Gauss-Newton method와 gradient descent method를 결합한 algorithm을 이용한 Generalized sidelobe cancellation beamformer

Abstract

A combined Gauss-Newton and gradient descent algorithm was proposed to speed-up low-frequency convergence rate by around 2~4 times in a generalized sidelobe cancellation (GSC) beamformer as in the Marquardt-Levenberg algorithm. An eigenvalue-based matrix inversion algorithm was proposed for the symmetric augmented Hessian matrix to reduce the matrix inversion time. The combined algorithm was verified using MATLAB code in PC, and the matrix inversion algorithm was verified using C code in android smartphone.