Towards Scalable Design of Future Wireless Networks

Wireless operators face an ever-growing challenge to meet the throughput and processing requirements of billions of devices that are getting connected. In current wireless networks, such as LTE and WiFi, these requirements are addressed by provisioning more resources: spectrum, transmitters, and baseband processors. However, this simple add-on approach to scale system performance is expensive and often results in resource underutilization. What then, are the ways to efficiently scale the throughput and operational efficiency of these wireless networks? To answer this, we explore several potential designs: utilizing unlicensed spectrum to augment the bandwidth of a licensed network; coordinating transmitters to increase system throughput; and finally, centralizing wireless processing to reduce computing costs.

First, we propose a solution that allows LTE, a licensed wireless standard, to co-exist with WiFi in the unlicensed spectrum. The proposed solution bridges the incompatibility between the fixed access of LTE, and the random access of WiFi, through channel reservation. It achieves a fair LTE-WiFi co-existence despite the transmission gaps and unequal frame durations. Second, we consider a system where different MIMO transmitters coordinate to transmit data of multiple users. We present an adaptive design of the channel feedback protocol that mitigates interference resulting from the imperfect channel information. Finally, we consider a Cloud-RAN architecture where a datacenter or a cloud resource processes wireless frames. We introduce a Tree-based design for real-time transport of baseband samples, and a processing framework that combines scheduling with fine-grained parallelism. The framework reduces processing times by migrating parallelizable tasks to idle compute resources, and thus, decreases the processing deadline-misses at no additional cost.

We implement and evaluate the above solutions using software-radio platforms and off-the-shelf radios, and confirm their applicability in real-world settings.