Age of Information Optimization for Timeliness in Communication Networks

With the emergence of technologies such as autonomous vehicular systems, holographic communications, remote surgery and high frequency automated trading, timeliness of information has become more important than ever. Most traditional performance metrics, such as delay or throughput, are not sufficient to measure timeliness. For that, age of information (AoI) has been introduced recently as a new performance metric to quantify the timeliness in communication networks. In this dissertation, we consider timely update delivery problems in communication networks under various system settings. First, we introduce the concept of soft updates, where different from the existing literature, here, updates are soft and begin reducing the age immediately but drop it gradually over time. Our setting models human interactions where updates are soft, and also social media interactions where an update consists of viewing and digesting many small pieces of information posted, that are of varying importance, relevance and interest to the receiver. For given total system duration, the number of updates, and the total allowed update duration, we find the optimum start times of the soft updates and their optimum durations to minimize the overall age. Then, we consider an information updating system where not only the timeliness but also the quality of the updates is important. Here, we use distortion as a proxy for quality, and model distortion as a decreasing function of processing time spent while generating the updates. Processing longer at the transmitter results in a better quality (lower distortion) update, but it causes the update to age in the process. We determine age-optimal policies by characterizing the update request times at the receiver and the update processing times at the transmitter subject to constant or age-dependent distortion constraints on each update. Next, different from most of the existing literature on AoI where the transmission times are based on a given distribution, by assigning codeword lengths for each status update, we design transmission times through source coding schemes. In order to further improve timeliness, we propose selective encoding schemes where only the most probable updates are transmitted. For the remaining least probable updates, we consider schemes where these updates are never sent, randomly sent, or sent by an empty symbol. For all these encoding schemes, we determine the optimal number of encoded updates and their corresponding age-optimal real-valued codeword lengths to minimize the average age at the receiver. Then, we study the concept of generating partial updates which carry less information compared to the original updates, but their transmission times are shorter. Our aim is to find the age-optimal partial update generation process and the corresponding age-optimal real-valued codeword lengths for the partial updates while maintaining a desired level of fidelity between the original and partial updates. Next, we consider information freshness in a cache updating system consisting of a source, cache(s) and a user. Here, the user may receive an outdated file depending on the freshness status of the file at the cache. We characterize the binary freshness metric at the end user and propose an alternating maximization based method to optimize the overall freshness at the end user subject to total update rate constraints at the cache(s) and the user. Then, we study a caching system with a limited storage capacity for the cache. Here, the user either gets the files from the cache, but the received files can be sometimes outdated, or gets fresh files directly from the source at the expense of additional transmission times which inherently decrease the freshness. We show that when the total update rate and the storage capacity at the cache are limited, it is optimal to get the frequently changing files and files with relatively small transmission times directly from the source, and store the remaining files at the cache. Next, we focus on information freshness in structured gossip networks where in addition to the updates obtained from the source, the end nodes share their local versions of the updates via gossiping to further improve freshness. By using a stochastic hybrid systems (SHS) approach, we determine the information freshness in arbitrarily connected gossip networks. When the number of nodes gets large, we find the scaling of information freshness in disconnected, ring and fully connected network topologies. Further, we consider clustered gossip networks where multiple clusters of structured gossip networks are connected to the source through cluster heads, and find the optimal cluster sizes numerically. Then, we consider the problem of timely tracking of multiple counting random processes via exponential (Poisson) inter-sampling times, subject to a total sampling rate constraint. A specific example is how a citation index such as Google Scholar should update citation counts of individual researchers to keep the entire citation index as up-to-date as possible. We model citation arrival profile of each researcher as a counting process with a different mean, and consider the long-term average difference between the actual citation numbers and the citation numbers according to the latest updates as a measure of timeliness. We show that, in order to minimize this difference metric, Google Scholar should allocate its total update capacity to researchers proportional to the square roots of their mean citation arrival rates. Next, we consider the problem of timely tracking of multiple binary random processes via sampling rate limited Poisson sampling. As a specific example, we consider the problem of timely tracking of infection status (e.g., covid-19) of individuals in a population. Here, a health care provider wants to detect infected and recovered people as quickly as possible. We measure the timeliness of the tracking process as the long term average difference between the actual infection status of people and their real-time estimate at the health care provider which is based on the most recent test results. For given infection and recovery rates of individuals, we find the exponentially applied testing rates for individuals to minimize this difference. We observe that when the total test rate is limited, instead of applying tests to everyone, only a portion of the population should be tested. Finally, we consider a communication system with multiple information sources that generate binary status updates, which in practical application may indicate an anomaly (e.g., fire) or infection status (e.g., covid-19). Each node exhibits an anomaly or infection with probability $p$. In order to send the updates generated by these sources as timely as possible, we propose a group updating method inspired by group testing, but with the goal of minimizing the overall average age, as opposed to the average number of tests (updates). We show that when the probability $p$ is small, group updating method achieves lower average age than the sequential updating methods