Cybersecurity Challenges in UAV Systems: IEMI Attacks Targeting Inertial Measurement Units

Abstract

The rapid expansion in unmanned aerial vehicles (UAVs) across various sectors, such as surveillance, agriculture, disaster management, and infrastructure inspection, highlights the growing need for robust navigation systems. However, this growth also exposes critical vulnerabilities, particularly in UAV package delivery operations, where intentional electromagnetic interference (IEMI) poses significant security and safety threats. This paper addresses IEMI attacks targeting inertial measurement units (IMUs) in UAVs, focusing on their susceptibility to medium-power electromagnetic interference. Our approach combines a comprehensive literature review and QuickField simulation with experimental validation using a commercially available 6-degree-of-freedom (DOF) IMU sensor. We propose a hardware-based electromagnetic shielding solution using mu-metal to mitigate IEMI’s impact on sensor performance. The study combines experimental testing with simulations to evaluate the shielding effectiveness under controlled conditions. The results of the measurements showed that medium-power IEMI significantly distorted IMU sensor readings, but our proposed shielding method effectively reduces the impact, improving sensor reliability. We demonstrate the mechanisms by which medium-power IEMI disrupts sensor operation, offering insights for future research directions. These findings also highlight the importance of integrating hardware-based shielding solutions to safeguard UAV systems against electromagnetic threats.

1. Introduction

In the rapidly evolving domain of unmanned aerial vehicles (UAVs), inertial measurement units (IMUs) play a pivotal role among the most versatile sensors with a wide range of potential use cases. IMUs are not just essential for navigation, data collection, and guidance in various applications. They also extend beyond mere flight control to include advanced functions such as attitude and heading reference systems (AHRSs), inertial navigation systems (INSs), and precise georeferencing for photogrammetry and LiDAR mapping. This variety underscores the critical importance of choosing the appropriate IMU technology and grade to meet the specific demands of a UAV mission, such as high-accuracy navigation, inspection, environmental monitoring, and package delivery. However, the widespread adoption of UAVs has also exposed them to new vulnerabilities, particularly cybersecurity threats such as intentional electromagnetic interference (IEMI) attacks. These attacks exploit the susceptibility of electronic components like the IMUs to high-power or low-power electromagnetic energy sources, posing significant threats to the security and safety of UAV operations, especially in package delivery applications. This research aims to provide practical insights into the vulnerabilities against medium-power IEMI attacks that may not completely destroy the IMU like high-power IEMI or have little to no effect like low-power IEMI, thereby enhancing the cybersecurity and safety of UAV operations in real-world applications.

The advancement of microelectromechanical system (MEMS) technology has led to the development of smaller, more affordable IMUs, broadening their use across different sectors, where they are also integrated into wearable devices like smartwatches for applications such as ergonomic risk assessment during manual tasks [1] and real-time biofeedback systems [2]. This versatility underlines the critical role IMUs play in measuring kinematic parameters across various applications, making them indispensable in the modern technological landscape.

IMUs in UAVs have become increasingly common due to their pivotal role in improving navigation accuracy and reliability. IMUs are vital components in UAVs for estimating the system state, offering high accuracy, fast update frequencies, and compact sizes [3]. IMUs prove particularly valuable in environments where global positioning system (GPS) signals may be unreliable or inaccessible, such as dense urban canyons or indoor settings [4]. The integration of IMUs with global navigation satellite systems (GNSSs) in UAV navigation algorithms has also been explored to enhance precision and robustness, highlighting the versatility and adaptability of IMUs in various UAV applications [5].

While they are a critical component of UAVs, these IMU sensors, if attacked by IEMI, can fail and have catastrophic consequences. IEMI attack can interfere with the sensor’s ability to accurately measure orientation and movement, causing deviations in the data. On the other hand, radio-frequency interference (RFI) can disrupt the communication between the IMU and other systems, potentially corrupting data transmission and reception. By understanding and mitigating these factors, the reliability and effectiveness of IMU sensors in UAVs can be enhanced, ensuring precise navigation and control in various operational scenarios.

Our study explores the impact of IEMI attacks on IMU sensors. Following an extensive review of the existing literature, we experimented with custom-built tools to detect these effects directly within a controlled laboratory setting. To facilitate this investigation, we specifically engineered a custom medium-power electromagnetic field (EMF) source for this research. In addition, we used a widely used commercially available IMU sensor for the experiment.

The structure of this paper is as follows. Section 2 provides a comprehensive review of related work on IEMI attacks and their impact on IMUs in UAV systems. Section 3 describes the materials and methods used in this study, including the experimental setup and the operational principles of the IMU sensor. Section 4 presents the results of the experiments, analyzing the baseline measurements and the effects of IEMI attacks at different distances and orientations. Section 5 discusses the proposed mitigation method, focusing on the electromagnetic shielding solution developed to counter IEMI attacks. Finally, Section 6 summarizes the key conclusions and offers suggestions for future research directions in enhancing cybersecurity for UAV navigation systems.

2. Related Work

As the deployment of UAVs continues to expand, so does the spectrum of vulnerabilities they face, particularly cybersecurity threats. Among these, IEMI attacks capable of exploiting the electronic susceptibilities of UAVs, with a particular focus on the IMU, have emerged as a significant concern. This section discusses the related work concerning IEMI attacks on UAVs, with a specific emphasis on the vulnerabilities of IMUs, and outlines the limitations of current research in this area.

