Research On Channel Estimation And Optimal Training Design In Cloud Radio Access Networks

Nowadays, as the fourth generation (4G) networks is moving towards a successful commercial stage, mobile operators are facing great pressure to deal with the overwhelming data traffic brought by the rapid progress of mobile Internet. It becomes quite difficult for current cellular networks to provide higher data rates and support various quality of service guarantees without increasing capital expenditure and energy consumption. Cloud radio access networks (C-RANs) are promising solutions to address this situation by exploiting the benefits of incorporating cloud computing technology into wireless networks. With this concept, signal processing functions are concentrated into the cloud server and radio access functions are hanged over to the distributed remote radio heads (RRHs). In this way, global resource scheduling and management can be achieved in an energy-efficient way along with providing the nearly ubiquitous provision of high data rate coverage.This thesis focuses on the research on channel estimation and training design in C-RANs. The main contributions and innovations include the following aspects:(1) The architecture of a uplink C-RAN is established consisting of multiple user equipments (UEs) and RRHs connected to the cloud based baseband unit (BBU) pool.The communication links in C-RANs are modeled as two kinds of links: one is the radio access link (ACL) for communications between the UE and the RRH, and the other is the wireless fronthaul link (WFLs) connecting the RRH to the BBU pool. In order to be closer to the practical channel model, the communication links are assumed to be correlated in time and space domains. Moreover, the Gauss-Markov model is used to track the time evolution of the communication links, which is an effective way to describe the dynamic behavior of wireless channel.(2) The properties of channel state information (CSI) for different communication links in C-RANs are investigated along with proposing corresponding CSI acquisition strategies. By analyzing the signal transmission in C-RANs and the different CSI properties of ACLs and WFLs in space and time domains, the centralized signal processing and CSI acquisition schemes at the BBU pool are established. By using the Kalman filter, the sequential minimum mean-square-error (SMMSE) estimator is developed through a prior knowledge of long-term channel correlation statistics and the latest radio channel state. Compared with the traditional linear minimum mean-square-error (LMMSE) estimator, the simulation results have shown that the SMMSE estimator can efficiently decrease the channel estimation error and improve the quality of data transmission in C-RANs. Moreover, in certain coherent time interval, the estimation accuracy of the SMMSE estimator is improved as the fading block increased.(3)The efficient segment training transmission scheme is developed for C-RANs to obtain individual CSI of ACLs and WFLs. In segment training scheme, the RRHs are liberated from additional signal processing as the CSI acquisition originally operated at RRHs is migrated to the BBU pool. For both ACLs and WFLs, the structure of the optimal training sequences is derived by minimizing the mean-square-error (MSE)of channel estimation by taking the power constraint into account. In particular, the lengths of training sequences to guarantee accurate estimation for ACLs and WFLs are different and the MSE minimization criterion is not effective to fully depict the overall system performance of C-RANs. Thus, the tradeoff between the lengths of two segment training sequences is optimized by maximizing the spectral efficiency in the uplink transmission. Simulation results have demonstrated that the optimal training design can effectively improve the estimation accuracy compared with the other non-optimal structure. Furthermore, the optimal training length design can further improve the overall system performance.