1 Introduction

Mobile ad hoc networks (MANETs) [1] in the environment are composed of mobile devices called nodes. They dynamically move in space and communicate with each other to form a communication network in which there is no centralized element to manage the devices. The properties and characteristics of MANETs imply that the autonomous behaviour of nodes, their constant movement at different speeds, their variable number and limited energy resources make these types of networks susceptible to resilient data transmission [2]. The resilient of data transmission is severely limited not only by these characteristics but also by the possibility of malicious or unreliable nodes joining the communication process [3]. For these reasons, there is a need to provide resilient data transmission and a resilient routing process to eliminate these characteristics and situations and to ensure resilient communication in the discovery and maintenance phases of the communication path, but also in the data transmission phase itself [4, 5]. To achieve these goals and requirements, a resilient routing algorithm based on decentralized blockchain technology and artificial intelligence (AI) has been proposed. The proposed algorithm is implemented in reactive and proactive routing protocols along with which it provides resilient communication. The working principle of such modified protocols is in the path according to the obtained network Quality of Service (QoS) and technical information of the nodes in the network. Based on this information stored using decentralized blockchain technology, which is also the input parameters of neural networks (NNs), a suitable resilient node was determined. From such nodes, a resilient communication path was then determined to provide more efficient transmission and routing of data, which were evaluated using QoS parameters [6,7,8]. Before using the algorithm itself and obtaining the information required for resilient routing, routing was secured through the chosen routing protocol [9].

As mentioned earlier, MANETs consist of mobile nodes that move freely in space and form a wireless link between themselves, in which each subscriber can be a receiver or sender of information. All such nodes operate on a peer-to-peer basis, so this type of network does not require a central element to manage individual devices and the entire network [10]. Some certain advantages and disadvantages arise from the theory of MANETs [11]. The main advantages of this type of network are their flexibility and their ability to be used in extreme situations where conventional wired networks cannot be used or would be too costly and impractical to use. On the other hand, there are also several drawbacks associated with MANETs that hinder their secure, efficient and stable use [10]. However, the resilient data transmission algorithm proposed by us is mainly intended for use in extreme situations in the form of military conflicts, natural disasters, etc. The definition of resilient implies that it is a set of multiple parameters (Fig. 1) that directly or indirectly affect the efficiency and stability of a given network [12, 13].

Fig. 1
figure 1

Example of concepts belonging to resilience [13]

Full size image

Frequent topology changes, changes in the number of nodes in a network and their different mobility rates directly affect network performance and disruption of established communication paths [14]. This results in packet loss and the need to start a routing process to find a new communication path. With re-routing, the network becomes flooded with signalling packets, which also reduces the efficiency of the network [15]. These problems and shortcomings can be observed when using reactive or proactive routing protocols, which work on the principle of finding the shortest path from the source node to the destination node. A change in topology or a deliberate disruption of the communication path leads to the need to re-find it through the routing process. A common problem in routing and communication according to available routing metrics is the situation where a node is included in the communication path that intentionally or unintentionally hinders communication and routing by dropping packets, breaking the communication path, etc. [14]. This scenario may arise when a malicious node arrives to the network with a specific goal of obstructing communication, or when a node is integrated into the network that, while not displaying the traits of a malicious node, is incapable of maintaining reliable and efficient communication due to technical constraints (such as connectivity issues, low energy levels, and so on). One way these losses could be avoided is by selecting nodes suitable for the routing process, path maintenance and communication itself [16, 17].

The decentralized blockchain technology (Fig. 2), based on its properties and working principle, allows for the storage of various data about individual nodes in the network, which can be used in our design for subsequent iterations of routing and maintaining the communication path [18, 19]. The special feature and the biggest advantage of this technology is the immutability of the stored data, which is based on the persistence of the individual blocks of the database [20]. This means that all the blocks are linked within the decentralized blockchain technology through a Hash value. This value is determined by a consensus algorithm of various kinds. However, in terms of security, the Proof of Work (PoW) [21] consensus algorithm appears to be the most secure even though its undeniable drawback is energy consumption, which is a very important parameter in MANETs [22, 23]. The advantage of the immutability of the data stored in the database during the routing process is to increase resilient of the network as a whole since it is possible to use this information to eliminate nodes that intentionally or unintentionally confuse efficient communication [24, 25].

