Sparse q-Laplace kernel online prediction for indoor localization in the Internet of Things

As an important component of IoT-oriented applications, the indoor positioning estimation is getting increasing concern with IoT’s rapid development. However, the performance of indoor positioning is heavily relied on the complexity of the environment, which is always full of noises, such as Gaussian noise mixed with impulsive noise. These noises can deteriorate the precision performance of indoor positioning identification systems. In order to attack this problem, we propose a new kernel called generalized q-Laplace kernel to produce a new q-Laplace kernel adaptive filtering algorithm (qLaKAF), which is combined with the recently proposed kernel mean p-power error criterion (KMPE). The proposed qLaKAF has two vital features. Firstly, the q-Laplace kernel is employed to combat the Gaussian noise together with abrupt noise in real-world scenarios. Besides, the KMPE is utilized to obtain higher-order information in addition to second-order information which facilitates to suppress mixed noise. Furthermore, a Strengthened Surprise Criterion (SSC) is applied to qLaKAF to reduce the size of neural networks. The qLaKAF algorithm assisted by SSC is called Strengthened Surprise Criterion q-Laplace kernel adaptive filtering algorithm (SSC-qLaKAF). Three experiments are carried out on two real-world scenarios to validate the effectiveness and accuracy performance. The experimental results demonstrate that the accuracy has been improved by at least 3.6%; meanwhile, the SSC-qLaKAF neural network size can be reduced by up to 12.5%, without much loss of accuracy performance compared to qLaKAF.