Convolutional Codes, Trellis Diagrams, Viterbi Decoding Algorithm, and Modern FEC

Unit 8 — Practical link and network engineering

Why this unit exists

The unit centers on emergency communications service, where a regional network carries mixed urgent control traffic and bulk data. Its controlling difficulty is that latency tails and observed delivery reliability are contractual requirements. The four lessons isolate different parts of the mechanism, yet the delivered system must satisfy all of them at once. You will link equations to implementation, design measurements and deliberately test situations in which a convenient idealization becomes misleading.

Concept sequence

Lesson 8.1 — Adaptive modulation and coding

The most aggressive modulation is rarely optimal for delivered payload. Higher spectral efficiency increases raw bits per symbol but also raises the required signal quality for a specified error target. The scheduler must weigh payload goodput, retransmission cost, interference exposure and channel-estimation delay. A threshold table can become unstable if it switches schemes on noisy near-threshold measurements. Hysteresis and smoothing help, but slow adaptation may lag a rapidly changing link. An outer-loop controller can adjust thresholds based on observed acknowledgments rather than assuming perfect channel calibration.

Key relation: A link-adaptation controller selects a modulation/coding scheme using estimated channel quality, predicted packet error and implementation-specific throughput tables.

Design boundary: Goodput figures are dependent on traffic pattern, failure correlation, headers and retransmission details. A controller optimized using mean SNR can fail severely during rare fades.

Lesson 8.2 — ARQ, hybrid ARQ and retransmission

Automatic repeat request uses error detection and feedback to resend missing or corrupted frames. Stop-and-wait schemes are simple but can waste a long round-trip time. Sliding windows allow several frames in flight; sequence numbers and buffering manage reordering. Hybrid ARQ combines forward-error correction with retransmission and may soft-combine repeated observations or transmit incremental redundancy. Receiver memory, feedback error and delay budgets constrain practical designs. Stronger FEC reduces retries but may reduce raw throughput, so end-to-end goodput is a coupled design problem.

Key relation: For independent packet attempts with success probability p, expected attempts to success are 1/p; HARQ may combine failed transmissions so attempts are not statistically identical.

Design boundary: A geometric retry model is wrong when fading remains correlated across attempts or incremental redundancy changes conditional success probability after each transmission.

Lesson 8.3 — Queueing, latency and quality of service

A communication system must deliver useful data on time, not merely maximize peak physical-layer rate. Packet queues arise when bursts arrive faster than a scheduler can transmit them. Average service rate exceeding average arrival rate is necessary for long-term stability in basic models but says little about tail latency during bursts. Real packet service times are not always exponential, so the M/M/1 formula is an instructive benchmark rather than a blanket guarantee. Prioritization helps urgent traffic but can starve other flows; admission control and shaping influence both reliability and fairness.

Key relation: For a stable M/M/1 queue with arrival rate λ and service rate μ>λ, mean system time is W=1/(μ-λ), and mean number in system L=λW.

Design boundary: Little's law is broadly valid for steady systems with consistent counting, but the exact M/M/1 mean waiting-time formula depends on its arrival and service assumptions. Report the 95th percentile rather than only mean latency for interactive services.

Lesson 8.4 — Validation, reproducibility and deployment boundaries

A communication study should separate physical model assumptions from claims about deployed hardware or real traffic. Use a fixed seed to reproduce experiments, but repeat across seeds to measure uncertainty. Log rate, coding parameters, received power reference, noise bandwidth, channel model, packetization and error policy. Establish unit tests for basic identities, numerical conservation checks and negative controls for failure cases. Report confidence intervals when estimating rare error rates. Regulatory compliance and product certification need actual measurements and applicable jurisdiction-specific requirements, not a synthetic chart or a clean source-code build.

Key relation: An end-to-end measured goodput G is accepted user payload divided by elapsed observation time; simulated symbol BER, packet error rate, EVM and goodput answer different questions.

Design boundary: A visually attractive plot can hide a normalization bug; in simulation, independent recomputation of selected numerical points is essential. Correlation across trials invalidates common binomial uncertainty calculations.

Integrated case study

An engineer reviews the architecture for emergency communications service. The first review identifies the performance or integrity quantity at the receiver. The second asks which subsystem can change that quantity. The third checks the impact on user payload, energy consumption, delay and operating cost. A favorable conclusion requires comparable conditions; changing both the channel model and signal power while comparing decoders makes it impossible to attribute improvement to one change.

Three-stage assessment plan

1. Analytical baseline. Derive a transparent relation using one of the four listed key models. Describe the allowed parameter region, units and what was held constant in the comparison. For emergency communications service, decide which model variable is realistically measurable and which must be inferred.

2. Synthetic controlled test. Isolate one impairment, vary its magnitude and compare to the baseline. Re-run with a different seed and a second numerical implementation. Build a plot with labeled axes and, for error probabilities, uncertainty intervals.

3. Engineering hand-off. Replace the idealized assumption with a more plausible impairment relevant to a regional network carries mixed urgent control traffic and bulk data. State which earlier conclusion remains reliable and which needs further data. Record what must be checked before a prototype is trusted outside the laboratory.

Unit questions

• Which of the four mechanisms is most likely to control acceptance for emergency communications service, and what measured evidence would establish that claim?

