Updated 28-Oct-2008. The circuit diagram has been updated. The most important change is that a 100n capacitor in series with a 1ohm resistor has been added to the output to improve stability with capacitive loads around 2nF, which could be the capacitance of unterminated speaker cables and so is a potential problem. The inductor damping resistor has also been reduced to 2R2. The explanation for these changes can be found on the MJR-7-Mk3 page.
With mosfet amplifiers there is no clear distinction between class-B and class-AB. For a bipolar output stage it is common to call a design class-B if the quiescent current, Iq, is adjusted for minimum crossover distortion, and class-AB if the current is increased well above this optimum level so that operation remains in class-A for small signals. (Not everyone agrees on this convention but it seems a sensible choice, for audio amplifiers at least, because these two classifications have rather different design problems. For example 'class-B' usually requires far more accurate quiescent current setting and thermal compensation for best results.) Class-AB causes higher distortion than class-B at higher signal levels where device switching starts to occur. For mosfets the distortion is generally highest at low quiescent current and reduces as Iq is increased, so there is not a similar optimum value giving minimum distortion. My choice of 100mA for the circuit presented here is because the mosfets used have zero temperature coefficient around this value, but with adequate heatsinks the current could be increased. With reduced supply voltage the circuit can even be biased into class-A for the lowest distortion, and here mosfets have some advantage because of their nearly square-law characteristics. If they were perfectly square-law then the peak output current in class-A would be 4 x Iq instead of 2 x Iq for typical bipolar output stages, and up to this level the distortion would be zero for ideal accurately matched complementary devices. Unfortunately well matched pairs are unobtainable, and they are not very accurately square-law, and they have non-linear capacitances, so even class-A still has its problems.
With 'class-B' the design problems are more difficult. I wanted to make a simple design, and decided that 6 transistors were the minimum needed for adequate performance. The first mosfet based circuit I tried is a little unusual, but is nothing really new. The results however are surprisingly good, third harmonic being about 0.006% at 20kHz at 300mV input and 7R5 load. (This level is chosen to reveal any crossover distortion.) Distortion is at about the same level as in my previous feedforward output design, but stability is found to be more easily predictable. Quiescent current is again not critical and crossover effects are very small, but without the need for any accurate component values. This design is far better than I had expected. A version with one extra transistor gave even lower distortion, though there is no real need for improvement, and this is described HERE.
I carried out music tests extracting the distortion using music signals and a speaker load (a Mordaunt-Short MS20 was used), using a cd player output taken direct via a simple two resistor attenuator. At a little above my normal listening level the 'error' signal extracted was amplified and recorded, and later listened to alone. It just sounded like noise together with fairly undistorted music just above the noise level which had not been totally nulled. Any distortion must have been well below the noise level. The noise level alone is inaudible in normal operation even with an ear held next to the speaker, so it seems safe to conclude that distortion is even further below audibility. Here also is a ten second extract as a wav file (1.69MB). (Although amplified considerably this is still at a very low level, so quantisation distortion could have some audible effect.)
The distortion extraction circuit used, which works with any test signal and load, even using music and driving a speaker, is actually built into this six transistor amplifier, with a low impedance output, and all you need to add is a resistor plus a low noise op-amp circuit to increase the extracted signal level to a more easily observable level. A perfect null is not achieved, but attenuation of the undistorted component relative to the distortion exceeds 80dB at 2kHz, falling to 65dB at 20kHz. There is no adjustment or accurate component selection involved, though a simple addition to the circuit can add an adjustment for an even better null. How this works is explained HERE.
It was an obvious choice to use mosfet output power devices. The low drive requirements can reduce driver stage complexity. A capacitor coupled output and single supply avoid the need for output switching relays and associated circuitry for speaker protection, and a single supply line fuse may then be all the protection needed in case of shorted output. To avoid problems of added distortion and reduced damping factor the output capacitor is included in the overall feedback loop.
INPUT STAGE: Mosfet amplifiers generally have higher open-loop distortion than bjt versions, but can have far higher feedback loop gain, and to reduce distortion to my usual aim of under 0.01% at 20kHz over 60dB loop gain is needed. With such a level of feedback the distortion added by any reasonably good input stage should be reduced to insignificant levels, so the design is not critical. The choice made is to use a complementary feedback pair, in which a low current input device has low noise, with a higher current second device. The emitter resistor needs to be very small, around 1 or 2 ohms, to give the required high open-loop gain.
DRIVER STAGE: To drive the mosfet output stage we need a large voltage swing, but the current only needs to be sufficient to charge and discharge the mosfet input capacitances, and for a low to medium power amplifier only a few mA are needed. The complementary feedback pair input stage alone is capable of the current level, but to avoid distortion caused by the Early-effect should have only a very low output voltage. The following driver stage therefore needs a high voltage gain, but a current gain of unity is sufficient, and a single common-base device in a cascode arrangement is used. The base voltage, about 3.3V, reduces the maximum output voltage swing, but the reduction is typically less than half a dB, so not important. The reduction from both this and the mosfet gate-source voltage requirements could be avoided using an arrangement of bootstrapped loads, but squeezing out the maximum possible power rating is not a high priority. The low voltage, around 2V, at the input transistor collector means the input signal voltage must also be low to avoid a relatively large collector-base voltage variation causing Early-effect distortion, so the amplifier is designed to be inverting, with a 'virtual-earth' input, and typically under 1mV signal at the input base. This has the fortunate side-effect described above, that a very simple method of distortion testing becomes possible.
OUTPUT STAGE: This is just a conventional mosfet output stage, and the capacitances of the mosfets limit the high frequency open-loop gain. The other component setting the open-loop gain is the emitter resistor of the input transistor, typicaly 1 or 2ohms, and this can be reduced to increase loop gain and reduce distortion, or increased to improve the stability margin if needed. The mosfets I used are claimed to include internal zener diodes to limit the gate-source voltage to safe levels, but the data sheets fail to include any maximum current rating for these diodes, so I decided to play safe and include the additional 12V zener diodes shown to limit gate-source voltages.
STABILITY: The 10p capacitor in parallel with the 220k overall feedback resistance is included because the resistor used is unlikely to have a predictable impedance up to the unity loop gain frequency, which may be 10MHz or more in a mosfet amplifier. This capacitor together with the 390p from the input base to earth define the feedback level at high frequencies more reliably than high value resistors, and also the input transistor capacitance now has little effect. Ringing with capacitive loads is almost entirely absent (I had to use a 4uF load to see any square wave ringing at all, and even then the signal before the output inductor had no ringing, so it was just an inductor resonance effect, nothing to do with stability). Having finally checked the inductance (13 turns 1.2mm enamelled copper wire with inside diameter 1cm) it is only around 0.4 uH, which will have practically no effect at audio frequencies, but all capacitive loads tried are handled well so I left it unchanged.
If the mosfet capacitances alone applied a 6dB/octave reduction from the -3dB open loop frequency around 5kHz the unity gain frequency would be 40MHz. The input cfp stage however starts to roll off above 5MHz, almost entirely because of the pnp transistor base-emitter capacitance. The 1n input filter capacitor reduces loop gain by 6dB above the audio range. Stability is considered in more detail HERE.
OUTPUT CAPACITOR: There is widespread mistrust of electrolytic capacitors, especially when used for signal coupling. I have seen them accused of reverberation, hysteresis and several other forms of unmusical behaviour. (Dielectric absorption is often mentioned, but the usual model of this effect is entirely linear. I demonstrated in my Measuring Capacitor Distortion article that for a coupling capacitor this could actually reduce phase errors.) The real effects include a little low-order harmonic distortion, and a capacitance and equivalent series resistance which depend on the frequency. My previous design demonstrated that low distortion was possible when including the capacitor in the overall feedback loop, and this is again done here, but with far higher loop gain. The advantages are considerable, including the need for only a single supply voltage, and avoidance of relay contacts in series with the speaker, which are often used to provide speaker protection in case of large output offset voltages resulting from fault conditions. The capacitor prevents any dc output other than a small leakage current reaching the speaker. Speaker protection is missing from many published designs, but is really essential. Reliability problems with relay contacts persuaded me to reconsider output capacitors, although it looked at first like a backward step. I have never experienced capacitor failure in audio circuits although I spent many years repairing electronic equipment and have an amplifier of my own design which has been in regular use for over 30 years. The 'lifetime' specifications for electrolytics are generally at the maximum temperature rating, and lifetime is often claimed to double for every 10deg.C reduction. Use of the 105deg.C rated types should increase lifetime typically 4 times compared to the 85deg.C types with the same lifetime rating. Electrolytics are of course not close tolerance components, and some variation and long term drift in characteristics must be expected and taken into account. Other output capacitor 'problems' include reduced bass damping factor and possible large switch-on thump as the capacitor charges through the speaker, which are both reduced by the overall feedback. The output capacitor used was actually two 2200uF in parallel. The larger total surface area compared to a single 4700uF will help keep the temperature down and further improve lifetime.
POWER RATING:With 63V supply capacitors the supply should be limited to about 60V, and then the amplifier can produce about 30W into 8ohms and 50W into 4ohms. High voltage output and driver transistors (160V) are used, and although not recommended, the power rating can be increased just by changing the supply voltage and using capacitors with adequate voltage rating, remembering to adjust the 10k preset to set the output stage dc level to half the supply voltage. This should only be tried using speakers with a medium or high impedance, with adequate heatsink, and for normal music signals rather than prolonged sine or square wave testing. I have tried a 94V supply, and using a toneburst signal (IHF-A-202 / EIA RS-490) measured maximum output power before clipping at 79W into 7R5. (A different design using obsolete TO3 mosfets reached 100W with the same supply, but this is only a 1dB difference.) The output coupling capacitors should be rated at the full supply voltage to survive undamaged under fault conditions. 100V types are far more expensive than 63V, but don't be tempted to use 63V types with 100V supplies, even though they then only operate at about 50V. An open circuit npn driver transistor would cause the output voltage to rise close to the positive supply and then capacitor damage becomes highly likely if the rating is insufficient. Most of the initial testing was carried out with only a 48V supply. Before assuming that this is too low it is instructive to check the power level being used in the intended application. With my MS20 speakers rated at 87dB/W at 1m. I have used an 85W amplifier for many years, in a room 7m x 5m, but recently measured peak output at my normal listening level, and was surprised to find it rarely went over 1W. This demonstrates that linearity at low power levels is important, and the present design is excellent in this respect as demonstrated on the 'More recent results' link below.
