Bricasti Design M1 D/A converter
Back in the late 1980s, it seemed a good idea: Separate a CD player's transport section from its D/A circuitry so that each could be optimally designed, and, as D/A technology improved, the sound of your CD player could be upgraded by replacing the outboard D/A processor. The catch was that the transport and D/A chassis needed to be connected with a serial data link: S/PDIF in optical or electrical flavors, or balanced AES/EBU. To minimize the number of cables required, the format of that link embedded the clock data within the audio data, which rendered the link sensitive to interface timing uncertainty, or jitter. (See "Bits Is Bits?," by Malcolm Hawksford and Chris Dunn.)
Overcoming the effects of jitter back then required heroic engineering at heroic prices. In February 1992, when we reviewed the Mark Levinson No.30 Reference D/A processor from Madrigal Audio Labs, which successfully addressed the interface problem, it cost a staggering $13,950. But the No.30 also illustrated an advantage of the standalone D/A processor: It was not limited to the CD's 44.1kHz sample rate and 16-bit word length. The No.30 was originally a 20-bit device operating at 44.1 and 48kHz; subsequent upgrades allowed it to handle 24-bit data at sample rates up to 96kHz, and to decode HDCD datastreams. But again, this all came at a price: the upgraded version, the No.30.5, released in 1994, cost $15,950; the final version, No.30.6, from 1999, cost $16,950.
I bought the 1992 review sample of the No.30, and purchased both the .5 and .6 upgrades. The No.30 was my reference D/A processor until 2009, when its separate power supply stopped working. The failure was hardly surprisingthe unit had been powered up almost continually for 17 years.
Unfortunately, although Madrigal had committed to supporting their Reference products indefinitely (footnote 1), Harman's purchase of Madrigal and the subsequent closings of first the Connecticut facility that used to manufacture Mark Levinson products, then of Lexicon's Massachusetts facility, where Levinson production had been moved, meant that Harman would no longer repair broken No.30.6sor any other old Mark Levinson gear.
I have since discovered a Massachusetts dealer that offers Mark Levinson repair by experienced staff (footnote 2) and have shipped them my No.30.6 for surgery. In the meantime, I've been auditioning 21st-century D/A processors that are claimed to offer state-of-the-art performanceincluding the subject of this review, the M1 ($7995) from Bricasti Design Ltd.
The Bricasti M1
I had been alerted to the existence of the M1 by the August 2011 installment of John Marks's column, "The Fifth Element." "Fast, detailed, effortlessly powerful, musically revealing. Fatigue-free listening," he wrote, concluding that the M1 offered "The best digital playback I have heard." John offered to send me the M1 for measurement and some listening. Well, he didn't so much offer as order.
Not that I needed much persuadingthe M1's lineage ties it to my Mark Levinson No.30.6. Bricasti's cofounders, Casey Dowdell and Brian Zolner, had respectively been a DSP-software engineer and international sales manager at Lexicon before Harman International closed its New England operations (footnote 3). While Bricasti writes all its own signal-processing software, it contracts out some of the hardware engineering to AeVee Labs, a company in New Haven, Connecticut, founded by Bob Gorry, who used to be Chief Engineer at Madrigal Audio Labs.
The M1 is Bricasti's second product, the first being the M7 reverberation engine, aimed at the pro-audio market. Housed in a single 17"-wide black chassis with brushed-aluminum control knob and pushbuttons and a central red alphanumeric LED display, the M1 bears a superficial resemblance to Mark Levinson products of yore. Clockwise from the bottom left, the six pushbuttons control Display brightness, input Status, Input select, Filter select, Aux input S/PDIF or word-clock select, and Enter. When one of the first five buttons is pushed, the rotary control allows you to scroll through the choices offered, which are then selected by pressing Enter.
On the center section of the rear panel are the IEC AC inlet, the power switch, and four transformer-isolated digital inputs: AES/EBU on an XLR, S/PDIF electrical on an RCA, S/PDIF optical on the usual TosLink jack, and a BNC jack that can be switched either to S/PDIF (default) or Word Clock, the latter allowing the M1 to be slaved to other digital devices. Unusually, there is no USB port, but there is a remote trigger jack. To the left and right of this central area are the analog output sections, each offering balanced signals via an XLR jack and unbalanced via an RCA. Each of these sections also has a small back-lit level-trim control for the balanced output, adjustable from +8 to +22dBm (1.959.75V). The default output level is +14dBm (3.9V).
Internal LEDs on the circuit board glow a soft red through the M1's perforated side panels; the effect is elegant and attractive. The M1 runs hot; after the review sample had been powered up for a day, its top panel stabilized at 103°F (39°C).
Other than a central signal-processing board and its switch-mode power supply, the M1's internal construction is dual-mono. The analog circuitry for each channel is constructed on an Arlon printed-circuit board (another echo of Mark Levinson gear); the layouts of the two channels' components mirror each other. Each channel is powered by its own toroidal AC transformer mounted behind the front panel, followed by extensive voltage regulation.