• How would the design fail if the assumption “Goodput figures are dependent on traffic pattern, failure correlation, headers and retransmission details. A controller optimized using mean SNR can fail severely during rare fades.” were ignored?

• Identify a numerical sanity check, an implementation test, and a separate real-world validation requirement.

• Where would additional redundancy help and where might it merely increase latency or occupy already-limited bandwidth?

Suggested learning order

Read the four lessons in order; complete the simulation, then submit the project with a result table and the associated limitations.

Unit 8, Lesson 1: Adaptive modulation and coding

Module: Practical link and network engineering. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain adaptive modulation and coding from a physical or algorithmic model rather than memorizing an isolated equation.

• Distinguish assumptions, symbols, measured quantities and synthesized values in a complete communications link.

• Perform an independent limiting-case check and identify an operating region where the model fails.

• Design and document a reproducible synthetic experiment that tests a stated prediction.

System context and motivating decision

Consider emergency communications service: a regional network carries mixed urgent control traffic and bulk data. In this setting, latency tails and observed delivery reliability are contractual requirements. The engineer therefore needs a model that is useful at the right reference plane, not an attractive formula copied from a different link. A quantity measured at the transmitter cannot automatically be used as if it were the receive-filter output, and uncoded symbol reliability cannot be silently treated as packet reliability. The purpose of this lesson is to connect adaptive modulation and coding with observable decisions in this chain.

Physical or algorithmic explanation

The most aggressive modulation is rarely optimal for delivered payload. Higher spectral efficiency increases raw bits per symbol but also raises the required signal quality for a specified error target. The scheduler must weigh payload goodput, retransmission cost, interference exposure and channel-estimation delay. A threshold table can become unstable if it switches schemes on noisy near-threshold measurements. Hysteresis and smoothing help, but slow adaptation may lag a rapidly changing link. An outer-loop controller can adjust thresholds based on observed acknowledgments rather than assuming perfect channel calibration.

Core mathematical relationship

A link-adaptation controller selects a modulation/coding scheme using estimated channel quality, predicted packet error and implementation-specific throughput tables.

Meaning and conventions

The central expression above describes adaptive modulation and coding. Before calculating, identify the set of variables, their units and whether the expression applies to discrete or continuous time. When the relevant quantities are probabilities, confirm the conditioning event and whether samples are independent. When complex amplitudes occur, remember that a conjugate, not a real transpose, normally defines an energy inner product. Where a ratio of powers is stated in decibels, 10 log10 rather than 20 log10 is the appropriate conversion; for voltage magnitudes with identical impedances, a 20 log10 form may be suitable. This model is one part of the signal processing chain, not a substitute for a performance measurement.

Derive, calculate and inspect

First mark the source of the input and output quantities in A link-adaptation controller selects a modulation/coding scheme using estimated channel quality, predicted packet error and implementation-specific throughput tables.. State which symbols are known, which must be estimated and which are chosen by design. Translate the equation into a reproducible step-by-step computation, then test zero or very small noise, an intermediate operating point and a stressed case. A result should remain physically interpretable under these changes. Where the relationship is approximate, determine the approximation error before interpreting small performance improvements as significant. Never add gains or losses twice because the same physical effect appears in two bookkeeping categories.

Fully explained example and verification

Suppose scheme A sends 1 Mbit/s with 1% packet failure, producing approximate one-attempt goodput 0.99 Mbit/s; scheme B sends 2 Mbit/s with 40% packet failure, yielding 1.2 Mbit/s before retransmission overhead. B seems better on this narrow metric, but if failed packets trigger long retries and latency constraints, A may be preferred.

How to verify: Recompute the numeric statement from the specified inputs without reading a plotted curve. Re-run after increasing the most influential input by 10%, and explain the direction and approximate size of the change. Then compare that outcome with the limiting case for the model. A plausible number is not sufficient if the observation plane, normalization or statistical assumptions have changed.

Boundary conditions and common mistakes

Goodput figures are dependent on traffic pattern, failure correlation, headers and retransmission details. A controller optimized using mean SNR can fail severely during rare fades.

An additional failure mode in emergency communications service arises when the local mechanism is interpreted as an end-to-end guarantee. For example, packets may be rejected by framing or replay checks even after an error-correcting decoder outputs a plausible bit sequence; conversely, a successful frame may be too late to serve a time-critical application. To separate causes, log the inputs and outputs of each major subsystem, retain the raw observation when possible, and name the exact stage at which a failure is declared.

Guided computational experiment

Create a synthetic channel-quality trace with three modulation profiles and measured success curves. Compare fixed selection, instantaneous threshold selection and hysteresis using cumulative payload and outage.

Reproducibility protocol. Record the independent variables, their increments and physical units; state random seeds, number of simulated trials and target metrics. Use two numerical implementations for at least one predicted point. When drawing a probability curve, use enough trials to estimate its uncertainty and report the confidence interval; small error counts produce unstable estimates. When a parameter sweep crosses a model limit, annotate the plot rather than truncating inconvenient results.

