An Adapted Mean Teacher Model for Semi-supervised Indoor Positioning using Channel Impulse Response

As an effective solution to alleviating the insufficiency of labeled data for indoor positioning, deep semi-supervised learning (DSSL) can be employed to lessen the dependency on labeled data by exploiting potential patterns in unlabeled samples. Inspired by the inherent similarity between image processing and indoor positioning, and the efficiency of the consistency regularization method for image classification, we propose an adapted mean teacher (AMT) model under the DSSL paradigm for indoor positioning by using channel impulse response. For enhancing the generalization of the trained model, we design an efficient implicit augmentation scheme for the training process in AMT model. In addition, we design a special residual network to efficiently extract location characteristics in the AMT framework. We conduct extensive simulation experiments for indoor scenarios with a heavy non-line-of-sight condition to demonstrate the effectiveness of our proposed AMT model. Numerical results illustrate that AMT model outperforms a number of consistency regularization methods and the pseudo-label method, in terms of accuracy and convergence.