Unlike classical cyberattacks such as deception and denial-of-service (DoS) attacks, which exploit software vulnerabilities, IEMI attacks represent a form of electronic warfare. They target the electronic components of a drone, especially the IMU, to induce malfunctions or catastrophic failures. These attacks do not require access to the system’s software, making them particularly dangerous. However, IEMI and cyberattacks could be combined, where deception attacks exploit IEMI-induced sensor errors to manipulate control systems.

The literature reveals a growing awareness of the potential for IEMI to disrupt UAV operations, with studies demonstrating how high-power electromagnetic pulses (EMPs) can permanently damage UAV electronic systems.

In [6], the authors developed a method to access a UAV microcontroller and performed a high-power IEMI test. They exposed it to continuous wave signals ranging from 100 MHz to 2 GHz. The results showed significant sensor distortions at field strengths above 55 V/m.

In [7], researchers examined the effects of high-power EMPs between 100 MHz and 3.4 GHz on a commercial drone. They developed a real-time monitoring system and used a ramp signal to increase pulse amplitude linearly. The findings showed that specific field strengths directly impacted the magnetometer, while the accelerometer and gyroscope were affected by motor speed vibrations. This confirms the need for an internal system to monitor drone sensors against such disruptions.

In [8], the authors explored the susceptibility of drones with MEMS gyroscopes to sound-induced disruptions. They tested 15 MEMS gyroscopes, identifying resonant frequencies in 7 that significantly affected their performance. Real-world tests and simulations revealed that drones with these vulnerable gyroscopes lost control and crashed in all trials when exposed to targeted sound noise. The findings underscore the importance of developing countermeasures against such cybersecurity threats to drone operations.

The study in [9] presents sensor deprivation attacks (SDAs) on UAVs, leveraging IEMI to intermittently reconfigure sensors rather than continuously spoofing signals. By altering sensor update rates or suspending feedback, SDAs introduce erroneous data into the extended Kalman filter, destabilizing flight operations. Unlike traditional attacks, SDAs require minimal interference and retain their impact even if the attacker is out of range. Testing on commercial UAVs showed that SDAs evade standard anomaly detection, underscoring the need for enhanced protections. However, SDA effectiveness can vary with sensor models and their control over outputs remains limited, often leading to unpredictable deviations.

The authors of [10] demonstrated how low-power EMI can disrupt the communication channel between a drone’s IMU and control unit, corrupting sensor data and causing crashes. By targeting susceptible frequencies specific to the control board, the attack simultaneously distorted the gyroscope, accelerometer, and magnetometer outputs, propagating instability through the control algorithm. The research underscores the vulnerability of sensor communication channels and the need for mitigation strategies like electromagnetic shielding.

While IEMI attacks target hardware vulnerabilities in UAV systems, communication delays also pose a significant challenge by reducing the system’s ability to respond to threats. Recent studies, such as [11], have proposed control strategies to maintain system stability despite these delays. Applying such strategies to UAVs could enhance resilience against both communication delays and IEMI attacks.

Table 1. Research on IEMI attacks.

Type of IEMI	Method	Target	Signal Type/Strength	Reference
Non-RF interference	Acoustic	MEMS Accelerometer	Speaker 2–30 kHz	[8]
		MEMS gyroscope	Bluetooth speaker Distance: 10 cm SPL: 113 dB	[12]
Low-power IEMI	Radiated EMI	Drone	Continuous wave 470–862 MHz 1.4–2.7 GHz 10 V/m	[13]
	GPS spoofing		GNSS emulator; emulated GLONASS	[14]
	RC jamming		Sweep jamming	[15]
	GPS jamming		5 types of jamming	[16]
	Sensor deprivation	UAV Controller (IMU, Kalman Filter)	Intermittent IEMI enabling sensor reconfiguration	[9]
	Antenna-based EMI injection	Drone’s IMU (all sensors simultaneously)	EMI targeting sensor-control communication, 10–100 W	[10]
High-power IEMI	Horn antenna	Quadcopter	UWB EMP FOM = E (V/m) × R (m) (1.6 kV, 3.5 kV, 49.5 kV)	[7]
	UWB EMP Radiator (4 TEM horns)	Drone	UWB EMP FOM = E (V/m) × R (m) (1.6 kV, 3.5 kV, 49.5 kV)	[17]
	Horn coil	Sensor network	Continuous wave 2–3 GHz Field peak: 0.24–0.36 kV/m	[18]
Mid-power IEMI	Antenna coil	Drone’s IMU	Transient wave 53 dB ≥2000 uT	This work

provides an overview summarizing the types of IEMI investigated, the methods employed, the targets, and the characteristics of the electromagnetic signals utilized in these research efforts.

Despite the increasing recognition of the threat IEMI poses, there remains a significant gap in the literature regarding comprehensive defense mechanisms. In addition, past research on IEMI primarily focused on the effects of high-power and low-power EMI, which involve using either very strong or weak electromagnetic signals [19]. Moreover, these studies often consider the UAV as a whole rather than isolating the effects on individual IMU sensors, leaving a gap in understanding how IEMI attacks directly impact these critical components.

In this paper, our experiment attempts to reduce that gap in the literature. We incorporate an IEMI attack that emits controllable magnetic fields to assess the impact on the IMU sensor-only readings. The experiment was conducted where the EMF antenna and the IMU sensor were stationary. The results from these experiments are then discussed in detail to show the impact of these low-power IEMI attacks on the IMU sensor readings and how they would affect UAV operations. Moreover, we contribute to closing this gap by introducing a shielding method designed to safeguard IMUs from IEMI attacks effectively.