Problem set with targeted response guidance

Question 1. Reference planes and dimensional consistency

Draw the component-level path through a regional network carries mixed urgent control traffic and bulk data. Label exactly where the inputs to “A link-adaptation controller selects a modulation/coding scheme using estimated channel quality, predicted packet error and implementation-specific throughput tables.” are acquired or estimated. Construct a three-column table of quantity, unit, and uncertainty source. Identify one tempting but invalid substitution and explain which physical interface it crosses.

Answer guidance: A complete solution has a block diagram and a clear distinction between physical measurement and algorithmic estimate. For adaptive modulation and coding, justify every normalization and explicitly say what is held constant when comparing two conditions.

Question 2. Rebuild the numerical case

Independently reconstruct the worked calculation for adaptive modulation and coding. Change the leading numerical parameter by 20%, holding other inputs fixed, and explain whether the result changes linearly or nonlinearly. Supply every intermediate numerical step.

Answer guidance: The answer must reproduce the worked statement under its stated assumptions, account for units, and include an inverse or limiting check. If a result disagrees, trace the exact operation; do not round intermediate values until the final display.

Question 3. Competing design changes

For emergency communications service, propose two ways to improve the outcome represented by A link-adaptation controller selects a modulation/coding scheme using estimated channel quality, predicted packet error and implementation-specific throughput tables.: one changes a transmitter or source parameter, the other changes receiver processing. List their respective energy, implementation, latency and spectral costs.

Answer guidance: Grade the response on tradeoffs, not a single optimization metric. For this topic, specify which impairment each intervention addresses and whether a gain in one stage increases overhead elsewhere.

Question 4. Break the assumptions

Create a deliberately unrealistic operating point for adaptive modulation and coding and explain a specific broken assumption. Then propose a synthetic observation that would reveal the failure. Describe what a poorly documented analysis might incorrectly conclude.

Answer guidance: The response must identify an assumption explicitly and design a measurement that detects the discrepancy. Replacing one uncertain parameter with a convenient nominal value is not an assumption test.

Further engineering synthesis

To write a defensible report for adaptive modulation and coding, separate evidence into three columns: a derivation, a reproducible numerical implementation and a system-level observation. The derivation establishes the modeled relationship; simulation checks implementation and explore cases that are hard to calculate by hand; field observation tests the relevance of modeling assumptions. Failure of a field measurement to match the derivation may reveal either a coding bug or a missing physical effect. Avoid the reflex to adjust free parameters until curves agree, because that conceals rather than explains uncertainty.

Finally trace the effect of adaptive modulation and coding onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for emergency communications service. Describe which of these metrics has a monotonic response to parameter changes and which displays a tradeoff. Explain the direction of the tradeoff with one numerical or plotted piece of evidence; where evidence is absent, label the claim as a hypothesis. Every numerical conclusion should be reproducible from recorded inputs.

Cross-reference: See the model for Unit 8, Unit 8 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 8, Lesson 2: ARQ, hybrid ARQ and retransmission

Module: Practical link and network engineering. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain arq, hybrid arq and retransmission from a physical or algorithmic model rather than memorizing an isolated equation.

• Distinguish assumptions, symbols, measured quantities and synthesized values in a complete communications link.

• Perform an independent limiting-case check and identify an operating region where the model fails.

• Design and document a reproducible synthetic experiment that tests a stated prediction.

System context and motivating decision

Consider emergency communications service: a regional network carries mixed urgent control traffic and bulk data. In this setting, latency tails and observed delivery reliability are contractual requirements. The engineer therefore needs a model that is useful at the right reference plane, not an attractive formula copied from a different link. A quantity measured at the transmitter cannot automatically be used as if it were the receive-filter output, and uncoded symbol reliability cannot be silently treated as packet reliability. The purpose of this lesson is to connect arq, hybrid arq and retransmission with observable decisions in this chain.

Physical or algorithmic explanation

Automatic repeat request uses error detection and feedback to resend missing or corrupted frames. Stop-and-wait schemes are simple but can waste a long round-trip time. Sliding windows allow several frames in flight; sequence numbers and buffering manage reordering. Hybrid ARQ combines forward-error correction with retransmission and may soft-combine repeated observations or transmit incremental redundancy. Receiver memory, feedback error and delay budgets constrain practical designs. Stronger FEC reduces retries but may reduce raw throughput, so end-to-end goodput is a coupled design problem.

Core mathematical relationship

For independent packet attempts with success probability p, expected attempts to success are 1/p; HARQ may combine failed transmissions so attempts are not statistically identical.

Meaning and conventions

The central expression above describes arq, hybrid arq and retransmission. Before calculating, identify the set of variables, their units and whether the expression applies to discrete or continuous time. When the relevant quantities are probabilities, confirm the conditioning event and whether samples are independent. When complex amplitudes occur, remember that a conjugate, not a real transpose, normally defines an energy inner product. Where a ratio of powers is stated in decibels, 10 log10 rather than 20 log10 is the appropriate conversion; for voltage magnitudes with identical impedances, a 20 log10 form may be suitable. This model is one part of the signal processing chain, not a substitute for a performance measurement.

Derive, calculate and inspect

First mark the source of the input and output quantities in For independent packet attempts with success probability p, expected attempts to success are 1/p; HARQ may combine failed transmissions so attempts are not statistically identical.. State which symbols are known, which must be estimated and which are chosen by design. Translate the equation into a reproducible step-by-step computation, then test zero or very small noise, an intermediate operating point and a stressed case. A result should remain physically interpretable under these changes. Where the relationship is approximate, determine the approximation error before interpreting small performance improvements as significant. Never add gains or losses twice because the same physical effect appears in two bookkeeping categories.

