Year: 2026
Type: article
Abstract: The current paper introduces a new deep learning model of the adaptive beamforming structure of smart antenna systems to enhance the spatial selectivity and interference rejection. The proposed model incorporates angular awareness into the recurrent learning process, known as the Angular‐Attention Bidirectional Gated Recurrent Unit (Attn‐BiGRU), which allows dynamic attention to be paid to beam and null directions, as shown in Figure 1. The model can be gradually taught more difficult angular situations through curriculum learning of the integration, which increases convergence and strength. The model was trained and tested using synthetically generated data that was based on a 10‐element linear dipole array at 1,800 MHz. Standard error measures: MSE, RMSE, MAE, and antenna‐specific measure, SLLmax were used to evaluate the performance. According to the comparative outcomes, the Attn‐BiGRU demonstrated better beam alignment, null depth, and sidelobe suppression than the traditional algorithms (VS‐LMS, RLS, SMI) and the deep learning ones (ANN, BiRNN, BiLSTM, BiGRU). The method can be applied to real‐life scenarios and be used in the area of vehicular networks, reconfigurable surfaces, and 6G beam management architectures. The findings indicate that the model is not very simple, and it would serve as an effective and precise solution to real‐time and time‐sensitive applications in the next generation wireless communication systems.
Source: Radio Science
DOI: 10.1029/2025rs008432
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7120090173