Optical Link Design and Performance Analytics

Introduction

Optical-link design converts service requirements—capacity, reach, error performance, availability and cost—into engineering decisions concerning fiber, transmitters, receivers, amplifiers, modulation, protection and operational margins. A viable link must satisfy several constraints simultaneously. Adequate received power alone does not guarantee operation if dispersion closes the eye, amplified spontaneous emission (ASE) reduces optical signal-to-noise ratio (OSNR), nonlinear interference (NLI) limits coherent performance, or failures violate the service-level agreement (SLA).

Modern design therefore combines analytical budgets, calibrated simulation, uncertainty analysis and optimization. Power and rise-time budgets remain important for intensity-modulation/direct-detection (IM-DD) systems, while coherent links are more commonly evaluated through OSNR, generalized signal-to-noise ratio (GSNR), bit-error ratio (BER), Q-factor and generalized mutual information (GMI). Reliability analysis must additionally represent switching delay, repair-duration distributions, common-cause failures and statistical uncertainty in rare-event estimates.

Learning Objectives

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

• construct optical power and engineering-margin budgets;
• develop rise-time and dispersion budgets under stated assumptions;
• quantify ASE, chromatic-dispersion, polarization-mode-dispersion and nonlinear penalties;
• relate launch power to the trade-off between ASE-limited and nonlinear operation;
• model steady-state and transient service availability;
• account for non-exponential repair times and correlated failures;
• evaluate rare outages using statistically defensible simulation;
• formulate capacity, reach, cost and reliability optimization problems; and
• interpret simulation results without confusing model precision with physical accuracy.

1. A Link-Design Framework

A defensible design normally follows an iterative sequence:

1. Specify the service: line rate, net information rate, latency, target BER, FEC, reach, availability and restoration requirements.

2. Define the physical route: fiber type, length, attenuation, dispersion, splice plan, connectors, amplifiers and shared-risk groups.

3. Select a transmission architecture: IM-DD or coherent detection, modulation format, symbol rate, channel spacing and amplification strategy.

4. Construct baseline budgets: received power, OSNR or GSNR, dispersion, rise time and latency.

5. Add penalties and margins: aging, temperature, repairs, component tolerances, filtering, crosstalk and nonlinear effects.

6. Evaluate reliability: failure, repair, switching and common-cause models.

7. Simulate uncertainty: component variation, traffic evolution, manufacturing spread and stochastic failures.

8. Optimize and validate: compare feasible alternatives by cost, capacity, energy and reliability, followed by field testing.

The process is iterative because improving one metric may degrade another. Increasing launch power improves the linear signal-to-noise ratio but eventually increases Kerr nonlinearity. Additional amplifiers extend reach but add ASE and cost. Protection improves availability but may duplicate equipment and consume spectrum.

2. Optical Power Budgets

2.1 Received-power equation

For an unamplified point-to-point link, all quantities expressed consistently in decibels give

\[

P_{R}=P_{T}+\sum_j G_j-\left(\alpha L+L_{sp}+L_c+L_{other}\right),

\]

where:

• \(P_T\) and \(P_R\) are transmitted and received powers in dBm;
• \(G_j\) is the gain of amplifier \(j\), in dB;
• \(\alpha\) is fiber attenuation in dB/km;
• \(L\) is route length in km;
• \(L_{sp}\), \(L_c\) and \(L_{other}\) are splice, connector and other insertion losses in dB.

The logarithmic equation must not be used with powers expressed in watts. In linear units, gains multiply and losses divide.

If the receiver requires at least \(P_{req}\) dBm under specified BER, modulation and FEC conditions, the raw power margin is

\[

M_{raw}=P_R-P_{req}.

\]

After allocating penalties and engineering allowances,

\[

M_{res}=M_{raw}-M_{disp}-M_{imp}-M_{age}-M_{repair}-M_{unc},

\]

where the terms respectively represent dispersion penalty, implementation penalty, aging, future repairs and residual uncertainty. A positive residual margin is necessary, but not sufficient, because some impairments do not behave as equivalent power losses.

Receiver sensitivity must be quoted with its measurement conditions. A value obtained back-to-back at a pre-FEC BER cannot automatically be applied after filtering, dispersion or wavelength multiplexing.

