Talks

I: Talk at the IEEE DTPI 2022

Title: An Introduction to Digital Twins for Meta-Surface Empowered Wireless Networks

To support the stringent requirements of emerging applications, such as augmented reality and brain-computer interface, wireless-based digital twin technology has emerged recently. Digital twinning creates a virtual representation of a wireless network at Edge devices. Edge-enhanced digital twining at wireless networks supports the creation of Intelligent Radios that facilitate real-time intelligent radio resource management. This talk aims to introduce the potential of reconfigurable intelligent surface (RIS) technology equipped with digital twining for ultra-reliable low-latency communication (URLLC). RIS technology has become a focal point of research due to its advantages to enhance the physical layer security, reliability, energy efficiency, and coverage by smart reconfiguration of the wireless environment that is enabled by the Digital Twin concept. The RIS panels have two-dimensional (2D) structures that are nearly passive and composed of metamaterial with macroscopic programmable physical characteristics. In this talk, we present a comprehensive review of the state-of-the-art RIS families, with emphasis on their operating principles from the viewpoints of both physics and communications, their passive beamformer design, and, ML-enabled RIS wireless networks. We will also discuss the compatibility of reinforcement learning (RL) with RISs to constantly control the dynamic nature of wireless propagation while reaching a seamless integrated framework for Communication, sensing, and computation. Last but not least, we will discuss the research opportunities of an amalgam of RISs with other emerging technologies such as edge computing and Quantum machine learning (QML) to satisfy the ambitious goals of WDT.

II: Talk at the 78th SWCS International Annual Conference

Talk 1: GPR System Design for Intelligent Root-Zone Soil Moisture Characterization and Optimal Farm Irrigation

Optimal farm irrigation is key to both conservation of water and soil resources. High-resolution soil moisture estimation is critical to optimal farm irrigation. This paper investigates high-performance ground penetrating radar (GPR) system design for intelligent root-zone soil-moisture characterization. Specifically, GPR transmitted waveform and signal processing at the receiver will be properly adjusted to facilitate high-resolution range profile (HRRP) estimation. This enables high-performance extraction of backscattered signal features received from different soil layers (with various moisture content). This signal feature information can be used for root-zone soil moisture extraction by applying machine learning (ML). Accurate high-resolution HRRP estimation needs high signal bandwidths, which contrasts with the higher Depth-of-Penetration (DOP) attainable via lower frequency bands. On the other hand, lower frequency bands only support narrower bandwidths. To address all the above concerns and achieve a properly received signal feature for ML via lower frequency bands, techniques of proper waveform designs capable of high-resolution HRRP estimation for different transmission schemes will be investigated. Examples of these transmission schemes include stepped frequency continuous wave (SFCW), Frequency Modulated Continuous Wave (FMCW), and Orthogonal Frequency Division Multiplexing (OFDM). Accordingly, appropriate signal processing schemes corresponding to each transmission mode are applied to integrate the signals received across all frequency bands. An inverse problem for end-to-end (starting from the transmission to reception) design will be offered by which we can jointly attain proper waveform, soil moisture imaging, and high-resolution HRRP. 

Talk 2: Enhanced Soil Moisture Estimation via Intelligent Full Waveform Inversion of GPR Data

Root Zone soil moisture characterization is key to optimum irrigation, healthy soil mineral content, and water conservation. Ground Penetrating Radar (GPR) has been widely used to estimate soil moisture via Full Waveform Inversion (FWI). The conventional FWI method, however, is susceptible to significant challenges that can cause large result errors, in particular, for complex models such as soil moisture. These challenges include large computational costs or large computational times, the dependency of the method on the starting model accuracy, and data challenges, such as missing low-frequency data and nonlinearity that can render local minima traps and cycle skipping.  Intelligent FWI methods using machine learning have been developed to address these challenges.  Common machine learning methods use deep learning, deep neural networks, and convolutional neural networks with generative adversarial algorithms.  GPRNet is an existing algorithm with an open source that is analogous to FWI and uses a five-layer convolutional neural network. However, GPRNet has been trained via simple models, not suitable for simulating radar data for complex soil moisture models. We use gprMax another open-source code to simulate datasets based on more complex models such as soil moisture to train the GPRNet.  Then, we enhance our intelligent FWI method by including the convolutional neural network to increase the accuracy and performance of the new and increasingly complex datasets. We achieve further enhancements by including different filtering methods, changing activation functions, manipulating the number of hidden layers, and altering other parameters and regularization in our algorithm.