ESP Logo
 Elliott Sound Products Project 217 

'Practice' Power Amplifier

© July 2021, Rod Elliott


In the case of this project, 'practice' has a different meaning from what readers may assume.  Practice amps are common for guitarists so they can practice without incurring the wrath of other householders and/ or the neighbours.  This is not that kind of practice amplifier!  In this case, it's an amp that beginners (or advanced hobbyists) can build to practice their construction skills, and learn many of the basics of how amplifiers work along the way.  Many constructors have never truly analysed the circuitry of projects they build, and if something goes wrong, they don't know where to look, and don't know what to look for.

This may seem like a very long article for such a simple project, but that's because I've included far more information than normal.  This project is designed so that the constructor has the opportunity (and is strongly encouraged) to experiment, take measurements, understand how the amplifier works, and examine the function of each section.  No PCB will ever be offered, for the simple reason that it would not provide any incentive for troubleshooting/ fault-finding.

The amp is not designed for high power, and it will normally provide around 6W output with a 24V supply, increasing to 20W with a 40V supply (the maximum recommended).  All transistors can be substituted with equivalent devices, taking note of voltage rating, maximum current and power dissipation.  The output transistors will dissipate less than 8W with a 40V supply, but peak dissipation may exceed 20W with a speaker load.  This is not a challenge for any transistor rated for 50W or more, making budget devices like the TIP/ MJE3055/ 2955 ideal.  The drivers are in the TO92 package, and no heatsink is needed for them.  Naturally, the power transistors need a heatsink, but nothing fancy or expensive.  Even a flat sheet of aluminium of around 50mm² will be enough with a 24V supply (although that is marginal).

There are many modifications that can be made to improve low-frequency response or change the gain or just for the sake of experimentation.  This is an amplifier that's designed to be modified and played with, so you get a better understanding of power amplifiers and how they work.  I've included a 'how it works' section and details of the design process.  While many amps use different topologies, the principles are not changed.  The input transistor is configured with current feedback rather than the (now) far more common voltage feedback.  For an experimental amplifier this has one major advantage, in that the amp has a very good phase margin, and it is unlikely to oscillate unless you do something silly.

Although the recommended supply voltage is 24V, the amp operates more-or-less normally with as little as 5V.  Power output is only around 50mW, and the input bias network has to be changed to get that.  However, this shows the flexibility of the circuit.  While it could easily be scaled up to provide 100W or more, most people have objections to using an output capacitor in a power amp, and it's hard to justify.  One (and probably the only) advantage is that if the amplifier fails 'DC' (output device short-circuit) it can't blow up your speakers.

A few specifications are in order, but they are relatively unimportant for this design.  I measured distortion at 0.1% at full power into an 8Ω load, and 0.03% with no load.  Full power bandwidth extends to 80kHz, and output noise measured less than 3mV using a switchmode power supply, and everything just lying on my workbench.  With a linear supply and a shielded enclosure, I'd expect output noise to be less than 1mV (-70dB referred to 1W output).

Circuit Description

The design uses no expensive components, and uses a single supply of between 12V and 40V (up to +48V is possible).  Output power is modest by design, and it uses the minimum number of parts possible while still giving good results.  The general design used to be very common, and was used in countless low power audio power amps before 'someone' decided that we all needed (much) more power.  One downside of many of these early capacitor-coupled amps is that there's a 'thump' at power-on, but it never caused anyone any problems.  Because power was limited anyway, the thump created as C5 charged could never damage a speaker.  Turn-on thump is audible but minimal with this design, and it reaches only 1V peak with a 24V supply.

Figure 1
Figure 1 - Practice Amplifier Schematic

This is a fairly thoroughly re-engineered take on the early designs.  All components are readily available at low cost, especially the silicon.  There's also a 'twist' in the design of the output stage that removes the requirement for a bias trimpot and thermal compensation to prevent thermal runaway.  The output transistor emitter resistors are deliberately a high value, so bias current is always a 'sensible' value.  Even quite drastic temperature changes for the output transistors won't affect it by very much.  Normally these high values would cause a significant loss of power, so they're bypassed with diodes.

