An Adaptive Secure and Efficient Bio-Inspired Routing Protocol for Effective Cooperation in FANETs

In FANETs, the communication is a highly important component for any application concerning cooperation between the UAVs. For this reason, problems regarding security must be considered as much as the problems related to efficiency or reliability, as there is a trade-off to establish between routing efficiency and security overheads. Many relevant research works have been conducted [7-13, 20-22, 27, 28] in the direction of designing routing protocols, these works mainly focus on the development of efficient and secure routing protocols in terms of specific metrics.

Indeed, all these works could be classified into three main categories: (i) Cryptography-based solutions [7, 8, 12, 13, 27, 28], (ii) Trust-based solutions [9-11], and (iii) Bio-inspired based solutions [20-22]. Some of the related previous works that have been carried out in order to make FANETs more trustworthy are explained in this section.

2.1 Main cryptography-based solutions

In this context, many solutions have been proposed to secure inter-UAV communications. They are primarily focusing on various security services such as authentication, access control, data integrity, availability, and privacy. These methods are the greatest defense against malicious insiders, but they are notoriously resource-intensive and computationally intensive, making them impractical for use in today's UAVs.

Haque and Chowdhury [7] suggested to investigate the feasibility of using Identity-Based Encryption (IBE) in a UAV network with limited resource availability. Where they have assessed the practicality and performance of IBE in the UAV network by measuring the energy used by key management procedures, assessing the feasibility of the technique, and presenting an efficient security framework for a resource restricted wireless UAV network.

The work featured in the study [8] describes the implementation of a blind signature method in a certificate-less environment for UAVs. There are no public-key certificates needed for the proposed scheme. It also does not have the problem of having to keep the keys safe. Nevertheless, the amount of data that is gathered from the platform that monitors the UAVs may be too large to be processed by the same UAVs that are involved in the monitoring activity. Therefore, the authors extend their model to the 5G mobile network by using Multi-access Edge Computing (MEC) in its architecture to ensure secure communication between drones and the base station (BS).

In the same context, a pairwise key establishment scheme is proposed in the study [12], in which nodes in FANETs are not susceptible to get compromised. The idea of this work is based on the phenomenon of the small-world, each node in the network establishes the initial keys with nodes that are no more than three-hops away from it. Then, a pairwise key is established when any two nodes on the path between them exchange partial keys with one another. The pairwise key is generated by adding the partial keys together at each node. During the process of obtaining partial keys, the nodes on the route make use of the initial keys to safely transfer their partial key to the other nodes on the route.

SUAP (Secure UAV Ad Hoc Routing Protocol) is a novel secure routing protocol proposed in the study [13]. The approach enables message authentication while also providing detection and prevention of wormhole attacks. SUAP is a reactive protocol that makes use of public key cryptography and hash chains to protect data. However, the size of the authentication messages and the amount of computing power needed are the main drawbacks of this work.

He et al. [27] proposed a secure communication scheme for UAVs network. The idea proposes that each node maintains and controls a region in which the authorized devices may get a broadcast key without the need for a centralized online authority, this could be achieved by using the hierarchical identity-based broadcast encryption and pseudonym method. All devices in this system are capable to anonymously broadcast encrypted communications and decrypting the lawful ciphertext.

The cryptographic approaches are deployed to transform readable information into meaningless one, so ensuring the confidentiality and integrity of this information. However, cryptographic protection is complex by the fact that UAVs have a limited energy capacity and they cannot efficiently perform cryptographic calculations. Therefore, generating different security codes has been always considered as a difficult problem.

2.2 Main trust-based solutions

The goal of these solutions is to find the misbehaving node when it forwards a packet in the network. Each node builds trust links with other nodes by giving trust values to its relations, and such systems give a way for a requesting node to evaluate the confidence he provides to the provider of a resource. Currently available solutions for FANETs that are based on trust were first presented for MANETs [6, 14], with just a few solutions being specifically designed for FANETs.

As described in the study [9], a new trust model for FANETs has been presented, in which an evolutionary algorithm is utilized to optimize the weights of different factors in order to calculate direct trust values. Direct trust is grouped together with a recommendation to calculate the final trust value of a node. Based on the evaluation of the final trust values, the nodes are either included in the recommended list or removed from the list.

