Robust Tensor Decomposition-Based Parameters Estimation Algorithm for Coherent Targets in Regular-Resource OFDM-ISAC Systems

Abstract

High-precision parameter estimation in regular-resource OFDM-based integrated sensing and communication (ISAC) systems is severely hindered by multipath coherence, which induces rank deficiency and neutralizes conventional methods. To overcome this, we propose a Tensor-based Robust Algorithm for Coherent Targets Estimation (TRACE). First, a dual-domain spatial-frequency smoothing and conjugate-reversal mechanism breaks the phase binding of coherent echoes to restore the effective signal rank. This enables a customized eigenvalue-variance metric to robustly detect the target count under low SNR and power mismatches. To avoid exhaustive grid searches, a non-iterative pre-estimation phase leverages shift-invariance and polynomial rooting for high-fidelity tensor initialization. These factors are then refined via an accelerated alternating least squares (ALS) approach, which exploits Khatri-Rao properties for dimensionality reduction and continuously enforces physical Vandermonde constraints via an inner gradient descent loop integrated with a rank-1 Hankel manifold projection. Final parameters are extracted using a combination of total least squares (TLS) and robust polynomial root-finding. Extensive simulations validate that TRACE effectively decouples coherent targets, achieving an optimal balance between super-resolution accuracy and computational efficiency.