2.2 Worked power-budget example

Consider a 40 km unamplified link with:

• transmitter power: \(0\) dBm;
• fiber attenuation: \(0.20\) dB/km;
• connector loss: \(1.0\) dB;
• total splice loss: \(0.5\) dB;
• receiver requirement: \(-14\) dBm.

Fiber loss is

\[

0.20\times40=8.0\ \text{dB}.

\]

The total physical loss is therefore

\[

L_{tot}=8.0+1.0+0.5=9.5\ \text{dB},

\]

and

\[

P_R=0-9.5=-9.5\ \text{dBm}.

\]

The raw margin is

\[

M_{raw}=-9.5-(-14)=4.5\ \text{dB}.

\]

Suppose 0.8 dB is assigned to dispersion and implementation effects, 1.5 dB to aging, and 1.0 dB to repairs and future splices. The residual margin is

\[

M_{res}=4.5-0.8-1.5-1.0=1.2\ \text{dB}.

\]

This link passes the stated power criterion, although the relatively small residual margin should be tested against uncertainty in actual component losses.

3. Rise-Time and Dispersion Budgets

3.1 Root-sum-square approximation

For a bandwidth-limited IM-DD link whose individual impulse responses are approximately Gaussian and statistically independent, the system 10–90% rise time may be approximated by

\[

t_{sys}=\sqrt{t_T^2+t_R^2+t_{CD}^2+t_{PMD}^2+t_{modal}^2},

\]

where the terms represent transmitter, receiver, chromatic-dispersion, polarization-mode-dispersion and modal contributions.

This root-sum-square rule is not universal. It becomes unreliable for strongly non-Gaussian responses, severe filtering, chirped transmitters, nonlinear propagation or significant intersymbol interference. Such cases require waveform simulation or measurement.

For conventional NRZ binary signaling, a commonly used engineering criterion is

\[

t_{sys}\lesssim \frac{0.7}{B},

\]

where \(B\) is bit rate in bit/s. The coefficient is a rule of thumb rather than a fundamental limit; transmitter extinction ratio, receiver equalization and target BER affect the permissible value.

3.2 Chromatic dispersion

For a narrow optical spectrum over a region in which dispersion is approximately constant, temporal spreading is estimated by

\[

\Delta t_{CD}=|D|L\Delta\lambda,

\]

where \(D\) is in ps/(nm·km), \(L\) in km and spectral width \(\Delta\lambda\) in nm. The result is in ps. The definition of spectral width—RMS, full width at half maximum or another measure—must be consistent with the pulse model.

For a modulated laser with chirp, the effective dispersion penalty depends on the time-dependent frequency as well as the nominal optical linewidth. The simple expression can then underpredict pulse distortion.

Chromatic dispersion can also be described through the propagation constant:

\[

\beta_2=-\frac{\lambda^2}{2\pi c}D,

\]

provided units are converted consistently. Coherent receivers estimate and digitally compensate deterministic accumulated chromatic dispersion, but compensation does not recover information lost through inadequate sampling, excessive noise, analog bandwidth limitations or nonlinear signal-noise interaction.

3.3 Polarization-mode dispersion

For first-order PMD in a long, randomly coupled fiber, mean differential group delay is commonly modeled as

\[

\tau_{PMD}=D_{PMD}\sqrt{L},

\]

where \(D_{PMD}\) is in ps/√km. PMD is stochastic and varies with temperature, stress and wavelength. The square-root law is an ensemble relation, not a deterministic upper bound. Outage calculations should use an appropriate delay distribution or measured PMD statistics.

3.4 Numerical rise-time assessment

For the 40 km link, suppose:

• \(t_T=25\) ps;
• \(t_R=30\) ps;
• \(D=17\) ps/(nm·km);
• \(\Delta\lambda=0.05\) nm;
• \(D_{PMD}=0.10\) ps/√km;
• modal dispersion is negligible.

Then

\[

t_{CD}=17\times40\times0.05=34\ \text{ps},

\]

and

\[

t_{PMD}=0.10\sqrt{40}=0.632\ \text{ps}.

\]

Thus

\[

t_{sys}=\sqrt{25^2+30^2+34^2+0.632^2}=51.8\ \text{ps}.

\]

At 10 Gbit/s, the approximate NRZ limit is

\[

\frac{0.7}{10^{10}}=70\ \text{ps}.

