Certified Dynamic Subspace Optimization for Bayesian ISAC Design
Full-space covariance optimization is costly in large-array Bayesian integrated sensing and communication (ISAC). A fixed transmit subspace can reduce this cost, but the resulting optimality loss depends on the posterior distribution. In this paper, a certified dynamic subspace method is developed for angle Bayesian Cramér-Rao bound (BCRB) minimization under joint angle and complex-gain uncertainty. For uniform linear arrays, exact response preservation over a nondegenerate posterior angular interval requires the full antenna space. Subspace sufficiency is therefore assessed through full-space objective-gap bounds. A communication-feasible basis is initialized from the user-channel span and local BCRB sensitivity. Restricted semidefinite programs are then solved. Primal restoration and dual repair yield gap bounds that remain valid after lossless user-beam recovery. When an accurate restricted solve does not certify the requested tolerance, the basis is expanded using a direction identified from the covariance dual slacks. Under a strict communication-power margin, joint uniqueness of the restricted optimal dual multipliers is established. With exact restricted solves, this uniqueness guarantees strict BCRB improvement after certificate-driven expansion. Numerical results validate the bounds against full-space references for the sampled problems. Compared with a fixed Fisher ordering, the proposed method can reach the same certificate target with fewer dimensions. On large arrays, total runtime is reduced relative to the full-space and factorized baselines at matched certificate targets.
Authors
- Cheng-Jie Zhao (ORCID: https://orcid.org/0000-0003-0852-571X)
- Zongyao Zhao
- Yuanwei Liu
- Zhaolin Wang
- Qingjiang Shi
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-12
- DOI
- https://doi.org/10.48550/arxiv.2611.16624
- Primary Topic
- Direction-of-Arrival Estimation Techniques
- Type
- preprint