Continuous blood pressure (BP) monitoring is essential for cardiovascular healthcare. While traditional cuff-based measurements are unsuitable for continuous tracking, emerging cuffless wearable sensors (e.g., PPG, bioimpedance) provide viable alternatives. However, existing estimation methods often overlook the inherent multi-periodicity and temporal dependencies of physiological signals. Furthermore, personalized modeling is severely hindered by the scarcity of subject-specific data. To address these challenges, we propose a novel Physics-Informed Temporal Network (PITN) integrated with adversarial contrastive learning for precise BP estimation across three modalities: bioimpedance, PPG, and millimeter-wave. Specifically, we first introduce the PITN to explicitly model the multi-periodicity and temporal variations of BP dynamics governed by cardiovascular cycles. To tackle data scarcity, we employ adversarial training to generate realistic physiological time series, enhancing model robustness. Additionally, we utilize contrastive learning to capture discriminative variations, aggregating signals with similar BP values in the latent space while separating dissimilar ones. Extensive experiments on three public datasets demonstrate the superiority and effectiveness of our proposed method over state-of-the-art approaches.