Binary Passband Signaling: BASK, BFSK, BPSK, and Coherent Matched Filters

Unit 2 — Modern coding and probabilistic decoding

Why this unit exists

The unit centers on coded low-power spacecraft downlink, where a remote probe sends diagnostic records through a noisy long-distance link. Its controlling difficulty is that bit energy, decode latency and packet loss each constrain the operating mode. 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 2.1 — Log-likelihood ratios and soft information

A hard bit has only a sign, while a log-likelihood ratio preserves confidence. Passing confidence from the demodulator to the decoder allows weak bits to be corrected using parity relationships with other bits. Prior probability enters as an additive log-prior odds term. LLR sign conventions vary across implementations; a decoder built for the opposite convention produces systematic errors. Clipping LLR magnitude may be necessary in fixed-point hardware, but overly aggressive clipping destroys reliability information.

Key relation: For equally likely BPSK symbols x in {+1,-1} over AWGN with noise variance sigma^2, L=log[p(y|+1)/p(y|-1)]=2y/sigma^2.

Design boundary: The formula assumes memoryless Gaussian noise and correctly scaled observations. Channel mismatch, clipping and non-Gaussian interference can make high-magnitude LLRs misleading.

Lesson 2.2 — Sparse parity checks and LDPC iteration

A sparse graph exposes parity structure without explicitly listing all codewords. In sum-product decoding, each variable node combines intrinsic channel LLR with extrinsic messages from neighboring checks; each check sends a constraint-informed update to one variable, excluding that variable's previous contribution. A practical min-sum approximation replaces expensive tanh operations with a signed minimum, often with normalization or offset corrections. Cycles make graph messages dependent, so iterations are approximations rather than independent Bayesian inference.

Key relation: For parity-check matrix H, an LDPC code uses sparse constraints Hc=0; belief propagation exchanges variable-to-check and check-to-variable messages on a Tanner graph.

Design boundary: Short cycles and absorbing sets cause error floors. More iterations can consume time and power without reducing error under poor channel estimates. Never compare LDPC and uncoded BER while silently changing information-bit energy.

Lesson 2.3 — Turbo codes and extrinsic-information exchange

Concatenated convolutional encoders with a permutation can transform relatively short-memory codes into powerful long-block constructions. The interleaver disperses localized input patterns so that a damaging low-weight pattern for one component is less damaging for the other. During decoding, a component separates systematic-channel evidence, input prior and newly learned extrinsic evidence. Sending the entire posterior as a new prior double-counts information and encourages false certainty.

Key relation: Turbo decoding iterates between soft-in/soft-out constituent decoders, exchanging extrinsic LLRs through an interleaver.

Design boundary: Iterative information transfer assumes meaningful extrinsic messages. Short interleavers, correlation between messages, poor termination and quantized arithmetic can harm convergence.

Lesson 2.4 — Polar transforms, frozen positions and CRC-aided decisions

Polar coding relies on the unequal reliability of synthesized bit channels at large block lengths. The transmitter places known values in unreliable positions and information in reliable positions, then applies a structured binary transform. Successive-cancellation decoding conditions successive decisions on earlier decisions, so an early wrong decision may propagate. List decoding keeps several candidates. A CRC appended to the message can help select among list candidates, but a CRC is not magic: incorrectly sized lists or unfavorable channels can still produce block errors.

Key relation: The kernel F=[[1,0],[1,1]] over GF(2) forms Kronecker-power polarization transforms; selected synthetic channels carry information while others are frozen.

Design boundary: A code optimized for one channel reliability ordering may perform poorly on another. Finite-length performance cannot be inferred solely from asymptotic capacity arguments.

Integrated case study

An engineer reviews the architecture for coded low-power spacecraft downlink. 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 coded low-power spacecraft downlink, 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 remote probe sends diagnostic records through a noisy long-distance link. 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 coded low-power spacecraft downlink, and what measured evidence would establish that claim?

• How would the design fail if the assumption “The formula assumes memoryless Gaussian noise and correctly scaled observations. Channel mismatch, clipping and non-Gaussian interference can make high-magnitude LLRs misleading.” 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 2, Lesson 1: Log-likelihood ratios and soft information

Module: Modern coding and probabilistic decoding. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain log-likelihood ratios and soft information 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 coded low-power spacecraft downlink: a remote probe sends diagnostic records through a noisy long-distance link. In this setting, bit energy, decode latency and packet loss each constrain the operating mode. 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 log-likelihood ratios and soft information with observable decisions in this chain.

