Deep Learning for High-Resolution Sensing Parameter Estimation in Bistatic ISAC
Published in IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), 2026
Authors: Salmane Naoumi, Kevin Mutuwira, Roberto Bomfin, Ahmad Bazzi, Marwa Chafii
Keywords: Deep Learning, Integrated Sensing and Communications, ISAC, Bistatic ISAC, High-Resolution Estimation, Parameter Estimation, Channel State Information, CSI, Angle of Arrival, AoA, Angle of Departure, AoD, Time of Arrival, Cramér-Rao Lower Bound
Key Takeaways
Develops deep learning architectures for high-resolution sensing parameter estimation (AoA, AoD, delay) in bistatic ISAC systems using communication CSI, achieving near-CRLB performance with orders-of-magnitude reduction in execution latency.
Abstract
Key Highlights & Quantitative Findings
- Real-Time Bistatic Sensing: Bypasses computationally expensive iterative parameter-scanning algorithms (such as 2D-MUSIC or maximum likelihood) by mapping communication channel state information (CSI) directly to target parameters using deep learning.
- Joint Super-Resolution Estimation: Estimates Angle of Arrival (AoA), Angle of Departure (AoD), and Time of Arrival (ToA) simultaneously with resolution exceeding the Fourier limit.
- Near-Optimal Estimation Accuracy: Numerical simulations show estimation variance approaching the theoretical Cramér-Rao Lower Bound (CRLB) even in low-to-moderate SNR regimes, with orders-of-magnitude faster online inference suitable for real-time 6G networks.
Abstract
Bistatic integrated sensing and communication (ISAC) offers substantial coverage and interference advantages over monostatic architectures by physically separating the transmitter and receiver. However, extracting high-resolution sensing parameters—such as angle of arrival (AoA), angle of departure (AoD), and propagation delay—from communication-centric reference symbols presents severe computational challenges under low signal-to-noise ratio (SNR) conditions and real-time latency constraints. In this paper, we propose a deep learning framework for high-resolution sensing parameter estimation in bistatic ISAC systems. By processing complex-valued channel state information (CSI) acquired from standard communication waveforms, our neural network architecture learns the underlying spatial-temporal multipath representations, bypassing the computational bottleneck of high-resolution parametric scanning methods (e.g., 2D-MUSIC or maximum likelihood). Extensive numerical simulations demonstrate that the proposed deep learning approach achieves super-resolution parameter estimation accuracy approaching theoretical Cramér-Rao lower bounds (CRLB) with orders-of-magnitude reduction in online execution time, enabling real-time sensing in 6G bistatic networks.
Recommended citation: S. Naoumi, K. Mutuwira, R. Bomfin, A. Bazzi, and M. Chafii, "Deep Learning for High-Resolution Sensing Parameter Estimation in Bistatic ISAC," accepted to IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN), 2026.
Show BibTeX
@article{naoumi2026deep,
title = {Deep Learning for High-Resolution Sensing Parameter Estimation in Bistatic ISAC},
author = {Salmane Naoumi and Kevin Mutuwira and Roberto Bomfin and Ahmad Bazzi and Marwa Chafii},
journal = {IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN)},
year = {2026},
month = {may},
publisher = {IEEE},
url = {https://therealbazzi.github.io/publication/deep-learning-high-resolution-sensing-bistatic-isac-icmlcn26},
}