Fully explained example and verification

If packet success probability is 0.8 on each independent attempt, expected attempts to deliver one packet are 1/0.8=1.25. If each attempt has 1000 payload bits and 100 overhead bits, the idealized payload efficiency is (1000/1100)/1.25≈72.7%, excluding acknowledgment time and contention.

How to verify: Recompute the numeric statement from the specified inputs without reading a plotted curve. Re-run after increasing the most influential input by 10%, and explain the direction and approximate size of the change. Then compare that outcome with the limiting case for the model. A plausible number is not sufficient if the observation plane, normalization or statistical assumptions have changed.

Boundary conditions and common mistakes

A geometric retry model is wrong when fading remains correlated across attempts or incremental redundancy changes conditional success probability after each transmission.

An additional failure mode in emergency communications service arises when the local mechanism is interpreted as an end-to-end guarantee. For example, packets may be rejected by framing or replay checks even after an error-correcting decoder outputs a plausible bit sequence; conversely, a successful frame may be too late to serve a time-critical application. To separate causes, log the inputs and outputs of each major subsystem, retain the raw observation when possible, and name the exact stage at which a failure is declared.

Guided computational experiment

Simulate 10000 packet transfers with a controllable attempt error rate and random seed. Measure attempt distribution, mean completion time and protocol goodput. Include a finite retry limit and unresolved failures.

Reproducibility protocol. Record the independent variables, their increments and physical units; state random seeds, number of simulated trials and target metrics. Use two numerical implementations for at least one predicted point. When drawing a probability curve, use enough trials to estimate its uncertainty and report the confidence interval; small error counts produce unstable estimates. When a parameter sweep crosses a model limit, annotate the plot rather than truncating inconvenient results.

Problem set with targeted response guidance

Question 1. Reference planes and dimensional consistency

Draw the component-level path through a regional network carries mixed urgent control traffic and bulk data. Label exactly where the inputs to “For independent packet attempts with success probability p, expected attempts to success are 1/p; HARQ may combine failed transmissions so attempts are not statistically identical.” are acquired or estimated. Construct a three-column table of quantity, unit, and uncertainty source. Identify one tempting but invalid substitution and explain which physical interface it crosses.

Answer guidance: A complete solution has a block diagram and a clear distinction between physical measurement and algorithmic estimate. For arq, hybrid arq and retransmission, justify every normalization and explicitly say what is held constant when comparing two conditions.

Question 2. Rebuild the numerical case

Independently reconstruct the worked calculation for arq, hybrid arq and retransmission. Change the leading numerical parameter by 20%, holding other inputs fixed, and explain whether the result changes linearly or nonlinearly. Supply every intermediate numerical step.

Answer guidance: The answer must reproduce the worked statement under its stated assumptions, account for units, and include an inverse or limiting check. If a result disagrees, trace the exact operation; do not round intermediate values until the final display.

Question 3. Competing design changes

For emergency communications service, propose two ways to improve the outcome represented by For independent packet attempts with success probability p, expected attempts to success are 1/p; HARQ may combine failed transmissions so attempts are not statistically identical.: one changes a transmitter or source parameter, the other changes receiver processing. List their respective energy, implementation, latency and spectral costs.

Answer guidance: Grade the response on tradeoffs, not a single optimization metric. For this topic, specify which impairment each intervention addresses and whether a gain in one stage increases overhead elsewhere.

Question 4. Break the assumptions

Create a deliberately unrealistic operating point for arq, hybrid arq and retransmission and explain a specific broken assumption. Then propose a synthetic observation that would reveal the failure. Describe what a poorly documented analysis might incorrectly conclude.

Answer guidance: The response must identify an assumption explicitly and design a measurement that detects the discrepancy. Replacing one uncertain parameter with a convenient nominal value is not an assumption test.

Further engineering synthesis

To write a defensible report for arq, hybrid arq and retransmission, separate evidence into three columns: a derivation, a reproducible numerical implementation and a system-level observation. The derivation establishes the modeled relationship; simulation checks implementation and explore cases that are hard to calculate by hand; field observation tests the relevance of modeling assumptions. Failure of a field measurement to match the derivation may reveal either a coding bug or a missing physical effect. Avoid the reflex to adjust free parameters until curves agree, because that conceals rather than explains uncertainty.

Finally trace the effect of arq, hybrid arq and retransmission onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for emergency communications service. Describe which of these metrics has a monotonic response to parameter changes and which displays a tradeoff. Explain the direction of the tradeoff with one numerical or plotted piece of evidence; where evidence is absent, label the claim as a hypothesis. Every numerical conclusion should be reproducible from recorded inputs.

Cross-reference: See the model for Unit 8, Unit 8 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 8, Lesson 3: Queueing, latency and quality of service

Module: Practical link and network engineering. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain queueing, latency and quality of service from a physical or algorithmic model rather than memorizing an isolated equation.

• Distinguish assumptions, symbols, measured quantities and synthesized values in a complete communications link.

• Perform an independent limiting-case check and identify an operating region where the model fails.