The selected digital input is fed to a large Analog Devices ADSP-21368 SHARC digital signal-processing chip. This appears to include the S/PDIF receiver, and also manages the front-panel display and controls, synchronizes the clocking of each channel's D/A converter, and provides the oversampled reconstruction filters. Also on this board are a Xilinx FPGA chip and another Analog Devices processor, an ADSP-BF532 Black Fin chip. Each channel's oversampled and low-passfiltered data are fed to its board via ribbon cable, where they're converted to analog with an Analog Devices AD1955 24-bit/192kHz chip. The AD1955 is a two-channel part, but in the M1, each chip is operated in differential-output mode for increased dynamic range. The D/A converter's differential current outputs are followed by four high-slew-rate (250V/?s) op-amps, these Analog Devices AD843s, which are followed in turn by two discrete-transistor output buffer sections, one balanced, one unbalanced.
The AD1955 includes a digital-domain volume control, but this is not used in the M1. The D/A converter chip also has an internal reconstruction filter, this available as Filter 0 in the Bricasti's Filter menu, and described in the manual as a "basic half-band filter" with 6dB attenuation at the Nyquist Frequency. The other six filters are Bricasti's own. They are:
- Filter 1: 20kHz bandwidth, stop-band at Nyquist Frequency with low ripple and high attenuation.
- Filter 2: Similar to 1, with a filter of slightly gentler slope.
- Filter 3: Steepest slope, highest bandwidth.
- Filter 4: Low-delay filter with full attenuation at Nyquist Frequency but higher passband ripple.
- Filter 5: Second version of a half-band filter with 6dB attenuation at Nyquist frequency.
- Filter 6: Low-delay filter starting at 18kHz and full attenuation at Nyquist frequency.
None of these filters is a minimum-phase or apodizing type, but given the flexibility of the M1's DSP platform, I don't see why such filters couldn't be offered as a firmware upgrade.
Because the M1 lacks a USB input, I used the Empirical Audio Off-Ramp4 USB-S/PDIF converter I reviewed in December 2011 to take the audio data from the Mac mini I use as a music server, connecting its output with a 75-ohm BNC cable to the Bricasti's Aux input. For silver discs, I connected the AES/EBU output of my Ayre Acoustics C-5xeMP universal player to the M1 with a 1m length of DH Labs Silver Sonic cable.
The first order of business was to choose a reconstruction filter. Like John Marks, I felt Filter 0 was flat and uninvolving. However, while John's favorite was Filter 4, I felt this tended to smooth over fine detail a little. I ended up doing almost all my auditioning with Filter 6, which I felt struck the right balance between the presentation of detail and the ability to throw a deep, well-defined soundstage. With Filter 6, Ray Brown's double bass in "Exactly Like You," from his Soular Energy (24/192 ALAC file ripped from DVD-Audio, HiRez Music HRM2011), had exactly the right balance between the body of the instrument's tone and the leading edges of the notes. And Gene Harris's piano on this recording had an almost crystalline clarity but without sounding forced.
That piano had presumably been close-miked, but the M1's sympathetic way with piano extended to more distantly miked instruments. For example, the 9' Steinway D used by Robert Silverman in his recording of Liszt's Sonata in B Minor (CD, Stereophile STPH008-2) had superb weight to its low frequencies, yet without sounding muddy. And the supportive acoustic of Albuquerque's First United Methodist Church was nicely resolved.
When I recorded the Elgar Piano Quintet at the 1998 Santa Fe Chamber Music Festival for release on Bravo! (CD, Stereophile STPH014-2), I had to mike the players closely because we weren't allowed to turn off the air-conditioning system of St. Francis Auditorium. In postproduction I used a Lexicon PCM 90 digital reverberator to produce a synthesized acoustic that matched, as closely as I could judge, the sound of the hall. As D/A converters have improved over the years, I worry that their increasing transparency will eventually unmask the difference between the artificial and real acoustics on this CD. The Bricasti surprised me with the Elgar; its reproduction of this performance pushed the string quartet and piano farther back in the soundstage compared with my expectations. While there was no audible conflict between the residual reverb of the auditorium and that produced by the Lexicon, the overall balance was more reverberantas if the reverberation tails were being more clearly resolvedthan I remembered hearing when I mastered the CD using the Levinson No.30.6.
Footnote 1: Props to McIntosh Labs for making the same commitment and living up to it. This, to me, is what high-end audio should be about: customer support of the same caliber as the product's performance.
Footnote 2: The Service Bench, 227 Carnegie Row, Norwood, MA 02062.
Footnote 3: Dowdell and Zolner presented a master class on high-end product design, "The Fine Line between Voicing and Design," at the Audio Engineering Society Convention in New York on October 20, 2011. From the abstract: "When the design brief of a digital to analog conversion system calls for performance regardless of cost, a series of known concepts can be put in place. When performance is the highest priority, the execution and fine tuning of these concepts changes the design brief and project schedule."