WiFiUS: Collaborative Research: Ultra-low latency and High Reliability for Wireless IoT

The goal of this project is to develop new architectures and control algorithms for wireless networks to support ultra low latency and high reliability IoT applications. We next give a high-level model for latency and reliability that will be used to guide our subsequent discussions and then give a summary of our proposed research. We consider a wireless network consisting a mix of different technologies (cellular, Wifi, etc.) and devices (handsets, IoT devices, etc). Though our focus is on supporting IoT applications with high reliability and low latency requirements, we will also consider other forms of traffic that must co-exist with this. For each such IoT application, we assume that it has to accomplish certain tasks triggered by the occurrence of external events. These tasks need to be accomplished within a given latency of �i seconds and with a specified reliability of qi, which gives the probability that the latency requirement is met, i.e., we want to ensure that the actual time to complete this task, W, satisfies Pr(W � �i) = qi: (1) The basic example we will consider is where an IoT task consists of sending a single message between two IoT devices or between one device and infrastructure, for example conveying a sensor reading to a monitoring center. Other more general examples might include sending a sensor reading from one of k possible sensors, sending multiple messages over time to, e.g., to track the state of an evolving process within the given delay constraint and with the given reliability, or interactively exchanging messages between two nodes (e.g., request/replies). In most of this proposal, we focus on the basic case but plan to also consider these extensions in our work. The latency of conveying this message to the destination then depends on if this message is sent in one or more packets, how these packets are mapped into physical layer transmissions, the number of hops the message needs to take to reach the destination, how many times it is re-transmitted, and any scheduling or queueing delays that are incurred. The reliability depends on losses due to channel noise and interference, failure to schedule a packet within the given deadline, and loss of a node (e.g., due to it running out of power or other forms of failure). Our proposed effort will address all of these factors so as to best exploit the interdependencies among them. To accomplish this and manage the overall complexity, we structure our proposed research according the the following natural time-scales at which key issues emerge: � The shortest time-scale we consider is the packet time-scale. Issues here include how messages are divided up into packets, what is the length of the basic time-slot chosen to send such a packet as well as other physical layer parameters (e.g. the form of error control coding, the bandwidth used, etc.). At the packet level, these impact the latency (via the time to send a packet) and the reliability via the probability that a packet is lost due to channel noise. For example, using longer time-slots can enable the use of more powerful codes increasing reliability but at the cost of increased latency. � The next time-scale we consider is the MAC time-scale. The main focus here is on scheduling traffic across different flows so as to meet the latency/reliability constraints of each IoT flow, while also serving any other traffic flows that may be present. This scheduler will account for the latency, reliability offered by the design choices at the packet time-scale. Here, our ma