Characterizing network-level rhythmic dynamics over multiple spatio-temporal scales can significantly advance our understanding of brain cognitive function and information processing. In this research, we propose a new switching state space model called latent dynamical coherence model or briefly LDCM. In the LDCM, we develop model inference and parameter estimation solutions that facilitate studying network-level rhythmic dynamics at scales. In the proposed framework, we incorporate both continuous and discrete state processes, helping us to capture dynamics of functional connectivity at various rates, such as slow, rapid, or a combination of both. We then demonstrate an application of our model in characterizing circuit dynamics of the anesthetic state in a sample data set, recorded from a patient under anesthesia using 64-channel EEG over the course of two hours.