Why the amplifier is warm before you play anything
Chapter 4 parked a transistor in the middle of its range so a signal could swing both ways without hitting a wall. That decision had a price, and the chapter mentioned it in passing: current flows continuously, with no signal present, purely to hold the transistor there. Time to find out exactly how large that price is.
Take the follower from chapter 3, feed its emitter resistor from a speaker instead of a resistor to ground, and bias it to sit at rest with a steady current ICQ flowing through it — a single transistor doing the whole job, top to bottom of the wave. This is a Class A stage: the device conducts for the full 360° of every cycle, signal or no signal.
Commit before you touch anything
A Class A follower is biased at ICQ = 0.8 A. You turn the input signal all the way down to nothing. What happens to the power drawn from the supply?
Where the power actually goes
The supply delivers a fixed bill. With the device conducting all the time, the average current it draws is set by the bias point, not by the music. PDC ≈ (total supply voltage) × ICQ, and that number does not move with the signal.
The load only collects what the swing delivers. Audio power is PAC = V̂2/(2RL) for a sine of peak V̂. It grows with the signal — but it can never exceed what the bias point allows the current to swing by, which is ICQ·RL at most.
Everything not delivered is dissipated as heat, right in the transistor. At idle, PAC is zero and PDC is not — so all of it turns into heat. That is why the amplifier is warm before you play a single note.
Now put a number on it for a real load. Drive an 8 Ω speaker this way at a sensible listening level and the standing current has to be large enough to swing the peaks you want — which means dozens of watts turning into heat to deliver a handful of watts of sound. A pocket radio can get away with this. A hi-fi amplifier cannot; it needs a heatsink the size of a small book and a fan in the bigger ones.
In the wild
Why tube guitar amps run hot enough to fry an egg on the chassis. Most classic designs are Class A by choice — the sound is prized more than the electricity bill — and a 15 W Class A amp can dissipate 60 W as heat doing it.
Why a Class A hi-fi power amp needs a heatsink the size of a radiator. A 20 W-per-channel Class A design can dissipate 150–200 W as heat, continuously, whether or not anyone is in the room.
Why your phone does not run its speaker amplifier in Class A. A battery has no interest in discarding 75% of its charge as heat. Chapter 8 and beyond will meet Class D, which pushes efficiency past 90% by switching instead of dissipating — but push-pull, this chapter's subject, is the step that comes first historically and conceptually.
Before you move on
A Class A stage is biased at ICQ and is currently playing silence. Where does all the power the supply delivers go?
The handoff that isn’t
The obvious fix for lesson 1’s waste: stop asking one transistor to do the whole wave. Use two — an NPN that only handles the positive half and a PNP that only handles the negative half — and let each one sit off, drawing nothing, while the other is working. Tie their bases together, feed both from the same input, join their emitters at the output. This is complementary symmetry, and each device only conducting half the time makes it a Class B stage.
Here is the assumption almost everyone makes the first time they see this circuit: the two transistors simply hand the wave to each other at zero, like a relay baton passing hand to hand without breaking stride.
Commit before you touch anything
NPN and PNP, bases tied together, no bias voltage between them at all. Feed a clean sine into the shared base. What comes out at the joined emitters, right around zero?
Why the baton actually gets dropped
The NPN's base–emitter junction needs to be forward biased by roughly 0.6 V before it conducts at all — chapter 1's diode threshold, showing up again. So does the PNP's, in the opposite direction. With the bases tied straight together and no bias voltage between them, there is a window well over a volt wide, centred on zero, where neither device is turned on: the input is above −0.6 V but below +0.6 V, and both junctions sit below their knee.
The output does not track the input through that window. It sits at zero, waiting, until the input climbs high enough to wake one transistor up — and then it jumps to catch up. Twice a cycle, at every zero crossing, the same dead spot repeats. This is crossover distortion, and it is not a subtle effect: it puts sharp, high-order harmonics into a signal that had none, which is exactly the kind of distortion the ear is most sensitive to.