3. Materials and Methods

IMUs are available in configurations such as 6 degrees of freedom (DOF) and 9 DOF, with the former consisting of a 3-axis accelerometer and gyroscope. These sensors are critical for navigation, particularly in environments where GPS signals are unreliable. Our work focuses on a 6-DOF IMU, as shown in Figure 1, commonly used in UAV applications to evaluate the impact of IEMI attacks.

3.1. Accelerometers

Accelerometers operate through an electromechanical sensor that detects static and dynamic acceleration forms. Static acceleration encompasses consistent forces exerted on an object, such as friction or gravity, which are largely predictable and uniform. For instance, the force of gravity is a constant 9.8 m/s², with gravitational pull being nearly identical across the Earth’s surface. In contrast, dynamic acceleration involves irregular forces, such as those produced by vibrations or impacts. A vivid illustration of dynamic acceleration is the abrupt change in velocity experienced during a vehicle collision. Accelerometers are designed to sense these variations and translate them into quantifiable data, typically in the form of electrical signals. This allows for precise monitoring and analysis of movement in various applications, from automotive systems to smartphones and aerospace technologies.

It is important to acknowledge that acceleration induces a force, which is detected by the force-detection mechanism intrinsic to the accelerometer. Thus, the accelerometer’s operational principle is based on directly measuring force rather than acceleration. It quantifies acceleration indirectly by applying force along one of its axes. Moreover, an accelerometer is an electromechanical apparatus comprising an intricate assembly of apertures, cavities, springs, and conduits crafted using microfabrication techniques. The construction of accelerometers involves a multilayer wafer process, where the quantification of acceleration forces is accomplished by observing the displacement of a mass in relation to stationary electrodes.

Accelerometers are integral in various fields, such as navigation, seismic sensing, and motion tracking. They are commonly used in INSs to estimate position and velocity, especially in environments where GPS signals may be unreliable. Their role in wearable technologies for health monitoring and human motion analysis is also growing, particularly for tracking posture and fall detection.

Despite their broad applications, accelerometers have inherent limitations. Sensitivity to noise and temperature fluctuations can affect their precision, especially in environments with high levels of mechanical vibrations. In addition, accelerometers may drift over time, resulting in cumulative errors in velocity or displacement when used in standalone inertial navigation systems. This is typically mitigated by combining them with other sensors, such as gyroscopes and magnetometers.

Referring to Figure 2, a reference model of an accelerometer, the vector $R$ represents the force vector that the accelerometer measures, whereas R_x, R_y, and R_z denote the projections of the vector onto the $X$, $Y$, and $Z$ axes, respectively.

The interrelationship among R, $R_x$, R_y, and $R_z$ is quantified through Equation (1). This equation facilitates the mathematical description of how the magnitude of the force vector $R$ is decomposed into its constituent components along the three orthogonal axes. Thus, it provides a framework for analyzing the directional intensities of the force experienced by the accelerometer, thereby offering insight into the device’s orientation and movement within 3-dimensional space. This decomposition is fundamental in inertial navigation and motion sensing, as it allows for precise calculations of orientation and acceleration by interpreting the accelerometer’s output.

$$R^2=R_x^2+R_y^2+R_z^2$$

(1)

$$R=\sqrt{R_x^2+R_y^2+R_z^2}$$

(2)

Considering Equation (2), Pythagoras’s theorem in 3-dimensional space, the angle of tilting of each axis can be calculated using Equations (3)–(5).

$$a_{xr}=arcos\left({\displaystyle \frac{R_x}{R}}\right)$$

(3)

$$a_{yr}=arcos\left({\displaystyle \frac{R_y}{R}}\right)$$

(4)

$$a_{zr}=arcos\left({\displaystyle \frac{R_z}{R}}\right)$$

(5)

EMI can introduce noise into the accelerometer’s output, particularly at medium power. In a static state, the accelerometer is not subjected to any motion or external forces apart from gravity. The expected output should primarily reflect gravitational acceleration, which acts along the vertical axis:

$$a_{static}=\left[\begin{array}{c}0\\ 0\\ 9.8\mathrm{m}/\mathrm{s}²\end{array}\right]$$

(6)

However, accelerometer measurements are also affected by bias and sensor noise. Thus, the actual output can be written as [21]:

$${\widehat{a}}_{static}=a_{static}+b+\eta$$

(7)

where $b$ is the sensor bias (a small, consistent error) and η is the sensor noise (modeled as a Gaussian process with mean zero and variance ${\sigma }^2$). When an IEMI attack occurs, the accelerometer is subjected to strong EMFs, introducing additional distortions into the measurement. The IEMI affects the sensor’s electronic circuitry, causing changes in bias and increasing the noise levels. The altered measurements can be expressed as:

$${\widehat{a}}_{static}^{IEMI}=\left[\begin{array}{c}{a}_x^{\prime }\\ a_y^{\prime }\\ 9.8{\displaystyle \frac{m}{s^2}}+a_z^{\prime }\end{array}\right]$$

(8)

where $a_x^{\prime },a_y^{\prime },{\mathrm{a}\mathrm{n}\mathrm{d}a}_z^{\prime }$ are the false readings caused by IEMI. IEMI-induced errors can propagate through velocity and position calculations. Integrating erroneous acceleration values over time results in larger errors in velocity and position:

$$v(t)={\int }_0^t\widehat{a}\left(\tau \right)d\tau$$

(9)

$$p(t)={\int }_0^{t}v\left(\tau \right)d\tau$$

(10)

In this case, false acceleration readings introduced by IEMI lead to incorrect velocity $v\left(t\right)$ and position $p\left(t\right)$ estimates, which can significantly disrupt the UAV’s navigation and control systems. These errors can lead to incorrect assumptions about the UAV’s orientation or motion, potentially causing the system to misinterpret its position or velocity.

