Research On A New HARQ Scheme Based On Spatial Gain In MIMO-OFDM Systems

The combination of MIMO and OFDM greatly increases the system capacity in the sense that OFDM and MIMO can respectively provide flat fading channel and spatial gain. The traditional MIMO system either uses BLAST to get spatial multiplexing gain or STC to get spatial diversity gain, but they both can be uniformly treated as spatial gain as they are obtained from the spatial dimension of multiple antenna. This paper proposes an HARQ method that mainly depends on the structure of STBC matrix to trade-off between the spatial gains. In this method the HARQ retransmission scheme is equivalent as transmitting different matrix rows, so by introducing the time dimension, the notions of multiplexing and diversity gains can be successfully unified.At first, this paper analyses the impulse response model of a multi-path fading channel. It gives an overview of the BLAST and STBC algorithms. Based on Alamouti's model, we also study a system that can get full spatial multiplexing and full spatial diversity gains.Secondly, this paper proposes a method combining BLAST and STBC called the B-S algorithm. As we already know, there are two types of STBC matrixes: Orthogonal and quasi-orthogonal matrixes. The orthogonal matrix can get full diversity gain but it cannot reach full rate and conversely, the quasi-orthogonal matrix can get full rate but can only get part of the diversity gain. The Alamouti model is the only one that can get full diversity gain and full rate at the same time. The basic B-S algorithm is based on Alamouti's model, but while trying to achieve full diversity, we derive an orthogonal B-S method using an orthogonal matrix; similarly for the full rate scheme, a quasi-orthogonal B-S algorithm is also derived. The simulation result shows that orthogonal B-S outperforms at middle and lower SNR but quasi-orthogonal B-S is better at higher SNR.Knowing that the decoding performance depends on the Euclidean distance between constellation points and that high level modulations have different Euclidean distances within their respective constellations, the overall performance can be improved by using different symbol mappings and soft-value decoding to get larger Euclidean distances at retransmissions. As the B-S algorithm relies on the transmitter matrix, constellation remapping can be divided in two types as well: The row-by-row remapping and the matrix remapping. The line remapping has better performance but higher complexity, it can make use of ML and Sphere Decoding. On the other hand, the second method based on the whole matrix retransmission keeps the structure of the matrix intact and so the decoding procedure becomes very simple. The LLR values can even be used directly to decode but the performance decreases slightly due to loss of diversity during remapping. Therefore, the system must trade-off between complexity and performance.Bases on the above analysis, this paper proposes a B-S-HARQ scheme, a low complexity approach based on matrix of LLR values constellation remapping and therefore a orthogonal matrix is used. At an HARQ retransmission, the matrix rows are kept inside the matrix and the constellation mapping remains unchanged until the next iteration where the following matrix is sent. This approach guarantees a low complexity decoding as it can use the matrix inverse and the LLR values between different matrixes to improve the system performance. This paper also proposes an optimization algorithm based on SD (Sphere Dedocing) where row and constellation remappings are applied. This method uses quasi-orthogonal matrixes, at an HARQ retransmission, each row of the matrix changes the constellation mapping and therefore guaranteeing the best system performance. The reason is because while the system gets spatial diversity, the highest remapping diversity gain is also attained. However, ML must be used to obtain such high performance therefore, this paper proposes the use of SD as it is less complex and its performance is relatively close to that of ML.Finally, simulations were conducted on every scheme indicated above, good results demonstrated their performance and on this basis, further research topics are proposed.