A Message Passing Approach to Acquire mm-Wave Channel State Information Based on Out-of-Band Data

Millimeter-wave (mm-wave) communication is a practicable scheme for big data communication, such as next-generation cellular communication. However, because mm-wave frequencies have an extremely large path loss, in order to mitigate the mm-wave path loss, a large number of antennas are packed for beamforming. Conventional multiple-input multiple-output (MIMO) beamforming uses digital processing, which leads to needing extremely giant energy. Hybrid analog/digital beamforming could serve as an awesome technique to reduce the cost. It is essential for achieving high beamfroming gain that accurate mm-wave channel information should be attained at the transmitter. This paper concentrates on the channel state information (CSI) acquirement problem in mm-wave communication systems with massive MIMO arrays. Because of the CSI acquirement, a method of significant overhead, we consider an accurate channel estimation scheme with low overhead. Not only do we in this paper propose using support information extracted at Sub-6GHz to aid the mm-wave CSI acquirement but also introduce a strategy used in the mm-wave channel estimation. We, in particular, formulate mm-wave CSI acquirement as a compressive sensing problem and obtain the CSI by using Bernoulli-Gaussian generalized approximate message passing (BG-GAMP) algorithm. We also extend the BG-GAMP algorithm with support distribution information from sub-6-GHz channel. Furthermore, based on the K nearest neighbor idea, we redesign the BG-GAMP algorithm depending on sub-6-GHz support distribution information. Details of BG-GAMP with the nearest neighbor learning algorithm that is built on support distribution information from sub-6-GHz channel would also be revealed. Simulation results show the out-of-band information aided mm-wave channel estimation is capable of reducing the pilot overhead greatly and channel estimation accuracy improved as well.

Abstract licence: CC BY-NC-ND

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.