3.2. Gyroscopes

Gyroscopes are devices that measure or maintain orientation and angular velocity. Unlike traditional mechanical gyroscopes, MEMS gyroscopes are fabricated using IC batch processing techniques, significantly reducing their size and cost and making them indispensable components in various applications. They operate on the principle of the Coriolis effect, where displacement is measured due to forces exerted on a vibrating element when the device rotates. This displacement is then converted into electrical signals, which can be interpreted to determine angular velocity.

$$A_{xz}=\mathrm{a}\mathrm{r}\mathrm{c}\mathrm{t}\mathrm{a}\mathrm{n}2(R_x,R_z)$$

(11)

$$A_{yz}=\mathrm{a}\mathrm{r}\mathrm{c}\mathrm{t}\mathrm{a}\mathrm{n}2(R_y,R_z)$$

(12)

Figure 3 illustrates a gyroscope reference model, adapted from an accelerometer model to demonstrate what the gyroscope measures in this context. In this model, $R_{xz}$ and $R_{yz}$ represent the projections of the inertial force vector $R$ onto the $XZ$ and $YZ$ planes, respectively. $A_{xz}$ is the angle formed between the $R_{xz}$ projection (on the $XZ$ plane) and the $Z$-axis, while $A_{yz}$ is the angle between the $R_{yz}$ projection (on the $YZ$ plane) and the $Z$-axis. These vectors create a triangular configuration, allowing the calculation of angles using trigonometry principles. Utilizing this model enables the computation of the rotation rate angle, providing valuable insights into the object’s angular motion.

The gyroscopic measurement is based on the following key equation that governs the Coriolis force:

$$F_c=-2m\left(v\times \Omega \right)$$

(13)

where $F_c$ is the Coriolis force experienced by the vibrating mass, $m$ is the mass of the vibrating element, $v$ is the velocity of the vibrating mass, and $\Omega$ is the angular velocity of the rotating reference frame.

During an IEMI attack, the Lorentz force caused by the EMF can add another term to the force equation. If a time-varying magnetic field is present due to IEMI, an additional force is introduced [21]:

$$F_L=q\left(E+v\times B\right)$$

(14)

where $F_L$ is the Lorentz force, $q$ is the charge on the vibrating element, $E$ is the electric field due to IEMI, and $B$ is the magnetic field due to IEMI. The total force acting on the vibrating element in a gyroscope under the influence of IEMI becomes:

$$F_{total}=-2m\left(v\times \Omega \right)+q\left(E+v\times B\right)$$

(15)

This combined force alters the gyroscope’s readings by adding disturbances to the rotational velocity measurement. The exact influence depends on the magnitude of $E$ and $B$ and the interference in the system’s signal processing.

In the static case, the gyroscope is not experiencing any rotational motion. This means that in normal operation (without interference), the output angular velocity should be zero, which means the Coriolis force is also zero. However, under an IEMI attack, the induced Lorentz force due to the EMF $B\left(t\right)$ will disturb the system, generating a false signal even though the IMU is stationary. Thus, the output angular velocity during the IEMI attack would be corrupted by the error introduced by the interference:

$${\Omega }_{error,static}={\displaystyle \frac{q\left[E\left(t\right)+v_{x}B\left(t\right)\right]}{2m{v}_x}}$$

(16)

One of the methods to mitigate the effect of IEMI on gyroscopes is to introduce electromagnetic shielding around the IMU system. This shielding can be modeled using Maxwell’s equations to reduce the influence of $\overline{E}$ and $\overline{B}$ fields on the system. Magnetic shielding involves using materials with high permeability, such as mu-metal or mu-ferro. It redirects flux lines around sensitive instruments like IMUs, ensuring their safety and operational stability, as shown in Figure 4 [22]. However, if the magnetic field is too strong, a single high-permeability shield may saturate, losing its ability to absorb flux. In such cases, a shielding design of two stages or more is employed. This combination provides comprehensive protection from intense magnetic bursts and residual interference, achieving high magnetic attenuation ratios (e.g., 500×, 1000×, or more).

3.3. Experiment Setup

In this study, we designed a specialized EMF antenna to create the magnetic fields necessary for our experiments. As shown in Figure 5, this antenna features a core enveloped in a copper coil capable of producing magnetic fields up to and beyond 2000 µT at different distances from the IMU, surpassing the requirements to generate a medium IMEI attack. Central to the design of our EMF antenna is a custom-built loop antenna measuring 13.7 cm, devised specifically to simulate magnetic conditions. The generator is powered by a Chroma Programmable AC Source, model 61601, capable of supplying a continuous transient current of up to 8 A. This mode facilitates a current boost to as much as 8 A and a voltage of 51.2 $V_{rms}$, yielding a power output of 204.4 watts at a power factor of 1.

The field strength was approximately 2000 µT at the closest distance to the IMU, with a frequency range of 50 Hz to 60 Hz. The waveform was a continuous sinusoidal signal, with variations introduced by the distance between the antenna and the IMU sensor. This setup was intended to simulate the type of EMI encountered in real-world scenarios, such as low-EMP attacks or unintentional interference from nearby power lines [23].

