Reviewer
: Srajan Ebaen
Financial interests: click here
Sources: 27" iMac with 5K Retina display, 4GHz quad-core engine with 4.4GHz turbo boost, 3TB Fusion Drive, 16GB SDRAM, OSX Yosemite, PureMusic 2.04, Tidal & Qobuz lossless streaming, COS Engineering D1, Metrum Hex, AURALiC Vega, Aqua Hifi La Scala MkII, SOtM dX-USB HD w. super-clock upgrade & mBPS-d2s, Apple iPod Classic 160GB (AIFF), Astell& Kern AK100 modified by Red Wine Audio, Cambridge Audio iD100, Pro-Ject Dock Box S Digital, Pure i20
Power & integrated amplifiers: Pass Labs XA30.8, FirstWatt S1, F6; Crayon Audio CFA-1.2; Goldmund Job 225; Gato Audio DIA-250; Aura Note Premier; Wyred4Sound mINT; AURALiC Merak [on loan]
Loudspeakers: EnigmAcoustics M1, Albedo Audio Aptica; soundkaos Wave 40; Boenicke Audio W5se; Zu Audio Submission; German Physiks HRS-120, Gallo Strada II w. TR-3D subwoofer
Cables: Complete loom of Zu Event; KingRex uArt double-header USB; Tombo Trøn S/PDIF; van den Hul AES/EBU; AudioQuest Diamond glass-fibre Toslink; Arkana Research XLR/RCA and speaker cables [on loan]
Power delivery: Vibex Granada/Alhambra on all components
Equipment rack: Artesania Audio Exoteryc double-wide 3-tier with optional glass shelves, Rajasthani hardwood rack for amps
Sundry accessories: Acoustic System resonators
Room: Irregularly shaped 9.5 x 10m open floor plan with additional 2nd-floor loft; wood-paneled sloping ceiling; parquet flooring; lots of non-parallel surfaces (pictorial tour here)
Review component retail in Europe (no VAT): €4'090


After our October 2014 post of Cees Ruijtenberg's Dac One, I'd asked whether, in multi-paralleled current-output with Ω-matching coupling transformer, this Transient module would inform Metrum's next converter. "Yes, we will use it for a new product to hopefully come in December. Prototypes already work very well and I am preparing pre-production units. Our biggest challenge is the housing. We use more aluminium than before and don't like any scratches. The new model will use 8 converters, four per channel in fully balanced operation. 24-bit data will split up over two DAC clusters. The first cluster runs from 0 to about 65dB and is taken over by the next cluster down to -144dB. Summing of these clusters is done in the analogue domain. This gives very low distortion over the entire range. Splitting up the data is the job for our FPGA module. By the way, for me the DAC module was the heaviest job to finish. It held all of my attention almost 24/7 for more than a year and I concentrated mostly on its I/V stage. The Hex DAC was my reference throughout and very hard to beat. Launching the separate Transient modules was due to the enormous R&D costs we incurred in the past. The Transient blocks will reduce our production costs when more companies get interested to use them for OEM applications.


"When using I/V transformers in the current-output scheme, we had to use a lot of electronics outside the DACs due to the two balanced data clusters per channel. Therefore we focused on a very good I/V section. The Lundahl iron on each board is based on Aurix technology to create the single-ended output. Compared to the Hex, there is almost no load on the balanced outputs due to this technique." When I asked Cees whether he'd beaten the Hex, "this will be up to the people and probably is a matter of taste. My first experience was with Norma Winstone's Distances and the cut "Ciant". There was more information present, hence many recordings followed. For that I only have one technical explanation. It's all about speed. The funniest thing was that this approach starting all over from scratch with our own converter chips produced, without yet the FPGA module, a first result very close to the Hex. As a matter of fact, there is not one shared part with the Hex. A recording engineer called it my signature. From my point of view, that is wrong. I am never busy with creating any kind of sound. I can only confirm if the result is closer to my experience with life performances; or whether some info is noticed which was inaudible before."


If you paid attention, something about this Cees chat should have stood out. To back up, consider this morsel from Chord's Rob Watts who explained his volume control in the Chord Hugo TT. For our purposes, I've italicized the relevant passage. "...The problem with digital volume is not the function itself. With the appropriate bit width and noise shaping/dithering strategies, a perfectly transparent distortion-free volume function is possible. The problem occurs in the DAC itself. DACs are not very good at resolving low-level signals accurately. Digitally attenuating the signal puts more stress on the low-level capability of the DAC..."


One needn't even attenuate signal prior to conversion (apply digital volume control) to challenge a DAC chip's low-level linearity. The less and least significant bits already do that at full signal. If we look at the Pavane's published SN/R of -144dB, it's some 15-20dB better than the current standard. Clearly Cees pulled a fast one; as well he should if speed was the thing. In ultra simplistic terms, our man splits his 24-bit data into two clusters. The first gets bits 1-12 as the most/more significant bits, augments them with empty bits below and processes this normally. Cluster two gets bits 13-24 as the less/least significant bits but moves them into its ladder DAC's top segment. This neat trick has them processed just like the first cluster. Once in the analog domain, our (sexist remark alert!) hotter half is attenuated back down to its proper loudness values. Only then is it recombined with the other half. And so our crafty Dutchman applies a temporary dynamic expander action in the digital domain to improve the low-level linearity of the D/A process which naturally wants to get less and less precise as recorded signal amplitude in the lower/lesser bits diminishes and gets mixed up with/in digital noise.


Before you slap your head in well-practiced "duh, dude" fashion of "why hasn't anyone else figured this out sooner", Cees could tell you a story. Which for obvious reasons he won't. "We are careful not to give too much info to the world as it took me two years to get the best forward-correction algorithm which above all was based on long listening sessions. But in fact, this is how it does work. The unused four bits of each module are supplied with zeroes so we use twelve. With both clusters it means that normal switching noise is immeasurable. By attenuating the LSB part to the correct analog level, we push switching noise down to about -170dB which is somewhere deep in the noise floor of parts and purely theoretical." The FPGA executing the code for the forward-correction logic operates with a 400MHz clock speed.