The LM301A is an error amplifier and integrator. It measures the difference error and adds integral feedback which reduces the steady state error to near zero.
The idea is to measure the differential voltage across the LT1083 device and develop an error signal to feed to the driver and reduce the controlled voltage error to zero (the voltage across the LT1083).
The lower LT1011 is the gated driver.
The upper LT1011 syncs the 2x line frequency to the driver so it turns on the SCR's at the right time relative to the line voltage phase. It is basically a zero cross detector plus a ramp generator, so it generates a ramp for each zero crossing of the input line.
The right side LT1004-1.2 is a reference diode that allows output adjustment down lower than with a direct connection to ground.
The center LT1004-1.2 along with the 1N914 diode creates a reference subtractor which subtracts about 2v from the input voltage of the LT1083, and is the upper left arm of a balanced bridge (the two 16k resistors are the lower arms of the bridge). The LT1083 is the upper right arm of the bridge, so the voltage across the LT1083 is compared to the voltage across the center LT1004-1.2 plus diode voltage, and the difference amplifier integrates the difference to develop an error voltage. The error voltage (with bias) is then fed to the driver which generates a low output when the ramp voltage becomes greater than the error voltage which of course means the output at pin 7 timing depends on the error voltage and the timing of the ramp. That low output couples through the pulse transformer T2 to turn the SCR on. The 1uf cap across the 560 ohm resistor provides a higher current gate surge to drive the gate with a higher current pulse to turn it on. The action of the error voltage and the ramp make up a pulse width modulated scheme where only the turn on point has to be controlled.
The pulse widths at the output of the SCR's get wider until the voltage across the LT1083 equals the voltage across the center LT1004-1.2 plus diode voltage.
MrAl, Good job. That circuit is complicated and net easy to understand. It shows that a 3-pin regulator can work well with a pre-regulator.
Now days I would replace the SCR section with a PWM. The idea of keeping 3 volts across the 3-pin regulator is good. I save some commercial power supplies that have a PWM and linear regulator working in series. That way you get the best of both worlds.
The LT1083 still uses it's own internal bandgap reference voltage to generate it's regulated output. The LT1004 can not improve that, all it does is shift the "reference point" of the LT1083 down by about 1.2V (to 1.2V below ground) which makes the LT1083 able to bring it's output down to 0V.
Bounty:
You know what? I think you're right. Thanks for catching that. I misinterpreted the connection of the ref diode. However, it is possible to improve the temperature characteristic with an external highly stable reference diode. That's just not the required connection for doing that so i removed it from the detailed circuit explanation i provided back a few posts. All fixed