Research On Frame Synchronization Based On Extreme Learning Machine

As an important components of wireless communication systems,different frame synchronization(FS)methods directly affect the performance of the wireless communication systems.Seeking a FS method with lower error probability is research interest of experts and scholars.However,a large number of devices inevitably cause nonlinear distortion in wireless communication systems,which reduces the error probability performance of FS.To improve the error probability performance of FS and reduce the bit error rate(BER)of communication systems,the main work of this thesis is as follows.First,an ELM-based FS with nonlinear distortion is investigated by employing the ELM network to learn nonlinearity.The coarse features of synchronization metric(SM)are first captured by preprocessing,and then ELM network is constructed to learn and estimate the offset of frame boundary.The analysis and results show that compared with the classical FS method,the ELM based FS scheme could significantly reduce the error probability of FS while improve the performance in terms of robustness with nonlinear distortion.Then,an improved ELM-based FS method is investigated by developing the inter-frame correlation.First,the multi-frame data is preprocessed by weighted superposition,followed by the preprocessed SM is fed into ELM network for offline learning.The simulation results show that,compared with the cross-correlation based FS,compressed sensing-based FS,and ELM based FS,the improved method can effectively improve the error probability performance of FS.Finally,aiming at the training sequence of ELM-based FS inevitably occupies the bandwidth resources,an ELM based FS using superimposed training is investigated to reduce the error probability of FS and avoid the occupation of bandwidth resources of training sequence used for FS.At the transmitter,the weighted training sequence is superimposed on the data symbols for transmission,the ELM network is constructed to learn and obtain the frame offset at the receiver.Compared with existing methods,this method not only saves bandwidth resources,but also improves the error probability performance of FS and BER of symbol detection.The ELM based FS in this thesis can be applied to burst-mode communication systems or other communication systems with higher requirements for machine learning,e.g.,the 5th generation(5G)wireless systems,the 6th generation(6G)wireless systems and others.