The vision of connected vehicles has stayed in conceptual development stage for years. A main hurdle towards its materialization is limited infrastructure support---the long latency and low throughput of current cellular infrastructure is insufficient to meet the projected application demands in connected vehicles. In this project, we will explore new network architectures and protocols to overcome the performance limitation of the cellular infrastructure for connected vehicles. The new system design will be validated on a small-cell testbed that we will build on the UCSD campus.

Multiple network standards, including 802.11p (DSRC) and LTE, already include specifications for vehicular wireless connectivity. However, the performance of state-of-the-art wireless networks still falls short of the requirements of demanding vehicular applications. For example, safety-critical connected-vehicle functions (e.g., lane assistance and forward collision warning) often require below 10 ms of latency; whereas context sensing applications (e.g., delivering bird’s eye view from road-side cameras to each vehicle) often require up to Gbps of throughput. In contrast, the current LTE wireless infrastructure still has end-to-end latency of 100 ms level, and throughput at Mbps level. Notably, the nominal link latency of LTE should be below 10 ms, and throughput on the order of 100 Mbps. The network-level performance attrition is mainly caused by the network management and interference management overhead.

This project will explore cooperative small-cells as a basic network architecture to realize high-rate, low-latency V2I networking and support demanding applications such as safety-critical driving assistance and real-time 3D camera-view sharing.

The cooperation among small-cell base-stations occurs at two levels: (i) PHY layer cooperation, which treats the small-cell base-stations as distributed MIMO nodes with carrier-level synchronization, and enables concurrent transmissions from/to base-stations through MIMO interference cancellation. (ii) Link layer cooperation, which uses the inter-base-station backhaul to coordinate the user association, handoff, and data synchronization among small-cell base-stations.

These cooperation mechanisms can build upon the coordinated multi-point transmission (CoMP) architecture defined in the LTE. Although CoMP has been explored extensively through analytical models and simulations, many practical issues remain open and its effectiveness is unclear. Our research will delve into the underexplored but critical aspects of CoMP, and tailor it for V2I networks.

Specifically, (i) We will explore a dynamic clustering mechanism that creates virtual base-stations through a cluster of small-cell base-stations. The small-cell base-stations within the same cluster maintains the same identity to the user, so that handoff becomes transparent. Each user is served by its own cluster, which changes dynamically as the user moves. (ii) We will install edge computing servers near each small-cell base-station, and explore the tradeoff between users' local computing versus offloading plus edge computing. The end goal is to optimize the application level latency for vehicle applications. (iii) We will extensively explore the "side-channel" information in future autonomous vehicles to facilitate the network functions. This includes, e.g., the GPS location information, and the 3D environmental information which can be used to facilitate channel prediction and network resource allocation.

A stretching goal of the project is to extend the small-cell architecture to millimeter-wave (mmWave) nodes. Due to high directionality of mmWave radios, operating the mmWave link, even between a single base-station and vehicle, becomes challenging. When multiple base-stations jointly serve mobile users, the problem becomes even more complicated, as user needs to determine not only which base-station to connect to, but also which beam to user among hundreds of directional beams. Our team has recently developed location-aware beam management schemes that simplifies the decision space using relative geometrical information. Although our work initially targeted scenarios with human mobility, we believe similar mechanisms can be extended to handle the V2I case.

Testbed construction and experimental validation will be one significant component in this research. We plan to work closely with industry partners to deploy a small-scale testbed. The testbed will comprise 2 to 4 sub-6GHz small-cell base-stations, deployed on poles or roofs, with either fiber or mmWave backhaul in between, and all the base-stations will be connected to a central EPC server for control and monitoring purposes. The base-stations will have distributed MIMO capabilities, either through carrier-level synchronization among themselves, or by distributing their RF front-ends as remote radio heads through front-haul links. In the second year of our project, we will also deploy 802.11ad-compatible 60 GHz mmWave small-cells. The base-stations will be 802.11ad outdoor access points from AirFide, and these base-stations will share the same backhaul and sites as the sub-6GHz ones. Due to hardware constraints, we expect these mmWave base-stations will only have link-level coordination to realize coordinated hand-off or beam steering, but will not have distributed MIMO capabilities.