\]

The design passes this approximate rise-time test. It should still be verified by an eye-diagram or BER simulation because the dispersion term has been represented by a simplified equivalent broadening.

4. Noise, OSNR and Coherent Performance

4.1 Amplified spontaneous emission

An optical amplifier adds ASE. For an amplifier with linear gain \(G\), spontaneous-emission factor \(n_{sp}\), optical frequency \(\nu\), and measurement bandwidth \(B_o\), the total ASE power across two polarizations is commonly approximated by

\[

P_{ASE}=2n_{sp}h\nu(G-1)B_o.

\]

The expression assumes an idealized flat gain and specified optical bandwidth. Cascaded systems require propagation of signal and noise through each subsequent gain and loss element. The high-gain noise-figure approximation \(F\approx2n_{sp}\) should not be treated as exact at low gain.

OSNR is

\[

OSNR=\frac{P_{signal}}{P_{ASE,ref}},

\]

where both powers must refer to the same location and the noise must be measured in a declared reference bandwidth, often 0.1 nm. OSNR values cannot be compared legitimately when bandwidth, polarization convention or measurement method differs.

4.2 BER, Q-factor and information metrics

For ideal binary detection with Gaussian decision statistics and equal variances, a Q-factor may be defined as

\[

Q=\frac{\mu_1-\mu_0}{\sigma_1+\sigma_0},

\]

with

\[

BER\approx\frac{1}{2}\operatorname{erfc}\left(\frac{Q}{\sqrt{2}}\right).

\]

This relationship is conditional on the assumed statistics and detector. It is not generally valid for all coherent modulation formats, nonlinear channels or mismatched DSP.

For coded coherent systems, pre-FEC BER alone may not predict decoding success. GMI or normalized GMI is often more informative because it accounts for soft-decision information. Vendor thresholds must be matched to the actual modulation, FEC code, decoder and required post-FEC error rate.

5. Nonlinear Transmission Penalties

5.1 Kerr effects

The fiber refractive index depends weakly on optical intensity. Consequences include self-phase modulation, cross-phase modulation and four-wave mixing. Their impact depends on launch power, effective area, dispersion, span length, channel spacing, modulation and amplification architecture.

In dispersion-uncompensated coherent systems, a calibrated Gaussian-noise-type model often represents per-channel nonlinear noise as

\[

P_{NLI}=\eta P_{ch}^{3},

\]

where \(P_{ch}\) is per-channel launch power in linear units and \(\eta\) is a system-dependent coefficient. The cubic law is a model approximation, not a universal physical law. It may require correction for modulation format, stimulated Raman scattering, spatial channels, dispersion maps and nonlinear signal-noise interactions.

A 2 dB reduction in launch power changes linear power by \(10^{-2/10}=0.631\). Under the cubic model, NLI changes by

\[

(0.631)^3=10^{-6/10}\approx0.251.

\]

Thus reducing launch power from 0 dBm to −2 dBm reduces modeled NLI power to approximately 25.1% of its former value. Signal power also decreases, so the overall SNR must be recalculated rather than inferred from NLI alone.

5.2 Optimum launch power

In a simplified single-channel representation,

\[

SNR(P)=\frac{P}{N_{ASE}+\eta P^3},

\]

where \(N_{ASE}\) is assumed independent of launch power. Differentiating gives

\[

\frac{d(SNR)}{dP}=\frac{N_{ASE}-2\eta P^3}{(N_{ASE}+\eta P^3)^2}.

\]

The stationary maximum occurs when

\[

P_{opt}=\left(\frac{N_{ASE}}{2\eta}\right)^{1/3}.

\]

At this point, \(N_{ASE}=2P_{NLI}\). This result applies only to the stated model. Real systems may have transceiver noise, power-dependent amplifier behavior, spectral loading changes and implementation penalties that shift the optimum.

A useful effective metric is

\[

\frac{1}{GSNR}\approx\frac{1}{SNR_{ASE}}+\frac{1}{SNR_{NLI}}+\frac{1}{SNR_{TRX}},

\]

when the represented noise contributions can reasonably be treated as independent additive impairments. Deterministic filtering penalties and cycle slips should not automatically be inserted as Gaussian noise terms.

