A channel state information and geomagnetic fused fingerprint localisation algorithm based on multi‐input convolutional neural network

Abstract

AbstractNumerous applications require indoor localisation, and one of the current research areas is how to leverage low‐cost ubiquitous signals for indoor localisation. This research designs a multi‐input convolutional neural network (Multi‐CNN) localisation approach to combine natural geomagnetic signals and universal 5G communication signals. To create the location fingerprint data, the geomagnetic three‐component data and channel state information (CSI) must first undergo independent preprocessing. Subsequently, the rebuilt CSI amplitude and geomagnetic intensity are employed for separate offline training to efficiently extract the corresponding data features. Lastly, Multi‐CNN is used to estimate the user's location online. The localisation outcomes for the conference room and hall demonstrate that the Multi‐CNN algorithm can achieve average localisation accuracies of 1.41 and 2.66 m, respectively. These are higher than the single‐input CNN algorithms by 21% and 15%, and higher than backpropagation network (BPNN) algorithm by 24% and 17%, and higher than the weighted K‐nearest neighbour algorithm by 34% and 28%. The Multi‐CNN‐based localisation approach successfully fuses the diverse data, potentially satisfying most indoor localisation applications.