Once the emitter current exceeds around 250mA, the diodes conduct and effectively bypass the resistance.  This allows the amp to deliver full current to the load.  While you may expect this to introduce distortion, it doesn't (well it does, but it's minimal).  The small amount of distortion is easily corrected by feedback.  Unlike a Class-B output stage, there is always (current) gain in the output stage because the output transistors cannot turn off completely, which is the primary cause of crossover (aka 'notch') distortion.

While not something you'll be able to verify easily (if at all), the circuit shown has an open-loop (zero feedback) gain of over 80dB up to 1kHz, with over 60dB of gain at 20kHz.  This is actually better than many modern power amplifiers that use a long-tailed pair for the input.  Without feedback, the discontinuity caused by the output diodes is visible, but once feedback is applied it disappears completely.  The output diodes will normally be 1N4004 or similar, but if you wish to run the amp with a voltage greater than +30V or use a 4Ω load, I suggest either two in parallel, or use the higher rated 1N5404 or similar.  If you use standard diodes (1N4004 or similar), you may experience some distortion at high frequencies.  Using high-speed diodes prevents that from happening if it's likely to make you uncomfortable.  UF4004 diodes (an 'ultra-fast' version of the 1N4004) or other fast (soft recovery) diodes capable of at least 1A will be perfect.  I used BYV26E diodes as I had them in stock.

How It Works

Q1 is the input transistor, which is configured (via feedback) as an error amplifier.  The AC signal on the base and emitter should be almost equal (typically less than 600µV RMS difference), and any discrepancy is used as an error correction signal for the remainder of the circuit.  The collector of Q1 is direct-coupled to the base of Q2, which provides all of the voltage gain for the amplifier.

Q2 is the 'VAS' - voltage amplifier stage.  The AC signal at the base of Q2 is only about 16mV peak-peak at full output, and is 'pre-distorted' to compensate for the distortion produced by the output stage.  The collector load for Q2 (R7, via D1 to D4) is bootstrapped, using C4 and R6.  C4 ensures that the voltage across R7 remains substantially constant, thereby ensuring a reasonably constant current.  The bootstrap circuit isn't perfect in this role, but it works very well in practice and has been used in countless amplifier designs.  Using an active current source will reduce distortion slightly, but adds several extra parts and reduces the maximum output voltage.

There are two feedback paths, with the primary feedback network comprising of R3, C2 and R4.  R3 provides DC feedback to maintain the output voltage at 0.65V above the DC voltage at the base of Q1.  This is set to half the supply voltage via R1 and R2, and while adjustment is not included, it can be added using a trimpot in place of R2.  The feedback network provides unity gain at DC, because C2 blocks the DC component.  The input impedance is 16k (close enough) because R1 and R2 are in parallel for the input signal.

AC feedback is a combination of the current through R3 and R12, developing a voltage across R4 (100Ω).  By including an additional 6dB of feedback from the speaker side of C5, premature rolloff caused by the low value (only 1,000µF) is mitigated.  This also removes some of the distortion normally associated with electrolytic capacitors (real or imagined).  Due to this secondary feedback, the low frequency response is -3dB at 12.5Hz with an 8Ω load.  This can be extended by using a higher value for C5.

note Note that I've described this amplifier as using current feedback (CFB).  The terminology is a little misleading though, since the feedback node (the emitter of Q1) has an AC and DC voltage present.  However, the emitter has a very low input impedance, and the current through the feedback resistor (R3) determines the bandwidth.  More current provides better high-frequency response.  This is in contrast to a voltage feedback amp (nearly all common opamps for example), where both inputs are high impedance, and the slew-rate (and hence the gain-bandwidth product) is determined by the dominant-pole capacitor, and isn't affected by the feedback network values.  Current feedback opamps are used for high-frequency operation (10MHz to 100MHz) where the CFB topology provides the best performance.  See High Speed Amplifiers in Audio for an in-depth analysis.

Phase compensation is the usual arrangement, using C3 (100pF) as a 'dominant pole'.  This has been included to limit the bandwidth, which is otherwise so great that the amplifier will oscillate at the slightest provocation.  Bandwidth is flat to 80kHz at full output, with a slew rate of about 6V/µs.  While it's easy to extend the bandwidth by reducing the value of C3, this becomes troublesome and the amp may oscillate.  I ran tests where the amp was essentially flat to 400kHz, and it became very sensitive to speaker lead capacitance, and would oscillate if the input was open-circuit.  The smallest capacitance between output and input is enough to cause enough positive feedback to allow oscillation.  As little as 2pF between input and output (with the input open) is enough to cause oscillation.

A side-effect of the wide bandwidth of this amp is the potential for instability with a 'typical' speaker lead and load.  If you have issues, the way to cure this is to add a Zobel network (C7 and R13).  The Zobel network is shown as optional, and with a good layout it won't be necessary.  However, the Zobel network costs peanuts and is always a good idea.  R13 doesn't have to be 2.7Ω as shown - 10Ω is a common value and is used in most ESP amp projects.

The bias current is set by the four 1N4148 diodes (D1 - D4) which provide a nominal bias voltage of 2.8V DC.  This is enough to ensure that the two driver transistors (Q3, Q4) and output transistors (Q5, Q6) conduct with no signal.  The bias current ensures that there is no crossover distortion.  The actual bias (quiescent) current in the output stage will be variable in practice, because transistor base-emitter and diode forward voltages vary from one device to the next, and also with temperature.

The high-value emitter resistors (2.7Ω) ensure that no combination of transistors or bias diodes will cause excessive current.  With typical devices, the output bias current will be between 20mA and 50mA.  To prevent excessive voltage drop at high output current, the diodes (D5 and D6) effectively bypass the resistors, both maximising the available output current and minimising the dissipation in the resistors.  'Conventional' low value resistors (e.g. 100mΩ) could be used, but then a 'bias servo' would be required, adding an extra transistor, a couple of resistors and a trimpot.  Bias stability requires that the bias servo transistor be mounted on the heatsink, which for this design would be an unwanted construction complication.

Because the amp runs from a single supply, the speaker is coupled via a capacitor.  Although it's shown as 1,000µF, this can be increased for improved bass response.  With the values shown, bass is -3dB at 12.5Hz, reduced to 8.7Hz if C5 is increased to 2,200µF.  The bass rolloff is mitigated to some extend by R12, which provides some additional feedback from after C5.  If you expect to drive a 4Ω speaker, the higher value is recommended.  You can increase it further, but it will cost more and it's unlikely to be audible with most programme material.

Design Process

Any power amplifier design starts from the output.  With a known supply voltage and load impedance, this means you can determine the peak output current required.  For this amp, the effective supply voltage is ±12V with a 24V supply, so an 8Ω load will demand 1.5A peak.  From the datasheet, the minimum hFE at 4A is 20, so the base current will be 75mA.  The BC639/ 640 driver transistors have a minimum hFE of 63, so their base current will be 1.2mA.

The VAS (voltage amplifier stage, Q2) needs a current of about twice the base current of the drivers, or 2.4mA.  The bootstrap resistors (R6, R7) will deliver about 2.7mA, so all conditions so far are satisfied.  The gain of Q2 is also a minimum of 63, so its base current will be 43µA, plus 650µA drawn by R5.  The input transistor has to deliver just under 700µA.  The DC voltage across R3 (2.2k) is therefore 1.54V.

The final calculation is for the voltage divider at the base of Q1.  Its base current is equal to the emitter current (700µA) divided by the minimum hFE for a BC559 (110), so 6.4µA.  The current through the voltage divider (R1, R2 and R11) should be at least five times the base current, or 32µA.  With 78k as shown (and a 24V supply) it's over 300µA, which is more than enough to ensure stable performance.  Note that all calculations are based on the 'worst case' (minimum) hFE for all transistors, and real examples will generally be higher.  By using the minima, we ensure that the circuit will perform as expected regardless of transistor gain variations.

The design process described above is pretty much standard for any amplifier design.  By starting from the output, we can be sure that the circuit will work even if the transistors all have the lowest gain specified in the datasheet.  One step that has not been included here is to work out the SOA (safe operating area) of the output and driver transistors.  It's not needed here, because the voltages and currents are all well below the maximum shown in the datasheet.  The datasheets for the MJE3055/ 2955 don't even include a SOA graph!  Presumably this is because the manufacturers know that 'most designers' will be aware of SOA limitations, and will only use these devices at low power.  I never recommend the TO-220 package for anything that dissipates more than 20W.

