The TCJ Unanticipated Amp in Blog 546 17 October 2021 has some pretty interesting ideas such as 1) the singleton input stage that John calls a Bastode (aka Rush pair) instead of a differential input stage that drives a push-pull VAS, and 2) a Class-AB+C output stage (aka DoubleCross).
My simulation, as close as possible to the original, is attached.
To get it stable I used shunt compensation (C8, R21) and minimised the feedback resistors shunt capacitor (was 300pF now 56pF). This gives roll-off at 1.5MHz. Clip recovery also looks good with no spike in the d(V(out)) plot (LHS corner). THD is low at 0.047% at full output (+/-24Vpk).
An interesting feature of the distortion profile is dominant 2nd harmonic which according to John gives "single-ended-quality sound". This feature appears to be due to the Bastode (Rush) input stage which does not cancel even harmonics like the usual LTP input stage.
The Class-AB+C (DoubleCross) output stage can be seen in action in the plot above. The idle current is quite high at nearly 200mA in Q9 and Q10 and covers the first watt in Class-A. Above 400mA the Class-C transistors Q11 and Q12 conduct and they blend in smoothly with no Gm-doubling effect above 1 watt. Nice.
One down side I noticed is a relatively low slew-rate of 3.3V/us (with compensation that I used). This only allows full output swing to 20kHz before it slew limits. I would prefer more like 100kHz so I tried some things to increase the slew rate. Mods are in following threads. I also tried adding autobias and Class-G.
My simulation, as close as possible to the original, is attached.
To get it stable I used shunt compensation (C8, R21) and minimised the feedback resistors shunt capacitor (was 300pF now 56pF). This gives roll-off at 1.5MHz. Clip recovery also looks good with no spike in the d(V(out)) plot (LHS corner). THD is low at 0.047% at full output (+/-24Vpk).
An interesting feature of the distortion profile is dominant 2nd harmonic which according to John gives "single-ended-quality sound". This feature appears to be due to the Bastode (Rush) input stage which does not cancel even harmonics like the usual LTP input stage.
The Class-AB+C (DoubleCross) output stage can be seen in action in the plot above. The idle current is quite high at nearly 200mA in Q9 and Q10 and covers the first watt in Class-A. Above 400mA the Class-C transistors Q11 and Q12 conduct and they blend in smoothly with no Gm-doubling effect above 1 watt. Nice.
One down side I noticed is a relatively low slew-rate of 3.3V/us (with compensation that I used). This only allows full output swing to 20kHz before it slew limits. I would prefer more like 100kHz so I tried some things to increase the slew rate. Mods are in following threads. I also tried adding autobias and Class-G.
The following mods includes Class-G and autobias with Schottky diodes.
First, the input stage uses a jFET so the servo can be moved to the inverting input and the input capacitor can be omitted (if desired). Since the servo is feeding the inverting input it is modified to non-inverting using a circuit shown in Bob Cordell's Power Amplifier book (2nd Ed).
Next, the VAS has a current sources added to the top side with cascoding. The bottom VAS transistor is common base so it is effectively cascoded already. The spreader transistors are now part of the autobias loop using Schottky diodes rather than emitter resistors. To get the Class-C transistors to start conducting at the desired crossover point I have used IRF610/9610 MOSFETs to lift the base voltage of Q11 and Q12. he cut-in point is adjusted using the source resistors relative to the base resistors (the base resistors need to be kept below 10 ohms).
The Class-G uses Schottky diodes D5,D6 and effectively a cascode of the inner MOSFETs M1 and M2. The outer MOSFETs M3 and M4 are bootstrapped from the output voltage with 8.2v zeners for 10V inner supply rails. Q17 and Q18 are for rapid charge removal of the outer MOSFETs. R28 and R35 across the Schottky's is for convergence in simulations; it provides a path for current into node 's3' back to the 10V rail and therefore to ground. The various MOSFET currents are shown below with a sinewave input near clip.
Six currents are plotted:
+10V rail is V11, the +37V rail is V2,
-10V rail is V12, the -37V rail is V3,
Class-C +ve side is Q11, the -ve side is Q12.
In this circuit changeover to Class-C occurs at the same time as the Class-G.
EG Class-G changes at 0.54ms is where I(V12) stops rising and fall to zero with I(V3) taking over.
Class-G is useful in this circuit to reduce the idle heat dissipation by a factor of 4, from about 12W to 3W for effectively an 80W amp with a 4 ohm load. But Class-G does not reduce the worst case heat dissipation with a sinewave so the heatsink size remains the same. But If you rate the heatsink for music only then it can be reduced by about 30%.i
An electrothermal version is attached giving the plot below
This shows the idle current at turn on rises to 242mA with no signal, then rises to 460mA at full output power. This is despite autobias and is due to the negative temperature coefficient of the Schottky diodes and the inadequate temperature coefficient of the spreader transistors (which are mounted on the power transistors).
A better outcome is obtained using MOS-diodes in the autobias loop instead of Schottky diodes and is covered in another thread.
The slew rate of this circuit is 4V/us with 3mA through the input stage to keep the jFET dissipation below 100mW with 37V rails. The SR can be doubled by doubling the input stage current to 6mA (as per the original) but the jFET then needs a cascode, then giving full output up to 50kHz.
First, the input stage uses a jFET so the servo can be moved to the inverting input and the input capacitor can be omitted (if desired). Since the servo is feeding the inverting input it is modified to non-inverting using a circuit shown in Bob Cordell's Power Amplifier book (2nd Ed).
