Feedback and Interference Alignment in Networks
- Author(s): Suh, Changho
- Advisor(s): Tse, David
- et al.
The increasing complexity of communication networks in size and density provides us enormous opportunities to exploit interaction among multiple nodes, thus enabling higher date rate of data streams. On the flip side, however, this complexity comes with challenges in managing interference that multiple source-destination pairs in the network may cause to each other. In this dissertation, we make progress on how we exploit the opportunities, as well as how we overcome the challenges.
In the first part, we find that feedback - one of the common ways to enable interaction in networks - has a promising role in improving the capacity performance of networks. Earlier results on feedback capacity were somewhat discouraging. This is mainly due to Shannon's original result on feedback capacity where he showed that in point-to-point communication, feedback does not increase capacity. Hence, traditionally it is believed that feedback has had little impact on increasing capacity of communication links. Therefore, the use of feedback has been limited to improving the reliability of communications, usually in the form of ARQ. In this dissertation, we show that in stark contrast to the point-to-point case, feedback can improve the capacity of interference-limited network. In fact, the improvement can be unbounded. This result shows that feedback can have a potentially significant role to play in mitigating interference. Also in the process of deriving this conclusion, we characterize the feedback capacity of the two-user Gaussian interference channel to within 2 bits, one of the longstanding open problems in network information theory.
In the second part, we propose a new interference management technique for widely deployed cellular networks. Inspired by a recent breakthrough, the concept of interference alignment, we develop an interference alignment technique for cellular networks. Our technique promises almost interference-free communication with the increase of the number of clients in cellular networks. It shows substantial gain (around 30% to 60%) as compared to one of the interference management techniques in current cellular systems. In addition, it comes with implementation benefits: it can actually be implemented with small changes to emerging 4G cellular standards and architectures at the base-stations and clients. In particular, the required signal-processing circuitry, software control, and channel-state feedback mechanisms are extensions of existing implementations and standards.
Lastly, we extend the interference alignment principle, developed in the context of wireless networks, into other fields of network research such as storage networks. In an effort to protect information against node failures, storage networks employ coding techniques, such as maximum distance separable (MDS) erasure codes, known as optimal codes in reliability with respect to redundancy. However, these MDS codes come with prohibitive maintenance cost when it comes to repairing failed storage nodes. While only partial information stored in the failed node needs to be recovered, the conventional MDS codes focus on the complete data recovery (including unwanted data, corresponding to interference) by downloading too much information from survivor storage encoded nodes, thus causing the high repair cost. Building on the connection between wireless and wireline networks, we leverage the interference alignment principle to develop a new class of MDS codes that significantly reduces the repair cost over the conventional MDS codes and also achieves information-theoretic optimal bound on the repair cost for all admissible code parameters.