Notice, too, that the size of the gap does not depend on the size of the signal. A loud sine and a quiet sine both lose the same fixed window — so the quiet one loses proportionally far more of itself. This is why crossover distortion is at its most audible on quiet passages, and nearly disappears at full volume — the opposite of ordinary clipping, which is why it fooled early designers who only tested loud.
In the wild
The characteristic “grainy” or “transistory” sound of early solid-state amplifiers. First-generation transistor hi-fi in the 1960s went straight to complementary Class B for the efficiency, and got a reputation for a harsh sound that tube amps did not have — this notch is a large part of why.
Why a cheap portable speaker sounds worst at low volume. Cost-cut designs sometimes skimp on the bias network from lesson 3. The distortion floor is nearly constant in absolute terms, so it dominates exactly when the music is quiet.
Zero-crossing distortion on an oscilloscope. If you ever probe a suspect amplifier and see a small flat notch stitched into an otherwise clean sine right at the axis, this is what you are looking at — and lesson 3 is the fix.
Before you move on
Why does a pure Class B push-pull stage distort near zero even though each transistor individually is a clean amplifier?
Splitting the difference
The dead zone exists because neither junction is forward biased near zero. So forward-bias them a little, on purpose, even at rest. Insert a small voltage source of about 1.3–1.5 V between the two bases — roughly two diode drops — and both transistors sit just barely on when the input is zero, instead of both sitting off.
Commit before you touch anything
You add exactly enough bias between the bases to make each junction sit right at its 0.6 V knee when the input is zero. What happens to the dead zone?
The bill for closing the gap
Nothing is free. With both junctions sitting at their knee even at rest, a small standing current now flows through both transistors continuously — the quiescent current ICQ is back, just far smaller than a full Class A design would need. That is the entire idea of Class AB: borrow a sliver of Class A's always-on current, only near the crossover, and get Class B's efficiency everywhere else.
The relationship between bias voltage and quiescent current is not gentle. It is the same exponential from chapter 1 and chapter 2 — IC = ISeVBE/VT — which means a bias voltage that is a few tens of millivolts too high does not raise ICQ a little. It raises it by a large multiplicative factor. Drag the bias slider past the sweet spot on the bench and watch the quiescent current climb steeply, then explosively.
In the wild
The bias trimmer inside almost every solid-state power amplifier. A technician servicing one measures ICQ directly and adjusts a single potentiometer until it lands in spec — too low and crossover distortion returns; too high and the output transistors run hot and age faster.
Why some boutique amplifiers are deliberately biased into heavier Class AB, or even Class A. Pushing ICQ well past the minimum needed to close the gap keeps both transistors further from cutoff for a larger part of the swing, at the cost of exactly the heat lesson 1 described. It is a real, audible trade some designers choose to make.
Why headphone amplifiers often run pure Class A. The load is much lighter than a speaker — tens to hundreds of ohms instead of 8 Ω — so the wasted heat from lesson 1 is small in absolute terms, and some listeners prefer paying it to avoid any trace of the crossover notch.
Before you move on
Why is the bias voltage between the bases trimmed on the bench with a real amplifier, rather than fixed at a calculated value?
The bias that has to run hot to stay cool
Lesson 3 trimmed VBB once, on the bench, at room temperature. But an output transistor driving a real load does not stay at room temperature — it is bolted to a heatsink that climbs from 25 °C at idle to 80 °C or more under sustained power. And VBE is not constant with temperature: chapter 1 already told you a diode's forward voltage falls by about 2 mV per degree Celsius at a fixed current. The output transistors' base–emitter junctions do the same thing.
Commit before you touch anything
VBB is fixed at the room-temperature sweet spot from lesson 3, by a voltage source that itself stays at room temperature. The output transistors' heatsink then heats up to 80 °C under load. What happens to ICQ?
Thermal runaway, named precisely
This is a positive feedback loop with no natural brake: heat lowers VBE, a lower VBE threshold means more of a fixed VBB becomes excess drive, excess drive raises ICQ, and higher ICQ means more power dissipated as heat in exactly the junction that just got more sensitive. Left unchecked, this destroys the output transistors — the failure mode that gives the phenomenon its name.
