Mark Levinson No.38S preamplifier

Even as Robert Harley was writing his Stereophile review of the $3995 Mark Levinson No.38 remote-controlled line preamplifier (it appeared in August '94, Vol.17 No.8, p.98), Madrigal Audio Laboratories announced an upgraded, cost-no-object version, the No.38S (footnote 1). At $6495, the 'S is significantly more expensive than the junior version; although it uses the same chassis, power supply, and circuit topology, it's in all other ways a different preamplifier.

The No.38S features more than three dozen refinements, all of which reportedly improve its sonic performance. The big change is the use of a four-layer printed circuit board in place of the '38's two-layer board. While the board's dielectric is not the superb-sounding Teflon used in the Levinson No.26S—you can't get multilayer Teflon pcbs—it is made from an unidentified material developed for VHF communications use, and is said to share some of Teflon's electrical characteristics. The inner two pcb layers are "power planes," while the top layer appears to be a ground plane, giving the voltage rails and ground reference very low source impedances—as if, Madrigal says, "there was a very fast bypass capacitor on each IC supply pin."

As a result of extended listening tests to passive components, Madrigal's engineers focused on a small number of critical components in the '38 circuit that would benefit from being replaced by premium parts (footnote 2). Fourteen resistors in each channel were replaced by close-tolerance bulk metal-foil types, and one capacitor per channel was replaced by a Teflon-dielectric type.

Interestingly, the actual gain elements in the '38S are Burr-Brown OPA2604 op-amp ICs, two per channel, combined with a current-mode stage to form a semi-discrete instrumentation amplifier. The OPA2604 is a premium chip, widely respected in the world of analog audio design.

For a detailed functional description, I refer readers to Bob Harley's August 1994 review of the No.38. But basically, the circuits for the two channels are separated by the power-supply and control circuitry, these screened by an internal can. This, together with the use of opto-isolators to connect control information to the volume-control and selector relays, results in a very low noise-floor. The input-selection relays are adjacent to the input jacks, minimizing the length of pcb trace the signal has to encounter, while the volume control uses a two-channel MDAC chip, one per channel.

The "M" in MDAC stands for "multiplying"; I first saw the use of a D/A converter as a digital-word–controlled volume control in a ReVox amplifier of the early 1980s. Usually, to convert digital data to analog, data words are fed to the MDAC's data input and control the arrangement of a precision resistor chain. A very steady reference DC voltage is fed to the chip's reference pin and is divided down by this resistor ladder to produce an output voltage proportional to the input data word. The MDAC can therefore be viewed as a digital-word–controlled variable resistance. It doesn't take a huge leap of imagination to envisage feeding an audio signal to the MDAC's reference voltage pin and using the MDAC as a variable resistor in a voltage-divider chain, say, or in the negative feedback loop of an amplifier circuit. Changing the digital words to the MDAC will then result in the analog signal being produced at different output levels—what we laymen refer to as a "volume control."

Although using an MDAC in this way requires the designer to be careful of unforeseen problems—that early ReVox amplifier sounded pretty poor—the advantages of it are: 1) very close matching between channels—DAC resistors have to be of inherently closer tolerance than even the expensive kinds you find in a parts bin; 2) it's a trivial matter for the designer to implement mute and balance controls or program different inputs to have different sensitivities; and 3) the designer can produce any kind of volume-control "law" he or she wants simply by selecting which set of data words is sent to the DAC.

The MDAC used by Madrigal in the '38 and '38S is, I believe, a 12-bit device—meaning it can produce 4096 equal-level steps. Madrigal's engineers have chosen to provide a control range of 92dB in precise 0.1dB steps. Below an indicated "73.2" on the '38S's front panel, the volume control acts as an attenuator; above that level, an active gain stage provides up to 18.9dB of voltage gain.

The control data are provided by an optical shaft encoder connected to the large knob on the front panel. This has been programmed to have a useful action: spin it fast and the volume changes very rapidly; spin it slowly and the volume changes very gradually. Nice!

Consider how similar in shape sharks and dolphins are. Yet one is one of the oldest species of fish, the other the descendant of a bear-like land mammal that returned to the sea. This process of very different organisms developing almost identical superficial physical qualities is termed convergent evolution.

The same phenomenon can be observed in audio: at the very high end of preamplifier design, it really doesn't matter whether a designer chooses to use tubes, FETs, or even op-amp chips—the goal is to produce a control center with absolutely no character of its own. As preamplifier performance improves, therefore, it becomes harder to characterize a product as sounding any one particular way. Thus it was with the No.38S.

Throughout the review period, differences between recordings and D/A processors all seemed much greater in magnitude than anything I could ascribe to the Levinson preamp. Rather than discuss its sonic signature in absolutes, therefore, I'll contrast it with both the No.38 and two other leading line stages and let what it does emerge from the comparisons (see Sidebar).

Against the No.38: Before the '38S took pride of place in my system, I used the sample of the Mark Levinson No.38 reviewed by Robert Harley for several months. The basic '38 is functionally the same as the 'S version, but soundwise, there was no contest. The balance of the less-expensive preamp is smooth, but in comparison with the 'S, it lacked detail and clarity. It's not that it sounded dark in, say, the way the Melos SHA-1 does, but the No.38 failed to get enough of a handle on the excitement embedded in the recordings. On the bright side, the '38 sounded very inoffensive, and no one who visited my listening room was bothered by anything wrong with the sound. But neither did they feel moved to comment overmuch on the things that were right with the sound.

Footnote 1: I do note that—unlike the Nos.30.5, 31, 33, and 20.6—Madrigal has not labeled the '38S one of their "Reference" products. Does this mean they think they could do even better one day? Or does it mean they see the days of the traditional analog preamplifier/control center as being numbered?

Footnote 2: See Ben Duncan's essay elsewhere in this issue on the effects of parts tolerance on sonic performance.

Mark Levinson division of the Harman Consumer Group
1718 W. Mishawaka Road
Elkhart, IN 46517
(516) 594-0300