Next, the VAS has a current sources added to the top side with cascoding. The bottom VAS transistor is common base so it is effectively cascoded already. The spreader transistors are now part of the autobias loop using Schottky diodes rather than emitter resistors. To get the Class-C transistors to start conducting at the desired crossover point I have used IRF610/9610 MOSFETs to lift the base voltage of Q11 and Q12. he cut-in point is adjusted using the source resistors relative to the base resistors (the base resistors need to be kept below 10 ohms).
The Class-G uses Schottky diodes D5,D6 and effectively a cascode of the inner MOSFETs M1 and M2. The outer MOSFETs M3 and M4 are bootstrapped from the output voltage with 8.2v zeners for 10V inner supply rails. Q17 and Q18 are for rapid charge removal of the outer MOSFETs. R28 and R35 across the Schottky's is for convergence in simulations; it provides a path for current into node 's3' back to the 10V rail and therefore to ground. The various MOSFET currents are shown below with a sinewave input near clip.
Six currents are plotted:
+10V rail is V11, the +37V rail is V2,
-10V rail is V12, the -37V rail is V3,
Class-C +ve side is Q11, the -ve side is Q12.
In this circuit changeover to Class-C occurs at the same time as the Class-G.
EG Class-G changes at 0.54ms is where I(V12) stops rising and fall to zero with I(V3) taking over.
Class-G is useful in this circuit to reduce the idle heat dissipation by a factor of 4, from about 12W to 3W for effectively an 80W amp with a 4 ohm load. But Class-G does not reduce the worst case heat dissipation with a sinewave so the heatsink size remains the same. But If you rate the heatsink for music only then it can be reduced by about 30%.i
An electrothermal version is attached giving the plot below
This shows the idle current at turn on rises to 242mA with no signal, then rises to 460mA at full output power. This is despite autobias and is due to the negative temperature coefficient of the Schottky diodes and the inadequate temperature coefficient of the spreader transistors (which are mounted on the power transistors).
A better outcome is obtained using MOS-diodes in the autobias loop instead of Schottky diodes and is covered in another thread.
The slew rate of this circuit is 4V/us with 3mA through the input stage to keep the jFET dissipation below 100mW with 37V rails. The SR can be doubled by doubling the input stage current to 6mA (as per the original) but the jFET then needs a cascode, then giving full output up to 50kHz.
The following mods includes Class-G and autobias with MOS-diodes instead of Schottky diodes.
The slew rate is unchanged. The bias stability is much better. Starting at 382.6mA and 382.1mA warmed up then 381.9mA at full power. Nice.
THD with autobias is not much affected by choice of bias and Class-C and Class-G crossover values and is around 0.05% at 1kHz into 4 ohms. At 1W 1kHz 4 ohms it is 0.007%.
The slew rate is unchanged. The bias stability is much better. Starting at 382.6mA and 382.1mA warmed up then 381.9mA at full power. Nice.
THD with autobias is not much affected by choice of bias and Class-C and Class-G crossover values and is around 0.05% at 1kHz into 4 ohms. At 1W 1kHz 4 ohms it is 0.007%.
Hi All,
Another mod for the TCJ Unanticipated amp, as the previous post with Class-G but now with McPherson Class-G.
MOSFETs M11,M12 and 10 ohm 20W power resistors R22,R23 are added. When Class-G MOSFET's M3,M4 conduct current flows through the 10 ohm resistors diverting about half the heat in M3,M4 into the resistors. Then at high currents the resistors are bypassed by M11,M12 to get the full output swing (less about 1V across M11,M12).
Mr McPherson used this resistor diverter method to reduce the heatsink size for a series-pass linear power supply back in the 70's or so. It was mentioned by Ian Hickman in an Electronics World article in the 90's. I successfully used it recently in a Capacitance Multiplier for a power amp. In this amp the McPherson resistors with Class-G reduce the heatsink size by 30% at worst-case dissipation. (In a practical circuit a gate diode may be needed to stop large reverse voltages appearing on M11, M12 gate).
Plots below show the currents and dissipation in the various parts with the McPherson resistors:
and the plots with Class-G and no McPherson resistors:
Files are attached including the electrothermal versions and the original circuit.
You can get the average dissipations of each plot using Ctrl-RtClk.
Then you add all power transistors dissipations to get the total heatsink dissipation.
BTW One interesting unique feature of the TJC Unanticipated amp is the input stage gives dominant H2 (but not sure which phase and if it can be made to give either phase). Most symmetry type input stages give H2 cancellation but this amp input stage does not.
Another mod for the TCJ Unanticipated amp, as the previous post with Class-G but now with McPherson Class-G.
MOSFETs M11,M12 and 10 ohm 20W power resistors R22,R23 are added. When Class-G MOSFET's M3,M4 conduct current flows through the 10 ohm resistors diverting about half the heat in M3,M4 into the resistors. Then at high currents the resistors are bypassed by M11,M12 to get the full output swing (less about 1V across M11,M12).
Mr McPherson used this resistor diverter method to reduce the heatsink size for a series-pass linear power supply back in the 70's or so. It was mentioned by Ian Hickman in an Electronics World article in the 90's. I successfully used it recently in a Capacitance Multiplier for a power amp. In this amp the McPherson resistors with Class-G reduce the heatsink size by 30% at worst-case dissipation. (In a practical circuit a gate diode may be needed to stop large reverse voltages appearing on M11, M12 gate).
Plots below show the currents and dissipation in the various parts with the McPherson resistors:
and the plots with Class-G and no McPherson resistors:
Files are attached including the electrothermal versions and the original circuit.
You can get the average dissipations of each plot using Ctrl-RtClk.
Then you add all power transistors dissipations to get the total heatsink dissipation.
BTW One interesting unique feature of the TJC Unanticipated amp is the input stage gives dominant H2 (but not sure which phase and if it can be made to give either phase). Most symmetry type input stages give H2 cancellation but this amp input stage does not.