Mark Levinson Reference Digital Processor No.30

[adapted from Audio magazine equipment profile, Jan. 1993, by Anthony H. Cordesman and Leonard Feldman].

Both the transistor and CD, despite offering many technical advantages, often sounded worse at first than the technologies preceding them. In each case, subjective critics overreacted by condemning the new technology. Technical critics overreacted by claiming that the new technology's measured performance was superior and that the subjective critics had to be wrong. The resulting dialectic led to improvements that proved both sides right.

In the case of CD, this process has led to the discovery of digital distortions, the design of D/A converters which provide major advances in both sound and technology, and a steadily improving level of performance which has gradually allowed the Compact Disc to live up to its inherent promise.

The Mark Levinson Reference Digital Processor No. 30 is clearly intended to embody this new state of the art. The No. 30 combines a wide range of advances in D/A converter technology and sound quality. It is superbly built, beautifully finished, and has some very unusual features.

==SPECS==

• Frequency Response: 10 Hz to 20 kHz, +0, -0.2 dB.
• THD: 0.003% at 1 kHz and 0 dB, A weighted.
• Dynamic Range: 98 dB.
• S/N: 105 dB.
• Channel Separation: Greater than 110 dB.
• SMPTE-IM Distortion: Less than 0.005%.
• Digital-to-Analog Conversion: Two custom 20-bit P/A converters.
• Digital Filter: Eight-times oversampling.
• Analog Filter: Bessel-tuned, linear phase to 40 kHz, Low-Level Linearity: To 70 dB, deviation unmeasurable; below -90 dB, approximately +1.7 dB (un dithered, referenced to 0 dB at 1 kHz).
• Output Impedance: Less than 6 ohms.
• Dimensions: Processor, 19% in. (49.2 cm X 18.4 cm X 39.2 cm); power supply, (39.9 cm X 13.3 cm X 37.8 cm).
• Weight: Processor, 34 lbs. (15.4 kg), power supply, 25 lbs. (11.4 kg).
• Price: $13,950.
• Company Address; Madrigal Audio Laboratories, P.O. Box 781, Middletown, Conn. 06457.

== ==

The No. 30 is in two pieces. The main unit has the digital circuitry and controls in the center, cradled between towers that each hold one channel’s D/A and analog circuitry. The other unit contains separate power supplies for the main unit’s digital center section and its two analog towers, each connected by a separate cable. My only complaint about the ergonomics is that the No. 30 is big and takes up a lot of space—though no more so than some preamps with separate power supplies.

The No. 30 also has as many inputs and outputs as many preamps. There are eight digital inputs—five balanced female XLR connectors, one ST (AT&T) optical connector, a two EIAJ Toslink optical connectors—enough for most professionals. The Levinson Reference Digital Processor No. 30 also has three digital outputs, two male XLRs for listening and recording plus one EIAJ optical output. The analog out puts include two balanced XLR male connectors and one unbalanced RCA jack per channel. Two “Communication Ports” (“Master” and “Slave”) hint at system ex tensions to come. All of the electrical connectors are gold-plated.

Besides input switching, the front panel has buttons for “Standby” (the unit is never turned completely off), “Digital Re cord Select” (which allows you to record any digital input while listening to another, effectively providing a digital tape monitor), polarity inversion, and display dimming. Indicator LEDs show status, blinking when the unit is in standby mode and glowing when the No. 30 detects copy-protection codes in the digital bitstream and when the digital input signal conforms to the AES/EBU standard. A large dot-matrix display offers a choice of four intensity levels; it also shows when signals are muted, locked, or absent and identifies the selected input and its sampling frequency. Internal switches allow you to rename any of the eight inputs as “CD,” “LD,” “DBS,” “DAT,” “VCR,” or “AUX” to match your system’s setup.

The input switching is highly sophisticated, using a cross-point switch on the DSP board, the uppermost of the three boards in the main unit’s center section. The digital input module, on the same board, has individual active termination, retriggering, and balanced driver circuitry for each input. This was done to avoid possible degradation of input signals, especially from unbalanced outputs. To pre vent transient noise when switching between inputs, the first input is gently muted, and the signal only returns when the No. 30 has locked onto the new signal.

