A Bayesian-Genetic Hybrid Method for Frequency Diverse Array Transmit Beampattern Optimization

This paper proposes a new method to find the optimum frequency offsets in a Frequency Diverse Array (FDA) radar transmit beampattern that minimizes the Peak Side Lobe Level (PSLL) and Half Power Beamwidth (HPBW). This method uses a modified Bayesian Optimization (BO) framework where Bayesian Neural Network (BNN) is used for surrogate modelling of FDA radar transmit beampattern objective function. By resampling the weights and biases of BNN, we get a distribution of predictions whose mean and standard deviation are used to compute the Expected Improvement (EI). A population of frequency offsets where the EI is higher is used to initialize Genetic Algorithm (GA). Unlike the traditional method of random initialization, this method guides GA, which in turn, searches and updates the BNN with function evaluations. Hence, each sub-optimal GA run trains the BNN, and this cycle is repeated until convergence. Simulation results show that this method yields PSLL and HPBW that are lower than state-of-the-art.