关键词: 欠定盲分离, 径向基函数网络, l₀范数, 修正牛顿法

Abstract: When the algorithms based on optimizing approximated l₀ norm are applied to source recovery in underdetermined blind source separation, the complexity is high and the recovery accuracy is greatly affected by the step size. An algorithm for source recovery in underdetermined blind source separation based on radial basis function (RBF) network (SRRBF) was proposed in order to solve these problems. Depending on RBF network, an alternate optimization is performed in the method proposed. Additionally, the approximated l₀ norm is optimized by modified Newton method to avoid inaccurate recovery caused by unsuited step size. Simulations verify that computational complexity of SRRBF is dramatically low while the recovery precision is high.

Key words: underdetermined blind source separation, radial basis function network, l₀ norm, modified Newton method