Zobrazit v EDS

Parallel Non-Iterative Cascaded Feedforward Neural Network Decoder for Low-Density Parity-Check Codes.

Uloženo v:
Podrobná bibliografie
Název: Parallel Non-Iterative Cascaded Feedforward Neural Network Decoder for Low-Density Parity-Check Codes.
Autoři: Ahire, Harshawardhan P.1 hpahire.p17@el.vjti.ac.in, Wagh, Sushama R.2 srwagh@ee.vjti.ac.in
Zdroj: IAENG International Journal of Computer Science. Mar2025, Vol. 52 Issue 3, p566-578. 13p.
Témata: Feedforward neural networks, Bit error rate, Low density parity check codes, Telecommunication systems, Decoders & decoding, Tanner graphs
Abstrakt: Low-density parity-check (LDPC) codes are widely used in modern systems because they are highly effective for error correction, nearing the Shannon limit performance across diverse communication channels. However, choosing an appropriate decoder for LDPC is crucial for accurate information retrieval. Traditional decoders, such as the message- passing sum-product algorithm (MP-SPA), message-passing belief propagation (MP-BP) and its variants, often suffer from computational complexity, error floors, and latency owing to their probabilistic and iterative nature. This experiment utilizes a cascaded feedforward neural network (CFNN) as a non- iterative decoder for quasi-cyclic (QC) LDPC codes. The CFNN completed 376 iterations within 43 minutes and 21 seconds, achieving 41.9% performance with a 3.62−02 gradient over six validation checks. The bit error rate (BER) of the CFNN decoder for the QC-LDPC improved by 10−0.24 compared to conventional decoders at a signal-to-noise ratio (SNR) of 3, with the CFNN reaching a BER of 10−4.5 at 5 dB SNR versus 10−3.6 for conventional decoders. The results show that the CFNN decoder excels. Overall, the CFNN proved to be an effective non-iterative alternative, overcoming the limitations of traditional decoders, such as suboptimal performance and error floors. The enhanced BER performance at various SNR levels demonstrates the efficacy of the CFNN decoder in improving decoding accuracy and reliability in communication systems. [ABSTRACT FROM AUTHOR]
Databáze: Supplemental Index
Popis
Abstrakt:Low-density parity-check (LDPC) codes are widely used in modern systems because they are highly effective for error correction, nearing the Shannon limit performance across diverse communication channels. However, choosing an appropriate decoder for LDPC is crucial for accurate information retrieval. Traditional decoders, such as the message- passing sum-product algorithm (MP-SPA), message-passing belief propagation (MP-BP) and its variants, often suffer from computational complexity, error floors, and latency owing to their probabilistic and iterative nature. This experiment utilizes a cascaded feedforward neural network (CFNN) as a non- iterative decoder for quasi-cyclic (QC) LDPC codes. The CFNN completed 376 iterations within 43 minutes and 21 seconds, achieving 41.9% performance with a 3.62−02 gradient over six validation checks. The bit error rate (BER) of the CFNN decoder for the QC-LDPC improved by 10−0.24 compared to conventional decoders at a signal-to-noise ratio (SNR) of 3, with the CFNN reaching a BER of 10−4.5 at 5 dB SNR versus 10−3.6 for conventional decoders. The results show that the CFNN decoder excels. Overall, the CFNN proved to be an effective non-iterative alternative, overcoming the limitations of traditional decoders, such as suboptimal performance and error floors. The enhanced BER performance at various SNR levels demonstrates the efficacy of the CFNN decoder in improving decoding accuracy and reliability in communication systems. [ABSTRACT FROM AUTHOR]
ISSN:1819656X