As the world becomes increasingly interconnected, the demand for faster, more reliable, and highly efficient wireless communication continues to accelerate. The impending arrival of sixth-generation (6G) wireless systems promises to reshape connectivity by operating at higher frequencies—including terahertz bands—and delivering data rates in the range of terabits per second. However, achieving these ambitious targets requires fundamentally new approaches to channel coding. Next-generation coding techniques are the backbone of 6G, enabling robust error correction, ultra-low latency, and energy efficiency across massive device networks. This expanded article explores the latest developments in coding for 6G, including polar codes, low-density parity-check (LDPC) codes, rateless codes, and machine learning–driven adaptive schemes.

Understanding 6G Wireless Systems

6G is envisioned to succeed 5G around 2030, supporting transformative use cases such as holographic communications, digital twins, autonomous systems, and pervasive artificial intelligence. Operating in the sub-terahertz (100 GHz to 300 GHz) and terahertz (300 GHz to 3 THz) bands, 6G will achieve unprecedented data rates—potentially exceeding 1 Tbps—while reducing end-to-end latency to sub-millisecond levels. This paradigm shift imposes stringent requirements on physical-layer coding:

  • Extreme throughput: Codes must support very high data rates with minimal computational overhead.
  • Ultra-reliable low-latency communication (URLLC): Applications like remote surgery and autonomous driving demand near-instantaneous decoding.
  • Massive connectivity: 6G may connect millions of devices per square kilometer, each with limited power budgets.
  • Efficiency in diverse environments: Terahertz channels are highly susceptible to absorption and blockage, requiring codes that can adapt to rapidly changing conditions.

Meeting these challenges requires moving beyond the coding schemes inherited from 5G (primarily LDPC for data and polar codes for control) and developing new families of codes that can be tailored for the 6G air interface.

Evolution from 5G Coding to 6G Requirements

5G NR (New Radio) introduced two pillars of channel coding: LDPC codes for the data channel and polar codes for the control channel. While these choices provided significant gains over 4G’s turbo codes, 6G demands even more flexibility. For instance, 5G’s LDPC codes are optimized for code rates around 1/3 to 8/9 and block lengths up to 8448 bits, but 6G use cases may involve extremely short or very long blocks. Similarly, polar codes in 5G rely on successive cancellation list (SCL) decoding, which offers excellent performance at moderate complexity but may not scale well to terahertz bandwidths. As such, the 6G community is actively researching coding schemes that can gracefully adapt to a wide range of block lengths, code rates, and latency constraints while maintaining low error floors.

Key Coding Challenges in 6G

Developing next-generation coding techniques for 6G involves surmounting several interrelated hurdles:

  • Ultra-high data rates with perfect reliability: At speeds beyond 100 Gbps, even minor decoding inefficiencies become costly. Codes must offer near-Shannon-limit performance with hardware-friendly parallel architectures.
  • Microsecond-level latency: Real-time control loops require codes that can be decoded in a few microseconds. This motivates low-complexity decoders and iterative early termination.
  • Energy efficiency for massive IoT: Billions of low-power devices will rely on simple encoders and decoders. Codes must minimize active circuitry and support duty-cycled operation.
  • Multi-service multiplexing: 6G will simultaneously serve eMBB (enhanced mobile broadband), URLLC, and mMTC (massive machine-type communications) with heterogeneous quality-of-service (QoS) requirements. A unified coding framework that can dynamically adjust is highly desirable.
  • Channel variability: Terahertz signals are affected by atmospheric absorption, rain, foliage, and even blockage from hand gestures. Channel state information must be captured and exploited by adaptive coding.

These challenges drive the exploration of novel coding paradigms, many of which are described below.

Next-Generation Coding Techniques

Researchers worldwide are investigating a variety of advanced coding strategies tailored to 6G. The following sections detail the most promising approaches, drawing on recent publications and ongoing standardization discussions.

Polar Codes: Beyond the 5G Baseline

Polar codes, invented by Erdal Arıkan in 2008, were adopted for 5G control channels due to their capacity-achieving property and low encoding complexity. For 6G, polar codes are being enhanced in several dimensions:

  • High-rate polar codes: New construction algorithms (e.g., convolutional polar codes, polarization-adjusted convolutional (PAC) codes) improve performance at short block lengths, critical for URLLC.
  • List decoding with early termination: Advanced SCL decoders can reduce average latency by stopping once correct frames are detected, meeting sub-100 microsecond deadlines.
  • Rate-compatible polar codes: Design methods that allow puncturing and shortening enable seamless support for multiple code rates without changing the encoder structure—essential for adaptive modulation and coding (AMC).
  • Belief propagation (BP) decoders: Parallel BP decoders can be implemented in hardware to achieve high throughput, though with a performance gap to SCL. Hybrid BP/SCL decoders balance throughput and error correction.

