We consider a downlink multiuser MISO system
with bounded errors in the Channel State Information at the
Transmitter (CSIT). We first look at the robust design problem
of achieving max-min fairness amongst users (in the worstcase
sense). Contrary to the conventional approach adopted in
literature, we propose a rather unorthodox design based on a
Rate-Splitting (RS) strategy. Each user’s message is split into
two parts, a common part and a private part. All common
parts are packed into one super common message encoded
using a public codebook, while private parts are independently
encoded. The resulting symbol streams are linearly precoded
and simultaneously transmitted, and each receiver retrieves its
intended message by decoding both the common stream and
its corresponding private stream. For CSIT uncertainty regions
that scale with SNR (e.g. by scaling the number of feedback
bits), we prove that a RS-based design achieves higher max-min
(symmetric) Degrees of Freedom (DoF) compared to conventional
designs (NoRS). For the special case of non-scaling CSIT (e.g.
fixed number of feedback bits), and contrary to NoRS, RS can
achieve a non-saturating max-min rate. We propose a robust
algorithm based on the cutting-set method coupled with the
Weighted Minimum Mean Square Error (WMMSE) approach,
and we demonstrate its performance gains over state-of-the art
designs. Finally, we extend the RS strategy to address the Quality
of Service (QoS) constrained power minimization problem, and
we demonstrate significant gains over NoRS-based designs.