• Design and document a reproducible synthetic experiment that tests a stated prediction.

System context and motivating decision

Consider emergency communications service: a regional network carries mixed urgent control traffic and bulk data. In this setting, latency tails and observed delivery reliability are contractual requirements. The engineer therefore needs a model that is useful at the right reference plane, not an attractive formula copied from a different link. A quantity measured at the transmitter cannot automatically be used as if it were the receive-filter output, and uncoded symbol reliability cannot be silently treated as packet reliability. The purpose of this lesson is to connect queueing, latency and quality of service with observable decisions in this chain.

Physical or algorithmic explanation

A communication system must deliver useful data on time, not merely maximize peak physical-layer rate. Packet queues arise when bursts arrive faster than a scheduler can transmit them. Average service rate exceeding average arrival rate is necessary for long-term stability in basic models but says little about tail latency during bursts. Real packet service times are not always exponential, so the M/M/1 formula is an instructive benchmark rather than a blanket guarantee. Prioritization helps urgent traffic but can starve other flows; admission control and shaping influence both reliability and fairness.

Core mathematical relationship

For a stable M/M/1 queue with arrival rate λ and service rate μ>λ, mean system time is W=1/(μ-λ), and mean number in system L=λW.

Meaning and conventions

The central expression above describes queueing, latency and quality of service. Before calculating, identify the set of variables, their units and whether the expression applies to discrete or continuous time. When the relevant quantities are probabilities, confirm the conditioning event and whether samples are independent. When complex amplitudes occur, remember that a conjugate, not a real transpose, normally defines an energy inner product. Where a ratio of powers is stated in decibels, 10 log10 rather than 20 log10 is the appropriate conversion; for voltage magnitudes with identical impedances, a 20 log10 form may be suitable. This model is one part of the signal processing chain, not a substitute for a performance measurement.

Derive, calculate and inspect

First mark the source of the input and output quantities in For a stable M/M/1 queue with arrival rate λ and service rate μ>λ, mean system time is W=1/(μ-λ), and mean number in system L=λW.. State which symbols are known, which must be estimated and which are chosen by design. Translate the equation into a reproducible step-by-step computation, then test zero or very small noise, an intermediate operating point and a stressed case. A result should remain physically interpretable under these changes. Where the relationship is approximate, determine the approximation error before interpreting small performance improvements as significant. Never add gains or losses twice because the same physical effect appears in two bookkeeping categories.

Fully explained example and verification

For λ=80 packets/s and μ=100 packets/s, the M/M/1 benchmark yields W=1/(100-80)=0.05 seconds and average L=4 packets. Increasing λ to 95 packets/s gives W=0.2 seconds even though the server still has nominal unused capacity. Latency increases nonlinearly near saturation.

How to verify: Recompute the numeric statement from the specified inputs without reading a plotted curve. Re-run after increasing the most influential input by 10%, and explain the direction and approximate size of the change. Then compare that outcome with the limiting case for the model. A plausible number is not sufficient if the observation plane, normalization or statistical assumptions have changed.

Boundary conditions and common mistakes

Little's law is broadly valid for steady systems with consistent counting, but the exact M/M/1 mean waiting-time formula depends on its arrival and service assumptions. Report the 95th percentile rather than only mean latency for interactive services.

An additional failure mode in emergency communications service arises when the local mechanism is interpreted as an end-to-end guarantee. For example, packets may be rejected by framing or replay checks even after an error-correcting decoder outputs a plausible bit sequence; conversely, a successful frame may be too late to serve a time-critical application. To separate causes, log the inputs and outputs of each major subsystem, retain the raw observation when possible, and name the exact stage at which a failure is declared.

Guided computational experiment

Generate exponential and bursty arrivals with equal average rate. Compare waiting-time distributions for first-in-first-out and a two-class priority scheduler. Verify Little's law numerically.

Reproducibility protocol. Record the independent variables, their increments and physical units; state random seeds, number of simulated trials and target metrics. Use two numerical implementations for at least one predicted point. When drawing a probability curve, use enough trials to estimate its uncertainty and report the confidence interval; small error counts produce unstable estimates. When a parameter sweep crosses a model limit, annotate the plot rather than truncating inconvenient results.

Problem set with targeted response guidance

Question 1. Reference planes and dimensional consistency

Draw the component-level path through a regional network carries mixed urgent control traffic and bulk data. Label exactly where the inputs to “For a stable M/M/1 queue with arrival rate λ and service rate μ>λ, mean system time is W=1/(μ-λ), and mean number in system L=λW.” are acquired or estimated. Construct a three-column table of quantity, unit, and uncertainty source. Identify one tempting but invalid substitution and explain which physical interface it crosses.

Answer guidance: A complete solution has a block diagram and a clear distinction between physical measurement and algorithmic estimate. For queueing, latency and quality of service, justify every normalization and explicitly say what is held constant when comparing two conditions.

Question 2. Rebuild the numerical case

Independently reconstruct the worked calculation for queueing, latency and quality of service. Change the leading numerical parameter by 20%, holding other inputs fixed, and explain whether the result changes linearly or nonlinearly. Supply every intermediate numerical step.

Answer guidance: The answer must reproduce the worked statement under its stated assumptions, account for units, and include an inverse or limiting check. If a result disagrees, trace the exact operation; do not round intermediate values until the final display.

