Learning-driven Adaptive Stochastic Resonance for Smoothing Time-frequency Ridges in Nonstationary Millimeter-wave Radar Signals

Addressing the issues of jumps, breaks, and drifts in time-frequency ridges from nonstationary millimeter-wave radar signals under noisy conditions, there is an urgent need to improve continuity and smoothness. This paper investigates a learning-driven adaptive stochastic resonance algorithm that achieves adaptive self-tuning by incorporating Hebbian learning theory into the stochastic resonance system. Experiments on drive shaft measurement using a millimeter-wave radar across four different variable-frequency conditions are designed, and the results of different methods are evaluated with the indicators of the continuity index and smoothness index. Quantitative results show that the learning-driven adaptive stochastic resonance dramatically improves average continuity index and smoothness index from 0.02 and 0.23 in the original signals to 0.984 and 0.977, respectively. Moreover, it outperforms time-varying scale adaptive stochastic resonance by 15–20% in both metrics. The findings demonstrate that the learning-driven adaptive stochastic resonance yields more continuous and smoother ridges with stronger robustness and adaptability. This work provides an effective solution for robust ridge extraction in non-contact vibration detection.