| Summary: | 博士 === 國立交通大學 === 電子工程系所 === 96 === This dissertation presents algorithm designs for sphere decoders and low-density parity check (LDPC) code decoders in multi-input multi-output (MIMO) systems from implementation point of view. Based on statistical techniques, complexity reduction schemes are proposed. Sphere decoders of hard-decision outputs and LDPC decoding algorithms in AWGN channel are discussed first. Then the sphere decoders with soft-decision outputs for channel-coded MIMO systems are investigated.
Sphere decoding algorithm is one realization of maximum likelihood signal detection for MIMO systems, and the computation can vary with channel due to the fading phenomena. Among several modified algorithms, K-best algorithm is suitable for hardware implementation for the constant computation and predictable hardware complexity. However, K-best algorithm has to be realized with the assumption of worst channel condition in order to maintain the system performance. For complexity reduction, an early pruning scheme combined with K-best algorithm is presented. According to the system model and channel statistics the expected complexity can be analyzed as well. Based on the complexity analysis, an early-pruned multi-K-best algorithm is proposed by which the lowest decoding speed can be further improved. The simulation results in 64-QAM 4 × 4 MIMO channel show that 90% complexity can be reduced with imperceptible degradation in both symbol error rate and bit error rate above 10−5.
For decoding LDPC codes, min-sum algorithm is one common simplification of Log-BP algorithm, but there is a performance gap due to the approximation inaccuracy. Normalization schemes are investigated to compensate the approximation error. Moreover, the normalization factor can be described by a function of the decoder inputs, noise variance, and the decoding iteration number. The data-dependent correction terms can be analyzed and derived by order statistic and density evolution. Simulated in DVB-S2 system, the dynamic normalization schemes effectively mend the SNR loss from Log-BP algorithm to min-sum algorithm with few normalization overheads, and 1.0dB SNR improvement, which is about 95% of the SNR loss from Log-BP to min-sum algorithm, can be achieved.
For channel coded MIMO systems, a sphere decoder is modified to a list sphere decoder that generates a candidate list for computing the soft inputs. Under iterative message passing decoding, the candidate list and the soft value generation impact the decoding convergence. Sufficiently large candidate list is required to avoid error floor. Thus, a path augmentation technique is proposed by which a larger and distinct list can be employed in computing the probabilistic information of each received bit. Compared with directly generating a larger list, path augmentation performs comparatively less operations. In our simulation based on a 64-QAM 4×4 MIMO system with LDPC codes defined in IEEE802.11n, the proposed augmented-list sphere decoder based on 64-best algorithm achieves the lowest error floor and saves about 50% computations, if compared to the standard list sphere decoder which is based on 128-best algorithm. Moreover, by the proposed early pruning scheme and multi-K-best algorithm, 94% reduction in sorting complexity can be achieved.
|
|---|
