UC Berkeley

We address the interesting question of how multiple dynamical systems should track each other in real-time over a shared channel. This dissertation covers a wide spectrum of results on this topic: our theories, engineering designs, computer simulations, prototype implementation and real-world evaluations. This research is motivated by the V2V (vehicle-to-vehicle) communications for active roadway safety, which is an important application in Intelligent Transportation Systems.

Our theoretical work is presented in Chapter 2. Two formulations of real-time tracking are analyzed. In the first formulation, we assume a scalar continuous-time continuous-state source and an AWGN (additive white Gaussian noise) channel without feedback. The analysis shows the MMSE (minimum mean-squared error) optimality of the innovation encoder and its tracking performance. In the second formulation, we assume a discrete-time continuous-state LTI (linear time-invariant) source and a G/G/1-infinity queueing network to deliver the information flow of the source from the encoder to the decoder. We derived a necessary and sufficient condition for real-time tracking stability in the form of a bound on the entropy rate of the LTI source. This bound is a function of the queueing server capability and other performance parameters. In both formulations, we extend the analysis to a multiple-access channel to understand the basic principle of designing encoders for real-time tracking multiple dynamical systems over a shared channel.

Our engineering designs for multiple dynamical systems to track each other over a shared channel are presented in Chapter 3. To model packet-switched networks, we assume that the source process state, e.g., a real scalar or a real vector, can be delivered in a message with negligible distortion. This assumption is valid since practical applications usually do not require infinite precision of the source state. In the analysis, we further assume a slotted ALOHA channel for tracking multiple scalar LTI systems. First, non-adaptive channel access schemes are analyzed for tracking stability and MSE (mean squared-error) performance. Adaptive transmission control schemes are then proposed for each node to adjust its message rate based on the tracking error of its own state at other nodes. With preliminary Matlab simulations, a comparison is made and a transmission rate control is proposed for vehicles to broadcast state information and track each other in real-time. This rate control has an on-demand nature and considers both tracking error and channel congestion. A microscopic traffic simulator and a network simulator are used to show that tracking performance of the proposed design is robust to variations in different traffic conditions.

We start Chapter 4 by stating our understanding of the V2V safety communications problem and our design approach. The error-dependent message generation control from Chapter 3 is then enhanced with a transmission power allocation for each out-going safety message. This power control uses sensed channel utilization as side-information used by each vehicle to infer the condition of the shared channel. Our proposed rate/power control responds to channel congestion by maintaining the same information intensity, i.e., the same message rate, to as many neighboring vehicles as possible while temporarily stopping communication to farther vehicles by reducing transmission power. The robustness of the proposed design is verified by large-scale network simulations. Its real-time tracking accuracy is shown to outperform the currently proposed 100-millisecond beaconing of messages with 20-dBm transmission power.

In Chapter 5, we present implementation and real-world evaluations of the V2V transmission rate and power control proposed in Chapter 4. The implementation details are described as message generation and power assignment functional blocks. We also summarize results from outdoor vehicle mobility and scalability evaluations conducted at the General Motors Technical Center. In addition, updated performance measures and simulation results are provided for challenging highway and intersection scenarios. Overall, the prototype evaluations and simulation results show the superior tracking performance of our design over the currently proposed 100-millisecond beaconing with 20-dBm power. Our design has been adopted by General Motors R&D and serves as a candidate for a national standard defining how vehicles should track each other for active roadway safety.