Least Squares Support Vector Machines for Channel Prediction in the MIMO System - Developing Concepts in Applied Intelligence - page 39

Least Squares Support Vector Machines for
Channel Prediction in the MIMO System

Jerzy Martyna

Institute of Computer Science, Jagiellonian University,
ul. Prof. S. Lojasiewicza 6, 30-348 Cracow, Poland

Abstract. A new LS-SVM method for a multiple-input multiple-output
(MIMO) channel prediction is presented. A least squares support vector
machine (LS-SVM) is proposed as a prediction technique. The LS-SVM
has nice properties in that the algorithm implements nonlinear decision
regions, converges to minimum mean squared error solutions, and can be
implemented adaptively. We also formulate a recursive implementation
of the LS-SVM for channel prediction in the MIMO system. The perfor-
mance of the new method is shown by a simulation of the bit error rate
in the given environment.

Keywords: channel estimation, multiple-input multiple-output (MIMO)
channel, least squares support vector machine (LS-SVM).

1 Introduction

During the last several years support vector machines (SVM) and kernel methods
[1], [10] have generated considerable interest and have been used in many appli-
cations from image processing to analyzing DNA data. These methods belong to
one of the preeminent machine learning statistical models for classification and
regression developed over the past decade.
The multiple-input multiple-output (MIMO) communication systems have
recently drawn considerable attention in the area of wireless communications.
Multiple antenna systems are one of the key technologies for the next genera-
tion wireless communications, especially in WCDMA-based 3G systems [5]. This
technique has attracted much attention due to its advantage in capacity as well
as the ability to support multiple users simultaneously [3]. In the MIMO system
it is usually assumed that the channel state information (CSI) is known at the
receiver, but not at the transmitter, and the transmitter has no knowledge about
the channel coecients. Although the increasing number of antennas improves
the capacity of the system, it increases its complexity. However, these methods
waste time while learning the channel when meaningful data can be sent.
The use of the SVM method and recently developed the least squares (LS)
SVM method [7] have been proposed to solve a number of digital communication
problems. Among others, signal equalization and detection for a multicarrier
(MC)-CDMA system based on SVM linear classification has been investigated by
Rahman et al. [4]. In the paper by Sanchez-Fernandez et al. [6] SVM techniques