Question 3. Competing design changes

For emergency communications service, propose two ways to improve the outcome represented by For a stable M/M/1 queue with arrival rate λ and service rate μ>λ, mean system time is W=1/(μ-λ), and mean number in system L=λW.: one changes a transmitter or source parameter, the other changes receiver processing. List their respective energy, implementation, latency and spectral costs.

Answer guidance: Grade the response on tradeoffs, not a single optimization metric. For this topic, specify which impairment each intervention addresses and whether a gain in one stage increases overhead elsewhere.

Question 4. Break the assumptions

Create a deliberately unrealistic operating point for queueing, latency and quality of service and explain a specific broken assumption. Then propose a synthetic observation that would reveal the failure. Describe what a poorly documented analysis might incorrectly conclude.

Answer guidance: The response must identify an assumption explicitly and design a measurement that detects the discrepancy. Replacing one uncertain parameter with a convenient nominal value is not an assumption test.

Further engineering synthesis

To write a defensible report for queueing, latency and quality of service, separate evidence into three columns: a derivation, a reproducible numerical implementation and a system-level observation. The derivation establishes the modeled relationship; simulation checks implementation and explore cases that are hard to calculate by hand; field observation tests the relevance of modeling assumptions. Failure of a field measurement to match the derivation may reveal either a coding bug or a missing physical effect. Avoid the reflex to adjust free parameters until curves agree, because that conceals rather than explains uncertainty.

Finally trace the effect of queueing, latency and quality of service onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for emergency communications service. Describe which of these metrics has a monotonic response to parameter changes and which displays a tradeoff. Explain the direction of the tradeoff with one numerical or plotted piece of evidence; where evidence is absent, label the claim as a hypothesis. Every numerical conclusion should be reproducible from recorded inputs.

Cross-reference: See the model for Unit 8, Unit 8 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 8, Lesson 4: Validation, reproducibility and deployment boundaries

Module: Practical link and network engineering. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain validation, reproducibility and deployment boundaries from a physical or algorithmic model rather than memorizing an isolated equation.

• Distinguish assumptions, symbols, measured quantities and synthesized values in a complete communications link.

• Perform an independent limiting-case check and identify an operating region where the model fails.

• Design and document a reproducible synthetic experiment that tests a stated prediction.

System context and motivating decision

Consider emergency communications service: a regional network carries mixed urgent control traffic and bulk data. In this setting, latency tails and observed delivery reliability are contractual requirements. The engineer therefore needs a model that is useful at the right reference plane, not an attractive formula copied from a different link. A quantity measured at the transmitter cannot automatically be used as if it were the receive-filter output, and uncoded symbol reliability cannot be silently treated as packet reliability. The purpose of this lesson is to connect validation, reproducibility and deployment boundaries with observable decisions in this chain.

Physical or algorithmic explanation

A communication study should separate physical model assumptions from claims about deployed hardware or real traffic. Use a fixed seed to reproduce experiments, but repeat across seeds to measure uncertainty. Log rate, coding parameters, received power reference, noise bandwidth, channel model, packetization and error policy. Establish unit tests for basic identities, numerical conservation checks and negative controls for failure cases. Report confidence intervals when estimating rare error rates. Regulatory compliance and product certification need actual measurements and applicable jurisdiction-specific requirements, not a synthetic chart or a clean source-code build.

Core mathematical relationship

An end-to-end measured goodput G is accepted user payload divided by elapsed observation time; simulated symbol BER, packet error rate, EVM and goodput answer different questions.

Meaning and conventions

The central expression above describes validation, reproducibility and deployment boundaries. Before calculating, identify the set of variables, their units and whether the expression applies to discrete or continuous time. When the relevant quantities are probabilities, confirm the conditioning event and whether samples are independent. When complex amplitudes occur, remember that a conjugate, not a real transpose, normally defines an energy inner product. Where a ratio of powers is stated in decibels, 10 log10 rather than 20 log10 is the appropriate conversion; for voltage magnitudes with identical impedances, a 20 log10 form may be suitable. This model is one part of the signal processing chain, not a substitute for a performance measurement.

Derive, calculate and inspect

First mark the source of the input and output quantities in An end-to-end measured goodput G is accepted user payload divided by elapsed observation time; simulated symbol BER, packet error rate, EVM and goodput answer different questions.. State which symbols are known, which must be estimated and which are chosen by design. Translate the equation into a reproducible step-by-step computation, then test zero or very small noise, an intermediate operating point and a stressed case. A result should remain physically interpretable under these changes. Where the relationship is approximate, determine the approximation error before interpreting small performance improvements as significant. Never add gains or losses twice because the same physical effect appears in two bookkeeping categories.

Fully explained example and verification

If zero errors appear in 1000 independent trials, the observed rate is zero but the one-sided approximate 95% upper bound is about 3/1000=0.003, not zero. A claim that BER is below 10^-6 cannot be supported by such a run. For very low BER, use analytically justified importance sampling or much larger experiments with explicit error bars.

How to verify: Recompute the numeric statement from the specified inputs without reading a plotted curve. Re-run after increasing the most influential input by 10%, and explain the direction and approximate size of the change. Then compare that outcome with the limiting case for the model. A plausible number is not sufficient if the observation plane, normalization or statistical assumptions have changed.