In our study, we employed an MPU−6050 [24], a commercially available IMU sensor with 6 DOF. The sensor combines a 3-axis gyroscope and a 3-axis accelerometer in the same package, augmented by an onboard digital motion processor (DMP) that executes 6-axis motion fusion algorithms. This configuration enables the device to interface with external magnetometers or additional sensors via an auxiliary master I²C bus. This facilitates the autonomous collection of a comprehensive dataset from various sensors, eliminating the need for direct oversight by the system’s central processor. The sensor specifications are listed in

Table 2. MPU−6050 sensor specifications.

IMU Model	Gyro. Range (dps)	Gyro. Sensitivity (LSB/°/sec)	Gyro. Rate Noise (mdps/rtHz)	Accel. Range (g)	Accel. Sensitivity (LSB/g)	Digital Output
MPU−6050	±250	131	0.005	0.005	16384	I2C
	±500	65.5		0.005	8192
	±1000	32.8		0.005	4096
	±2000	16.4		0.005	2048

. To ensure accurate monitoring of movements across a broad velocity spectrum, the components are equipped with a customizable gyroscopic sensitivity scale spanning ±250, ±500, ±1000, and ±2000 degrees per second alongside an adjustable accelerometer sensitivity range of ±2 g, ±4 g, ±8 g, and ±16 g. Furthermore, it includes an integrated temperature sensor and an on-chip oscillator, exhibiting a maximum variance of ±1% across the operational temperature spectrum. The IMU was mounted on a stable platform to ensure that it stayed stationary throughout the experiment. By eliminating movement, we were able to isolate the effects of EMI on the sensor’s output without any additional noise introduced by any physical motion. The IMU was positioned at a fixed distance from the EMF antenna, with the distance varied in subsequent tests to measure the effect of proximity on the interference strength.

An Arduino Uno was used to connect the IMU and collect sensor data. The sensor was connected to the Arduino over the I²C protocol using the QWiic interface [25], a simplified I²C protocol. Figure 6a shows the IMU attack experiment setup, while Figure 6b shows the I²C connection to the sensor over the Qwiic interface. This connection includes power (3.3 Vdc), ground, and the data (SDA) and clock (SCL) lines required for I²C communication. The Arduino was programmed to read the sensor data over I²C every 10 ms. The data were then sent over the serial interface for collection and analysis. As our goal was to monitor the effects of magnetic fields on the sensor readings, we used minimal code to reduce any overhead created by reading and processing data and avoided using data post-processing functions available in the IMU driver library or any post-filtering.

4. Results

4.1. Baseline Measurement

Before activating an IEMF attack, a baseline measurement was performed with the IMU in its stationary position. This allowed us to record the expected output under normal conditions. These tests evaluated the baseline error and variability between multiple separate but identically configured tests. This baseline allowed us to better understand the inherent error in our setup, which would be considered in later experiments involving EMF interference. Any deviations from these values under baseline conditions would indicate potential sensor drift or noise. Figure 7 presents the accelerometer and gyroscope readings for test 1 and test 2, showing the comparison of results derived from these tests. However, the figure does not represent the expected output during dynamic flight conditions, such as when a drone drops altitude. In such cases, accelerometer readings, particularly on the $Z$-axis, would change to reflect the movement, while the gyroscope would register angular velocity depending on the drone’s rotation.

To ensure the reliability of the comparison between the tests, all equipment was powered down after completing test 1 and then restarted before conducting test 2. This minimized potential residual effects that could have influenced the results of test 2. Figure 8 shows the percentage error between the two accelerometer and gyroscope reading tests.

Due to the sensor’s lack of prior calibration, the error margin between the two independent tests varied significantly. The error ranged from −6% to 14% in the $X$-direction, while errors in the $Y$ and $Z$ directions were much smaller, between −1.5% and 1.5%. Ideally, the error should approach zero in all directions, with the $Z$-direction showing a steady gravitational acceleration of approximately 9.8 m/s². On the other hand, for gyroscope readings, the ideal condition would be zero readings when the sensor remains stationary, as it measures rotational velocity. However, substantial errors were observed due to the lack of calibration and uncontrolled sensitivity, with significant deviations from the ideal zero readings. On average, the errors recorded were between −3.5% to 5% for the $Y$-axis, −5% for the $Z$-axis, and no error for the $X$-axis. Similar results were recorded across multiple baseline tests for the accelerometer and gyroscope, with the error between test 1 and test 2 representing the worst-case scenario.

Although the IMU sensor used in the experiments was not calibrated prior to testing, steps were taken to account for this limitation through baseline measurements. Baseline measurements were conducted under normal conditions, where the IMU was stationary and free from external EMI. These measurements provided a reference point that allowed us to quantify the inherent errors associated with the sensor’s lack of calibration, such as drift, bias, and noise. The baseline data revealed variations in the sensor’s performance, particularly in the accelerometer’s $Z$-axis, which consistently deviated from the expected gravitational acceleration of 9.8 m/s², and notable errors in the gyroscope’s zero-rate output. To ensure the validity of the experimental results, we used the baseline data as a control against which subsequent measurements taken during the IEMI attack could be compared. By doing so, we were able to isolate the effects of the IEMI attack from the preexisting sensor errors. Specifically, we employed a differential analysis approach, where the discrepancies between the baseline and IEMI-affected data were attributed mainly to the presence of EMI. This approach allowed us to mitigate the influence of sensor drift and bias that would otherwise obscure the true impact of the IEMI attack. Furthermore, no post-processing filters were applied during data acquisition to minimize the sensor’s intrinsic noise, and the raw sensor outputs were used. This ensured that the experimental results reflected the sensor’s original response to the IEMI attack.

