MOSFET power amplifier with error correction
When designing a linear power amplifier, we have a choice of bipolar or MOSFET power transistors at the position of output devices. Bipolar (BJT) transistors are used much more frequently, but please let me also discuss and show a design with MOSFET power transistors. I would start with a quote from excellent classic design book by Bob Cordell: Designing Audio Power Amplifiers:
-----------------------------------------------------------------------------------------------------------
“MOSFET class AB biasing tends to be simpler and less critical. MOSFETs do not have an optimum class AB bias current. Instead, they operate better with greater bias as long as thermal objectives are met. For this reason, typical MOSFET power amplifiers operate at higher bias current per output pair and have a larger class A region of operation for small signals. However, their lower transconductance tends to result in higher measured values of static crossover distortion. Because of their high speed, MOSFET amplifiers are less prone to dynamic crossover distortion as a result of switch-off characteristics. MOSFETs do not suffer from beta droop and fT droop at high currents and are generally able to handle high peak currents better than BJTs. The ease with which MOSFETs are driven also means that there is less stress on driver transistors when the amplifier is delivering high current to the load.
MOSFETs are a bit more prone to high-frequency parasitic oscillations as a result of their inherently higher speed. For this reason, circuit design and layout can require more care than for BJT designs.
Biasing power MOSFETs for class AB operation is different from that for BJTs in two ways. First, MOSFETs require greater forward bias at the gate than BJTs do at the base. Vertical MOSFETs can require up to 4 V. Some vertical MOSFETs like the 2SK1530 require only about 1.7 V, however. If MOSFETs requiring 4 V forward bias are combined with emitter follower drivers, the total bias spreading voltage will be on the order of 9.2 V. This compares to a typical bias spreading voltage of about 4.0 V for a BJT output Triple. This difference in required bias spreading voltage is not a problem for the traditional Vbe multiplier or minor variants of it, but it does imply that the driver circuitry will require more voltage headroom in a MOSFET design. This is sometimes dealt with through the use of boosted power supplies for the circuits preceding the output stage.
Crossover distortion is one of the most insidious distortions in class AB power amplifiers. It occurs at fairly low signal levels and often contains a high-order distortion spectrum that is more dissonant and difficult to remove with negative feedback. It is a result of the changing gain of the output stage as the signal current delivered to the load goes through zero (the crossover). MOSFET output stages are also subject to crossover distortion.
Static crossover distortion in BJT output stages is a result of the output impedance changing as the output current goes through zero. The output impedance forms a voltage divider with the load impedance; as a result the gain of the output stage changes. The lower the value of the output impedance, the smaller the crossover distortion will be for a given percentage change in the output impedance.
The same is true for power MOSFET output stages, but the output impedance is generally quite a bit higher for a given amount of bias current. This is because the transconductance of a MOSFET is much smaller than that of a BJT. The transconductance of a BJT at Ic = 100 mA is about 4 S. The transconductance of an IRFP240 biased at Id = 150 mA is about 1 S. As a result, the sum of the transconductances of the upper and lower MOSFETs dips in the crossover region. This is referred to as transconductance droop.
Conventional negative feedback is not the only way to reduce distortion. Various error-correction techniques can be used in place of, or in connection with, negative feedback.
In virtually any well-designed power amplifier the output stage ultimately limits performance. It is here where both high voltages and large current swings are present, necessitating larger, more rugged devices that tend to be slower and less linear over their required operating range. The performance-limiting nature of the output stage is especially evident in class AB designs where the signals being handled by each half of the output stage have highly nonlinear half-wave-rectified waveforms and where crossover distortion is easily generated. In contrast, it is not difficult or prohibitively expensive
to design front-end circuitry of exceptional linearity.
Overall negative feedback greatly improves amplifier performance, but it becomes progressively less effective as the frequency or speed of the errors being corrected increases. High-frequency crossover distortion is a good example. The philosophy here is based on the observation that only the output stage needs extra error correction and that such local error correction can be less complex and more effective.
While the power MOSFET has many advantages, it was pointed out that the lower transconductance of the MOSFET will result in moderate crossover distortion unless rather high bias currents are chosen.”
-----------------------------------------------------------------------------------------------------------
So, to utilize MOSFET power transistor advantages and keep reasonably low distortion, we need to apply error-correction circuit in the output stage. One of the error-correction circuit attempts is my power amplifier called PM-AB2. The amplifier is very simple, it uses an opamp as a voltage amplifying stage and an output stage with N-MOSFET/P-MOSFET pair, that's all.
Its basic schematics can be seen below.
The error-correction circuit is constituted by Q4 and Q7 transistors, R4-R6 and R15,R16 resistors and C1, C2 capacitors. These parts monitor the output voltage and create the variable bias voltage between gates of M1 and M2, to minimize non-linearity of the M1, M2 output stage with the load used. In the image below, we can see that the bias voltage between gates of M1 and M2 is not constant, due to action of the error-correction circuit, but tries to minimize output stage non-linearity to keep the low distortion. Without this error correction, amplifier distortion is about 10x higher.
