Field-Effect Transistors

There are 3 main kinds of transistors - Bipolar, Junction Field-Effect (JFETs), and Metal-Oxide semiconductor FETs (MOSFETs). Both come in N and P varieties with opposite polarities. This page is about JFETs and compares them to bipolar transistors to illustrate why JFETs work better for simple preamp stages. MOSFETs are not discussed here.

There's a lot of talk about why tubes sound better than transistors, often attributed to how tubes clip gently and transistors clip with hard edges. Roughly accurate but there's more to it than that - transistors don't necessarily clip hard and tubes can clip with hard edges when pushed. Rather, the main difference is a tube (and a JFET) is a voltage-driven device that's fairly linear when biased properly, whereas a transistor is a current-driven device that tends to be more nonlinear and usually requires negative feedback to correct, that's where the hard clipping edges come from, especially in the extreme case of opamps.

Here's a simple bipolar transistor preamp stage with varying amounts of drive...




Before clipping it looks ok but the distorted waveform is.. well yuk. The top of the wave rounds and the bottom squares, and when distorted the base rectifies the input signal causing the duty cycle to shift, basically losing power. Some duty cycle shift is OK but with this circuit it only gets worse the harder it's pushed and there's no way around it because the base is biased positive to the emitter, any distortion that causes the emitter to stop following the input signal turns the base into a diode. The resistor values are fairly high in this example but reducing all the resistors by a factor of 10 (and increasing the caps by 10) makes little difference other than increasing the current drain.

Here's an almost identical circuit using a JFET instead of a transistor...




Much better! There's still a hint of rounding on the top of the wave but it's partially because it's not biased perfectly centered. Compared to the bipolar transistor, the FET stays cleaner until it reaches clip then it flattens more or less the same on both sides, and there is no duty cycle shift. Eventually the gate will conduct and shift the duty cycle but it takes over a volt of input as the gate is negative to the source, the signal has to exceed the bias point plus the juction drop before rectification can occur. Tubes do a similar thing (to a greater extent as the signal voltage vs rectification voltage is greater) but the solution is the same - add some resistance between the output of a stage and the gate of the next stage (or in series with the gate) so if the gate conducts at least it won't load the previous stage and cause further issues. But keep in mind that resistors are noise makers by the laws of the universe, use only as many ohms as needed.

The 2N4393 shown in the simulation is available but the common J113 has practically identical specs (Vgs threshold from -0.5V to -3.0V) and works about the same in this circuit. To minimize unit-unit variation I made the source resistor the same as the drain resistor, typically the source biases about 1.3V above the gate with a typical variation of +/- 0.4V (but not guaranteed!), because the source and drain resistors are the same value this limits the drain bias shift to the same amount. There will be some unit-unit variation but it's less than in typical JFET example circuits that omit R3 and use a smaller resistor for R4, those circuits usually require a trimmer for R4. The gain is determined roughly by R6 divided by the parallel combination of R4 and R5 (minus a bit) so if R4 is a trimmer it's going to also change the gain (and also the output level). Also note that dialing up more gain by reducing R5 increases the clean distortion and causes more pre-clip rounding on the top side of the wave. Biasing the gate positive with a larger source resistor limits the output swing but it's worth it to reduce variability, and if trimming is needed it can be done using R3 or R4 with less gain variation. It also simplifies the design process - whatever the supply voltage is, divide it by 3 and bias the gate to that voltage minus the FET's typical operating gate voltage, which for the J113 is about 1.3V. So 1.6V on the gate results in 2.9V on the source, 2.9V across the drain resistor and 3.1V across the FET, in the ballpark. It's a bit more complicated than that, the source follows the signal (unless R5 is 0) so the bottom clip point gets pushed up by the signal on the source - so having a bit more voltage across the FET is a good thing - but there's also loading effects which tends to need less voltage across the FET to balance out the clipping.

Now to analyze the distortion characteristics using FFT plots. Before distortion there isn't much difference...



For both the transistor and the FET the 2nd harmonic is about 32db down, or roughly 2% distortion, the other harmonics fall off about the same. Running clean both these circuits will sound about the same - transistors can be "warm" sounding, at least much better than chips.

Running the circuits into clipping better shows the difference...

 


Besides the FET waveform looking better, the higher harmonics drop off more rapidly. The transistor's spectrum looks harsh and this is fairly mild clipping.