The No. 30 has clearly been de signed for updating. The circuit boards are modular for effective upgrades through board-swap ping, and key solid-state devices (including the EPROM that holds the No. 30’s software routines) can be easily replaced.

Madrigal Audio Laboratories, which makes Levinson components, believes that purity and regulation of the power supply are critical to superior digital performance, and has put at least as much emphasis on the design of the power supply as on the digital and analog signal circuitry. A switching power supply is used for the digital section, providing the high power, freedom from variations in input power, isolation, and thermal stability that digital circuits re quire. The output of this supply is filtered and also has seven separate regulators to reduce noise on the d.c. line. The left and right analog power supplies begin with a.c. filtering and are said to be improvements of the designs in the latest Mark Levinson preamps. These power supplies also have an extremely low rejection ratio, which is rated at —80 dB in the power supply and another -40 dB in the main unit.

Madrigal contends that the data jitter caused at the digital interface receiver (DIR) that locks onto the incoming digital signal is more important in limiting sound quality than the problems dealt with by digital filtering. The company claims that most existing D/A converters have too much jitter to let the rest of the system resolve at 16 bits, much less the higher bit resolutions used by some D/A converters. Some DIRs, for example, are accurate to approximately 200 nS; the No. 30 has a rated accuracy of better than 100 p5.

The Levinson version of the DIR, called a Digital Audio Interface Receiver,(DAIR), uses three separate narrow-aperture phase- locked loops (PLLs), each optimized for a single sample frequency (32, 44.1, or 48 kHz), instead of a single wide-aperture PLL that can cover all three sampling rates. The DAIR’s jitter is said to be more than low enough for the No. 30’s 20-bit resolution at eight-times oversampling.

Madrigal believes that the theoretical advantages of using general-purpose DSP chip sets with proprietary algorithms and high sampling rates are offset by the resulting requirement for much more critical jitter specifications. They also feel that existing transports and circuitry cannot use more than eight-times oversampling with out a loss of sound quality because of jitter- related distortion and that, in any case, there is no practical advantage to oversampling rates higher than eight-times.

According to Madrigal, application-specific ICs (ASICs) are more desirable than general-purpose DSP computers because integrating the software into the silicon lets the ASICs run faster. Thus, from their point of view, only if the desired algorithms aren’t available in an ASIC is a general- purpose chip set needed. An added benefit is that the best custom silicon devices have more accurate digital filters than have yet been implemented with general-purpose DSP circuits. After testing a number of devices and algorithms, the company found an application-specific DSP chip from Nippon Precision Circuits that met their requirements.

Each channel module of the No. 30 holds two linear 20-bit D/A converters, one for each polarity of the balanced digital signal. The positive and negative digital audio signals are processed separately, maintaining differential operation in the digital do main and providing differential analog signals to the analog output buffers. This eliminates the need for inversion circuits in the analog domain.

--- NO. 30 MEASURED BETTER, IN NEARLY ALL THE MAIN PARAMETERS, THAN ANY D/A UNIT SO FAR TESTED.---

The final analog filter in the No. 30 uses a Bessel-tuned active filter in the output stage for maximum phase linearity. Instead of the voltage op-amps used in the feed back loops of most such circuits, Madrigal uses new and faster current op-amps that are said not to be affected by musical dynamics. Their output buffer has discrete components, with a d.c. servo—rather than coupling capacitors—to improve harmonic accuracy.

The No. 30 has two separate signal paths to enhance channel separation. Madrigal claims that the analog circuitry in each tower of the No. 30 represents an advance over the design in any of its preamps and amplifiers, partly due to the use of Teflon boards (whose dielectric constant is low and doesn’t change with frequency) for the analog circuits.

The instruction manual is well written, fun to read, and easy to follow—a comment that rarely applies to high-end literture. But despite its sophistication, I was able to hook up the No. 30 before I read the manual; both connections and operation are straightforward and intuitive.

My subjective impressions of the No. 30’s sound are in the usual place, at the end of this review. But to see how this technology measures on the test bench, as well as for a brief listening report, I turn you over to Len Feldman.

