Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

Tsinghua Science & Technology

Fast and Scalable GPU-Based RPA Decoder for Reed-Muller Codes

Kairui Tian, He Sun, Zhanxian Liu, Rongke Liu

Published Aug 5, 2025• 2 citations
Publisher

Abstract

Abstract The emerging Massive Communication (MC) highlights the need for efficient short-length error-correction coding schemes with effective decoders. Classical Reed-Muller (RM) codes combined with the recently developed Recursive Projection-Aggregation (RPA) decoder present a promising solution, as the RPA decoder demonstrates near Maximum Likelihood (ML) performance and supports highly parallel implementation. To address the speed and flexibility requirements of Cloud Radio Access Networks (C-RANs) across various MC applications, this paper proposes a fast and scalable RPA decoder on Graphics Processing Units (GPUs). By leveraging a thread-per-projection mapping strategy, we develop an optimized thread block architecture for the RPA decoding of second-order RM codes, which can be easily extended to construct a multi-dimensional block array for decoding higher-order RM codes. Additionally, we introduce a stationary projection pruning technique that seamlessly adapts the RPA decoder kernel to simplified variants, facilitating flexible trade-offs between error-correction performance and implementation complexity. Experimental results show that the pruned RPA decoder kernel on the NVIDIA A100 GPU achieves throughputs of 1.69 Gbps and 1.33 Gbps for the RM (6, 2) and RM (7, 2) codes, respectively, delivering speedups of 2.95× and 3.69× compared to a state-of-the-art software-based Successive Cancellation List (SCL) decoder.

DOI
10.26599/tst.2025.9010001

Related papers

Machine Learning-Aided Efficient Decoding of Reed–Muller Subcodes

Mohammad Vahid Jamali, Xiyang Liu, Ashok Vardhan Makkuva +3 more

Reed-Muller (RM) codes achieve the capacity of general binary-input memoryless symmetric channels and are conjectured to have a comparable performance to that of random codes in terms of scaling laws. However, such results are established assuming maximum-likelihood decoders for general code parameters. Also, RM codes only admit limited sets of rates. Efficient decoders such as successive cancellation list (SCL) decoder and recently-introduced recursive projection-aggregation (RPA) decoders are available for RM codes at finite lengths. In this paper, we focus on subcodes of RM codes with flexible rates. We first extend the RPA decoding algorithm to RM subcodes. To lower the complexity of our decoding algorithm, referred to as subRPA, we investigate different approaches to prune the projections. Next, we derive the soft-decision based version of our algorithm, called soft-subRPA, that not only improves upon the performance of subRPA but also enables a differentiable decoding algorithm. Building upon the soft-subRPA algorithm, we then provide a framework for training a machine learning (ML) model to search forgoodsets of projections that minimize the decoding error rate. Training our ML model enables achieving very close to the performance of full-projection decoding with a significantly smaller number of projections. We also show that the choice of the projections in decoding RM subcodes matters significantly, and our ML-aided projection pruning scheme is able to find agoodselection, i.e., with negligible performance degradation compared to the full-projection case, given a reasonable number of projections.

arXiv:2301.06251•IEEE Journal on Selected Areas in Information Theory•Jan 1, 2023•3 citations

Pipelined Architecture for Soft-Decision Iterative Projection Aggregation Decoding for RM Codes

Marzieh Hashemipour-Nazari, Yuqing Ren, Kees G. W. Goossens +1 more

The recently proposed recursive projection-aggregation (RPA) decoding algorithm for Reed-Muller codes has received significant attention as it provides near-ML decoding performance at reasonable complexity for short codes. However, its complicated structure makes it unsuitable for hardware implementation. Iterative projection-aggregation (IPA) decoding is a modified version of RPA decoding that simplifies the hardware implementation. In this work, we present a flexible hardware architecture for the IPA decoder that can be configured from fully-sequential to fully-parallel, thus making it suitable for a wide range of applications with different constraints and resource budgets. Our simulation and implementation results show that the IPA decoder has 41% lower area consumption, 44% lower latency, four times higher throughput, but currently seven times higher power consumption for a code with block length of 128 and information length of 29 compared to a state-of-the-art polar successive cancellation list (SCL) decoder with comparable decoding performance.

arXiv:2210.11069•IEEE Transactions on Circuits and Systems I Regular Papers•Aug 31, 2023•11 citations

Reduced Complexity RPA Decoder for Reed-Muller Codes

Jiajie Li, Syed Mohsin Abbas, Thibaud Tonnellier +1 more

The recursive projection-aggregation (RPA) decoder is a recently proposed near maximum likelihood (ML) decoder for Reed-Muller (RM) codes with low rates and short code lengths. However, the high computational complexity of RPA decoding is a major bottleneck for using RPA in applications that have a limited resource and energy budget. In this work, syndrome-based early stopping techniques as well as a scheduling scheme are proposed for the RPA decoder, which help in reducing the computational complexity while keeping similar decoding performance. Comparing to the baseline RPA decoder, the proposed techniques result in a 69−98% reduction in the average computational complexity for a target frame error rate (FER) of 10−5. Additionally, this work introduces hardware-friendly approximation functions to replace the RPA’s computationally expensive transcendental projection function.

