The Avalanche Photodiode Receiver

Swap the BPW34 for the silicon APD, keeping everything else the same. An APD needs typically 100 to 200 V of reverse bias — check YOUR device's datasheet, because the breakdown voltage is a per-part number and exceeding it destroys the diode. Use an adjustable high-voltage bench supply in place of the -9 V bias, with the APD's anode to its negative output, and current-limit it hard, to a few tens of microamps. At 100 to 200 V the bias is dangerous to touch: switch it off and let it discharge before changing anything. An APD in breakdown draws runaway current and dies in milliseconds; the series R_BIAS is the only thing standing between a mistake and a dead part. Sweep the bias from 50 V upward in 5 V steps with a constant light level, and record the output at each. The photocurrent will be flat, then start to climb, then climb steeply. That curve IS the gain M against voltage, normalised to the flat region where M = 1. Now warm the diode gently with your fingers and watch the output FALL at constant bias. APD gain drops as temperature rises, because hotter carriers scatter off lattice vibrations before they gain enough energy to ionise. That is why a real APD receiver servos its bias against a thermistor.

HOW MULTIPLICATION WORKS. In a high field a carrier gains enough energy between collisions to break a bond and make a new electron-hole pair; both accelerate and can do it again. M is the average number of carriers collected per carrier generated. The randomness of WHERE each ionisation happens is what makes the output noisier than simple shot noise. WHY k IS THE MATERIAL'S WHOLE STORY. If only one carrier type ionises, the cascade runs one way and the statistics are tight. If both do, it feeds back on itself, the gain becomes wildly variable and F(M) heads towards M. Silicon k is about 0.02, InGaAs/InP about 0.45, germanium close to 0.9. Silicon APDs are superb — and silicon is transparent above about 1100 nm, exactly where fibre wants to work. The best APD material and the best fibre wavelength do not overlap, which is one reason the amplifier in the next blueprint mattered so much. THE SIBLING TABLE. PIN photodiode: gain 1, no high-voltage supply, no temperature servo, flat response, cheap; all the gain and all the noise come from the amplifier. APD: gain 10 to 200, needs 100-200 V current-limited and temperature-compensated, adds F(M) of its own, and buys typically 5 to 10 dB of sensitivity in a thermally limited receiver — at a best M that is not the biggest M. The deciding question is never which is better, it is whether the receiver is THERMALLY LIMITED. If it is, the APD wins by putting its gain ahead of the thermal noise. If the signal is strong enough that shot noise already dominates, the APD only adds F(M) and makes things worse. WHAT THIS SHARES WITH THE CCD. Both are silicon collecting photo-generated charge. The CCD integrates it in a capacitor for milliseconds then shifts it out; this receiver reads the current continuously at a hundred megahertz. Long integration against wide bandwidth, from identical physics — which is why a camera sensor can see a single star and cannot carry a gigabit. COMMON FAILURES. Ringing on the step response means Cf is too small or the summing node is too long. A DC offset that swamps everything is ambient light, not a fault: AC-couple after the amplifier or put an optical filter in front. A dead APD is almost always a bias supply that was not current-limited.