Photodetectors and Optical Receiver Design

Introduction

An optical receiver converts a weak, modulated optical signal into reliable digital information. Its performance depends jointly on the photodetector, front-end electronics, filtering, sampling and decision algorithm. Even a detector with high responsivity may yield poor sensitivity if its capacitance, dark current, multiplication noise or associated amplifier is unsuitable.

Direct-detection receivers infer transmitted intensity without recovering optical phase. They dominate access networks, data-centre links, passive optical networks and many short- and medium-reach systems. This chapter develops the physical and statistical models required to design such receivers, with particular emphasis on PIN and avalanche photodiodes (APDs), noise, decision thresholds, sensitivity and bit-error rate (BER).

Learning Objectives

After studying this chapter, a reader should be able to:

• explain photocarrier generation in PIN and avalanche photodiodes;
• calculate quantum efficiency and responsivity;
• model shot, thermal, dark-current and avalanche multiplication noise;
• compare PIN and APD receiver performance;
• analyse transimpedance receiver bandwidth and noise;
• derive decision thresholds for Gaussian receiver statistics;
• calculate BER, sensitivity and quantum-limited performance; and
• evaluate practical direct-detection receiver designs.

1. Photodetection Fundamentals

A photon of frequency f carries energy

Eₚ = hf = hc/λ,

where h is Planck's constant, c is the speed of light and λ is the vacuum wavelength. In a semiconductor, a photon can generate an electron–hole pair if its energy exceeds the material band-gap energy E_g. The approximate long-wavelength cutoff is

λ_c = hc/E_g.

Absorption alone is insufficient: generated carriers must be collected before recombination. Detector material, absorption-layer thickness, electric field and carrier transit time therefore affect efficiency and speed.

1.1 Quantum efficiency

External quantum efficiency η is the ratio of collected electron–hole pairs to incident photons:

η = number of collected carriers / number of incident photons.

If incident optical power is P, the incident photon rate is P/(hf). The average primary photocurrent is consequently

Iₚ = ηqP/(hf),

where q is the elementary charge. Detector responsivity is

R = Iₚ/P = ηq/(hf) = ηqλ/(hc). (1)

Responsivity has units of amperes per watt. When λ is expressed in micrometres,

R ≈ ηλ/1.23984 A/W.

This expression assumes that each collected photon contributes one elementary charge and that there is no internal multiplication. Reflection, incomplete absorption and recombination reduce η.

Worked Example 1: Responsivity of an InGaAs detector

An InGaAs PIN photodiode has quantum efficiency η = 0.80 at 1550 nm. Its responsivity is

R = 0.80 × 1.55/1.23984 ≈ 1.00 A/W.

For an incident power of 2 μW,

Iₚ = RP = 1.00 × 2 μW = 2.00 μA.

The result is physically plausible because the unity-quantum-efficiency responsivity at 1550 nm is approximately 1.25 A/W.

2. PIN Photodiodes

A PIN photodiode contains p-type and n-type regions separated by a relatively wide intrinsic region. Reverse bias establishes an electric field across this region. Photons absorbed there produce carriers that drift toward the contacts.

2.1 Advantages

PIN photodiodes offer:

• high quantum efficiency;
• low dark current;
• high linearity;
• low excess noise;
• relatively low bias voltage; and
• bandwidths from several gigahertz to tens of gigahertz in optimized devices.

They are widely used when received power is sufficient or when low-noise transimpedance amplifiers provide adequate sensitivity.

2.2 Bandwidth limitations

The response is limited by carrier transit time and circuit capacitance. If B_t is the transit-time bandwidth and B_RC is the RC bandwidth, a useful approximation for independent first-order limitations is

1/B_d² ≈ 1/B_t² + 1/B_RC². (2)

This is an engineering approximation, not a universal device law. The RC bandwidth is approximately

B_RC = 1/[2πR_eq(C_d + C_a)],

where R_eq is the effective resistance, C_d is detector capacitance and C_a includes amplifier and parasitic capacitances.

Increasing absorption-layer thickness improves absorption but lengthens carrier transit time. Device design therefore involves an efficiency–bandwidth trade-off.

3. Avalanche Photodiodes

An APD includes a high-field multiplication region. A primary carrier can acquire sufficient energy to create additional electron–hole pairs by impact ionization. The average multiplication factor is

M = I_M/Iₚ,

where I_M is multiplied photocurrent and Iₚ is primary photocurrent. The effective responsivity is

R_APD = MR.

