The first time I heard the Sonic Frontiers SFD-2 D/A converter, I said to myself, "Got to get one of those machines to try out and, I hope, to review." Sonic Frontiers was kind enough to let me try one out, and after I listened to and enjoyed the unit for a number of months, the gods finally aligned the universe in such a manner as to allow me to review it.
I think the SFD-2's price/performance ratio makes it a breakthrough product.
With some input from UltraAnalog Inc., Sonic Frontiers has designed a D/A converter that is right up there among the very best. Among its circuit features are a digital filter with eight-times over-sampling, dual UltraAnalog converters in a balanced configuration, and a fully balanced tube output stage.
Starting out with a heavy-gauge steel enclosure and stainless steel front panel, the approximately 30-pound package is beautiful, rugged, and functional. Three front-panel toggle switches, spaced across the front panel, set output polarity, select in put source, and select "Standby" or "Operate" mode.
Three LED indicators in the left half of the panel show "Power," signal lock-in, and de-emphasis. Toward the right, three more LEDs show the sampling frequency of the selected input's signal. Digital inputs on the rear panel include an XLR jack for AES/EBU signals, a phono connector for coax, and an AT&T (ST) glass-fiber connector. Conspicuous by its absence is the Toslink optical connector that many audiophiles consider sonically inferior. A coaxial output (another phono jack) carries the selected digital input for connection to other equipment. Neutrik silver XLRs are used for balanced analog outputs; Kimber Kable phono connectors are used for the unbalanced outputs as well as for the digital in put and output phono connectors. An IEC a.c. power connector and line-fuse holder round out the rear-panel features.
The interior space is divided roughly into thirds. The left-hand section, as seen from the rear, houses the power-supply p.c. board, which feeds the digital processor board and provides plate and heater power for the analog output board. An example of the many nice touches in the SFD-2 is the use of power transformers with end bells instead of the usual open-construction board-mount transformers. In the middle section is the digital processor board, surrounded by a metal shield to keep the digits contained therein. Analog circuitry (consisting of output-stage voltage regulators, a passive output-filter assembly, muting relays, and the tube output stage) is on the third p.c. board to the right. The fourth p.c. board, mounted be hind the front panel, interconnects the LEDs and digital input selector, and pro vides the power on/off time-delay and muting-control circuitry. High-quality parts, such as Caddock and Holco resistors and MIT film capacitors, abound. Build quality is superb. Truly, this is a first-rate component!
The SFD-2 is one of the first products, if not the first, to use the new UltraAnalog AES 20 input-receiver module. Convinced that jitter in the incoming S/PDIF data stream and the jitter produced in the input receiver itself are potential causes of audible degradation in the final audio output of a D/A converter, UltraAnalog set about de signing an input receiver that would reduce these effects. The AES 20 contains a low-jitter input-receiver chip and fast phase-locked loop (PLL) combined with a second PLL that reduces the overall jitter-attenuation cutoff frequency to about 1-kHz. Some unusual techniques are employed in the AES 20, including brass shielding of the phase-locked loops, to insure ultra-low jitter in the receiver module's outputs.
-------------
Specs:
Sampling Rates: 32, 44.1, and 48 kHz.
Frequency Response: 5 Hz to 20 kHz, +0.5, -0.1 dB.
S/N, A-Weighted: Greater than 110 dB.
THD: Less than 0.05%.
Jitter: Intrinsic jitter, less than 40 picoseconds; jitter rejection from 1 kHz, up.
Channel Separation: At 1 kHz, greater than 105 dB; at 10 kHz, greater than 85 dB.
Output Voltages: Unbalanced, approximately 3.5 V; balanced, approximately 7.0 V.
Dimensions: 19 in. W x 4 in. H x 13 in. D (48.3 cm x 10.2 cm x 33 cm).
Weight: 26 1/2 lbs. (12 kg).
Price: $4,695.
Company Address: 2790 Brighton Rd., Oakville, Ont., Canada L6H 5T4.
---------------
Digital outputs of the AES 20 module are fed to an NPC SM5803 digital oversampling low-pass filter. This in turn drives two UltraAnalog D20400A dual D/A converter modules in a complementary, or push-pull, signal arrangement, one for each channel. The digital data is fed, with normal and inverted polarity, into the two inputs of each channel's D/A module.
Within each stereo channel, this tends to cancel any common even-order error characteristics in the difference between the two phases of each module's audio out puts. Richard Powers of UltraAnalog might contest this by countering that the D20400A measures so well, as normally-used (one module used for both channels), that this differential operation using two modules may be needlessly gilding the lily.
Perhaps so, but the proof of the pudding is in the listening, and in my own and others' opinions, the SFD-2 definitely sounds better when listening to the differential output, as opposed to using the single-ended output. (This is, perhaps, getting ahead of myself.)
