The Classroom

Part 3: The Linear Power Supply

Bob Harper

Issue 10, April 2018

For almost as long as there has been electricity, there have been linear power supplies – but sometimes one is not enough.

Some projects require more than one voltage, and sometimes the second voltage has to be negative with respect to the first. One reason for this is that voltages around 0V are impossible to control if that is also your most negative voltage. Instrumentation and audio amplifiers that amplify true AC voltages also require current flow in both directions, requiring a true dual polarity supply.

One solution is to buy two commercial power supplies and, after making sure they are both independent of the real Earth, connect them together with a common voltage.

Using two supplies allows the experimenter to see what happens if the two voltages change independent of one another, as might happen to a battery powered device when the battery on one side of the supply goes flat before the other side.

So while a dual fixed-voltage linear-regulated supply is good for production, sometimes a lab supply should be able to have the voltages controlled independently; and at other times, controlled together.

This month we look at some of these requirements, and the basis of a project to do it all.


Although an operational amplifier (op-amp) can be used with a single supply voltage, some projects require the op-amp to be supplied by two voltages: one above ground and one below ground (i.e., a negative voltage).

power supplies connected together
Circuit is useful for op-amps and devices requiring +/- 12VDC

One method is to use two 9V batteries, connected with one positive terminal to ground and the other battery negative terminal to ground. Together, the two batteries will produce 18V across the two batteries, and supply the op-amp with +/-9VDC to ground, for portable use. [1] shows how you might use two ordinary power supplies connected together, to form a dual rail supply.

The two voltages should be equal, but of opposite polarity. To achieve this using linear regulators, IC makers have both positive and negative regulators such as the LM7805/LM7905 complimentary devices. The extensive series includes +/- 9V, +/-12V, +/-15V, +/-18V, and even +/-24V. You are probably more familiar with the LM7805, which has a partner, LM7905.

The first two letters denote the manufacturer’s preferences, such as “LM”, standing for “linear monolithic”. Other devices may use µA, ON, etc. The numbers all share the format 78xx, positive voltage regulator, or 79xx, negative voltage regulator, where “xx” represents the voltage of the devices, LM7818 would be an 18V positive regulator, for example.

The circuit shown here [2] is useful for op-amps and devices requiring +/- 12VDC, in a minimal parts count project that can be added to if required. An LED and ballast resistor could be placed across the input of the LM7812 to the input of the LM7912 to indicate raw power in, or two LEDs and ballast resistors added across either output to indicate that power output is available.

The regulators have internal over-current and over-temperature protection, so no need to add more protection to the circuit; although a 2A fuse might be added in each of the AC inputs, not ground though! It all depends on what you use the circuit for, and what you are really protecting.


Our next step toward a useful lab power supply is a dual polarity supply with both voltages variable. We will use a complimentary pair of regulators: the LM317 and LM 337. Datasheets are readily available online, and easily found.

The first point to note about the two regulators is the connections are not the same. Although the people who use these regulators could probably agree on which pins should be which, very likely using the LM7805 as a likely model, the designers apparently had a compelling reason for not making them the same, and not making them symmetrical, mirror image or, well anything that could be argued as designed on purpose. Go figure!

The figure below shows the pin-out of the LM317 and LM337. The LM317 pins are, in order left to right as viewed looking at the labelled face: Adjustment pin, Output pin and Input pin. There are reasonable design considerations to make this arrangement sensible. The output transistor possibly being connected to both the centre pin, and the heat sink mount would arguably provide the best heat dissipation.

pin-out of the LM317 and LM337

Why then would the LM337 have the pins as Adjustment, Input, Output? Maybe somebody can make an argument for it, but I find it to be poor engineering design. It’s like letting the artists take over. Instead of making a mirror image PCB for a negative polarity version of the positive PCB, the negative side of the PCB has to be designed separately. Apart from that, the PCB is straight forward. Two separate pots can be used to adjust each polarity, or a dual ganged pot could be wired to adjust both pots together, so long as you understand the voltages will only match as well as the dual gang pot you use.

A sneaky tweak that has been used in some quite expensive old power supplies, is to place a smaller pot, one tenth of the resistance of the main pots, between the two (ganged) pots with the wiper going to ground. It is usually labelled as the “Fine Volt Adjust”, or “Balance Adjust”, or similar.

Here we have one variation [4] with the trim pot of 100Ω connecting the wipers of the two voltage pots to the common ground. Whether the two voltage pots are ganged or not, the Fine Balance pot should allow enough adjustment to balance the voltages of either polarity to the same value.

trim pot

Adjusting the pot has the effect of raising the voltage on one polarity while dropping the voltage on the other. A little tweaking and you have matched voltages. A nuisance for a lab technician to have to use, but a simple cheap trick for hobbyists.


The circuit [5] has the same regulation circuit as before, but with the addition of an op-amp, which does the tracking so only one adjustment potentiometer is required. Of course, a second pot and a switch can be added to allow separate adjustments, if required.

regulation circuit


The transformer supplies a center-tapped AC source to the input connector, which is shown as screw connectors; but any form of suitably insulated and current rated connection can be used. In the simplest method, the wires from the transformer could be soldered directly to the PCB.

