In this project, I developed a multitask deep learning model that performs beamforming selection and user localization simultaneously, inspired by Integrated Sensing and Communication (ISAC) for future 6G networks. The model uses a Convolutional Neural Network to extract features from Channel State Information (CSI) and a shared network head to output both the optimal beam index and estimated user position.

Features:

• Multitask Learning: One model performs two tasks at once — beamforming prediction and coordinate localization.
• Weighted Loss Optimization: Allows the network to balance and prioritize both objectives effectively.
• High Antenna Array Support: Designed and tested on up to 64-antenna configurations.

Results:

• Top-1 Beamforming Accuracy: 78.2%
• Top-3 Accuracy: 99.21%
• Localization Error: as low as 2.11 meters

Performance Gain:

• +7% improvement over single-task beamforming models
• 81% reduction in localization error compared to baseline approaches

Impact:

This work demonstrates how multitask learning can significantly improve efficiency and performance in ISAC systems, contributing to practical 6G wireless deployment strategies.