Deep Complex-valued Convolutional Neural Networks (DCCNN) for Joint Spectrum Sensing and Channel Estimation
Abstract
Publicly Releasable Topic #: CR-01 Technical POC: Bryan Woosley Increasing utilization of spectrum by friendly and adversary parties leads to a dynamic spectrum environment that is nontrivial to model. Furthermore, the impacts of the spatial environments are site-specific and highly complex to capture through conventional modeling tools. While applications of Deep Learning in the physical layer of wireless communication have shown remarkable performance, it is unclear if these solutions can capture complex real-world scenarios by learning the dynamic spectrum and spatial environment. This proposal focuses on reconfigurable spectrum monitoring with a new type of complex-valued neural networks that are shown to capture complex wireless signal representations and learn transformations in wireless signals without explicit modules. It is well known that spectrum detection cannot be decoupled from channel estimation, and treating each subproblem separately leads to suboptimal performance. The goal of this project is to transform deep complexvalued convolutional neural networks (DCCNN) from their infancy to practical operation in the wild for joint spectrum sensing and channel estimation. The proposed design will be expanded for operation in the wild such that unique features of the wireless channel can be captured. To this end, the proposal focuses on three main research thrusts: DCCNN for dynamic spatial environments, DCCNN for dynamic spectrum environments, and reconfigurable distributed sensor management. The proposed solutions will be evaluated in a city-scale gigabit wireless network testbed developed on the UNL campus and the city of Lincoln, Nebraska. To enable persistent electromagnetic spectrum awareness, this project focuses on DCCNN-based joint spectrum sensing and channel estimation algorithms. This is the first step towards predicting and characterizing adversary spectrum utilization for exploitation and EMS maneuver space. The project results will support communication activities for spectrum-denied areas and dynamic spectrum sharing for a large group of terminals.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 12, 2025
- Source ID
- N001742310007
Entities
People
- Mehmet Can Vuran
Organizations
- United States Navy
- University of Nebraska–Lincoln