Complete schematics of the amplifier follows:
One of my goals is to have high input impedance, then the amp makes light load for the DAC or preamp and works well even with passive “preamp” or tube preamp. The input impedance is close to R3 value, 100 kohm. To keep this high input impedance, the IC1 opamp must have JFET input stage. This excludes bipolar input opamps from considerations. And, I like JFET input opamps also because of their much higher immunity to EM interference. OPA134, OPA627, LT1122 and OPA445 were tested in this amplifier. OPA445 would allow for higher power supply voltage and thus higher voltage swing/output power, but its noise and high frequency linearity is worse. The results with the other three opamps were almost identical.
I use Hitachi 2SK413/2SJ118 power transistors for the reason that I have a good stock of these parts. So the error-correction resistor network is optimized to these devices. Idle current is set to 80 – 100mA.
Measurements
Square wave response
This is one of the most important measurements, on power amplifiers, at least to me. At lower repetition frequency, it directly shows the low frequency extension of the frequency response, at higher repetition frequencies it shows rise time (Tr 10%-90%) and is directly related to -3dB high frequency corner (Fc = 0.35/Tr). The flatness of square wave response top and bottom is related to frequency response flatness. Square response also shows the stability of the amplifier and possible high frequency oscillations, that are impossible to disclose by a slow spectrum analysis in the audio band.
We can see that rise time/fall time is below 2us, corresponding to at least 170kHz/-3dB bandwidth. The response is aperiodic, free of overshoots or oscillations or ringing.
Harmonic distortion
ASR asks for 5W/4ohm/1kHz SINAD measurement, so here it goes:
THD+N = 0.0032% equals to SINAD = 90dB. This is not a record breaker, but is more than enough to be inaudible. Mains related residuals are below -108dBr. There is no hiss, no hum, no buzz audible even with the ear at the tweeter/midrange/bass drivers of the speaker with 90dB/2.83V/1m sensitivity.
THD and THD+N as a function of output power at 1kHz/4ohm is shown below
and THD/THD+N as a function of frequency at 16W/4ohm
CCIF Intermodulation distortion 19kHz+20kHz
is again of of the important measurements to me, as it shows possible issues in high frequency linearity
As we can see, intermodulation products are below -90dBr.
Amplifier construction
The following images show amplifier construction. It uses a linear power supply and it is built in a standard 3U x 19” case with side heatsinks. It is one of my prototype cases. Though it is a linear class AB amplifier with a linear power supply, its power consumption is 14W at idle or low volume.
When designing a linear power amplifier, we have a choice of bipolar or MOSFET power transistors at the position of output devices. Bipolar (BJT) transistors are used much more frequently, but please let me also discuss and show a design with MOSFET power transistors. I would start with a quote from excellent classic design book by Bob Cordell: Designing Audio Power Amplifiers:
-----------------------------------------------------------------------------------------------------------
“MOSFET class AB biasing tends to be simpler and less critical. MOSFETs do not have an optimum class AB bias current. Instead, they operate better with greater bias as long as thermal objectives are met. For this reason, typical MOSFET power amplifiers operate at higher bias current per output pair and have a larger class A region of operation for small signals. However, their lower transconductance tends to result in higher measured values of static crossover distortion. Because of their high speed, MOSFET amplifiers are less prone to dynamic crossover distortion as a result of switch-off characteristics. MOSFETs do not suffer from beta droop and fT droop at high currents and are generally able to handle high peak currents better than BJTs. The ease with which MOSFETs are driven also means that there is less stress on driver transistors when the amplifier is delivering high current to the load.
MOSFETs are a bit more prone to high-frequency parasitic oscillations as a result of their inherently higher speed. For this reason, circuit design and layout can require more care than for BJT designs.
Biasing power MOSFETs for class AB operation is different from that for BJTs in two ways. First, MOSFETs require greater forward bias at the gate than BJTs do at the base. Vertical MOSFETs can require up to 4 V. Some vertical MOSFETs like the 2SK1530 require only about 1.7 V, however. If MOSFETs requiring 4 V forward bias are combined with emitter follower drivers, the total bias spreading voltage will be on the order of 9.2 V. This compares to a typical bias spreading voltage of about 4.0 V for a BJT output Triple. This difference in required bias spreading voltage is not a problem for the traditional Vbe multiplier or minor variants of it, but it does imply that the driver circuitry will require more voltage headroom in a MOSFET design. This is sometimes dealt with through the use of boosted power supplies for the circuits preceding the output stage.
Crossover distortion is one of the most insidious distortions in class AB power amplifiers. It occurs at fairly low signal levels and often contains a high-order distortion spectrum that is more dissonant and difficult to remove with negative feedback. It is a result of the changing gain of the output stage as the signal current delivered to the load goes through zero (the crossover). MOSFET output stages are also subject to crossover distortion.
Static crossover distortion in BJT output stages is a result of the output impedance changing as the output current goes through zero. The output impedance forms a voltage divider with the load impedance; as a result the gain of the output stage changes. The lower the value of the output impedance, the smaller the crossover distortion will be for a given percentage change in the output impedance.
