If left uncontrolled, harmonic currents and voltages in power system present series operational problems both to operators and consumers of electricity ranging from efficiency drop to disruption. Proper monitoring of harmonics on the other hand depends on the accuracy of its estimation. In this paper a more noise resilient family of stochastic gradient based algorithms called Generalized Least Mean Mixed Norm (GLMMN) is applied for amplitude and phase angle estimation of a harmonic power system signal corrupted with a random Gaussian noise. The proposed GLMMN algorithm combines two stochastic gradient algorithms through a mixing parameter that updates online based on a sigmoid function of the instantaneous estimation error. To evaluate the performance of the proposed algorithm, seven distinct mixed norm cases has been examined for power signal containing harmonics in two lower signal to noise ratio (SNR) environments. Comparison of the results obtained using these mixed norm algorithms is made with the respective single error norm algorithms such as LMS, LMAT, LMF. The simulation results demonstrate that the estimation accuracy obtained using combination of two error norms is better than estimations based on single error norm algorithms under impulsive noise interferences.