UC San Diego
Using statistical information to improve communication in MIMO networks
- Author(s): Ghosh, Sagnik
- et al.
This dissertation discusses the use of statistical information about channels, or Channel Distribution Information (CDI), in multiuser MIMO networks. There are three main topics covered here. The first is on how to efficiently quantize CDI for feedback in a multiuser environment. The second is to develop a new outage framework for multiuser MIMO networks, and to provide a near-optimal solution for beamforming and power control in the network. The third is to add limited CSI feedback to the CDI scheme to reduce outage, perform better power control, and provide a more reliable network. In the first part of this work, we develop techniques to efficiently quantize channel covariance matrices in MIMO Rayleigh fading environments. While these covariance matrices change less frequently than the channel matrices themselves, this information needs to be updated when it does change. Furthermore, these covariance matrices have significantly more parameters to quantize than their channel matrix counterparts. Since many applications focus on utilizing the strongest eigenmodes of the channel covariance matrix and since these matrices tend to be low- rank, we focus on efficiently quantizing the dominant eigenvectors of these matrices. We develop Lloyd-type algorithms based on training data from the environment to develop our codebooks. We also develop an algorithm based on the reduced-parameter Kronecker and Weichselberger models to generate codebooks with reduced real-time codeword search. In the second part of this work, we examine single user and multiuser MIMO beamforming networks with CDI. Since CDI changes infrequently compared to CSI, algorithms based on CDI can achieve significant savings in feedback compared to algorithms based on CSI. With CDI, we can only guarantee quality of service for a specified outage probability in the network. Assuming correlated Rayleigh fading on all the links, we derive a closed-form expression for the outage probability. Then, using this expression, we derive algorithms for joint transmit/receive beamforming and power control to minimize the weighted sum power in the network while guaranteeing these outage probabilities. For both single-user and multiuser MIMO scenarios, we present optimal algorithms under the Kronecker model assumption, and we present near- optimal algorithms assuming general correlation structures on the links. We then show that using these algorithms based on CDI, if we are willing to accept given outages on the links, we can achieve comparable power usage in the network relative to algorithms based on CSI. We then extend this framework to the broadcast channel with multiple receivers and transmitters. In the third part of this work, we extend the previous outage framework based on CDI to utilize CSI of the direct links of the channels. The CDI framework considered here looks at minimum Signal- to-Interference-plus-Noise (SINR) requirements from each user that is met with a specified outage probability. This outage on the links is undesirable. However, in many standards, CSI is measured at the receiver via training data from its corresponding transmitter. The CSI on the interfering links is typically unavailable. If this CSI can be measured, this information can be fed back to the transmitters so they will not transmit when a given link is in outage. In addition, CSI feedback can be used for power reduction and thus potentially enable links that were previously in outage to transmit. Algorithms discussing these ideas are developed and discussed. Furthermore, an SINR quantizer is developed to minimize power usage, and combined CDI and CSI results are also discussed for the case of the broadcast channel with multiple transmitters and receivers