Adaptive communication power control for enhancing attack resilience in UAV networks

1. Introduction

Unmanned Aerial Vehicles (UAVs), commonly known as drones, have attracted great attention in various fields due to their versatility and convenience. They are widely used in various fields such as surveillance,¹ environmental monitoring,² agriculture,³ disaster management⁴ and logistics.⁵ UAVs are highly valuable for their ability to operate in challenging environments, collect real-time data, and perform tasks without human intervention. However, the utility of a single UAV is often limited by its operational range, battery life, and susceptibility to interference or damage.⁶ These limitations can severely impact mission success, especially in complex or large-scale operations. To address these limitations, researchers are increasingly focusing on UAV swarms. A drone swarm is a group of coordinated drones that work together to accomplish tasks that would be difficult or impossible for a single drone to accomplish. Drone swarms offer several advantages over single drones, including enhanced coverage, redundancy and resilience. For example, in surveillance missions, a UAV swarm can cover a larger area faster and more thoroughly than a single UAV. If one UAV malfunctions, the other UAVs can continue the mission, thus increasing the overall success rate.⁷

The successful deployment and operation of UAV swarms depend on robust and efficient communication networks. Due to the highly variable nature of UAV swarm networks, UAV swarms use self-organizing networks in which nodes can autonomously establish and maintain connections with each other without a network center for integrated planning.⁸ This leads to the phenomenon of “fragmentation” of the UAV swarm network once some of the UAV nodes in the UAV network fail due to hardware failures, environmental disturbances or intentional attacks.⁹ Figure 1(a) shows a UAV swarm network topology with close connectivity between nodes. However, if we remove some of these nodes, the UAV swarm network structure becomes, as shown in Fig. 1(b). Areas of communication gaps can occur, where parts of the network are isolated, posing a significant risk to UAS mission success. Some method of ensuring that connectivity and information exchange are restored to the swarm network is then required, which is critical to improving network resilience and achieving mission success.

In this paper, we propose a complex network theory-based approach to enhance the robustness of UAV swarm networks. It enables the UAVs in the swarm to dynamically adjust their communication power levels according to the network conditions. By increasing the communication range when necessary, UAVs can remain connected even in the event of a node failure. Adaptive power control algorithms can help balance energy consumption and connectivity requirements to ensure efficient and stable communication.

2. Related Work

2.1. UAV network anti-attack method

In recent years, many researchers have proposed methods to enhance the resilience of communication networks against attacks, aiming to improve the security of various networks and systems. Below is an overview of some key research achievements, along with their limitations when it comes to restoring overall network connectivity.

Wang et al. proposed a game-theoretic collaborative defense method that employs low/medium interaction honeypots for active protection of mobile IoT systems.¹⁰ Sánchez-Patin˜o et al. developed an anti-attack model based on dynamic compensation and reverse penetration processes in P2P networks.¹¹ Xu et al. introduced a DDoS attack defense strategy utilizing Software-Defined Network Function Virtualization (SDNFV) architecture combined with a traffic classification approach.¹² Liu et al. designed observers based on normalization and estimation to gather observations of unknown states and faults and formulated a distributed fault-tolerant consensus strategy resistant to attacks.¹³ Eldosouky et al. proposed a Stackelberg game dynamic model to simulate interactions between UAV pilots and navigation signal jammers, aimed at protecting UAVs from GPS spoofing attacks.¹⁴ Ge et al. introduced a source-aware distributed trust model named UAV-pro, which facilitates accurate trust assessment of UAV networks in resource-constrained environments.¹⁵

However, most of these methods focus on non-UAV domains or propose defense strategies only against spoofing attacks, failing to address how to restore the overall network’s communication capability after the loss of communication ability in some nodes.

2.2. Resilience evaluation method

Resilience and its assessment techniques have been developed and used in a variety of real-world complex systems such as infrastructure,¹⁶ ecosystems¹⁷ and complex engineering systems.¹⁸ Literature¹⁹ outlines many resilience assessment methods.

