The privacy issue under the motion sensor-based side channel is a fundamental and critical research topic with many challenges. The existing solutions do not solve some significant problems in practice, for example, the protection mechanism should balance user experience with defensive effectiveness. Moreover, extra settings should not be required. As an effort towards this issue, the common pattern of motion sensor-based side-channel attacks is analyzed, and it finds that the key step of these side-channel attacks is learning the mapping relationship among user behavior, device status, and sensor reading. In addition, a protection method is proposed which applies differential privacy scheme and injects random noise to sensor readings indiscriminately to reduce the effect of learning mapping relationship. This defense method is implemented in system framework, thus it is transparent to both users and attackers. Moreover, the mechanism of proposed defense method is analyzed theoretically to demonstrate how this method decrease the attack success rate and prove that this method can work for any other known and unknown motion sensor side-channel attacks. Finally, the proposed schema is evaluated by conducting experiments against 11 typical motion sensor-based side-channel attacks.