More recent results using improved measuring techniques.
Details of component types and recommended UK suppliers.
Power supply and input controls, plus the board layout.
Photos of stages in the construction. This is a version I built myself, just to use for my computer sound output, and designed for low cost, so the appearance is not great, but it works well.
TYPICAL SPECIFICATION.
INPUT IMPEDANCE: 10.4k
VOLTAGE GAIN: 25.9dB
FEEDBACK LOOP GAIN: 81dB at 1kHz, 65dB at 20kHz.
HARMONIC DISTORTION at 300mV input:
2nd HARMONIC: 0.00033% at 1kHz, 0.0012% at 5kHz.
3rd HARMONIC: 0.0001% at 1kHz, 0.0004% at 5kHz.
INTERMODULATION DISTORTION:
Input 20kHz 150mV + 19kHz 150mV, Distortion product at 1kHz, -90dB relative to 300mV total input.
DAMPING FACTOR: This is included to demonstrate that the output capacitor effect is not only eliminated by the feedback loop, but the output resistance even becomes slightly negative. The damping factor is an almost entirely unimportant specification. The resistance of the speaker voice coil in series with the amplifier output impedance limits the level of speaker damping possible, so provided the damping factor is more than about 20 further increase makes practically no difference. The values measured for an 8ohm load are:
380 at 1kHzOUTPUT POWER WITH Vs=60V: 30W average sine-wave power into 8ohms.
820 at 30Hz
1500 at 20Hz
Negative resistance -0.01ohm at 15Hz, -0.06ohm at 10Hz.
Footnotes:
1.). The general circuit arrangement, as I said earlier, is nothing new, and I have found something similar dating back to 1964, now described on the Elliot Sound Products site. This is a simple and not particularly good design (though not bad for 1964). It does include the output capacitor partly in the feedback loop, but the low frequency stability is not convincing, with three capacitors adding phase shifts, one of which depends on the load impedance, also very little feedback at 20kHz, so the distortion will be high. The idea of adding a buffer stage at the input is a good point which may be worth copying if a higher input impedance is needed, but with a more advanced circuit.
2.). The MJR-6 is included in simulation tests here and is also mentioned on some of the links. It is in French, and I only understand a little, so some of my criticisms here may be unfair. There is a xls file giving simulation results for more than 20 designs, with points awarded for the various results. The MJR-6 appears to do well apart from having 4 points deducted for the output stage mid-point voltage temperature drift, resulting from variations in Vbe and current gain of the input transistor, mostly caused by slow changes in air temperature, any self-heating related to the input signal being extremely small. For a capacitor-coupled design this has no effect on output offset which measured somewhere under 0.1mV, and depends on the leakage current of the capacitors. The drift only affects the maximum peak output level, and should have no effect on sound quality below clipping. Unless a stabilised supply is used the supply voltage is not constant, so the question of whether the output stage voltage is at the mid-point is meaningless, and it could even be argued that if the temperature is high this must be because the power output is high, and so the average supply voltage will be reduced, and a reduced 'mid-point' voltage at high temperatures could therefore be an advantage. (If I ever make a commercial version I must remember to mention this excellent design feature!)
Another point is reduced because supply rejection is only 90dB, which I guess is calculated at 100Hz. This could easily be improved if required, but the supply noise may already be 40dB below maximum output, so then the actual supply effect at the output is at about -130dB. The threshold of hearing is at sound level 20dB at 100Hz, so assuming 100dB maximum sound level, the supply effect is probably 50dB below the threshold of hearing. The highly regarded JLH class-A amplifier was found to have only 62dB supply rejection, but even this is more than adequate at 100Hz.
On the plus side the distortion figures are among the best, with only two others having lower third harmonic, and both of those have far higher quiescent current, operating in class-A at the signal level simulated. From the present top 16 (Jan 2007) I believe only the MJR-6 and one other design are not simulated in class-A. The MJR-6 distortion, both simulated and measured, is primarily second harmonic.
The MJR-6 was never intended to be the best possible design, I just wanted to make a simple and reliable amplifier suitable for less experienced constructors, so maybe I should be happy with their rating. Another design included is one of my old favourites, the JLH class-A, which achieves adequately good results without unnecessary complexity. I once made one myself, and was perfectly happy with it for many years.