Tran et al. proposed a methodology for assessing resilience that takes into account system performance data before and after a disruption event.²⁰ This approach can also be used to evaluate systems experiencing multiple disruptions. Li et al. introduce the concept of heterogeneous graphs and first construct a topological model of a heterogeneous communication network for UAV swarms, followed by an innovative resilience model. Based on these two models, a mission-oriented UAV swarm baseline-resilience assessment method is proposed.²¹ Two types of resilience behaviors are simulated based on UAV actions: formation transition and redeployment, Liu et al. For the former, a semi-Markov-based model is used to represent the changes in the probability distributions of the IE link states during such resilience behaviors; and for the latter, a resilience-based optimization method for the reconfiguration of the swarm’s IE topology is proposed. The model combining the two resilient behaviors helps to understand the recovery process of UAV swarms and can be used to select appropriate recovery strategies, which further support mission planning and improve the resilience of UAV swarms.²²

By combining complex networks and resilience, the navigational signaling capability of UAV swarms is assessed. Sun et al. proposed a baseline assessment method for UAV swarm resilience based on complex networks in that paper.²³ Zhang et al. investigated a resilience assessment method for self-organized networks of UAV swarms. First, a two-layer coupled UAV swarm network model is constructed based on complex network theory, and three network topology metrics (average node degree, average clustering factor, and average network efficiency) are used to characterize resilience. Thus, two resilience assessment methods under dynamic evolution are designed: a resilience assessment algorithm considering dynamic reconfiguration and a resilience assessment algorithm considering information relevance.²⁴ Cerabona et al. proposed an innovative approach to disruption and resilience management based on the principles of physics, where disruptions are regarded as forces affecting the performance of the supply chain, and disruptions are considered as forces that change and deviate the trajectory of supply chain performance in a supply chain performance framework.²⁵ Liu et al. introduce the Community-based Backward Generating Network (CBGN) method for efficiently identifying influential nodes in complex networks. This method combines community detection and a novel graph traversal technique to enhance the accuracy and computational efficiency of node selection, demonstrating improved performance in spreading influence across networks.²⁶ Wang and Wang employ multi-scale cross-sample entropy and generalized complexity synchronization metrics to assess the cross-correlation complexity and synchronization, offering insights into the resilience of financial markets by highlighting how they adapt to and recover from external shocks and systemic risks.²⁷

3. Problem Statement and Model

3.1. Mobile model description

The scenario we consider is a vast and empty strip in which a large number of UAVs are evenly distributed and performing tasks at different locations. Due to the huge size of the strip, the vertical distance between UAVs is relatively small, while the horizontal distance accounts for the vast majority of the communication. Therefore, the UAV distribution model in this study is simplified to a two-dimensional plane.

There are n drones distributed in an $m\ast m$ two-dimensional space. These UAVs move continuously and pause occasionally according to a pre-assigned task. In a real-world application, the task affects the UAV’s movement path, but in this paper, we simplify this factor by assuming that the UAVs move randomly through space. Each UAV uses a unique identifier (${\mathrm{\text{ID}}}_i$). The velocity of each UAV at moment t can be described as ($v_{xi}$, $v_{yi}$). The position is ($x_i$, $y_i$). All this information is obtained by the GPS positioning system.

3.2. Communication model description

Each UAV has an identically configured multi-speed transmit power communication device that enables it to adjust its communication power based on the distance of surrounding UAVs to achieve energy savings. UAVs may enter low-power modes during different mission phases to save battery loss, a feature that is particularly important in large-scale UAV networks.

We adopt the communication connectivity model proposed in the literature.²⁸ The model innovatively takes communication distance into account, thus effectively avoiding the appearance of isolated nodes in the connectivity graph. This feature is particularly important for UAV clusters, as maintaining network connectivity is key to ensuring mission collaboration and information sharing. The specific model is as follows :

where $N_t$ denotes the current number of nodes in the network, $\epsilon$ denotes the adjustment factor, and $k_i$ denotes the degree of freedom of node i. This degree of freedom can be considered as the communication ceiling of a UAV. The function $F(d_{j\to i})$ is formulated as follows :where $r_c$ denotes the communication range, $d_{j\to i}$ denotes the distance between node j and i, and $\eta$ denotes the distance impact factor. In practice, as the communication distance increases, some nodes may not be able to connect in the communication edge region. At this time, we can use $\eta$ to adjust the size of this edge region, as shown in Fig. 2. Note that when ${\sum }_{l=1}^{N_t}F(d_{j\to l})=0$, the probability that node j connects to all existing nodes is 0, which implies that there is no node within the maximum communication range of node j.

