Here I will try to explain some aspects of Douglas Self’s amplifier design briefly.
The circuit diagram can be seen below (I will have to assume that Self will not be unduly mad at me making the design publically available, as he has already done so himself in this article)
As we can see the design is a three stage architecture, with a negative feedback loop from the output back to the first stage.
- Input stage
The input stage is a differential gain amplifier, which handles the subtraction of the feedback network signal from the input signal. The amplifier is a standard differential pair with a current source at the tail and a current mirror active load which converts the output to a single ended signal for input to the next stage. The feedback network is designed to roll off for higher frequencies to avoid HF instability. - Gain stage
The central stage handles the voltage gain of the amplifier. A darlington configuration is used to make the gain higher, and some positive feedback compensation in the form of a miller capacitance is used. - Output stage
The output is used to buffer the output signal from the gain stage, and must be capable of driving the large currents required for an 8 ohm speaker load. The stage utilizes a double emitter follower configuration to do this.
I will discuss some peculiarities of the design below.
Switch-off distortion
Switch-off distortion is a phenomenon that occurs when a transistor has its base driven from a high impedance source, and is suddenly pushed to a lower base voltage. In the ideal case, the transistor would react immediately to this change. However, there will be some charge left over in the base region of the transistor, that has nowhere to go except into the transistor channel and through the emitter. The consequence is that the transistor keeps on conducting for a while even after being switched off.
The remedy for this is to provide an easy discharge path for the base charge. This can be seen in the schematic in the form of resistor R9 (in the darlington pair), and in the decoupling of R15 with capacitor C6 in the output stage.
Output stage biasing
In order to have a properly class-B biased output stage, there needs to be a voltage differential between the output stage inputs of four diode junction voltages, at DC (four is due to the fact that double emitter followers are used). One way of doing this would of course be to just put four diodes in series between the input terminals inside the gain stage. A better way however, is the one that Self uses here. The circuit is called a Vbe-multiplier and does exactly that. The Vbe of the transistor TR13 in the gain stage is multiplied using the surrounding resistances so that the input points at R12 and R24 are exactly 4Vbe apart at DC. The whole biasing circuit is then decoupled using capacitor C4 so that at AC conditions, a slight amplitude change from 0 will turn on one side of the output stage.
Another important aspect of this biasing network is that it is temperature stable. A temperature change in the output transistors will change their base-emitter junction voltage slightly. If the bias network remains constant, this would either mean more crossover distortion, or more static power dissipation (depending on which way the temperature is changing). By using a multiplier that includes the actual Vbe voltage of a transistor, these temperature changes will be compensated for automatically (as long as the bias transistor has the same temperature and similar tempco of the output transistors).
Zobel network
The damped inductor and the RC circuit to ground on the output is called a Zobel network. The circuit is used to isolate the feedback network of the amplifier from the inductive load that the loudspeaker coil represents, making the speaker look more like a resistive load. This is done to keep the amplifier stable. For more reading, see this link.
Decoupling capacitors
C7-10 are all decoupling capacitors, where C7 and 9 are to be placed near the small signal part of the design, while C8 and 10 (which are larger) will stabilize the supply near the output transistors.
A note of some importance here is that the ground return for these capacitors should be separated from the signal grounding network (because the charging pulses of current can affect the signal network otherwise). Therefore C7-10 have a separate reservoir ground return connection which is to be connected to the power supply at a grounding point which is closer than the star point to the transformer ground.