I doubt whether the sort of idea suggested here would make a worthwhile improvement to the standard circuit given in the manufacturers data sheet, so I never tried it. It is nothing really new, but I include a few details here for anyone who wants to experiment.
Although excellent results are achieved with the mosfet designs there can be, in some parts of the world, difficulty in obtaining the lateral mosfets specified at a reasonable price. Integrated circuit amplifiers are so widely used that prices and availability are often far better. This by itself may not be sufficient reason for most of us to abandon discrete component designs, but there are other advantages also. I mentioned in my article on overload protection that the sole purpose of complex load line protection circuits is to prevent the output power device junction temperature from exceeding its maximum safe value, typically 170deg.C. This is only difficult because we have no direct means of measuring the temperature, and are reduced to estimating it by detecting voltage, current, heatsink temperature, and duration of power dissipation. Designers of integrated circuits however can place a temperature detection device immediately next to the output device on the same silicon chip, and achieve reliable protection with no external components. The National Semiconductors 'Overture' series have this sort of protection built in. The LM3886T appears to be a good choice, having a peak current output of 7A or more.
Looking at the LM3886T data sheet there are a few limitations. The input noise is typically 2uV (IHF-A-weighted), but a maximum of 10uV, which is adequate, but not good. The distortion at 30W into 8ohms is low, about 0.003% at 1kHz, but rises at low and high frequencies. The low frequency rise is typical of IC amplifiers because of the thermal coupling between the output stage and the low level stages, and the high frequency increase is as expected from the fall in feedback loop gain at higher frequencies, effects of non-linear capacitances and so on. Also, the power supply rejection is good for the positive supply, but relatively poor for the negative supply.
The supply rejection could be less of a problem for a single supply design rather than plus and minus supplies. The use of an output capacitor for speaker coupling then becomes necessary, which avoids the need for speaker switching relays to protect against fault conditions. The obvious approach is then to add an input stage. This will enable the use of a circuit similar to my previous designs, with output capacitor inside the feedback loop. A gain of about 10 in the input stage can reduce closed-loop distortion by the same factor, reduce noise if this is a problem, and avoid common-mode input distortion being added by the IC. Adding too much gain in the input stage would make stability difficult to achieve, but 10 seems reasonable.
Again there is nothing really new in this approach, I first saw it suggested for use in a pre-amp circuit using a 741 op-amp with an input stage to improve performance (Wireless World, Dec 1972, page 575, letter to the editor by D.R.S.Hedgeland, and improvement March 1973 p.119 by M.L.G.Oldfield.) This used a differential long-tailed-pair input stage, which is useful for driving the differential input of the IC, and can also reduce one small problem of my previous designs, which is the effect of temperature on d.c. operating level.
The general idea is illustrated in the following circuit diagram. This is not a complete or final design, just a first approximation. In a final version one of the LM3886T pins needs to be connected to a point at half the supply voltage, which is conveniently supplied here by the input stage supply, and there is a mute pin which needs an input current, and other additions are needed. The 1V2 bias voltage needs to have very low noise and a LM371LZ regulator could be used here. The 24V supply could also use a regulator, type L7824, but this has a maximum input of 40V so a zener of about 33V ahead of the regulator could be used. Two regulators and a zener may seem a little excessive, but they only cost about the same as the low noise input transistors.
Again I must stress that this is an incomplete and untested circuit, and anyone who wants to try it needs to be able to carry out the necessary tests for stability. An output inductor similar to the one in my other designs is certainly a good idea, and separate ac and dc feedback paths as used before are probably better than the simple voltage divider across the output capacitor shown here. Supply smoothing capacitors also need to be added. The 10pF capacitor across the feedback resistor plus 390pF from input base to earth as in my mosfet designs should be included, and estimates for the resistor and capacitor between the two input transistor collectors are 330R and 220pF. A Zobel network at the output consisting of 2R7 plus 0.47uF is suggested in the I.C. data sheets, which should be consulted for details of how to use the mute input and other design information.