6. Penalty and Margin Engineering

Margins should reflect uncertainty rather than conceal incomplete design. Typical categories include:

• design margin: model and parameter uncertainty;
• aging margin: laser degradation, connector contamination and component drift;
• repair margin: additional splices or route modifications;
• operational margin: wavelength loading, temperature and gain-control variation;
• restoration margin: performance on a longer or differently loaded backup route.

Blindly adding every quoted penalty in decibels can be incorrect. Independent noise sources should first be combined in linear noise power. Worst-case insertion losses may be added in dB, while random tolerances may require statistical convolution or Monte Carlo analysis. Double counting occurs when a vendor’s required OSNR already includes an implementation penalty that is added again by the network designer.

Best practice is to maintain a margin register specifying the physical cause, statistical basis, confidence level, owner and validation method of every allowance.

7. Availability and Outage Modeling

7.1 Steady-state availability

For an alternating renewal process with finite mean up-time \(E[U]\) and repair time \(E[R]\), long-run availability is

\[

A=\frac{E[U]}{E[U]+E[R]}.

\]

If mean time between failures denotes mean operating time, this becomes

\[

A=\frac{MTBF}{MTBF+MTTR}.

\]

The ratio does not require exponential up- and down-times, although exponential assumptions are required for a simple continuous-time Markov chain with constant transition rates.

For rare failures with rate \(\lambda\) per year and mean repair duration \(r\) hours,

\[

U\approx\frac{\lambda r}{8760},

\]

where \(U=1-A\). The approximation neglects overlapping failures and is suitable when unavailability is small.

For statistically independent series elements,

\[

A_{series}=\prod_i A_i.

\]

For two independent parallel elements that both must fail to interrupt service,

\[

U_{parallel}=U_1U_2.

\]

Independence must be demonstrated, not presumed.

7.2 Switching time and transient interruption

Protection does not imply uninterrupted service. Failure detection, hold-off timers, signaling, optical-switch movement, laser settling and routing convergence create a switching interval. If switching events occur at rate \(\lambda_s\) and each causes interruption \(t_s\), the average fractional unavailability contributed by successful switching is approximately

\[

U_{sw}=\frac{\lambda_s t_s}{T_y},

\]

where \(T_y\) is one year expressed in the same time unit.

Average availability alone may hide SLA violations. For example, many 100 ms interruptions may produce little annual downtime but disrupt latency-sensitive industrial control. Availability studies should therefore report interruption frequency and duration distributions, not only a percentage.

7.3 Repair-duration distributions

Two repair processes with identical MTTR can have different outage risk. Fiber cuts may have a right-skewed duration distribution because of permit delays, weather and inaccessible ducts. Lognormal, Weibull or empirical distributions may be more credible than an exponential model.

Distribution choice matters especially for protected systems: the probability that a second route fails before the first is restored depends on the entire repair-time distribution. Non-exponential models can be analyzed using semi-Markov methods, renewal simulation or discrete-event simulation. A Markov chain is appropriate only when state residence times are exponential or when non-exponential durations are represented by justified phase-type approximations.

7.4 Correlated and common-cause failures

Apparently diverse paths may share ducts, bridges, power supplies, control software, buildings or flood zones. Such shared-risk link groups create correlated failures. Let \(U_c\) denote common-cause unavailability. For rare events, a two-route approximation is

\[

U_{service}\approx U_1U_2+U_c+U_{sw},

\]

provided terms are small and double-counting is negligible. More detailed fault trees, Bayesian networks, copulas or event-based simulations are needed when dependence is complex.

Geographic information systems should be used to verify physical diversity. Logical route diversity shown in a network-management system is insufficient if both fibers occupy the same trench.

7.5 Rare-event simulation precision

For ordinary Monte Carlo estimation of outage probability \(p\), using \(N\) independent trials, the estimator \(\hat p\) has variance

\[

\operatorname{Var}(\hat p)=\frac{p(1-p)}{N}.

\]

For small \(p\), relative standard error is approximately

\[

\frac{SE(\hat p)}{p}\approx\frac{1}{\sqrt{Np}}.

\]

If \(p=10^{-6}\), achieving about 10% relative standard error requires \(Np\approx100\), or approximately \(10^8\) independent trials. A simulation that observes no outage does not prove zero risk.