3.1 Excess multiplication noise

Avalanche multiplication is stochastic. If k_A is the ratio of hole to electron ionization coefficients for electron-initiated multiplication, a commonly used excess-noise model is

F(M) = k_A M + (1 − k_A)(2 − 1/M). (3)

This expression is based on a local, uniform-field multiplication model. Real heterostructure and thin multiplication-region APDs can deviate from it.

The mean signal increases by M, but the mean-square shot noise increases approximately by M²F(M). Thus APD gain is not equivalent to noiseless electronic amplification.

3.2 Gain–bandwidth trade-off

Higher avalanche gain increases carrier build-up time. Many APDs exhibit an approximate gain–bandwidth product over part of their operating range:

M B_APD ≈ constant.

The approximation should be applied only within the manufacturer's specified regime. APDs also require high, temperature-dependent bias and protection from breakdown or optical overload.

Worked Example 2: APD gain and excess noise

An APD receives 1 μW at 1550 nm. Its primary responsivity is 1.0 A/W, multiplication is M = 10, ionization ratio is k_A = 0.20 and noise bandwidth is 1 GHz.

The primary and multiplied currents are

Iₚ = 1.0 × 1 μW = 1 μA,

I_M = 10 μA.

From Equation (3),

F(10) = 0.20(10) + 0.80(2 − 0.1) = 3.52.

Ignoring dark current, the avalanche shot-noise current is

σ_APD = √[2qB_nM²F(M)Iₚ]

= √[2(1.602 × 10⁻¹⁹)(10⁹)(100)(3.52)(10⁻⁶)]

≈ 0.336 μA rms.

Although the mean current is multiplied by ten, the noise is also enhanced by stochastic multiplication.

4. Receiver Noise

Receiver noise must be evaluated over the equivalent noise bandwidth B_n, which need not equal the nominal 3 dB bandwidth. The value of B_n depends on the complete electrical filter response.

4.1 Shot noise

Photocarrier arrivals are approximately Poisson distributed when photons arrive independently. For a PIN detector, the shot-noise variance is

σ_sh² = 2q(Iₚ + I_b + I_d)B_n, (4)

where I_b is background-induced current and I_d is dark current. The formula assumes stationary current and a bandwidth much lower than the inverse microscopic pulse duration.

For an APD, a simplified output-referred model is

σ_sh,APD² = 2qB_n[M²F(M)(Iₚ + I_b + I_db) + I_ds], (5)

where I_db is multiplied bulk dark current and I_ds is unmultiplied surface leakage. Manufacturer conventions must be checked because dark-current partitioning differs among data sheets.

4.2 Thermal noise

For a resistor R_L at absolute temperature T, the mean-square thermal-noise current over B_n is

σ_th² = 4k_BT B_n/R_L, (6)

where k_B is Boltzmann's constant. This expression represents an ideal resistor with white noise over the relevant band.

A practical transimpedance amplifier (TIA) also contributes input current noise, voltage noise acting through source impedance, feedback-resistor noise and subsequent-stage noise. A complete design integrates the input-referred noise spectral density over frequency:

σ_n² = ∫₀∞ S_i,eq(f)|H_n(f)|² df.

Worked Example 3: Correct calculation of PIN receiver noise

A PIN receiver has total average detector current Iₚ + I_d = 10 μA, noise bandwidth B_n = 100 MHz, load resistance R_L = 10 kΩ and temperature T = 300 K. Background noise and amplifier noise are neglected.

Shot noise is

σ_sh = √[2(1.602 × 10⁻¹⁹)(10 × 10⁻⁶)(100 × 10⁶)]

= 1.790 × 10⁻⁸ A = 0.0179 μA.

Thermal noise is

σ_th = √[4(1.381 × 10⁻²³)(300)(100 × 10⁶)/(10⁴)]

= 1.287 × 10⁻⁸ A = 0.0129 μA.

Because the components are independent, the total rms noise is

σ_n = √(σ_sh² + σ_th²)

= 2.205 × 10⁻⁸ A = 0.0221 μA.

For a 10 μA signal current, the electrical amplitude ratio is 10/0.0221 ≈ 453. The corresponding current-squared SNR is approximately 2.06 × 10⁵, or 53.1 dB.

