Optimizing UWB CIR Truncation for NLOS Identification on MCU Deployment

Truncating the channel impulse response (CIR) is a critical step for non-line-of-sight (NLOS) identification within ultra-wideband (UWB) systems to manage data complexity. This letter formulates the process as a constrained and quantifiable optimization problem, thereby achieving an optimal balance between computational performance and classification accuracy. To implement this methodology, we first designed a lightweight attention-assisted network to serve as an effective evaluation tool. Leveraging this network, we then conducted a systematic search that identified an optimal input configuration: a CIR subset containing only 63 sample points. This co-optimized input model pair achieves a remarkable NLOS recall of 93.68%. More importantly, this result provides a critical a priori determination for deploying high precision localization algorithms on resource constrained MCUs. This co-optimized system has been successfully quantized and deployed on an STM32 MCU, verifying its high efficiency and feasibility for industrial grade applications.