Borrowing the title of the Alice Cooper song seems appropriate. This is a classic, and hot, topic: vacuum tubes versus transistors. Why does a tube amplifier sound different than a solid-state amplifier? This article attempts to solve that mystery looking at the component and system level in a comprehensive—and agnostic—way. (For more, see Part 2.)
This will require a discussion of not only the active devices but also passive devices and transducers and basic audio circuits. These discussions will challenge the stuff we’ve blindly accepted for the last five or six decades. The baseline assertion is that if we can hear a difference between something and something else, we can measure said difference.
The probability that this article will shatter all myths and end the discussion once and for all is absolutely nil. As creatures of habit, we like what we like. That part is subjective. I’ll try my best attempt to stay away from hearsay “noise” and stick to the facts.
In fact, there are distinct differences that must be reinforced throughout the discussion. The primary difference is in the application of the amplifier. If the amplifier is on stage, used in a performance, certain distortions sound good. They enhance the performance in perhaps a bass amplifier or guitar amplifier. We wouldn’t want an absolutely pure amplifier in these applications.
On the other hand, if we are in a playback room listening to a recording, or perhaps in a home theater environment looking for the utmost in accuracy, we seek a different amplifier altogether. In that case we want something that is absolutely pure. Most applications lie between these two points, but both extremes must be discussed.
What can we actually hear? The answer may startle you. At low frequencies, we can discriminate phase to within thousandths of a degree in our heads. This is ultimately what drove all of that low-jitter digital-to-analog converter (DAC) stuff in CD players a few years ago.
Our ears are amazing. It turns out that we can hear jitter at the low end down to fractions of a microsecond. What our brain does with that information is even more amazing. The brain interprets this as a wishy-washy soundstage and muddy bass. We can also hear subtle differences in music passages that are 130 dB down from the highest signals. Then there’s the matter of bandwidth.
This brings us to the statement, “If we can hear it, we can measure it!” I wholeheartedly agree. The trouble is that if our head has 130-dB dynamic range and can discriminate phase at or near phase-locked loop (PLL) levels, that 90-dB notch filter-type distortion analyzer isn’t even close. But there are folks that make wonderful instrumentation for audio measurements. It’s not cheap, but their equipment can measure anything we can hear—if used properly.
So what are we actually hearing? The answer is complicated because when you compare a solid-state amplifier and a tube amplifier in a given application, there are many interactions in the equipment beyond the quantum, small signal, and large signal events in the active devices.
The capacitors, transformers, speakers, resistors, gain stages, inverter stages, cabling, and damping factor all have an impact on sound quality. In most cases, the vacuum tube circuit topology is so different that we are hearing stuff well beyond the active devices. Let’s explore these things one by one (see the figure).
Any discussion of resistors needs to go back to the fundamental resistor noise equation:
where K is Boltzman’s constant (1.374e-23 joule/degree Kelvin), T is the temperature in degrees Kelvin, R is the resistance, and ?f is the bandwidth. This is the noise contribution of the ideal resistor—in the absence of any other effects. But there are other effects.
Carbon composition resistors aren’t that great for audio. At high voltages, they are horrible. The carbon material has all sorts of micro-arcing issues that make huge noise contributions well beyond the fundamental resistor noise. In terms of noise temperature, carbon comp resistors should be about the worst type.
Also, the carbon film material in the input attenuators and pre-amplifier potentiometers is horrible for noise performance. But sometimes there’s just not much we can do about that. If somebody upgraded my Fender Twin tube guitar amp, retrofitted with JBL E120 speakers, to digital potentiometers instead of knobs, I’d tar and feather them. Those knobs, and quite possibly the noise they make, are part of the bravado and character of that amplifier. I wouldn’t play Beethoven’s fifth through it from a pure digital source with outboard DACs, but it does wail nicely.
For high-power applications, if you must consider a wirewound resistor, choose the Dale wirewound series in the aluminum blocks, with part numbers that end in “-NH” or someone else’s equivalent. These devices are “counterwound,” which I’ve always interpreted as perhaps winding one layer clockwise (CW), then winding the next layer counterclockwise (CCW), keeping an even number of layers.
Winding the resistors in this fashion minimizes inductance. The flux from the CW solenoid nearly cancels the flux from the concentric CCW solenoid. If you absolutely need minimal inductance, consider the metal-oxide slab resistors akin to those found in dc-to-daylight RF dummy loads. These components are available as snubber resistors in very low resistance values from Ohmite and others.
Capacitors are an important topic in the audio world, though not insanely complex. Consider a simple example. Say you have a component with a dielectric material with a large voltage coefficient of capacitance, perhaps a multi-layer ceramic capacitor (MLCC). Should you use it in a pre-amplifier or sensitive audio circuit? No! That voltage coefficient of capacitance means that as the voltage across the two-terminal device changes, the capacitance changes too. We can see this from the fundamental capacitor equation:
I = C(dV/dt) + V(dC/dt)
The V(dC/dt) part looks new, but it’s simply the chain rule of differentiation. In most applications it is zero and forgotten. For a small signal coming into this capacitor, the minor voltage fluctuations will cause minor capacitance fluctuations. These fluctuations produce unwanted distortion currents that get exponentially worse with higher throughput signal levels.
That’s easy enough to see. But we can’t help but wonder: For a few microvolts from an electric guitar or condenser microphone, can we really hear that distortion current? The answer is yes and the levels are nothing short of amazing. It turns out that we are hearing differences that are 90 to 130 dB down.
Piezoelectric effects also can add distortion. If the MLCC housing structure deflects at all, there is distortion. For fun, try this: Replace a polypropylene cap on the input stage of a good pre-amplifier with a large, high-voltage MLCC disc capacitor of the same value. Then coil up your index finger and ping it in situ with a little gain on the output.
Don’t do this with speakers that you like. You’ll be surprised. The voice coil jumps out pretty hard and makes a loud popping sound. This is purely the piezoelectric effect under “macrophonic” excitation. (Normally, these components are much less aggravated by microphonic excitation from the transducer output making them wiggle slightly.)
As to which capacitors are best, this gets subjective once again. However, I strongly recommend that anyone looking for “the truth” on audio capacitors—whether in the signal path, crossover, or bypassing stages—take a serious look at Cyril Bateman’s “Capacitor Sound.” This first appeared in Electronics World starting in July of 2002 (www.scribd.com/doc/2610442/Capacitor-Sound). A more comprehensive listing with better images can be found at www.proaudiodesignforum.com/forum/php/viewtopic.php?f=6&t=153&start=2.
Seeking the truth on capacitors, Bateman rolled up his sleeves and developed oscillators and analyzers to find it. He goes through polypropylene caps, electrolytics, and many others for coupling as well as decoupling/bypassing. This is the best work I’ve seen to date on the subject, and he remains agnostic throughout the work. He also gets into the details and intricacies of building the analyzer to measure capacitor distortion products.
Some of his later works were on a standalone distortion meter, distortion versus bias in capacitors, and real-time analysis methods. They all appeared in Electronics World between roughly 2002 and 2003. (I tried finding a contact for Bateman and I could not find it. If anyone has this, please post it in the comments below.)
Any inductor in the audio path, whether the B+ filter choke that might be integrated into the magnetic structure of an old juke box loudspeaker, feeding the plates of some 6550s or 6L6s, or the inductors in the crossovers, must be carefully considered at design. When we saturate the core material, we are pounding the structure into and out of saturation at twice the frequency of the program material applied. This can give rise to offensive distortion products that will increase sharply with increasing amplitude of the driven signal. This sharp rise results from the fundamental equation that describes the inductor:
V = –L(di/dt) + I(dL/dt).
In most applications, dL/dt is zero, so we simply ignore it. In the audio arena, however, when the inductor saturates, the inductance drops. Thus, dL/dt creates a voltage distortion product. Some take this to mean, “only use air-core inductors,” even where that’s not practical. Others do a wonderful job designing structures on laminations, amorphous tapes, and powdered iron materials and simply stay away from saturation and/or use materials that saturate gradually. Others just use ferrites and stay well away from BSat.
Audio output transformers are tough to design. For an output transformer, we need a device that does not saturate at 20 Hz and has low enough leakage inductance to allow 20-kHz signals with absolutely minimal attenuation. Magnetically, this means that we need a huge magnetizing inductance for the 20-Hz end of the spectrum and really low leakage inductance for the 20-kHz end. We also need a core and materials that exhibit very low eddy-current losses and skin-effect losses at the frequencies at hand.
The windings must be arranged such that the throughput signal response is as flat as possible between the two band edges. This isn’t easy. The windings also have to have minimal skin-effect and proximity-effect losses, as well as minimal capacitance from turn to turn and bank to bank. The design further has to provide acceptable insulation between primary and secondary in tube circuits.
There are many schools of thought, books, and resources on the subject. The most notable one that comes to mind is by Meeno Van Der Veen (www.next-tube.com/articles/Veen2/Veen2EN.pdf). However, Jenson, McIntosh, and many others have done great work. Some go with toroidal structures due to their near-zero leakage flux. Some go with EI lamination stacks and painstakingly wound bobbins. Some wind both the bobbin and the core.
Consider a first-order shot at a practical example. If we have a 200-W (sine) output stage with devices that can swing between roughly 0 and 500 V in a push-pull configuration, we have ±500 V of excitation at the primary. We need to get this signal to an 8-Ω speaker. The output impedance then looks close to
(500 V/1.414)2/200 W
or about 625 Ω (assuming the impedance looking into the plates is equivalent to that of the transformer). We need to match this to 8 Ω. So, the turns ratio must be about 9:1.
If the magnetizing and leakage inductance are to add less than 1% to this impedance, the leakage inductance impedance contribution must be 625 Ω/0.01 or 62.5 k Ω at 20 Hz. This works out to roughly 500 H. The leakage inductance needs to be 1% of 625 Ω or 6.25 Ω at 20 kHz. This works out to 50 µH. So we have a structure than needs an Lmag of 500 H and an Llk of 50 µH. That makes Lmag/Llk = 1x107. That’s a chore!
In the switch-mode power supply (SMPS) world, we get happy when Lmag/Llk approaches 102 or 103. The people that design those 765-kV transformers get giddy when Lmag/Llk = 104 at 60 Hz. The purists’ transformers are 60 dB on that at three decades of bandwidth!
And we haven’t even discussed the capacitive interactions between those two points and how to minimize that to keep the overall transfer function flat within fractions of a dB. The output transformers used in guitar and bass amplifiers are sometimes run in saturation on low notes or passages. Again, this would be for an on-stage application.
Audio transducers range from arrays of electrostatic panels to magnetostatic panels, to electrostatic speakers that modulate a flame, to magnetic structures with voice coils, ribbons, and diaphragms. They also represent the wildest, most varied section in this discussion. Transducers are one of the many subjects that can provoke the “you’re an idiot” response, with the closed-minded originators tossing a few expletives around and sticking their fingers in their ears and screaming “la, la, la. I can’t hear you.” But it’s not that bad.
Asking for an audio transducer is like asking a machinist for a bolt. What kind? What are you doing with it? What are the limits? Where does it need to fit? What obstructions are there? Each of these questions has driven incredible innovations in transducer (and bolt) design. To go through all of the transducers would require a book. But the salient points of any driver selection are the desired frequency response, required enclosure volume, excursion, maximum sound pressure, compression point, distortion, bandwidth, and efficiency.
The secret sauce of most high-end drivers lies in the second-order attributes, like the damping discussion below. Live audio monitors will tend toward high sound pressure level (SPL) designs, compromising a little directivity, and perhaps distortion to bring a full program to every seat in the house. Instrument amplifiers are designed much the same. Studio monitors will tend to focus on the brutal truth and keep the playback as accurate as possible. These drivers will compromise efficiency for accuracy. Ear-bleeding SPL isn’t required in a playback environment—at least theirs. In mine, I can go really loud!
The issue of damping is another tough discussion. Damping is a second-order interaction between the amplifier and the loudspeaker. At low frequencies, critical damping sounds and feels tight. Underdamping can sound great at live or loud venues, although if we have a very underdamped response in a high-Q transducer, it just ends up sounding like a tuned organ pipe that only knows one note. A slightly underdamped reproduction of a snare drum hit can be felt in your chest. Overdamped reproduction sounds muddy at low frequencies.
At higher frequencies, we hear artifacts of damping. This gets interesting fast. In a midrange or midbass driver, assuming the structure has a stiff suspension and significant dc flux in the gap, the voice coil is of substantial size to apply enough flux to the gap to make the cone and suspension system move fairly efficiently, and the driver has a fairly low total Q, there will be bell modes on the cone. These bell modes are simple harmonics. Some transducer designers have boosted this effect by adding aluminum dustcaps to allow higher-frequency material a path of less resistance to the ether.
A solid-state amplifier with lots of feedback and an ultra-low resultant output impedance will not allow these bell modes to originate. It will damp them, usually with a very high damping ratio. Some advertise numbers on the order of 104. Yet I can’t figure out how they actually get there when the speaker cable has a dc resistance on the order of 20 mΩ (round trip for 10AWG, 10-ft tandem at room temp or 20-ft total path length, not taking into account inductance or capacitance). This would then make the damping factor into an 8-? load around 400 assuming the output impedance of the amplifier was zero.
From a practical standpoint, a damping ratio higher than six or seven can’t really be discerned at any audible frequency in A/B (user comparison) testing, so we need not get terribly worried about the claimed exponent. On the other hand, the higher output impedance of the basic tube amplifier, whether push-pull or single-ended, allows these bell modes to radiate.
Nelson Pass did a lot of research on this subject, but he’s certainly not the only one. John Eargle alluded to this all the time, as did James Bullough Lansing and even Cyril Bateman. The main idea of the research that was done on this subject was that high-end midrange drivers need to “speak for themselves.” That is, damping the bell modes is a bad idea. It results in a bad sound compared to not damping these modes.
This can be verified quite simply. Take your favorite high-SPL, low-Q midrange driver in a decent, properly designed enclosure. Play some material into the driver from a solid-state amplifier with ultra-low output impedance and a high damping ratio. Then put a fair amount of resistance—equal to or greater than the transducer impedance—in series with the driver. This makes the source impedance look much higher.
Play the same program again. (Recompensate your phon/Fletcher-Munson curves if you are a loudness fanatic.) You will hear a much warmer, sweeter reproduction. This comes directly from the bell modes of the cone and or dustcap of the driver being allowed to resonate as opposed to being damped.
If you try this with a lesser quality driver, you will notice the opposite effect. The lower damping factor sounds worse. This is strictly related to the transducer design and the transfer function from voice coil flux in the magnetic gap to acoustic output in the ether.
There are also higher-frequency resonant modes in horn/driver configurations. In most cases, these modes must be damped or the reproduction will “honk” somewhere in the midrange or high end. In other words, the resonance colors the transducer output to add a ring or a honk to notes and passages that were otherwise inert.
The discussion to this point shows how complex the issue of tubes versus transistors is even before getting into the active devices themselves. Part 2 will delve into the differences between the active devices and the amplifier topologies and make clear why the material on the passive devices is necessary. The comparison is not straightforward at all, and when we forced it to be a straightforward comparison the results were surprising.