Design of Learning Based MIMO Cognitive Radio Systems

TL;DR

This paper proposes a practical multi-stage learning and transmission strategy for MIMO cognitive radio systems, optimizing power and time allocation to maximize capacity while protecting primary users.

Contribution

It introduces a novel three-stage CR transmission scheme with environment learning, channel training, and data transmission, including capacity bounds considering estimation imperfections.

Findings

Derived a lower bound on ergodic capacity for CR with imperfect channel knowledge.

Identified a fundamental tradeoff among learning, training, and data transmission stages.

Optimized power and time allocation to enhance CR system capacity.

Abstract

This paper addresses the design issues of the multi-antenna-based cognitive radio (CR) system that is able to operate concurrently with the licensed primary radio (PR) system. We propose a practical CR transmission strategy consisting of three major stages: environment learning, channel training, and data transmission. In the environment learning stage, the CR transceivers both listen to the PR transmission and apply blind algorithms to estimate the spaces that are orthogonal to the channels from the PR. Assuming time-division duplex (TDD) based transmission for the PR, cognitive beamforming is then designed and applied at CR transceivers to restrict the interference to/from the PR during the subsequent channel training and data transmission stages. In the channel training stage, the CR transmitter sends training signals to the CR receiver, which applies the linear-minimum-mean-square-error (LMMSE) based estimator to estimate the effective channel. Considering imperfect estimations in both learning and training stages, we derive a lower bound on the ergodic capacity achievable for the CR in the data transmission stage. From this capacity lower bound, we observe a general learning/training/throughput tradeoff associated with the proposed scheme, pertinent to transmit power allocation between training and transmission stages, as well as time allocation among learning, training, and transmission stages. We characterize the aforementioned tradeoff by optimizing the associated power and time allocation to maximize the CR ergodic capacity.