Canonical decomposition of multidimensional noisy MIMO channel estimation

In modern MIMO wireless communication systems with a large number of antennas (Massive MIMO), the accuracy of channel state estimation plays a key role in ensuring high data throughput (spectral efficiency). Accurate estimation of channel parameters allows the base station to form a spatially directed transmission filter, compensating negative effects such as multipath propagation, inter-symbol interference, and additive white Gaussian noise. In such systems, low signal-to-noise ratio in the uplink can significantly limit the throughput in the downlink channel. Traditional channel estimation methods often prove insufficiently effective under strong noise and multipath propagation conditions. A major limitation of conventional estimation techniques consists in the independent processing of each MIMO channel, which fails to exploit the spatial properties of the filtered signal for interference and noise suppression. This paper proposes an approach based on transforming the multidimensional MIMO channel estimation signal into a multimodal tensor for joint processing across all dimensions, allowing more effective utilization of the data structure. As demonstrated in the study, tensor-based processing methods for multidimensional signals can potentially provide up to a 10% improvement in MIMO channel estimation accuracy under low SNR conditions (up to -20 dB).