A Trust Based Clustering Scheme (TBCS) for FANETs was suggested in the study [10]. This scheme employs a multi-criteria fuzzy method for classification, which is based on the behavior of the node in a fuzzier and more complicated environment. In this proposal, there is a reward and punishment mechanism that tries to turn the node's behavior into trustful state, and to separate bad and malicious nodes from the rest of the nodes in the FANETs.

Barka et al. [11] proposed a new trust-aware Monitor-based communication architecture for Flying Named Data Networking (FNDN). In this work, the selection of monitors is based on their reliability and stability, they are then tasked with the responsibility of disseminating the interest packets in order to prevent the issue of broadcast storms. Once the data producer receives the interest, the data is sent to the requester through the quickest and most trusted method possible, depending on the circumstances.

Most of these models are based on the notion of recommendation, which, however, suffer from the drawbacks of sluggish convergence and excessive complexity of trust calculations, as well as the huge overhead of network traffic, leading to an inaccurate evaluation of trust value.

2.3 Main bio-inspired based solutions

As it was have mentioned above, all of the current solutions from both pervious categories are suffering from high rates of packet loss, which is considered an intrinsic property of the FANETs. Actually, no previous work for FANETs has focused on bio-inspired routing protocols and the manner they can be used to improve network security. Nonetheless, some papers which are considered to be among the worth noting work in the field have specifically treated FANET communication architecture design challenges and issues. Among these papers, some of them could be quoted, like:

It has been suggested [20] to apply a bio-inspired clustering strategy for FANETs (BICSF), which employs a hybrid mechanism of glowworm swarm optimization (GSO) and krill herd optimization (KH). The researchers suggested an effective cluster management algorithm for the optimal position of the UAV based on the behavioral research of (KH) and genetic operators such as mutation and crossover. In order to guarantee effective communication between the UAVs, a path recognition algorithm based on the weighted residual energy, the number of neighbors, and the distance between the UAVs has been developed to aid in route selection.

With the use of a clustering strategy, the work referenced in the study [21] tries to resolve the problem of the short lifetime of the FANETs. In order to do this, two alternative clustering techniques are used: the first scheme is a mix of the k-means clustering algorithm and the firefly algorithm, while the second scheme is based on the glowworm swarm optimization (GSO) algorithm and the firefly algorithm.

Several FANET algorithms, as well as algorithms based on swarm intelligence, are discussed briefly in the study [22], including a brief summary of the current algorithms in use (ant and bee colonies). It was shown via the experimental research carried out by the authors that it is possible to use bio-inspired algorithms in a cost-effective manner.

Because a hacked or disloyal node offers a significant risk to the network, the path that avoids the malicious node may be more important than the shortest route. As a matter of fact, most routing protocols used by the FANET community are designed to discover the shortest route from a source to a destination without taking into consideration the existence of any malicious nodes along the route. In order to enhance the routing protocol regarding security issues, high interest is focused on the creation of a solution that will make the routing protocol more resistant against malicious nodes actions.

This section depicts a novel reliable security framework based on AIS and PeSOA algorithms, which may be subdivided into two phases: Penguin-AIS-Routing and Penguin-AIS-Security.

4.1 A routing protocol based on PeSOA (Phase 1)

Generally, Ad hoc routing protocols, in their most basic form, follow a common procedure. When the RREQ (Route REQuest) procedure is initiated (step 1 in the Figure 2), the sender node waits for RREPs (Route REPlay) from the other nodes, after that, the routes reply (step 2) is conveyed to the source through a variety of different paths. Nodes provide information about their residual energy, routing load, and hop-count in the reply packets they send back to the sender. Once the reply packets have been received from multiple routes, the source node assesses the data and deduces the fitness value for each path (step 3).

In order to accomplish this, the source node selects the number of paths "n" having the best fitness value and ranks them in its routing table according to the fitness values of each path in descending order, and then it transmits data over the path with the highest fitness value that is ranked at the top of the routing table (step 4). In the event that the route fails, the data transmission is resumed using the path with the highest fitness value after failing. If all "n" routes fail to transmit data, the algorithm begins the process over again, searching for the best fitness multi-paths and repeating the process until the data is successfully sent.