The audio outputs from the D/A converter modules are led out of the digital board via shielded cables to the input of the analog board. Here they are filtered in a proprietary passive low-pass circuit that combines inductors and capacitors with an adjustable twin-T notch filter, presumably to null out the eight-times oversampling frequency (352.8 kHz). Finally, a tube cathode follower, utilizing a Sovtek 6922 dual triode in each channel, buffers the passive filter outputs and provides a low-impedance final audio output. These dithode-follower circuits utilize positive and negative 120-V d.c. supplies in order to eliminate a coupling capacitor between the DAC/filter output and the tube's control grid. Four separate solid-state voltage regulators sup ply the positive and negative voltages for each channel. Note, there are two followers for each channel, one for each signal phase.
There is no provision for combining the two phases into one composite unbalanced output; rather, the unbalanced output is the output of the positive-polarity signal.
The three transformers on the power-supply p.c. board are, logically, assigned to powering the tube output circuitry, the digital +5 V loads, and separate positive and negative 16-V supplies for the Ultra Analog D20400As in each signal channel.
Power for the tube heaters is sourced from a full-wave-rectified, unregulated d.c. supply. The front-panel "Standby/Operate" switch connects or disconnects the tube heaters from this supply. When the unit is plugged into the a.c. line, all the digital circuitry and high-voltage supplies to the tube output buffers are operational.
Measurements
Frequency response for both channels, with instrument loading on the balanced outputs, is plotted in Fig. 1. Also shown is the effect on one channel's response of loading the output with the IHF load (10 kilohms in parallel with 1,000 pF). The main effect is a reduction of low-frequency response caused by the high-pass filter action between the 3.3-µF output coupling capacitors and the effective load per phase of 5 kilohms (or 10 kilohms, phase to phase). Also seen is the wideband drop in output of the 10-kilohm load against the intrinsic output impedance of the tube cathode-follower stage, some 0.31 dB.
High-frequency response is essentially un changed by the IHF loading. Best low-frequency response (which has an effect on overall space, not just bass) will be obtained with balanced loads of 50 kilohms or more. Frequency response for the un balanced outputs (not shown) was about the same as for the balanced outputs, except that the low-frequency loss with the IHF load was only about half as great. Response with de-emphasis switched in was perfectly flat and therefore is not shown.
Fig. 1--Frequency response.
Fig. 2--THD + N vs frequency for 0-dB digital signals.
Fig. 3--Deviation from linearity.
Total harmonic distortion plus noise, as a function of frequency at digital full-scale, is shown in Fig. 2 for balanced outputs with instrument loading. Distortion is low up to about 1 kHz, because of second-harmonic cancellation in the push-pull out put. For some reason, the balance degrades above about 1 kHz, so the overall distortion rises as shown. Results are shown for two measurement-filter bandwidths, 22 and 80 kHz. The increased high-frequency distortion measured in the 80-kHz band width is partially due to harmonic distortion itself, but it also includes some leakage of the eight-times oversampling frequency into the measurement.
Putting the IHF load on the balanced outputs is kind of rough on these tube cathode-follower output stages. We have already seen that it unacceptably (in my opinion) affects low-frequency response.
In the case of full-scale harmonic distortion, the IHF load (not shown) raised the amount of distortion to about 0.03% or 0.04% up to 2 or 3 kHz, with the curve then blending into the levels seen in Fig. 2.
Not a bad result, really. With the unbalanced outputs, THD + N was approximately twice as much as shown in Fig. 2.
Looking at the distortion in the DAC out put before the filter and tube stage, I could see that the tube output stage was the dominant cause of the output distortion. k Measurements of distortion versus digital signal level (not shown) indicated that distortion decreases rapidly with decreasing signal level and is down in the noise by about-12 dB full-scale (dBfs). However, in reality, the higher numbers discussed above for full-scale signals are not the distortion levels that the system produces with music, as the average level of most digital pro gram material is much lower than full-scale.
Deviation from linearity as a function of digital signal level is plotted in Fig. 3 for 500 Hz. The data shown is for the unbalanced outputs, but there was no notice able difference with the balanced outputs. Using the CBS CD-1 fade-to-dither test (which measures the same kind of thing but uses a CD transport to drive the D/A instead of the Audio Precision digital generator) yielded essentially the same result.
I was curious to see if using a balanced AES/EBU feed from my digital signal source instead of my usual unbalanced coax feed would make any noticeable difference in the measurements discussed thus far. Repeating some of the measurements with a balanced cable from the Audio Precision digital generator output to the AES/EBU input of the SFD-2 yielded essentially the same results.
Another look at low-level linearity is the noise-modulation test devised by Richard Cabot of Audio Precision. This test causes a low level 40-Hz signal to be presented at input levels of -60, -70, -80, -90, and -100 dBfs. For each of these input levels, the output is analyzed by sweeping a third-octave filter from 300 Hz to 20 kHz.
