Title:	PRESTO: Predictive Service Quality Management for Transport Services
DNr:	Berzelius-2022-195
Project Type:	LiU Berzelius
Principal Investigator:	Oscar Stenhammar <ostenh@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2022-09-23 – 2023-04-01
Classification:	20105
Homepage:	https://www.kth.se/profile/ostenh
Keywords:

Abstract

Recent advances in wireless communications, real-time control, sensing, positioning, collaborative spectrum management and artificial intelligence are enabling the transport sector to become more cost-efficient, secure, and sustainable. Due to new requirements arising in road, railway, air and maritime transport, reliable wireless communications between vehicles, road infrastructure and road users are no longer a "nice to have", but are integral parts of cooperative intelligent transportation systems and smart cities. To deploy trustworthy mobile networks, which are capable of delivering both mobile broadband and mission critical services to the transport sector, network operators must deal with the problem that the performance of wireless networks vary in time and space. Conceptually, these performance variations can be addressed by improving service reliability and coverage and/or by enabling applications to foresee the dynamics of network performance and changes in the service quality. While techniques that enable the first approach (improving service reliability and coverage) are well-known, predicting the temporal and spatial variations of service performance metrics poses largely hard-to-answer and unsolved questions. Service performance metrics that are of high interest to mission critical applications include data throughput between users and the network as well as between users, and latency and service fulfillment probability. Indeed, if applications are provided with reliable predictions on these performance metrics, they can react proactively to changes in space and time. The idea of the PRESTO project is therefore to provide predictions of spatio-temporal network capacity, coverage and quality-of-service along roads/streets. To realize such predictions, PRESTO will build on recent advances in the sensing capabilities of vehicles and the capacity increase of wireless networks, that enable the collection of large amount of real-time measurement data. Specifically, the basic idea of PRESTO is to investigate, use and extend machine learning (ML) techniques to predict throughput and quality of service such that the following requirements are met: 1) The predictions have to be done at locations and times at which only noisy or unbalanced (partial) measurements are available. This is because in wireless vehicular data sets missing data commonly exist, since vehicular wireless data suffer from a number of impairments such as highly varying antenna placements and propagation effects. 2) The predictions have to be done on the basis of distributed data sets that are connected over public communication networks. The ML services must be provided by distributed algorithms, where the coordination and computation procedures of ML algorithms are constrained by bandwidth limitations, varying latency and message losses of public networks. 3) The predictions have to be done and possibly revised in real-time or quasi real-time typically in the order of hundreds of milliseconds, seconds or tens of seconds. This is because the performance of the network may be affected by sudden changes in the propagation environment, network load, software or hardware impairments etc. When a sudden change occurs, applications benefit from the quick updates of predicted values.