End-to-End DAE-LDPC-OFDM Transceiver with Learned Belief Propagation Decoder for Robust and Power-Efficient Wireless...

Area of Science:

• Wireless Communication Engineering
• Machine Learning for Communications
• Signal Processing

Background:

• Conventional wireless systems often treat encoding, modulation, and decoding independently.
• Existing systems struggle with dynamic adaptation to varying channel and noise conditions.
• There is a need for robust, energy-efficient, and adaptive communication architectures for future networks.

Purpose of the Study:

• To propose a novel Deep Autoencoder-LDPC-OFDM transceiver architecture.
• To integrate a learned belief propagation decoder for enhanced performance.
• To achieve robust, energy-efficient, and adaptive wireless communication.

Main Methods:

• End-to-end joint optimization of encoding, modulation, and decoding components.
• Integration of a learned belief propagation (BP) decoder with trainable parameters.
• Iterative message-passing process for adaptive refinement of log-likelihood ratio (LLR) statistics.

Main Results:

• Achieved a Bit Error Rate (BER) of 1.72% and Block Error Rate (BLER) of 2.95% at 10 dB SNR.
• Outperformed state-of-the-art models (Transformer-OFDM, CNN-OFDM, GRU-OFDM) by 25-30%.
• Demonstrated superior performance over traditional LDPC-OFDM systems by 38-42% across datasets.
• Achieved 26.6% Peak-to-Average Power Ratio (PAPR) reduction and low inference latency (3.9 ms).

Conclusions:

• The proposed Deep Autoencoder-LDPC-OFDM architecture offers high performance, power efficiency, and scalability.
• It provides superior reliability and low-latency communication suitable for 6G and beyond.
• The system demonstrates robustness in realistic wireless environments including time-varying and multipath fading channels.