A method for correcting ultra-wideband measurement distance using multi-channel neural networks

Abstract Ultra-wideband (UWB) ranging is adversely affected by non-line-of-sight propagation, resulting in significant errors in indoor positioning. To mitigate these errors, we propose a multi-channel neural network model named MultiChannelNet (MCN), which integrates three complementary features: measured distance, channel impulse response, and Received Signal Strength Indicator (RSSI). Each feature is processed through a dedicated network branch and fused to achieve robust distance correction. To address the challenge of hyperparameter tuning in neural networks, we introduce an improved whale optimization algorithm (IWOA) for efficient optimization. The IWOA-MCN method was combined and evaluated in two open-source datasets. The experimental results show that, in the case of existing environmental data, IWOA-MCN can significantly reduce the ranging error and positioning error. In cross-validation scenarios without prior data and cross-environment experiments, it can still achieve a relatively good error reduction effect. Compared with the existing methods, this method has strong adaptability and superior performance, providing a practical solution for high-precision indoor positioning based on UWB.