Importance sampling, splitting, subset simulation and conditional Monte Carlo can improve efficiency. Importance sampling must include likelihood-ratio weights; otherwise the estimate is biased toward the deliberately over-sampled failure conditions. Results should include confidence intervals, effective sample size, convergence diagnostics and model-validation evidence.

8. Simulation and Multiobjective Optimization

8.1 Simulation hierarchy

Different questions require different model resolution:

• analytical screening: rapid elimination using loss, dispersion and approximate OSNR equations;
• link-level simulation: waveform propagation, filtering, DSP, BER or GMI estimation;
• network simulation: routing, wavelength assignment, restoration and traffic blocking;
• reliability simulation: stochastic failures, repair logistics and switching;
• techno-economic simulation: capital cost, operating cost, energy and upgrade timing.

Model calibration should use component data and field measurements. Sensitivity analysis should vary attenuation, noise figure, nonlinear coefficient, repair time, traffic growth and equipment price. Validation against only one operating point is inadequate.

8.2 Optimization formulation

Let design vector \(x\) contain modulation format, symbol rate, launch power, channel count, amplifier spacing, FEC, route and protection architecture. A multiobjective problem may be written as

\[

\min_x\{C_{life}(x),E(x),P_{out}(x)\},

\]

subject to

\[

C_{net}(x)\ge C_{demand},\quad GSNR(x)\ge GSNR_{req},

\]

\[

A(x)\ge A_{SLA},\quad L_{latency}(x)\le L_{max},

\]

plus spectrum, equipment and regulatory constraints.

Because cost, energy and reliability conflict, the solution is often a Pareto frontier rather than a single optimum. Robust optimization protects against bounded uncertainty, while stochastic optimization minimizes expected cost or risk under probability distributions. Chance constraints such as

\[

\Pr[GSNR\ge GSNR_{req}]\ge0.999

\]

are useful only when the underlying uncertainty model is credible.

Surrogate models and machine learning can accelerate computationally expensive physical simulations. They should be constrained to validated domains, monitored for extrapolation and accompanied by uncertainty estimates. An accurate surrogate of an incorrect physical model remains incorrect.

9. Integrated Illustrative Case: Regional Enterprise Link

Background

This fictional case considers a 40 km enterprise link carrying 10 Gbit/s NRZ traffic, with a future coherent upgrade under evaluation. The operator requires positive physical-layer margin and improved service availability through route protection.

Problem

The engineering team must verify loss and rise time, assess the effect of launch-power reduction on modeled NLI for the upgrade, and estimate protected service availability.

Analysis and approach

The physical data are:

• attenuation \(0.20\) dB/km;
• 40 km route length;
• connector loss \(1.0\) dB;
• splice loss \(0.5\) dB;
• transmitter power \(0\) dBm;
• receiver requirement \(-14\) dBm.

The fiber loss is 8.0 dB and total physical loss is 9.5 dB. Received power is −9.5 dBm, giving 4.5 dB raw margin and 1.2 dB residual margin after the stated 3.3 dB allowances.

The calculated rise time is 51.8 ps, below the approximate 70 ps criterion at 10 Gbit/s. Waveform testing is nevertheless specified because linewidth and chirp can vary with temperature.

For the future coherent system, a calibrated planning model predicts excessive nonlinear contribution at 0 dBm per channel. Reducing launch power to −2 dBm changes cubic NLI by

\[

10^{-6/10}=0.251.

\]

The team does not claim an automatic SNR improvement; it recomputes GSNR because signal power also falls by a factor of 0.631.

For availability, each diverse route has failure rate 0.4 per year and mean repair duration 8 h. Its rare-event unavailability is

\[

U_r\approx\frac{0.4\times8}{8760}=3.65\times10^{-4}.

\]

Independent dual-route overlap is approximately

\[

U_r^2=1.33\times10^{-7}.

\]

A common-cause hazard occurs at 0.02 per year with mean restoration time 8 h:

\[

U_c\approx\frac{0.02\times8}{8760}=1.83\times10^{-5}.

\]

With a 50 ms interruption for each of approximately 0.4 protection switches per year,

\[

U_{sw}=\frac{0.4\times0.05}{31{,}536{,}000}=6.34\times10^{-10}.

\]

Hence, under the rare-event approximation,

\[

U_{service}\approx1.33\times10^{-7}+1.83\times10^{-5}+6.34\times10^{-10}

=1.84\times10^{-5}.