Notice what this means in practice: the bias diodes are chosen and mounted specifically to run hot when the output transistors run hot, and to track their exact VBE behaviour as closely as possible. A bias network sitting on a cool part of the chassis, wired for convenience rather than thermal contact, is a well-known way to build an amplifier that works perfectly on the bench and fails in the field.
In the wild
The Vbe multiplier bolted directly to the output transistor's heatsink tab. Open almost any solid-state power amplifier and you will find a small transistor or diode pair physically clamped against the same metal as the power devices — not near it, touching it — purely so it shares their temperature.
Emitter resistors as a second line of defence. A small resistor, often under 1 Ω, in series with each output transistor's emitter adds negative feedback: if one device's current rises, the voltage drop across its own resistor rises too, which reduces its own drive. Real amplifiers use both defences together — thermal tracking to prevent the drift, emitter resistors to limit the damage if it happens anyway.
Why a failed output transistor often takes its neighbour with it. If runaway starts in one device, it can also unbalance the current sharing between the two halves of the pair, stressing the other side until it fails too — which is why a blown power amp is so often a matched-pair repair, not a single part.
Before you move on
Why are the bias diodes in a real push-pull amplifier mounted on the same heatsink as the output transistors, in physical contact with them?
The worst moment is not the loudest one
Class AB earns back most of what Class A gave away. With each transistor conducting for only a little over half the cycle instead of all of it, the standing loss from lesson 1 mostly disappears, and efficiency rises a great deal — close to the ideal Class B limit, since the small crossover bias from lessons 3 and 4 costs very little once it is set correctly.
Commit before you touch anything
A push-pull stage is driven from silence up to full clipping. At what drive level does each output transistor dissipate the most heat?
Why the peak sits below full swing
For an ideal Class B stage driven by a sine of peak amplitude V̂ on a supply VCC, two standard results follow from the sine's own shape:
Differentiate that and the peak sits at V̂ = (2/π)VCC ≈ 0.637·VCC — about 64% of full swing, not 100%. At that exact point, each transistor is dissipating the most heat it will ever see from a sine wave, and it is not the point anyone would guess by ear, because it does not sound like the loudest setting.
Real program material makes this worse, not better. A sine wave is the best case for a power amplifier — it has the highest ratio of average to peak power of any common waveform. Music and speech have a much higher peak-to-average ratio, so their long-term average sits well below the 64% danger zone most of the time, but a sustained tone or a clipping guitar power chord can park an amplifier right at its worst point for seconds at a stretch. Amplifier heatsinks are sized with this in mind, not for a brief peak.
In the wild
Why amplifier datasheets quote a continuous power rating well below what a peak meter shows. The continuous rating reflects sustained dissipation near the worst-case drive point, not the highest number the amplifier can produce for an instant.
Why a compressed, near-constant-level radio signal is harder on an output stage than dynamic music. Heavily compressed audio spends far more time near peak drive levels, which means more time near the 64% danger zone specifically — one reason broadcast and PA amplifiers are built to a tougher thermal standard than hi-fi amplifiers of the same power rating.
Why bench-testing an amplifier with a sustained sine tone is considered a harsher test than listening to music at the same wattage. A steady sine held at the 64% point is close to the worst thermal case an amplifier will ever see; music at the same average power rarely dwells there.
Before you move on
Why does maximum transistor dissipation in an ideal Class B stage occur below full output swing?
The stage nobody draws on the first pass
Every bench so far drew the output transistors as if the small-signal stage before them could drive their bases directly. It cannot, not for long. A power transistor delivering an amp of output current still needs a meaningful fraction of that current at its base — β falls at high current, and a small-signal transistor from chapter 2 was never designed to source that much base drive on its own.
Commit before you touch anything
Why does a real push-pull power stage almost always add a smaller pair of transistors ahead of the output pair, rather than driving the output pair directly?