Physical or algorithmic explanation

A hard bit has only a sign, while a log-likelihood ratio preserves confidence. Passing confidence from the demodulator to the decoder allows weak bits to be corrected using parity relationships with other bits. Prior probability enters as an additive log-prior odds term. LLR sign conventions vary across implementations; a decoder built for the opposite convention produces systematic errors. Clipping LLR magnitude may be necessary in fixed-point hardware, but overly aggressive clipping destroys reliability information.

Core mathematical relationship

For equally likely BPSK symbols x in {+1,-1} over AWGN with noise variance sigma^2, L=log[p(y|+1)/p(y|-1)]=2y/sigma^2.

Meaning and conventions

The central expression above describes log-likelihood ratios and soft information. 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 equally likely BPSK symbols x in {+1,-1} over AWGN with noise variance sigma^2, L=log[p(y|+1)/p(y|-1)]=2y/sigma^2.. 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 y=0.3 and noise variance sigma^2=0.2, the observation LLR in the +1-over--1 convention is 3.0. With prior odds P(+1)/P(-1)=1/3, the posterior log-odds is 3+ln(1/3)=1.9014, still favoring +1. The prior changes confidence but not the observed sample itself.

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

The formula assumes memoryless Gaussian noise and correctly scaled observations. Channel mismatch, clipping and non-Gaussian interference can make high-magnitude LLRs misleading.

An additional failure mode in coded low-power spacecraft downlink 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

Plot logistic probability 1/(1+exp(-L)) against L and compare it with a one-bit hard detector. Inject a wrong variance estimate to observe overconfidence and calibrate using bins of predicted reliability.

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 remote probe sends diagnostic records through a noisy long-distance link. Label exactly where the inputs to “For equally likely BPSK symbols x in {+1,-1} over AWGN with noise variance sigma^2, L=log[p(y|+1)/p(y|-1)]=2y/sigma^2.” 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 log-likelihood ratios and soft information, 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 log-likelihood ratios and soft information. 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 coded low-power spacecraft downlink, propose two ways to improve the outcome represented by For equally likely BPSK symbols x in {+1,-1} over AWGN with noise variance sigma^2, L=log[p(y|+1)/p(y|-1)]=2y/sigma^2.: 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 log-likelihood ratios and soft information 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 log-likelihood ratios and soft information, 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 log-likelihood ratios and soft information onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for coded low-power spacecraft downlink. 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 2, Unit 2 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 2, Lesson 2: Sparse parity checks and LDPC iteration

Module: Modern coding and probabilistic decoding. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain sparse parity checks and ldpc iteration 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 coded low-power spacecraft downlink: a remote probe sends diagnostic records through a noisy long-distance link. In this setting, bit energy, decode latency and packet loss each constrain the operating mode. 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 sparse parity checks and ldpc iteration with observable decisions in this chain.

Physical or algorithmic explanation

A sparse graph exposes parity structure without explicitly listing all codewords. In sum-product decoding, each variable node combines intrinsic channel LLR with extrinsic messages from neighboring checks; each check sends a constraint-informed update to one variable, excluding that variable's previous contribution. A practical min-sum approximation replaces expensive tanh operations with a signed minimum, often with normalization or offset corrections. Cycles make graph messages dependent, so iterations are approximations rather than independent Bayesian inference.

Core mathematical relationship

For parity-check matrix H, an LDPC code uses sparse constraints Hc=0; belief propagation exchanges variable-to-check and check-to-variable messages on a Tanner graph.

Meaning and conventions

The central expression above describes sparse parity checks and ldpc iteration. 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 parity-check matrix H, an LDPC code uses sparse constraints Hc=0; belief propagation exchanges variable-to-check and check-to-variable messages on a Tanner graph.. 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 a single parity check c1 XOR c2 XOR c3=0, if a hard preliminary decision is 1,0,0 then its syndrome is 1, so at least one of these tentative bits violates the constraint. The syndrome alone does not identify which bit; confidence values distinguish likely corrections. An all-zero syndrome is a useful stopping condition but does not prove transmission was error-free.

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

Short cycles and absorbing sets cause error floors. More iterations can consume time and power without reducing error under poor channel estimates. Never compare LDPC and uncoded BER while silently changing information-bit energy.

An additional failure mode in coded low-power spacecraft downlink 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

Make a small sparse H matrix, draw its Tanner graph and carry out two LLR message-passing rounds manually. Compare flooding and layered schedules on the same synthetic observations.