==Measurements==

While I will also have a few words to say about the superb sound quality of this incredible processor, my main task is to discuss its electrical performance. I did not see Mr. Cordesman’s evaluation before writing this, but I must state, unequivocally, that the No. 30 performed better, in nearly all important parameters, than any D/A converter or digital processor I have yet evaluated—and that the measurements were consistent with what I heard in my abbreviated listening tests.

Fig. 1—Frequency response, both channels, using signals from CD player (A) and from test generator (B).

As I’ve done with other stand alone D/A converters, I measured the No. 30 with digital signals from both a CD transport and my Audio Precision test equipment. The AP equipment generates 24-bit data words, then sets the dither levels and assumes that the device under test will simply truncate at the set level.

Figure 1A confirms that the frequency response of the processor, when fed from the digital output of a CD player, is flat within the —0.2 dB claimed by Madrigal, from 10 Hz to 20 kHz. Results are substantially the same for signals generated by the AP equipment (Fig. 1B).

The first difference to show up between CD player-fed signals and signals generated by the Audio Precision test gear occurred when I measured THD + N versus frequency (Fig. 2). While the results shown for signals fed from a CD player’s digital out put are the best I have ever obtained (approximately 0.0015% at mid-frequencies), results for the signals generated by the AP test equipment are even better, an incredibly low 0.0008% or so at mid-frequencies. In fact, my usual scale had to be expanded downward to “capture” this plot. Notice, too, that even at the treble extreme of either sweep, the increase usually seen in THD + N is present but minimal, never rising above 0.0025%.

Fig. 2—THD + N vs. frequency, both channels, relative to maximum output level.

Fig. 3—THD + N vs. signal amplitude, both channels.

Fig. 4—Spectrum analysis of residual noise when playing “no-signal” track of test CD; both channels shown.

Figure 3 shows how THD + N varies with signal amplitude. For practical reasons, this test was conducted using only the digital output from the reference CD player, with signals that ranged in amplitude from 0 dB (maximum recorded level) to —90 dB. Even at maximum recorded level, the reading is only about —97.5 dB, corresponding to an equivalent THD + N of 0.00133%. At levels below —20 dB, the readings (referred to maximum recorded level) are even lower, —98.5 dB or better.

To isolate the actual THD from the residual noise of the system, I used the FFT spectrum-analysis capabilities of the Audio Precision DSP circuitry while the CD player fed the No. 30 a l-kHz steady-state tone. By “capturing” the results 16 times, the noise peaks were effectively averaged out, and only two significant actual harmonics of the l-kHz tone were discernible, at 3 and 5 kHz. Their dB values, when calculated as a percentage of THD, worked out to only 0.00107%.

Channel separation exceeded the published specification by far and was greater than I have measured for any previous CD player or D/A converter. Separation was nearly 137 dB at 1 kHz and was still greater than 124 dB at 16 kHz. What’s more, it was virtually identical for both channels.

Signal-to-noise ratio was the same whether I used the “no-signal” track from my CD-i test disc or a “no-signal” digital output from the Audio Precision gear. In both cases, S/N was 117.8 dB for the left channel and 117.9 dB for the right channel. While these results are remarkable, the real surprise came when I ran a spectrum analysis of the residual noise (Fig. 4) and found absolutely no noise peaks at 60, 120, or 180 Hz from the 60-Hz line frequency. This is the only DIA converter I have ever tested that showed no measurable hum-related components. Of course, part of this “miracle” arises from the fact that the Mark Levinson No. 30’s power supply is a separate component, connected to the processor via cables; no a.c. voltages ever reach the processor itself.

Figure 5 shows deviation from linearity, using undithered signals from 0 to —90 dB and dithered signals from —70 to —100 dB. The slight deviation of less than 1 dB for an undithered signal at —90 dB is better than claimed by the manufacturer and about as good as I have seen with any CD player or DIA converter. Bear in mind that the D/A converters used in this processor are not the increasingly popular one-bit variety that boast near-perfect linearity at the expense of some added noise (when improper noise-shaping is employed, as it some times is). Here we are dealing with 20-bit converters, which means that the converters used in the Mark Levinson No. 30 either have near-perfect low-level linearity or have been carefully calibrated externally during production.