If things are in order, these five curves basically overlie each other; this was the case for the SFD-2.
Interchannel crosstalk generally in creased at a rate of 6 dB per octave after it rose above the noise level, which occurred at about 500 Hz to 1 kHz. Crosstalk for both directions, with balanced or unbalanced outputs, was better than 100 dB down at frequencies up to 2 or 3 kHz and rose to a worst-case amount of about -75 dB at 20 kHz.
Test results for S/N, quantization noise, and dynamic range are shown in Table I for the balanced outputs. Data was similar for the unbalanced outputs, except the digital zero (0-dBfs) noise levels were 2 to 3 dB noisier. Either output was quiet enough to easily resolve the three-state waveshape of an undithered -90 dBfs input signal.
Output resistance measured about 175 ohms for the unbalanced outputs and about 350 ohms for the balanced outputs.
The a.c. line draw was 280 mA in "Stand by" and 440 mA when the SFD-2 was warmed up and fully operating.
above: Table 1--Signal-to-noise ratios. Quantization noise was -94.7 dB in either
channel; dynamic range was 96.1 dB in the left channel and 96.0 dB in the right.
Use and Listening Tests
Phono sources in my system during the review period included an Oracle Audio turntable fitted with a Well Tempered Lab arm and JVC X-1 moving-magnet pickup (used with my own tube phono preamp or a Quicksilver Audio preamp). Counter point DA-11A or PS Audio Lambda CD transports were used to feed the Sonic Frontiers SFD-2, the Stax DAC-Talent BD, and other (experimental) D/A converters.
Other signal sources included Nakamichi's ST-7 FM tuner and 250 cassette recorder, and a Technics open-reel recorder. Preamplifiers used included a Quicksilver Audio, Forssell tube line drivers, and an AR Limited Model 2. Power amplifiers used were a Crown Macro Reference, Quicksilver M 135s, and an Arnoux MB300A digital switching design. Loudspeakers used were B & W 801 Matrix Series 3s augmented in the range from 20 to 50 Hz by a pair of subwoofer systems, each using a JBL 1400Nd driver in a 5-cubic-foot ported enclosure.
To get the full benefit of the differential (or push-pull) operation of the SFD-2, one needs to feed its differential output into a component (normally a line-level preamp) with a balanced input having an impedance of at least 20 to 30 kilohms, and preferably 50 kilohms or higher. The in put-impedance consideration is relevant to getting the best low-frequency response, in view of the 3-0 output coupling capacitors in the output of the SFD-2. If the contemplated system preamp is to feed an un balanced output to a power amp, it is essential that this preamp combine both input phases, in a differential-amplifier manner, into the single output phase to be used. Some balanced-input preamps do not do this but separately pass both input phases on to the two output phases and include only one of these phases in their un balanced outputs. Such a preamp will get best results from the SFD-2 if-and only if-the following power amplifier has balanced inputs.
I was impressed with the sound I got from the SFD-2 right from the start, using its unbalanced outputs to feed the unbalanced inputs of my Forssell line driver, which, in turn, ,drove my various power amplifiers. I also modified the Forssell to accept a balanced source, combine both phases, have control of the volume, and then drive the amps unbalanced. The result was extremely good sound. This line driver permitted the SFD-2 to sound its best and has formed the basis for long-term listening with it.
The sound is characterized by a great sense of space and dimension, killer bass, great transparency and delicacy with low irritation levels, and simply the best digital sound reproduction I've had so far. Toward the end of the review period, I began using another, newer, Forssell line driver that is fully balanced from input to output and combines both input phases to either out put phase; the sound became better yet.
Using it to drive the Crown Macro Reference in balanced mode produced an exceptionally wonderful and musical sound.
One thing that drove me nuts, until I found out what caused it, was that after I used the SFD-2 for a number of months, the sound became a bit edgy and irritating (sort of like a phono cartridge going bad).
I tried changing everything, but nothing really helped until I changed the tubes in the SFD-2 and that wonderful sound came back! I personally think (but cannot conclusively prove) that the way the "Stand by/Operate" switch works in the SFD-2 makes the tubes slowly deteriorate, with resultant sonic degradation. So I would simply advise owners of this otherwise outstanding piece of gear to either leave it on in the "Operate" mode and/or have some spare new tubes on hand to put in whenever the sound degrades (if, indeed, it does so for anybody else). Other than this, the SFD-2 operated flawlessly.
In conclusion, do I like the SFD-2? You had better believe it. I think it is the best D/A converter I have had so far, and it has brought me untold musical pleasure and delight. Do go out and buy one of these, and discover the delights of good digital sound for yourself?
-Bascom H. King
(Adapted from Audio magazine, Jan. 1995)
Also see:
Sonic Frontiers Processor 3 D/A Converter (July 1998)
Sonic Frontiers Power-3 Mono Amp (Aug. 1996)
= = = =