While the direct approach saves on parts, and reduces the risk of a poor connection, or of a connection coming loose, it does become an issue for maintenance. A PCB can only be “unsoldered” and “resoldered” so many times before the PCB copper track itself de-laminates from the board. For beginners, soldering it on the first time can have a notable failure rate.

The ground goes directly to the output ground, which is not only a reliable reference point, but helps ensure that the voltage at that point is known. While it is not always connected directly to Earth, the electrical distribution “ground”, it is often connected via a resistor in the kiloOhm range, or via a 600V 1µF capacitor, to help ground any RF, or even audio interference, that finds its way into the circuit.

Most instrument technicians and even electricians have stories about being bitten by a supposedly grounded wire. So it is good practice to follow Mulder’s rule: “trust no-one”.

Each side of the centre-tapped winding passes via a diode bridge to the raw-positive and raw-negative rails. Two of the diodes connected cathode-to-cathode, to the positive rail, and the remaining two connected anode-to-anode, to the negative rail.

Current passes in the direction of the arrowhead in each diode symbol, and if the two polarity loads are equal (i.e., taking equal load current), no current will pass down the ground track. However, if one load becomes open-circuit, the load current of the remaining load will pass down the ground track. Therefore the three tracks should all be made suitable for full load current of the power supply.

The diodes cause a 0.6V drop that must be subtracted from the peak voltage of the transformer winding, one diode forward voltage per side.

The transformer windings also have resistance, 0.2Ω according to my multimeter, but I don’t currently have a suitable low resistanceΩmeter. However, at 0.2Ωs (estimated?) and 1.5A maximum current, there will be 0.3V drop per winding.

The transformer I brought home from Jaycar is Toroidal, with a high efficiency, 160VA rating, which means 160 Volt-Amps, NOT Watts. Transformers and AC motors may cite a maximum power rating in Watts, but the VA rating is the more important of the two.

The Vrms for my transformer is measured at 12.48VAC on a given day at a given location (i.e., the voltage depends on factors I cannot control, but mostly the voltage at my GPO at any given time of day). Therefore I can expect 160VA/12.48VAC = 12.82Amps total current from the two windings. This is much more than the 1.5A per side I want, but it has plenty in reserve for other additions to my proposed workbench power supply.

It is a common mistake of hobbyists to calculate the minimum component for a job, and try to make it do more than it is designed for. A different transformer, having half the capacity, probably costs just a few dollars less.

This transformer will provide +/-12.48 Vrms, or 17.65Vpeak, minus 0.6V drop across the diode, or ~17VDC with full smoothing (or ~34VDC if both polarities are used in series for a higher voltage supply).


C1 and C2 are electrolytic capacitors, wired across the positive supply, and negative supply respectively. Their job is to reduce the ripple to a value that can be controlled by the regulators without causing the ripple to appear on the output. This is deceptively easy on no load, so the load current is one important factor that determines the required capacitance. For the sake of this exercise, the maximum current can be taken as 1.5A.

The frequency also determines the capacitance requirement, as does the maximum allowable ripple voltage. For a full wave power supply on a 50Hz system, “time” will be a maximum of one half cycle, half x 1/50Hz = 10mS.

Some engineers prefer to use the ripple frequency, which for a full wave rectified sine wave is 2 x f, or 100Hz in Australia and other 50Hz countries.

Finally a low ripple voltage allows a higher usable output voltage. Let’s hope for 1V ripple, and see how the calculations go.

From an earlier column, the capacitance is determined by the formula:

C = It/V

where “C” is the capacitance in Farads, “t” is the time in seconds, and “V” is the maximum acceptable ripple voltage.

Therefore: C = 1.5A x 0.01s /1V = 15,000µF, or 15mF.

Note: The term mF is rarely used; however, before the term microFarad was accepted, the term mmF (milli-milli-Farad) was used, even on capacitor cases, rather than accept the metric ‘µF’.

Those who prefer using the ripple frequency would calculate this using:

C = I/2fV, = 1.5/(2 x 50 x 1) = 15mF.

My needs are not for a 1.5A supply, nor for the full available voltage, and as the capacitor requires some time to charge, the 10mS period will be somewhat less anyway, even at full current. So although 15,000µF is calculated as optimum, I have used 4700µF in my supply, and listed 2200µF as suitable for this circuit for op-amp experiments.

One of my past power supplies had been built using 10 x 10,000µF wired in parallel for an amateur radio transceiver, so large capacitance banks are possible. My suggestion is to leave some space around them and make sure you have some airflow for ventilation to cool the capacitors. They do heat up under load and if they go over their recommended thermal limits (i.e., temperature), they may vent electrolyte – sometimes even, explosively.

It also helps to have a small resistance in series with each capacitor to encourage them to share the current. Ten @ 1Ω resistors effectively in parallel is only 0.1Ω, and it helps smooth the ripple as well.

C3 and C4, with C7 and C8 protect the regulator ICs from spikes and EMI noise that might be amplified by the regulators internal circuitry; they help keep the circuit immune to interference.

