This review page is supported in part by the sponsors whose ad banners are displayed below

Srajan Ebaen
Financial Interests: click here
Source: 2TB iMac 27" quad-core with 16GB of RAM (AIFF) running OSX 10.8.2 and PureMusic 1.94g in hybrid memory play with pre-allocated RAM, Audirvana 1.5.10 in direct/integer mode 1, Metrum Hex, SOtM dX-USB HD with Super-clock upgrade & mBPS-d2s, AURALiC Vega, Apple iPod Classic 160 AIFF-loaded, Cambridge Audio iD100, Pure i20, RWA-modifed Astell&Kern AK100
Preamplifier: Nagra Jazz, Bent, Audio Tap-X, Esoteric C-03
Power/integrated amplifier: FirstWatt S1 monos, F6 stereo; Crayon Audio CFA-1.2, Bakoon AMP-12R, Gato Audio DIA-250, Goldmund/Job 225, Clones Audio AP1/55pm [on loan], AURALiC Merak [on loan]
Loudspeakers: soundkaos Wave 40, Boenicke Audio W5, German Physiks HRS-120, AudioSolutions Rhapsody 200, Zu Audio Submission
Cables: Complete Zu Event loom, KingRex uArt and Light Harmonic Lightspeed split USB cables, Tombo Trøn S/PDIF, Van den Hul AES/EBU, AudioQuest Diamond glass-fiber Toslink
Powerline conditioning: GigaWatt PF-2 + Vibex Two 1R on amps, Vibex Three 11R on front-end components
Equipment rack:
Artesania Exoteryc double-wide three tier with optional glass shelf, Rajasthani hardwood rack for amps
Sundry accessories: Extensive use of Acoustic System Resonators, noise filters and phase inverters
Desktop system: iPod/AK100 digital transport, Aura Vita integrated, Gallo Strada II + TR-3D, beyerdynamic T5p
Room size:
Irregularly shaped 9.5 x 9.5m open floor plan combines the living/listening room, kitchen and office. Added to this space the speakers see the air volume of the entry hall and a long corridor plus the 2nd-storey 6 x 9.5m loft. Wood-panel ceiling slopes up to the loft. Parquet flooring. Lots of non-parallel surfaces ('vertical gable' windows, twin-angle ceiling, spiral staircase enclosure, fireplace enclosure).
Review component retail in EU: starting at €4'900 + regional VAT for Platinum single-ended version, €11'900 + VAT for balanced Signature version; €699 for Hydra X+

Whilst the current PCM/DSD debate gets more confusing with each passing day—experts disagree, manufacturers disagree, digital converter and recording equipment specs disagree on whether 'pure' DSD and PCM even exists on the recording never mind playback side—there's far more consensus about the ideal way to decode PCM. In theory. Say hello to multi-bit ladder-type aka R2R converters (click here for some tech talk on the subject). But why are so few actually doing it in practice today? Because such silicon has fallen out of popularity with the chip manufacturers whilst remaining NOS inventories are being depleted. Curiously the same is true for bona-fide 1-bit DSD converter chips. They've vanished for good. It's peculiar that given the current format debate moving briskly through the audiophile specialty press, purpose-engineered conversion silicon for either format is no longer being made. (Which isn't 100% true. Cees Ruijtenberg of Metrum Acoustics for example keeps building R2R DACs with currently made silicon, albeit from the non-audio industrial sector to remain unidentified.)

That's why truly resourceful digital manufacturers on either side of this fence roll their own. Using field-programmable gate arrays or FPGA, Chord in the UK replace off-the-shelf PCM encoding with their own custom code. Meitner, Playback Designs and of late PS Audio apply the same code writing chops to 1-bit DSD. Back on PCM, MSB have championed the discrete resistor-ladder approach well before a Frenchman under the TotalDAC banner launched his own take on building an R2R DAC with hundreds of precision-matched resistors.

For those not wishing to spend their own years and discretionary funds in R&D of the same, MSB's discrete converter modules are available for OEM (we've previously encountered them in the Bulgarian Thrax Maximinus) just as are Andreas Koch's FPGA-driven DSD algorithms. With the mention of MSB and OEM—the latter bidirectional—we now arrive at today's subject. Welcome the Romanian Rockna Audio WaveDream DAC. Whether single-ended or balanced, in Reference or Signature guise, it gets two or four 2nd-gen modules sourced from MSB "capable of a sustained maximum sample rate of 3MHz". Beyond that designer Nicolae Jitariu pursues his own paths just as past their custom ceramic Koch board Nagra do with their HD DAC. Interestingly Nicolae has provided his own OEM/ODM assistance to Audio Mirror, Acoustic Precision, Fase, Jadis, Nonsolomusica, PS Audio (PerfectWave transport), Wadia, Goldmund and, yes, MSB. Though his name might still be news—Rockna Audio have been around for 15 years—you could already be familiar with his work.

