In modern connected healthcare applications, wearable devices supporting real-time monitoring and diagnosis have become mainstream. However, wearable systems are exposed to massive cyberattacks that threaten not only data security but also human safety and life. One of the fundamental security threats is device impersonation. We therefore propose PHY-IDS; a lightweight real-time detection system that captures spoofing attacks leveraging on body motions. Our system utilizes time series of physical layer features and builds on the fact that it is non-trivial to inject malicious frames that are indistinguishable with legitimate ones. With the help of statistical learning, our system characterizes the signal behavior and flags deviations as anomalies. We experimentally evaluate PHY-IDS's performance using bodyworn devices in real attack scenarios. For four types of attackers with increasing knowledge of the deployed detection system, the results show that PHY-IDS detects naive attackers with high accuracy above 99.8\% and maintains good accuracy for stronger attackers at a range from 81.0% to 98.9%.