2026 ASEE Annual Conference & Exposition

The rapid advancement of machine learning (ML) has created new opportunities within engineering disciplines; however, integrating practical ML techniques into Radio Frequency (RF) curricula remains inadequately addressed. Traditional RF engineering programs build upon a series of signal processing, circuits, and communications theory courses, often reaching core RF implementation several years into the program with little or no exposure to modern ML topics. While RF education tracks are highly specialized, students are increasingly entering the workforce where they encounter projects requiring ML, or even Artificial Intelligence (AI), integration into existing radio systems, yet most lack hands-on experience bridging these domains. This gap in instruction propagates challenges for future wireless engineers, as ML and AI knowledge can significantly boost their career opportunities. This work presents a series of laboratory-based exercises that bridge this divide through progressive, hands-on RF/ML experimentation building up to neural network automatic modulation classification (AMC). We provide eight practical labs that can be integrated into existing RF courses, offering educators options to incorporate ML techniques into their existing curricula.

The labs presented here are designed to reinforce foundational concepts of both RF and ML before attempting their integration. These exercises address unique challenges posed by applying ML to wireless signals, such as the deviation between synthetic and captured signals that often invalidate lab-only trained models when applying them to over-the-air (OTA) signals.

The lab sequence begins with foundations of RF, including spectrum visualization, where students generate and visualize synthetic signals. Students then create Software Defined Radio (SDR) procedures to transmit and receive signals using GNU Radio, understanding how OTA RF signals propagate. This is followed by spectrum analysis and signal classification where signals are identified via feature extraction as a basis for what will be automated with machine learning. Progressive modules introduce neural network development where students will implement a 1D Convolutional Neural Network (CNN) on synthetic datasets, advancing to captured signals mirroring real RF propagation challenges. The labs progress to use TorchSig, an open-source signal processing ML toolkit, to introduce sophisticated signal generation capabilities, including realistic channel impairments and signal-to-noise ratio (SNR) values. Students implement and compare multiple architectures, from basic CNNs to advanced Cross-Covariance Image Transformer (XCiT) models, evaluating their performance under varying RF conditions. The final lab culminates with the real-time use of trained models for live signal classification, demonstrating practical applications of ML in RF systems.

The presented sequence of labs follow a structured progression that reinforces prerequisite knowledge of core RF education where each lab builds upon concepts from previous exercises. Educators can reorder or skip any labs where students have appropriate prerequisite knowledge. Additionally, each lab includes configurable parameters for adjusting complexity levels based on course requirements. These labs can be adopted in undergraduate or graduate programs with clear pathways for extending content based on learning outcomes.

Our contributions provide educators with reproducible labs for integrating ML and RF education, which prepares students for emerging challenges in next generation wireless systems where dynamic signal processing will be fundamental to system optimization.