Reasearch And Implementation Of Direction Of Arrival Algorithm Suitable For High-Speed Railway Wireless Communication

The high-speed railway(HSR)is an important part of achieving China transportation power strategy,and has become the preferred way for people to travel long distances.To date,a "four vertical and four horizontal" HSR network has been comprehensively built,while an "eight vertical,eight horizontal" national HSR network is being established,thus China has become the country who has the most developed HSR network around the world.However,due to the high speed of the HSR trains,the HSR mobile communication system suffers from serious Doppler effect and frequent inter-cell handover,which often results in Internet lag and frequent drop call.Therefore,improving users’ Internet experience is a major challenge for HSR wireless communication systems.The smart antenna is an effective technique to solve this problem,and the direction of arrival(DOA)estimation,as an important component of smart antenna,has an important impact on the performance of smart antenna.Therefore,this thesis will investigate the DOA estimation technique for the HSR scenario.Firstly,this thesis elaborates on the principles of subspace class DOA estimation algorithms,and compares the estimation accuracy of various algorithms and the impact of various factors on the performance of the algorithms.In addition,for the coherent signal sources,the spatial smoothing algorithm is used for decoherence to complete the DOA estimation.The analysis shows that these algorithms can achieve super-resolution estimation,but they are based on eigenvalue or singular value decomposition,and the MUSIC algorithm also needs to perform a spectral peak search to obtain signals’ angle information,which has a high time complexity and cannot track the rapidly changing signal sources in real-time.Secondly,radial basis function neural network(RBFNN)has a simple learning rule and strong nonlinear approximation capability,which is very suitable for DOA estimation.DOA estimation is performed by learning from a large number of samples,using training samples collected based on a priori information in the actual environment to build a suitable mathematical model,and it can take into account the actual array model,signal model,signal-to-noise ratio and other factors,so it has better environmental adaptability.In the HSR scenario,the speed and the position of the train at any moment are known.Based on this prior information,the DOA estimation model in the HSR scenario is built,and a suitable training and test set can be constructed.The simulation analysis proves that the DOA estimation based on RBFNN can achieve real-time tracking in high-mobility scenarios,and the computation time is much shorter than the subspace class DOA estimation algorithms.In addition,in the RBF neural network,the number of neurons is always determined at the initial stage,so there is a risk that either too many neurons or too few neurons lead to large errors in the test results.To address this problem,this thesis studies an improved RBFNN for DOA estimation,which can dynamically adjust the number of neurons to build a better network structure and thus improve the performance of the network.Through simulation comparison,the improved RBFNN has a simpler structure,higher estimation accuracy,and less time complexity.Finally,this thesis uses universal software radio peripheral(USRP)to build the DOA estimation platform,completes the algorithm flow chart by using GNU Radio,and verifies the algorithm proposed in this thesis on the platform.This thesis uses USRP X310 with two Twin RX daughter boards to build a quadratic uniform line array,and it firstly corrects the antenna array for phases,then uses the algorithms studied in this thesis for DOA estimation.The experiments demonstrate that the algorithm studied in this thesis can distinguish the direction of the signal in real-time by moving the position of the signal sources.