International Symposium on Turbo Codes and Iterative Information Processing•Aug 30, 2021•16 citations

Efficient RPA Decoding of Reed-Muller Codes with Structured 2-D Subspace Projections

Vallabhaneni Kushal, Shiv Pratap Singh Rathore, Navin Kashyap

The recursive projection-aggregation (RPA) decoder achieves near-maximum likelihood (ML) performance for lowrate Reed-Muller (RM) codes but suffers from high computational complexity when applied to high-rate codes. To address this limitation, we propose a structured subspace projection framework tailored to the code order: specifically, 2-dimensional subspaces for third-order RM codes. We demonstrate that decoding via projections onto these carefully chosen subspaces achieves better performance compared to an equal number of projections onto randomly selected subspaces. Further, we integrate automorphism ensemble decoding (AED) with our pruned projections, leveraging the rich symmetry group of RM codes to recover performance degradation from subspace pruning.

National Conference on Communications•Feb 26, 2026•0 citations

Automorphism Ensemble Decoding on GPU: Achieving High Throughput and Low Latency for Polar and RM Codes

Yansong Li, Kairui Tian, Rongke Liu

Automorphism ensemble decoding (AED) is a highly parallel approach that enables decoding of polar and Reed-Muller (RM) codes with automorphisms, offering a practical solution with near-maximum likelihood (ML) performance and manageable computational complexity. To meet the growing demands for high throughput and low latency in cloud and virtual random access networks, this paper presents a graphics processing unit (GPU)-based AED architecture for polar and RM codes, utilizing low-complexity successive cancellation (SC) and small list SC (SCL) decoders as the constituent of AED. The proposed architecture exploits the inherent parallelism of AED to optimize decoding tasks on the GPU, significantly enhancing throughput by efficiently harnessing the extensive parallel processing capabilities of the GPU. Additionally, improved thread mapping and data management techniques markedly reduce latency for automorphism ensemble SC (Aut-SC) decoding, while a low-latency sorting mechanism further accelerates automorphism ensemble SCL (Aut-SCL) decoding. Experimental results on an NVIDIA RTX 4090 demonstrate that the proposed Aut-SC decoder, with an ensemble size of 8, achieves a throughput exceeding 17 Gbps. Compared to the state-of-the-art softwarebased SCL decoders, the proposed GPU-based Aut-SC and Aut-SCL architectures outperform existing solutions by factors of up to 28× and 10×, respectively, in normalized throughput while maintaining the same or even superior error correction performance.

IEEE Transactions on Signal Processing•Jan 1, 2025•1 citation

Sparse Multi-Decoder Recursive Projection Aggregation for Reed-Muller Codes

Dorsa Fathollahi, Nariman Farsad, Seyyed Ali Hashemi +1 more

Reed-Muller (RM) codes are one of the oldest families of codes. Recently, a recursive projection aggregation (RPA) decoder has been proposed, which achieves a performance that is close to the maximum likelihood decoder for short-length RM codes. One of its main drawbacks, however, is the large amount of computations needed. In this paper, we devise a new algorithm to lower the computational budget while keeping a performance close to that of the RPA decoder. The proposed approach consists of multiple sparse RPAs that are generated by performing only a selection of projections in each sparsified decoder. In the end, a cyclic redundancy check (CRC) is used to decide between output codewords. Simulation results show that our proposed approach reduces the RPA decoder's computations by up to 80% with negligible performance loss.

