Page 85 - Kaleidoscope Academic Conference Proceedings 2021

OPTIMAL PILOT SEQUENCE DESIGN FOR MACHINE LEARNING BASED CHANNEL
ESTIMATION IN FDD MASSIVE MIMO SYSTEMS

1
1
Hayder AL-Salihi , Mohammed Al-Gharbawi and Fatin Said 2
1
Iraqi Communications and Media Commission, Baghdad, Iraq
2
Department of Informatics, King’s College London, London, UK

ABSTRACT impractical when employing a very large number of antenna
arrays at the base station. [5–8].
In this paper, we consider the problem of channel estimation Pioneering research papers have shown that the inevitable
for large scale Multiple-Input Multiple-Output (MIMO) problem of pilot overhead drawback can be resolved by
systems, in which the main challenge that limits the Compressed Sensing (CS) approaches. Whereby exploiting
functionality of massive MIMO is the acquisition of precise the sparse nature of the wireless channel, the CS algorithms
Channel State Information (CSI). We introduce an efficient can obtain the CSI from a small number of the channel
channel estimation approach based on a block Sparse coefficients. In spite of that fact, the CS-based algorithms
Bayesian Learning (SBL) that exploits the temporal common experience practical restrictions (i.e., CS-based channel
sparsity of channel coefficients. Furthermore, an optimal estimation requires prior knowledge of channel sparsity which
pilot approach to reduce the pilot overhead is derived. The is not feasible practically) that can be addressed through the
optimal pilot is obtained by minimizing the Mean Square Error state-of-the-art of a machine learning (ML) approach i.e.
(MSE) of the proposed SBL estimator using Semi-Definite Sparse Bayesian Learning (SBL). The SBL approaches have
Programming (SDP). Simulation results demonstrate that received much attention recently in wireless communication
the SBL-based approach is more robust than conventional systems since they generally achieve an optimum recovery
methods when fewer training pilots are used. performance and overcome the CS shortages. Thus, in
this paper, we propose a novel channel estimation technique
Keywords - Channel estimation, massive MIMO, based on SBL to reduce the pilot overhead of massive
semidefinite programming, sparse Bayesian learning MIMO systems. The proposed SBL considers the temporal
characterization of wireless channels in contrast to the
1. INTRODUCTION previous published works. Also, the proposed method
models the temporal correlation structure using a probabilistic
Massive Multiple-Input Multiple-Output (MIMO) is structure to avoid overfitting resulting from the limited
considered as the enabling technology for 5G and beyond of knowledge of the channel correlation structure [9–11].
the mobile communication system, thanks to its high system Furthermore, in this paper, in order to obtain an accurate
capacity, high spectral and energy efficiency and high data estimate of CSI and to reduce the pilot overhead further,
rate [1], [2]. Massive MIMO systems have been suggested we perform an enhanced approach to the proposed SBL
to employ tens or hundreds of antennae at the base station; method using the optimal pilot design. The optimal pilots
accordingly, a significant beamforming can be achieved, and are designed to minimize the Mean Square Error (MSE)
the system can serve a vast number of users [3], [4]. subject to the transmit power constraint based on optimization
The major challenge that limits the massive MIMO potential problem formulation. The numerical results show that, with
features is the acquisition of precise Channel State Information optimal training, the MSE can be reduced when compared
(CSI) at the base station. In general, based on the operating with a non-optimal estimation algorithm. This indicates that
duplex mode, the acquisition of CSI can be classified into two we can achieve a better performance with the aid of limited
categories, i.e., Time Division Duplex (TDD) and Frequency pilot resources. Therefore, we can reduce the number of the
Division Duplex (FDD) approaches. While the massive employed pilots to reduce the pilot overhead.
MIMO system has been investigated in (TDD) mode, since the The remainder of this paper is organized as follows. The
channel reciprocity property allows for simpler acquisition of multi-cell massive MIMO system model is presented in
CSI. However, the FDD mode is the standard duplexing of the Section 2. The SBL-based channel estimator is analyzed
majority of the current commercial cellular networks. In FDD in Section 3. Section 4 presents the optimal pilot design
systems, the antennas at the base station send orthogonal pilots approach, while section 5 presents simulation results, and the
to the mobile stations and the channel will be estimated by the conclusions are drawn in Section 6.
mobile station. The estimated channel will be then fedback The following notations are adopted throughout this paper:
to the base station. Hence, the number of orthogonal pilots for any matrix G, 8, 9 denotes the (i,j)th element, while
)
is proportional to the number of antennas that makes FDD the superscripts (.) , (.) −1 and (.) denote the transpose

978-92-61-33881-7/CFP2168P @ ITU 2021 – 23 – Kaleidoscope