User clustering in cell-free massive MIMO NOMA system: A learning based and user centric approach

For future wireless communications, Cell-free Massive Multiple-Input Multiple-Output (CF-mMIMO) systems and Non-orthogonal Multiple Access (NOMA) schemes are considered potential candidates to meet the greater coverage and capacity demands. Nevertheless, a traditional CF-mMIMO system faces scalability issues and poses numerous challenges in handling the expanding number of user equipment and ensuring their dependable connectivity, particularly in larger geographical areas. To address this challenge, a user-centric (UC) approach is implemented in a CF-mMIMO system, wherein a designated subset of access points (APs) serves a specific number of users from the entire pool of available APs. To implement a NOMA aided CF-mMIMO system, users must be grouped using a suitable clustering scheme to achieve greater spectral efficiency (SE), sum-rate, and reduced bit error rate (BER). For efficient user clustering, unsupervised machine learning (ML) algorithms, such as k-means, k-means++, and improved k-means++ are employed. In this paper, a multiuser NOMA aided CF-mMIMO system with a UC approach is investigated and closed-form expressions for intra-cluster interference and SINR are derived and the performance of the proposed system is analyzed in terms of achievable sum-rate and BER. The proposed system with the UC approach and three ML algorithms namely k-means, k-means++, and improved k-means++ demonstrate 12%, 10%, and 17% higher achievable sum-rate as compared to the NUC approach with same ML algorithms respectively. Similarly, the proposed system with UC and ML approaches exhibits 52%, 55% and 61% improved achievable sum-rate respectively, as compared to far pairing, random pairing, and close pairing schemes. Moreover, the system model is validated through the conformity of the theoretically derived bit error rate with the simulation results for a three-user scenario.