Even with a supply voltage of 40V and a 4Ω load, the average dissipation is under 10W, with a worst-case peak (reactive load) of 50W.  No changes are necessary if the supply voltage is increased, but it must not exceed 40V.  In theory, it can handle more than 40V, but dissipation in the driver and output transistors will increase, placing them at risk of failure.  Despite the amp's simplicity, it will provide less than 0.1% THD (total harmonic distortion plus noise) into an 8Ω load.  Voltage gain (Av) is determined by the ratio of the feedback resistors, in this case ...

Av = ( R3 || R12 / R4 ) + 1
Av = 1.1k / 100 + 1 = 12

That means that an input voltage of 560mV will give close to full power with a 24V supply.  The gain can be changed (within reason) by altering the value of R4, with a lower value giving higher gain and vice versa.  It's unlikely that this will need to be changed, as it's a fairly 'sensible' gain for a low-power amplifier.  Since this design is specifically intended for experimentation, there's a lot to be said for making changes.  Details of modifications you may wish to try are described next.

One part of the design process that is extremely difficult to calculate is the dominant pole compensation cap (C3).  In most cases it will be far easier to determine the value during testing.  Simulators can help, but the models used must be extremely accurate, but most are 'functional', and can only give an approximation.  The circuit used here was simulated and most of the functionality was easily proven, but the simulator failed to provide accurate phase information.  This is common!

One thing that this project shows with zero ambiguity is the frequency response of electrolytic capacitors.  With a 2,200µF output cap, response was measured to be flat to over 400kHz (with a much smaller dominant pole cap than shown).  This didn't change if the load was connected or disconnected, demonstrating that the response of the capacitor does not intrude on the audio at high frequencies.  A point that I've made countless times is that adding a small plastic film (or snake oil) capacitor in parallel with electros is a completely pointless exercise for audio, and this amp provides an opportunity for you to prove it for yourself.  It does no harm, but unless you are working with radio-frequency amplifiers (> 1MHz) it serves no purpose.

Making Changes

As already noted, you can use any transistors you have on hand that satisfy the voltage, current and power criteria of those suggested.  They can be rated for higher voltage, current or power, but preferably not less.  For example, you can use TIP31/ 32 transistors in place of the MJE3055/ 2955, although they will be operating close to their limits.  TIP41/ 42 are also suitable.  You could even use TIP141/ 145 Darlington transistors and you won't need the drivers (Q3, Q4).  If you have BD139/ 140 transistors on hand, they can be used instead of the BC639/ 640.  They would be overkill, but they'll work just fine.

The emitter resistors don't have to be 2.7Ω, and anything from 2.2Ω to 3.9Ω will also work.  Lower values will cause greater quiescent current and vice versa.  The diodes in parallel with the emitter resistors can be low voltage types, but not Schottky!  The latter will conduct too early and bias current may be far higher than expected.  High-speed (conventional) diodes are also quite alright, and reduce distortion at high frequencies.  The output capacitor can be reduced if full bass response isn't required, or increased if you want 'better' bass or just have something suitable on-hand.

To get the maximum possible undistorted output level, R2 can be replaced with a 22k fixed resistor in series with a 20k trimpot.  With a load connected, adjust the input level and trimpot setting to get the maximum undistorted output level.  The difference will be well below 1dB compared to the fixed ratios provided, but it gives you the ability to see the effects of varying the DC output voltage.  The power difference is academic at best, amounting to less than 0.5dB.

Figure 2
Figure 2 - DC Voltages At Strategic Nodes (No Signal)

Because this is a project designed specifically so that constructors can experiment with it, I've included the relevant voltages at each circuit node where there is something 'interesting' or useful.  Not all voltages will be exactly as shown due to component variations, but they are all a good representation of the voltages you'll measure.  Because the amp uses feedback to set and stabilise many of the voltages shown, a wiring error will disrupt nearly all of the voltages, and it can be very difficult to determine the error just by taking measurements.  This is the reason that very few ESP circuits show measured voltages - they will only be correct when everything is working properly.  If the amp works as it should, you don't need to measure the voltages (an interesting conundrum!).

Feedback causes problems for anyone who is not skilled in the art of servicing (and it really is an art).  For reasons that escape me, many people seem to assume that servicing equipment is 'easy', and anyone can do it.  Nothing could be further from the truth.  It takes a lot of skill and many years of experience to be a good repair tech, with many being able to teach designers a thing or two about how products should be designed and constructed in production.

