Non-orthogonal multiple access (NOMA) has been recognized as a potential technique to achieve higher spectral efficiency (SE) for future 5G systems. However, the anticipated thousand-fold increase in wireless data traffic urgently calls for new designs of power-efficient communication systems. In this paper, we investigate NOMA from the perspective of energy efficiency (EE) in a multiuser downlink system. Firstly, we generalize an existing cognitive radio (CR) inspired NOMA scheme by introducing multiple antenna techniques and extending the number of primary users to an arbitrary number. Then, we aim at optimizing the EE of this generalized CR inspired NOMA scheme subject to an individual quality of service (QoS) constraint for each primary user, which leads to a non-convex fractional programming problem. For this challenging problem, we propose an efficient algorithm based on the sequential convex approximation (SCA) method. Our numerical results show that NOMA has superior EE performance in comparison with conventional orthogonal multiple access (OMA).