In any case, when I used dithered signals in the range from —70 dB to —100 dB to check out linearity at low levels, results were even better. As seen in Fig. 5, I could detect no deviation from perfect linearity whatsoever! My usual fade-to-noise test results, shown in Fig. 6, further confirm the No. 30’s excellent low-level linearity as well as its low residual noise level and high EIA dynamic range (which I calculated as just over 100 dB). The EIAJ dynamic range exceeded the published claim, measuring 99.95 dB for the left channel and 99.91 dB for the right channel.

For further confirmation of the superb linearity of this processor, I plotted output versus input, using progressively lower dig ital input signals generated by the Audio Precision (Fig. 7). I could not easily translate this plot into deviation from perfect linearity, but close examination of the graph reveals that even at —100 dB, output matched the input perfectly. What little deviation there is occurs at levels below —110 dB, where the deviation from perfect linearity is still less than 1.0 dB!

Fig. 5—Deviation from linearity.

Fig. 6—Fade-to-noise test, using dithered signals from

Fig. 7—Output vs. input linearity at 500 Hz, for test generator signals fed via AES/EBIJ digital input.

At this point, eager to get the processor off the test bench and into my music system, I made but one additional measurement: A check of frequency (or pitch) accuracy determined by the crystal clock in the processor. That accuracy was within —0.0003%, which means that a middle A musical note recorded at the standard 440 Hz would be reproduced as a frequency of 439.99868 Hz. I suspect that even persons possessing the most perfect “perfect pitch” aren’t likely to be upset by this minuscule “discrepancy:’

==Use and Listening Tests==

Listening tests were conducted with a minimum amount of electronic equipment in the signal path. The digital output of my reference CD player was fed directly to one of the balanced digital in puts of the Mark Levinson No. 30 processor. The No. 30’s analog outputs were fed directly to an amplifier equipped with its own input level control. The amplifier, in turn, fed my reference KEF 105.2 speakers. To be completely honest about it, I did not conduct any comparison tests against other state-of-the-art digital reproducing equipment. I did, however, listen to a fairly wide selection of my favorite CDs and can attest to the fact that they never sounded better. I attribute the open, airy sound quality of this setup to the No. 30’s superb low-level linearity and to its ultra-low levels of noise and distortion.

Of course, I would not have expected a processor costing some $14,000 to deliver sound that was in any way flawed. Using the Levinson in my listening setup, I realized yet again that when the digital circuitry of an audio system is as perfect as this, the limiting factor becomes the source material.

Comparing some of my earliest CDs with late releases proved highly revealing. Where software flaws were previously masked by hardware imperfections, suddenly they were easily identified. By the same token, well-produced recent CDs de livered the kind of superb sound that digital naysayers have maintained is inherently impossible in a 16-bit digital audio system with a 44.l-kHz sampling rate. If the claim for CD of “perfect sound, now and forever” was an overstatement back in 1983, it’s closer to the truth when playing CDs and DATs using a processor such as the No. 30.

I’m interested to see if Tony Cordesman agrees!

--Leonard Feldman

= = = =

The No. 30 is a truly outstanding reference unit that has led me to reappraise some of my initial judgments of CD sound quality. Many discs sounded far better with the No. 30 than they had with other units.

I evaluated the sound of the No. 30 by comparing it to other top DIA converters like the Theta Digital DS Pro Generation III, the D/A section of a Sony DTC-77ES DAT deck, and to some good CD players such as the Sony CDP-X777ES and Philips LHH500.

I used two different reference systems for these comparisons. The first used a Classé Audio DR-6 preamplifier and DR-25 power amplifiers with Apogee Diva speakers, and the second used a Krell KBL preamplifier, Krell MDA-300 power amplifiers, and Thiel CS5 speakers. As the two systems are in different rooms, with different acoustics, they help me avoid judgments based on a single set of room interactions and listening conditions.

While the Theta did compete with the No. 30 in many ways, the No. 30 outperformed it as well as any competing D/A converter that I could beg or borrow, in three critical respects: The ability to resolve low-level musical and harmonic information, to handle subtle and complex dynamics and musical changes, and to preserve the overall coherence of the music.