As a practice/ demonstration amplifier, this design is a little less difficult to troubleshoot than many others, but a small error will still cause a cascade of 'strange' or 'impossible' voltages throughout the circuit.  Also, be aware that fault finding in something that's just been built (and has never functioned) is a lot more difficult than finding a faulty part in a circuit that did work at some stage.  The most common 'new build' errors are parts in the wrong place, bad solder joints and/ or missing or shorted connections.  Since this amp will be built using Veroboard or similar, forgetting to cut a track or link between tracks will cause malfunction.  You may hate me for not providing a complete layout for you to follow, but you'll learn far more than you would if there were a nice PCB with all the part locations marked!

Figure 3
Figure 3 - Modified Output Stage (Quasi-Complementary)

Another change that could be useful is to use a quasi-complementary output stage.  This is handy if you happen to have NPN power transistors, but no PNP equivalents.  Perhaps surprisingly (perhaps??), this has the potential to reduce distortion, where one may expect it to be worse.  The reduction is only small, and was simulated rather than built, as I didn't feel like making more than one amp for testing.  The DC voltages are changed ever so slightly, but not enough to warrant another drawing.  The main change is the voltage across the biasing diodes (D1-D3) which is lower because one diode has been removed.  There's also an extra resistor (R14) that's necessary with this output stage.


This amplifier will ideally be built on Veroboard, and no PCB is planned.  The amp is meant for you to learn 'stuff', and just plugging parts into a circuit board isn't going to teach you anything.  You will make mistakes, especially with transistor orientation and neglecting to cut Veroboard tracks and/ or add links where needed.  That is all part of the learning process, and it may be annoying, but the lessons are valuable.  I've been using Veroboard for many, many years, and almost all prototypes for published projects are built using it.  Despite many years of experience, I still make mistakes, but that's part of life (the person who makes no mistakes makes nothing at all).

I have included the top view of the transistors, as this is not always apparent from the drawings in datasheets.  The way they are shown is (IMO) pretty bloody useless much of the time, as it's almost always a bottom view shown, when every process (including designing or loading a PCB) requires the top view.

Figure 4
Figure 4 - Transistor Pinouts, Top View

Using the above diagrams will help ensure that the transistors are oriented properly.  If they are wrong, the amp won't work and the transistor(s) may be damaged.  Note that the TO-220 transistors (TIP/ MJE3055/ 2955) have the metal tab connected to the collector.  This means that only the 3055 needs to be insulated from the heatsink (a silicone pad will be fine for the low power dissipated).  By connecting the 2955's collector directly to the heatsink (with thermal compound), the heatsink is grounded.  The heatsink I used is larger than necessary, but it was lying around in my workshop, and was begging to be used for something.  It's pure coincidence that it's almost exactly the same width as the Veroboard.

Figure 5
Figure 5 - Prototype Amplifier Using Veroboard

For reference, the above photo shows my layout.  It's as compact as I could make it, but that's not essential.  To save space, the output capacitor (C5) is not mounted on the board, and I found the Zobel network (C7, R13) unnecessary (hence the reference to it being optional).  The left-hand pin is for feedback via R12 and the middle pin is the input.  The MJE2955 is screwed directly to the heatsink, with the MJE3055 insulated as described above.  If you look, you'll see that I used high-speed diodes.  These are BYV26E (fast, soft recovery) diodes, rated for 1A average current or 10A repetitive peak.

There are only two links under the board, with both used for ground connections.  The bias diode 'string' is enclosed in clear heatshrink tubing.  Make sure that the diode string is 100% reliable - if it goes open circuit, both output transistors will turn on hard, effectively short-circuiting the power supply.  Damage to the amplifier or the power supply (or both) is almost a certainty!

There are some tracks that must be kept short, and you need to ensure minimal capacitance between a few points on the board.  The input is the main one of those, and if you look carefully at the prototype, you'll see that the two 33k resistors are connected so that the the leads going to the base of Q1 are short, with most of the lead going to ground or C6.  If there's any capacitive coupling from the output to the input, the amp will probably oscillate, either continuously or at some part of the output waveform.  Most of the other connections are benign, and C3 limits the bandwidth to a 'sensible' maximum.

