A complementary emitter-follower (Class-B push-pull) output stage uses an NPN transistor to source current on positive swings and a PNP to sink current on negative swings. The problem: each BJT needs ~0.6 V of V_BE before it turns on. Near the zero-crossing, both transistors are off simultaneously, creating a dead zone of roughly ±0.6 V where the output sticks at zero. The audible result is harsh, buzzy crossover distortion — especially obnoxious at low listening levels where the signal spends most of its time near zero.

The fix is Class-AB biasing: hold both transistors just barely conducting at idle, so when the signal crosses zero, one device is already taking over from the other. The classic technique places two diode drops between the NPN and PNP bases — typically two forward-biased diodes or, better, a V_BE multiplier (a transistor with a resistor divider between collector and base, giving an adjustable, thermally-tracking bias of n·V_BE).

Why the V_BE multiplier matters: output transistors heat up under load, and V_BE drops about 2 mV/°C. If your bias source doesn't track that drop, quiescent current rises with temperature, the transistors heat further, and you get thermal runaway — a dead amplifier with melted solder. Mounting the bias-multiplier transistor on the same heatsink as the output devices forces its V_BE to fall in lockstep, holding idle current stable.

Real-world example: A 20 W into 8 Ω hi-fi amplifier output stage. Peak output current is √(2·20/8) ≈ 2.2 A. You'd choose TIP41C/TIP42C complementary BJTs, set quiescent current around 20–50 mA (enough to keep both devices in conduction through the crossover region, low enough that idle dissipation is reasonable), and add 0.22 Ω emitter degeneration resistors in series with each output transistor. Those small resistors add local negative feedback that further stabilizes bias against temperature and forces current sharing if you parallel devices.

Rule of thumb for quiescent current: set I_Q so the voltage across each emitter resistor is roughly 10–25 mV. For 0.22 Ω resistors, that's 45–115 mA — the sweet spot where crossover artifacts disappear without wasting power. Adjust the V_BE multiplier's trim pot while measuring this voltage with the input shorted.

Wrap the whole output stage inside a global op-amp feedback loop and residual crossover distortion drops by another 40+ dB, but the feedback can't fix what the bias gets badly wrong — start with a properly biased stage, then let feedback polish it.