Fig. 2
figure 2

Example of block composition in decentralized blockchain technology [24]

Full size image

Resilient data transmission in MANETs is one of the biggest challenges that arise when using them. To achieve resilient data transmission it is necessary to provide a resilient communication path preceded by a resilient routing process. The definition of MANETs implies that nodes are constantly moving at different speeds and in arbitrary directions. However, the routing process needs to consider not only this fact but also the technical parameters of the nodes in the network in the form of energy consumption, stability of the Internet connection and the degree of network QoS parameters. It is also necessary to consider the situation in which malicious nodes join the network and deliberately drop signalling or data packets, thereby undermining resilient of the network. These and many other facts affect the stability of the connection, the security and resilient of the whole network. Based on this, it is necessary to design a mechanism that will prefer the nodes that perform best within the mentioned parameters and situations in the routing process based on previous experience and information gained.

In this paper, the Ad Hoc On-Demand Distance Vector(AODV) [26] and Optimized Link State Routing (OLSR) [27] routing protocols were used to initially discover the communication paths from the source node to the destination node. To provide a resilient routing process, a new algorithm has been proposed that uses decentralized blockchain technology and AI in the form of deep neural networks (DNNs). Such a algorithm has been implemented in the above routing protocols Resilient - Ad Hoc On-Demand Distance Vector and Resilient - Optimized Link State Routing (R-AODV and R-OLSR). Both these protocols considered packet delivery metrics, throughput, total End-To-End (E2E) delay, the ratio between signalling and data packets, and technical parameters of individual nodes in the network as key parameters for resilient node selection and subsequent resilient communication path construction. The main contributions of this paper are the design of a new resilient routing algorithm based on AI and decentralized blockchain technology, the design of resilient routing protocols R-AODV and R-OLSR, as well as the comparison of the obtained experimental results.

Recently, the various options and principles of communication within mobile wireless networks have increasingly come to the fore, bringing with them new challenges and requirements for efficiency, security and especially resilience of communication. The communication itself, but also the routing process, is subject to many processes and situations that affect the network as a whole, such as dynamic environments, unstable infrastructure, the possibility of unauthenticated and malicious nodes joining, technical limitations of nodes, and many other characteristics and situations that make the design of a suitable solution in this area more challenging. This is why care must be taken to have a robust routing process and thus carefully select the communication nodes themselves during the communication of many nodes in the network. Based on these facts, there are a large number of challenges and problems associated with resilient routing itself in MANETs. Therefore, our work is motivated in this direction. The proposed resilient routing model is based on decentralized Blockchain technology and Artificial Intelligence, which together with routing protocols provide resilient and efficient selection of communication nodes based on technical and QoS parameters participating in a given network.

The main contributions of this paper are:

  • Resilient routing approach in MANETs based on decentralized blockchain technology and DNN.

  • Design of modified resilient reactive and proactive routing protocols R-AODV and R-OLSR.

  • Utilization of decentralized blockchain technology in a resilient routing process.

