Evaporative Angle: A Generative Approach to Mitigate Jamming Attacks in DOA Estimation

Current Direction of Arrival (DOA) estimation methods are unable to provide reliable estimates when faced with jamming attacks. To address this issue, we propose Direction of Arrival Estimation via Conditional Generative Adversarial Networks (DOA-CGAN), the first generative approach to remove the jamming component from the received signal covariance matrix. In our model, we input the received signal covariance matrix to an unsupervised generator that filters it to generate a matrix that can be deemed legitimate by a supervised discriminator. After training, we leverage the generator as a filter able to remove the jamming component from the received signal covariance matrix and feed its output to classical DOA estimation algorithms. Numerical results demonstrate that our proposed method delivers robust DOA estimation compared with other machine learning methods with a root mean squared error smaller than 0.2°.