This amp was used to verify operation and to take measurements of frequency response and distortion.  High frequency response can be quite astonishing if the 'dominant pole' cap (C3) is reduced, and it can exceed that of most amplifiers of far greater complexity.  It can be extended to over 400kHz (-1dB!) with no evidence of increased distortion, but the amp becomes very sensitive to speaker lead capacitance and can oscillate easily.  I do not recommend that C3 be reduced from the value shown (100pF), but up to 220pF is fine.  At 1kHz, I measured the distortion at 0.06% (any power, 8Ω load), which is far better than I was expecting for such a simple design.  Distortion is predominantly 2nd harmonic.  A listening test didn't indicate any problems with sound quality, but the tests were done in my workshop which is not an ideal listening space.  One thing that is important is a good-sized supply bypass capacitor.  For initial tests I used my bench supply with ~1m long leads, and without a decent (2,200µF) bypass, distortion rose to 0.1%, almost double that with very short (~50mm) leads from the PSU cap to the board..

Because the amp is intended as a learning exercise, none of the usual criteria apply, but it's perfectly suitable as a small amp for powering external speakers for a TV set or as a 'mini' hi-fi.  It will out-perform most other 'simple' amplifiers, and I'd certainly be quite happy to listen to it.  With a 'mid-range' 36V power supply, it has ample power for casual listening, and with reasonably efficient speakers it is surprisingly loud.

One thing that's immediately obvious is that using an LM1875 or similar power amp IC is far less complex than the design shown.  However, you don't have access to any of the internal circuit nodes, and it's just a case of installing the IC, a few resistors and a couple of capacitors and away you go.  You won't learn anything about circuit design using an IC, and that would defeat the purpose of this exercise.

I learned circuit design in my teenage years by analysing existing designs and working out why this or that value of resistor was used, calculating -3dB frequencies based on impedances within the circuit, and building (and modifying) suitable candidates to examine performance and test their functionality.  This design is intended to let you do the same.  It's also worth noting that an oscilloscope was one of my first purchases, having built my own audio oscillator and designed and built a 'full function' transistor tester.

Power Supply

The power supply rejection ratio (PSRR) is better than you might expect, with 2V p-p supply ripple attenuated by over 90dB with the input shorted.  The primary reason is C6, which ensures that the input bias voltage is free from ripple.  C1 assists when the input is shorted, but with the input left open-circuit, PSRR is reduced to a little over 44dB - not a particularly good result.  Therefore, the supply needs to be free of ripple, or hum will be audible with high impedance sources (including a volume pot if used).  You can also increase the value of C6, with a value of 220µF reducing output hum (input open) from 18mV to about 3mV RMS.

Because the current drain is fairly modest, hum should not be a problem, even with a fairly basic power supply.  An alternative is a switchmode 'plug-pack' (aka 'wall wart') supply, and 2A versions are available for use with LED lighting products.  The average current draw at full power into an 8Ω load is about 400mA, with the peak current being a little over 1.2A.  C7 will help to smooth out the peak current, and a 2A supply will usually be quite alright with 8Ω loads.  Predictably, if you use a 4Ω load the current is doubled (both peak and average).

Current is drawn from the power supply only during positive peaks.  This is different from amps that use a dual supply.  During positive peaks, current passes to the load via Q5 and C5, and in the process, C5 is charged slightly.  For negative half-cycles, C5 is discharged via Q6 (and the load of course).  The RMS current through C5 is ~850mA at full power into 8Ω with a 24V supply.  It's important to ensure that C5 is rated for a ripple current of at least 1A RMS, or 2A if you intend to use a 4Ω load.

If you build a linear ('conventional' transformer, bridge rectifier and filter capacitor), I suggest a minimum of 4,700µF for the filter cap.  You could use a regulated supply or a capacitance multiplier, but that adds complexity to what is intended to be a very simple amplifier.  However, if you decide to build a pair of these amps to perform a full listening test, then it's worthwhile to build a decent supply that won't compromise the amp's performance.

Figure 6
Figure 6 - Transformer-Based Regulated Power Supply

If you don't want to use a commercial switchmode supply (they are readily available and not expensive), then the supply shown above will work perfectly.  It will cost far more than a switchmode supply, but it will also be quieter.  Because there's no high-frequency switching, the noise level will be very low.  The TIP3055 requires a heatsink, and it must be electrically insulated from it (the usual stuff).  With a 2A current draw, the transistor will dissipate around 7W, but with music the average will be a lot less.  Note that the mains input must be via a switch and fuse as shown.  Output ripple should be less than 1mV RMS at any sensible output current.

