This interdisciplinary research project addresses active-passive radio frequency (RF) spectrum coexistence with very stringent interference thresholds for passive users. The research challenges are tackled across multiple layers of the protocol stack (physical, medium access control, and application). The Mississippi State University team developed a one-of-a-kind testbed for collecting remote sensing datasets in conjunction with in-band 5G communication transmission with ground truth in an isolated RF environment. Using this testbed, we collected datasets with a L-band radiometer at 1.4 MHz with a bandwidth of 28 MHz, assimilating NASA's satellite-borne passive radiometer system SMAP, for different types of RF interference (RFI) from 5G systems. This data set, shared with the research community through common software and data repositories along with our testing methodology, software, and hardware tools, which are accessible through the project Website https://sites.google.com/view/swift-ai-spectrum, enables research, training, optimization, and benchmarking of spectrum coexistence solutions. The radiometer design, 5G transmission schemes, testbed configurations, and the findings from our experiments are reported in peer-reviewed research papers. The data collection has enabled innovative solutions for characterizing, minimizing, and removing RFI and has, thus far, resulted in two PhD dissertations and one MS Thesis at Mississippi State University:

• Walaa H. Alqwider, "Deep reinforcement learning for advanced wireless networks enabling service and spectrum coexistence," PhD Dissertation, Mississippi State University, May 2024. This dissertation develops machine learning methods to optimize 5G transmission parameter selection for minimizing RFI in active-passive spectrum coexistence scenarios. Leveraging data collected from measurements on the active-passive RF coexistence research testbed at Mississippi State University, the research demonstrates how radiometer feedback can guide dynamic adjustments in transmission parameters to achieve spectrum coexistence without compromising sensing applications. The results show that  even low-power and sparse 5G transmissions can induce significant RFI, resulting in measurable biases in temperature readings captured by the radiometer. However, the study also shows that by intelligently adjusting transmission parameters, the residual bias can be mitigated, preserving the integrity of passive RF sensing data. Further research is identified to further reduce the residual interference as opposed to discarding the RF contaminated samples.

• Md Mehedi Farhad, "Estimating surface reflectivity with smartphone and semi-custom GNSS receivers on UAS-based GNSS-R technology and surface brightness temperature using UAS-based L-band microwave radiometer," PhD Dissertation, Mississippi State University, May 2024. This dissertation explores innovative techniques for remote sensing of surface soil moisture (SM) using unmanned aircraft systems (UASs), focusing primarily on a dual-polarized L-band microwave radiometer. The radiometer measures surface brightness temperature to estimate emissivity and SM, with real-time onboard processing of in-phase and quadrature signals to detect and mitigate RFI. This reduces post-processing time and enhances data reliability. Additionally, the study investigates GNSS reflectometry (GNSS-R) using cost-effective U-blox GNSS receivers mounted on UAS platforms. These receivers estimate surface reflectivity by capturing reflected GNSS signals, providing high spatio-temporal resolution SM data. A complementary approach utilizing smartphone-based GNSS receivers further demonstrates the potential of accessible and affordable remote sensing technologies. Together, these methods highlight advancements in UAS-based L-band radiometry and GNSS-R, offering scalable, efficient solutions for precision agriculture and environmental monitoring.

• Ahmed M. Alam, "Deep learning-based radio frequency interference detection and mitigation for microwave radiometers using 2-D spectral features," MS Thesis, Mississippi State University, August 2024. This thesis addresses the critical challenge posed by RFI to the effectiveness of passive microwave radiometry in climate studies and Earth science. Despite operating in protected frequency bands, radiometers are increasingly affected by RFI from sources such as air surveillance radars and unmanned aerial vehicles. Traditional RFI detection methods, which rely on handcrafted algorithms designed for specific interference types, often lack adaptability and generalization. To overcome these limitations, this thesis proposes a deep learning framework that employs convolutional neural networks to detect diverse RFI types on a global scale. By learning directly from radiometric data, including raw moment data and Stokes parameters dynamically labeled using quality flags, this approach enhances detection accuracy, efficiency, and scalability. The results highlight the potential of this data-driven solution to significantly improve RFI mitigation in passive remote sensing applications.

Last Modified: 01/28/2025
Modified by: Vuk Marojevic