August 2001

Audiopax Designer Eduardo de Lima has introduced his new statement 3880 tube monoblocks. SoundStager John Potis and I thoroughly enjoyed our prior sessions with Audiopax designs. As any two good-natured collegial competitors would, we vied for the 3880's world-premiere review honors. Lady Luck smiled upon John -- look for his review toward the end of the year. My contribution will be a brief sidebar to John's review -- and this month's column, which deliberately omits sonic impressions. Instead, it investigates the engineering solutions that enable the 3880's unique combination of traits.

Traditional options

The new Audiopax amps are single-ended tube designs. However, they're neither single-ended triode, pentode nor ultralinear. With two KT-88s per channel, they're not paralleled single ended either. Instead, they work in an advanced version of a proprietary circuit dubbed LM3. Spelled out, it becomes low mu triode with higher raw efficiency emulator. To understand the significance of this basic circuit, we'll first revisit the traditional options for connecting a pentode in class-A triode operation before looking at the evolutionary changes in the 3880.

The conventional triode-connected pentode scheme hooks the screen grid to the plate. Including power efficiency, this has the original pentode work very much as a triode. It now exhibits its own fixed characteristics. For example, a triode-connected KT-88 will have very similar plate resistance and power efficiency to a 300B, but a little more gain and higher distortion levels.

Enhanced triode mode uses the grids of a pentode in a different manner. When compared to conventional triode connection, the final result usually behaves as a triode with lower levels of distortion but much increased plate resistance and naturally much higher output impedance. The plate resistance in this connection is just about the same as regular pentode operation.

Depending on the amount of screen coupling, the distributed load connection called ultralinear has the pentode behave anywhere between a pure pentode and a pure triode. At some optimized point in between -- and depending on the specific circuit architecture -- the best characteristics of each may be combined. For example, a KT-88 as pentode has a plate resistance of around 11,000 ohms. In triode mode, this lowers to around 650 ohms, while the typical ultralinear scheme increases again to 2000 ohms. (These are average values that discount specific bias settings.) In a single-ended circuit, the differential between 650 ohms and 2000 ohms of plate resistance drastically affects how an amplifier will sound. Attempts to lower plate resistance much below 2000 ohms will end up once again with a basic triode-connected KT-88.

Another operating mode called partial cathode coupling was first introduced in the original push-pull Quad II amp. This circuit allows considerable reduction of plate resistance/output impedance and distortion levels while keeping the power efficiency as high as the pentode. Since there's no such thing as a free audio lunch, the amount of coupling for optimal distortion behavior is not equal to what's required for best output impedance. (When calling it "best" output impedance, de Lima reminds us that this does not necessarily equate to "lowest," as convention would have us believe.)

A novel architecture

The Audiopax LM3 output stage is a variation on partial cathode coupling. It also is a sophisticated model of local feedback. Local feedback is limited to just one amplifier stage, while global feedback is applied around more than one stage. An un-bypassed cathode resistor is considered local feedback. The loop from the output tap to the input-stage cathode resistor is global feedback. De Lima pointed out that within this definition, it's perfectly valid to call a triode a pentode with intrinsic feedback (low mu triodes then possessing the highest amount).

His LM3 circuit employs screen derivation, a tertiary cathode winding and a resistor without bypass in the cathode leg. (This resistance could also be imbedded in the tertiary windings of the output transformer.) De Lima warned that his particular scheme involves much more than just combining an ultralinear tap with a tertiary winding for cathode coupling -- he hinted at how controlling the actual resistance in the cathode leg is just one of many additional and important variables. The main difference to the classic partial-cathode-coupling implementation is said to be not so much the technique itself, but how and towards what specific ends it is employed. In the LM3 circuit, the undisclosed control parameters are focused on achieving a very specific combination of attributes -- distortion levels, output impedance and varying distortion spectra aligned to different power levels -- to realize a very particular sound.

Another slap in the face of convention

Of course, no SET circuit is complete without specialized output transformers. Already their size makes those in the 3880 visibly different. As one goes upscale from affordable to more and more expensive single-ended-triode amps, the traditional tendency is for increased transformer mass and size to prevent saturation and enhance low-frequency performance. The Audiopax transformers remain positively Lilliputian to minimize primary inductance, which, in conjunction with other circuit-design parameters, is said to actually give much better bass than larger transformers would. De Lima also proclaims their construction devilishly complex and perfectly safe from any reverse-engineering attempts. His transformers contain more than 20 different coils and feature a special grain-oriented iron core with a highly unusual treatment.