So, the optimal route between source and destination is carried out using the characteristics of the PeSOA algorithm, by treating groups of penguins as UAVs. The area-based search may be mapped using PeSOA, the inter-group migration of UAVs based on explored routes is analogous to penguin inter-group movement based on fish presence. Each group of UAVs will be led by a drone (leader) that will transmit data to the central command, such as, position and discovered targets, and thus, data may be shared with the other groups.

2.png

Figure 2. Architecture of Penguin-AIS model

4.1.1 Objective function

In most cases, there are three major reasons making the route broken between sources and destinations in FANETs networks: The first one is that the UAVs will be rendered inoperative due to a shortage of energy, the second one is UAV mobility, and the third one is traffic congestion, as traffic increases, there will be greater congestion, which may result in packet droppage in certain instances. In this study, the three aspects used to verify the consistency and fidelity of a route are: The Remaining Energy level, the Link Firmness, and the Congestion level.

The best route should have a balanced UAVs energy distribution, excellent connection reliability and, a balanced congestion distribution. Alternatively, this may be stated as the integrated cost function of routing between two UAVs, as shown in the following relationship:

$F_{i j}=\sum_{\substack{i, j \\ i \neq j}}\left(\omega_1 *\left(E_{l e v}(i)+E n r_{\operatorname{lev}(j)}\right)+\omega_2 * L_{F r m(i, j)}+\omega_3 * 1 / \operatorname{Cong}_{\operatorname{lev}(i, j)}\right)$        (5)

where, F_ij represents the objective function, ω denotes the parameter for weight adjustment, such that (ω₁+ω₂+ω₃=1), Enr_lev(i) and Enr_lev(j) indicate the Remaining Energy level of the UAVs (i) and (j) respectively, L_{Frm(i, j)} represents the Link Firmness between UAVs (i) and (j), and Cong_{lev (i, j)} represents the Congestion level between UAVs (i) and (j).

The section below explains all relevant expressions in relation with the objective function:

Remaining Energy Level. It is necessary to estimate and update the remaining energy of each UAV regularly in order to improve the endurance of each UAV.

$E n r_{R e m}=E n r_{\text {Initial }}-E n r_{E x h a u s t e d}$           (6)

where, "Energy_Rem" represents the residual energy level, "Enr_Initial" shows the initial level of energy, and "Enr_Exhausted" indicates the level of energy used when conveying a message. "Enr_Exhausted" can be given as [39]:

$E n r_{\text {Exhausted }}=E n r_{\text {Send }}+E n r_{R e c}$            (7)

where, "Enr_Send" is the quantity of energy consumed while sending the data, and "Enr_Rec" is the quantity of energy consumed while receiving the data. The "Enr_Send" and "Enr_Rec" may be calculated using Eq. (8) and Eq. (9), respectively.

$\begin{gathered}E n r_{\text {Send }}=\left(E n r_{\text {Trans }} * N b r_{\text {bits }}\right) +\left(E n r_{S N R} * N b r_{\text {bits }} * d^2\right)\end{gathered}$               (8)

$E n r_{\text {Rec }}=\left(E n r_{\text {Tranc }} * N b r_{h i t c}\right)$             (9)

where, "Enr_Trans" represents the quantity of energy used to convey a message in one unit, "Enr_SNR" stands for the quantity of energy required to achieve a certain Signal to Noise Ratio (SNR), "Nbr_bits" indicates the number of bits present in the message, and "d" describes how a UAV is far from another (See Eq. (13)).

The energy level has a value that varies in [0, 1]. If the energy level value is greater than the threshold value, a UAV could be associated with other UAVs in order to facilitate the routing process in an efficient manner, else the UAV is isolated and will not participate in the routing process.