One significant advantage of polar codes is their deterministic structure, which permits efficient hardware implementation. However, their latency at very long block lengths remains a concern, as successive cancellation decoding is inherently serial. Emerging designs, such as improved BP scheduling and list-based BP, aim to overcome this limitation. As discussed by researchers from IEEE, PAC codes have demonstrated significant performance gains over conventional polar codes in the short-block regime, making them a strong candidate for 6G control and short-packet applications.

LDPC Codes: Evolution for Massive MIMO and Terahertz

Low-density parity-check (LDPC) codes, already the workhorse of 5G data channels, are being refined for 6G. Their quasi-cyclic (QC) structure facilitates high-speed decoders using iterative message passing, supporting throughputs of 100 Gbps and beyond. Key research directions include:

  • Optimized degree distributions for terahertz channels: LDPC ensembles can be designed to match the fluctuating signal-to-noise ratios (SNRs) typical of wideband terahertz links, minimizing error floors.
  • Protograph-based codes: Protograph LDPC codes offer a compact representation and can be scaled for very high code rates (e.g., R > 0.9) with minimal performance loss, useful for high-SNR links.
  • Spatially coupled (SC) LDPC codes: These codes approach the Shannon capacity more closely and display excellent thresholds, making them attractive for low-latency applications when combined with windowed decoding.
  • Hardware-aware designs: To reduce power consumption in massive MIMO base stations, LDPC decoders are being co-designed with the receiver chain to share pipelining and memory resources.

One challenge for LDPC in 6G is the need for flexible rates and block lengths. While 5G LDPC supports a large number of base graphs, 6G may require even finer granularity. Techniques such as lifting and base graph extension can provide rate-compatibility without excessive complexity. Recent work from Qualcomm highlights how LDPC codes can be extended to support sub-100 μs latency while maintaining stringent BLER (block error rate) targets.

Rateless Codes: Fountain and Raptor Codes for Dynamic Channels

Rateless codes—also known as fountain codes—are ideal for scenarios where channel conditions are unpredictable or where a single transmitter must serve multiple receivers with varying SNRs. Unlike fixed-rate codes, rateless codes generate a potentially infinite stream of encoded symbols; a receiver can decode once it has collected enough symbols, regardless of which ones were lost. For 6G, this property offers several benefits:

  • Adaptive rates without feedback: In high-mobility scenarios (e.g., vehicle-to-everything), feedback loops may be too slow. Rateless codes automatically adjust to the instantaneous capacity.
  • Efficient broadcast/multicast: A single encoded stream can serve users with different channel qualities, as each receiver stops when it has enough symbols—ideal for content distribution over 6G multicast.
  • Combined with HARQ: Hybrid automatic repeat request (HARQ) schemes that use rateless codes can reduce retransmission overhead by transmitting incremental redundancy until successful decoding.

Luby transform (LT) codes and Raptor codes are the most common rateless families. For 6G, research focuses on reducing their decoding complexity—currently O(K log K) for RaptorQ—to meet terahertz speeds. Moreover, integration with polar and LDPC codes as outer codes is being explored to create hybrid rateless/fixed-rate schemes. A comprehensive survey by ACM examines how fountain codes can be deployed in beyond-5G networks, noting their potential for massive IoT and sporadic traffic.

Machine Learning–Based Codes: Adaptive and Learned

The rise of deep learning has opened the door to neural network–based coding and decoding, where the traditional code structure is replaced or augmented by learned components. Machine learning (ML) codes for 6G are particularly promising due to their ability to adapt to channel statistics that are difficult to model analytically. Key approaches include:

  • End-to-end learned coding: Autoencoder architectures treat the entire communication chain as a learned mapping, jointly optimizing the encoder, channel, and decoder. This can yield codes that outperform classical designs on specific channels (e.g., non-linear or multi-user channels).
  • Neural decoders for existing codes: Instead of designing new codes, ML can improve decoding of polar and LDPC codes. For example, convolutional neural networks (CNNs) can replace iterative message passing, offering faster convergence at the cost of training overhead.
  • Meta-learning for rate adaptation: A lightweight model can predict the best coding scheme (code type, rate, and modulation) based on real-time channel measurements, enabling dynamic code switching without explicit feedback.
  • Attention-based sequence models: Transformers and vision transformers have been adapted to decode short blocks near the optimal maximum a posteriori (MAP) bound, a feat that traditional suboptimal decoders cannot match.

