Abstract. Channel noise is a key factor that affects the measurement accuracy of narrowband wireless indoor positioning. To reduce noise effects, it is often required to effectively filter the received spread spectrum signals. However, current filtering methods in the time or frequency domain can only filter high frequency band noise, making the measurement accuracy low. In this paper, a low frequency band noise reduction algorithm based on the threshold segmentation of the rectangular window is proposed. The method is based on thresholding on time-frequency wavelet domain by variance analysis and signal despreading is achieved by wavelet autocorrelation at low frequency band after denoising with reference to locally generated spread spectrum signal without added noise at the same subband. Simulation results show that compared with other filtering methods, the proposed method by wavelet autocorrelation and denoising can eliminate the noise effects at the receiver, leading to improving the positioning accuracy significantly.

As most systems are inherently nonlinear in nature, many efforts have been made to improve the understanding of complicated nonlinear models. However, current research has indicated that it is still a challenge to accurately model and identify nonlinear systems by conventional methods such as machine learning. This paper investigates a complex nonlinear system with three parameters identification by training a Deep Neural Network (DNN) to model the system based on Fourier series theory. The DNN with 10 layers is constructed such that it can model any nonlinear system, and the parameters identification is performed by the trained neural networks. The proposed method has been evaluated by applying to a nonlinear system for multiple parameters measurement by interferometric fiber sensors. Experimental results demonstrate that the DNN can accurately model the nonlinear system and identify the corresponding parameters, leading to a solution to complex nonlinear system approximation with minimized error.