Link Firmness. This measure is composed of three components: connection quality, safety degree and mobility prediction factor. If UAVs "i" and "j" are within communication range of one another, the link firmness between "i" and "j" may be described as a combination of link quality "L_{Qty (i,j)}", safety degree S_{Deg (i,j)}, and a mobility prediction factor "M_{Prd (i, j)}", that could be described as:

$\left\{\begin{array}{c}L_{F r m(i, j)}=\lambda_1 * L_{Q t y}+\lambda_2 * S_{\text {Deg }(i, j)}+\lambda_3 * M_{\operatorname{Prd}(i, j)} \\ \text { with }\left(\lambda_1+\lambda_2+\lambda_3\right)=1\end{array}\right.$              (10)

where, λ₁, λ₂ and λ₃ are the weighting factors that are determined by the considered state. It is worth noting that all three costs are standardized to a unit cost. As a result, the value of Link_{Frm (i, j)} is restricted to the range of [0, 1]. After that, we'll go through each of the measures that are mentioned in Eq. (10).

(1) Link quality: The forward and reverse delivery ratios are used in order to determine the link quality. Its computation is referred to [40], and the function expression is provided in Eq. (11).

$L_{Q t y(i, j)}=\gamma^f * \gamma^r$                 (11)

where, "γ^f" stands for the forward delivery ratio, it is defined as the ratio of a packet being successfully sent from UAV "i" and received by UAV "j". Similarly, "γ^r" stands for the reverse delivery ratio.

(2) Safety degree: Assume that (x_i, y_i, z_i) is the coordinates of UAV "i". The degree of safety S_{Deg (i,j)} denotes the proximity between the UAVs "i" and "j" based on their current space, as shown in the following Equation:

$S_{D e g(i, j)}=\frac{R-d}{R}$                   (12)

where, "R" is the communication range of a UAV, and "d" denotes the distance between two UAVs, such as:

$d=\sqrt{\left[x_j(t)-x_i(t)\right]^2+\left[y_j(t)-y_i(t)\right]^2+\left[z_j(t)-z_i(t)\right]^2}$                 (13)

(3) Mobility prediction factor: Real-world kinematic and dynamic constraints prohibit a UAV from moving in an unexpected way, which is undesirable. For the purpose of evaluating the movement trend throughout the duration of the next period of time, the relative velocity between UAVs "i" and "j" at each given point of time, the following formula should be used [40]:

$v_{\text {relative }}=\frac{d_t-d_{(t-\Delta t)}}{\Delta t}$                (14)

where, "t" and "t-∆t" are the times of the most recent and second-to-last respectively of the received “Hello” messages. "d_t" and "d_t-∆t" are the distances between UAVs "i" and "j" respectively. If both UAVs move apart in the opposite direction at a maximum velocity “v_max”, then "v_relative" is up to the maximum value "2v_max". If they are near to each other while traveling at their maximum velocity, "v_relative" will approach the minimum value, to become "-2v_max". So, the mobility prediction factor is defined as follows:

$M_{\text {Prediction }(i, j)}=e^{1-\frac{v_{\text {relative }}}{2 * v_{\max }}}$               (15)

Congestion Level. The congestion level measure is calculated using the traffic mass (T_mass) that happens between the UAVs routes. When the volume of traffic increases, the specific data will either not be transferred or transmitted at a very low transmission rate, resulting in incomplete transmission of the data, which leads to stagnation. The data may be lost as a result of a UAV's buffer Queue being overloaded. So, T_mass is a metric that is used to evaluate the buffer queue degree of each UAV.

To make the routing procedure more efficient, a route with the least quantity of traffic will be picked. The initiator UAV will always keep an eye on the queue state of the neighboring UAV. Every neighboring UAV will regularly update its queue length, and send it to the source. The traffic mass and traffic mass level may be expressed as:

$\left\{\begin{array}{c}T_{\text {mass }}=\frac{\sum_{i=1}^n P_i}{\text { Overall }_{\text {Queue }}} \\ T_{\text {mass-level }}=\frac{T_{\text {mass }}}{\text { Queue }_{\max }}\end{array}\right.$               (16)

where, P_i, is the i^th pattern value that reflects the size of the buffer queue at any given time, Overall_Queue represents the total number of buffer queue, that expanses patterns for a certain time period, Queue_max reflects the buffer queue's maximum length, and "n" represents the number of UAVs along the route. In this way, the congestion level could be given according to the following formula:

Congestion $_{\text {level }}=1-T_{\text {mass-level }}$            (17)

The Congestion_level value must be in the [0, 1] range. If it is more than 1, indicating that there is more traffic, and hence a greater chance of packet loss, indicating also, that UAV should not be used as a forwarding UAV.

