Wireless communication is the preferred and practical mode of communication in a wide range of situations. In a typical wireless transmission, there is a transmitter that transmits the signal, and a receiver that receives the signal. Safety-critical operation, high-throughput, and low-latency are very important in current and future wireless systems. The objective of radiowave propagation modeling is to establish the correlation between the signal at transmission & the reception or, in other words, to determine characteristics of the transmission channel.
What is the biggest limitation of the existing modeling methods?
The dichotomy between computational efficiency and accuracy of the propagation models. It means that when we try to improve on one parameter (either computational efficiency OR accuracy), the other parameter invariably takes a hit. How do we overcome this challenge?
With Machine Learning-Driven Modeling!
What is Machine Learning-Driven Modeling?
Let’s assume an input x to the ML model is mapped to output y. The goal of the ML model is to learn an unknown function f that accurately correlates x to y in all situations.
The research paper by Aristeidis Seretis, Costas D. Sarris discusses various ML-based radio wave propagation modeling techniques, gives an overview of various relevant research papers & also discusses the limitations of the modeling techniques. It also goes further and classifies various models based on their approach to each of these limitations. Here, scientists have established the three main building blocks of any ML radio propagation model: The Input, the ML model itself, and the output.
Various propagation models were analyzed in this research paper based on their Input, the ML Model & the Output. In the words of the authors, the following conclusions substantiate the advantage of ML-driven modeling techniques against existing methods:
The authors of this research say that the near-future advances in the field of machine learning will make it possible to reduce the required amount of training data and time required to complete modeling even further, thus essentially making the model input data simpler, while also improving accuracy. Reinforcement learning and application of GANs for electromagnetic wave propagation modeling also looks very promising.
Research Paper: Aristeidis Seretis, Costas D. Sarris “An Overview of Machine Learning Techniques for Radiowave Propagation Modeling“