3.3. Attack model description

To simulate the attack scenarios that UAV clusters may face in actual operation, we have designed three attack models:

(i)	Random Attack Model: n nodes out of m nodes are randomly selected to disable them. This attack simulates the effect of random hardware failures or environmental disturbances on the UAV network.
(ii)	Attack the node with the highest degree of freedom: the node with the highest degree of freedom (number of connections) is selected for the attack. This strategy aims to maximize the impact on the overall network connectivity by destroying key nodes in the network.
(iii)	Attack the nodes with the highest betweenness centrality: n nodes with the highest betweenness centrality are selected for attack. Nodes with high betweenness centrality play an important role as bridges in the network, and attacking these nodes can significantly affect the data transmission efficiency and robustness of the network.

The first method can simulate a UAV failure event caused by an unexpected situation, and the latter two attacks can simulate a scenario suffering from a human attack.

4. Proposed Recover Method

In this section, we will propose a method named Adaptive Communication Power Control (ACPC) that allows UAV clusters to turn the broken UAV network back into a whole when they are damaged, ensuring communication between UAVs without affecting the mission accomplishment. In the previous section, we mentioned that UAVs have multiple communication energy consumption gears to save energy for communication, and similarly, in times of danger, UAVs can increase their communication power to gain a larger communication radius to get in touch with more distant UAVs. However, in the event of an attack, all drones do not need to increase their communication power, and the key is to minimize the increase in power.

4.1. Communication power calculation model

Since UAVs usually work at an altitude of 100–1000m, there are only a few obstructions to block communication. So, we use the Free Space Path Loss (FSPL) communication model to calculate the increase in energy consumption due to the rise in communication distance. FSPL is a metric in wireless communication that describes the attenuation of signal strength when a radio wave propagates in free space. It is the reduction in signal power caused by a radio wave as the propagation distance increases without the influence of any obstacles, reflections, or scattering. FSPL is a frequency and distance function commonly used to calculate the signal loss between two antennas in free space. The following equation can express FSPL :

where d denotes the communication distance in meters (m), c is the speed of light (m/s), and f is the Signal frequency in hertz (Hz). This formula can be transformed into the format of communication power and communication distance as follows :where $G_t$ is the transmit antenna gain in decibels (dB), $G_r$ is the receive antenna gain in decibels (dB), and $P_{\mathrm{\text{sens}}}$ is the receive sensitivity in decibels milliwatts (dBm). To convert transmit power from dBm to milliwatts (mW), use the formula $P_t(\mathrm{\text{mW}})=10^{P_t(\mathrm{\text{dBm}})∕10}$.

It can be seen that the increase in communication power becomes more and more significant as the distance increases.

4.2. Recovery model

After an attack, all nodes are divided into connectivity slices and each connectivity slice performs self-detection. We use percolation theory²⁹ as a basis for restoring connectivity to an attacked UAV network. In percolation theory, we need to add random edges between many small segments of peer-to-peer networks in the network. When the number of such random links increases to a certain threshold, the different components of the network must start combining until a huge connected component emerges, and thus these huge connected components contain most of the peer networks in the attacked network. In UAV networks, such connections are often not generated randomly, but according to the node’s positional speed, and communication distance. That’s why we use a signal transmitter with multiple gears of transmit power. In daily use, the stalls are at smaller places, and when a large communication hole occurs under attack, the stalls are gradually increased until the threshold is satisfied.

(i)	All nodes that detect an ego degree less than 2 increase their communication power by one notch.
(ii)	All that detect an ego degree equal to the upper limit of freedom decrease their communication power by one notch.

