Sonic Frontiers SFD-2 D/A processor

I feel privileged to have followed the remarkable evolution of digital processors over the past four-and-a-half years. Since my first digital review—a survey of three modified CD players back in August 1989—I've been fascinated by the developments that have inexorably improved the quality of digitally reproduced music.

The climb has been slow and steady, with most newer digital products outperforming older ones. New input receivers, better DACs, and an increasing body of accumulated knowledge have pushed the edge of the digital envelope. This gradual progress has been marked by sudden leaps forward in performance: the Theta DS Pro Generation III, the Mark Levinson No.30, and the Meitner IDAT all broke new ground in different aspects of musical presentation. Along the way, we've also seen products that redefine what we can expect at certain price levels: the $2000 PS Audio UltraLink and $895 Meridian 263 are prime examples of high value for the dollar. These exemplary products set the pace for improvements in digital, fueled by the competitive marketplace. Designers are constantly being pushed to make better and better converters to stay competitive.

Unfortunately, the best in digital has been inordinately expensive. Some digital front-ends cost more than a fully loaded Honda Accord, which is fine—if you can afford it. Even at these stratospheric price levels, however, no processor is the best in all areas. For example, the No.30's bass and dynamics are eclipsed by the Meitner IDAT and Theta Gen.III; the Gen.III's treble is on the etched and forward side compared to the No.30 and IDAT; the IDAT lacks the sense of ease and resolution of recorded detail revealed by the No.30. No single processor does it all, although the compromises are vastly lower with the best units.

Given this situation, could it be possible that the upstart $4650 Sonic Frontiers SFD-2 digital processor—the first digital product from a tube amplifier company—manages to get almost everything right musically, yet cost far less than the current contenders for the state of the art?

A wolf in sheep's clothing
The SFD-2 looks unassuming, even plain. The front panel is Sonic Frontiers' typical ¼"-thick stainless-steel inner panel covered with a gold overlay panel. Three toggle switches run along the panel bottom: one selects between three digital inputs, one inverts absolute polarity, and the third powers the tube filaments. (The rest of the unit is turned on whenever the SFD-2 is plugged in.) A row of three LEDs indicates the incoming sampling frequency, and another row of LEDs shows filament supply on, if the unit is locked to the incoming data, and whether the de-emphasis circuit is engaged.

The rear panel has three inputs: AT&T ST-type optical, AES/EBU (an XLR connector), and coaxial on an RCA jack. An RCA jack provides a digital output. Analog output is via a pair of RCA jacks (single-ended) and XLR connectors (balanced). All three digital inputs and balanced outputs are included as standard. An IEC AC power-cord jack finishes off the rear panel.

Once the sheepskin (the top cover) has been removed, however, the SFD-2's inner wolf is immediately evident. The SFD-2 is clearly a serious piece of audio engineering. The parts quality, design, and execution are absolutely first-rate. The massive and sophisticated power supply, the first use of two new UltraAnalog parts (the AES20 input receiver and D20400A DACs), and a beautifully implemented tubed output stage all suggest that the SFD-2 is not just an attempt at making a good-sounding processor for under $5000, but an all-out attempt at the state of the art. There no hint of compromise evident anywhere in the circuitry or parts.

Fully half the SFD-2's chassis is filled with power-supply components. The main power-supply board consumes the right-hand third of the unit, with additional supply components distributed near the circuits they supply. Three separate transformers are used: one for the +5V digital electronics and housekeeping circuitry, one for the input receiver and DACs, and one with dual secondary windings for the tubes' high-voltage and filament supplies. The transformer outputs are rectified by four full-wave bridge rectifiers made from high-speed diodes. A fifth rectifier is a more common integrated bridge, probably used in the +5V digital supply. Filter capacitors are the Panasonic HFQ electrolytics and custom-made Cornell-Dubilier types. The Panasonic HFQ is a new—and increasingly popular—cap designed for power-supply filtering in audio components. All power-supply electrolytics are double-bypassed with Wima polypropylene types.

The SFD-2 has a total of 16 regulation stages for six supply voltages. Some power-supply rails are supplied from cascaded regulation (one regulation stage feeding a second regulation stage), and others use separate regulation stages to feed different circuits. For example, the ±12V supplies to the AES20 input receiver and D20400A DACs are regulated on the power-supply board, then further regulated on the digital board. In addition, the input receiver's ±12V regulation stages are completely separate from the DACs' ±12V stages. In all, eight regulation stages are used just to supply ±12V to the DACs and input receiver.

These regulation stages are built around Linear Technology LT337T and LT317T ICs, which are quieter than the popular (and less expensive) 7800/7900 series regulators. The LT regulators also have lower output impedance at high frequencies. Further, the regulation stages are doubly bypassed with Wima polypropylene caps. The +5V supply for the housekeeping functions (input muting, LED drivers, etc.) is separately regulated from the input receiver's +5V supply.

