Norouzi, SajedehSajedehNorouziRahmani, MostafaMostafaRahmaniBraun, TorstenTorstenBraun0000-0001-5968-7108Burr, AlisterAlisterBurr2024-12-102024-12-102024https://boris-portal.unibe.ch/handle/20.500.12422/191619In modern wireless communication systems, accurate channel estimation (CE) is essential for coherent signal detection, ensuring reliable reconstruction of transmitted signals. While deep neural networks have significantly advanced CE accuracy, especially in the context of 5G New Radio (NR), they often suffer from high model complexity. This paper presents GraNet, an innovative graph-based neural network specifically designed for channel estimation. By leveraging the intrinsic graph structure of MIMO communication systems, GraNet enhances the learning process, leading to more efficient and accurate CE. In a comparative analysis with ChannelNet, a prominent deep learning-based CE method, simulation results show that GraNet not only improves estimation accuracy but also substantially reduces computational complexity. Although ChannelNet achieves better accuracy at higher SNR levels, GraNet remains competitive while offering lower computational complexity, making it a more efficient solution. These advantages make GraNet a highly promising candidate for practical deployment in 5G NR MIMO systems.enGraph Neural NetworkMIMO5G NRChannel Estimation000 - Computer science, knowledge & systemsconference_item10.48620/77344