While calibration reduces error margins, this study relied on baseline measurements to validate results and reflect real-world conditions for commercial IMUs, which are often uncalibrated to save costs and time. Even calibrated IMUs require periodic recalibration due to drift from environmental factors, aging, or stress.

4.2. Impact of IEMI Attacks on IMU Sensor Performance

4.2.1. Exposure at 0 Inches from the Antenna

To investigate the effects of an IEMI attack, the IMU was placed directly adjacent to the EMF antenna. Figure 9a displays the accelerometer readings before and after the EMF antenna was activated, while Figure 9b shows the percentage of error observed due to the magnetic field.

The IEMI attack had a significant impact on the velocity readings. Once the EMF antenna was activated, significant errors were recorded across the $X$- and $Y$-axes. The $X$-axis in particular showed large deviations, with errors ranging from −170% to 100% during certain time samples. Errors on the $Y$-axis ranged from −55% to 42%, while the $Z$-axis exhibited smaller errors, approximately between −1.5% and 2.5%.

Similarly, Figure 10a shows the rotational velocity readings before and after the EMF antenna was activated. In contrast, Figure 10b shows the percentage of error observed due to the magnetic field on gyroscopic data.

The EMF activation affected the gyroscope’s angular velocity readings. When the antenna was activated, the gyroscope displayed altered readings across all three axes. The error ranged from −80% to 110% in the $X$-direction. Errors on the $Y$-axis ranged from −65% to 50%, while the $Z$-axis exhibited larger errors, approximately between −200% and 100%.

4.2.2. Exposure at 10 Inches from the Antenna

The IMU sensor was positioned 10 inches away from the EMF antenna. Figure 11a shows the accelerometer readings before and after the EMF antenna was activated, while Figure 11b shows the rotational velocity readings before and after the EMF antenna was activated.

The increased distance between the IMU sensor and the EMF generator reduced the impact of the IEMI attack compared to the earlier experiment at 0 inches.

4.2.3. Exposure at 5 Inches with the IMU Rotated 90°

The IMU sensor was positioned 5 inches from the EMF antenna with the IMU rotated 90°. Figure 12a displays the accelerometer readings before and after the EMF antenna was activated, while Figure 12b shows the percentage of error observed due to the magnetic field.

The IEMI attack significantly impacted the velocity readings. Once the EMF antenna was activated, large errors were recorded across the X-axis followed by the Y-axis. The $X$-axis showed large deviations, with errors ranging from −38% to 70% during certain time samples. Errors on the $Z$-axis ranged from −14% to 15%, while the $Z$-axis exhibited smaller errors, approximately between −1% and 1%.

Similarly, Figure 13a shows the rotational velocity readings before and after the EMF antenna was activated. In contrast, Figure 13b shows the percentage of error observed due to the magnetic field on gyroscopic data.

The EMF activation affected the gyroscope’s angular velocity readings. When the antenna was activated, the gyroscope displayed altered readings across all three axes. The error ranged from 0% to 20% in the $X$-direction. Errors on the Y-axis ranged from −50% to 50%, while the $Z$-axis exhibited larger errors, approximately −100% during certain time samples.

A comparison of the standard deviation ($\sigma$) and coefficient of variation (CV) with and without IEMI attacks for the MPU−6050 accelerometer and gyroscope is shown in

Table 3. Comparison of standard deviation and coefficient of variation for IMU sensor performance with and without IEMI attack.

IMU Model	Accelerometer Value									Gyroscope Value
	Without IEMI			With IEMI			Ratio			Without IEMI			With IEMI			Ratio
	${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}}}$				${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$
MPU−6050	$0.021$	$0.017$	$0.029$	$0.18$	$0.38$	$0.3$				$\mathrm{N}/\mathrm{A}$	$0.00064$	$0.00045$	$0.0084$	$0.0036$	$0.0032$
	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$
	23	3.8	0.32	200	80.	3.2	8.7	21	10.	$\mathrm{N}/\mathrm{A}$	2.1	$4.6$	$17$	12	$33$	$\mathrm{N}/\mathrm{A}$	5.7	$7.2$

. The results show a significant increase in $\sigma$ of both the accelerometer and gyroscope outputs under IEMI conditions. For the accelerometer, $\sigma$ increased by up to 8.6 times on the $X$-axis, 22 times on the $Y$-axis, and 10 times on the $Z$-axis when exposed to IEMI. On the other hand, the CV values showed large increases, with the CV for the accelerometer’s $X$-axis rising from 23% without IEMI to 200% with IEMI. The gyroscope followed a similar trend, with an increase in both σ and CV under IEMI exposure. The ratios of CV with IEMI to CV without IEMI for the gyroscope were as high as 5.7 for the $Y$-axis and 7.2 for the $Z$-axis. These results highlight the impact of IEMI on IMU sensors, with standard deviations changing by a large factor and CV ratios showing clear degradation in signal stability. This shows the need for effective mitigation strategies to protect UAV systems from IEMI.