5. Evaluation Method

Sharing and interacting with information is a crucial aspect of UAV swarm networks. This study employs the same system performance metric as detailed in Tran et al.,²⁰ which assesses the total volume of information received by the UAV swarm. Each UAV generates information periodically and sends it to the target UAV using the shortest path. The likelihood of a UAV generating a message at time t is given by the following :

where ${\mu }_0$ represents the initial probability of message generation, N indicates the total number of UAVs, and $N_t$ denotes the current count of UAVs. Tran defines system performance as the total number of messages received by the UAV swarm network at time t. This definition arises from the fact that the shortest path length of each message affects the real-time performance of message exchange within the UAV swarm. The total count of messages received in the network is presented as follows :where $R_i(t)$ denotes the number of messages received by node i, $d_j^i$ is the shortest path length obtained by the shortest path algorithm (e.g. Dijkstra’s algorithm³⁰), $N_t$ denotes the current number of UAVs, and $\mathrm{\Delta }$ denotes the time-sensitive coefficient. $\mathrm{\Delta }$ has a value in the range of $0<\mathrm{\Delta }\le 1$ given according to the task requirements. The data to be sent by each drone is a random number. Multiple simulations were performed and averaged as the original value of $y(t)$.

The original method does not penalize the case where nodes i and j are not reachable to each other. The reduction in the number of UAV nodes when attacked will itself bring about a reduction in the communication volume, which cannot directly measure the connectivity of a network. So this paper proposes a new formula for evaluating network connectivity at moment t as follows :

where $d_j^i$ and $N_t$ have the same meaning as in the above equation and ${\mathrm{\text{hop}}}_{\mathrm{\text{max}}}$ is the maximum number of hops allowed by the network. We use an activation function to evaluate the connectivity of the network based on the number of hops the message is forwarded. We removed the random way of generating the information and used fixed time generation instead, this way we can reduce the error caused by random numbers. Using this formula, we can evaluate how well the entire network is connected based on the number of hops the message is forwarded. where $\sigma$ denotes the performance factor, $\delta$ denotes the absorption factor, $\rho$ denotes the recovery factor, $\tau$ denotes the recovery time factor, and $\zeta$ denotes the fluctuation factor. These coefficients use a comparison of pre-recovery and post-recovery capacities. The individual values are calculated as follows :where $y_D$ denotes the average level before the network is compromised, $y_{\mathrm{\text{min}}}$ denotes the average level of network performance after the compromise without using the recovery strategy when the network performance is damaged, $y_R$ denotes the average level at which the network performance stabilizes after using the recovery strategy. $\zeta$ is the signal-to-noise ratio of the original data and the smoothed data calculated after obtaining the smoothed data by using Savitzky–Golay (SG) filter³¹ on the obtained performance data $y(t)$. Then, the signal-to-noise ratio of the original and smoothed data will be calculated. Details are shown in Fig. 3.

6. Results and Discussion

In this paper, MATLAP 2017a is used for simulation. UAV swam is used to perform a task like forest surveillance, etc. However, drones can be rendered useless by certain attacks or by energy depletion. It is our goal to prevent the emergence of isolated drones without changing the planned paths of the drones. We assume that there are N drones distributed in a fixed range, the drones’ action path adopts the Random Waypoint action mode, and the communication radius of the drones is initially $r_c$, which will be changed later. We assume a time point $t_{\mathrm{\text{attack}}}$ at which an attack is initiated and a time point $t_{\mathrm{\text{recovery}}}$ at which the drone detects that an attack has been generated and adopts a recovery mode, and our simulation is at $t_{\mathrm{\text{final}}}$ end. The specific configuration parameters are shown in Table 1.

We will analyze the performance of our proposed method ACPC in terms of network connectivity, and power required for communication, and compare the R-value of the UAV swarm network before and after using ACPC.

First, we conducted an experimental analysis of the connectivity levels of UAV networks under different attack scenarios. Utilizing Eq. (7), we obtained the UAV network connectivity performance metric, denoted as $y(t)$. In this experiment, we employed three types of attack strategies: random attack, degree-based attack and betweenness-based attack. Detailed explanations of these attack strategies can be found in the attack model section of this paper. The number of UAV nodes in the network was set to 200, with varying numbers of UAVs subjected to the experiment. In Fig. 4, the three experimental results correspond to UAV networks where 40, 60 and 80 UAVs were attacked, respectively. The time interval from 0s to 20s represents the normal operation phase of the UAV swarm, where the fluctuations in $y(t)$ are mainly due to the internal movements of the UAV nodes. From 21s to 30s, the UAV swarm was under attack, during which a certain number of UAV nodes lost communication capabilities, leading to a sharp decline in the connectivity metric $y(t)$. From 31s to 50s, the UAV swarm employed the proposed method to restore connectivity. The results indicate that when the number of attacked nodes is relatively small, using ACPC can effectively restore UAV network connectivity. Even under large-scale attacks, the method still exhibits a certain degree of recovery capability. Moreover, the proposed method demonstrates resilience against all three types of attacks.

