Channel Estimation for Pinching Antennas Systems using Deep Learning

Pinching antennas (PAs) are novel type of antenna that leverage dielectric waveguides and electromagnetic coupling to mitigate large-scale path loss. Channel estimation in PA systems is highly challenging, as it represents an ill-posed problem due to the underdetermined nature of the channel characteristics. This creates a strong need for efficient estimation techniques. In this work, a deep learning framework is proposed that integrates convolutional neural networks (CNNs) with artificial neural networks (ANNs) to estimate channel coefficients on a per-antenna basis, thereby providing greater flexibility and adaptability to varying numbers of PAs in practical deployment scenarios. While the proposed CNN-ANN method delivers efficient performance compared to the conventional approach, it also has low computational complexity.