5. Receiver Electronics

5.1 Transimpedance front end

A TIA converts detector current into voltage. For feedback resistance R_f, the low-frequency output is approximately

v_o = −i_pR_f.

A large R_f provides high conversion gain and low feedback-resistor current noise, but detector and amplifier capacitances can reduce bandwidth or compromise stability. Practical designs use feedback capacitance, inductive peaking, regulated cascodes or distributed architectures to manage this trade-off.

5.2 High-impedance and integrated receivers

A high-impedance front end can provide low noise but may require equalization because its RC time constant causes intersymbol interference. Integrate-and-dump receivers accumulate charge over a symbol interval and are closely related to matched-filter detection for rectangular pulses.

Monolithic integration reduces parasitic capacitance. Silicon photonics commonly integrates germanium photodiodes with CMOS or BiCMOS TIAs, although optical coupling loss, thermal effects and process compatibility remain important.

5.3 Filtering, equalization and clock recovery

The electrical filter should suppress out-of-band noise without excessively distorting pulses. Insufficient bandwidth causes intersymbol interference; excessive bandwidth admits unnecessary noise. Feed-forward equalizers, decision-feedback equalizers and maximum-likelihood sequence methods can extend performance, but may increase power, latency and error propagation.

Clock-and-data recovery determines the optimum sampling instant. Timing jitter reduces eye opening and creates an additional sensitivity penalty not represented by a static amplitude-noise model.

6. Statistical Decision Theory

Let the sampled decision variable X have conditional Gaussian distributions

X|0 ∼ N(μ₀, σ₀²), X|1 ∼ N(μ₁, σ₁²),

with μ₁ > μ₀ and prior probabilities P₀ and P₁. Gaussian modelling is usually accurate when many thermal and shot-noise contributions combine, but it can be inaccurate in photon-starved or strongly nonlinear regimes.

6.1 Maximum a posteriori decision

The maximum a posteriori (MAP) receiver selects bit 1 when

p(X|1)P₁ > p(X|0)P₀.

Likelihood crossings satisfy

(x − μ₀)²/(2σ₀²) − (x − μ₁)²/(2σ₁²)

= ln[(P₀σ₁)/(P₁σ₀)]. (7)

If σ₀ = σ₁ = σ, Equation (7) is linear and gives

γ = (μ₀ + μ₁)/2 + [σ²/(μ₁ − μ₀)]ln(P₀/P₁). (8)

For equal priors, γ is the midpoint of the means.

When σ₀ ≠ σ₁, Equation (7) is quadratic and can have zero, one or two real likelihood crossings. Therefore, a single between-means threshold is not universally equivalent to exact MAP detection. Two crossings may produce a bounded decision region for one class and two tail regions for the other. In many optical receivers, one root is outside the physically useful current range, so a single between-means threshold remains a practical approximation.

For example, let μ₀ = 0, σ₀ = 1, μ₁ = 2, σ₁ = 2 and P₀ = P₁. Equation (7) becomes

x²/2 − (x − 2)²/8 = ln 2,

with roots approximately −2.571 and 1.238. Because the bit-1 distribution has the larger variance, exact MAP detection assigns both extreme tails to bit 1 and the interval between the roots to bit 0.

6.2 BER for a single threshold

If the receiver decides 1 for X > γ, its BER is

P_e = P₀Q[(γ − μ₀)/σ₀] + P₁Q[(μ₁ − γ)/σ₁], (9)

where

Q(x) = (1/√(2π))∫ₓ∞ exp(−u²/2)du.

Equation (9) applies only to the specified single-threshold decision rule. If exact MAP detection has two thresholds, error probabilities must instead be integrated over the corresponding multiple decision regions.

A widely used engineering quality factor is

Q_R = (μ₁ − μ₀)/(σ₁ + σ₀). (10)

For equally likely bits, choosing the threshold so that the two normalized distances are equal gives approximately

BER ≈ Q(Q_R).

This is exact for equal Gaussian variances and a midpoint threshold. For unequal variances, it is a useful approximation rather than the general MAP result.

Worked Example 4: Threshold and BER

A binary receiver has equally likely bits, μ₀ = 0.05 V, μ₁ = 0.50 V and equal standard deviations σ₀ = σ₁ = 0.05 V.

