Background Sleep fragmentation is a persistent problem throughout the course of Parkinson's disease (PD). However, the related neurophysiological patterns and the underlying mechanisms remained unclear. Method We recorded subthalamic nucleus (STN) local field potentials (LFPs) using DBS with real-time wireless recording capacity from thirteen PD patients undergoing a one-night polysomnography recording, one month after DBS surgery before initial programming and when the patients were Off-Medication. The STN LFP features that characterized different sleep stages, correlated with arousal and sleep fragmentation index, and preceded stage transitions during N2 and REM sleep were analyzed. Results Both beta and low gamma oscillations in NREM sleep increased with the severity of sleep disturbance (arousal index (ArI)-betaNREM: r=0.9,p=0.0001) and sleep fragmentation index (SFI)-betaNREM: r=0.6,p=0.0301; SFI-gammaNREM: r=0.6,p=0.0324). We next examined the low-to-high power ratio, which was the power ratio of theta oscillations to beta and low gamma oscillations, and found it to be an indicator of sleep fragmentation (ArI-LHPRNREM: r=-0.8,p=0.0053; ArI-LHPRREM: r=-0.6,p=0.0373; SFI-LHPRNREM: r=-0.7,p=0.0204; SFI-LHPRREM: r=-0.6,p=0.0428). In addition, long beta bursts (>0.25s) during NREM stage 2 were found preceding the completion of transition to stages with more cortical activities (towards Wake/N1/REM compared with towards N3 (p<0.01)) and negatively correlated with STN spindles,which were detected in STN LFPs with peak frequency distinguishable from long beta bursts (STN spindle: 11.5Hz, STN long beta bursts: 23.8Hz), in occupation during NREM sleep (beta = -0.24, p<0.001). Conclusion Features of STN LFPs help explain neurophysiological mechanisms underlying sleep fragmentations in PD, which can inform new intervention for sleep dysfunction.