\]

Availability is approximately 99.99816%, corresponding to about 9.7 minutes of expected downtime per year. Common-cause risk dominates independent dual-route failure.

Outcome

The 10 Gbit/s link passes initial power and rise-time screening. The coherent upgrade requires launch-power optimization using GSNR rather than a power-only rule. Route protection substantially improves availability, but additional investment should target common-cause hazards, such as shared facilities and power, rather than merely reducing optical switching time.

Learning points

• Route length and attenuation must be arithmetically consistent.
• NLI reduction does not equal total SNR improvement.
• Small switching delays may matter to applications even when annual downtime is negligible.
• Common-cause failures can dominate nominally protected architectures.
• Analytical results should be followed by measurement and uncertainty analysis.

10. Engineering Practice, Governance and Emerging Directions

Optical design affects safety, contractual compliance and public infrastructure. High-power optical systems require laser-safety procedures, connector inspection and controlled access. Engineers must comply with applicable laser classifications, electromagnetic compatibility requirements, rights-of-way rules and environmental obligations.

SLA reporting should define whether planned maintenance, switching transients and partial-capacity degradation count as outages. Selectively excluding unfavorable events is both technically misleading and ethically problematic. Failure databases should preserve provenance and avoid exposing sensitive network topology.

Contemporary developments include probabilistic constellation shaping, flexible-rate transponders, spatial-division multiplexing, coherent pluggables, open line systems and telemetry-assisted control. AI-based power optimization and anomaly detection can improve operation, but automated changes require bounded control ranges, rollback mechanisms, audit logs and human accountability. Security is integral: compromised telemetry or software-defined control can invalidate both performance and availability assumptions.

Chapter Summary

Optical-link design is a multi-constraint process integrating power, bandwidth, noise, nonlinearity, reliability and economics. Power budgets compare received power with a condition-specific receiver requirement and retain explicit allowances for aging, repairs and uncertainty. Rise-time budgets provide useful screening for IM-DD systems when their Gaussian and independence assumptions are reasonable. Coherent systems are more appropriately assessed through OSNR, GSNR, BER and information-based metrics.

ASE increases with amplifier gain and bandwidth, while Kerr nonlinear interference is often approximated as cubic in channel launch power. This creates an optimum launch-power region rather than a rule that more power is always beneficial. Availability analysis must extend beyond steady-state averages to switching delays, repair-time distributions, common-cause failures and interruption frequency. Rare-event simulation requires confidence analysis and variance-reduction methods when outage probabilities are extremely small. Simulation and multiobjective optimization finally enable explicit trade-offs among capacity, reach, cost, energy and reliability.

Key Terms

ASE: Broadband spontaneous-emission noise added by an optical amplifier.

Availability: Long-run fraction of time for which a service is operational.

Chance constraint: A design constraint required to hold with a specified probability.

Chromatic dispersion: Wavelength-dependent group delay that broadens modulated optical signals.

Common-cause failure: A single event capable of disabling multiple nominally redundant elements.

Engineering margin: Reserved performance allowance for uncertainty, aging and future change.

GMI: Generalized mutual information used to assess achievable coded performance with soft decisions.

GSNR: Effective signal-to-noise metric combining relevant additive noise contributions.

Importance sampling: Rare-event simulation method that changes the sampling distribution and applies likelihood weights.

Insertion loss: Optical power reduction caused by a component or connection.

MTTR: Mean time required to restore a failed element or service.

Nonlinear interference: Signal distortion represented as noise arising from fiber nonlinearities.

OSNR: Ratio of optical signal power to ASE noise power in a specified reference bandwidth.

Outage probability: Probability that service or performance falls below a defined requirement.

PMD: Stochastic pulse spreading caused by polarization-dependent group delay.

Power budget: Accounting of transmitter power, gain, loss, receiver requirement and margin.

Residual margin: Margin remaining after all declared penalties and allowances are deducted.

Rise-time budget: Combination of transmitter, channel and receiver temporal responses.

Shared-risk link group: Links exposed to a common physical or operational hazard.

Switching time: Interruption interval associated with detecting a failure and transferring service.

