Adjustable DC Regulated Power Supply (0 - 30 V 2 mA - 3 A)With Digital Control

[Sideshow Bob]

in my opinion there is a flaw in your design:
you are measuring the total current with R3, including the currents that are used by the LM358 / LM393 und the current through Q1.

Yeah it would be better to move R3 on the high side. It can be inserted between  LM338 pin 2 and and R1. This way the regulator will compensate for any voltage drop in the current sense resistor. However also this will introduce some errors. As it will pick up the current that flow into LM338 pin 1 and R13//RV1

[vern]

you can keep R3 on the low side, you just have to move it to the output so that it is only measuring the output current.
Of course you have to change the circuitry for the voltage regulation to compensate for the voltage drop.

[PM3295]

Yes, there will be an error offset current of about 11 mA coming mainly from the regulator idling current. I don't see this as a problem since most of the time such a 5 A capable power supply will be used and operated at higher currents where the 10 mA error can be ignored. The 120R (R1) min load resistor value can be increased to 220R, and that will drop the error current by 6 mA.  It is usually recommended to use a value that will ensure the 10 mA (worst-case) loading current (data sheet) for the LM338. If you use the higher spec LM138, it will have a worst-case of only 5 mA. It is meant to be a simple and cheap power supply, not precision type. The error can easily be subtracted with a little more effort and complexity.

Other changes to reduce the error current:
1) R1 - 220R
2) R2 -27k (need 1 mA min)
3) R10/11 - 10k or more
4) RV2 - 10k or more
5) R8 - 6.8k
6) LM138 ( 5 mA )
7) Use TLV431 (0.1 mA) with R2 =270k, but need to scale R10/R11 ~ 3; 33k/11k

Alternative to LM358 (LM7332 with R-R output)

[Vineyards]

Makes sense but ideally, a device must be versatile enough to operate in a broad range of conditions. For example, many DC-to-DC converters cannot be stable if the load current is under a certain limit. We need to add a resistor to let them always face a tiny amount of current. Without these little tricks here and there they would not last very long.

[PM3295]

By adding a separate negative regulated voltage source of -5 V @ 20 mA, we can make it adjust down to zero. It also fixes the near-zero error of the opamp, and compensates out the steady error reading (adjusting P2) caused by the idling current of the regulator and other parts. Now the read-back of the true load current is linear down to zero.

It shows using the LM317 with current bypass transistor, but will work the same with the original LM338 without any bypass.

This auxiliary supply can be obtained from a small isolated DC-DC converter, connected to the 12 V rail, or from the usual capacitor coupled, charge pump used by other designs.  https://www.mouser.com/datasheet/2/670/pyse1_d-2821163.pdf

The DC-DC modules are not that tightly regulated (10% - load dependent), so you will need something with a 9V output and then add a small 5 V regulator, (LM78L05 or LM79L05).  

[Vineyards]

Charge pump devices will fall short of satisfying any sort of current requirement other than that of a buffer stage anyway. With DC-DC converters there are regulated and unregulated ones. TI raves about one of theirs like this:

It is increasingly difficult for DC/DC converters to achieve tight output voltage tolerances driven by smaller
process geometries for new FPGAs, DSPs, and ASICs. As process technology advances, the performance
processors require tighter voltage accuracy and lower core voltages. Multiple factors need to be taken into
consideration when determining the output accuracy of a DC/DC converter. Most engineers stop at the
initial accuracy on the front page of the datasheet of the DC/DC converter. However, designers must
consider the tolerance of the voltage feedback resistors used by the following:
• DC/DC converter
• The ratio of the resistor divider
• Routing distance and trace losses of the circuit board
• Input voltage variations
• Voltage ripple
• Temperature swings
• Load transients
Additionally, many designers will want headroom or margin to make sure the solution is always within the
tolerance expectation of the processor. The TPS546D24A was developed to help designers achieve
higher output voltage accuracy by actually specifying the output voltage accuracy, rather than the initial,
reference, or VFB accuracy.

[PM3295]

I don't think that the 20 mA requirement is too much for a charge pump. You will need at least 100 uF caps to do the job and 5 V regulator.

One important requirement of this type of charge pump connection, is that you need to have some load current on the output of the bridge, in excess of the maximum aux load current needed. It is a good practice to make this min current at least twice (40 mA in this case) of what the pump output should provide. Below it can be seen that if we sweep the load current (IG1) up from zero, you only get proper operation around 40 mA. With 0 mA load current, there is insufficient output.

Using dual diode full wave, with center-tap transformer you won't have this min current limitation. 

[sadman]

here is some idea's from my side on LM317 CV/CC power supply

sadman

[PM3295]

Using the cheap LM358 at high gain (x100) will be a problem because of the input offset voltage spec of this part. This figure can be between 0.3 - 3 mV. So you may get a part giving an output error voltage between 30 - 300 mV. To help cancel this out, you may need one of the more expensive zero-drift opamps that is available. This particular model from National comes with about 1.7 mV programmed in.

[Sideshow Bob]

Using the cheap LM358 at high gain (x100) will be a problem because of the input offset voltage spec of this part. This figure can be between 0.3 - 3 mV. So you may get a part giving an output error voltage between 30 - 300 mV. To help cancel this out, you may need one of the more expensive zero-drift opamps that is available. This particular model from National comes with about 1.7 mV programmed in.

How often do you need your lab power supply to deliver below say .3 volt?

[PM3295]

The problem is with the current sensing circuit, not the power-supply  output voltage. With no load current, it may sense it as 172 mA. This reading can be as low as 30 mA or as high as 300 mA depending on the manufacturing spread on the part you get.

[sadman]

Using the cheap LM358 at high gain (x100) will be a problem because of the input offset voltage spec of this part. This figure can be between 0.3 - 3 mV. So you may get a part giving an output error voltage between 30 - 300 mV. To help cancel this out, you may need one of the more expensive zero-drift opamps that is available. This particular model from National comes with about 1.7 mV programmed in.

there is well optimized LM317 power supply schematic in side the power board zip file also look at that

[sadman]

Please, see: https://www.youtube.com/watch?v=2w4cR8D8WJ4

here is Elektor answer to this video

https://www.elektormagazine.com/news/0-30v-ua723-based-regulated-power-supply