DIY Discrete and Simple Voltage Regulator

Some time ago I was playing around quite a bit with vintage audio amplifiers/receivers, and in many of them I was improving the power supply portion for the low current differential amplifier stages. This was always a simple and cheap task, that proved well worth the time when it came to sound. In a desire to “do this differently”, I didn’t want to use an IC for this, but rather wanted to go with a discrete yet simple design. The circuit I came up with was very well suited for such applications, and I therefore decided it would be a good idea to make an independent regulator PCB out of it for general use in audio stuff I build. At the time I also had limited experience with PCB design, so this seemed like a great project to start with. There’s no better way to learn than simply giving it a try.

Fig. 1. Toli’s Minimal Voltage Regulator Schematic

The simplest form of the regulator is shown in Fig. 1. These are older schematics I drew with TinyCAD before using KiCAD, and are therefore not as nice looking as schematics on my more recent circuits. The circuit is quite straight forward to follow, but its different than most voltage regulators. Unlike most regulators where an explicit voltage reference is available and then buffered by an error amplifier and pass transistors, this circuit uses a different mechanism. The “reference” voltage in this case is the VGS needed for M3 to pass the current provided by J2. All the components to the left of J2 are no more than a full wave rectifier and a bulk filter capacitor. J2 in this case operates in saturation and therefore acts as a constant-current-source (CCS). This current charges the gate node of M2, which in turn pulls the output node high. This causes the gate node of M3 to rise and M3 conducts the current of J2 into the ground node. There equilibrium state is when the current of M3 is equal to that of J2. Therefore, the output voltage can be described as:


This (regulator portion, without the rectifier) is a circuit that I have used more than once as local regulators inside vintage amplifiers. It can be build on a small prefboard and placed close to the point of load. The fact its exact output DC value isn’t determined before measurement (due to VTH variations of M3), is of low significance in audio applications. However, it is a fairly good regulator, with very low noise.

However, when using it as a standalone regulator line regulation is much more significant. One part of the circuit that degrades line regulation in this circuit is the sensitivity of J2 to voltage changes over it. Therefore, to improve this, one can modify the circuit somewhat to that of Fig. 2. In this circuit, J2 is no longer operated directly from the rectified voltage. Instead D5 zener is used to provide J2 with a clean supply. J1 is used again as a CCS to limit current variations through the zener. J1 must obviously support the current of J2 and the zener diode. Since we now have an additional voltage that is somewhat higher than the gate of M2, we can use it to control the gate of an additional NMOS, M1. It can act as a cascode for M2, limiting VDS variations over it, and improving line regulation further. This modified circuit obviously comes at the cost of increased headroom requirement of the regulator (Vin-Vout needed for proper operation).

Fig. 2. Improved Line Regulation

A possible trade-off between the two (high headroom, and good AC line regulation) can be made by replacing these additional devices with a simple low-pass-filter (LPF), as in Fig. 3.

Fig. 3. LPF for Reduced 100Hz/120Hz Ripple

However, I wanted to see how far I could push this basic circuit by extending the circuitry around it to give extra functionality. Some of the things I was hoping to achieve were reduced headroom requirements without sacrificing performance, as well as some basic form of current limit. I prefer not to build circuits that don’t have some for of current limit included, simply as a measure of minimizing damage in case of a problem somewhere. After a couple of iterations I’ve arrived at the circuit of Fig. 4, which I’ve favorably named “ToliReg” 🙂

Fig. 4. Final Version of the Voltage Regulator (ToliReg)

This circuit has quite a bit more parts, and doesn’t really follow the original intent of “keeping it as simple as possible”, but it does offer quite a bit more functionality. Device designation has been changed from the first schematic, but its still easy to recognize the same mechanism the sets the output voltage. M1 is now the feedback device (error amplifier), and M2 the pass transistor. Raa has been added as an optional trim for output voltage to make it less sensitive to parameters of M1. The CCS is now implemented using CRD1 (which can still be a JFET, which is exactly what a CRD is). However, to have a lower headroom requirement, this current is not sourced directly, but rather through a current mirror made up of Q5/Q6. Q7 is used as a cascode for Q6, to limit thermal differences between Q5/Q6. D6 is only there to provide the bias for Q7’s base.
To provide a quite supply for this biasing circuit,a LPF (R1 and C4) is used, buffered by Q4 as a capacitance multiplier. D5/C3 act as a ‘peak hold’ circuit, which is a very effective addition. Using this topology, significant voltage drop over the bulk capacitor is allowed without affecting the operation of the error amplifier, which in turn means better regulation even with higher output current.
Finally, a current limiting feature is added around Rsense. This resistor is placed at the  drain of M2, so it won’t affect output impedance of the regulator. As the voltage over this resistor rises to ~0.6V, Q1 will conduct. This in turn will make Q2/Q3 active. Q2 will pull the gate of M2 lower and will limit the output voltage (and current). Q3 is optional and can be used to operate a LED for visual notification of the current limit condition.

Fig. 5. ToliReg V3 Board Layout

I have also designed a dual rail version of the same regulator, to use in my own projects. I have made some noise measurement of the regulator (on an earlier version of the circuit board as shown in the figure below. The noise measurement was done by using a LNMP from tangent (see link for more info), with a 100KHz -3dB BW. The total integrated noise at the output of the regulator set to 24V was measured at 10uVrms.

Fig. 6. PCB Used for Noise Measurement

As with some of the other projects I  did back at the day, I have organized most of the needed information to build one of these in a PDF file to post in a couple of forums. I’m attaching this file here too, with the schematic, BOM, and a few additional notes about the circuit.


Leave a Reply

Your email address will not be published. Required fields are marked *