Abstract An alternative weighted least-squares (WLS) implementation of a two-channel linear-phase quadrature mirror filter (QMF) banks can be efficiently solved in a parallel manner based on the Hopfield neural network (HNN). The author extends an improved structure of feedback neural network to formulate the error function in the optimization of QMF banks as a Lyapunov energy function to find the Hopfield-related parameters. Once these parameters are obtained and input to the network, the optimal filter coefficients of the QMF banks are therefore derived when the network achieves its final state. As compared to the previously developed neural-based method, the proposed framework can be easily applied to the design of QMF banks without incurring convergence problems. In additions, the architecture of the proposed technique is regular and simple with inherent parallelism and can be easily implemented by using analog VLSI technology in real-time. Simulation results are included to illustrate the proposed neural-based approach can achieve the same performance as the method of WLS.