The MAP threshold is

γ = (0.05 + 0.50)/2 = 0.275 V.

The normalized decision distance is

(0.275 − 0.05)/0.05 = 4.5.

Therefore,

BER = Q(4.5) ≈ 3.40 × 10⁻⁶.

The equal variances and equal priors are essential to the midpoint result.

7. Receiver Sensitivity

Receiver sensitivity is the minimum average received optical power needed to achieve a specified BER under stated conditions. A complete specification should include wavelength, data rate, modulation format, extinction ratio, temperature, overload limits and test pattern.

For on–off keying (OOK), let P₁ and P₀ be the optical powers for bits 1 and 0. The average power for equally likely bits is

P_avg = (P₁ + P₀)/2.

The extinction ratio is ER = P₁/P₀. Finite extinction ratio reduces the current difference R(P₁ − P₀), thereby degrading sensitivity.

Under equal Gaussian noise σ and ideal P₀ = 0,

R P₁ = 2σQ⁻¹(BER),

so

P_avg = σQ⁻¹(BER)/R. (11)

This special-case expression must not be used when noise depends strongly on signal level, as often occurs with shot noise or APD multiplication.

7.1 Quantum limit

For an ideal photon-counting OOK receiver with zero photons in bit 0, no dark counts and a decision of 1 whenever at least one photon is detected, the only error is receiving zero photons during a transmitted 1. If N₁ is the mean detected photon count in a one bit,

BER = 0.5e^(−N₁).

The mean photon count averaged over both equally likely bits is

N_avg = N₁/2 = −0.5ln(2BER). (12)

At BER = 10⁻⁹, N_avg ≈ 10.02 detected photons per bit. At 1550 nm and 10 Gb/s, this corresponds to

P_avg = N_avg(hc/λ)R_b ≈ 12.8 nW ≈ −48.9 dBm.

This is an idealized limit. Practical receivers require substantially more power because of finite quantum efficiency, thermal noise, dark current, bandwidth constraints, synchronization and implementation penalties.

8. PIN–APD Design Comparison

The approximate APD output SNR can be represented as

SNR = (MI_s)²/{2qB_nM²F(M)(I_s + I_b + I_db) + σ_T² + σ_s²},

where σ_T² represents electronic noise and σ_s² includes unmultiplied surface-current noise.

At low M, electronic noise may dominate, so multiplication improves sensitivity. At high M, excess avalanche noise and reduced bandwidth offset further gain. Consequently, an optimum finite multiplication often exists.

PIN receivers generally provide better linearity, lower bias, simpler temperature control and lower excess noise. APDs are attractive where their internal gain overcomes substantial front-end thermal noise. The final choice should be based on system sensitivity, bandwidth, cost, reliability and operating environment rather than responsivity alone.

9. Applied Case: Designing a 10 Gb/s PIN Receiver

Background

This illustrative case concerns a 1550 nm, 10 Gb/s OOK link using a PIN-TIA receiver. The detector responsivity is 0.90 A/W, the extinction ratio is 10 dB, and measured rms decision noise is 2.5 μA for both symbols. The required BER is 10⁻¹², for which Q⁻¹(10⁻¹²) ≈ 7.034.

Problem

Determine receiver sensitivity and assess a link delivering −13 dBm. The engineering specification requires at least 4 dB of link margin.

Analysis

An extinction ratio of 10 dB gives

P₀/P₁ = 0.1.

For equal noise standard deviations,

I₁ − I₀ = (σ₁ + σ₀)Q_R

= 5.0 μA × 7.034 = 35.17 μA.

Because

I₁ − I₀ = R(P₁ − P₀) = 0.90(0.9P₁),

P₁ = 35.17 μA/0.81 ≈ 43.42 μW,

P₀ = 4.342 μW.

Thus

P_avg = (43.42 + 4.342)/2 = 23.88 μW

or approximately −16.22 dBm.

At −13 dBm, the available margin is

−13 − (−16.22) = 3.22 dB,

which is below the required 4 dB.

Approach and Solution

Reducing input-referred rms noise to 2.0 μA through lower detector capacitance and an improved TIA gives

I₁ − I₀ = 4.0 μA × 7.034 = 28.14 μA.

