Tubes Versus Solid-State Audio Amps—The Last Word (Or “House Of Fire,” Part 2)

Fig 1. This basic source-follower circuit is the kind of amplifier you would find in an electric bass guitar amplifier. R6 provides the feedback. On the other hand, low-distortion applications require extensive drive circuitry and usually run common-source.

Fig 2. The basic building block of most linear audio amplifiers is the emitter-follower circuit. Q7 and Q8 sink base drive current on over-current events (once the ballast resistors overcome Vbe). High-fidelity purist versions of this amplifier may use cascoded elements in the drive stage for higher bandwidth.

Fig 3. The most popular vacuum tube is the triode, which works a lot like a JFET, with the cathode analogous to the source, the plate or anode analogous to the drain, and the grid analogous to the gate.

Fig 4. Tubes also come in tetrode and pentode versions, with added grids to control the electron beam.

Fig 5. While probably not dead-on accurate, this schematic gives an idea of how Dunipace created a drop-in experiment by setting up a bias point on the MOSFETs, providing necessary thermal compensation, and reducing the gain enough to show that that the solid-state output devices performed like tubes.

In an attempt to figure out why a vacuum-tube amplifier sounds different than a solid-state amplifier, Part 1 considered what we can hear, what we can discern, and some of the attributes of passive devices that affect audio design (see “‘House Of Fire’: Firebottles And Groove Tubes Versus Devices That Find Their Origins In Sand (Part 1)”).  

The article discussed two extreme applications: a live performance with a guitar amplifier and one that required absolute accurate reproduction. Part 2 examines the active devices, the amplifier topologies, and, lastly, an experiment that shattered the myth that tubes sound better than transistors—all other things the same.

Active Devices: MOSFETs

There simply aren’t many MOSFETs available for linear amplifiers in the audio world. Linear MOSFETs are typically lateral devices that have no intrinsic body diode. The higher-gain MOSFETs used in switch-mode power-supply (SMPS) applications often won’t work in linear amplifiers due to hotspotting at low currents and high voltages in linear-mode operation. This was discovered at International Rectifier by a researcher named P. Spirito, and consequently named the Spirito Effect (see “The Spirito Effect Improved My Design—And I Didn’t Even Know It”).

Some older planar-technology vertical devices have larger features and lower gains. They are available in complementary form and can be used in linear amplifiers. A few vendors still build linear amplifiers with MOSFET power stages. These are conventional, low-output-impedance, voltage-sourced amplifiers, though. They have a lot of feedback, and the output devices handle the output current directly.

The MOSFET’s square-law transfer characteristic creates some subtle differences in sound quality that arise from its tendency to produce more even harmonics on overdrive and clipping program excursions. The drive stages to the MOSFET output stage usually have to be markedly different. For a common-source configuration, all drive signals must be translated up or down to within a few volts of the rail. For a source-follower topology, the drive is easier, but the bias point requires extensive consideration. IR application note AN-948 depicts a source follower amplifier (Fig. 1).

Bipolar Junction Transistors

Admittedly, I’ve spent most of my time working on, and with, BJT-based (bipolar junction transistor) audio amplifiers. The BJT has come a long way from that hetero-junction, point-contact device that Haynes and Schockley demonstrated. That device was later made into a monolithic device by Schockley and eventually brought into mass production by masterful minds like Gus Mellick. BJTs continue to be exploited into wonderful, novel applications by ingenious folks like Richard Dunipace at Fairchild Semiconductor.

The audio transistor used in a linear amplifier’s output stage typically is a fairly high-voltage device. Most substantial amplifiers use 250-V transistors, rated for somewhere around 15 A. These devices are designed for high linearity of Ib versus Ic. The Ft of these devices ranges from a couple of megahertz clear up to 30 MHz. They are designed for a large safe operating area (SOA) and fairly high power dissipation.

There are two dominant output-stage topologies that use BJTs. The most popular is the emitter-follower stage. In this configuration, the collectors of the NPN and PNP devices connect to the B+ and B– supplies, respectively, and the emitters are ballasted by a small resistance value (Fig. 2). Compound devices (predominantly Darlington) are used in modern high-power amplifiers due to the need for very high current gain. Many devices are used in parallel to accommodate the current required by the low load impedance of the output transducer array. The ballast resistors balance the currents in the devices and overcome the negative temperature dependency of Vbe versus temperature.

This type of output stage is purely a current-gain stage. Therefore, it requires a symmetric voltage-gain stage as a driver. This arrangement offers a bandwidth advantage: Because the emitter-follower stage has a voltage gain of 0 dB, Ft can be applied directly. The output impedance of the emitter follower stage is that of the base circuit divided by the gain of the device (actually β + 1). This is a small number in most applications.

As noted, the transistors are typically Darlington-connected devices with current gain. If the lumped impedance of the base circuit is 1 kΩ or so, the output impedance in open-loop configuration is in the range of a few ohms. The global feedback circuitry used in these types of amplifiers reduces this output impedance substantially.

Saturation is by and large avoided by having the emitter follow the load voltage. But saturation (and the recovery from saturation) does need to be considered if short-circuit loads are to be sustained. However, most designs circumvent this with over-current protection placed across the ballast resistor to reduce the base drive to the Darlington element beyond a given IR drop in the ballast resistor. In Figure 2, this is Q7 and Q8.

The disadvantage of this output stage is the temperature versus Vbe stability. As temperature goes up, Vbe drops. Higher-gain devices will see a sharper drop and, thus, a tendency to handle disproportionally large currents. Locally, we can force the devices to share with ballasting resistors in the emitters. However, globally, we have to hold up a bias point in either class AB mode or class A mode of operation. To keep this bias point stable, we need to use a Vbe multiplier circuit, a diode array, a thermistor compensation network, or some combination of these three approaches. In Figure 2, this is Q9, a Vbe multiplier. This output stage can be viewed as a near ideal voltage source.

The other predominant approach used in BJT output stages is the common-emitter configuration, where the emitters of the PNP and NPN devices are tied to B+ and B–, respectively, again with appropriate ballasting. This stage provides both voltage and current gain. It usually requires an emitter-follower stage as a driver. The advantages of this stage are higher output impedance in open-loop mode, improved biasing stability, and improved short-circuit performance compared to the emitter-follower stage. The downside is that Ft comes into play, limiting the amplifier’s bandwidth for a given gain level. Again, these amplifiers are typically built in a closed-loop configuration with a lot of global feedback from input to output.

Tubes

The most popular vacuum tube is the triode, which consists of a cathode, a grid, and an anode (Fig. 3). It also has a heater (not found in solid-state devices!), which creates an electron flux that is modulated by the grid and results at the cathode. Sound familiar? Like the lateral drift region in a field-effect transistor modulated by the gate voltage? An easy model to keep in your head is one of a JFET, where the grid is analogous to the gate, the cathode is analogous to the source, and the plate or anode is analogous to the drain.

The vacuum tube will conduct a lot of anode (or drain) current for a near zero voltage applied to the grid with respect to the cathode. As the grid voltage is driven negative with respect to the cathode, the tube conducts less and less current—very similar to a JFET—with a pinchoff voltage of –8 V or so.

That said, there are some differences. There are tetrode vacuum tubes and pentodes (Fig. 4). These devices feature added screen and suppressor grids. Most of our favorite beam power tubes are pentodes, yet they are most often configured as triodes, with the suppressor grid tied to the cathode and the screen grid held at a neutral potential.

A tube’s impedances are much higher than what we are accustomed to in the solid-state world. Consider a practical example. If the tube is “Off” at –8 V Vgc, and “On” at 0 V Vgc, we have something less than an 8-V operating range for a linear amplifier, say 5 V. If the transconductance is 4000 µMhos (Siemens) or 0.004 A/V, we can then cause a maximum change in plate current of 20 mA. The anode or plate voltage in this situation would be on the order of 250 to 500 V.

A typical transconductance for a modern MOSFET is 200 Siemens or 200 A/V. This is 94 dB hotter than a vacuum tube in terms of gain. And that does make sense. The gate in a silicon MOSFET is a very short distance from the channel, usually fractions of a micron. The channel forms a conducting “slab” that might have resistivities on the order of 7 x 10–8 ohm-meter in the On state and 7 x 10–2 ohm-meter in the Off state.

In a vacuum tube, the grid is about 1500 µm away from the cathode, separated by a 3D gas. If we view this gas as a conductive slab, as in the MOSFET case, the tube’s “slab” is much larger. The slab will have a resistivity on the order of 2 x 102 ohm-meter in the On state and 2 x 1010 ohm-meter in the Off state.

Considering the spatial relationships, we’d expect the dynamic capacitances of the vacuum tube to be far less than the silicon-based devices and indeed they are. We’d also expect the SOA to be able to sustain peaks well outside of the designed operating range. Perhaps this is severe overdrive in a push-pull audio amplifier, or perhaps a very high voltage standing-wave ratio (VSWR) in a tuned RF amplifier. Tubes can handle things like that until the anode dissipates enough power to soften the glass envelope and cave it in. Solid-state devices can’t handle that kind of SOA violation for very long. There’s just not enough space to contain it.

As the tube contaminates, the Off state resistivity of the “slab” drops. As the heater ages, the on state resistance of the “slab” goes up.

Clearly, with output characteristics and impedances like this, the vacuum tube needs a matching transformer to drive an 8-Ω load. We discussed that in detail in part one of this article. The transformer brings in its own subtleties, the most predominant of which is damping, assuming the transformer was well designed for the application.

A vacuum tube is almost a square-law device. The fundamental transfer characteristic for a tube is Ia = KE1.5, where Ia is the anode current or plate current, K is a constant for the tube geometry, and E is the plate-to-cathode voltage.

Another aspect of the triode circuit biasing is often exploited. The grid bias voltage is derived from the input power supply, which is seldom regulated. As the input line droops, the bias voltage on the grid approaches zero. Remember that this is approaching the On condition.

As the voltage input to this type of amplifier droops, the bias current goes up. This will often take a lightly biased class AB amplifier into hard class A operation and offer different sound characteristics, namely more even-harmonic content. Folks like Eddie Van Halen have exploited this extensively in various works by “dimming” the amplifier with a variac and forcing this hard class A operation.

The Drop-In Experiment

A while back, Richard Dunipace and I worked together on non-audio circuitry. But we shared a similar passion for sound and measuring the oddities that set amplifiers apart. I came up with a straightforward idea—replace the output tubes in a high-end, push-pull, triode-type tube amplifier with high-voltage MOSFETs. We studied and conferred at length, and Richard developed the necessary ballasting, biasing, level shifting, and local feedback circuitry to emulate a triode-connected 6L6 vacuum tube with 1200-V MOSFETs (Fig. 5).

What we found, which led to subsequent investigations on damping, passive components, transformers, inductors, and transducers, was that the MOSFET replacement sounded and measured identical to the tube within 1/2 dB. Further, the MOSFET version measured the same on THD plots, burst modes, square-wave response, and other dynamic testing. The output tube was not the reason that the amplifier sounded the way it did. Most of the tonality that we heard came from the pre-amplifier, the inverter stage that drove the output devices, the output transformer, and the transducer.

We later proved this a second time by taking horizontal output transistors (low-gain, high-voltage BJTs) and plugging them into the same application, with their own thermal compensation, level shifting, and local feedback. Again, the amplifier sounded and measured within 1/2 dB. We never got around to testing insulated-gate bipolar transistors (IGBTs). We concluded that the sound difference wasn’t dominated by the devices used. Hats off to Richard for working up these circuits, usually dead bug style, and painstakingly stabilizing and measuring the results.

