRs as well as unbalanced quarter‑inch jack sockets. The alignment between analogue and digital levels is set internally with no provision for user adjustment, but the differing reference levels appointed to the XLRs and jacks should cater for most installations. The unit is calibrated at the factory such that a peak level of +18dBu on the XLRs equates with a 0dBFS digital signal, whereas the same digital level relates to +4dBu on the unbalanced jack inputs (and equates to +2dBV). Recalculating these figures for the nominal analogue operating levels means that the professional alignment level of 0dBu corresponts to ‑18dBFS (the EBU standard); and the semi‑pro level of ‑10dBV to ‑10dBFS.
The digital interconnectivity is simple, but fairly flexible. Both the digital output and the digital input are equipped with XLR, S/PDIF and optical TOSlink interfaces; all three outputs carry the same signal, with the input being selected using a front‑panel button. For the high‑sample‑rate modes the interfaces operate in the single‑wire (double‑fast) format and are therefore compatible with most other high‑rate digital devices, such as those from TC Electronic, for example.
Also on the rear panel is another XLR input socket, provided specifically to accept an AES11‑standard word clock signal (unfortunately, there is no provision for a standard TTL word clock signal on the familiar BNC socket). External clocking is necessary when locking the converter to a house reference, or when using multiple 2402s together, for example. The review model had no word clock output facility at all (other than via the AES digital output), but I gather that all future models imported from the manufacturer will incorporate a BNC socket supplying a word clock output — and hopefully without any change to the unit price given below. Finally, the ubiquitous IEC mains inlet features an integrated power switch and fuse holder. The rack case measures a modest 275mm deep and weighs just 3.3kg.
The front panel is just as straightforward as the rear, with only seven buttons to configure the whole unit. The white painted panel has clear black legends alongside every status LED, selections being controlled through the smart grey buttons. The machine is simple and utterly intuitive to use, although I would have liked to have seen some different‑coloured LEDs employed to make the current configuration a little more obvious from across the room — all the indicators are orange in the present version.
The first button on the extreme left is the analogue input selector for the A‑D section, which chooses between the rear‑panel XLR and jack connectors, the current status being indicated through a pair of the LEDs. An unlabelled function accessed by double‑clicking this button activates a permanent peak‑hold facility on the metering — the selected input LED flashes to indicate that this condition is enabled, while a second double‑click cancels and resets the peak hold mode.
The LED bar‑graph meter spans a range from ‑60 to 0dBFS, with 2dB steps from ‑12dBFS and an automatic one‑second peak hold when the level exceeds this point. Interestingly, instead of following the usual approach of showing the true converted digital peak levels on the meter, Digital Audio Denmark have chosen to implement the IEC268‑10 PPM standard. This requires a 10mS integration time, which means that the meter will, inevitably, under‑read on the briefest of transient peaks. Of course, metering tends to be a personal thing with different engineers preferring various meter ballistics and scales, so I'm sure as many will favour this approach as those who dislike it. In reality, I would suggest that few people rely on the meters of their converters anyway, and probably make more use of those on their computer or hardware recorders, consoles, or even bespoke metering systems. In practice, I found that as long as I kept signal peaks in the first couple of yellow LEDs just above ‑12dBFS, everything worked just fine.
The next button along from the meter is the digital input selector for the D‑A section; the associated LEDs illuminate to show whether the AES‑EBU, S/PDIF or Optical input has been selected. The fourth lamp (labelled 'carrier') lights when the machine has recognised and synchronised to the selected input signal.
The digital output from the unit would, under normal circumstances, be supplied from the A‑D converter, but another button allows a cross‑link to be made to the selected digital input (LEDs again indicate the status). Not only does this allow the converter to serve as a useful digital (and analogue) router and format converter, but as the digital through path incorporates a sample‑rate converter, material of incompatible sampling rates can also be accommodated with ease. The output sampling rate is always that of the A‑D section, which is determined by the next button. There are six possibilities, all with their own LED indicator: 32, 44.1 and 48kHz and then doubled rates of 64 (a very non‑standard and potentially problem‑making sample rate), 88.2 and 96kHz. Changing sample rate takes roughly three seconds, as an automatic recalibration sequence has to be completed each time the rate is changed.
The A‑D stage uses a CS5397 delta‑sigma chip (see box) running at 64/128x oversampling to generate 24 bits per sample, and the digital inputs also all accept 24‑bit signals. However, there are many occasions where a lower resolution is required — such as when recording to DAT or CD‑R, for example — so the next button allows the digital output resolution to be redithered to 16, 18, or 20 bits using the various noise‑shaping curves built in to the CS chip, or to pass the full 24 bits.
The penultimate button selects the word clock sync source between internal, external and digital input. The internal clock generator is highly accurate and specified as better than 5ppm — an order of magnitude better than that of the Apogee PSX100! However, no figures are given for the (arguably more important) jitter performance. The external sync source listed on the panel comes via the AES11 input socket on the rear — the sample‑rate LEDs indicating the detected incoming sample rate in this mode — and the digital input source is that selected by the digital input button. If no external clock source is available the machine will not allow that source to be selected and, should the external clock fail, it will automatically revert to the internal clock until the external source reappears. Finally, the power button on the extreme right of the panel is actually a 'standby' switch, the mains breaker being on the rear panel.
In Use
Connecting and configuring the 2402 is child's play and I quickly discovered that the machine performs extremely well. Although it costs significantly less, I compared it in a head‑to‑head with my own Apogee PSX100 which provides similar core converter and I/O routing facilities. However, the Apogee also incorporates ADAT and TDIF interfaces, with bit‑splitting options and both dual‑ and single‑wire high‑sample‑rate options. The DAD does not share these features but does provide a sample‑rate converter instead.
