NOMA Decoding: Successive Interference Cancellation or Maximum Likelihood Detection?

Is successive interference cancellation (SIC) decoding always the optimal choice in non-orthogonal multiple access (NOMA) systems? While the answer is positive based on Shannon theory, which is applicable to infinite-length codewords drawn from a Gaussian distribution, this may not universally hold for systems with finite-alphabet inputs. Specifically, in this paper, we demonstrate that for quadrature amplitude modulation (QAM)-based NOMA, SIC decoding fails for certain values of power allocation coefficient a:, used to divide power among NOMA users. With this observation, we propose employing maximum likelihood (ML) detection to decode QAM-NOMA. While SIC decoding for QAM-NOMA requires allocating higher power to the user with a weaker channel to prevent symbol crossing in super-constellations, ML detection can successfully handle a broader range of power allocation coefficients. We then derive closed-form symbol error rates for quadrature phase shift keying-based NOMA systems across any a: and validate them through simulations. The results demonstrate the effectiveness of ML detection, particularly in scenarios where SIC decoding fails.