4.3. IEMI Impact on MPU−6050: Key Vulnerabilities

During our IEMI attack, the MPU−6050 may have been affected in several ways, primarily due to the nature of its internal components and their operation principles. The gyroscope, for example, operates based on the Coriolis effect, where rotation causes a vibration in the MEMS structure that is detected by capacitive pickoffs. The accelerometer detects the displacement of proof masses due to acceleration forces, again measured by capacitive sensors. These sensors rely on precise capacitance measurements and are sensitive to changes in the EMF around them. Also, the gyroscope and accelerometer signals undergo amplification, demodulation, and filtering to produce a voltage proportional to the measured quantity. This voltage is then digitized using on-chip 16-bit analogue-to-digital converters (ADCs). Strong EMFs can induce currents and voltages within the signal conditioning circuitry and the ADCs, leading to erroneous readings. The MPU−6050 features a DMP that processes data from the accelerometers and gyroscopes [20]. The DMP is designed to offload computation from the host processor and typically operates at high frequencies for accurate results. An IEMI attack could have disrupted the timing and processing within the DMP, leading to incorrect motion-processing outputs. In addition, the MPU−6050 communicates with the host processor via I²C. IEMI may have corrupted the data being transmitted over these communication lines, leading to further inaccuracies in the data received by the host processor. The internal clock of the IC chip synchronizes the operation of the signal conditioning, ADCs, DMP, and other control circuits (Figure 14). The clock is generated from the MEMS oscillators of the gyros. As shown in the figures above, IEMI attacks may have disrupted the clock generation, leading to timing errors affecting sensor data calculations. Below are the key vulnerabilities that may have contributed to these disruptions.

• The MPU−6050’s power-on sequencing is sensitive to the proper ramp-up of VDD and VLOGIC. IEMI could have disrupted these power ramps, causing improper initialization of the IMU’s internal clocks and sensor conditioning mechanisms. Such disturbances likely contributed to erratic sensor behavior, especially in gyroscope measurements.
• The MPU−6050’s internal MEMS oscillators operate at 27–33 kHz for gyroscopes. IEMI, especially in the frequency range of 100 MHz to 3.4 GHz, can introduce noise that disrupts these oscillators, causing deviations in the gyroscope’s output. Furthermore, the ADCs, which digitize the analogue signals, are prone to high-frequency IEMI noise, leading to distortion in the recorded angular velocity and linear acceleration.
• The DMP, which is responsible for fusing data from the gyroscope, accelerometer, and external sensors, relies on precise timing and synchronization. IEMI attacks likely introduced timing mismatches within the DMP, disrupting its motion processing algorithms and causing delays or inaccuracies in the orientation and motion data output.
• Capacitive sensing in the accelerometer is particularly vulnerable to IEMI, as EMPs can mimic displacement, resulting in false acceleration readings. The accelerometer’s zero-g calibration could also be altered by IEMI, leading to non-zero measurements even when no external forces are present. This could explain the anomalies observed in the accelerometer data during the tests.

5. Proposed Mitigation Method

To mitigate IEMI attacks on the IMU, we proposed a hardware-based approach using electromagnetic shielding. The proposed solution involves designing and implementing an electromagnetic shielding system to reduce the impact of IEMI on the IMU’s operation. The shielding unit was constructed from mu-metal, a material with high magnetic permeability that attenuates both low- and medium-frequency EMFs. The mu-metal was shaped into a foil with multiple layers and wrapped around the IMU sensor to form a rectangular enclosure, ensuring that the sensitive components of the IMU were protected from external EMFs. The shielding enclosure was designed to be lightweight and compact, making it suitable for integration into UAV systems without significantly increasing payload weight.

The effectiveness of the shielding was evaluated by comparing the IMU’s performance with and without the shielding under identical IEMI attack conditions. First, baseline measurements were recorded with the unshielded IMU to establish a reference for sensor drift, noise, and bias. Following this, the unshielded IMU was exposed to a medium-power IEMI attack, and the resulting data were analyzed to assess the impact of the EMFs on sensor accuracy. Finally, the shielding was applied to the IMU, and the same IEMI conditions were replicated to evaluate the shielding’s ability to attenuate the EMI. The effectiveness of this shielding is quantified by the shielding effectiveness (SE) parameter, which measures how well the shield reduces EMF strength. SE is expressed in decibels (dB) and is calculated using the following equation:

$$SE=20{\log }_{10}\left({\displaystyle \frac{H_0}{H}}\right)$$

(17)

where $H_0$ is the magnetic field strength without shielding and $H$ is the magnetic field strength with the shield in place. A higher SE value indicates better shielding performance. To ensure the shield adequately protected the IMU, we designed it to achieve an SE value greater than the required threshold. This threshold was determined based on the IMU’s susceptibility to IEMI, ensuring that the magnetic field strength at the IMU was reduced below its tolerance level.

A simulation was conducted using QuickField software 6.6 [26] to further validate the shielding design, enabling precise modeling of the magnetic field distribution within and around the proposed shield. The simulation parameters included a shield with dimensions of 2 inches by 1.5 inches, a relative magnetic permeability of 20,000 for the mu-metal, and an external magnetic field of 2000 μT. The magnetic flux density inside the closed rectangular shield was reduced to 15 μT, as seen in Figure 15a. However, when a small side opening was introduced to accommodate wires connected to the IMU, the magnetic field at the location increased to approximately 110 μT, as seen in Figure 15b. These simulation results demonstrate the superior attenuation properties of mu-metal against the expected external field and support the use of mu-metal as a practical and efficient solution for mitigating magnetic interference.

Next, we conducted an experiment to evaluate the shielding effectiveness of various materials, including mu-metal and mu-ferro, across different magnetic field intensities and foil layer configurations. The attenuation of the magnetic field, measured in decibels (dB), was analyzed to assess each material’s performance in mitigating electromagnetic interference, as shown in Figure 16.

The experimental results closely matched the simulation data, showing that a four-layer mu-metal shield—matching the total thickness specified in the simulation—achieved an attenuation of approximately 52 dB at a magnetic field intensity of 2000 μT. This result highlights the superior shielding performance of mu-metal compared to mu-ferro, particularly in configurations with multiple layers.

