Research On PAPR Reduction In Underwater Acoustic OFDM Communication System

The recent increasing applications in the underwater environment require huge data sharing networks and efficient wireless communication systems,but communication in underwater undergoes major issues.Due to robustness against multi path propagation,orthogonal frequency division multiplexing(OFDM)becomes the only choice.After successful terrestrial applications OFDM is also implemented for underwater and acoustic communications.But unlike terrestrial applications,implementation of OFDM technology in underwater environment faces many difficulties due hydro acoustic channel parameters in the underwater i.e.Doppler effect,beam forming,and many heavily powered devices are required,which are more costly than terrestrial networks.In this dissertation we have introduced a deep learning based method,which is not only cost effective but can reduce the peak to average power ratio(PAPR)by using intelligent learning algorithms which avoids the non-linearity of power amplifier,hence additional high power devices are eliminated.After observing different PAPR reduction techniques,the OFDM communication systems found some complexities in the network which decreases the speed due to high computation,and also found accuracy problems.So the machine learning algorithms have resolved these issues,because they are more fast,reliable and cost effective.Using deep neural network(DNN)learning algorithm in underwater acoustic OFDM communication system can have improved bit error rate(BER)performances.The proposed deep learning method has replaced the normalization layer with “Tanh” layer.In the first step this method mitigates the PAPR,and after that improves the performances of power amplifier,that undergoes non-linearity behavior.This method performs efficient learning of auto-encoders for efficient searching of phase factors and improves the BER performances.Moreover this dissertation also discusses convolutional neural networks(CNN)method with soft feed-back is employed.It is important to find out phase factors with least PAPR like other techniques such as selective mapping(SLM).But for the same reason,we have designed a neural network method with the help of neurons,which then can find the optimum search of phase factors and allowing the phase factors with minimum PAPR.In the neural network based method the PAPR can be decreased by adding “k” variable also defining the range of the distribution i.e.+1 to-1,which is to get optimum searches and eliminating the lowest point.The method is then compared with Hopfield Neural networks(HNN)and is verified with the results illustrating the efficiency and PAPR performances are according to the requirement and also resolved other power issues.