Accelerating Neural BP-Based Decoder Using Coded Distributed Computing

While neural BP-based (NBP) decoders exhibit superior error correction performance compared to belief-propagation (BP) decoders, the NBP decoder's high computational and memory requirements impede its practical deployment in communication systems. To overcome this challenge, we propose a Coded...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE transactions on vehicular technology Ročník 73; číslo 9; s. 13997 - 14002
Hlavní autori: Han, Xuesong, Liu, Rui, Li, Yong, Yi, Chen, He, Jiguang, Wang, Ming
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: New York IEEE 01.09.2024
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Predmet:
BP-based decoder
Codes
coding theory
Communications systems
Complexity theory
Decoders
Decoding
Distributed computing
Distributed processing
Encoding
Error correction
neural network
Neurons
straggler resilience
ISSN:0018-9545, 1939-9359
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:While neural BP-based (NBP) decoders exhibit superior error correction performance compared to belief-propagation (BP) decoders, the NBP decoder's high computational and memory requirements impede its practical deployment in communication systems. To overcome this challenge, we propose a Coded Neural BP (CNBP) scheme to accelerate the NBP decoder in distributed environments, while considering storage constraints and providing resilience to stragglers. The key idea is to reformulate the primary operations of the NBP decoder as matrix-vector multiplications by introducing weight matrices and transformations. Based on this, the acceleration of the NBP decoder is achieved by speeding up matrix-vector multiplications using coded distributed computing. Extensive experiments conducted on Amazon EC2 cluster demonstrate that CNBP achieves notable acceleration and scalability performance without any loss in error correction performance.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9545
1939-9359
DOI:10.1109/TVT.2024.3391836