CrossNet: Joint Channel Estimation and Localization in Deep Learning Method
Chongyang Li, Tianqian Zhang, Shouyin Liu · IEEE Communications Letters · 2025
This letter proposes CrossNet, a novel deep learning (DL) approach for joint channel estimation and outdoor localization. Similar to fingerprint methods that utilize features such as angle of arrival (AoA) and receive signal strength indicator (RSSI), CrossNet leverages neural networks to extract positional information from channel state information (CSI). However, instead of relying on direct matching within a database, CrossNet learns the implicit relationship between CSI and location through training, enabling more accurate and robust localization. The purpose of joint channel estimation and localization is to obtain more precise positioning information from more accurate channel estimation. We built a single-input single-output (SISO) downlink communication system on the DeepMIMO dataset and generated the necessary data for our experiments. We conducted multiple comparative experiments to evaluate the performance of CrossNet. Extensive comparative experiments demonstrated that CrossNet effectively utilizes pilots for user equipment (UE) localization and significantly improves localization accuracy through joint channel estimation.