Non-orthogonal multiple access (NOMA) is an important candidate for 5G radio access technology. In NOMA transmitter, different users' signals are superposed on the same radio resource with different power allocation factors. The receiver removes other users' signals from the received signal before decoding its own signal. In this work, an iterative gradient ascent-based power allocation method is proposed for downlink NOMA transmitter. It maximizes the geometric mean of the throughputs of users who share the same radio resource to provide proportional fairness between users. Simulation results show that the method achieves theoretical best results in terms of the suggested metric. Also it is shown that it increases the efficiency as much as 80% when compared to orthogonal multiple access (OMA) and it gives better results than NOMA that uses fixed and fractional power allocation methods.