Abstract
The research work invokes the long short-term memory
(LSTM) deep learning model for combating the issue of outdated
channel state information (CSI) during channel estimation
on the wireless medium. For demonstration of the concept, the
downlink free space optical (FSO)/radio frequency (RF) relaying
strategy with outdated CSI has been contemplated. In the considered
amplify-and-forward (AF) cooperative relaying, the channel
gain has been extracted based on the available outdated CSI at the
relay node. Of course, the performance of the FSO/RF downlink
system is inferior due to low correlation between the previous
(original) channel state and the measured CSI during the next
time interval. Since the LSTM can use larger input data sequentially
to predict the next target probabilities based on correlation
among input variables, they become most suited for CSI estimation
from the available outdated CSI. The trained LSTM model
becomes accomplished to estimate the previous state, thus serving
the relay node to adjust the gain more accurately. The trained
LSTM model in the present research work is highly accurate with
mean square error (MSE) and root mean square error (RMSE) of
MSE = −43.49 dB and RMSE = −13.42 dB, respectively.
The performance of the downlink FSO/RF relay has been presented
in terms of outage probability, ergodic capacity and bit error rate
(BER). It has been shown in the paper that using the trained deep
learning LSTM model, the performance of the relaying system can
be made equivalent to that when timing delay exists between the
original and the estimated sample values.