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 remote probe sends diagnostic records through a noisy long-distance link. Label exactly where the inputs to “For parity-check matrix H, an LDPC code uses sparse constraints Hc=0; belief propagation exchanges variable-to-check and check-to-variable messages on a Tanner graph.” 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 sparse parity checks and ldpc iteration, 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 sparse parity checks and ldpc iteration. 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 coded low-power spacecraft downlink, propose two ways to improve the outcome represented by For parity-check matrix H, an LDPC code uses sparse constraints Hc=0; belief propagation exchanges variable-to-check and check-to-variable messages on a Tanner graph.: 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 sparse parity checks and ldpc iteration 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 sparse parity checks and ldpc iteration, 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 sparse parity checks and ldpc iteration onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for coded low-power spacecraft downlink. 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 2, Unit 2 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 2, Lesson 3: Turbo codes and extrinsic-information exchange

Module: Modern coding and probabilistic decoding. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain turbo codes and extrinsic-information exchange 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 coded low-power spacecraft downlink: a remote probe sends diagnostic records through a noisy long-distance link. In this setting, bit energy, decode latency and packet loss each constrain the operating mode. 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 turbo codes and extrinsic-information exchange with observable decisions in this chain.

Physical or algorithmic explanation

Concatenated convolutional encoders with a permutation can transform relatively short-memory codes into powerful long-block constructions. The interleaver disperses localized input patterns so that a damaging low-weight pattern for one component is less damaging for the other. During decoding, a component separates systematic-channel evidence, input prior and newly learned extrinsic evidence. Sending the entire posterior as a new prior double-counts information and encourages false certainty.

Core mathematical relationship

Turbo decoding iterates between soft-in/soft-out constituent decoders, exchanging extrinsic LLRs through an interleaver.

Meaning and conventions

The central expression above describes turbo codes and extrinsic-information exchange. 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 Turbo decoding iterates between soft-in/soft-out constituent decoders, exchanging extrinsic LLRs through an interleaver.. 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 a demodulator gives L_channel=0.8, while a component decoder contributes L_extrinsic=1.1 and the prior is -0.2. Its posterior is 1.7 in the common additive approximation. The quantity sent to the other constituent decoder should be 1.1, not 1.7. That difference matters over repeated iterations.

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

Iterative information transfer assumes meaningful extrinsic messages. Short interleavers, correlation between messages, poor termination and quantized arithmetic can harm convergence.

An additional failure mode in coded low-power spacecraft downlink 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

Experiment with a small deterministic permutation and track a set of LLRs through deinterleaving. Plot the difference between exchanging extrinsic and complete posterior information as iterations increase.

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 remote probe sends diagnostic records through a noisy long-distance link. Label exactly where the inputs to “Turbo decoding iterates between soft-in/soft-out constituent decoders, exchanging extrinsic LLRs through an interleaver.” 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 turbo codes and extrinsic-information exchange, 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 turbo codes and extrinsic-information exchange. 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 coded low-power spacecraft downlink, propose two ways to improve the outcome represented by Turbo decoding iterates between soft-in/soft-out constituent decoders, exchanging extrinsic LLRs through an interleaver.: 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 turbo codes and extrinsic-information exchange 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 turbo codes and extrinsic-information exchange, 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 turbo codes and extrinsic-information exchange onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for coded low-power spacecraft downlink. 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 2, Unit 2 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 2, Lesson 4: Polar transforms, frozen positions and CRC-aided decisions

Module: Modern coding and probabilistic decoding. Prerequisites: Communication Systems I, introductory probability, linear systems, complex algebra and elementary Python or JavaScript.

Learning outcomes

• Explain polar transforms, frozen positions and crc-aided decisions 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 coded low-power spacecraft downlink: a remote probe sends diagnostic records through a noisy long-distance link. In this setting, bit energy, decode latency and packet loss each constrain the operating mode. 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 polar transforms, frozen positions and crc-aided decisions with observable decisions in this chain.

Physical or algorithmic explanation

Polar coding relies on the unequal reliability of synthesized bit channels at large block lengths. The transmitter places known values in unreliable positions and information in reliable positions, then applies a structured binary transform. Successive-cancellation decoding conditions successive decisions on earlier decisions, so an early wrong decision may propagate. List decoding keeps several candidates. A CRC appended to the message can help select among list candidates, but a CRC is not magic: incorrectly sized lists or unfavorable channels can still produce block errors.

Core mathematical relationship

The kernel F=[[1,0],[1,1]] over GF(2) forms Kronecker-power polarization transforms; selected synthetic channels carry information while others are frozen.

