FEEL stands out as a promising enabling technology for performing training and inference at network edges, bypassing the need of sharing raw data with central servers. To facilitate FEEL, federated averaging (FedAvg), as a first-order algorithm, has been widely adopted. A crucial step in utilizing FedAvg is model aggregation, which requires substantial communication overhead between edge devices and the central server. However, limited radio resource remains a significant bottleneck for aggregating locally computed model updates. To address this issue, this chapter investigates a novel over-the-air computation (AirComp) based approach for low-latency global model aggregation. This approach involves joint device selection and beamforming design, which is modeled as a sparse and low-rank optimization problem. To achieve efficient algorithm design, we provide a difference-of-convex-functions (DC) representation for the sparse and low-rank functions, enhancing sparsity and accurately detecting the fixed-rank constraint during device selection. Furthermore, we develop a DC algorithm to solve the resulting DC program with global convergence guarantees. The advantages and superior performance of the proposed algorithm are demonstrated through simulations.