Research On Indoor Location Algorithm Based On Deep Learning

Indoor localization techniques are the focus of the research focus of localization techniques in the recent few years. With the development of wireless communication techniques, the demands of localization services in many fields are increasing rapidly. However, the environment is highly complex, which makes the transmission of the signal affected by the multi-path effect and the time varying channel. As a result, there are many problems to be solved to achieve precise localization in indoor environment. The key point of this paper is to apply a noise reduction deep belief network(DBN) fingerprinting localization algorithm to an indoor localization task. The research work of this paper can be divided into the following aspects:Firstly, the fingerprinting database is acquired by simulation experiment. In order to acquire the signatures which constitute the fingerprinting database, the channel of environment need to be modelled. In this paper bouncing ray method is used to model the channel of a specific indoor environment. The signal transmitter is abstracted to a point launching thousands of rays which are traced during the transmission. And the signal receiver is abstracted to a receiving ball which is used to receive the rays. The channel is modelled by estimating the parameters by analyzing the rays received by a receiving ball.Secondly, the regression model used in the fingerprinting localization algorithm is extremely important. The regression model used in this paper is DBN which is a typical model in deep learning models. DBN is a model that can be used to fit the complex relationship between the signatures to the position information. In this paper, considering the factor that the indoor environment is unstable, a noise reduction pre-training method is used in the pre-training phase of DBN, which makes DBN reduce the effect of noise. This noise reduction pre-training phase makes the model work well with a low signal to noise ratio and adapt to the unstable indoor environment.Thirdly, in order to increase the robustness of the noise reduction DBN location algorithm, the model is considered to be deployed to a distributed environment. The noise reduction DBN can be applied to a localization task even when the number of anchor points is reduced to one. In this paper, the noise reduction DBN models are trained respectively for each anchor point. As a result, every anchor can finish the localization task independently and get the positions of the transmitters. The results of the anchor points are gathered to calculate the final result. The distributed reduction DBN localization algorithm is more robust. Even when some anchor points are invalid, the algorithm can also be used to finish its localization task.