4.1.2 Design of the fitness function

The fitness model presented for the PeSOA algorithm is based on the three parameters mentioned above. The solution that offers the greatest quantity of energy remaining, the greatest amount of link firmness intensity, and the lowest degree of congestion, results in the highest fitness function value and represents the best solution. The fitness function may be expressed as follows:

$\operatorname{Fit}\left(F_{i j}\right)=\operatorname{Max}\left(F_{i j}\right)$             (18)

4.2 AIS module (Phase 2)

In order to prevent these kinds of attacks from occurring in the network, a technique that mimics the Artificial Immune System based on the AIS features is used. This objective is accomplished by inspiring the Dendritic Cell (DC) model from the HIS, allowing it to recognize and detect all kinds of malicious attacks in FANETs. Before moving into the construction of the suggested algorithm, it's important addressing the following section to demonstrate how this cell operates.

4.2.1 Dendritic cells

Dendritic Cells (DCs) are immune system antigen-presenting cells (also known as accessory cells). Their primary role is to digest antigen material and deliver it to immune system T-Cells on the cell surface. They serve as conduits for information between the innate and adaptive immune systems. DCs perform anomaly detection on newly discovered antigens and generate adaptive immunity in order to identify the antigen's precise response.

DCs have three primary functions: collecting, processing and immune response regulation. At the collection step, each DC gathers an input antigen and its associated signals. When the first antigen is collected, the processing step starts. As many antigens as possible should be gathered and processed, DCs should operate in tandem to combat the invasion of antigens with varying structural characteristics.

DCs may be classified into three distinct different states: immature, semi-mature and fully mature. In the presence of sufficient input signals, immature DCs develop into either semi-mature or mature DCs, depending on the concentration of particular kinds of these input signals. Immature DCs are exposed to four kinds of input signals: PAMP, danger, safe, and inflammatory signals.

Each signal concentration affects the terminal differentiation state of immature DC when it is subjected to these input signals (either mature or semi-mature). Immature DC processes its contents and produces the cytokine interleukin-12 (IL-12) in response to PAMP and danger signals. Interleukin-10 (IL-10) production is also triggered by safe signals. As a result, incoming input signals provide information about the behavioral context of digested antigens, which may determine whether they are benign or malicious. IL-12 and IL-10 are often used to determine if a migrating DC is semi-mature or mature. In adaptive immunity, DCs move to meet T-Cells after processing the incoming antigen and its associated signals.

To be activated, each T-Cell in the lymph node needs two impulses. The first signal is produced when T-Cell epitopes attach to peptide-MHC on the surface of DC in both danger and safe situations. The second signal will be released either from completely mature DCs like IL-12 in order to encourage naïve T-Cells (NT-Cells) to combat in the danger state, or by semi-mature DC in the form of IL-10 to inhibit NT-Cell in the safe state [41].

As soon as the NT-Cell forms a bond with a mature DC and begins to receive IL-12, the NT-Cell begins to differentiate via a series of events known as clonal expansions. Clones are then separated into memory T-Cells (MT-Cells) and suppressor T-Cells (ST-Cells). MT-Cells store the malicious pathogen identified in the body as soon as it is found, to take a rapid reaction to this pathogen. Further, complete mature DCs will migrate into the lymph node NT-Cells.

4.2.2 Secure routing using DCs algorithm “SR-DCA”

This section describes the techniques used to develop the suggested security model, dubbed (SR-DCA: Secure Routing using Dendritic Cell Algorithm). SR-DCA is a stand-alone algorithm that secures the routing protocols in each UAV in FANETs. It mostly relies on the ability of DCs to serve as abnormality detectors and immune response controllers, respectively.

