Low-density parity-check (LDPC) codes have become a cornerstone of modern digital communication systems, enabling reliable data transmission near the Shannon limit. First introduced by Robert Gallager in his 1963 doctoral dissertation and later rediscovered in the 1990s, these codes are now integral to standards such as DVB-S2, Wi-Fi (IEEE 802.11n/ac/ax), 5G NR, and deep-space communications. The ability to adjust the code rate—that is, the ratio of information bits to total transmitted bits—is essential for adapting to varying channel conditions and quality-of-service requirements. Two primary techniques for code rate adjustment without redesigning the entire code structure are puncturing and shortening. Both methods alter the effective code rate while leveraging the same underlying LDPC matrix, offering flexibility for rate-compatible systems. This article examines the principles, implementation details, and practical trade-offs of puncturing and shortening in the context of LDPC codes.

Fundamentals of LDPC Codes

An LDPC code is a linear block code defined by a sparse parity-check matrix H of size m × n. The sparsity of H means that it contains very few ones, which enables efficient iterative decoding algorithms, typically belief propagation (also known as sum-product algorithm) performed on a Tanner graph representation. In the Tanner graph, variable nodes correspond to the n bits of a codeword, and check nodes correspond to the m parity-check equations. Edges connect variable nodes to check nodes according to the ones in H. Valid codewords satisfy all parity-check equations: H · cT = 0.

The code rate R is given by (n – m) / n for a full-rank matrix, representing the fraction of a codeword that carries original information. A higher code rate means less redundancy (more throughput), while a lower code rate provides stronger error correction (more reliability). However, a fixed LDPC matrix yields a fixed rate; to vary the rate efficiently without storing multiple matrices, puncturing and shortening are employed.

Why Adjust Code Rates?

Communication channels vary over time due to fading, interference, and noise. A single code rate cannot optimally serve all conditions: under good conditions, a high rate maximizes data throughput; under poor conditions, a low rate ensures reliable reception. Adaptive coding and modulation (ACM) systems therefore need rate-compatible codes. Puncturing and shortening allow the same encoder and decoder to support multiple rates, reducing hardware complexity and memory requirements.

The Puncturing Technique

Puncturing is a method to increase the effective code rate by selectively removing (not transmitting) some bits of a codeword after encoding. Typically, a high-rate mother code is first generated using an LDPC matrix; then a subset of the parity bits (or occasionally systematic bits) is discarded before transmission. The receiver, aware of the puncturing pattern, treats the missing bits as erasures and attempts to recover the original codeword using the remaining bits and the full parity-check matrix.

Mathematically, if the mother code has length n and k information bits (rate k/n), puncturing p bits results in a transmitted codeword of length n – p, giving an effective rate of k / (n – p), which is higher than the mother code rate. The decoder, however, still operates on the full n-bit variable node set—the punctured variable nodes are initialized with zero log-likelihood ratios (LLRs), indicating maximum uncertainty (i.e., an erasure). During iterative decoding, messages from check nodes may provide information about the erased nodes, often allowing successful recovery if the number of punctured bits is not too large relative to the code’s error-correcting capability.

Puncturing Patterns and Design Considerations

The performance of a punctured LDPC code depends heavily on which bits are punctured. Random puncturing can degrade performance because it may break the structure of the code, especially for short block lengths. Systematic design approaches exist, such as:

  • Regular puncturing: Remove bits according to a fixed periodic pattern. This is simple but may not be optimal.
  • Column-weight-based puncturing: Prioritize puncturing of variable nodes with lower column weight (fewer connections to check nodes) because they contribute less to decoding. Puncturing higher-weight nodes tends to cause more severe performance loss.
  • Recovery-aware puncturing: Design the pattern so that punctured nodes can be recovered quickly—typically by ensuring that each punctured node is connected to at least one check node whose other neighbors are all transmitted (i.e., a “recovery check node”).

One important concept is the recovery tree: after a certain number of decoding iterations, the LLR of a punctured variable node may become non-zero due to extrinsic information from check nodes. The deeper the recovery tree, the more iterations needed. For rate-compatible families, "puncturing levels" are often precomputed to achieve a range of target rates with minimal performance degradation. For example, the DVB-S2 standard defines a set of puncturing patterns for its LDPC codes to produce multiple code rates from a single mother code.

Impact on Decoding Complexity

Puncturing does not increase decoding complexity per iteration because the same number of variable and check nodes are present in the Tanner graph. However, the number of iterations required for convergence may increase due to the initial erasures. Additionally, the decoder must know the puncturing pattern to appropriately initialize LLRs. Overall, puncturing is a low-complexity method to achieve higher rates, making it popular in adaptive systems.

The Shortening Technique

Shortening is used to decrease the effective code rate, making the code more robust. In shortening, some of the information bits are set to a known value (typically zero) before encoding, and those bits are not transmitted. Only the remaining information bits and all parity bits are sent. The receiver, knowing which positions were shortened, treats the known bits as additional side information during decoding.

Consider a mother code with k information bits and n total bits (rate k/n). If we shorten s bits, the actual transmitted codeword consists of k – s information bits and n – k parity bits, for a total of n – s transmitted bits. The effective rate becomes (k – s) / (n – s), which is lower than the mother code rate. The decoder uses the full parity-check matrix but initializes the LLRs of the shortened variable nodes to a very high (or very low) value representing the known bit—effectively providing perfect channel information for those positions. This strengthens the parity-check constraints and improves error correction.

Implementation and Performance

Shortening is straightforward in practice: the encoder simply sets the chosen information-bit positions to zero (or any known value) and then encodes as usual. At the decoder, the LLRs for those variable nodes are set to +∞ (for a known zero in binary phase-shift keying) or the equivalent large magnitude. Because these bits are known with certainty, they serve as anchor points that help the iterative decoder resolve uncertainties in other bits.

