MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641050605 A) filed by Srinivasa Ramanujan Institute Of Technology; Mr. M. Sivalingamaiah; S. Sai Sushma; J. V. Sreenath Reddy; and C. Ajay Kumar, Ananthapuramu, Andhra Pradesh, on April 21, for 'machine learning-driven channel state information estimation for next generation networks.'
Inventor(s) include Srinivasa Ramanujan Institute Technology; Mr. M. Sivalingamaiah; S. Sai Sushma; J. V. Sreenath Reddy; and C. Ajay Kumar.
The application for the patent was published on May 1, under issue no. 18/2026.
According to the abstract released by the Intellectual Property India: "Machine Learning-Driven Channel State Information Estimation for Next Generation Networks Abstract: The invention relates to a machine learning-based framework for efficient channel state information (CSI) estimation in frequency division duplex (FDD) massive MIMO systems. Traditional CSI feedback methods suffer from excessive overhead and limited reconstruction accuracy. This invention proposes an enhanced CsiNet architecture integrated with a spatial attention mechanism to emphasize dominant channel features during compression and improve reconstruction accuracy at the base station. The decoder is strengthened with increased feature maps to capture complex spatial correlations. Performance evaluation using normalized mean square error (NMSE) and cosine similarity demonstrates superior reconstruction accuracy and beamforming consistency compared to conventional methods. The invention enables scalable, low-complexity CSI estimation suitable for next-generation 5G and 6G wireless communication systems. The spatial attention mechanism allows the encoder to dynamically prioritize high-energy angular-delay regions, improving the informativeness of the compressed representation. The decoder's expanded feature maps enhance its ability to refine CSI reconstructions, especially under low compression ratios. Unlike iterative compressive sensing techniques, this framework offers real-time performance with minimal latency. The architecture is trained end-to-end using supervised learning on standardized channel datasets, ensuring generalizability across diverse propagation environments. The proposed method supports adaptive beamforming, interference mitigation, and robust operation in bandwidth-constrained feedback channels. It is particularly effective in mmWave and THz bands, where channel sparsity and directionality are pronounced. Applications include ultra-reliable low-latency communications (URLLC), mobile edge computing (MEC), and massive machine-type communications (mMTC). The invention contributes to intelligent wireless networks by enabling dynamic CSI estimation aligned with AI-driven optimization strategies."
Disclaimer: Curated by HT Syndication.