The two regulators have different pinouts, so don’t get caught copying the top-side to the bottom-side of the PCB. Each IC has an Input, an Output, and an Adjustment pin, which easily connects to the Input from the transformer, and the Output to the load terminals.

We have also allowed for D5 and D6, a diode from output back to input of each polarity IC, to protect the ICs from Inductive loads, and sudden loss of input power. They are optional but cheap insurance, just note their polarity and orientation.

D9 and D10 have a similar purpose, this time protecting the Adjustment pins of each IC from excessive and/or reverse voltages. This is unlikely, but also cheap to include.


IC1, an LM317, is adjusted by a 2 potentiometer, although a 1kΩ can be used, as long as the Adj. Pin can get 50µA min from the Voltage Divider without loading the bias circuit, therefore causing voltage changes.

Assuming 15V is the required maximum voltage out, which I will probably not achieve from my choice of transformer, since the regulator head voltage can be 2.5V at full load. The calculated current through the 2kΩ pot will be I = V/R = (15-1.25)/2,000 = 6.875mA, a little lower than 10mA recommended; and R1 will then be R = V/I = 1.25V/6.875mA = 182Ω. So we’ll use 180Ω for R1, and 2,000Ω for VR1.

For a dual variable supply without tracking, we’d use the same values for R2 and VR2, and keep R2 = 180Ωs for the tracking version as well. In the circuit we have presented, a switch, SW1 can be added to switch between dual output controls and tracking controls.

SW1 should not be changed while the load is connected as the output could spike to maximum.


Completing the circuit, without tracking for now, C7 and C8 are often included in power supplies as a kind of final filter. First they help to further smooth the voltage to the load, but they also help remove load fluctuations from the supply circuit. They can be much larger, 1000µF or more, but we used only 100µF, and suggest they are not required if the load is constant and without load ripple.


That leaves us with the tracking amplifier using an LM741, other op-amps can be used instead, such as the TL071, LF351 etc. The LM741 has a capacitor on each supply pin, 10µF each. The only other components required are three resistors: R3, R4 and R5.

R3 connects the LM741 non-inverting input, pin 3, to ground as a reference. Ideally R3 should be equal to R4 and R5 in parallel, which makes it 5kΩ as we used 10kΩ for R4 and R5. The circuit shows R3 as a 5k1, which is the nearest standard value available.

R4 and R5 are in series across the +ve and -ve outputs, and if they are exactly equal in resistance, and the two outputs are exactly equal in voltage, but of opposite polarity, then the centre connection between R3 and R4 would present 0V to the inverting input of the LM741; and it’s output, if it were not connected to R2, would be 0V.

Connected to R2, as it is, the output of the LM741 would only be 0V if the output of both regulators were 1.25V of opposite polarity.

If the +ve voltage were 6.25V, meaning the voltage on the adjustment pin of the LM317 were 5V, then any voltage less than -6.25V at the -ve output would cause R4 and R5 to make the inverting pin of the LM741 to be more positive than 0V.

The output of the LM741 would become more negative, because of the inverting input, until the output of the LM741 became -5V, thus causing the LM337 to generate -6.25V at the -ve output. At that moment, the inverting and non-inverting pins would be equal, and the output of the LM741 would be stable.

If load changes caused the LM337 output voltage to become more negative, the inverting input of the LM741 would become too negative, causing the LM741 output to become more positive, until again, +ve output is equal to -ve output.

Changing the voltage on the LM317 by adjusting VR1, will cause the LM337 to follow along, tracking the LM317 with a negative output of the same voltage.


While the aim of The Classroom article series is to help you understand how the electronics works, we also want to share enough information so you can apply the techniques yourself. You should be aware however, that commercial power supplies will often have unexplained circuitry, that at times seems to be there to do nothing more than complicate the circuit.

I trust the added circuitry makes some contribution, under certain circumstances, but commercial circuits do contain more parts than hobby circuits, most of the time. For example, a capacitor across VR1 will make the voltage across VR1 more stable, but when making adjustments to VR1, the capacitance will allow a very small delay measured in milliseconds that allows the LM741 to “keep up”. Sometimes a basic, and cheap power supply, can be modified to do more than intended. Some power supplies were built out of battery chargers, but modern chargers are different beasts.

There are often mods for commercial gear to help the gear perform to higher standards, with better accuracy, control, or power handling. Sometimes the mods are there to allow a commercial piece of equipment to be used for other purposes. Just be sure the mod is well documented, and that you understand what is being done and why.

We have not produced a project of this power supply, simply because there are plenty of cheap commercial supplies already out there. By the time you gather a case, transformer, PCB full of components, voltmeter and ammeter, terminals and wiring, switches and knobs, and no doubt other costs I have not thought of, you will have spent more than a reasonable power supply might cost to buy off the shelf. However, if you need a special power supply, for a specific purpose, and you cannot simply pick one up on your way home, it’s nice to know that you can, not only build something that works, but also diagnose faults and fix them, or extend it’s functionality, without losing control.

Part One

Part Two