Repeating the calculation gives P₁ ≈ 34.74 μW and P_avg ≈ 19.11 μW, or −17.19 dBm. The revised margin is 4.19 dB.

Outcome

The lower-noise design meets the margin requirement, subject to laboratory validation over temperature, process variation, timing jitter and representative data patterns.

Learning Points

• Extinction ratio must be included in sensitivity calculations.
• Electrical noise reduction can be more valuable than increasing detector responsivity.
• Nominal sensitivity is not sufficient; design margin is required.
• Signal-dependent shot noise should be included if it is no longer negligible.

10. Practical Challenges and Best Practices

Receiver models can fail because of intersymbol interference, pattern-dependent baseline wander, nonlinear TIA response, optical reflections, mode-partition noise, relative-intensity noise, timing jitter and overload recovery. Good practice includes:

• using the equivalent noise bandwidth rather than assuming it equals data rate;
• separating multiplied and unmultiplied APD dark currents;
• extracting noise from measured spectra or eye statistics;
• validating Gaussian assumptions at low photon counts;
• specifying sensitivity and overload together;
• testing across wavelength, temperature and manufacturing variation;
• reserving implementation margin for aging and jitter; and
• verifying BER using sufficiently long, representative sequences.

Machine-learning methods are increasingly used for adaptive thresholding, equalization and anomaly detection. They can compensate for drift but should not replace physical noise modelling. Training-data coverage, computational latency, explainability and failure under unfamiliar conditions require governance.

Optical equipment must comply with laser-safety requirements such as IEC 60825-1. Receiver production should also address electrostatic-discharge protection, hazardous-substance regulations, electromagnetic compatibility and secure control interfaces. Reported sensitivity results should state methods and assumptions transparently to avoid misleading system integrators.

11. Contemporary Developments

Current research includes germanium-on-silicon detectors, silicon–germanium APDs, waveguide photodiodes, uni-travelling-carrier structures and single-photon avalanche diode arrays. Co-designed photonic–electronic integration reduces parasitic capacitance and energy per bit. Digitally assisted TIAs and adaptive equalizers extend bandwidth, while probabilistic and machine-learning detectors address nonlinear or non-Gaussian channels.

These advances do not eliminate fundamental trade-offs among quantum efficiency, bandwidth, capacitance, multiplication noise, linearity and power consumption. Future receivers will increasingly be optimized as integrated optical, electronic and algorithmic systems.

Chapter Summary

Photodetectors convert photon flux into electrical current. Quantum efficiency determines the fraction of incident photons that produce collected carriers, while responsivity relates photocurrent to optical power. PIN photodiodes provide low-noise, linear detection; APDs add internal gain but introduce excess multiplication noise, gain–bandwidth constraints and bias complexity.

Receiver performance depends on shot noise, thermal noise, dark current, amplifier noise and filtering. A TIA must balance conversion gain, noise, capacitance, bandwidth and stability. Gaussian decision theory provides useful BER models, but unequal variances can produce two MAP likelihood crossings, so a single threshold is not universally exact. Sensitivity must be stated for defined BER, data rate, extinction ratio and operating conditions. Reliable design combines physical modelling, statistical detection, circuit optimization and margin-based system validation.

Key Terms

Avalanche photodiode: A photodiode using impact ionization to provide internal current gain.

Bit-error rate: The probability or measured fraction of incorrectly detected bits.

Dark current: Detector current present without intentional incident light.

Decision threshold: A decision-variable value separating symbol decisions.

Direct detection: Detection based primarily on received optical intensity rather than phase.

Equivalent noise bandwidth: The bandwidth of an ideal rectangular filter passing the same noise power as the actual filter.

Excess-noise factor: The factor describing additional APD noise caused by random multiplication.

Extinction ratio: The ratio of optical one-level power to zero-level power.

Impact ionization: Carrier multiplication caused by energetic carriers generating new electron–hole pairs.

MAP detection: Selection of the symbol with the greatest posterior probability.

Noise-equivalent power: Optical input power producing signal equal to noise in a specified bandwidth.

PIN photodiode: A p–intrinsic–n detector that converts absorbed photons into collected carriers.

Quantum efficiency: The ratio of collected photocarriers to incident photons.

Q-function: The upper-tail probability of a standard normal random variable.

Receiver sensitivity: Minimum average received optical power required for a specified performance.

