If you want a linear PSU, use a
LM723.
I posted a LTspice sim of the regulator section of a Philips PE 1535-00 0V - 40V, 0A - 0.5A. single output PSU,
[here]. Given a big enough heatsink, it would be fairly easy to extend to 5A:
* To beef up the pass transistor to handle 5A, add three 2N3055 transistors B and C paralleled with 0.1R individual emitter resistors for current sharing, to V26, the BDY20 to make a Darlington pair, with an extra 47R resistor from the 2N3055 bases to the lower end of the emitter resistors.
*Replace V111, V112 with fast >100V, >7A diodes.
*Decrease R137 to 0.22R to increase the current limit to 5A.
Caution: if the peak input voltage is >45V, increase the Zener voltage of V109. It needs to be Vin_peak-16V, or a few volts more. If you use a 50Hz or 60Hz transformer and a bridge rectifier to feed it, you will probably need a preregulator as the PSU cant tolerate more than about a 12V swing between the peak unloaded input voltage and the troughs of the ripple at full load. Another pair of paralleled 2N3055 transistors should do the job in a Darlington capacitance multiplier circuit with a >1A driver transistor, with the voltage across the capacitor clamped to 45V by a chain of 15V zeners.
The heatsink needs to be *MASSIVE* and probably fan cooled, as worst case, with the output shorted it will be dissipating 225W + whatever the preregulator dissipation is. e.g. with a 44V 8.33A secondary feeding it, at 10% high mains voltage that could be another 115W.
A better option would be to find a used 48V, 300W or greater telecoms SMPSU, and either hack it to drop the output voltage to a fixed 38V, or trim it down to 45V and use it to feed the modded Philips ciruit I have just described if you need a fully variable output.