Godkända
SOVA Based LTE Turbo Decoders
ANG LAY HONG () och LIM WEE GUAN ()
Start
2009-02-01
Presentation
2009-08-31
Plats:
Avslutat:
2009-09-23
Examensrapport:
Sammanfattning
The Max-Log-MAP algorithm is commonly used in a constituent decoder for turbo coding applications. In this thesis, the use of Soft-Output Viterbi Algorithms (SOVAs) as an alternative to Max-Log-MAP for use in the 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) standard is investigated. The decoding performance of the Battail Rule SOVA (BR-SOVA) was found to be comparable to that of the Max-Log- MAP although there is a high price to be paid in terms of hardware complexity. A Simplified Battail Rule SOVA (SB-SOVA) algorithm that is suitable for hardware implementation is proposed, and the decoding performance of SB-SOVA is similar to BR-SOVA. A proposed hybrid- SOVA architecture that combines both Hagenauer Rule SOVA (HR-SOVA) and BR-SOVA to reduce the computational complexity of SB-SOVA is described, and the decoding performance was found to be within 0.1 dB of SB-SOVA. To improve throughput and reduce the latency, the input data block is divided into windows to allow for parallel processing. The -stage warm-up method to determine the initial state of a window was found to give better performance as compared to the next iteration initialize (NII) method, with the performance of a windowed decoder performing within 0.1 dB of the ideal Max-Log-MAP. Hardware architectures for the SB-SOVA and hybrid- SOVA decoders are presented and the memory requirements for SOVA were assessed to be around 10 % of that required for Max-Log-MAP on a per-window basis. The SB-SOVA and hybrid-SOVA decoders consume 87 % and 30 % more hardware resources per window respectively as compared to the Max-Log-MAP decoder. For the same degree of parallelization, the SOVA-based decoders provide 34 % higher throughput than a Max- Log-MAP decoder. The increased throughput of the SOVA architecture may enable the SB-SOVA based decoder to have lower hardware requirements as compared to the Max-Log-MAP as the the degree of parallelization required for a given data rate will be lower
Handledare: Matthias Kamuf (Ericsson)
Examinator: Joachim Rodrigues (EIT)