arXiv:2011.12882•IEEE International Symposium on Information Theory (ISIT)•Jul 12, 2021•31 citations
Abstract The emerging Massive Communication (MC) highlights the need for efficient short-length error-correction coding schemes with effective decoders. Classical Reed-Muller (RM) codes combined with the recently developed Recursive Projection-Aggregation (RPA) decoder present a promising solution, as the RPA decoder demonstrates near Maximum Likelihood (ML) performance and supports highly parallel implementation. To address the speed and flexibility requirements of Cloud Radio Access Networks (C-RANs) across various MC applications, this paper proposes a fast and scalable RPA decoder on Graphics Processing Units (GPUs). By leveraging a thread-per-projection mapping strategy, we develop an optimized thread block architecture for the RPA decoding of second-order RM codes, which can be easily extended to construct a multi-dimensional block array for decoding higher-order RM codes. Additionally, we introduce a stationary projection pruning technique that seamlessly adapts the RPA decoder kernel to simplified variants, facilitating flexible trade-offs between error-correction performance and implementation complexity. Experimental results show that the pruned RPA decoder kernel on the NVIDIA A100 GPU achieves throughputs of 1.69 Gbps and 1.33 Gbps for the RM (6, 2) and RM (7, 2) codes, respectively, delivering speedups of 2.95× and 3.69× compared to a state-of-the-art software-based Successive Cancellation List (SCL) decoder.
Reed-Muller (RM) codes achieve the capacity of general binary-input memoryless symmetric channels and are conjectured to have a comparable performance to that of random codes in terms of scaling laws. However, such results are established assuming maximum-likelihood decoders for general code parameters. Also, RM codes only admit limited sets of rates. Efficient decoders such as successive cancellation list (SCL) decoder and recently-introduced recursive projection-aggregation (RPA) decoders are available for RM codes at finite lengths. In this paper, we focus on subcodes of RM codes with flexible rates. We first extend the RPA decoding algorithm to RM subcodes. To lower the complexity of our decoding algorithm, referred to as subRPA, we investigate different approaches to prune the projections. Next, we derive the soft-decision based version of our algorithm, called soft-subRPA, that not only improves upon the performance of subRPA but also enables a differentiable decoding algorithm. Building upon the soft-subRPA algorithm, we then provide a framework for training a machine learning (ML) model to search forgoodsets of projections that minimize the decoding error rate. Training our ML model enables achieving very close to the performance of full-projection decoding with a significantly smaller number of projections. We also show that the choice of the projections in decoding RM subcodes matters significantly, and our ML-aided projection pruning scheme is able to find agoodselection, i.e., with negligible performance degradation compared to the full-projection case, given a reasonable number of projections.
The recently proposed recursive projection-aggregation (RPA) decoding algorithm for Reed-Muller codes has received significant attention as it provides near-ML decoding performance at reasonable complexity for short codes. However, its complicated structure makes it unsuitable for hardware implementation. Iterative projection-aggregation (IPA) decoding is a modified version of RPA decoding that simplifies the hardware implementation. In this work, we present a flexible hardware architecture for the IPA decoder that can be configured from fully-sequential to fully-parallel, thus making it suitable for a wide range of applications with different constraints and resource budgets. Our simulation and implementation results show that the IPA decoder has 41% lower area consumption, 44% lower latency, four times higher throughput, but currently seven times higher power consumption for a code with block length of 128 and information length of 29 compared to a state-of-the-art polar successive cancellation list (SCL) decoder with comparable decoding performance.
The recursive projection-aggregation (RPA) decoder is a recently proposed near maximum likelihood (ML) decoder for Reed-Muller (RM) codes with low rates and short code lengths. However, the high computational complexity of RPA decoding is a major bottleneck for using RPA in applications that have a limited resource and energy budget. In this work, syndrome-based early stopping techniques as well as a scheduling scheme are proposed for the RPA decoder, which help in reducing the computational complexity while keeping similar decoding performance. Comparing to the baseline RPA decoder, the proposed techniques result in a 69−98% reduction in the average computational complexity for a target frame error rate (FER) of 10−5. Additionally, this work introduces hardware-friendly approximation functions to replace the RPA’s computationally expensive transcendental projection function.
The recursive projection-aggregation (RPA) decoder achieves near-maximum likelihood (ML) performance for lowrate Reed-Muller (RM) codes but suffers from high computational complexity when applied to high-rate codes. To address this limitation, we propose a structured subspace projection framework tailored to the code order: specifically, 2-dimensional subspaces for third-order RM codes. We demonstrate that decoding via projections onto these carefully chosen subspaces achieves better performance compared to an equal number of projections onto randomly selected subspaces. Further, we integrate automorphism ensemble decoding (AED) with our pruned projections, leveraging the rich symmetry group of RM codes to recover performance degradation from subspace pruning.
Automorphism ensemble decoding (AED) is a highly parallel approach that enables decoding of polar and Reed-Muller (RM) codes with automorphisms, offering a practical solution with near-maximum likelihood (ML) performance and manageable computational complexity. To meet the growing demands for high throughput and low latency in cloud and virtual random access networks, this paper presents a graphics processing unit (GPU)-based AED architecture for polar and RM codes, utilizing low-complexity successive cancellation (SC) and small list SC (SCL) decoders as the constituent of AED. The proposed architecture exploits the inherent parallelism of AED to optimize decoding tasks on the GPU, significantly enhancing throughput by efficiently harnessing the extensive parallel processing capabilities of the GPU. Additionally, improved thread mapping and data management techniques markedly reduce latency for automorphism ensemble SC (Aut-SC) decoding, while a low-latency sorting mechanism further accelerates automorphism ensemble SCL (Aut-SCL) decoding. Experimental results on an NVIDIA RTX 4090 demonstrate that the proposed Aut-SC decoder, with an ensemble size of 8, achieves a throughput exceeding 17 Gbps. Compared to the state-of-the-art softwarebased SCL decoders, the proposed GPU-based Aut-SC and Aut-SCL architectures outperform existing solutions by factors of up to 28× and 10×, respectively, in normalized throughput while maintaining the same or even superior error correction performance.
Reed-Muller (RM) codes are one of the oldest families of codes. Recently, a recursive projection aggregation (RPA) decoder has been proposed, which achieves a performance that is close to the maximum likelihood decoder for short-length RM codes. One of its main drawbacks, however, is the large amount of computations needed. In this paper, we devise a new algorithm to lower the computational budget while keeping a performance close to that of the RPA decoder. The proposed approach consists of multiple sparse RPAs that are generated by performing only a selection of projections in each sparsified decoder. In the end, a cyclic redundancy check (CRC) is used to decide between output codewords. Simulation results show that our proposed approach reduces the RPA decoder's computations by up to 80% with negligible performance loss.