However, deploying ML codes in practice comes with challenges: a) the need for large amounts of training data and retraining when channel conditions change, b) high computational cost during inference (especially for transformers), and c) lack of explainability and guaranteed performance bounds. Despite these obstacles, the 3GPP Technical Report TR 38.843 on 6G studies indicates that AI/ML-based physical layer components are being considered as a key enabler. An in-depth review from IEEE Communications Magazine discusses the trade-offs between learned codes and traditional ones, emphasizing that hybrid approaches—where classical code structures are combined with learned receivers—may dominate early 6G deployments.

Emerging Coding Paradigms for 6G

Beyond the four main families described above, several other coding concepts are gaining traction in the 6G research community.

Non-Orthogonal Multiple Access (NOMA) Coding

NOMA allows multiple users to share the same time-frequency resources by superimposing their signals in the power domain. This requires joint coding and decoding that separates users based on their codebooks. Sparse code multiple access (SCMA) and low-density spreading (LDS) are two prominent NOMA coding techniques. For 6G, NOMA codes need to be scalable to hundreds of simultaneous users while keeping the receiver complexity manageable. Machine learning is again being applied to learn optimal codebooks for multi-user detection.

Intelligent Reflecting Surface (IRS)-Aided Coding

IRS panels can programmatically control the propagation environment, but they also introduce phase noise and symbol distortions. Emerging work treats the IRS as part of the encoder, using space-time coding across the reflecting elements. This creates a coupled coding and beamforming problem. Recent results show that exploiting the IRS’s additional degrees of freedom can improve diversity and coding gain, especially for non-line-of-sight terahertz links.

The Critical Role of Machine Learning in Code Design

ML’s influence on 6G coding extends beyond learned codes. It also facilitates real-time optimization of coding parameters, such as rate, block length, and modulation order, based on environmental sensing. For example, a reinforcement learning agent can tune the coding rate to maximize throughput while respecting latency constraints. Another application is in channel estimation: accurate CSI (channel state information) is necessary for good coding performance, and deep neural networks can predict the effective channel from limited pilots.

Furthermore, ML is being used to automate the design of protograph LDPC codes or polar code construction for specific channels. By training a generative model on a dataset of good codes, researchers have discovered new rate-compatible LDPC patterns that outperform manually optimized designs. However, such codes often lack theoretical optimality, making their adoption in standards a lengthy process. The balance between data-driven flexibility and provable performance remains an active debate.

Future Outlook and Practical Implementation

As we approach the 2030 milestone for 6G commercialization, the coding techniques described here will need to transition from academic papers to real-world silicon. Key milestones include:

  • Standardization: 3GPP is expected to start the 6G study phase around 2025-2026, with coding schemes being a central topic. Likely, a modular framework will be adopted, enabling different codes for different use cases (e.g., polar-like codes for short packets, LDPC for streaming, rateless for broadcast).
  • Prototyping at Terahertz frequencies: Testbeds operating around 140 GHz are already being built. These will evaluate the practical throughput and decoder complexity of candidate codes. For instance, a 100 Gbps LDPC decoder in 7 nm CMOS has been demonstrated, but terahertz processing introduces new digital architecture challenges.
  • Energy-aware design: 6G base stations will consume huge amounts of power if not optimized. Codec hardware must be energy-proportional, scaling down with traffic load. Rateless and adaptive codes can help by reducing redundant transmissions.
  • Integration with AI-native air interface: The next-generation air interface may include an AI/ML layer that interacts with coding to perform joint source-channel coding, improving end-to-end reliability for applications like holographic video.

In summary, the transition from 5G to 6G coding is not merely an incremental improvement but a paradigm shift. The codes will become more adaptive, more tightly integrated with the physical channel, and in some cases, learned from data. While polar and LDPC codes will likely remain core, rateless and ML-based codes will expand their roles, especially in challenging multipoint and dynamic environments.

Conclusion

The successful deployment of 6G wireless systems hinges on the development of next-generation coding techniques that can handle extreme data rates, ultra-low latency, massive connectivity, and highly variable channels. This article has examined the evolution from 5G coding to the promising frontiers of polar codes, LDPC codes, rateless codes, and ML-based codes. Each family brings unique strengths—polar codes for short-block control, LDPC for high-throughput data, rateless codes for broadcast and dynamic links, and ML codes for adaptive performance. As research progresses, hybrid approaches that combine these techniques in a flexible framework will likely define the 6G standard. The path forward requires continued innovation in algorithm design, hardware implementation, and system integration, but the rewards—a truly connected world with seamless, instantaneous communication—are well worth the effort.