Precedents

De Lima's initial engineering goal for the LM3 circuit's earlier incarnations was the optimization of the relationship between screen coupling, cathode coupling and cathode resistance. He wanted to coax a pentode into behaving like a low-mu 300B triode in the areas of gain, output impedance and overall distortion characteristics. A push-pull precursor of this circuit called super ultralinear was first mentioned in a 1950s article by Thomas Burroughs. It used a transformer by the Chicago Standard Transformer Company. This circuit never materialized as a successful commercial project precisely because the necessary transformers were forbiddingly complex and expensive.

Evolution

With his 3880 monoblocks, de Lima set his sights beyond the LM3's original design goal to add a very specific distortion spectrum. When combined with actual loudspeakers, it should produce the immediacy and timbral purity of 45 or 2A3 micro-power triodes, but also dispense with the output power of 211 or 845 power triodes. Since no existing plug-in tube combines these disparate traits, de Lima pursued his dream the same way he had gone after counter-intuitive results before -- through clever circuit implementation. His Audiopax 3880s are the first expressions of this concept and an evolutionary culmination of earlier expressions of the LM3 model.

Each 3880 monoblock features two entirely separate single-ended amplifiers inside its chassis. Thus made up of two half-channels, each monaural amplifier consists of two pi-filter chokes, two output transformers, one dual-windings filament transformer (with separate feeds for the 12AT7 input and KT-88 power tubes) and one shared power transformer with separate windings for each half channel. And no, it’s not a balanced circuit. What's truly radical are the two output transformers that aren't exactly identical. This minor asymmetry enables de Lima to control the harmonic-distortion tuning of his KT-88 duo to an extent beyond previous designs.

On the surface, this arrangement might be mistaken for a classic paralleled SET model. It's not. To begin with, the use of two completely independent power supplies is said to guarantee far superior stability. It prevents one tube from dominating another -- a problem apparently very common to paralleled schemes. Minor differences in tubes -- due to original condition and later aging -- are amplified by power-supply interactions. This requires precisely matched tubes and renders paralleled circuits inherently susceptible to bias drift and related problems. For Audiopax, more important than avoiding these limitations was the ability to control the average distortion behavior (as it ties in complex patterns to power and frequency) much more tightly than paralleled tube interactions would allow.

Instead of paralleling his KT-88s, de Lima therefore combines the two non-identical amplifier halves at the transformer outputs in series. This produces what he calls an "aggressively conservative" output rating of 28 watts. It realizes the first of his three design challenges that demanded 845 or 211 power-triode drive. Another challenge had specified the use of easily obtainable non-exotic tubes that would remain affordable and reliable throughout the usage cycle of his product. The four KT-88 output and four 12AT7 input tubes sourced from Penta Lab fulfill that second requirement. Did his third come to fruition, the sonic finesse and recognizable aural aroma of 45 or 2A3-type micro-power sound? We'll await John's review to determine that.

What we can already agree upon is this: to even stand a chance of making a KT-88 pentode sound like a 45 triode requires a very fundamental insight into why a 45 triode sounds exactly like it does. De Lima's six-years-and-counting mathematical research into tube-distortion behavior (and how it interacts in the real world with speakers, not load resistors) provides the foundation upon which his specific design approach rests. He employs circuit and transformer models to duplicate or clone the very complex distortion signature of the tube he wants to emulate. Simultaneously, he takes full advantage of the greater linearity and output power of the tube type he's identified as the most suitable substitute.

Added wrinkles

Such a design brief in and of itself already appears mighty ambitious and without precedent. However, de Lima thought up yet another piece of uniqueness, as though he couldn't leave well enough alone. Two top-panel-accessible trim pots on the 3880 allow bias-current adjustments (from 2.6 to 3.2 on a volt meter's arbitrary 20V scale). Depending on the final distortion behavior of a specific amp/speaker interface, optimum results may be obtained by running the KT-88s at identical settings within a preset window or -- and this is the unique part -- at staggered bias values. Such an offset does not affect frequency response. Rather, it's claimed to fine-tune tonality purely in the distortion domain, and as a function of component interaction. My Triangle Ventis XS speakers clearly indicate that 2.7 and 3.0 are my optimal settings. The Avantgarde Duos sound best at 3.05 and 3.20. The differences aren't subtle, either.

More to come

Pray tell then, what do these amps sound like? Sorry, but I've just enough pixels left to spit out two more provocative words: cliffhanger time!

...Srajan Ebaen
srajan@soundstage.com 

• The Y-Files Current Issue