Boundary conditions and common mistakes

A visually attractive plot can hide a normalization bug; in simulation, independent recomputation of selected numerical points is essential. Correlation across trials invalidates common binomial uncertainty calculations.

An additional failure mode in emergency communications service arises when the local mechanism is interpreted as an end-to-end guarantee. For example, packets may be rejected by framing or replay checks even after an error-correcting decoder outputs a plausible bit sequence; conversely, a successful frame may be too late to serve a time-critical application. To separate causes, log the inputs and outputs of each major subsystem, retain the raw observation when possible, and name the exact stage at which a failure is declared.

Guided computational experiment

Write a reproducible experiment manifest with model assumptions and seed. Produce a benchmark showing BER, packet error, payload goodput and latency separately. Add a failure-injection test and write down what remains unvalidated.

Reproducibility protocol. Record the independent variables, their increments and physical units; state random seeds, number of simulated trials and target metrics. Use two numerical implementations for at least one predicted point. When drawing a probability curve, use enough trials to estimate its uncertainty and report the confidence interval; small error counts produce unstable estimates. When a parameter sweep crosses a model limit, annotate the plot rather than truncating inconvenient results.

Problem set with targeted response guidance

Question 1. Reference planes and dimensional consistency

Draw the component-level path through a regional network carries mixed urgent control traffic and bulk data. Label exactly where the inputs to “An end-to-end measured goodput G is accepted user payload divided by elapsed observation time; simulated symbol BER, packet error rate, EVM and goodput answer different questions.” are acquired or estimated. Construct a three-column table of quantity, unit, and uncertainty source. Identify one tempting but invalid substitution and explain which physical interface it crosses.

Answer guidance: A complete solution has a block diagram and a clear distinction between physical measurement and algorithmic estimate. For validation, reproducibility and deployment boundaries, justify every normalization and explicitly say what is held constant when comparing two conditions.

Question 2. Rebuild the numerical case

Independently reconstruct the worked calculation for validation, reproducibility and deployment boundaries. Change the leading numerical parameter by 20%, holding other inputs fixed, and explain whether the result changes linearly or nonlinearly. Supply every intermediate numerical step.

Answer guidance: The answer must reproduce the worked statement under its stated assumptions, account for units, and include an inverse or limiting check. If a result disagrees, trace the exact operation; do not round intermediate values until the final display.

Question 3. Competing design changes

For emergency communications service, propose two ways to improve the outcome represented by An end-to-end measured goodput G is accepted user payload divided by elapsed observation time; simulated symbol BER, packet error rate, EVM and goodput answer different questions.: one changes a transmitter or source parameter, the other changes receiver processing. List their respective energy, implementation, latency and spectral costs.

Answer guidance: Grade the response on tradeoffs, not a single optimization metric. For this topic, specify which impairment each intervention addresses and whether a gain in one stage increases overhead elsewhere.

Question 4. Break the assumptions

Create a deliberately unrealistic operating point for validation, reproducibility and deployment boundaries and explain a specific broken assumption. Then propose a synthetic observation that would reveal the failure. Describe what a poorly documented analysis might incorrectly conclude.

Answer guidance: The response must identify an assumption explicitly and design a measurement that detects the discrepancy. Replacing one uncertain parameter with a convenient nominal value is not an assumption test.

Further engineering synthesis

To write a defensible report for validation, reproducibility and deployment boundaries, separate evidence into three columns: a derivation, a reproducible numerical implementation and a system-level observation. The derivation establishes the modeled relationship; simulation checks implementation and explore cases that are hard to calculate by hand; field observation tests the relevance of modeling assumptions. Failure of a field measurement to match the derivation may reveal either a coding bug or a missing physical effect. Avoid the reflex to adjust free parameters until curves agree, because that conceals rather than explains uncertainty.

Finally trace the effect of validation, reproducibility and deployment boundaries onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for emergency communications service. Describe which of these metrics has a monotonic response to parameter changes and which displays a tradeoff. Explain the direction of the tradeoff with one numerical or plotted piece of evidence; where evidence is absent, label the claim as a hypothesis. Every numerical conclusion should be reproducible from recorded inputs.

Cross-reference: See the model for Unit 8, Unit 8 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 8 project — Practical link and network engineering

Design brief

Create an original computer-based engineering investigation of emergency communications service: a regional network carries mixed urgent control traffic and bulk data. The stakeholder fears that latency tails and observed delivery reliability are contractual requirements. Deliver a short technical report that combines the four models in Unit 8, an independent numeric calculation, a simple synthetic implementation and a comparison between two candidate solutions.

Required artifact A — model assumptions and operating points

Write a system block diagram and a table defining the reference points for all quantities in the four lessons. Include a benign operating point and two stresses, one concerning arq, hybrid arq and retransmission and the other concerning validation, reproducibility and deployment boundaries. Give symbolic units and numerical values. Make clear which values are synthetic; do not label simulated observations as measured data.

Required artifact B — two methods, one controlled comparison

Implement an equation or algorithm from Adaptive modulation and coding, then independently recompute a selected point. Next vary an input related to Queueing, latency and quality of service while holding the baseline model fixed. Record the dependent metric at at least ten settings; if the output is an estimated failure probability, run multiple seeds and show sampling uncertainty. Report a parameter setting at which the second candidate outperforms the first and one at which it does not.