The paper consists of several main sections. In the first part, the theoretical background of MANETs, blockchain technology and resilient of data transmission in multi-hop mobile networks have been described. In the next section, contributions and research works in the area of resilient data transmission in different types of multi-hop mobile networks using different technologies providing communication efficiency enhancement were described. The problem of a resilient routing process in MANET was outlined in the paper along with the proposed solution using AI and decentralized blockchain technology. Section 3 described the proposed resilient routing algorithm implemented in reactive and proactive routing protocols. The description of the proposed algorithm includes a description of the methodology of the algorithm itself, the determination and calculation of the QoS and technical parameters that served as input parameters for the DNN, and how decentralized blockchain technology was used in the routing process. In Section 4, experimental simulation scenarios were proposed to evaluate the proposed resilient routing algorithm. This section included the analysis of the obtained results, which were evaluated in terms of QoS parameters. These parameters included the signal-to-data packet ratio, the total E2E delay, and the overall throughput of the tested network. This analysis evaluated the average degradation of routing protocols without the resilient routing method implemented compared to routing protocols with the resilient routing method. The paper concludes by evaluating the proposed resilient routing algorithm along with outlining further challenges and directions in the area of resilient data transmission in multi-hop mobile networks. Table 6 lists the notions used later.

2 Actual state of the art

In the following section, a brief review of related literature and research activities aimed at improving the security and resilient of mobile multi-hop networks has been presented. Individual works have focused on improving the above properties mainly through blockchain technology along with the implementation of various encryption, and decryption-based technologies as well as different types of routing protocols and algorithms. At the end of this section, a brief comparison of the selected papers with our proposed resilient routing algorithm based on decentralized blockchain technology and DNNs has been presented.

Vanitha et al. [21] proposed a two-channel adversarial capsule generation network optimized by arithmetic optimization method for a trusted distributed routing scheme based on blockchain technology within MANETs. The purpose of designing such a network was to distribute a trust module for each routing node in the MANET network that provides authority based on blockchain technology. The flow of packets within the network is dependent on the creation of a token which contains a secret key that is distributed within each routing node, thus the authors ensured the provision of an optimal communication path within the network. To analyze the results obtained and evaluate the proposed model, the authors tested the model through the parameters of throughput, delay, energy consumption, latency and efficiency. From the analyzed results, the authors report a significant improvement in delay and energy consumption compared to existing methods.

Naresh et al. [28] proposed a method of multiple shared keys within a group of nodes based on blockchain technology, using smart contracts that serve as two-party controllers within each node in the network. The main contribution of the authors was the creation of a multi-party key agreement that provides privacy and enhances network security. They achieved significant improvements in the security and privacy of large-scale wireless ad hoc networks under per-key attacks.

Liu et al. [29] addressed the problem of traffic management and monitoring in large cities and the related problem of incident reporting due to technical and non-technical factors. They proposed an identity authentication and privacy protection scheme based on blockchain technology and circular signatures to address these issues and enhance network resilient. The authors used blockchain technology to preserve the anonymity of the network participants while providing traceability, meaning that the identity of a particular network participant was anonymous, but a particular message could be traced back to the source node if necessary. In this paper, the authors also presented a new circular signature scheme based on the SM2 algorithm that allows a group of users to anonymously sign a message. For this scheme, they analyzed its security level using smart contracts in the Ethereum network. Based on the results obtained, it was possible to conclude that the authors achieved an efficient approach that is resistant to a wide range of common attacks.

Yang et al. [30] addressed the importance and enhancement of security in wireless sensor networks by saving energy, which is a key factor for a dynamic network. To address this challenge, they proposed a trust-aware dynamic routing algorithm based on augmented AODV for secure communication in Wireless Sensor Networks (WSNs) Trust-Aware Dynamic Routing - Extended AODV (TADR-EOADV). The proposed algorithm works with direct trust, recommended trust, link strength, energy level, and fitness score, which have been used to measure the distributed security level of nodes in routing. According to the analyzed results, the authors reported that the TADR-EAODV protocol improved the average packet transmission rate by 6% to 17% compared to the existing state-of-the-art routing algorithms Low-Energy Adaptive Clustering Hierarchy (LEACH-TM), Time-Aware Opportunistic Routing Protocol (TAAO-SDTIM) and TAOSC-MHR. The results related to denial-of-service attacks were analyzed and according to the results obtained, the TADR-EADOV algorithm was able to successfully identify the attackers in the simulated network according to their abnormal behavior.

