Compressed sensing based channel estimation for multiuser massive multiple input multiple output systems

Spatial multiplexing advantages of Massive MIMO systems which have an important part of the future generation cellular communication systems come to the forefront thanks to large number of antennas at base station. Base station must obtain downlink channel state information in order to fully benefit from spatial multiplexing in Multi-User Massive MIMO. In FDD Massive MIMO systems which do not have reciprocity like TDD systems, CSIT estimation with conventional methods like least squares needs long pilot symbols proportional with antenna numbers. CSIT estimation with long pilot symbols proportional with antenna numbers causes disadvantage for FDD Massive MIMO systems by decreasing the throughput of data packets. In the literature, there are studies for resolving this disadvantage of FDD Masssive MIMO by using channel vectors contained within sparsity. In consequence of increasing the number of base station antennas and with limited number of scatterers, it can be observed that sparsity in channel vectors belongs to the users. Sparsity behaviour of channel vectors makes it possible to use compressed sensing methods for CSIT estimation. In the literature, JOMP reconstruction algorithm which uses joint sparsity information among users was investigated. Within the scope of thesis, researches were carried out for improving the CSIT estimation performance of JOMP and conventional OMP reconstruction algorithms. Studies suggesting that there can be an angular reciprocity between uplink and downlink channels in FDD systems were investigated. Further studies are carried out about data based upon angular reciprocity from uplink channel can consistute a-priori information to reconstruction algorithms. Logit weighted function formed by a-priori information was adapted to JOMP algorithm. Thus, LW-JOMP algorithm was proposed for improving CSIT estimation of Multi-User FDD Massive MIMO system. Thanks to channel vectors that were created from MATLAB based QuaDRiGa radio channel generator, reconstruction algorithms were simulated. CSIT NMSE values, common non-zero elements' recovery rates and individual non-zero elements recovery rates were investigated versus different parameters. In simulations, it was observed that reconstruction algortihms that use a-priori information provides performance improvement in CSIT estimations.