As though to underscore his engineering credentials, Nicolae's €699 Hydra X+ under his second lower-tier brand Audiobyte performs its battery-powered USB bridge functions with a custom-coded Xilinx Spartan 6 as another FPGA custom job. To it he adds premium Crystek clocks and I²S output via HDMI. The Hydra X1 is thus direct competition for Audiophilleo and SOtM. Obviously the upscale WaveDream DAC gets the same proprietary USB solution, not a generic M2Tech, XMOS, Amanero & Co module. But our DAC get a lot more that's not off the rack. This includes selectable digital filters written by Jitariu's team. Those exploit 15 GMACS of processing power from 58 DSP blocks and "avoid standard Nyquist rate filters" in favour of "an advanced 2000-tap convolution filter created from a Parks-McClellan equiripple combined with a raised cosine type". Even his Femtovox clocking scheme is different*. The linear power supplies get three transformers and 20 linear regulators "designed from scratch for low impedance and ultra-low noise".

* "The heart of the WaveDream clocking system is a digitally programmable low-jitter clock. Its precision is set by a 38-bit control word which gives an accuracy of <1ppb (parts per billion). This clock generation creates a very low jitter of 0.3ps or 300fs on any input sample rate. In absolute terms there are claims of lower jitter figures on the market which can be obtained only by using oven-controlled crystals at single frequencies. With sample rates varying from 44.1 to 384kHz, it’s unlikely that a DAC can work with a single master-clock frequency unless it uses an ASRC with its own set of problems. Our digitally controlled clock allows bit-transparent operation without the smearing caused by an ASRC block. But the master-clock generation technique isn’t all. To properly adjust the digitally controlled oscillator frequency, our system measures the incoming sample rate with high accuracy. Whilst that could simply program the DCO, we use a complex algorithm that takes care of the real frequency that's written in real-time to the DCO. The algorithm keeps frequency variations to a minimum to yield a fixed frequency in the short term and to track source frequency deviations over the long term. The resulting loop corner frequency of the system is as low as 0.5Hz."

To rattle off more hard figures which are so popular in DAC discussions, the WaveDream is good for 24/192 via its coax and AES/EBU inputs and does 32/384 PCM and 11.28MHz DSD via USB and its two HD links on HDMI. All incoming data of 44.1kHz or multiples thereof upsample to 705.6kHz. 48kHz and its multiples arrive at 768kHz. A-weighted dynamic range is 125dB, S/NR 140dB, output impedance 5Ω. Asynchronous input reception occurs at 560MHz sampling and the "memory-based digital-domain PLL" again operates at 300fs jitter. Mathematical precision is 68-bit integer, the volume control is a 256-step custom multiplier array over 0.5dB increments implemented as quad 63-bit DSP inside the FPGA. Selectable dither is "4-bit Gaussian ultrasonic". The 128x65 yellow Oled display is dimmable over 8 steps, the firmware is USB upgradeable—which by altering the FPGA coding actually upgrades hardware—and there's remote control. The user can invert phase, select between linear, minimum and hybrid filters, activate the dither generator and alter the digital PLL settings between local clock and clock stream.

On competitive coin, early sightings in Hong Kong had potential shoppers there call the single-ended (basic) Platinum WaveDream priced 35% below the equivalent MSB Analog DAC, the balanced Signature version (top model) the same 35% beneath the MSB Signature IV+. After having followed the WaveDream's online tracks as far as I could to find it most interesting, I asked Nicolae about his interest in a review. "It will be my pleasure to see it happen. However, give me a little time to see how I will set this up. My lab has rather limited production capacity." That answer reflected my expectations for a busy ODM designer working out of Romania. But all good things are worth waiting for.

If all this digital talk had you dizzy wondering about analog—after all, we listen to the DAC's output in the analog domain—there's no traditional output stage. "We initially planned on a tube stage but eventually had insufficient space for something proper. Then we experimented with a solid-stage analog volume control stage. After extensive listening tests we realized that our digital volume control was truly exceptional. So finally we opted for taking the output signal directly off the DAC modules. This must have been a good decision because a few of our customers have already sold off their preamps." With 5Ω output impedance, direct drive would certainly not be an issue. If all the talk about digital volume, discrete R2R conversion and DSD compatibility had you confused... yes the WaveDream must convert DSD to PCM; at 384kHz to be precise. If digital volume control had you say lossy out loud, "ours has 63-bit internal resolution and a very low noise floor. Being digital domain, beyond a certain level it becomes lossy of course. There's always a trade-off between losing bits and signal-path purity." On where on their dial this kicks in, Nicolae had this: "It's not as straightforward as one might think. Mathematically any digital volume control throws away 1 bit every 6dB. But in practice it depends how you are doing it because you basically get the signal level closer to the noise floor. In our particular case we have large dynamic range (32 bits = 192dB) so even if we start to substract bits, there's still a lot of dynamic range left to match even very good analog attenuators."