Due to the structural limitations of UAVs, energy is both crucial and limited, making it essential to assess how much our proposed method ACPC increases the energy consumption of the UAV system. We used [insert method] to convert the communication radius of UAV nodes into the corresponding communication energy required. As shown in Fig. 5, the initial smooth portion represents the total communication energy consumption of the UAV swarm before the attack. Following the attack, some UAV nodes lost their communication capabilities, reducing overall communication energy consumption, reflected in the decrease observed in the figure. Thus, from Fig. 5, we can see that the more UAV nodes are attacked, the more pronounced this drop becomes. The final segment of increased energy consumption is due to some UAVs increasing their communication power to restore network connectivity. It is evident that the overall energy consumption did not significantly increase, remaining at a level similar to that before the attack. However, as the number of attacked nodes grows, the increase in energy consumption becomes rapid. When the UAV swarm experiences an excessive number of attacks, ACPC may lead to excessive energy consumption, which is an area we need to improve in future work. We use three colors and three different line styles to represent the type of attacks on nodes and the number of nodes attacked: red for random attacks, green for degree attacks and blue for betweenness centrality attacks. Solid lines indicate that 40 nodes are attacked, dashed lines represent 60 nodes, and denser dashed lines denote 80 nodes attacked. It is evident that when the same number of nodes are attacked, the green lines are consistently higher than the other two colors, indicating that degree attacks require more energy for recovery compared to random and betweenness centrality attacks. This could be because degree attacks directly disrupt network connectivity, but are also more recoverable, allowing for restoration through increased energy input. In contrast, networks find it harder to fully recover from betweenness centrality attacks, leading to lower energy consumption.

Finally, using Eq. (8), we evaluated the resilience of the UAV network’s connectivity under different attack scenarios. In Fig. 6, the number of UAVs attacked in the three experimental results is 40, 60 and 80, respectively. The blue sections represent the resilience values of the UAV swarm network when no recovery measures are taken after an attack. In contrast, the orange sections depict the resilience values when ACPC is applied. It is evident that the network’s robustness significantly improves with our recovery method. The most pronounced increase in resilience is observed under betweenness attacks, with resilience values increasing by 5.3612, 7.4177 and 8.9435 times. Our method also shows substantial effectiveness against random and degree attacks, with resilience improvements of 2.5857, 3.0313 and 3.6188 times, and 2.6322, 4.1868 and 6.3443 times, respectively.

7. Conclusion

Due to the unique nature of UAV missions, these UAVs often operate in areas far from base stations, necessitating a continuous connection with the UAV network to ensure mission completion. This paper introduces a method to maintain the connectivity of a UAV swarm, proposing a novel evaluation approach to assess the overall network connectivity. The results indicate that by adjusting the communication power of UAVs, the system can demonstrate significant recovery performance when under attack, without a substantial increase in total energy consumption. However, it should be noted that this study does not integrate task completion with UAV path planning. Altering the UAVs’ paths can also impact the network’s connectivity. Future work will aim to incorporate task-driven path planning with the current approach, calculating the overall energy consumption required for such an integrated strategy.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (No. 72201268), the Natural Science Foundation of Sichuan Province (No. 2022NSFSC1902), the Key Laboratory of Flight Techniques and Flight Safety of CAA (No. FZ2022ZX46), the Safety Foundation of CAAC (No. FY2024MHBZ-12), the Special Project for Improving Scientific Research Capabilities of the Suining Branch of CAFUC.

Funding

This research was funded by Research and Practice on the Construction of Modern Industrial College for Large Airplanes (MHJY2024008).

ORCID

Linfeng Zhong  https://orcid.org/0000-0001-8277-4277

Lei Zhang  https://orcid.org/0009-0004-5276-8780

Hao Yang  https://orcid.org/0009-0003-8234-1799

Pengfei Chen  https://orcid.org/0009-0008-4422-2197

Qingwei Zhong  https://orcid.org/0000-0003-1541-0347

Fei Hu  https://orcid.org/0009-0005-7783-6118

Jin Huang  https://orcid.org/0009-0006-8346-4039