Post Doc - Distributed Learning Estimation Techniques for new 6th Generation Wireless Communication Technologies
Postdoc open position
Distributed Learning Estimation Techniques for new 6th Generation Wireless Communication Technologies
Funding project: “6th Generation Wireless Communications Networks: New Concepts, Algorithms and Applications” (6th. Geração: Novos Conceitos, Algoritmos e Aplicações), FAPESP / MCTI / CGI – 2022
Supervisor: Prof. Vitor H. Nascimento
Institution: University of São Paulo, Brazil
Contract duration and salary: up to 3 years with a fellowship of approximately BRL 12,570.00 per month (tax free) and funds for travel and research expenses. The start date can be as early as June 2026.
Pre-requisites: Candidates must hold a PhD degree in electrical engineering (with a focus on signal processing) or applied mathematics. Having a strong mathematical background and/or computer literacy skills (Python, C++, Matlab, Julia etc.) is a definite plus.
How to apply: Candidates should contact the supervisor via email (vitnasci@usp.br) providing a short motivation and a detailed academic-style CV including the list of publications. Applications will be received until the position is suitably filled.
Research summary
For 6th Generation Wireless Communication (6G), networks will have to intensively use multiple antenna systems, such as massive MIMO, to increase significantly their information rate to meet the 6G performance requirements. Therefore, new technologies, such as Reconfigurable Intelligent Surfaces (RISs) (also known as intelligent reflecting surfaces), are envisaged as a promising solution with low cost and complexity. For both conventional and RIS-assisted MIMO scenarios, the performance of the system architecture strongly relies on the accuracy of the instantaneous channel state information. Thus, the channel estimation accuracy plays an important role in the design of new systems. In the traditional centralized learning (CL) schemes, the channel model parameters are estimated in the base station (BS) with data collected by the users, and then the model parameters are sent back to the users, which can perform channel state information. This approach, however, involves huge communications overhead. In order to deal with the overhead of the CL scheme, distributed learning techniques can be employed, where parameter estimation is made locally at the users, and shared among them. This approach can be partially distributed, such that the BS still participates as the main node responsible for the estimation, or fully distributed, where all the learning occurs at the user nodes. Therefore, distributed estimation techniques, such as distributed least mean squares (LMS), recursive least squares (RLS) and Kalman filters can play an important role in presenting cost-effective solutions to this new setting. Also, distributed machine learning techniques, also known as Federated Learning, are strong candidates.
Objectives
This project will require a researcher dedicated to the topic of channel estimation using robust and low-cost algorithms. The researcher must develop cooperation techniques between algorithms, determine how to efficiently distribute different algorithms to be used at each node of the network, in order to create diversity and thus achieve robustness. The researcher should also develop performance and robustness measures, and compare the results with those in the literature. Furthermore, the researcher must develop mathematical models, new concepts, tools to theoretically analyze the performance of algorithms, and create simulations to verify the predictions. The results will be published in conference proceedings and international journals.
Workplan
The postdoctoral fellow must be included in the project to contribute to the items “i) Innovative techniques for signal processing” in section “d.4) Scientific and technological challenges and the means and methods to overcome them.” Thus, the algorithms to be investigated in the first and second year include the development of distributed algorithms for channel estimation, considering both ideal versions (without data quantization) and practical versions (considering data quantization and limiting the information exchanged between the network nodes). The problem of limiting the necessary a priori information must also be considered, so that the network can operate with adequate performance even in the presence of strong uncertainties in the parameter evolution models.
During the third year the fellow must work mainly on establishing the robustness of the proposed methods, in particular comparing the results obtained with those available in the literature.