An Iterative BP-CNN Architecture for Channel Decoding (1707.05697v1)

Abstract: Inspired by recent advances in deep learning, we propose a novel iterative BP-CNN architecture for channel decoding under correlated noise. This architecture concatenates a trained convolutional neural network (CNN) with a standard belief-propagation (BP) decoder. The standard BP decoder is used to estimate the coded bits, followed by a CNN to remove the estimation errors of the BP decoder and obtain a more accurate estimation of the channel noise. Iterating between BP and CNN will gradually improve the decoding SNR and hence result in better decoding performance. To train a well-behaved CNN model, we define a new loss function which involves not only the accuracy of the noise estimation but also the normality test for the estimation errors, i.e., to measure how likely the estimation errors follow a Gaussian distribution. The introduction of the normality test to the CNN training shapes the residual noise distribution and further reduces the BER of the iterative decoding, compared to using the standard quadratic loss function. We carry out extensive experiments to analyze and verify the proposed framework. The iterative BP-CNN decoder has better BER performance with lower complexity, is suitable for parallel implementation, does not rely on any specific channel model or encoding method, and is robust against training mismatches. All of these features make it a good candidate for decoding modern channel codes.

• The paper presents an iterative BP-CNN architecture that interleaves belief propagation with CNN-based noise refinement to boost SNR and reduce BER.
• It employs a novel dual-objective loss function integrating noise estimation accuracy with the Jarque-Bera test to align residuals with AWGN assumptions.
• The approach enhances decoding performance under correlated noise while offering flexibility for various channel models and supporting parallel computation.

An Iterative BP-CNN Architecture for Channel Decoding

The paper at hand introduces a novel approach to channel decoding using a combination of belief propagation (BP) and convolutional neural networks (CNNs). The iterative BP-CNN architecture targets the decoding challenge posed by correlated noise in communication channels. By integrating deep learning with traditional BP decoding, the authors aim to improve decoding performance under complex noise conditions.

Core Contributions

1. Iterative BP-CNN Structure: The proposed system is an iterative architecture that interweaves a trained CNN with a BP decoder. The decoding process consists of alternating between BP, which estimates the coded bits, and CNN, which refines noise estimation. This interleaving is designed to gradually enhance the signal-to-noise ratio (SNR) and improve bit error rates (BER).
2. Novel Loss Function for CNN Training: The paper defines an innovative loss function for CNN training, which integrates noise estimation accuracy with a statistical normality test, specifically the Jarque-Bera test, of the residuals. This dual-objective approach not only reduces the power of the residual noise but also tailors the noise distribution to better suit the BP algorithm's assumptions, which are often optimized for AWGN channels.
3. Flexibility and Robustness: The proposed architecture does not depend on specific channel models or encoding methods, thereby being adaptable to various noise characteristics. Additionally, it supports parallel computing, which enhances its suitability for implementation in contemporary communication systems.

Numerical and Practical Insights

The authors present extensive experimental results that demonstrate the efficacy of the BP-CNN decoder in reducing BER compared to traditional BP decoding. Under strong noise correlation, significant performance improvements are reported. Notably, the enhanced BP-CNN, with its specialized loss function, performs better than the baseline BP-CNN in handling residual noise, showcasing the benefit of incorporating statistical distribution considerations in the loss function design.

Theoretical Implications and Future Directions

The fusion of deep learning techniques with traditional graph-based decoding introduces new avenues for improving decoding algorithms, especially in non-ideal channel conditions. By leveraging the feature extraction capabilities of CNNs, the iterative BP-CNN decoder marks a step towards adaptive and efficient channel decoding solutions. This method can be further explored with more complex and varied neural network architectures, potentially incorporating other types of neural networks such as recurrent neural networks (RNNs) which might provide benefits for temporal correlation handling.

Conclusion

The iterative BP-CNN architecture represents a promising direction for accommodating correlated noise in channel decoding. By effectively marrying BP's iterative message passing abilities with CNN's feature learning strengths, this framework opens up robust solutions for next-generation communication environments. Future work should continue to explore the scalability of this approach, its computational efficiency, and its application across diverse and challenging communication scenarios.