The SR-DCA is designed to be a monitoring point for verifying certain routing packet types, such as RREQ, RREP or Hello packets (as shown in Figure 3), before proceeding to the packet management of the routing protocol.

This approach seeks to eliminate the attack when it reaches the nearest legitimate UAV. At first sight, A stream of fusible information represented in packets with different IP source addresses and different behaviors of each packet appears to be difficult to identify an attacker. Indeed, if a group of anomalous and normal packets arrives at the queue of the legitimate UAV at the same time of detection, an intrusion detection system may mistakenly associate the behavior of anomalous packets with that of a legitimate UAV, which leads to ignore the aggressor identity.

The SR-DCA is made up of three major units (Figure 3): Monitoring and security unit, Innate unit, and Adaptive processing unit. Each unit consists of different components that interact with one another, and have a link to other components in outer units. In addition, each unit is in charge of performing a specified task in the intrusion detection process.

3.png

Figure 3. SR-DCA model

Monitoring and security unit. It is comprised of two main components: a Queue of Routing Packets (Q.R.P) and an Antigens Manager.

(1) The Q.R.P has three primary functions: collecting input packets from the routing protocol, storing the input packets in a First-In-First-Out (FIFO) queue, and delivering a packet from the front of the queue to the Antigens Manager as required.

(2) The Antigens Manager acts as the nerve center of SR-DCA's operations management. It is in charge of controlling the reception and delivery of inputs and outputs from and to the routing protocol, respectively.

Innate unit. In response to a Q.R.P entry, the Antigens Manager extracts an antigen from the packet. Each retrieved antigen reflects the packet's IP address. In the next step, the Antigens Manager determines whether or not the antigen is present in the entire gene list (Global-List).

4.png

Figure 4. Genes store component

The Genes Store component (Figure 4) contains a list of various genes, which is represented by the (Global-List) variable. For each gene, a sub list from the Global-List is defined as (Sub-List), which consists of one antigen "a_i" and many associated signals (PAMP, safe, danger, and inflammatory) that are specific to that gene "s_j".

where, "i" is the number of antigens in the Global-List and "j" is the number of signals per Sub-List. The signals in each gene, on the other hand, are calculated from a variety of effects produced by the antigen's packet in terms of both its rate and the breaks of link it caused.

The Antigens Manager examines the input antigen, if an antigen is found in the Global-List, then the index of the stored antigen is got and the input antigen needs to be checked. Otherwise, a new gene list for that antigen and its associated signals is generated. The Antigens Manager checks if the input antigen has ever been classified as malicious in the Malicious Antigen List (M.A.L).

Adaptive processing unit. In order to make use of the M.A.L, it must be placed in the MT-Cells (Memory T-Cells). M.A.L is a collection of malicious antigen profiles. A malicious antigen and its corresponding existence counter "C" are included in each profile. This counter calculates how many times the antigen is identified and classified as malicious in the SR-DCA. If the Antigens Manager discovers a counter associated with the antigen in M.A.L that is higher than a specified threshold "t", the input antigen is deemed malicious and a context of "1" is used to symbolize it. A consequence of this, is that the associated routing packet is regarded as an abnormal packet. The routing protocol subsequently loses the packet and does not reply to the source node, causing the package to be deleted.

For the input antigen, two different scenarios could happen in the abnormal detection of DCs.

(1) The first scenario occurs, when the input antigen is not detected in MT-Cells.

(2) The second scenario occurs when the input antigen is discovered, but its counter of existence is less than a particular threshold.

It should be noted in this part that the MT-List threshold "t" value is very significant and difficult to be calculated, because if "C" surpasses the tested input antigen, the antigen is regarded dangerous each time it appears in SR-DCA. Although the use of MT-Cells may expedite response times, its implementation should be done with caution.

The immature DC differentiates into a semi-mature DC, when the antigen context is "0". On the other hand, the DC differs from the mature DC, if the context is "1". Regardless of the state, by following the maturity, the DC should move directly to the adaptive processing unit, where it may regulate the immune response of the NT-Cell, which is controlled by the NT-Cells component.