A key advantage of shortening is that it does not require any modification of the parity-check matrix or the decoding algorithm. The code structure remains intact, but the code rate is effectively lowered. Unlike puncturing, which introduces erasures that must be resolved, shortening adds perfect information. As a result, performance improvements are generally smoother than those achieved by lowering the rate via other methods (such as using a lower-rate mother code directly).

Trade-Offs and Interactions with Puncturing

Shortening and puncturing are often combined to achieve fine-grained rate adaptivity. For instance, a system may use a mother code with rate 1/2, then apply puncturing to raise the rate toward 2/3 or 3/4, and simultaneously apply shortening to lower the rate toward 1/3 or 1/4. This dual approach allows a single decoder to cover a wide range of rates. However, the combination must be designed carefully: if many bits are both shortened and punctured (which would be redundant), performance can suffer. Typically, systems either shorten information bits to lower the rate or puncture parity bits to raise the rate, but not both on the same bits.

One common scenario where both techniques are used is in incremental redundancy (IR) hybrid automatic repeat request (HARQ). Initially, a high-rate punctured codeword is sent. If decoding fails, additional parity bits (which were previously punctured) are transmitted in subsequent retransmissions—this is essentially an IR scheme based on puncturing. Shortening can be employed to reduce the effective rate if the original mother code is too high for the channel.

Comparison of Puncturing and Shortening

AspectPuncturingShortening
Effect on code rateIncreases rate (less redundancy)Decreases rate (more redundancy)
Bits removed/not transmittedParity (or systematic) bits are omittedInformation bits are set to known value and omitted
State at decoderPunctured bits are erasures (LLR=0)Shortened bits have perfect LLR (large magnitude)
Performance trade-offHigher rate → less error correction; may need more iterationsLower rate → stronger error correction; increased overhead
ComplexitySame decoder graph; increased iterations possibleSame decoder graph; no extra iterations needed
Design challengePuncturing pattern must be optimized to avoid error floorMinimal design effort; works well for small shortening amounts
Common useAdaptive rate increase, ARQ, rate-compatible familiesFine rate reduction, code shortening for specific block sizes

Applications in Modern Standards

Both puncturing and shortening are deployed in major communication standards. In DVB-S2 and its extension DVB-S2X, LDPC codes of rate 1/4 to 9/10 are supported. These are derived from a common set of mother codes using puncturing and shortening. The standard defines specific patterns to achieve each target rate, ensuring near-optimal performance. Similarly, 5G NR uses LDPC codes for the data channel, with base graphs (BG1 and BG2) designed to support rate-compatibility. Puncturing is used to achieve the highest rates, while shortening (combined with repetition) helps reach the lowest rates. The 5G standard also employs a “block CRC” and HARQ that relies on incremental redundancy using punctured parity bits.

In Wi-Fi 802.11n/ac/ax, LDPC codes are optional but widely implemented. The standard defines a set of code rates (1/2, 2/3, 3/4, 5/6) for each block length. These rates are achieved by starting from a rate-1/2 mother code and puncturing to obtain higher rates, or by shortening to obtain lower rates (though shortening is less common in this context). The IEEE 802.11 working group published performance evaluations that validate the use of punctured LDPC codes for high-throughput scenarios.

Deep-space missions, such as those by NASA using the CCSDS standard, also rely on LDPC codes with puncturing and shortening for telemetry and command links. The ability to adjust code rate without changing the hardware decoder is invaluable for long-duration missions where channel conditions are unpredictable.

Performance Metrics and Optimization

When designing a rate-compatible system using puncturing and shortening, engineers evaluate bit error rate (BER) and frame error rate (FER) as functions of signal-to-noise ratio (SNR). The goal is to approach the Shannon limit for each effective rate. Puncturing typically introduces a small penalty compared to a dedicated mother code of the same rate, but the penalty can be minimized through careful pattern design. Shortening, because it adds known bits, often performs very close to a natural low-rate code of the same dimension.

Another important metric is the error floor—the region at high SNR where the error rate flattens due to trapping sets or small cycles in the Tanner graph. Puncturing can elevate the error floor if poorly designed, especially for LDPC codes with moderate block lengths. Shortening, on the other hand, rarely introduces new trapping sets because it only reduces the number of information bits without altering the graph topology.

Researchers have proposed many algorithms for jointly optimizing puncturing patterns, including density evolution, extrinsic information transfer (EXIT) charts, and genetic algorithms. For instance, the work by Ha, Kim, and McLaughlin ("Rate-Compatible Punctured Low-Density Parity-Check Codes With Short Block Lengths", IEEE Transactions on Information Theory, 2004) provides a foundational framework for designing good puncturing patterns. Similarly, for shortening, the effect on the decoder’s extrinsic messages can be analyzed using density evolution to ensure no degradation.

Conclusion

Puncturing and shortening are essential tools for adjusting the code rate of LDPC codes without requiring multiple distinct encoder and decoder implementations. Puncturing increases the effective rate by omitting bits, transforming them into erasures at the decoder, while shortening decreases the rate by fixing information bits to known values and not transmitting them. Both methods preserve the core structure of the LDPC matrix, enabling flexible rate compatibility for adaptive communication systems. Modern standards such as DVB-S2, 5G NR, and Wi-Fi rely on these techniques to provide robust performance across a wide range of channel conditions. With continued advances in pattern optimization and decoder design, punctured and shortened LDPC codes remain a vibrant area of research and a practical cornerstone of digital communications.