The No. 30 manages to combine very flat and extended upper octaves with bass and midrange that do not reveal any special coloration or anomalies. You can check this for yourself by playing back any voice re cording where the singer’s voice has not been heavily processed, such as Victor Braun’s baritone in Schubert’s Winterreise (Dorian DOR-90 145) or Jan DeGaetani’s mezzo-soprano on a recording of Schubert’s songs (Elektra Nonesuch 9-79263-2).

The bass is tight and controlled, superbly defining and revealing differences in bass frequency, and the mid-bass is equally superb. I noted this especially on the bass drum on a recording of the Hoist Suite No. 1 in E Flat (Reference Recordings RR 39CD); the amazing mix of bass drum and orchestra on “Olympic Fanfare,” in the National Symphonic Winds’ Center Stage (Wilson Audio WCD-8824), and the mix of bass drum, soprano voice, chorus, and full orchestra on Mahler’s Eighth Symphony by Robert Shaw and the Atlanta Sym phony Chorus and Orchestra (Telarc CD- 80267). The No. 30 is equally impressive with all the usual organ spectaculars, and is particularly revealing with the final pas sages of Eugene Ormandy’s recording of the Saint-Saens Symphony No. 3, “Organ” (Telarc CD-80051).

The No. 30 may lack a certain richness or warmth found in some competing units, but it is yet another advance in terms of accurate timbre, low-level detail, transparency, and coherence. For accurate rather than emphasized upper bass and lower midrange, I have yet to hear an equal.

The midrange is not only flat and de tailed, it is richly harmonic and’ musical. One of my favorite CDs for testing this is the Scott Reiss and Hesperus recording on Golden Apple (GACD 7750) for recorder, violin, viola, and harpsichord. These are very demanding instruments, and playback with most digital processors can be very fatiguing. The Theta Generation III is the only other unit I have found that makes this recording equally listenable, but it does not provide equal resolution of midrange detail in the softer passages.

The upper midrange and treble of the No. 30 are very flat and extended. This can present a slight problem with mediocre recordings, particularly those made with microphones placed too close to the performers or with older digital equipment. Some listeners may prefer sweeter and warmer sound over detail and accuracy. Any serious audiophile, however, is going to want that detail with good recordings, and particularly the recent recordings made with 20-bit equipment. Here the No. 30 is clearly the state of the art.

The Levinson provided the best performance to date with two recordings I use to evaluate the upper midrange and treble in digital processors: The soprano voice in the Julianne Baird/Ronn McFarlane disc The English Lute Song (Dorian DOR-90109) and the sound of the harpsichord in Cohn Tilney’s recording of Bach toccatas for harpsichord (Dorian DOR-901 15).

The No. 30 reproduces the spatial information on the recording and not some ideal in the designer’s mind. To hear this, compare the soundstage of virtually any close-miked or multitrack CD—Jennifer Warnes’ Famous Blue Raincoat ( Cypress 661 111-2) or Willie Nelson’s Always on My Mind (CBS CK-7464-37951-2) are particularly good CDs to try—with the more natural soundstage on a Telarc or Dorian re cording. The No. 30 lets the differences come through clearly without changing the apparent size of the soundstage.

Similarly, the No. 30 makes no attempt to provide sharper imaging than is heard in live performance, sin that often trades an apparent improvement in left/right definition for an artificial narrowing of individual images and/or loss of depth. The overall merits of accuracy over artificiality are particularly striking in a CD with natural orchestral sound, like the Malcolm Arnold recording of Arnold: Overtures (Reference Recordings RR-48CD), as well as in Dorian, Telarc, and Wilson CDs recorded in concert halls or churches. I also prefer the No. 30’s natural imaging with recordings that were produced with more defined imaging, such as the guitar and percussion in Bruce Dunlap’s About Home (Chesky JD 59); the mix of voice, instruments, and rain sounds on “I Get the Blues When It Rains” from Eileen Farrell’s It’s Over (Reference Recordings RR-46CD), and the complex percussion imaging in Jonathan Haas’ Virtuoso Timpanist (CRD 3449).