In the first instance (context=0), DC causes the NT-Cell to become an ST-Cell by causing it to a process of differentiation. ST-Cells are notified to examine MT-Cells for a benign antigen and reduce "C" if such a situation occurs. Conversely, in the second instance, the NT-Cell develops into an MT-Cell inside the MT-Cells component. It is in both instances that the stimulated NT-Cell returns the antigen context to the Antigens Manager located in the Monitoring and security unit. Due to this, the Antigens manager provides results (either 1 or 0) to the routing protocol. The result of "1" should be interpreted by the routing protocol as the presence of an attack produced by the antigen packet, and the result of "0" should be interpreted as the absence of an attack.

Algorithms 2 and 3 present SR-DCA pseudo-code to describe Figure 3's model steps.

4.3 Description and functioning

4.3.1 Optimization-PeSOA procedure in FANETs

The specifics of the proposed model and its pseudocode are described in the steps below: (Figure 5 and Algorithm 4)

(1) Step 1 (Area Definition): The first step is to determine the entire region (the total area) covered by the network and split it into Sub-areas.

(2) Step 2 (Drone population): Define the population of UAVs and the area.

(3) Step 3 (Group Formation) (the first characteristic of PeSOA): Is the process of dividing the UAVs into groups that were determined in step 2. The division of the group takes place as follows:

a) Divide UAVs equally according to the number of sub-areas, for example, 6 UAVs with 2 sub-areas implies that 3 UAVs are required for each area.

b) Otherwise, create as many equally dispersed groups as possible and place the remaining UAVs in the final group, for example, 7 UAVs with 3 sub-areas equals group formations of: Group 1=2 UAVs, Group 2=2 UAVs and Group 3=3 UAVs.

(4) Step 4 (Group Head): Define the UAV-leader for each group to communicate with the central command and to communicate within intra-group.

(5) Step 5 (Boundaries for velocity and position): For each UAV group and sub-area, define the minimum and maximum limits for velocity (v_max, v_min) and position (x_max, x_min).

(6) Step 6 (Search start): Coordinates for a beacon are provided to the UAV-leaders for their particular sub-area, from which the UAV group will launch its search operation.

(7) Step 7 (Search within a group): In order to carry out its communications, each drone conducts a search inside its group to find the target drones in a sub-area. The search is carried out using the Eq. (1) from paragraph 3.2.2.

(8) Step 8 (Information sharing): At the end of each time interval "t", the UAV-leader gathers certain information (such as remaining battery capacity and recorded track coordinates) from neighboring UAVs in the group and sends it to the command center (PeSOA inter-group information sharing strategy).

5.png

Figure 5. Flowchart of the Penguin-AIS-Routing model

(9) Step 9 (Inter-group movement of UAVs) (the second characteristic of PeSOA): The inter-group exchange is dependent on the battery capacity and the criteria of the area search completion. The steps needed to complete this operation are as follows:

a) Following the conclusion of a search in a particular sub-area, the command center is responsible for determining whether or not the battery capacity is accessible in the UAVs belonging to that specific group.

b) Ensure that the battery is not completely exhausted (as in the case of penguin oxygen) and then verify the status of the other groups, otherwise, send the drones with low battery capacity for recharging.

c) First, determine if other groups are still doing searches, and if so, instruct the group-head to distribute all of the UAVs, including itself, to the new coordinates so that it may join the other groups still conducting searches.

d) After joining the new groups, UAVs will begin operating under the supervision of the new UAV-leader of that group.

(10) Step 10 (Finalizing research): After all of the groups have completed their particular sub-area searches, and the overall search for the whole area has been completed, all of the UAVs will return to their starting points and the search will be completed.

4.4 Secure and efficient bio-inspired routing protocol

4.4.1 Integration with AODV Routing Protocol

Although Penguin-AIS can be used with a variety of FANET routing protocols, it is considerably more compatible with reactive routing systems since RREQ is only transmitted when a UAV intends to communicate with a destination UAV. As a consequence, it produces superior outcomes, particularly in terms of security, performance and routing efficiency.

Actually, the major emphasis is on network layer implementation, the protocol presented here is an extension of the Ad hoc On-Demand Distance Vector (AODV) protocol [42], a thorough description of this protocol “Penguin-AIS-AODV” is provided below.

This section focuses on the use of the SR-DCA algorithm in order to improve the security of Penguin-AIS-AODV, in particular to prevent most attacks in FANETs, such as packet-loss attacks and resource-consumption attacks. When the Penguin-AIS-AODV gets a RREQ, it must first examine the context of the RREQ in order to establish its source, that is, whether the RREQ was provided by an attacker or a legal source node, before proceeding. Therefore, the SR-DCA protects the Penguin-AIS-AODV by processing the fictitious request for response packets (RREQs), which take resources from the destination nodes and from the network itself. The SR-DCA also conducts local anomaly intrusion detection in each UAV, without requiring any information about collaboration between the UAVs in the network.

The SR-DCA puts the received RREQs in the Q.R.P and processes them in the order in which they were received, using the FIFO queuing system. When this is done successfully, the SR-DCA extracts the RREQ antigen and its input signals with the help of the Antigens Manager. Each recovered antigen has a unique IP address that corresponds to the packet's original IP address. At the same time, the Antigens Manager computes the PAMP, safe, danger, and inflammatory signals depending on the processed RREQ's rate and the link breaks it induced.

The antigen of each RREQ and its associated signals are combined in the Genes Store's gene list. The presence of a new gene causes the SR-DCA to generate a new DC in the Innate unit, which continually checks the gene list for new input genes. Each DC manipulates a single antigen but not a single gene. It is possible that the Genes Store will have genes that have the same antigen but distinct signals. However, the technique does not require that an input antigen be repeated in several genes on the gene list. Once a DC has been generated, it manipulates the newly produced gene in the list of genes.

Finally, the context of the RREQ is determined depending on "C" and "t" according to the SR-DCA algorithm, as indicated in the part 3 of section 4.2.2. If a suspect UAV is determined to be a malicious UAV, it will be added to the list of malicious UAVs and the findings are shared with all neighbors. New routes won't include the rogue node, so it can't communicate with the rest of the network.

4.4.2 Complexity analysis of the Penguin-AIS model

It is vital to note that complexity is a key criterion for evaluating the performance of algorithms. The time and space complexity of the suggested model are detailed in the following section.

Time Complexity. The following processes are the most important factors influencing the time complexity of the Penguin-AIS model:

(1) The process of initializing a population takes O(n) time, where "n" is the size of the population to be initialized.

(2) Assuming that Max_Iteration is the maximum number of evolutionary iterations required to simulate the proposed algorithm, each group's fitness needs $O\left(\operatorname{Max}_{\text {Iteration }} * n * e\right)$ time, where "e" shows the entropy function.

(3) The method of position updating has a time complexity $O\left(\operatorname{Max}_{\text {Iteration }} * n * \operatorname{dim}\right)$, where "dim" denotes the space dimension.

(4) The Penguin-AIS-Routing algorithm is run until the termination requirements are fulfilled (stopping criterion), which takes O(k) time.

(5) In the Penguin-AIS-Security algorithm, the “SR-DCA” procedure requires O(k') as complexity time. It seems to be less complicated, since it just includes one “WHILE”.

Hence, the overall time complexity needs:

$\begin{aligned} & O(\text { Initialization })+O(\text { Fitness function })+ O(\text { Position update })+O(\text { Penguin }- \text { AIS - Routing })+ O(\text { Penguin - AIS - Security }), \text { meaning: }\end{aligned}$

$\begin{aligned} & O(n)+O\left(\operatorname{Max}_{\text {Iteration }} * n * e\right)+O\left(\text { Max }_{\text {Iteration }} * n *\right.  \operatorname{dim})+O(k)+O\left(k^{\prime}\right) \Rightarrow O\left(k * n\left(1+\operatorname{Max}_{\text {Iteration }}(e+\right.\right.  \left.\operatorname{dim}))+k^{\prime}\right) .\end{aligned}$

Space complexity. An algorithm's space complexity is the maximum amount of space used at any moment that is taken into account during the initialization step. Thus, Penguin-AIS algorithm's overall space complexity is O(n*dim).