I. Introduction
In millimiter-Wave (mm-Wave) underground mine communications, the channel characteristics are a key factor which affects the mining communication applications such as telemontoring, and video tele-surveillance [1]. The first step to combat the channel adverse effect, more particularly at low signal-to-noise ratio (SNR) range, is to accurately and efficiently characterizing and estimating this channel whatever is its nature, thereby ensuring a reliable data recovery [2], [3]. One of the most attractive approaches in channel estimation field, which exploits the sparse temporal structure of the channel, are the recently investigated greedy pursuit methods, such as in compressive sensing (CS) [4]. CS techniques allow accurate channel parameters estimation with less training load. However, the conventional orthogonal matching pursuit (OMP)-based channel estimation approach requires a-priori knowledge of the sparsity level, which is not available in practical scenarios. To overcome this drawback, a novel channel recovery algorithm, referred to as forward-backward (FB) based mm-Wave channel estimator, is proposed in this paper for the Multiple-Input Multiple-Output Orthogonal Frequency-Division Multiplexing (MIMO-OFDM) systems. The estimation is performed upon the channel measurements carried out in real MIMO underground mine environment at 60 GHz band. Unlike conventional methods, such as OMP, the constraint of knowledge of the sparsity level is not necessary in this algorithm, in which two consecutive processing steps are carried out to accurately identify the sparse components of the channel. In the forward step, the support estimate is enlarged. Then, the obtained path index of the channel is used as the initialization set in the backward process to allow the feedback cancellation of the possibly incorrect indices [5]. These forward/backward stages are repeated until the significant channel components are identified.