Review Questions

Short Answer Questions

1. Distinguish raw power margin from residual engineering margin.

2. Under what assumptions is root-sum-square rise-time addition appropriate?

3. Why must an OSNR value include its reference bandwidth?

4. What is a common-cause failure in a protected optical network?

5. Why can ordinary Monte Carlo simulation be inefficient for rare outages?

Long Answer Questions

1. Explain a complete workflow for optical-link design from service specification to field validation.

2. Derive the optimum launch power for \(SNR=P/(N_{ASE}+\eta P^3)\) and discuss the model’s limitations.

3. Compare power, rise-time, OSNR, GSNR, BER and GMI as optical-system design metrics.

4. Discuss how switching time and non-exponential repair distributions affect availability analysis.

5. Explain how robust, stochastic and multiobjective optimization support optical-network planning.

Analytical / Application Questions

1. A 60 km link has attenuation 0.22 dB/km, 1.4 dB connector loss, 0.8 dB splice loss, transmitter power 2 dBm and receiver requirement −16 dBm. Calculate received power and raw margin.

2. For \(D=16.5\) ps/(nm·km), \(L=80\) km and \(\Delta\lambda=0.04\) nm, calculate chromatic-dispersion broadening and assess its significance relative to a 100 ps bit period.

3. Design a simulation study for a protected link with correlated geographic hazards, lognormally distributed repair duration and target outage probability below \(10^{-6}\). Specify the estimator, precision target and variance-reduction approach.

Multiple Choice Questions with Answers

Question 1

A 40 km fiber has attenuation of 0.20 dB/km, connector loss of 1.0 dB and splice loss of 0.5 dB. What is its total physical loss?

A. 8.0 dB

B. 8.5 dB

C. 9.5 dB

D. 17.5 dB

Correct Answer: C. 9.5 dB

Question 2

If transmitter power is 0 dBm and total link loss is 9.5 dB, what is the received power?

A. −14.0 dBm

B. −9.5 dBm

C. 0.0 dBm

D. 9.5 dBm

Correct Answer: B. −9.5 dBm

Question 3

Under the model P_NLI = ηP³, reducing launch power by 2 dB multiplies nonlinear interference power by approximately:

A. 0.251

B. 0.501

C. 0.631

D. 0.794

Correct Answer: A. 0.251

Question 4

For a 10 Gbit/s NRZ link, the approximate 0.7/B rise-time criterion gives which maximum system rise time?

A. 7 ps

B. 10 ps

C. 70 ps

D. 100 ps

Correct Answer: C. 70 ps

Question 5

For SNR = P/(N_ASE + ηP³), which launch power maximizes SNR under the stated model?

A. (N_ASE/η)^(1/2)

B. (N_ASE/(2η))^(1/3)

C. (2N_ASE/η)^(1/3)

D. N_ASE/(3η)

Correct Answer: B. (N_ASE/(2η))^(1/3)

Question 6

Which statement about A = E[U]/(E[U] + E[R]) is correct?

A. It is valid only for deterministic repair times.

B. It requires all failures to be correlated.

C. It is a long-run alternating-renewal result when the means are finite.

D. It excludes repair duration from availability.

Correct Answer: C. It is a long-run alternating-renewal result when the means are finite.

Question 7

For a rare outage probability p, ordinary Monte Carlo has approximate relative standard error equal to:

A. 1/(Np)

B. 1/√(Np)

C. p/√N

D. √(N/p)

Correct Answer: B. 1/√(Np)

Question 8

Which factor most directly invalidates the formula U_parallel = U1U2?

A. Independent route failures

B. Small route unavailability

C. A shared duct exposed to one excavation

D. Equal repair times

Correct Answer: C. A shared duct exposed to one excavation

Question 9

Why must OSNR be reported with a reference bandwidth?

A. Signal power is always proportional to reference bandwidth.

B. Measured ASE noise power depends on the optical measurement bandwidth.

C. Fiber attenuation becomes zero at narrow bandwidths.

D. Reference bandwidth determines route availability.

Correct Answer: B. Measured ASE noise power depends on the optical measurement bandwidth.

Question 10

In the illustrative protected-link case, which contribution dominates service unavailability?

A. The 50 ms optical switching interval

B. Independent simultaneous route failures

C. Common-cause hazards

D. Chromatic-dispersion broadening

Correct Answer: C. Common-cause hazards