The original MJR-6 is already excellent and I have been using one myself for over 6 months with no problems. Anyone who wants a good simple amplifier can ignore all the 'improvements'. Extracting the distortion when playing music through speakers and amplifying it 100 times I can still hear nothing unpleasant, so I believe it is safe to say that regarding non-linear distortion the MJR-6 is at least 100 times better than it needs to be for playing music at normal domestic levels with a reasonable load impedance. Any suggested 'improvements' are therefore not essential.
If I ever arrive at a final version I may add a suitable board layout, but until then these should be regarded as untried suggestions, and for some time this will remain 'work in progress.'
A published simulation of the MJR-6 suggested that supply rejection at 100Hz was less than I expected at -90dB. This is certainly good enough, taking into account that the supply ripple may already be 40dB below maximum output, and the threshold of hearing is at 20dB sound level at 100Hz, so assuming maximum output 100dB the actual supply effect will then be 50dB below the threshold of hearing. The MJR-7 should already be better because the driver stage current source, which is one potential source of breakthrough, feeds a far lower impedance. The overall feedback should reduce any effect in the output stage to very low levels, but in the input stage the original 3V6 zener, operated at 1mA, had an impedance about 400 ohms, and so the regulation at this point was less than perfect. Using a better regulated voltage is easy, but the zener plus 1n capacitor were chosen to improve recovery from clipping in the input stage. On positive input clipping the input stage current is limited by the zener voltage falling, so this voltage must still be allowed to fall and recover quickly.
Keeping the impedance low at this point is also important at higher frequencies, where feedback through the collector-base capacitance of the 2SC2911 needs to be kept low. Increasing the 1nF capacitor will affect overload recovery, but another way is to use two red leds instead of the 3V6 zener, and these have a voltage drop around 1V7 each at 1mA and impedance about 40R, and according to some published measurements have lower noise than a typical zener, though noise is not a serious problem at this point in the circuit. Another red led is used to bias the driver stage current source instead of the original two diodes. The 220uF capacitor is partly to slow down the switch-on of the current source to prevent a switch-on thump at the output before the overall feedback becomes active to protect against this.
The d.c. operating level of the output stage depends on the base-emitter voltage of the input transistor, and this has a temperature variation of -2mV per deg.C. The output coupling capacitor prevents any slow thermal drift from reaching the speaker, and the only problem is that at high temperatures the change in operating level could be 2 or 3 volts, and so the clipping level will become more asymmetric. A big heatsink or a well ventilated case would reduce this problem. I originally showed a version on this page using the LM19 temperature sensor to provide temperature compensation, but have now decided that the benefits are so small that this is almost entirely pointless. There is even an argument that the temperature drift could be a positive benefit if an unregulated supply is used, because the temperature will only rise during operation at continuous high power, which will reduce the average supply voltage, and so a small fall in 'mid-point' output stage voltage could then help increase the available positive output before clipping. This will of course not be very exact, but neither will a fixed voltage, so compensation to achieve this is now omitted. If clipping is common you need to turn down the volume, or use a higher power amplifier, rather than worry about a temperature drift of the clipping level which is unlikely to exceed 1dB.
A high temperature cut-out is a more important addition. A possible method of adding a 70deg thermal cutout is to switch off the driver current source at high temperatures. This is not as reliable as switching off the power supply, and if the overheating was caused by faulty mosfets there is no certainty that they would be switched off effectively. I have read that when these mosfets fail they invariably go open-circuit, while if they became shorted the supply fuse should blow, so maybe this is safe enough. Switching the supply at the board is a problem because after the temperature falls the switch will reconnect the full supply voltage to the onboard 1000uF capacitor, with a high probability of damaging both the switch contacts and the capacitor. A possibility is to use a thermal switch to switch off the driver current, but also include a thermal fuse operating at a higher temperature to switch off the supply to the board if the switch fails to stop a temperature rise.
The original 15R resistor in the positive supply to the driver has now been abandoned and instead a 12V zener used to reduce the effect of supply noise on the current source. The point of this is to ensure the driver stage always clips before the output stage, which may be important for stability near clipping. I have also left out the 12V gate protection zeners, but I still have no certainty that the internal zeners have adequate current rating. I will try to find out about this, or if no information is to be found I may sacrifice a mosfet to determine its maximum safe current.
The 0.1uF capacitors used in parallel with the output and supply smoothing electrolytics are probably either completely unnecessary, or at least non-optimum. I have a Phillips application note from many years ago (early 1970s) which used 0.1uF capacitors across the supply but also added in parallel a 0.68uF in series with 1R, the resistor being to damp ringing caused by the supply capacitors resonating with the supply cable inductance. This was said to reduce output distortion near crossover in a class-B amplifier when the current in one half switches off. That design had no electrolytic on the amplifier board, but if these are used damping will be added by the internal resistance of a typical electrolytic. This varies with frequency, temperature, age, and varies between samples, and in any case the effect will depend on layout, which determines wiring inductance, so it is fortunate that extreme accuracy is not essential.
A series R-C Zobel at the amplifier output is included, and I just started with typical values used in other designs, with the intention of experimenting later to improve on this, but there appear to be no problems and I have left the R-C unchanged.
The purpose of this R-C is to ensure stability into any primarily inductive load at high frequencies, which could cause instability, typically a local output stage problem, when various stray reactances are taken into account. It may happen that some designs are already stable into any inductive load without this addition, but this will depend on layout and is difficult to test. It is easy to try different load capacitors. e.g. from 1nF up to several uF, but few of us have a wide range of inductances to experiment with. The original Hitachi application note for a 50W amplifier using the earlier 2SK133/2SJ48 family of lateral mosfets used 0.022uF and 4R7, but also a 10uH undamped inductor in series with the output, which seems a little excessive, and probably a bad idea.
The output inductor plus damping resistor are to ensure overall feedback loop stability with a capacitive load, and can be decided independently of the R-C network, though I have seen it stated that they should be related in some way. They solve different problems however, and so I suggest that in principle each should be related to its own worst case load.
As mentioned in the MJR-6 construction notes, the spikes found at the amplifier output for a particular alignment of the supply transformer were inaudible to start with, and easily reduced below the noise by rotation of the transformer, but the radiated signal could cause problems for other nearby equipment. It is not clear whether this is an unavoidable problem of the transformer used, or whether modification to the rectification and smoothing circuit could help, e.g. by adding a small series resistance or inductance to reduce the peak capacitor charging current. I have seen all sorts of other suggestions for improving power supplies, such as using soft-recovery diodes and adding various capacitors and resistors. My amplifier designs have good supply rejection, and the signal nulling distortion extraction method I used would have revealed any real effects on the amplifier output when playing music, but steps to reduce radiated interference may be worthwhile.
One more idea, to reduce the distortion another 20dB or more by the addition of a resistor. (Because of the capacitor coupled output in the MJR-7 it would also need a capacitor in series with the resistor, e.g. 100uF, but a direct coupled design may only need the resistor.) Suppose the distortion at the output of the original MJR-7 was 0.0002% at 1kHz. Most of this will be caused by output stage non-linearity, and the effect of the feedback will be to apply a signal to the input of the output stage sufficient to cancel most of the open-loop distortion at the output. If we measured the closed-loop distortion at the input of the output stage it would be something like 2%, around the same level as the open-loop output stage distortion. The really useful thing about this 2% distortion is that if the input and driver stages are linear it will be identical to the output distortion except for being ten thousand times greater, and also inverted. This is exactly the sort of thing we need if we want to apply error feedforward. Connecting a resistor, about 1k, at one end to the driver stage output, and at the other end to the output after the inductor will feed forward an opposite phase error signal, and will give output stage distortion nulling if the inductor has a small series resistance, about 0.1ohms, and is increased to about 3uH to compensate for falling driver stage gain at higher frequencies. With D applied to the load via 0.1ohms and -10,000D applied via 1k the distortion reaching the load will be zero.
This is more or less the same method used in the Quad 'current dumping' amplifier, just another way of looking at it. The Quad version used a feedback capacitor to accurately define the gain of the driver stages rather than use the open-loop gain as in my example. (The open-loop gain is not exactly predictable, so for accurate nulling a trimmer would be needed for the feedforward resistor, and the inductor can be adjusted by pulling the turns further apart or compressing them.) To work well we need input and driver stages with distortion far lower than the output stage - not too difficult - and a driver stage with low output impedance so that the non-linear input impedance of the output stage adds little further distortion. The loading effect on the driver stage from the 1k resistor may upset the nulling, and if I was going to use this technique with the MJR-7 I would add a buffer stage to drive the 1k resistor. An example of this is shown later.
In practice a larger output inductor will pick up more distortion from stray magnetic fields, and without careful layout could even add more distortion than we started with. The damping factor will also be reduced. Using a higher current buffer stage and reducing the feedforward resistor we could reduce the inductor and its series resistance.
Any amplifier using negative feedback round a loop including the output stage can in principle use this technique. Output stage distortion is reduced at its output and increased at its input as the feedback is increased, though not by the same factor. Unless very high feedback loop gain is used the feedforward resistor needed may have an inconveniently low value, and a high driver stage current could then be needed to drive this resistor. The Quad 405 used a 3uH inductor as in my example above, but needed a 47R feedforward resistor and a driver stage current of 50mA.
Here is a possible addition to the MJR-7, with a buffer stage to drive the 1k resistor. The addition has one transistor, two resistors and a capacitor, and the diagram shows where these are connected. The output inductor is increased to 3uH with 0.1ohm series resistance. The feedforward resistor is the 1k connected to the buffer transistor emitter. The lower 2k determines the buffer stage current, and with a 60V supply this will be about 10mA. At this voltage and current the transistor could be a 2SC2911 or similar.
There was rather more point in the Quad version, which used an unbiased output stage, and so needed feedforward to achieve good results, but with distortion under 0.001% up to 20kHz the original MJR-7 is already very good without any improvements.
(I wrote earlier that commercial use of this feedforward method may be restricted by the Quad patent, but I doubted whether this would still apply when using an open-loop driver stage and feedforward via a buffer stage. I have since seen a claim that the Quad patent has expired, but don't know if this is true.)
I specified an output inductor 0.4uH for my amplifiers, estimated by measuring the voltage drop when in series with a resistor.
When I checked using the formula
L (uH) = D2N2 /(L + 0.45D) for N turns, dia. D, length L (metres).
this gave a value 0.8uH. This is no problem, and anyone who has used my description of 13 turns 1cm inside diameter will have the same value I used. (D in the formula is the diameter at the centre of the wire, not the inside diameter.)
I have now checked the 'ringing' frequency with capacitive loads, and again this suggests a figure around 0.4uH, so I have doubts about the equation. I know inductor equations are notoriously inaccurate, but a two to one error? The exact value has no great importance, so I will leave this as an unsolved mystery for now.
There is some advantage in keeping the inductor value as small as possible. There is an article here by Rod Elliot in which crossover distortion only became measurable when an output inductor was added (see the updates at the end of his page). One explanation I would guess is that the highly non-linear output stage currents in a class-B amplifier will generate a varying magnetic field which can induce an emf in the output coil and add a distortion voltage to the output, a point I made earlier on this page. In addition to keeping the value small the location and orientation are important for reduction of magnetic field effects. The board layouts I have shown took this into account and are found to work well. With two channels and a power supply in the same case the layout becomes even more critical, and transformer orientation is also worth experimenting with as demonstrated here.
The inductor value I used was intended to give unconditional stability with capacitive loads up to about 4uF. For a general purpose amplifier any combination of speaker and cable could be used, so we have no certainty what is the worst case load we need to design for, and my own choice of 4uF as a maximum is fairly arbitrary and maybe a little over-cautious. The problem is not that the amplifier will immediately oscillate above this value, but a potential problem is around 100kHz where the phase shift becomes excessive, which could cause instability if the unity gain frequency falls, e.g. near clipping.
Keeping the inductance small adds a few problems. The optimum parallel damping resistance is smaller for a smaller inductance, and with a capacitive load the total load impedance at high frequencies is lower and can have unwanted effects on stability. The 7R5 damping resistor shown for my mosfet designs is too high, but that was the value I did all my testing with, and there were no obvious problems so I have left it unchanged for now. With the MJR-7 some further thought is needed, because the 300R mosfet gate resistors may then need to be reduced to avoid additional phase shift with a low impedance load at high frequency, but the 300R is also required to limit driver stage current with a shorted load.
An alternative to the inductor is a low value resistor, e.g. 0.22 ohms as used in some Linsley-Hood designs, which is not as effective as an inductor, but better than nothing. It is possible to design for stability without the inductor or resistor, but achieving the 0.001% distortion figure plus stability into 4uF with a simple 7 transistor design may not then be so easy.
Today's music. 3.June.2007.
This may not be the ultimate in high-end performance, but sounds pretty good to me. The CD/MP3 player is a Panasonic SL-CT700 (bought for £5 on Ebay), the headphones are Sennheiser PX100 (available new at £18 on Ebay, but watch out for fakes!), and the CD is 'The Classical Album 2005' borrowed from the local library.HOME.