Work in progress, not built and tested.
The plan was to eliminate some of the 'unfashionable' features of the MJR7 without sacrificing too much of the performance. The output coupling capacitor had to go, as did the input capacitor and any electrolytics, and there is at least some appearance of symmetry in the driver stages. The 'symmetry' does have a point, which is to cancel the distortion from the base-collector capacitances of the 4 transistors at least for small signals. For each Vcb increase in one transistor there is an equal Vcb reduction in an identical type, so variations in capacitance cancel to some extent. A jfet input is included, but with a single input device and a bootstrapped cascode jfet stage, plus a bjt cascode to allow a higher supply voltage, and also improve supply rejection. This input stage is unable to accurately control output offset voltage, so servo control becomes necesary. The mosfets are driven from a class-AB stage so that plenty of current is available to charge the gate capacitances and a high slew rate limit is possible. The open-loop -3dB is a fairly high 10kHz. Using dual-die mosfets a power rating over 200W into 4R is possible.
One problem is frequency compensation, this is more difficult than for the MJR7 not because of the extra stage, but because of the simpler VAS input which is a single transistor T4. I originally chose this as a BC560C just because I have lots of them. After a few simulations I found that it was difficult to get a good stability margin, and I realised the problem with the BC560C is its base-spreading resistance, which I took to be 160R. Assuming the 200MHz fT specification from Onsemi is correct (others specify 100MHz) the Cbe value at 5mA collector current is about 200pF, and the 160R plus 200pF add enough phase shift to be a problem, about 45 degrees at the loop unity gain frequency, so we need a lower rbb'. Something like the 2SA1085 with rbb' = 2R would be good, and I have a few of these, but they are now almost entirely unobtainable, so not ideal for a new design. Looking at easily available types the MPS2907 looks possible, with rbb' around 40R and 200MHz fT. The BC327 should be even better with rbb' around 30R and 260MHz fT, so for now that is the chosen type. T7 is less critical, a BC550 or anything similar should be ok, it should increase the current source impedance at least compared to just using a resistor.
That 10p compensation capacitor looks wrong, I expected about 100p would be needed, I need to check my simulation. One of the 220R presets is to adjust output offset so that the servo doesn't have too much work to do, and the other preset adjusts output stage quiescent current.
My aversion to relay speaker protection is based on personal experience with a few unreliable examples. There are of course alternatives to relays, for example using mosfets for the switching. These have a built in rectifier from drain to source, which means we need two in series connected in opposite direction, but provided we use types with very low on-resistance that should be no problem. There are plenty of examples published, so no need to design anything new. Driving the gates from a photovoltaic type opto-coupler simplifies the circuit requirements, the Panasonic APV1122 looks good, it has a relatively fast switch-off time of 0.1 ms, which is useful for high power applications, but the current output may be too low for driving the mosfet gate capacitance fast enough without additional complexity. I thought of including the mosfets inside the feedback loop, as I did with the MJR7 output capacitor, but this probably adds problems without being a significant improvement. Switching the supply rather than the output has some advantages, but may have problems with switch-on thumps or clicks, so I prefer the output switching option. (Switching both output and supply would be even safer, and need not add much to the cost. Incorporating a heatsink temperature detector to switch off and prevent overheating is also a good idea.) Here is a version with mosfet speaker switching. I have not included the 'control circuit' for the mosfet output switch, there are a few obvious variations, but without trying a few I don't want to make any recommendations.
The circuit is not a final design, just an example to show the sort of thing we could end up with if we try to 'improve' the MJR7. The only real improvement is the noise level, but the MJR7 noise is low enough for almost anything other than an ultra-high efficiency horn loaded speaker.
Of course we could improve the MJR7 noise level without giving up on the capacitor coupled output and all the advantages that brings. The same jfet input stage could be used as in the next example. The anti-thump circuit is a bit more difficult to implement, but maybe it is not necessary, or there may be alternative solutions.
The slew rate limit should also be 'improved' compared to the MJR7, but that is of no value, as I demonstrated in my 'Slew Rate-2' article. The gain-bandwidth is of more importance for transient accuracy, and this should be about the same for MJR11 and MJR7. Supply rejection is more difficult to predict, but should still be good.
There is one serious flaw in this circuit, which is the temperature coefficient of the output stage operating voltage. I had hoped to avoid servo control, but maybe that is the easiest solution.
A final idea to keep the anti-thump circuit is shown next as the MJR12.
To keep the same high feedback as the MJR7 to maintain low distortion the single BC337 in the VAS can be replaced with a CFP similar to the MJR7 input, but some experiment with the 1R5 resistor value may be necessary for good stability. The BF245C is becoming hard to find, so maybe something like the J111 could be substituted, the requirement being high Idss and high Vgsoff so that the BF862 is guaranteed a sufficiently high drain voltage, typically 3V. The BF862 could also be replaced by a J111, or maybe better a CPH3910. An alternative feedback arrangement is also used, but I have not even tried a simulation let alone actually built these circuits, so they are just 'ideas'.