The output tubes' high-voltage supply is generated on the power-supply board and sent to the analog board on a braided lead. The supply is then regulated on the analog board with four discrete regulation stages, two for each tube. Unusually, the 6922 output tubes are fed from a ±120V supply rather than a single B+ supply (more on this later). The discrete regulation stages supplying the analog output section are elaborate and sophisticated, using fully complementary devices developed for driver stages in power amplifiers. The high-voltage regulation stages are bypassed with Solen, Wima, and MIT MultiCaps.

The filament supplies for each tube are independently IC-regulated. The front-panel "Operate" switch turns on the filament supplies; all other voltages are present whenever the SFD-2 is plugged in. This keeps the power-supply electrolytic caps formed at all times.

The amount of regulation and separation of power-supply stages, along with the copious and expensive bypassing capacitors, is unusual and elaborate. This is an extraordinary power supply by any measure.

Moving on to the digital board, the SFD-2 uses the ubiquitous NPC 5803 digital filter—nothing unusual here. But the input receiver and DACs are another story. The SFD-2 is the first product to use UltraAnalog's AES20 input receiver, a sophisticated module that replaces the Crystal CS8412 or Yamaha YM3623 chip. Unlike these two purely monolithic chips, the AES20 is a combination of monolithic and discrete devices encapsulated in a potted module measuring 1.5" by 2". The AES20—which spent more than a year in development—greatly reduces jitter in the digital processor in which it is used.

The AES20 does this in two ways. First, it has very low intrinsic jitter, meaning that the AES20 adds very little jitter of its own to the clock recovered from the incoming digital datastream. According to the spec sheet, the AES20 has less than 40ps of intrinsic jitter, one fifth the Crystal CS8412's intrinsic jitter (specified at 200ps), and orders of magnitude less jitter than the Yamaha YM3623. Second, the AES20 rejects jitter in the incoming datastream from a CD transport instead of passing the jitter to the recovered clock. The frequency at which an input receiver begins to reject incoming jitter is called its "jitter attenuation cutoff frequency" (JACF). The Crystal CS8412's JACF is specified at 25kHz, meaning that jitter below 25kHz will be passed to the recovered clock, and jitter above 25kHz will be attenuated. The AES20's JACF is 1kHz, meaning that transport jitter above this frequency will be attenuated. Note that jitter from 0-40kHz causes audible problems (in multi-bit DACs). I'll have some measurements later on how well the AES20 works.

Not only is the SFD-2 the first product to use this promising new input receiver, it is also the first product to employ UltraAnalog's new D20400A DAC. The D20400A is the first revision of UltraAnalog's popular D20400 DAC since the device was introduced nearly six years ago. The two DACs have identical architecture, manufacturing procedures, and calibration techniques, but use slightly different internal parts. The revision was based on listening tests and feedback from designers who had been using the D20400 (footnote 1).

Further, the SFD-2 uses two of the dual-channel D20400 DACs to give true balanced operation for each channel. Many digital processors balance the signal in the analog domain after the DACs and analog stage. This is an inexpensive method of achieving a balanced output, but not nearly as good as converting the digital datastream to + and - signals for each channel, then converting those four digital signals to analog separately. This technique requires four DACs, four I/V converters, four analog output stages, and four output filters. The advantage is that any distortion, noise, or spuriae common to both channels (called "common-mode noise") will cancel (footnote 2).

The output stage is based on the Sovtek 6922, the Russian designation for the 6DJ8 dual-triode tube. One tube per channel is used, with each half of the dual triode handling each phase of the balanced signal. The tube stage is configured as a cathode follower, a circuit that provides high input impedance, low output impedance, and a voltage gain slightly less than one. The tubes therefore act as output buffers rather than gain stages. Unusually, the 6922s are supplied from ±120V rather than a single "B+" plate supply that holds the cathode at ground. Because the cathode isn't referenced to ground in the SFD-2's implementation, grid voltage swings can't force the output to hit ground. Moreover, ripple in the ±V supply cancels, further improving noise performance. Incidentally, the analog ground is decoupled from both the power-supply ground and the chassis. Finally, expensive ceramic tube sockets with silver contacts house the tubes.

The analog filtering uses a unique "twin-T" passive notch filter to remove the spurious components above the audio band. The filter is implemented with 1% polystyrene caps and Caddock resistors, all enclosed beneath a thick copper shield. The filters are individually trimmed and tested. Sonic Frontiers claims that the analog filter has a large effect on a product's sound, and thus spent much effort on designing one that sounded good.

The output stage bristles with extremely expensive MIT MultiCaps (a total of 14) and Vishay and Caddock resistors. The circuit board uses 3.5oz copper plating (most boards use 1 or 2oz plating) with bare copper traces which reportedly sound better than solder-plated traces. Connection to the RCA jacks is via Kimber AGSS silver hookup wire. The RCAs are Kimber Ultraplate types, and the XLRs are Neutrik silver-contact types, both of which are of the highest quality.

The single-ended outputs are simply driven by one half of the balanced signal. The Theta Gen.III also uses this approach. The Mark Levinson No.30, however, combines the two balanced signal halves differentially inside the processor to derive the single-ended output, maintaining the advantages of differential DACs even when using the single-ended outputs.

An output muting relay prevents noise from appearing at the analog outputs. The muting relays are engaged for about 45 seconds when the filament supplies are turned on, allowing the tubes time to warm up.

The SFD-2's build quality and execution are outstanding. Although I wouldn't call the SFD-2's chassis and front panel lavish, the circuitry and parts quality are certainly on a level with—or beyond—any other digital product I've reviewed.

Although I like the idea of keeping a product's cost down by putting topnotch electronics in a moderately priced chassis, I wasn't crazy about the front-panel toggle switches. They seem fragile, as if they would break off if bumped (this is also a liability on Theta processors). Finally, the first sample had an intermittent noise at the output of one channel; Sonic Frontiers sent a second sample, which was auditioned for this review. The second sample's polarity-inversion switch was either wired backward or installed upside down, meaning that the unit inverted polarity when in the 0-degree position, and did the opposite in the 180-degree position.

Music
I'll cut to the chase: The SFD-2's musical performance was absolutely stunning by any standard. In some ways—in many ways—the SFD-2 is the best-sounding digital processor I've heard.

The SFD-2's most surprising characteristic was its extraordinary bass reproduction. I say "surprising" because all the previous UltraAnalog-based processors I've auditioned have had less-than-full measures of bass depth, control, and slam. They've tended to be lean and somewhat lightweight, lacking the visceral bass punch that is so important to some music. Although the bass reproduction of many UltraAnalog-based processors has been articulate and agile, they have nevertheless lacked a sense of physical power. This is one area where the Theta Gen.III is king; the III's power and drive in the bass are unrivaled. In fact, Martin Colloms—whose ears I hold in high regard—has argued that, for this reason, the Gen.III is a more musically satisfying processor than the No.30, despite the latter's admittedly cleaner treble reproduction, greater sense of ease, and higher resolution of inner musical detail. A second reason the SFD-2's bass performance surprised me was its tube output stage. Tubes aren't noted for their tight, rock-solid bass.

Back to the SFD-2: Its bass was wonderfully full, rich, warm, deep, and powerful. The presentation had a nice weight without being slow or bloated. In fact, the SFD-2 had extraordinary resolution of pitch and detail. Bass notes were clearly articulated and differentiated, never degenerating into a slow blur. Moreover, the bass had a sense of realism, palpability, and musical rightness not heard from any other processor except the Meitner IDAT. Low-frequency-rich instruments just sounded more real and present than through other processors. Plucked acoustic bass had a roundness that beautifully conveyed the instrument's character. The SFD-2's bass was the antithesis of flat, sterile, or "cardboardy." John Patitucci's acoustic bass work on Kei Akagi's wonderful new Playroom (Bluemoon/Moo R2 79342) exemplified the SFD-2's strengths: every note was clear and precise, yet the instrument had a full measure of weight and warmth—a rare achievement. I greatly enjoyed the SFD-2's combination of fullness and precise pitch articulation.

Further, the SFD-2 possessed an awesome "center-of-the-earth" solidity at the very bottom end. The kick drum on Michael Ruff's superbly recorded Speaking in Melodies (Sheffield CD-35) was presented with stunning power and authority. The SFD-2's bass was like the bass from Krell power amplifiers: effortless, deep, tight, and powerful. Going back to the Speaking in Melodies disc, the SFD-2 beautifully conveyed the rhythmic power of bass guitar and kick drum working together to drive the music along. The sense of physical involvement was extraordinary. In this regard, the SFD-2 was significantly better than the No.30. Even compared with the Gen.III, the SFD-2's bass was more musically rewarding. Astonishingly, the SFD-2 had more depth, power, and bass dynamics than the Gen.III. Again, I wasn't expecting this kind of bass performance from a tubed, UltraAnalog-based processor.



Footnote 1: See my "Industry Update" of last August for a description of the AES20 and D20400A, and of how the DACs are individually calibrated.—Robert Harley

Footnote 2: Digital processors with true balanced operation include those from Meitner, Mark Levinson, Krell, Theta, and Kinergetics. Examples of processors that balance the signal in the analog domain after a single DAC are designs from PS Audio, EAD, Audio Research, and Proceed.—Robert Harley

X