Figure 17 shows the effectiveness of the proposed mu-metal shielding in mitigating interference from the EMF antenna during IEMI attacks. The results demonstrate significant reductions in sensor disturbances for both the accelerometer and gyroscope when the shield was in place.

The results above highlight the shield’s capacity to minimize external electromagnetic interference. Compared to the unshielded IMU, the shielded IMU’s acceleration and rotational velocity readings across all axes were noticeably stabilized.

To evaluate the effectiveness of the shielding solution, the standard deviation (σ) and coefficient of variation (CV) for the IMU sensor outputs were compared under IEMI conditions with and without the shielding applied.

Table 4. Comparison of standard deviation and coefficient of variation for IMU sensor performance with and without IEMI attack.

IMU Model	Accelerometer Value									Gyroscope Value
	Without IEMI			With IEMI			Ratio			Without IEMI			With IEMI			Ratio
	${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}}}$				${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathsf{\sigma }}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$
MPU−6050	$0.021$	$0.017$	$0.029$	$0.020$	$0.018$	$0.028$				$\mathrm{N}/\mathrm{A}$	$0.00064$	$0.00045$	$\mathrm{N}/\mathrm{A}$	$0.00064$	$0.00045$
	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{X}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Y}}_{\mathbf{w}\mathbf{o}}}$	${\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}}}/{\mathbf{C}\mathbf{V}}_{{\mathbf{Z}}_{\mathbf{w}\mathbf{o}}}$
	23	3.8	0.32	21	4.0	0.31	0.91	1.1	0.97	$\mathrm{N}/\mathrm{A}$	2.1	$4.6$	$\mathrm{N}/\mathrm{A}$	2.1	$4.5$	$\mathrm{N}/\mathrm{A}$	1.0	$0.98$

shows the results of these comparisons, indicating that the ratio of CV values between the “with IEMI” and “without IEMI” cases are close to 1. This shows that the implemented shield successfully mitigated the effects of IEMI, protecting the IMU from significant performance degradation. Moreover, these results represent a substantial improvement compared to Table 3, where the unshielded IMU exhibited much higher ratios, reaching values as high as 21 for the $Y$-axis accelerometer CV. This confirms the shielding solution’s practical utility in enhancing sensor reliability under IEMI conditions.

The proposed hardware-based shielding solution offers a distinct advantage over existing software-based approaches, such as sensor fusion and filtering algorithms, which attempt to mitigate the effects of EMI at the data processing stage. While software-based solutions are useful in rejecting erroneous data caused by interference, they do not prevent the initial disruption of the sensor’s operation. One limitation of software-based approaches is their reliance on the sensor’s ability to produce sufficient valid data for filtering algorithms to work effectively. For example, in sustained medium-power IEMI attacks, the sensor may become overwhelmed, leading to complete data corruption, which software solutions may be unable to correct. In contrast, our shielding method attenuates the EMFs before they reach the IMU, providing a more effective defense against medium-power IEMI attacks. By addressing the problem at the hardware level, our shielding solution provides a more robust defense against such attacks.

While the proposed electromagnetic shielding solution has proven effective in mitigating the effects of medium-power IEMI attacks, several limitations should be noted. First, the shielding is primarily designed to attenuate medium-power EMFs. Higher-power EMP may still penetrate the shield or induce currents in unshielded components, potentially causing system-wide failures. As such, the shielding may not offer complete protection against all forms of EMI, particularly in extreme environments where EMPs are a concern.

In addition, using mu-metal for shielding adds weight and complexity to the UAV design. While the shielding is relatively lightweight, its integration into weight-sensitive UAVs, such as small drones used in package delivery, may impose constraints on payload capacity and flight endurance. The increased complexity of integrating the shielding into UAV housing may also raise manufacturing costs and reduce accessibility for consumer-grade UAVs.

A combination of both strategies—hardware shielding and software mitigation through sensor fusion—could offer a comprehensive solution, providing redundancy and improving the overall resilience of UAV systems in environments prone to EMI.

6. Conclusions

This study has underscored the crucial role of IMUs in navigation UAVs, highlighting the significant challenges posed by medium-power IMEI attacks on the reliability and performance of these systems across various sectors. Through a thorough review of the literature and detailed experimental analysis, which involved testing a 6-degree-of-freedom IMU sensor with a custom EMF antenna, we investigated the mechanisms by which IEMI affects IMU performance, explored potential sources of interference, and proposed effective mitigation strategies. Our findings indicate that magnetic fields can significantly degrade the accuracy of IMU data, compromising UAV operational reliability. Based on our experimental outcomes, we have proposed and tested a hardware-based electromagnetic shielding solution using mu-metal. The shielding system significantly reduced the impact of IEMI on IMU performance, achieving attenuation of up to 52 dB under controlled conditions.

Further research is required to examine the influence of magnetic fields on IMU performance, especially under multiple attack scenarios. Future work will deepen our understanding of how such fields affect IMU sensors, ultimately aiding the development of more robust navigation systems for UAVs. Future research should also explore developing lighter materials with high permeability that can offer similar or enhanced shielding effectiveness without compromising the UAV’s weight and performance. In addition, the design of modular shielding units that can protect not only the IMU but also other critical electronic components could provide a more holistic solution to IEMI attacks. Our ongoing efforts aim to enhance the resilience and reliability of UAV operations, ensuring their safe and efficient deployment across a wide range of applications.