The differences between the true Nyquist anti‑aliasing filter in the DAD and the design used in the Apogee (see box on page 190) are audible, thanks to the resulting slight reduction in extreme high frequencies in the former, when compared directly using harmonically rich analogue source material (12‑string acoustic guitar is a good example). However, I didn't feel this was, necessarily, a negative thing — the unit had a more 'analogue' quality to its sound in some ways, possibly as a direct result of this drooping top end!
My casual listening tests were rather inconclusive as to whether Digital Audio Denmark's claims regarding AID represent a serious failing of conventional converters. I was not aware of any problems from the 12‑string guitar when comparing the acoustic source, the output from a Mackie mixer, or the A‑D‑A chain through the Apogee (at 24‑bit, no dithering and internal clock, 48kHz sampling). The same can be said of the 2402 configured in the same way, albeit with a marginally softer top end. In both cases, operating at 96kHz produced noticeably better results than at 48 or 44.1kHz, with a greater sense of air and clarity, and nothing to choose between the DAD and the Apogee in terms of sound quality. Stereo imaging was rock solid on both units, confirming the stability of the internal clocks. The 2402 also seemed to perform equally well when locked to an AES‑EBU clock output from the Apogee.
Groovy Dad
Whichever way you look at it, the DAD 2402 represents very good value for money. It boasts an excellent dynamic range and noise performance in both A‑D and D‑A stages, being virtually identical to the Apogee, yet costs significantly less. The sample‑rate converter could be a very useful facility in many applications, and the flexibility of input and output formats and routing is well worth having. The absence of ADAT/TDIF interfaces and bit‑splitting on the 2402 means that the Apogee will remain the converter of choice in some situations, but the attractive price of the DAD machine makes it a very affordable and cost‑effective alternative.
The avoidance of intermodulation distortion by using the correct Nyquist filtering is technically valid, but I'm not entirely convinced that there is a real problem with the conventional 45/55 filters used in so much equipment anyway. The extreme HF harmonics tend to be relatively weak in the majority of music and those aliased as a result of the imperfect filtering are weaker still, so I think any intermodulation products will be inaudible in most circumstances. In any case, it is just as likely that the high harmonic components of complex music could intermodulate in exactly the same way, even in a completely analogue system. The down side of the DAD approach, although technically laudable, is a slightly curtailed high‑frequency response which can be audible at 32 and 44.1kHz sample rates when compared with the original, or the output of another converter. However, many may see this as an advantage or an irrelevance — it certainly shouldn't put you off an audition of this impressive machine.
Aliasing Intermodulation Distortion
One of the defining elements of any digital audio system (along with the sample rate, bit resolution, and jitter) is the response of the anti‑alias filter. It has become custom and practice in the audio industry to employ delta‑sigma converters operating at a very high sampling frequency, followed by a digital decimation filter to reduce the sample rate to something more practical, whilst correspondingly increasing the bit resolution. The kind of digital filter response commonly used is referred to as an 0.45/0.55 filter, which means that the 'brick wall' filter slope starts at 45 percent of the sample frequency and achieves full attenuation by 55 percent. In terms of a 44.1kHz system that means the filter starts to turn over at 19.85kHz but doesn't reach full attenuation until 24.26kHz. At the Nyquist frequency (half the sampling rate), it will only have provided around 8.5dB of attenuation.
Digital Audio Denmark argue that this kind of filter slope may be acceptable for D‑A reconstruction filters, but is inappropriate for high‑quality A‑D anti‑alias applications because in failing to reject high‑frequency audio material above half the sampling rate (the Nyquist Frequency being 22.05kHz in the previous example), it allows the possibility of 'Aliasing Intermodulation Distortion'.
Essentially, their argument is that any input signals with frequencies between 22.05 and 24.26kHz will be aliased, or mirrored, into the area between 19.85 and 22.05kHz. Naturally, these high‑frequency signals will be relatively low in amplitude (both because their level in the source signal is likely to be low, and because they have been attenuated by the effects of the filter slope) and probably not directly audible at all. However, any non‑linearities of the recording and replay chain — the main culprit being the monitoring loudspeakers — will allow intermodulation between these aliased signals and other high‑frequency components in the wanted part of the audio band. This intermodulation will result in non‑harmonically related distortion, typically at relatively low frequencies, which may well become audible!
Digital Audio Denmark claim this potential problem of AID is one reason why 96kHz systems are often thought to sound better than 44.1kHz systems. There are negligible signals near the Nyquist Frequency in a 96kHz system and, consequently, there are no audible intermodulation products. Their solution to the problem is to adhere strictly to the Nyquist rules, and employ a digital filtering algorithm which produces maximum attenuation at half the sampling rate. Currently, Crystal manufacture a suitable decimation filter system (the CS5397 chip) where (for a 44.1kHz sample rate) the filter slope starts at 18.1kHz to reach full attenuation by 22.05kHz. In other words, it provides a 0.40/0.50 filter and is roughly 12dB down by 20kHz.
In practice, although this device provides a more accurate anti‑alias filter response, hopefully avoiding AID, it also has a greater impact on the wanted audio, with a measurable tail‑off towards the top end of the band (‑3dB by 19.3kHz). A different set of compromises....
Pros
- Good performance/price ratio.
- Flexible routing and digital format conversion.
- Internal sample rate converter.
- Precise anti‑alias filtering.
- High sample rate facilities.
- Precision internal clock.
Cons
- No bit‑splitting capability.
- Marginally dull top end at 44.1/48kHz.
- Dodgy 64kHz sample rate!
- No BNC word clock input.
Summary
A very cost‑effective, high‑performance A‑D and D‑A, housed in a neat 1U rackmount box. Flexible internal signal routing and format conversion are useful, as is the sample‑rate converter, but there are no bit‑splitting options or ADAT/TDIF interfaces.