Here are the LTspice (version 4) files for the above simulations...
transistor.asc.txt - the basic bipolar circuit
transistor1.asc.txt - modified for FFT analysis
tfet.asc.txt - the basic JFET circuit
tfet1.asc.txt - modified for FFT
(remove the .txt extension or copy to an .asc file)

LTspice is available from Linear Technology's software page, I use version 4 (under wine) but a newer version is available. Simulation isn't perfect but for what it does it's almost like magic. Just keep in mind that it doesn't model parasitic effects from the circuit layout and the components are assumed to be perfect unless otherwise specified. Usually if a circuit works in LTspice it'll work in the real world although sometimes it might need a few tweaks to compensate for effects that are not modeled. For audio usually the simulation is spot-on.

A 3-way Tone Network using a JFET

Disclaimer - I haven't actually built this circuit... it "should" work pretty much like the simulation but the component values in the tone network will likely need "by ear" adjustment depending on the application.

This circuit implements a 3-band cut/boost-type tone control network using a JFET for the gain stage. With the controls centered the gain is roughly unity (minus about a db). The bass control (on top) has a range of about +/-10db at 80hz (more at lower frequencies), the mid control has a range of about +/-7db at 1khz, and the treble control has a range of about +/-10db at 7khz (more at higher frequencies). This type of tone network probably wouldn't make for good guitar amp tones by themselves (too flat, guitar amps typically scoop the mids), the application I had in mind is putting it after an overdrive circuit that's already equalized to sound about right to allow better control over the overdrive tone.

The circuit has a fairly low input impedance (about 15K) that varies with frequency so it should be driven from a low impedance source. For high impedance sources use a buffer such as this circuit...

              .----------*------< 9V
| |
2.2M |
| |---'
in >--0.01u---*----->| J113
| |---.
| |
| *---1u---> to EQ
1.8M |
| 10K
_|_ _|_

Because the circuit cuts and boosts, keep the input level fairly low (around 100mV RMS) to avoid clipping at extreme settings. The circuit has a fairly low output impedance (about 5K) but if driving an input with an impedance of less than 100K increase the size of C7 to avoid bass loss.

Here's the circuit response with the tones flat and driven slightly into clipping...


Maximum undistorted output is roughly 1.7V RMS, FET variation might reduce this somewhat if not biased perfectly centered. The bottom bias resistor is slightly bigger than the previous examples because the higher current means less gate-source voltage difference and also because the source resistor is fully bypassed so it won't be going up and down with the signal. Note how the gate signal is distorted as the feedback corrects the distortion of this configuration. Controls U1 to U3 are 300K linear-taper controls, 250K linear controls should also work OK. In the simulation the controls ended up being backwards, wiper=0 for all the way up and 1 for all the way down, the correct orientation is for the wiper to travel to the left (towards the input) as the control is increased.

C4 sets the bass control frequency, C6 sets the mid control lower frequency, C5 sets the mid control high frequency, and C2 and C3 control the treble control frequency (could just be a single capacitor in series with R10 but this arrangement permits setting the cut and boost frequencies separately). R5 and R6 set the bass control range, R8 and R7 set the mid control range, R10 sets the treble control range. R9 is fairly large compared to the other resistors to minimize interaction.

Here are AC response plots for various control positions. Flat and with the bass control 90% up (sim wiper setting at 0.1)...

Bass at 10%, mid at 90%...

Mid at 10%, treble at 90%...

Treble at 10%, the last plot is with the bass at 90%, the mid at 10% and the treble at 80%...


Here's the tonecb_ac.asc.txt simulation file and the potentiometer model it needs to run.

An Old-Style Guitar Preamp using JFETs

This is similar to the preamp used in ShoBud amps with more of a rock-n-roll twist - I used a 50K mid control (with an extra capacitor) instead of the usual 10K, and added a master volume between the 2nd stage FET and the output buffer. The original ShoBud preamp had no master and the base of the output buffer transistor (a regular bipolar NPN) connected directly to the drain of the 2nd stage, an arrangement that wasn't designed to be clipped. Although provisions are made to let it get a bit dirty, it's not an overdrive preamp and is more like the clean channel of a typical tube amp. The output of the circuit can be run into any typical power amplifier that needs a volt or less for full output. Or run through a reverb circuit or an effects loop first. This is another circuit I've only "built" in simulation in this exact form but I've used variations of this many times over the last decades and it sounds very much like a tube amp.

Here's the circuit running clean and clipped...



The circuit runs at 18V, the simulation schematic shows 50V because that's what's typically available from the power amp section, reduce R20 if running from a lower voltage to ensure that the zener diode conducts and properly regulates the supply. The simulation shows 2N4393 but almost any N-channel JFET with a reasonably low Vgs threshold should work - J113, MPF102, 2N5457 etc. If clipping is off-centered tweak the values of R5 and R12 (with a J113 model I needed 75K for symmetrical clipping). R6 and R13 set the gains of the first and second stages, reduce for more gain. I configured the first stage has less gain to avoid overloading the input stage.

The tones resemble a normal guitar amp tone stack but I used somewhat different capacitor values, made the mid control bigger to make it more useful for boosting the gain. The extra capacitor C13 keeps it from boosting the highs along with the mids. C6 and R9 add a high treble boost at lower volume settings, R9 controls how much brightness is added. R9 is often a switch for full bright. The simulation shows some rolloff above 8khz, not sure what's up with that since there are no added high-cut caps (probably the "miller" capacitance of the 2nd FET). Guitar speakers cut off around 7khz so it doesn't matter much but this might be one of those areas where simulation deviates from reality.

Here's the AC response with the controls centered (20% log controls)...


The following plots show the response with the treble, mid and bass each at 10% and 90% with the others centered...







Something resembling flat is obtained with the treble at 50%, the mid all the way up and the bass at 20%...


Here's the fetgpre_ac.asc.txt LTspice simulation file, requires the potentiometer_standard model.

Dealing with unit-to-unit variation

JFETs have very loose specs when it comes to gate threshold, for a 1uA threshold the J113 has a specified Vgs from -0.5V to -3.0V. Other units are no better, it's just something the designer has to deal with when using JFETs. Biasing the gate positive and using a big source resistor helps (in that the circuit will usually work) but it does not eliminate the variability, especially when running from 9V as with a guitar pedal.

Here's a J113 model that can be used with LTspice to explore the effect of varying gate thresholds...

.MODEL J113 NJF (VTO=-1.29 BETA=9.25964E-003
+ LAMBDA=3.03839E-002 RD=1.30170 RS=1.30170 IS=9.86870E-016
+ CGS=1.05000E-011 CGD=1.20000E-011 PB=5.04493E-001 FC=0.5)

To use this model select the text icon and select spice directive, paste the above text into the text box and place anywhere on the schematic, then the 2N4393 text can be changed to J113 (right-click and change the part label text itself, it won't show up in the part select dialog). The default VTO is -1.29 but this can be changed to explore the effects of variability.

Here's the simple preamp circuit with the default threshold...

Here's the waveforms with the VTO=-0.9 and with VTO=-1.7 (about +/- 0.4V)...


...yea that's a problem. Both will "work" but will sound different.

Here's the guitar preamp with the J113 model (source resistors changed from 82K to 75K)...


Here's the response with VTO=-0.9V and VTO=-1.7V...



...not as bad because it's running at twice the supply voltage, but still some variation.

So what to do? One thing that can be done is replace the source resistors with trimmers, especially the last stage where the effect is most noticeable. But that adds cost and there's the additional labor step of connecting it to a signal generator and scope to make the adjustment. Another way to compensate is to grade the FETs - Here's a test circuit that replicates the bias of a preamp circuit...

         G O-----------*---2.2M----------.
FET | |
under D O-----------|----*----47K-----*------O +
test | | | 9V battery
S O--47K--. 470K `-> - + <-' .--O -
| | to meter |
`---*---------------------'

Since the source and drain resistors are the same, for proper unloaded response the meter should read about 3V, but typically you'll need a bit more voltage to account for loading effects, 3.1V or so with a 150K load. Regardless go through a bunch of FETs to see what the average is then in your circuit size the top bias resistor (or the source resistor) to produce the desired clip characteristics. Mark all the FETs close to the average say yellow, these are the "good" ones. If the meter reading is less the FET draws less current, paint it green and to optimize it use a smaller resistor for the top bias resistor. If the meter reading is greater paint it red and use a bigger resistor to optimize. Or just use the yellow ones for the last stage and put the reds and greens earlier in the preamp where exact clipping symettry isn't an issue.


Last modified November 22, 2016
Terry Newton (wtn90125@yahoo.com)