More and more CPU-based SDR systems appear in recent two years. Such system requires high speed real-time signal processing. In this paper, we present our effort on the speed optimization of Turbo decoder, the most co...
详细信息
ISBN:
(纸本)9781618399267
More and more CPU-based SDR systems appear in recent two years. Such system requires high speed real-time signal processing. In this paper, we present our effort on the speed optimization of Turbo decoder, the most computation-demanding module in all baseband modules. We jointly consider the algorithm parallelism and the processor architecture. Single Instruction Multiple Data (SIMD) instruction is used for software implementation. The evaluation results show that this proposed design can achieve a maximum of 124 Mbps throughput for single Soft Input Soft Output (SISO) module with Max-Log-map algorithm.
To decrease the complexity of map algorithm, reduced state or reduced search techniques can be applied. In this paper we propose a reduced search soft output detection algorithm fully based on the principle of M a...
详细信息
To decrease the complexity of map algorithm, reduced state or reduced search techniques can be applied. In this paper we propose a reduced search soft output detection algorithm fully based on the principle of M algorithm for turbo equalization, which is a suboptimum version of the Lee algorithm. This algorithm is called soft output M algorithm (denoted as SO M algorithm), which applies the M strategy to both the forward recursion and the extended forward recursion of the Lee algorithm. Computer simulation results show that, by properly selecting and adjusting the breadth parameter and depth parameter during the iteration of turbo equalization, this algorithm can obtain good performance and complexity trade off.
Parallel decoding of Turbo codes is vital to the applications with very high data rates. There are mainly two existing methods for the parallel decoding of turbo codes: one is to have the sub-blocks being overlapped (...
详细信息
Parallel decoding of Turbo codes is vital to the applications with very high data rates. There are mainly two existing methods for the parallel decoding of turbo codes: one is to have the sub-blocks being overlapped (OL);the other is to store the intermediate information of last iteration at the sub-block boundary (SBI). The overlapping in OL methods will slow down the decoding speed while the SBI requires some extra memory. In this paper, we present a new method which is essentially the combination of the OL and SBI but with two new features being introduced: (1) instead of storing the full information of the boundary distribution as SBI does, the proposed method only stores the index of the most probable state at the boundary and a reliability metric for initialization;(2) the boundary positions of the sub-blocks are moving in each iteration of the turbo decoding process. The proposed method outperforms the existing methods and is flexible in design and implementation.
暂无评论