The Research Of The Sphere Detection Algorithm For MIMO System

MIMO technology can increase the system capacity and spectrumefficiency without increasing the channel bandwidth and transmit power. It isconsidered to be one of the most significant technological breakthroughs in thewireless mobile communications. The performance of the signal detection algorithmhas a deep impact on the performance of the MIMO systems. Therefore the signaldetection algorithm is the current focus of the study. The sphere detection algorithmcan achieve the ML performance, and its computational complexity is polynomial-class in a certain range of system parameters. However, in low SNR range, or highorder of signal modulation, its computational complexity is still close to exponential,which has been a currently pressing problem.This thesis introduces the background of the MIMO system and severaltraditional MIMO detection algorithms in chapter one and chapter two. In chapterthree, the related theories of the sphere detection algorithm are detailed introduced.In chapter four and chapter five, this thesis proposes some improvements of thesphere detection algorithm. Finally, chapter six is the summary and outlook of thisthesis.The computational complexity of the sphere detection algorithm is still high inlow SNR range or high order of signal modulation. Therefore this thesis proposes anew sphere detection algorithm based on the ordering scheme, which is named RSIC.This algorithm is a combination of the SIC scheme and the RB scheme, it combinesthe advantage of the both two. However, when we calculate the reliability of everyelement, it is divided by the row norm of the pseudo-inverse of the channel matrix inorder to eliminate the noise amplification. In addition, this algorithm needs tocalculate the zero forcing solution in every iteration, so the interference caused bythe earlier detected signals in the SIC algorithm can also be offset. Simulation resultsshow that compared to the SIC、RB、BSQR algorithms, the SER performance of theproposed RSIC algorithm is almost exactly the same with other algorithms, but the computational complexity has been greatly reduced. For example, when SNR5dB,in a16QAM modulated44MIMO system, compared to the BSQR algorithmwhich has the lowest computational complexity among the other ordering schemes,the computational complexity of the RSIC algorithm is decreased by1/3, while themodulation order and the number of the transmit antennas is increased to64QAMand6respectively, the computational complexity is decreased by1/2and3/8or so.Moreover, two threshold algorithms are proposed in this thesis. It contrasts theaccumulated metric with the threshold when reaching a leaf node, and judge whetherthe maximum likelihood solution is found. The two proposed thresholds are all basedon the basic idea of statistics, named ST1and ST2. ST1is obtained through thecumulative probability distribution of the correct solution. In order to get the besttradeoff between the performance and the computational complexity, we can choosedifferent threshold according to different SNR region. ST2is the intersection of thecorrect solution and the wrong solution, because the closed-form solution can’t beobtained, it can only be estimated by linear, and it changes adaptively with thechannel matrix. Then the two strategies are combined with the ordering schemes,simulation results show that the performance of the algorithm is almost the samewhen using the threshold ST1, while the performance has a certain loss when usingST2. In the high SNR region, after using the two thresholds, the algorithm’s visitnodes all converge to m, which overcomes the problem that the visited nodesconverge to2m1in high SNR region due to the restriction of the SE search strategy.However, on the whole, the computational complexity of the algorithm whenadopting ST2is lower than the algorithm when using ST1.Finally, we combine the RSIC scheme with the threshold ST1、ST2. Simulationresults show that after combined with ST1, the performance is almost the same, thecomputational complexity has a certain degree of reduce, the converge value reducesfrom to. After combined with ST2, the performance has a certain loss, butthe computational complexity has been greatly reduced, when the SNR is0to30dB,it is almost close to.