Cell-Free Massive MIMO With Multi-Antenna Users: Do We Really Need Downlink Pilots?

Cell-Free Massive MIMO (CF-mMIMO) networks are set to revolutionize wireless communication, overcoming fifth-generation (5G) limitations and paving the way for sixth-generation (6G). This paper explores channel estimation strategies in CF-mMIMO networks with multi-antenna users, filling a key research gap. A pilot-based linear minimum mean squared error (MMSE) estimator and a novel blind estimation approach leveraging the sample average power of the received signal vector at the mobile stations (MSs) are proposed assuming the transmission on spatially correlated Ricean fading with phase shifts. Rigorous capacity lower bounds are derived for various detection/decoding strategies, including the classical hardening-based detection (HBD), which relies on statistical channel state information (CSI), the pilot-based detection (PBD), the blind-based detection (BBD), and the genie-aided ideal CSI-based detection (IBD) across different precoding configurations. Numerical results reveal that with effective pilot contamination management, the classical HBD scheme performs close to optimally with advanced centralized precoding. However, in user-centric (UC) CF-mMIMO networks with distributed precoders, the BBD strategy consistently outperforms the other schemes. For multi-antenna users facing high pilot contamination, DL detection schemes using instantaneous CSI are preferable, with the blind-based scheme being the one providing the best performance metrics when jointly implemented with advanced precoding strategies. This study underscores the trade-offs between statistical and instantaneous DL channel estimation, pilot overhead, and precoding strategy, offering insights on how to design and manage a CF-mMIMO network with multi-antenna users and guiding future research on this particular topic.