Responsivity: Average photocurrent generated per unit incident optical power.

Shot noise: Fluctuation caused by the discrete and random arrival of charge carriers.

Thermal noise: Electrical noise arising from thermal agitation of charge carriers.

Transimpedance amplifier: A circuit that converts input current into output voltage.

Review Questions

Short Answer Questions

1. Define external quantum efficiency and responsivity.

2. Why does an APD have an excess-noise factor greater than unity?

3. Distinguish equivalent noise bandwidth from 3 dB bandwidth.

4. What physical mechanisms limit PIN photodiode bandwidth?

5. Why must extinction ratio be included in OOK sensitivity calculations?

Long Answer Questions

1. Derive the responsivity of a unity-gain photodiode from photon energy and quantum efficiency.

2. Compare PIN and avalanche photodiodes with respect to sensitivity, noise, bias, bandwidth and reliability.

3. Explain the major noise sources in a direct-detection receiver and develop their mean-square current expressions.

4. Derive the Gaussian MAP decision equation and explain why unequal variances may produce two thresholds.

5. Discuss the architecture and design trade-offs of a high-speed PIN-TIA optical receiver.

Analytical / Application Questions

1. A PIN detector has η = 0.75 at 1310 nm and receives −20 dBm. Calculate responsivity and photocurrent.

2. For μ₀ = 0.1 V, μ₁ = 0.7 V, σ₀ = σ₁ = 0.08 V and equal priors, calculate the optimum threshold and BER.

3. Design an OOK sensitivity budget for a receiver with finite extinction ratio, signal-dependent shot noise and a specified implementation margin.

Multiple Choice Questions with Answers

Question 1

A photodiode has quantum efficiency 0.80 at 1550 nm. Its responsivity is approximately:

A. 0.50 A/W

B. 0.80 A/W

C. 1.00 A/W

D. 1.55 A/W

Correct Answer: C. 1.00 A/W

Question 2

For the receiver in Worked Example 3, the combined rms shot and thermal noise is approximately:

A. 0.0221 μA

B. 0.221 μA

C. 0.0308 μA

D. 1.82 μA

Correct Answer: A. 0.0221 μA

Question 3

For equally likely Gaussian symbols with means 0.05 V and 0.50 V and equal standard deviations, the MAP threshold is:

A. 0.225 V

B. 0.250 V

C. 0.275 V

D. 0.347 V

Correct Answer: C. 0.275 V

Question 4

When two Gaussian symbol distributions have unequal variances, exact MAP detection may require:

A. No decision threshold

B. Two likelihood-crossing thresholds

C. A threshold fixed at zero

D. A threshold equal to the larger mean

Correct Answer: B. Two likelihood-crossing thresholds

Question 5

For an APD with M = 10 and k_A = 0.20, the excess-noise factor from the stated model is:

A. 1.90

B. 2.00

C. 3.52

D. 10.0

Correct Answer: C. 3.52

Question 6

Which expression gives the responsivity of a unity-gain photodiode?

A. R = ηqλ/(hc)

B. R = hc/(ηqλ)

C. R = ηhc/(qλ)

D. R = qhc/(ηλ)

Correct Answer: A. R = ηqλ/(hc)

Question 7

The rms thermal-noise current of an ideal resistor R over noise bandwidth B_n is:

A. √(2qIB_n)

B. √(4k_BT B_n/R)

C. √(4k_BTRB_n)

D. 4k_BT/(RB_n)

Correct Answer: B. √(4k_BT B_n/R)

Question 8

Which change generally reduces the sensitivity penalty caused by finite OOK extinction ratio?

A. Increasing zero-level optical power

B. Reducing detector responsivity

C. Increasing the ratio P₁/P₀

D. Increasing equivalent noise bandwidth

Correct Answer: C. Increasing the ratio P₁/P₀

Question 9

For ideal photon-counting OOK at BER = 10⁻⁹, the average detected photon requirement is approximately:

A. 1 photon per bit

B. 5 photons per bit

C. 10 photons per bit

D. 100 photons per bit

Correct Answer: C. 10 photons per bit

Question 10

The primary function of a transimpedance amplifier is to:

A. Convert detector current into voltage

B. Convert voltage into optical phase

C. Multiply photon energy

D. Eliminate shot noise

Correct Answer: A. Convert detector current into voltage