Depth is neither enhanced nor restricted by the No. 30. If you listen to the latest Telarc symphony recordings (try track 1 on Mahler’s Eighth Symphony), the Dorian recordings in the Troy Music Hall, or a good baroque recording like Ton Koop man’s version of the Handel concertos for orchestra (Erato 4528-2), you will hear all of the depth that you expect to hear and a remarkably lifelike performance.

The No. 30 brought a new degree of harmonic coherence to the individual notes of guitar, harpsichord, and violin and to the upper frequencies in piano and cymbals, particularly in low-level passages. I believe this superior performance is traceable to two specific advances in sound quality. One is sheer resolving power; the other is transient speed and detail that are excellent at any level of dynamic energy.

The Mark Levinson No. 30 combines excellent transparency with musical credibility and coherence. It also has a remarkable ability to separate and define instruments and voices in complex pieces of music. This ability became clear with every decent re cording I have of grand opera or massive orchestral music. It came through beautifully on the Telarc recording of Mahler’s Eighth Symphony, in the much lighter fare on the Mackerras and Welsh National Symphony Orchestra recording of The Mikado (Telarc CD-80284), and on good organ and choir recordings such as the Proprius label’s classic Cantate Domino (PROP CDP 7762) or the Bach Choir of Bethlehem’s recording of Bach’s Cantata No. 140, “Wachet Auf” (Dorian DOR-90127).

As might be expected from its superior ability to reproduce low-level passages, the No. 30 has the lowest apparent noise I have heard in a digital processor. It extends the noise floor you hear in good recordings by removing an additional layer of haze from low-level passages.

The No. 30 performed best with the Madrigal MDC- I Pro digital interconnect, a three-conductor balanced cable for AESI EBU connections. It performed nearly as well when connected via a version of the MDC- 1 Pro cable with an RCA plug at the source end, although I would advise using the AES/EBU connection on those many high-end CD transports and professional DAT decks that have them. Optical cables are a definite second best.

Outstanding resolution in the upper midrange makes this unit more revealing of problems in recordings that relied on poor production values, some of the older and harsher analog-to-digital recording equipment, and aggressively close miking. There is nothing sweet or forgiving about the No. 30.

As is the case with every digital processor I have heard to date, the No. 30 also sounded better with some good recordings than other good ones, and there were a few odd cases where it did not perform as well as I expected with a good recording. No audio component is ever all things to all recordings. There also were a few moments where I would have liked just a touch more bass energy, particularly in the deep bass, and felt other units provided a preferable amount of musicality—although always at the cost of low-level detail.

I believe that the No. 30 really does achieve a level of natural resolution and detail that is more faithful to the original recording than any competing D/A converter, CD player, or digital tape unit I know of. I do not want, however, to leave the impression that it is absolutely without competition. At the very top, the balanced version of the Theta DS Pro Generation III competes in musicality, being warmer and more dynamic.

Madrigal Audio Laboratories would be among the first to agree that it is not necessary to pay $13,950 to have many of the advances in digital sound quality pro vided by the No. 30. The company now offers a No. 35 processor, and it costs just under $8,000. Madrigal’s less expensive Proceed line offers a less sophisticated mix of products with many of these advances. Firms like Audio Research, Krell Digital, Theta Digital, and Wadia Digital offer excellent mid-priced units. Many mainstream manufacturers have top-of-the-line digital products that sound far better than their offerings of one or two years ago, and many of the lesser known high-end companies, such as Audio Alchemy or PS Audio, offer products with outstanding performance.

My subjective comments have tended to focus on recent recordings, including a number made with 20-bit mastering. However, you do not have to listen to audiophile recordings to benefit from units like the No. 30; you will get a similar improvement in listening pleasure with virtually any good recording. While I still do about half of my listening to LP records, I believe that unbiased listening to the Mark Levin son Reference Digital Processor No. 30 reveals that digital sound can now provide a level of information and musicality that analog almost always cannot. If you haven’t heard this unit in a state-of-the-art system, playing the very best in recent CDs, you do not know what the state of the art is or what digital sound can achieve.

Also see: