Introduction
Electric guitarists have been long enamored with distortions made from discrete transistor stages. This relationship isn't going away anytime soon, given the recent rise in popularity of this kind of circuit. I'm not talking about the many fuzzes which fall into this category, but of "discrete transistor overdrives".
Tracing an history from current pedals through the most memorable transistor pre-amps from Roland and Peavey, we can say that this similar arrangement have accompanied electric guitar from the beginnings, being a direct transposition from circuits using vacuum tubes. Whether the purpose was to have distortion or to have enough amplitude to drive what comes next, usually two or more amplification stages are present, with an interstage attenuator controlling signal gain and provisions for equalization.
I haven't mentioned so far JFETs, which are usually the preferred device for these circuits. Usually, a reason for this is that they are depletion devices like vacuum triodes, but as "The Art of Electronics" made me realize, this just means that the linear range of operation is shifted from 0V down, whereas it's up from 0 for enhancement devices. This is hardly important as long as biasing is done correctly, and enhancement mode actually makes biasing easier on single supply. There's no denying the importance of the Fetzer valve concept in motivating the first adaptations of historical guitar preamplifiers and successive derivations; I'm not contesting the conclusions reached, which follow naturally from JFET theory. I'm claiming that they missed the point in focusing on the linear transfer function of common source amplifiers, when saturation behavior is what matters most. It's easy to claim that for a guitarist, a circuit analyzed this way, that doesn't clip even a tiny bit, will seem useless, and the few % of THD, or more exactly the different distribution of the harmonic contribution to this quantity, will go almost fully unnoticed. Once clipping is reached, they do go fully unnoticed, because the waveform is affected (distorted) on a much larger scale; we then are concerned with things like symmetry of the clipping, the range of the waveform that isn't affected (with the hard clipping threshold approximation), the shape of the clipping and other behaviors like rectification and blocking distortion.
By starting with the trivial assumption that the goal is some amount of clipping, from subtle to blatant, we can reframe the objective of matching the input sensitivity of one stage with the amplitude it receives. The choice between JFET and BJT is then mostly between arbitrarily high input impedance and more transconductance. I'm leaving noise outside of the comparison because, depending on the bias current, the source impedance and the specific part, either category could be the winner. If source impedance could be ignored, then some BJTs would be the winners, but it's not the case for guitar effects. I'm also not considering MOSFETs as a serious option, because of their irredeemably high noise and because, despite being "FET", their transconductance makes them perhaps closer to BJTs.
Since JFETs have been thoroughly explored, here I want to explore the capabilities of BJTs in this kind of circuit, recognizing their advantages and their wider availability, and shed all preconceptions about them possibly being less suited, or less similar to vacuum tubes.
Conventional circuits, be they JFET or tube, make use of multiple stages to reach the desired distortion, due to the voltage gain limitations, or more generally of the ratio of gain to headroom. This has indirect advantages, in making the clipping function composite, and any filter networks between amplifier stages pre-emphasis networks. These are probably some of the most quantifiable differences of a "discrete" circuit over just using a single op amp to achieve the desired distortion, together with the important effect of blocking distortion, which is uniquely linked to the interaction of large coupling or bypass caps with the bias of amplifiers, especially if those feature some DC gain; while often undesired, blocking distortion is probably responsible for short-term dynamics that could go under the terms "tube", "sponge", "loose". In larger amounts, blocking distortion is a fundamental component of fuzz distortion.
Given the generous transconductance and hence voltage gain of BJTs, it might seem challenging to translate conventional designs to them, but if we go by our assumptions and follow the criteria previously set, degeneration or feedback linearising the stage while reducing gain are fine, as long as the desired headroom/sensitivity/amount of clipping is achieved. It should also be remembered that many JFET circuits employ mu-amp topologies (with another JFET as active load) to achieve higher voltage gain from one stage anyway, regardless of the effects this has on linear operation harmonic distortion.
Before setting off, a couple of differences between BJT and FET have to be acknowledged:
While a JFET common source can only pull the output as hard as a resistor of value Rds(on), BJTs can potentially pull down much harder than this in saturation, given enough base current; the positive output half-wave is only pulled up by a resistor in both cases, but the asymmetry in how hard they clip is potentially higher.
Unless Vgs goes positive, a JFET gate isn't much of a load on the previous stage, and one could argue that when the gate diode clamps the source, the FET is already well into saturation so it doesn't really matter. The Vbe diode of a BJT is a load on the source at all times, even if the current exponentially decreases away from saturation. Again, this is more of a curiosity than a concern, since base current stays small before saturation, when the output can't change more.
The circuit(s)
In the spirit of giving options in redeeming BJTs, I'm showing multiple circuits. I have my favorite, which I'll show first, but that doesn't mean the alternatives are worse or less interesting necessarily. More often than not, a circuit goes through many changes before being posted here, sometimes changing radically, but this is even more the case, since the goal was something as vaguely defined as "an overdrive-type distortion that sounds good to me". The idea started even before what's presented here, with experimentations around a split-pot gain control aimed at reducing noise, instead of the usual attenuator.
Bijou #1
You'll notice that I'm not specifying transistor parts. Normally I'd write "2N3904" for a generic part, or "2N5088" for something with higher beta, but there's nothing special about those parts, and that's the whole point! I am specifying minimum beta, just because it matters a bit for the input stage, but for the others use what you have!
Some parts of this circuit, which we can call "Bijou #1", will be repeated without changes in the others, so will only be explained once:
The input stage is one of my favorite common emitter topologies, featuring DC-only feedback through a T network; care must be taken with the capacitor values so there's no unexpected resonance at low frequencies. The input impedance should be over 700k at 1k, since there's no AC feedback to lower it. Gain is modest, around 13 dB when loaded, and the output is reasonably clean. I've tried to lower the bias current of this stage, because this is closer to optimal for the source impedance of a passive pickup for lowest noise.
Before the second stage, we have a somewhat classic pair of "Gain" and shelving "Bass cut" controls, with a couple of details to write about. The stopper R6 only prevents the output to be muted completely, but still allows the full range of distortion available. The arrangement of "Bass cut" after "Gain" minimizes the interaction on the cutoff by the effect the input impedance of the second stage would have, being put in parallel with "Gain" as it is adjusted; there's still a tiny bit of interaction, but the cutoff is mostly just set by Q2's input impedance, and is around 850 Hz; the control is wired as "cut" to allow the use of a Log pot, and a value of 500k should still provide enough range.
Q2 and Q3 are similar clipping stages developed in conjunction with AOTMR. Degeneration helps to lower the gain, together with AC feedback and consequent low input impedance; the same degeneration and feedback, operating down to DC, also help to give predictable bias without trimming. The idea is that the stage is purposefully biased hotter than normal so that the LED doesn't conduct at idle, but with signal, the LED applies limiting reminiscent of the Bazz Fuss, or a DC coupled Big Muff, before the transistor has a chance to cut off. Like in the Bazz, this clipping is applied only on one half-wave, but unlike in that circuit and more similar to the latter, the diode doesn't set the bias. The use of a LED helps to keep bias high enough, giving good output, and prevents the possibility of the diode pulling the output low (rectification) with a large input signal and low source impedance. This stage is repeated twice, the second time with less gain, to add more clipping in a more symmetrical way. An interstage resistor lowers the gain, instead of more degeneration, to prevent excessive rectification of the output peaks because of it. Whether it is because of the saturation behavior, or because of the low number of capacitors, preventing most blocking distortion effects, I really like the results of these stages, giving a smooth distortion.
The output is taken from a "tap" on the collector load for two reasons: the emitter follower is DC-coupled, so the base voltage should be increased a bit, and to simultaneously attenuate the output a bit, because clipping in the next filter stage is undesirable. Since I knew I had to buffer the output after low-passing it anyway, I've tried something different which I've used to great effect in that overdrive, which is an active two-pole Sallen-Key style low-pass. Two-pole output lowpass filters aren't that uncommon in overdrives, since it is desirable to cut some very high frequencies regardless of the treble setting, but are usually passive. Since the parts were all there, I had the chance to use an active filter instead, and one that is resonant at the lowest setting, hoping to keep some brightness when the cutoff is low and extend the useful range. The gain of the BJT follower limits the amount of peaking achievable, but this keeps it sounding natural. Using a single-gang pot to tune a second-order lowpass like this is called "half-tuning": unlike with a dual-gang, Q isn't constant but peaks at a certain setting, in this case at the lower end.
The power supply to the whole circuit is filtered with a RC, since the stages are susceptible and the current draw is very low. The "Volume" control is linear, because I've found this gives the useful range closer to the center, and having to set the volume higher than half makes people uncomfortable; this is because the off-center bias and LED of Q3 don't give the maximum possible output, and this is further reduced by the filter. R19 isolates the buffer from cable capacitance.
Demo
Bijou #2
The second circuit swaps the clipping stages for some based on a trick from the underrated Vulcan: the collectors are this time center-biased, and the emitter bypassed; unlike in the Vulcan, they are feedback biased and with smaller resistors. The idea is that, with a biased diode in the middle of the base network, the bases can be pulled down, but not higher than what the 470k resistor would by itself. This prevents the transistor from saturating, which gives the harder half-wave clipping. The interstage resistor attenuates the signal, allowing the two stages to operate in similar conditions for roughly symmetrical clipping. The result is close to that of #1, even if the 100k ends up lowering the maximum gain and could be adjusted, and the output is larger. I still somehow prefer #1 though, personally.
Demo
Bijou #3
This one is an earlier version, and mostly reported here as curiosity. It doesn't sound bad and some might like it, but the crackling on the tails isn't what I was going for. Still, it doesn't mean the circuit isn't interesting.
The idea was to have a main "amplification" stage in the second stage, and a dedicated low-gain "clipping" stage, with the idea that the third will always clip before the second. AC feedback was cut for the third stage, because that's another potential path for rectification otherwise. This didn't end up mattering, and I'm still not sure what causes larger or smaller amounts of crackle, other than maybe blocking distortion, even when limiting bandwidth. D4 might not do much, but the idea was again to limit saturation by not allowing the base to go too positive.
Demo
The quality of this demo isn't as good, but there's still one.
Conclusion
I think the original goal has been reached, in showing that BJT transistors are suitable devices for pre-amp based overdrive pedals, without necessarily falling into the "vintage" or "fuzzy" clichés. More compact circuits based on feedback pairs are already popular, such as those from Colorsound, the Hudson Broadcast (or its charming brother the Narrowcast), or my own Two-transistor, but circuits based on stacking multiple low-gain stages with inter-stage controls have mostly been the realm of FETs so far.
My thanks go to AOTMR for feedback and discussion, and to the Electronics Audio DIY discord server, which has taken part into my "blind listening" test while revising the circuit.