The photodiode is just the front of the receiver. After it comes a careful chain of analog and digital electronics that recovers bits from microamps of photocurrent.
7.1 The transimpedance amplifier (TIA)
The first job is to convert the photodiode current into a voltage. The standard architecture is a transimpedance amplifier: an op-amp (or differential transistor pair) with a feedback resistor from output to input. The photodiode injects current into the inverting input, and the op-amp servos its output to keep the input at virtual ground. Output voltage is .
The TIA is the dominant noise contributor in a PIN receiver. A high minimizes thermal noise but limits bandwidth (the input-capacitance pole moves down). A low gives bandwidth but adds noise. Datacenter TIAs run at 25 to 100 GHz with of 200 to 1000 ohms.
Older "high-impedance" front ends used a very large followed by a post-amp that explicitly equalizes the bandwidth. They had lower noise but were complex; the TIA architecture won the practical battle.
7.2 Limiting amplifier and clock and data recovery (CDR)
After the TIA, a limiting amplifier boosts the signal to logic levels independent of input amplitude, preserving timing while clipping amplitude variations. Then a CDR (clock and data recovery) circuit extracts the bit clock from the data stream and uses it to sample each bit at the eye center.
Most CDRs are phase-locked loops driven by data transitions. They lock onto the average transition rate and provide a clock. Modern CDRs operate at the symbol rate (e.g., 53 Gbaud for 100G PAM-4) and recover both clock and data.
7.3 Forward error correction
Modern optical receivers always include FEC (Chapter 12). The transmitter encodes the data with redundancy; the receiver corrects errors before passing the data on. With FEC, the system runs at a much higher pre-FEC error rate (around ) and corrects to a post-FEC rate of . This relaxes the OSNR requirement by 6 to 10 dB, doubling the reach for free.
The standard FECs in optical:
- G.709 hard-decision Reed-Solomon (255, 239): 7% overhead, 6 dB net coding gain. Original choice.
- Soft-decision LDPC and TPC: 15-25% overhead, 11-12 dB net coding gain. Standard in coherent systems since 2012.
The DSP that implements soft-decision LDPC is itself a major chip, often consuming as many transistors as a small CPU.