Cramér-Rao Bound Optimization for Bistatic ISAC: Transceiver Design and Attention-Based ISACNet

• Published in: IEEE journal on selected areas in communications p. 1
• Main Authors: Mao, Weihao, Lu, Yang, Pan, Gaofeng, An, Jianping, Ai, Bo, Ng, Derrick Wing Kwan
• Format: Journal Article
• Language: English
• Published: IEEE 08.09.2025
• Subjects: Array signal processing; Bistatic ISAC; Covariance matrices; CRB; Inference algorithms; Integrated sensing and communication; ISACNet; joint transmit and receive beamforming design; Performance metrics; Radar; Signal processing algorithms; Signal to noise ratio; Uplink; Vectors
• ISSN: 0733-8716; 1558-0008
• DOI: 10.1109/JSAC.2025.3607866

This paper investigates the joint transmit and receive beamforming design for a bistatic integrated sensing and communication (ISAC) system, where a transmit base station (BS) and a receive BS are coordinated to simultaneously serve multiple downlink and uplink users as well as estimate target positions. The closed-form expression for the Cramér-Rao bound (CRB) for target positions and reflection coefficients is derived and minimized subject to constraints of the transmit power budget and communication requirements. To address the considered problem, the closed-form expression for the optimal receive beamforming vectors is derived, facilitating the development of a successive convex approximation (SCA)-based algorithm for optimizing the transmit information beamforming vectors and sensing covariance matrix. In addition, a learning-based approach named ISACNet, trained in an unsupervised manner, is proposed to handle the considered problem. The ISACNet incorporates multi-head self-attention and cross-attention mechanisms to significantly enhance its expressive capability. Simulations validate the effectiveness of the proposed SCA-based algorithm and ISACNet. It is observed that the sensing performance is predominantly affected by the downlink communication more than the uplink communication. Furthermore, our ISACNet generates an effective solution in millisecond-level response times with only a marginal performance degradation compared to the SCA-based algorithm.