Abstract:
In numerous WiFi sensing applications, such as passive human localization, the precision of sensing is often influenced by the estimation accuracy of multipath parameters...Show MoreMetadata
Abstract:
In numerous WiFi sensing applications, such as passive human localization, the precision of sensing is often influenced by the estimation accuracy of multipath parameters. Several existing algorithms leverage pulse-shaping filter information to enhance multipath and channel estimation. However, WiFi chips do not disclose this filter information, and no current research has focused on measuring or estimating these pulse-shaping filters. In this paper, we introduce a new deep learning approach for the accurate estimation of pulse-shaping filters using channel state information (CSI), which incorporates both multipath channel information and pulse-shaping filter information. Specifically, we construct a convolutional neural network consisting of an encoder-regressor architecture, where the encoder translates the CSI into a latent representation, and the regressor subsequently estimates the pulse-shaping filter from this representation. Our proposed model's efficacy is demonstrated through its low normalized root mean squared error (NRMSE) in a variety of channel conditions, highlighting its ability to accurately estimate pulse-shaping filters.
Date of Conference: 09-13 June 2024
Date Added to IEEE Xplore: 20 August 2024
ISBN Information: