Coordinated MultiPoint Transmission with Incomplete Information
Doctoral thesis, 2015
The demand for higher data rates and efficient use of various resources has been an unquenchable thirst across different generations of cellular systems, and it continues to be so. Aggressive reuse of frequency resources in cellular systems gives rise to intercell interference which severely affects the data rate of users at the cell-edge. In this regard, coordinated multipoint (CoMP) is one of the ways to mitigate interference for these cell-edge users. In the downlink, joint transmission (JT) CoMP involves the cooperation of two or more geographically separated base stations to jointly transmit to these users by treating the interfering signal as useful signal.
To realize the gains of JT-CoMP in a frequency division duplex system, the users need to feedback the channel state information (CSI) to its serving base station. This needs to be aggregated at the central coordination node for mitigating interference via precoding. However, the process of aggregation poses tremendous burden on the backhaul. One of the ways to reduce this burden is to use relative thresholding, where the users feed back the CSI of only those links that fall within a threshold relative to the strongest base station. The side effect of thresholding results in limited or incomplete CSI for precoding. Efficient backhauling is achieved when the quantity of CSI available for certain links at the central coordination node be correspondingly equivalent to the quantity of precoding weights generated for the same links. The incomplete CSI poses problems for the simple zero-forcing precoder to mitigate interference and also achieve efficient backhauling.
In this thesis, the main problem of simultaneously mitigating interference and achieving efficient backhauling is addressed with a layered approach. Our physical (PHY) layer precoding approach solves the problem and allowes the medium access control (MAC) layer scheduler to be simple. The PHY layer precoding algorithms such as successive second order cone programming are proposed using convex optimization in [Paper A], and particle swarm optimization based on stochastic optimization is proposed in [Paper B]. Also, we exploit the use of long term channel statistics for the incomplete CSI and characterize the promising performance of the proposed precoder using numerical bounds. Based on our results, we observed that the swarm algorithm struggles with the increase in the problem size. The MAC layer approach exploits scheduling to solve the problem keeping a simple PHY layer zero-forcing precoder [Paper C]. Our proposed constrained scheduling approach provides the best tradeoff in terms of average sum rate per backhaul use compared to other MAC layer techniques. These results can be applied to a variant of the baseband hotel, a centralized architecture.
In a distributed architecture, the CSI is exchanged periodically between the base stations over the backhaul for JT-CoMP. Any CSI feedback update from the user must be immediately exchanged over the backhaul to preserve the gains of JT-CoMP. We propose an improved decentralized local precoder design where the base station with new local CSI can design the local precoding weights in between the CSI exchange between base stations [Paper D]. With our approach some of the gains of JT-CoMP can still be preserved without the need to burden the backhaul.
particle swarm optimization