Three practical problems, three fixes
Not enough base current: driver transistors. A small NPN driver feeding the output NPN's base forms a Darlington pair — combined β near β1·β2. On the PNP side, a common trick pairs an NPN driver with the PNP output device in a Sziklai (complementary-feedback) pair, which behaves like a single high-β PNP but is built from parts that were both easier to manufacture well.
Matched complementary devices are hard to guarantee: quasi-complementary output. For decades, good PNP power transistors were harder to make than good NPN ones. The common workaround uses two NPN output transistors, with the bottom one wired in a Sziklai pair with a small PNP driver so the whole assembly still behaves like a complementary pair from the outside. Modern designs increasingly use genuinely matched complementary pairs now that PNP power devices have caught up, but quasi-complementary output remains common and is worth recognising on a schematic.
A shorted output could destroy the output transistors in microseconds: current limiting. A small sense resistor in series with each output emitter, plus a transistor that turns on and steals base drive away the moment the voltage across that resistor implies too much current, clamps the output current to a safe maximum regardless of how hard the load is shorted. This keeps the pair inside its safe operating area — the region of current and voltage a transistor can survive simultaneously, which is smaller than either limit alone would suggest.
In the wild
The classic 1970s–80s solid-state power amp schematic. Small-signal differential input, a voltage-gain stage, then exactly this driver-plus-quasi-complementary-output block. Recognising this block is most of the way to reading the whole amplifier.
Why some vintage amplifier repairs replace “just one” transistor and it still misbehaves. Output pairs are factory-matched for VBE and β specifically because lesson 3 showed how sensitive ICQ is to exactly those numbers. An unmatched replacement can leave the bias trim unable to find a safe point at all.
The little resistor that saves the amplifier when a speaker cable shorts. That is the current-sense resistor from fix three, usually the least glamorous component on the board and the reason the amplifier survives a mistake that would otherwise take out both output transistors at once.
Before you move on
In a quasi-complementary output stage, why can two NPN output transistors stand in for a true complementary NPN/PNP pair?
Chapter checkpoint
Question 1
A Class A stage is biased at a high ICQ and is currently playing silence. Roughly how much of the power the supply delivers is being dissipated as heat?
Question 2
Two complementary transistors, bases tied directly together, no bias voltage between them. What does the output look like near a zero crossing?
Question 3
Why does a small forward bias voltage between the two bases close the crossover gap?
Question 4
Reaching back to chapter 1: why does mounting the bias diodes on the same heatsink as the output transistors prevent thermal runaway?
Question 5
An ideal Class B push-pull stage is driven by a sine wave from silence up to full clipping. At what drive level is each transistor's heat dissipation highest?
Question 6
Reaching back to chapter 2: why can't a small-signal driver transistor's own base current usually supply a power output transistor directly?
Every word this chapter introduced
Where this goes next
About the simulations. Benches 2, 3 and 4 solve the complementary output stage exactly, at every sample point: each transistor's collector current comes from the same Ebers–Moll transport equation used since chapter 2, and the output node voltage is found by bisecting the shared-node current balance — IC,NPN(VOUT) − IC,PNP(VOUT) − VOUT/RL = 0 — to a tight tolerance at each of two hundred points around the cycle, rather than being drawn from the textbook piecewise formula. Bench 4's temperature dependence comes from the same IS(T) law introduced in chapter 2, applied to both the output devices and, in tracked mode, to the bias diodes themselves. Bench 1's power figures and bench 5's efficiency curve use the standard closed-form Class A and Class B relations rather than a point-by-point solve, which is why they are exact rather than approximate — the underlying assumptions are the same ones stated in chapter 5: β held constant with current, ideal capacitors, and a single VA = 75 V. Two further simplifications specific to this chapter: the output devices share chapter 2's small-signal IS, so this chapter's currents and dissipation figures are illustrative of the shape of the physics rather than a specific power transistor's datasheet; and driver-stage loading, covered qualitatively in lesson 6, is not included in any bench's numbers.