The regulator is based on the same principles as the amplifier.  It uses discrete parts, and it's designed for simplicity rather than high performance.  Regulation is secondary to noise (ripple) rejection, and it can provide 2A output current with ease.  The output ripple should be less than 1mV RMS at 2A output, and that should all but eliminate audible hum from the amplifier.  In case you were wondering, the two 1N4148 diodes boost the output voltage by roughly the same amount that the Darlington series-pass transistor reduces it.  Without the diodes, the output voltage will only be ~22.5V.

It is a fairly easy matter to design a regulator with much higher performance than the version shown, but it's not necessary.  The voltage can be changed by using a higher voltage transformer and a suitable zener diode.  For example, to obtain a 40V DC output, you'd use a 40V transformer, and a pair of 20V zener diodes in series.  No other changes should be needed, but naturally the capacitors need to be rated for the higher voltage (at least 60V).

There's an interesting paradox in the power supply decision-making process.  Despite the complexity of a switchmode supply, they can be bought for less than just the transformer for a linear supply.  A suitable transformer will typically cost at least AU$30.00, but may cost AU$100.00 or more, depending on the supplier.  With the transformer, you still need the rectifier, filter capacitors, regulator, boost transistor and heatsink, which increase the cost even further.  The switchmode supply is ready to go out of the box, needing nothing more than a power lead (and an on/ off switch).  Even the fuse is included internally.  I used a switchmode supply for most tests, but I did use my bench PSU for testing with a 40V supply.

One advantage you do have with the linear power supply is greater flexibility.  The output voltage can be greater than 24V DC, simply by using a higher voltage transformer, capacitors and zener diodes.  Don't go above 48V though, as the amplifier is not designed or intended to be used with more than that.  A 40V supply is preferred, and will give 20W output into 8Ω.

Regardless of the power supply you use, don't apply power directly to the amp when it's just been built.  Some errors can cause transistor damage, and may also damage the power supply.  Use a 27Ω or 33Ω resistor in series with the amp's positive supply.  The amp should draw a quiescent current of about 36-50mA with a 24V supply.  The voltage across a 33Ω 'safety' resistor should be between 1.1V and 1.7V with no load or input voltage.  The amp should bias itself normally, with around 12V at the output (before the output cap of course).  If you get a high voltage across the safety resistor (or no voltage at all), then something is amiss, and you have to find and fix the error before continuing.  Use Figure 2 to compare what you get to what you should get at each point of interest.

The 'Science' Behind Capacitor-Coupled Amplifiers

A capacitor-coupled power amp draws current only during positive half-cycles.  There's nothing mysterious going on, although it may appear to be somewhat confusing.  When a capacitor coupled amp is turned on, the output cap is charged to ½ the supply voltage, and most have a gentle 'thump' when turned on.  The output current charges and discharges the cap ever-so-slightly, but the voltage across the cap doesn't change by very much at high frequencies.  The AC component of the capacitor current is the current into the load.  It's also ripple current, and the capacitor must be able to handle the worst case ripple current.  While this may exceed its ratings, the average is much lower, so the cap isn't stressed with programme material.

At low frequencies, there's more voltage across the cap as its reactance increases with reduced frequency.  With the suggested 1,000µF output cap, the output is 3dB down at just under 20Hz with an 8Ω load.  At 20Hz, the voltage across the capacitor and the load is the same, at 0.7 of the applied voltage before the capacitor.  There's no appreciable difference in the current provided by the upper and lower transistors, because they both have to provide the same peak current into the load.  It doesn't matter if that current comes from the power supply or the output capacitor, the load 'sees' no difference.

The same principle applies to all capacitively coupled circuits, including (or perhaps especially) valves and single-ended transistor circuits.  The cap gains a small extra charge when the output is greater than ½Vcc, and it loses the same when the output is less than ½Vcc.  This keeps the coupling cap's charge essentially neutral.  Supply current is only drawn during positive ½ cycles, but it's not double the load current as you may expect.  The peak DC current is the same as the peak load current.  This has to be the case, as it's a series network (power supply, upper output transistor, capacitor and load), and the current through all sections must be the same.