Required artifact C — intentionally broken assumption

The limitations you must investigate include: (1) Goodput figures are dependent on traffic pattern, failure correlation, headers and retransmission details. A controller optimized using mean SNR can fail severely during rare fades. (2) Little's law is broadly valid for steady systems with consistent counting, but the exact M/M/1 mean waiting-time formula depends on its arrival and service assumptions. Report the 95th percentile rather than only mean latency for interactive services. Choose the more relevant limitation for emergency communications service, inject an impairment that invalidates the original assumption, and explain in technical language why the ideal calculation no longer predicts the result. An analysis that merely says “noise got worse” will not pass this requirement.

Required artifact D — user-level consequence

The application is a regional network carries mixed urgent control traffic and bulk data. Define payload size, raw resource rate, protocol overhead, retransmission rule and the deadline for an accepted message. Trace low-level failure through framing and decoding into delivery success, delay and useful goodput. Prefer a table containing both one-attempt performance and end-to-end outcomes. Explain what real deployment measurements would be required.

Reproducibility and reporting rubric

Submit the source code, version or runtime description, seed list, parameter ranges and generated CSV. Use descriptive filenames, include a short README with instructions for a fresh run, and test one identity or limiting case. Cite the assumptions that make each formula valid. Allocate 30% to physics and mathematical correctness, 25% to experiment design, 20% to interpretation and failure analysis, 15% to reproducibility, and 10% to clarity. An unsupported claim of production readiness receives no credit.

Oral defense prompts

Explain why ARQ, hybrid ARQ and retransmission was included or excluded in the final design. Which numerical result is most sensitive to a poorly known parameter? What single new piece of hardware or field evidence would most change the conclusion? Finally, explain why the synthetic model is adequate for learning but does not establish a radio license, an optical product certification or safety compliance.

Unit 8 assessment — Practical link and network engineering

Context: a regional network carries mixed urgent control traffic and bulk data

Problems (at least three required per unit)

Question 1: Adaptive modulation and coding

Problem: Starting with A link-adaptation controller selects a modulation/coding scheme using estimated channel quality, predicted packet error and implementation-specific throughput tables., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: Goodput figures are dependent on traffic pattern, failure correlation, headers and retransmission details. A controller optimized using mean SNR can fail severely during rare fades.

Mark scheme (10 points): 3 for model and assumptions, 3 for correct arithmetic or logical steps, 2 for limiting-case check, 2 for explaining why the stated limitation changes the conclusion. A correct final number without units and reference plane earns no more than 5 points.

Question 2: ARQ, hybrid ARQ and retransmission

Problem: Starting with For independent packet attempts with success probability p, expected attempts to success are 1/p; HARQ may combine failed transmissions so attempts are not statistically identical., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: A geometric retry model is wrong when fading remains correlated across attempts or incremental redundancy changes conditional success probability after each transmission.

Mark scheme (10 points): 3 for model and assumptions, 3 for correct arithmetic or logical steps, 2 for limiting-case check, 2 for explaining why the stated limitation changes the conclusion. A correct final number without units and reference plane earns no more than 5 points.

Question 3: Queueing, latency and quality of service

Problem: Starting with For a stable M/M/1 queue with arrival rate λ and service rate μ>λ, mean system time is W=1/(μ-λ), and mean number in system L=λW., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: Little's law is broadly valid for steady systems with consistent counting, but the exact M/M/1 mean waiting-time formula depends on its arrival and service assumptions. Report the 95th percentile rather than only mean latency for interactive services.

Mark scheme (10 points): 3 for model and assumptions, 3 for correct arithmetic or logical steps, 2 for limiting-case check, 2 for explaining why the stated limitation changes the conclusion. A correct final number without units and reference plane earns no more than 5 points.

Question 4: Validation, reproducibility and deployment boundaries

Problem: Starting with An end-to-end measured goodput G is accepted user payload divided by elapsed observation time; simulated symbol BER, packet error rate, EVM and goodput answer different questions., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: A visually attractive plot can hide a normalization bug; in simulation, independent recomputation of selected numerical points is essential. Correlation across trials invalidates common binomial uncertainty calculations.

Mark scheme (10 points): 3 for model and assumptions, 3 for correct arithmetic or logical steps, 2 for limiting-case check, 2 for explaining why the stated limitation changes the conclusion. A correct final number without units and reference plane earns no more than 5 points.

Question 5: Joint design

Combine two lessons in the emergency communications service case. Propose a before/after experiment where only one controlled variable changes and show how you will report both a link-level metric and a delivered-payload metric.

Mark scheme: prioritize controlled comparison (4), technically justified outcome (4), reproducible evidence (2).

Question 6: Critical evidence review

Find a plausible but unsupported claim of improvement, identify which additional variable may account for the effect, and design one negative control to disprove your own conclusion.

Mark scheme: claims supported by computed or experimentally generated data receive credit; appeals to authority or unsupported statements do not.

Submission policy

Use explicit parameter values, code or workings, and a conclusion that distinguishes verified arithmetic from engineering assumptions.