We present a new approach for signal separation from an undetermined instantaneous mixture of two BPSK (Binary Phase Shift Keying) signals in AWGN (Additive White Gaussian Noise) channel. The method uses frequency diversity of the mixture (frequency shift between carries) and the fact that signals of interest are binary variables. We compare separation results of our method to a theoretical BER (Bit Error Rate) of unmixed signals, which reveal algorithm's performance for different communications scenarios.