Shah et al. [31] proposed an adaptive routing protocol based on a biologically inspired genetic algorithm that selects the best path from source to destination using the AOMDV-FG fitness function. They compared the proposed algorithm with AOMDV-TA and EHO-AOMDV protocols in the parameters of E2E delay, throughput, energy consumption, packet delivery ratio (PDR) and routing overhead. Comparing the proposed genetic algorithm with AOMDV-TA and EHO-AOMDV, the authors achieved some improvement in all measured performance parameters, however, due to the high mobility of the network, the problem of collision of transmitted data persists which directly results in the reduction of the network throughput and hence the reduction of its ability to transmit data.

The Blockchain System for Security Data Collection (B4SDC) blockchain system, which is used to collect security-related data in MANETs, was proposed by Liu at al. [32]. This system optimizes the number of Route Request Packet (RREQ) and Route Replay Packet (RREP) packets sent to achieve higher network efficiency. The optimization is done using blockchain technology and digital signature which provides higher security against different types of attacks etc. From the results obtained, it was possible to conclude that the authors achieved higher resistance against spoofing and collusion attacks and also against excessive forwarding.

A trust management model based on blockchain technology was proposed in Vehicular Ad Hoc Networks (VANETs) to preserve privacy by Li et al. [33]. The privacy-preserving process is done by requesting a certificate without having to divulge one’s private information. Based on the experimental simulations and the analysis of the obtained results, it was possible to conclude that the proposed model showed some improvement in resilient to attacks that violate trust in the network.

A dynamic decentralized reputation system [20] using blockchain technology has been proposed to provide reputation information within the network. Compared to other methods and proposals for obtaining reputation information, such a system is not limited to obtaining information from individual nodes and their neighbors but obtains feedback from the entire network. According to the results obtained, it was possible to conclude that such a system provided a slightly higher availability of unchanged reputation information and also achieved better results for the loss parameter of this information.

Lin et al. [34] proposed a new protocol Blockchain-Based Conditional Privacy-Preserving Authentication (BCPPA) that provides secure communication in VANETs using Public Key Infrastructure (PKI) and Eliptic Curve Digital Signature Algorithm (ECDSA) digital signature. From the achieved results, the authors presented the improvement of network security.

Gutierrez [35] created a MeshSim simulator designed for real-time network simulation, which was used to experimentally test the proposed Reactive Gossip Routing protocol family. The proposed protocols were designed to provide resilient routing by mitigating the characteristics of MANETs that create failures, packet loss, and link breakage. However, in this paper, the author concludes that the proposed protocols from the Reactive Greedy Reactive (RGR) family that work on the principle of distance vector routing are not suitable for use in MANETs due to the effect of geographical dispersion based on which the desired properties of such networks are not achieved.

The previous studies done by different research teams have focused on improving network security through the utilisation of blockchain technology, as well as incorporating technologies like ring signatures, digital signatures, or integrating this technology directly into the routing protocol, among other approaches. Thus, from the above outputs, it can be seen that the aforementioned works have mainly focused on security, which is a subset of the term ’resilience’. The principle of application of the described works was based on the previous behaviour of the different network participants, but they did not work with the current state of the network in their designs. Our manuscript describes the design of a resilient routing algorithm, which is mainly aimed at increasing resilience in the routing process itself by using decentralized blockchain technology and DNNs to determine the appropriate routing node based on the current QoS state and technical parameters of each node in the network. The routing node or resilient communication path thus selected is stored in an immutable decentralized blockchain database, from which this data is used in subsequent routing iterations. Thus, our proposed algorithm has not only improved the network security based on previous experience, but also increased the resilience of the routing process and hence the network itself, by leveraging the current information and information gathered from previous communications.

