Position sensors, such as the gyroscope, the magnetometer and the accelerometer, are found in a staggering variety of devices, from smartphones and UAVs to autonomous robots. Several works have shown how adversaries can mount spoofing attacks to remotely corrupt or even completely control the outputs of these sensors. With more and more critical applications relying on sensor readings to make important decisions, defending sensors from these attacks is of prime importance.

In this work we present practical software based defenses against attacks on two common types of position sensors, specifically the gyroscope and the magnetometer. We first characterize the sensitivity of these sensors to acoustic and magnetic adversaries. Next, we present two software-only defenses: a machine learning-based single sensor defense, and a sensor fusion defense which makes use of the mathematical relationship between the two sensors. We performed a detailed theoretical analysis of our defenses, and implemented them on a variety of smartphones, as well as on a resource-constrained IoT sensor node. Our defenses do not require any hardware or OS-level modifications, making it possible to use them with existing hardware. Moreover, they provide a high detection accuracy, a short detection time and a reasonable power consumption.

位置传感器，例如陀螺仪，磁力计和加速度计，存在于从智能手机，无人机到自主机器人的各种设备中。几项工作表明，对手如何进行欺骗攻击以远程破坏甚至完全控制这些传感器的输出。随着越来越多的关键应用程序依赖传感器读数来做出重要决策，保护传感器免受这些攻击至关重要。

在这项工作中，我们提出了基于实用软件的防御措施，以抵御对两种常见类型的位置传感器（特别是陀螺仪和磁力计）的攻击。我们首先描述这些传感器对声学和磁性对手的敏感性。接下来，我们提出两种纯软件防御：基于机器学习的单个传感器防御和利用两个传感器之间的数学关系的传感器融合防御。我们对防御措施进行了详细的理论分析，并在各种智能手机以及资源受限的IoT传感器节点上实施了防御措施。我们的防御措施不需要任何硬件或操作系统级别的修改，因此可以将其与现有硬件一起使用。而且，它们提供了高检测精度，短检测时间和合理的功耗。