In practical downlink networks, instantaneous channel state information at the transmitter is often unavailable and it is costly to acquire, frequently delayed, and unreliable in multi-user or multi-target scenarios. This forces ISAC systems to rely instead on statistical channel knowledge, raising a fundamental question: what are the theoretical limits of integrated sensing and communication when the transmitter operates without full Channel State Information? We characterize these limits through an information and estimation-theoric lens, while also addressing a key gap in sensing benchmarks — the classical Cramér–Rao bound is only reliable at high SNR and unreliable in low-SNR regimes. To handle this, we explore the Ziv–Zakai bound for ISAC, which incorporates prior information and remains accurate across all SNR conditions.

The FR3 band (7–24 GHz) is a promising candidate for 6G, offering wide bandwidth; but this very breadth introduces a problem: fragmented band, propagation loss, antenna aperture, and hardware behavior vary significantly across the band, making conventional single-band MIMO architectures ineffective. Sensing and communication signals must also coexist with incumbent services already occupying parts of this spectrum, further constraining system design. This project develops a framework that exploits the fragmented, frequency-partitioned structure of FR3 rather than fighting it, using reconfigurable antenna architectures to adaptively reshape beam patterns and enable efficient multi-band operation within a single hardware platform.

Classical subspace methods like MUSIC give robust direction-of-arrival estimates but break down at low SNR and with few snapshots — exactly the regime ISAC systems must operate in. Rather than throwing data-hungry learning at the problem, this project develops hybrid schemes whose decisions are grounded in the geometric structure of the signal subspace rather than learned purely from data. The aim is sample-efficient estimators that recover performance from very limited, noisy measurements while preserving the interpretability of classical methods.