If the load draws (say) 1A peak (positive and negative) the supply current peaks to 1A too.  There's no violation of energy conservation - the average input current × supply voltage is suitably greater than the power delivered to the load.  Consider the amp described here, with a peak input voltage of 800mV.  The amp has a gain of 12, so 800mV (peak) input gives 9.6V peak output.  When this is positive, the amp provides ±1.2A peak into the load (9.6V and 8Ω).  The supply current is also 1.2A peak, or an average of 386mA.

If we calculate input power (V × Iaverage) this results in an input power of 9.24W.  The output power is 5.76W, giving an efficiency 62%, pretty much what we expect from a Class-AB amplifier.  The current provided by the upper and lower output transistors is almost identical, to the point where it can be simulated but you won't be able to measure it without very sophisticated equipment (the simulator claimed a peak difference of 2mA on 1.2A peaks).

None of this is immediately obvious, and it might seem that the positive peak current should be greater than the negative peak current (to 're-charge' the output cap).  This doesn't happen though.  The charge on the output capacitor has a net zero change, but it does pass the full AC load current.  As noted above, this applies with single supply small-signal amplifiers too, and it doesn't matter whether they use transistors, JFETs or valves (vacuum tubes).  All of this may sound a little strange, but it is real, and countless amplifiers rely on capacitive coupling.


This is an unusual project, not only because it's a low-power amplifier, but because it's designed specifically to allow as much experimentation as possible.  The recommended supply voltage is only 24V, which means you'll get just under 6W output, which is low by modern standards.  This level of output power was common in the early days of audio, with most low-cost systems providing even less.  Typical mantel radios (more commonly called 'wireless' at the time) would have struggled to get 2W without excessive distortion.  Even 'high-end' consoles didn't fare much better, as most used a single-ended pentode output stage.  Despite the low power output, you may be surprised how loud it is with a fairly efficient speaker.

One thing that will come in handy for other projects is the power supply.  If you buy a switchmode unit, aim for around 4A output, as this will allow you to power other projects later on, assuming that you don't wish to keep using the amplifier.  If you use a 40V supply, the output power is in line with many small commercial amps that were built in the 1970s, with a rated output of around 20W/ channel.  This was usually enough for 'normal' domestic listening at the time, since most speaker systems will provide at least 85dB SPL with only one watt, leaving plenty of headroom for transients.

One thing that you may have noticed is that the voltage rating for the capacitors is not shown.  This is deliberate, as it requires you to work that out for yourself.  Figure 2 will be helpful here, because voltages are shown for all the important nodes.  Naturally the voltage ratings will be different if you choose to use a higher voltage supply. but none of it is hard.

Unfortunately, I don't expect that too many people will build this amp, which is a shame.  There's so much to learn about audio and feedback systems in particular, and a simple, well behaved amp that won't self destruct if you run it at full power with a 100kHz signal is an excellent learning tool.  It will be a little irksome to build using Veroboard, but that is also a source of education.  The ability to make changes and experiment isn't something that you'll do if the semiconductors are expensive and an amplifier that's designed to run with a simple (cheap) switchmode power supply should give hours of fun as you learn.  Despite my many years of experience with amps of all types and sizes, I had fun building, measuring and experimenting with it.


This is a project where the only references are other ESP pages.  While I did look at a few other designs (primarily on the interweb), none was used (modified or otherwise).  Some of those shown elsewhere are very good, while others are incapable of even passable performance, with some that almost certainly won't work at all.  The design was created using the method described in the 'Design Process', and as such it's original.  The general principles are well known, and were very common prior to the 1980s.

HomeMain Index ProjectsProjects Index
Copyright Notice. This article, including but not limited to all text and diagrams, is the intellectual property of Rod Elliott, and is © 2021.  Reproduction or re-publication by any means whatsoever, whether electronic, mechanical or electro-mechanical, is strictly prohibited under International Copyright laws.  The author (Rod Elliott) grants the reader the right to use this information for personal use only, and further allows that one (1) copy may be made for reference.  Commercial use is prohibited without express written authorisation from Rod Elliott.
Change Log:  Page published and © Rod Elliott, July 2021./ Updated Jan 2022 - added 'The Science' section.