Font Size: a A A

Research On WiFi Localization System Based On Virtual Antenna Array And Deep Learning

Posted on:2021-04-11Degree:MasterType:Thesis
Country:ChinaCandidate:F LiuFull Text:PDF
GTID:2428330611957115Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Accurate indoor target location information has become an indispensable part of people's daily life.High-precision indoor positioning information is required for the location of specific stores in large shopping malls,the specific location of a department in a strange hospital,and smart home applications.Because GPS signals are weak indoors and cannot work normally,WiFi signals have become a hot signal for researchers to compete with due to their absolute advantages such as widespread deployment and low cost.In order to improve the localization accuracy,the traditional WiFi localization technology mostly adopts the method of increasing the antenna array,and sacrifices the hardware to improve the positioning accuracy.In practice,most commercial WiFi devices only support 3 antennas.If you increase the number of antennas,you must modify the device hardware and increase deployment costs.This loses the unique advantages of using WiFi.Therefore,in this paper,combined with the rapid development of deep learning and virtual antennas in recent years,without requiring hardware changes,a DOA high-precision estimation method combining convolutional neural networks and virtual antenna arrays is proposed,which effectively improves the DOA indoor localization accuracy.This article mainly studies the following three parts:1.Aiming at the problem of high physical cost and difficult deployment of multiple antennas in WiFi localization system,a system framework(called Wi-VAL)based on virtual antenna array and deep learning,using DOA estimation to implement WiFi localization is proposed.Based on the mathematical model of DOA estimation,the relationship between the resolution of DOA estimation and the antenna array is theoretically analyzed by formula derivation,and the "voting method" is used to verify the experiment.Finally,based on this relationship,the Wi-VAL localization system was proposed.2.In order to achieve the sensing effect with a larger number of antenna arrays without increasing the actual number of antennas,a virtual antenna expansion method based on channel impulse response is proposed.First,by studying the propagation characteristics of wireless signals and the transmission characteristics of indoor wireless channels,a statistical channel model of WiFi multi-subcarrier characteristics based on OFDM modulation is established to generate channel impulse response information.Furthermore,the transformation matrix required for virtual antenna expansion is designed using the idea of interpolation transformation,and this matrix is combined with the channel impulse response to obtain virtual antenna array data for DOA estimation.Finally,compared with the original WiFi AP with four fixed antennas,the performance of this method in DOA estimation is significantly improved.3.In order to better improve the accuracy of DOA estimation,a neural network is introduced on the basis of the virtual antenna array,and a DOA estimation method that combines a convolutional neural network and a virtual antenna array is proposed.The Wi-VAL system is implemented.First,the DOA estimation problem is transformed into a classification problem in deep learning,and then a deep learning architecture with a convolutional neural network as the front end is constructed.After training,a CNN-based DOA estimation model CNN-DOA is obtained.We use the CNN-based DOA estimation model to optimize the virtual expanded perception matrix of the antenna array,and finally obtain a high-precision DOA.The final experiment shows that the accuracy of the Wi-VAL positioning system based on CNN-DOA method has also been significantly improved.In the corridor scene,the error is 0.65 m.
Keywords/Search Tags:Indoor WiFi localization, wireless channel modeling, CNN, DOA estimation
PDF Full Text Request
Related items