Sinusoidal-assisted synchrosqueezing transform: Algorithms and biomedical applications

Abstract

Intermittent oscillation signals (IOSs) exist widely in biomedical systems. They are always contaminated by various randomnesses like noise and artifacts. Compared with traditional time–frequency analysis (TFA) methods, adaptive TFA methods show significant advantages in separating IOSs. However, these methods are implemented by assuming that the input signal is in a weakly noisy environment and have narrow-band limitation for extracting IOSs. Besides, it is difficult for these methods to alleviate the noise in the gap between time where IOSs happen. Therefore, in this paper, a new method called the sinusoidal-assisted synchrosqueezing transform (SASST) is proposed for separating IOSs in strongly noisy environments. The SASST method locates the IOSs with the aid of a sinusoidal signal. Such a sinusoidal signal fills the energy blank in the gap between time where IOSs happen. This enables the ridge detection approach, which is the necessary tool for mode retrieval of SST-based methods, to focus on the assisted sinusoidal signal rather than noise. Thus, the background noise in the gap is suppressed. Moreover, this sinusoidal-assisted framework can be introduced into any SST-based TFA method. Depending on the characteristics of different biomedical signals, different SST-based methods are chosen to achieve better separation. As a result, SASST is capable of extracting both narrow- and wide-band IOSs in strongly noisy environments. The effectiveness and advantages of the proposed algorithm are verified by both numerical simulations and real-world cases. Even in a strongly noisy environment of −10 dB, SASST achieves output SNR of 12.66 dB and 10.53 dB in narrow- and wide-band simulation, respectively.