WITHDRAWN: Machine learning based wireless acoustic communication with enhanced performance

Abstract

Abstract Wireless Acoustic Communication (WAC) is an area that has been developed with different applications. The acoustic channel underwater is very unique and depends on the environment. The WAC channel has limited bandwidth and a strong Doppler effect in relation to wireless radio communication. Machine learning based channel estimation and equalisation has been shown in recent literature to have advantages over conventional (equalising) technology in WAC communication. Since it was hard to build algorithms for WAC communication, ML can be advantageous, because it was difficult to find general channel models. This research is intended to study whether ML-based estimation and equalisation can give enhanced efficiency as part of an advanced physical layer structure. In the analysis the supervised ML is used to increase the bit error rate using a deep neural network and recurring neural network. For the analysis of a wide variety of channels a channel simulator with environmental feedback is used. The simulations are used for the testing of ML environments. It is demonstrated that ML conducts conventional techniques in extremely time-changing channels when trained by the channel beforehand. However, the use of ML without previous channel knowledge did not improve efficiency.