Meaning and conventions

The central expression above describes polar transforms, frozen positions and crc-aided decisions. 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 The kernel F=[[1,0],[1,1]] over GF(2) forms Kronecker-power polarization transforms; selected synthetic channels carry information while others are frozen.. 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 N=2, the transform sends u=(u0,u1) to x=(u0 XOR u1,u1). If u0=0 is frozen and u1=1 carries data, x=(1,1). For longer N, the ordering of bit reversal and the kernel product must be stated, because different conventions produce apparently different encoders.

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 code optimized for one channel reliability ordering may perform poorly on another. Finite-length performance cannot be inferred solely from asymptotic capacity arguments.

An additional failure mode in coded low-power spacecraft downlink 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

Encode all possible pairs for the N=2 kernel and verify invertibility. Extend to N=8, freeze selected input indices and enumerate the minimum Hamming distance for each frozen set.

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 remote probe sends diagnostic records through a noisy long-distance link. Label exactly where the inputs to “The kernel F=[[1,0],[1,1]] over GF(2) forms Kronecker-power polarization transforms; selected synthetic channels carry information while others are frozen.” 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 polar transforms, frozen positions and crc-aided decisions, 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 polar transforms, frozen positions and crc-aided decisions. 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 coded low-power spacecraft downlink, propose two ways to improve the outcome represented by The kernel F=[[1,0],[1,1]] over GF(2) forms Kronecker-power polarization transforms; selected synthetic channels carry information while others are frozen.: 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 polar transforms, frozen positions and crc-aided decisions 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 polar transforms, frozen positions and crc-aided decisions, 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 polar transforms, frozen positions and crc-aided decisions onto throughput, occupied bandwidth, energy per delivered information bit and deadline failures for coded low-power spacecraft downlink. 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 2, Unit 2 project brief and its assessment. The simulation depicts a deliberately simplified subproblem; it does not certify a live communication system.

Unit 2 project — Modern coding and probabilistic decoding

Design brief

Create an original computer-based engineering investigation of coded low-power spacecraft downlink: a remote probe sends diagnostic records through a noisy long-distance link. The stakeholder fears that bit energy, decode latency and packet loss each constrain the operating mode. Deliver a short technical report that combines the four models in Unit 2, 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 sparse parity checks and ldpc iteration and the other concerning polar transforms, frozen positions and crc-aided decisions. 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 Log-likelihood ratios and soft information, then independently recompute a selected point. Next vary an input related to Turbo codes and extrinsic-information exchange 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) The formula assumes memoryless Gaussian noise and correctly scaled observations. Channel mismatch, clipping and non-Gaussian interference can make high-magnitude LLRs misleading. (2) Iterative information transfer assumes meaningful extrinsic messages. Short interleavers, correlation between messages, poor termination and quantized arithmetic can harm convergence. Choose the more relevant limitation for coded low-power spacecraft downlink, 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 remote probe sends diagnostic records through a noisy long-distance link. 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 Sparse parity checks and LDPC iteration 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 2 assessment — Modern coding and probabilistic decoding

Context: a remote probe sends diagnostic records through a noisy long-distance link

Problems (at least three required per unit)

Question 1: Log-likelihood ratios and soft information

Problem: Starting with For equally likely BPSK symbols x in {+1,-1} over AWGN with noise variance sigma^2, L=log[p(y|+1)/p(y|-1)]=2y/sigma^2., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: The formula assumes memoryless Gaussian noise and correctly scaled observations. Channel mismatch, clipping and non-Gaussian interference can make high-magnitude LLRs misleading.

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: Sparse parity checks and LDPC iteration

Problem: Starting with For parity-check matrix H, an LDPC code uses sparse constraints Hc=0; belief propagation exchanges variable-to-check and check-to-variable messages on a Tanner graph., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: Short cycles and absorbing sets cause error floors. More iterations can consume time and power without reducing error under poor channel estimates. Never compare LDPC and uncoded BER while silently changing information-bit energy.

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: Turbo codes and extrinsic-information exchange

Problem: Starting with Turbo decoding iterates between soft-in/soft-out constituent decoders, exchanging extrinsic LLRs through an interleaver., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: Iterative information transfer assumes meaningful extrinsic messages. Short interleavers, correlation between messages, poor termination and quantized arithmetic can harm convergence.

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: Polar transforms, frozen positions and CRC-aided decisions

Problem: Starting with The kernel F=[[1,0],[1,1]] over GF(2) forms Kronecker-power polarization transforms; selected synthetic channels carry information while others are frozen., produce a dimensionally consistent worked example and one limiting case, then identify the violated assumption in this scenario: A code optimized for one channel reliability ordering may perform poorly on another. Finite-length performance cannot be inferred solely from asymptotic capacity arguments.

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 coded low-power spacecraft downlink 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.