And There’s More

Other factors to consider when comparing one sound to another are the speaker cables, electromechanical connections, op amps in the pre-amp stage, and perhaps long-tail differential pairs, the type of feedback, the gains, and a virtually endless list of things that make contributions to sound quality in the 1- or 2-ppm range (120 dB down).

I once sat in a listening room trying to discern ppm interactions between an input coupling capacitor and the metal cover a few inches above it. As we were working, the sound engineer casually got up and tied a half-hitch knot in one of the speaker cables. He replayed the program material (Supertramp, “Bloody Well Right”), and we both agreed that there was a subtle difference in the cymbals and high vocal passages.

I suspect we were hearing a few extra picofarads-to-earth ground and a few extra nanohenries of series inductance from transducers and crossovers whose inductances were in the millihenries, with capacitances in the microfarad range—again, ppm levels. You wouldn’t hear that on stage with a bass guitar amp, but in a pristine listening environment, it was very clear.

Nonlinear amplifiers have caught up to where linear amplifiers left off. Basically, all negative attributes of switching amplifiers have been overcome by Jun Honda’s work on the IR class D audio amplifiers. Please give the IRAUD AMP7S a spin if you are skeptical. This circuit features the IRS2092 driver IC. Yes, there is a lot of global feedback, akin to most any other voltage-sourced amplifier, but the amplifier is extremely accurate with wonderful tonality. The switching frequency is typically around 400 kHz. The evaluation board is set up for accurate reproduction with low distortion. 

Conclusion

This article resulted from an article by Communications Editor Louis E. Frenzel comparing tubes and transistors (see “Tubes Are Still Better Than Transistors For Audio Amplifiers”).

I wanted to delve into the subject and try and explain some of the sound differences that were wrongly perceived as being purely related to the vacuum tube in both the purist extreme application and that of live audio and guitar amplifiers. We proved that the active devices are a fairly small portion of the observed audible differences by a drop-in experiment, which led to investigating the rest of the signal path and the output damping factor.

You can easily approach the “tube-amp sound” by simply placing a substantial resistance in series with the output of a modern voltage-source audio amplifier. This decreases the damping factor and allows the bell modes on the transducer cone and dust cap to radiate as opposed to being clamped by an output impedance that is effectively zero.

If you are serious about audio design, take a moment and peruse the references below. They contain a lot of useful information well beyond what I can offer in a few thousand words. They will help you, whether you work with doped devices made from sand or fire bottles in guitar amplifiers, purist monoblocks, or anything in between.

Bibliography

1. Bateman, Cyril, “Capacitor Sound,” series of articles in Electronics World, 2002 to 2003. Reproduced at: http://www.proaudiodesignforum.com/forum/php/viewtopic.php?f=6&t=153&start=2 or http://www.scribd.com/doc/2610442/Capacitor-Sound.
2. Terman, Frederick Emmons, Radio Engineering, McGraw-Hill Electrical and Electronic Engineering Series, 1947.
3. RCA Receiving Tube Manual, Technical Series RC-25, Radio Corporation of America, Copyright 1966.
4. Eargle, John M., Loudspeaker Handbook, ISBN 0-412-09721-4, Kluwer Academic Publishing, 1997.
5. Self, Douglas, Audio Power Amplifier Design Handbook, ISBN 0-7506-5636-0, Tag McLaren Audio, Elsevier.
6. Self, Douglas, Self on Audio, ISBN 978-0-7506-8166-7, Elsevier.
7. Van der Veen, Meeno, “New Push-Pull Amplifiers,” article on tube amplifier topologies with a great discussion, http://www.next-tube.com/articles/Veen2/Veen2EN.pdf.
8. Dunipace, Richard, “Audio,” [email protected].