This paper proposes a new codebook-based SDMA scheme, which employs a variable number of active beams, i.e., a dynamic multiplexing order, to target a good tradeoff between the multiplexing gain and the inter-user interference. The new transmit scheme consist of a modification of the conventional feedback information and the development of a novel scheduling algorithm, where the new scheduling algorithm utilizes a merging of the channel correlation with limited feedback information, and is based on a proposed ML estimate of the possible user interferences. Simulation results demonstrate that in sparse networks the proposed SDMA scheme considerably outperforms some conventional SDMA schemes in terms of the sum throughput, with a comparable feedback overhead.