MathWorks is advancing the use of artificial intelligence in radar system development through a range of workflows built around MATLAB and Simulink, enabling engineers to apply machine learning and deep learning techniques to radar signal processing, target classification, waveform analysis, and spectrum sensing applications. The company’s AI for Radar framework combines simulation, synthetic data generation, signal labeling, and neural network training to support modern radar systems used in aerospace, defense, automotive, and wireless communications.
The approach focuses on integrating AI into the radar development pipeline, particularly in areas where traditional signal processing methods face challenges from increasingly crowded RF environments, complex target behaviors, and large-scale sensor datasets. Using MATLAB and Simulink, engineers can generate synthetic radar waveforms and echoes, label radar data, train AI models, and deploy algorithms for radar target recognition and signal classification.
One of the core capabilities highlighted by MathWorks is the synthesis of radar signals for training machine learning and deep learning models. Engineers can simulate radar echoes from objects with varying radar cross-sections, including cylinders, cones, helicopters, pedestrians, bicyclists, and hand gestures. These synthetic datasets help train neural networks without requiring extensive real-world radar collection campaigns. The company said the workflows support applications such as waveform modulation classification, occupied spectrum sensing, and micro-Doppler signature analysis.
MathWorks also supports radar target classification workflows using both conventional machine learning and deep learning architectures. According to the company, machine learning implementations can use wavelet scattering feature extraction with support vector machines, while deep learning methods include transfer learning using SqueezeNet as well as Long Short-Term Memory (LSTM) recurrent neural networks. These approaches are designed to classify radar returns and identify targets based on radar signatures and environmental conditions.
Another area of development involves waveform and spectrum analysis using deep neural networks. MathWorks demonstrated workflows that combine time-frequency analysis methods such as the Wigner-Ville distribution with convolutional neural networks (CNNs) for radar and communications waveform classification. These capabilities are intended for cognitive radar systems, intelligent RF receivers, and software-defined radio applications where automatic signal recognition is increasingly important.
The company’s Radar Toolbox and Deep Learning Toolbox provide additional AI-enabled workflows for radar data processing, including clutter suppression, weather radar estimation, pedestrian and bicyclist classification, anomaly detection, and sensor fusion. Recent examples introduced by MathWorks include end-to-end “From ADC to AI” workflows that guide engineers through raw radar data processing, dataset creation, labeling, and neural network training.
MathWorks said its radar AI workflows are also intended to support system-level radar design and simulation. Engineers can model monostatic, bistatic, multifunction, airborne, automotive, shipborne, and space-based radar systems using realistic synthetic environments before deploying algorithms to DSPs or FPGA hardware. The tools additionally support phased-array radar simulation, RF component integration, and testing under dynamic operating conditions.
AI-based radar processing is increasingly being adopted in applications such as autonomous systems, electronic warfare, spectrum management, surveillance, gesture recognition, and advanced driver assistance systems (ADAS). MathWorks stated that integrating AI with radar enables improved object detection, motion analysis, target identification, and adaptive spectrum awareness in environments where traditional rule-based processing may be limited.
Click here to view the original article by Mathworks on “AI for Radar”.