AI for 5G/6G Air Interface

3GPP’s Release 18/19 AI/ML air-interface study items set out to find where a trained model actually beats classical signal processing once training-data collection, generalization across deployments, and model-monitoring overhead are accounted for. Beam prediction, CSI compression, and positioning surfaced as the three use cases with the clearest gains.

I lead the AI/ML model development and 5G-Advanced system-level simulations behind this work: spatial and mobility-aware beam prediction using CNN and Transformer architectures, deep-learning CSI feedback frameworks that skip full channel reconstruction, and the AI/ML positioning models covered under Positioning in Cellular Networks. What began as pre-standardization feasibility studies at CEWiT, IIT Madras continues today at Tejas Networks, where I drive the model design, simulation platform, and patent portfolio that feed 3GPP Release 18/19 standardization.

Selected publications

• PosNet: A CNN-based Transformer Model for UE Localization in NLOS Dominated Scenarios — IEEE Communications Letters (submitted 2026). Transformer-backed CNN that attends to the most informative CIR taps, reaching 5.4 cm accuracy for 90% of UEs in NLOS-dominated channels.
• SwitchConvNet: Deep Learning Framework for Mobility-Aware Spatial Beam Prediction in Beyond 5G and 6G Systems — IEEE Transactions on Vehicular Technology (accepted 2026). Mobility-aware spatial beam prediction with an adaptive spline/Gaussian switching filter, cutting beam-acquisition time by over 80% with 77% top-1 and 97% top-3 accuracy.
• A Hybrid CNN-Transformer Framework for Precise Indoor Positioning in 6G Networks — IEEE MeditCom 2026. Hybrid CNN-Transformer architecture pairing local feature extraction with long-range attention for robust indoor positioning under severe NLoS.
• SuperConvNet: Super Resolution Orchestrated Deep Convolutional Neural Network for Spatial Beam Prediction in B5G and 6G Systems — IEEE Transactions on Machine Learning in Communications and Networking (submitted 2026). Super-resolution CNN for spatial beam prediction that cuts beam-acquisition latency by over 80% while improving throughput, with 84% top-1 and near-100% top-3 accuracy.
• PrecoderNet: A Feedback Framework for Massive MIMO-OFDM Systems — IEEE COMSNETS 2026. Deep-learning encoder/decoder that estimates the SVD-based precoding matrix directly, cutting CSI feedback overhead without full channel reconstruction.
• AI-ML Models for Wireless Positioning Using Channel Information — Indian Patent Application 202541112619 (2025). AI/ML models that extract location-sensitive features directly from channel information to estimate device position.
• Performance evaluation of AI-based CSI feedback schemes compliant with 3GPP standards — Physical Communication (2025). M-CsiNet, an AI-based CSI compression and reconstruction model benchmarked against 3GPP Type-II codebooks, delivering 10-15 dB SNR gain with two orders of magnitude less feedback overhead.
• AI/ML-Based Downlink Beam Prediction for Enhanced 5G-Advanced and 6G Networks — Indian Patent Application 202541083874 (2025). AI/ML techniques that predict suitable downlink beams to cut the measurement and signaling cost of beam selection and tracking.
• Efficient Data Collection and Model Architecture for AI-ML-Based Positioning — Indian Patent Application 202541009543 (2025). Data-collection and model-architecture techniques that streamline training and operation of AI-ML positioning models.
• Low Overhead AI-ML based Positioning Enhancements — Indian Patent Application 202441085443 (2024). AI/ML-assisted positioning that cuts the signaling, measurement, and data-collection overhead of positioning operations.
• AI-ML based Positioning Enhancements — Indian Patent Application 202441074530 (2024). AI/ML mechanisms that enhance positioning procedures using information already available within the wireless network.