The same is true for power MOSFET output stages, but the output impedance is generally quite a bit higher for a given amount of bias current. This is because the transconductance of a MOSFET is much smaller than that of a BJT. The transconductance of a BJT at Ic = 100 mA is about 4 S. The transconductance of an IRFP240 biased at Id = 150 mA is about 1 S. As a result, the sum of the transconductances of the upper and lower MOSFETs dips in the crossover region. This is referred to as transconductance droop.
Conventional negative feedback is not the only way to reduce distortion. Various error-correction techniques can be used in place of, or in connection with, negative feedback.
In virtually any well-designed power amplifier the output stage ultimately limits performance. It is here where both high voltages and large current swings are present, necessitating larger, more rugged devices that tend to be slower and less linear over their required operating range. The performance-limiting nature of the output stage is especially evident in class AB designs where the signals being handled by each half of the output stage have highly nonlinear half-wave-rectified waveforms and where crossover distortion is easily generated. In contrast, it is not difficult or prohibitively expensive
to design front-end circuitry of exceptional linearity.
Overall negative feedback greatly improves amplifier performance, but it becomes progressively less effective as the frequency or speed of the errors being corrected increases. High-frequency crossover distortion is a good example. The philosophy here is based on the observation that only the output stage needs extra error correction and that such local error correction can be less complex and more effective.
While the power MOSFET has many advantages, it was pointed out that the lower transconductance of the MOSFET will result in moderate crossover distortion unless rather high bias currents are chosen.”
-----------------------------------------------------------------------------------------------------------
So, to utilize MOSFET power transistor advantages and keep reasonably low distortion, we need to apply error-correction circuit in the output stage. One of the error-correction circuit attempts is my power amplifier called PM-AB2. The amplifier is very simple, it uses an opamp as a voltage amplifying stage and an output stage with N-MOSFET/P-MOSFET pair, that's all.
Its basic schematics can be seen below.
The error-correction circuit is constituted by Q4 and Q7 transistors, R4-R6 and R15,R16 resistors and C1, C2 capacitors. These parts monitor the output voltage and create the variable bias voltage between gates of M1 and M2, to minimize non-linearity of the M1, M2 output stage with the load used. In the image below, we can see that the bias voltage between gates of M1 and M2 is not constant, due to action of the error-correction circuit, but tries to minimize output stage non-linearity to keep the low distortion. Without this error correction, amplifier distortion is about 10x higher.
Complete schematics of the amplifier follows:
One of my goals is to have high input impedance, then the amp makes light load for the DAC or preamp and works well even with passive “preamp” or tube preamp. The input impedance is close to R3 value, 100 kohm. To keep this high input impedance, the IC1 opamp must have JFET input stage. This excludes bipolar input opamps from considerations. And, I like JFET input opamps also because of their much higher immunity to EM interference. OPA134, OPA627, LT1122 and OPA445 were tested in this amplifier. OPA445 would allow for higher power supply voltage and thus higher voltage swing/output power, but its noise and high frequency linearity is worse. The results with the other three opamps were almost identical.
I use Hitachi 2SK413/2SJ118 power transistors for the reason that I have a good stock of these parts. So the error-correction resistor network is optimized to these devices. Idle current is set to 80 – 100mA.
Measurements
Square wave response
This is one of the most important measurements, on power amplifiers, at least to me. At lower repetition frequency, it directly shows the low frequency extension of the frequency response, at higher repetition frequencies it shows rise time (Tr 10%-90%) and is directly related to -3dB high frequency corner (Fc = 0.35/Tr). The flatness of square wave response top and bottom is related to frequency response flatness. Square response also shows the stability of the amplifier and possible high frequency oscillations, that are impossible to disclose by a slow spectrum analysis in the audio band.
We can see that rise time/fall time is below 2us, corresponding to at least 170kHz/-3dB bandwidth. The response is aperiodic, free of overshoots or oscillations or ringing.
Harmonic distortion
ASR asks for 5W/4ohm/1kHz SINAD measurement, so here it goes:
THD+N = 0.0032% equals to SINAD = 90dB. This is not a record breaker, but is more than enough to be inaudible. Mains related residuals are below -108dBr. There is no hiss, no hum, no buzz audible even with the ear at the tweeter/midrange/bass drivers of the speaker with 90dB/2.83V/1m sensitivity.
THD and THD+N as a function of output power at 1kHz/4ohm is shown below
and THD/THD+N as a function of frequency at 16W/4ohm
CCIF Intermodulation distortion 19kHz+20kHz
is again of of the important measurements to me, as it shows possible issues in high frequency linearity
As we can see, intermodulation products are below -90dBr.
Amplifier construction
The following images show amplifier construction. It uses a linear power supply and it is built in a standard 3U x 19” case with side heatsinks. It is one of my prototype cases. Though it is a linear class AB amplifier with a linear power supply, its power consumption is 14W at idle or low volume.
Last edited: