SETUP AND TESTING.
PHOTOS OF CONSTRUCTION.
Latest update: The compensation capacitor has now been reduced from 220p to 100p and its series resistor increased to 470R. To keep adequate stability margins with reactive loads two other changes then became necessary. The capacitor from input base to earth is increased from 330p to 470p (this is part of the compensation, not an input filter - the impedance at that point is just a few ohms because the feedback creates a 'virtual earth' - the other 470p is the only input filter), and also the resistor in parallel with the output inductor is reduced from 2R2 to 1R5 - this also reduces 'ringing' with a 2uF load.
Much of the information on the original MJR6 and MJR7 pages, plus the later MJR7-Mk2, MJR7-Mk3 and MJR7-Mk4 is still relevant. The MJR6 page includes distortion extracted using the nulling method with a speaker load and a music signal to demonstrate that even the most basic of the designs has no audible distortion in normal use.
The Mk5 differs very little from the Mk4, they both include two channels on one standard size board (6" x 4") with a single star earth included on the board. One important difference is the use of a LM234 current source in place of a resistor, which was found to be responsible for some of the second harmonic distortion. (The LM234 is not essential, see note further down the page, a 470R resistor in its place makes little difference.) The mosfet bias control then needed changing a little to guarantee sufficient operating voltage for the LM234, and for this one more red LED is added. Another reason for these changes was to enable the addition of feedforward error correction in the later MJR9 version, which can reduce audio frequency distortion components to even lower levels. From a sound quality point of view there is almost certainly no point in further distortion reduction even beyond the original MJR6, but the additions needed are fairly simple, and it is interesting to see how far these circuits can be taken.
I have included an anti-thump circuit, at the top left of the diagram, which consists of BAV20 diodes plus a resistor and capacitor. I reduced the capacitor across the LED biasing the 2SA1209 current source to 1n to improve recovery from clipping, and this makes the switch-on pulse at the output worse, typically 8V. With 470uF in the anti-thump circuit the pulse is then not much over 1V, which makes just a small unobtrusive sound with my own speakers. ( Even this may still be a problem if directly driving mid-range or high frequency drivers in a bi-amp or tri-amp system. For this purpose the 4700uF output capacitor can be reduced, but then the 10uF in the feedback network and the 2u2 at the amplifier input must also be reduced in the same proportion. e.g. the values could be 470uF, 1uF and 0.22uF giving a -3dB low frequency gain around 60Hz.) The pulse can be reduced further by increasing the 470uF anti-thump capacitor, but there is then a longer delay before the amplifier reaches normal operation, it is about 15sec for each 470uF, though in practice there should be very little audible effect. At switch-on the capacitor starts to charge, which slows down the amplifier output capacitor charging pulse and so reduces the charging current through the speaker, but once fully charged the BAV20 diode connected to the 2SA1209 is reverse biased, so there is no further effect on the circuit. The capacitor needs a voltage rating greater than the supply voltage used because in normal operation it is charged close to the full supply voltage.
The NPN input transistor Tr1 was originally a 2SC2547E, but this is becoming increasingly hard to find, and those I have bought recently include some obvious fakes far below the specified current gain range. I used a 2SC2547E input stage in one channel and a BC549C in the other, operating at 200uA instead of the original 500uA, and confirmed that the noise levels at the amplifier outputs are then almost identical. Unfortunately reducing the input transistor current has some effect on stability, and without the 100n plus 1R at the output the amplifier is unstable with small capacitance loads around 1nF to 5nF. Relying on the output network for stability is not necessarily a problem, but in previous versions this was not essential, it was just included as extra security. Even reverting to 500uA operating current but using the BC549C or MPSA18 still adversely affected stability, and for now my only guess is that the higher base-spreading resistance rbb' ( 160R for the BC549C ) in combination with the input capacitance adds sufficient further phase lag to be a problem. More readily available types such as the 2SC2240BL (rbb' = 40R) although not as good as the 2SC2547E (rbb' = 2R) may be good enough.
The PNP is shown as a BC560C, which I thought would have higher current gain than the 2SA1085E, but testing a few samples I found little difference, so either type should work here.
Initial distortion test results show that the LM234 makes only a small improvement in the Mk5 circuit compared to a 470R resistor in its place. There is therefore little point including this unless the feedforward option is to be included to make a MJR9, in which case the LM234 will enable better distortion nulling. The circuit section with the simpler 470R option is shown next.
GAIN AND PHASE SHIFT.
This simulation result is identical to the Mk4 version, none of the changes have much effect on gain or phase shift. There is an almost perfect linear phase response from 1kHz to 20kHz, having the same effect in that frequency range as a constant time delay of 3.3usec, so having practically no effect on wave shapes. The -1dB frequency range is about 12Hz to 30kHz.
To avoid the mosfet mounting bracket used in earlier versions the mosfets are connected near the edge of the board so that they can be fixed directly to a heatsink. They could be soldered onto the board, but then there is some difficulty in ensuring they all line up correctly with the heatsink surface, and so the layout has been changed to allow the option of using terminal blocks as shown in the next photos. The blocks should be soldered in place before the 300R gate resistors so that these can be routed correctly to avoid obstructing the blocks. The inputs can also be taken to a terminal block if required.
The coil, shown next, is air-cored, and is made using 18-gauge enamelled copper wire (1.2mm dia.) which can be made by winding it on a 1cm dia former, for example an AAA battery as shown in the photo. There are 13 turns, giving an overall length about 17mm. The other photo shows how the gate protection zeners are fitted. To save space only two holes are provided and the zeners have two ends soldered together off the board. The connected ends are the same polarity, in the photo the cathodes (indicated by a dark band) are connected, but it works just the same if anodes are connected together instead. Don't try using a normal diode in series with a zener, that works in some mosfet circuits, but not in this one.
The next photo shows the final MJR7-Mk5.