3 Proposed algorithm of resilient routing for MANET

To provide resilient data transmission in MANETs in 5 G networks, a resilient routing algorithm has been proposed to optimize the process of the selection of mobile nodes suitable for routing and transmitting data packets. The routing process has been optimized using DNN, decentralized blockchain technology and by analyzing the parameters that are the input parameters of DNN. These parameters include the delivery ratio of signalling, data and relevant packets and also include the data obtained from the hardware of each node, these parameters include energy consumption (V\(_B\)(t)), computational power (V\(_{CPU}\)(t)), RAM load (V\(_M\)(t)) and the current location of the node (V\(_P\)(t)). This resilient routing process has been applied to reactive and proactive routing protocols. In addition to resilient routing within MANETs themselves, this mechanism can also be used in the cloud or inter-cloud communication (Fig. 3). These solutions have been conceived and designed for crises caused by natural and technological disasters and for situations where any communication is impossible or needs to communicate through resilient nodes and a resilient communication path.

Fig. 3
figure 3

Main idea of proposed solution of the resilient routing and deciding

Full size image

3.1 The main idea of proposed solution

MANET in a 5 G networking environment has been proposed to solve the above problem. In this network, a resilient routing mechanism has been implemented which is based on decentralized blockchain and DNN technology. The resilient routing procedure itself consists of AI-based routing and computation of routing parameters, which are the input parameters of the DNN and were also stored in the blockchain database that was distributed throughout the network. Based on the output of DNN, a suitable node was selected, and then the whole communication path was found which satisfied the conditions of the resilient communication path.

Since the proposed resilient routing algorithm was implemented in the framework of reactive and proactive routing protocols, it was necessary to split the process into a communication path discovery part and a communication path maintenance part (Fig. 4). In the initial phase, it was necessary to discover the communication path using the routing protocol and calculate the input parameters for the DNN, which were then distributed to the network via the decentralized blockchain technology. Once these parameters are distributed, a resilient routing process can be maintained and ensured using the proposed algorithm.

Fig. 4
figure 4

Cooperation between AI and decentralized blockchain technology during routing process in MANET

Full size image

The whole process of the proposed algorithm had to be divided into three main parts, in the first step, a network was considered in which the blockchain database and DNN are not distributed. After the subsequent distribution of these technologies, the proposed mechanism must ensure resilient transmission through unidirectional and multidirectional communication. Figure 6 shows a network that describes a situation where the decentralized blockchain database and AI are not distributed in the MANET to ensure resilient routing using the proposed algorithm. In such a situation, the initial routing is provided by routing protocols that fall into the category of reactive and proactive protocols to best demonstrate the suitability of such a solution. As can be seen from Fig. 6, the routing and distribution principle of decentralized blockchain and AI consists of three basic steps:

  1. 1.

    In the first step, the routing protocol itself provided the routing. If the routing process was successful and a communication path was found from the source node to the destination node, data communication took place. At the same time, the network and technical parameters were determined, which will be one of the basic parameters to determine a resilient communication path, the method of calculating these parameters was described in more detail in Section 3.2.

  2. 2.

    After successful routing and data communication, the distribution of the decentralized blockchain technology among the nodes that were part of the communication path was carried out. The blocks of the blockchain contained data and parameters resulting from this technology, such as the source node, the destination node, the timestamp, the nonce, the transmitted data, the hash of the previous block (in the case of the first block, it is a genesis block, which does not contain the hash of the previous block), as well as the parameters needed for a resilient routing process. These parameters were the input parameters to the DNN. The method of inserting this data into the blockchain block is shown in Fig. 5.

  3. 3.

    In the last step, the DNN update and calculation were performed. The input parameters of DNN are QoS and technical parameters. The output of the DNN was the node that satisfied the resilient conditions as much as possible and best. Subsequently, a resilient communication (Fig. 6) path was formed from such nodes. Similar to the QoS and technical parameters, the output parameter of DNN was embedded into a blockchain database which was then distributed to the MANET.

Fig. 5
figure 5

Inserting AI input data into the blockchain block for proposed algorithm

Full size image
Fig. 6
figure 6

Distributing blockchain technology and input AI parameters during routing in MANET

Full size image
Fig. 7
figure 7

Selected resilient single-path communication in MANET

Full size image

The example shown in Fig. 7, describes communication via a single communication path within the R-AODV and R-OLSR routing protocols. In this situation, the same MANET as in Fig. 6 was shown, but this network had a larger number of nodes among which the decentralized blockchain database was distributed. The topology presented shows that multiple communication paths have been found in the routing process. The principle of reactive and proactive routing protocols implies that the basic parameter of finding a communication path is its length, or the number of hops between the source and destination nodes. Our proposed resilient routing algorithm implies that the length of the communication path is not the fundamental parameter based on which the final path is determined. Thus, in terms of the number of hops, a resilient communication path could have been found that was larger or longer, but in terms of data transmission and routing resilient, this path was more suitable for data transmission.

In a MANET, a multi-path routing situation can occur. In Fig. 8, a situation was shown in which the blockchain database was distributed among all the nodes in the network, i.e., each node in the network had information about the QoS and technical parameters of each node in the network. In this case, resilient routing was possible through the proposed mechanism because each node had information about every node in the network. Due to this fact, the output of DNN could be a set of multiple communication paths, in our case they were denoted as the first path and the second path. The final routing or communication path was constructed from both of these paths, with the individual nodes in the selected path determined by the best QoS values and technical parameters. Thus, the final communication path could have a significantly higher number of hopping nodes but was more suitable in terms of resilient.

Fig. 8
figure 8

Resilient multi-path communication

Full size image

3.2 Optimization of deep neural network for MANET

DNN (Fig. 9) have various hyperparameters that affect how a given DNN will behave in terms of efficiency. These parameters can be divided into two basic groups, hyperparameters related to the structure of the DNN and hyperparameters related to the training algorithm. The hyperparameters related to the structure of the DNN include, for example, the number of hidden layers and the number of neurons.

Fig. 9
figure 9

Proposed algorithm of deep neural network implemented to the resilient routing algorithm

Full size image

The hyperparameter’s number of hidden layers and the number of neurons describe the actual number of hidden layers (HL) of the proposed DNN and the number of neurons (NoN) in these layers. The chosen values of these parameters affect the complexity and the actual capabilities of the designed network, in our case, two hidden layers (HL = 2) with five and four neurons in these layers (NoN\(_{HL1}\) = 5; NoN\(_{HL2}\) = 4) were designed. The number of layers and neurons was determined based on the number of input and output parameters of the DNN and experimental testing of that network with different ratios of these hyperparameters.

The hyperparameters of the training process take into account the learning coefficient, the error function and the batch size. The learning coefficient determines the rate at which the network weights are updated during the training process. It represents the size of the step the network takes in optimizing the weights based on the gradients of the error of the power function. The batch describes the number of samples that were processed in a single training iteration, which has a direct impact on the speed and stability of the network learning. The error function is a hyperparameter that is minimized during training in order to achieve optimal values of the weights to improve the performance of the DNN. The mean square error (MSE) is used as a secondary parameter in our proposed robust algorithm, which tracks the mean square difference between the predicted and actual data during the simulation processes. The value of MSE is given in the range of 0 to 1 and is determined by the relation Eq. (1):

$$\begin{aligned} E = \frac{1}{n}\sum _{i=1}^{n}(d_i - o_i) \end{aligned}$$
(1)

where \(d_i\) represents the expected output, \(o_i\) represents the actual output and n describes the number of outputs.

When optimizing multiple hyperparameters, a multidimensional space is created. In addition, it was necessary to take into account that the individual hyperparameters may interact with each other. To eliminate the mutual influence of the parameters, various optimization algorithms (manual tuning, random search, sequential search, etc.) are used to efficiently traverse the resulting multidimensional space to find the optimal combination of these parameters. In our proposed resilient routing algorithm, the random search method was used. This method allows us to reduce the number of learning instances compared to sequential search. This optimization method was chosen mainly because in our proposed use cases it was necessary to limit the number of learning instances from which randomly chosen hyperparameter tuples were evaluated. However, these parameters needed to be generated to uniformly cover the entire space of such parameters. The generation itself was provided by a random number generator defined by a probability uniform distribution.

To design a resilient routing algorithm in MANETs, it was necessary to determine the input parameters for the AI, based on which a node suitable for routing was determined. Subsequently, a resilient communication path was discovered from such nodes. The method of calculating the parameters V\(_B\)(t), V\(_{CPU}\)(t), V\(_M\)(t), V\(_P\)(t), PDR\(_{dat}\), PDR\(_{sig}\) and PRR has been described in detail below. The evaluation of these parameters was performed through continuous monitoring for the duration of the simulation.

The V\(_M\)(t) parameter represents the computation of the value of RAM used on a node (each node is unique with a unique ID value) within the routing. The relation for calculating V\(_M\)(t) is shown in Eq. (2):

$$\begin{aligned} V_M (t)=1-\left( \frac{Act_M(t)}{Max_M} \right) ~~[MB] \end{aligned}$$
(2)

where Max\(_M\) represents the maximum RAM capacity available for a particular node. This parameter describes the hardware capabilities of each node in the network, in our case this parameter was set to a fixed value of 4096MB. Act\(_M\)(t) represents the actual RAM usage at a given time, which was represented by random values from the range 0 to Max\(_M\)-1.

V\(_{CPU}\)(t) Eq. 3 determines the value of computational power over time for a particular node. The computation of this parameter is very important, especially because of the implemented decentralized blockchain technology and its consensus algorithm PoW, which is the most demanding just for computational power,

$$\begin{aligned} V_{CPU} (t)=1-\left( \frac{Act_{CPU}(t)}{Max_{CPU}} \right) ~~[MHz] \end{aligned}$$
(3)

where Act\(_{CPU}\)(t), describes the current value of computational power over time, which was represented by random values from the range 0 to Max\(_{CPU}\)-1, Max\(_{CPU}\), specifies the maximum computational capacity. This parameter describes the hardware capabilities of each node in the network same as in Max\(_M\), in our case this parameter was set to a fixed value of 3000 MHz.

The parameter V\(_B\)(t) Eq. (4), which describes the energy consumption of a particular node, is also important in the context of computing the consensus PoW algorithm and maintaining the MANET link,

$$\begin{aligned} V_B (t)=1-\left( \frac{Act_B(t)}{Max_B} \right) ~~[W] \end{aligned}$$
(4)

where the parameter Act\(_B\)(t) specifies the actual power consumption, which was determined in different modes, active mode, passive mode, sleep mode and computation mode. In the active mode, a particular node is involved in the communication or routing process, i.e., it sends, receives and forwards routing and data packets, in addition to monitoring the network and performing the activities belonging to the implemented routing protocol. The battery consumption in this mode was set to 25 W. The passive mode describes the state when a particular node is connected to the simulated network but does not perform any communication processes, the value of this parameter was set to 5 W. In computation mode, the mobile node was used to verify the data inserted into the decentralized blockchain technology through the PoW consensus algorithm. Since the PoW property implies that it is demanding the computational power of a particular node and hence on the energy consumption, the value of battery consumption was set to 28 W. Max\(_B\) determines the maximum energy or capacity of the battery, in our case, this parameter was set to a fixed value of 43 890 J i.e. 100%.

We consider the simulated MANET network, so to determine the position of a node and its neighbouring nodes, we need to calculate the ratio between the neighbouring nodes and the current node using X and Y coordinates. The calculation of V\(_P\)(t) Eq. (5) is shown below:

$$\begin{aligned} V_P (t)=\frac{(X_{1p}-X_{2p})(Y_{1p}-Y_{2p})\ldots (X_{np}-X_{n+1p})(Y_{np}-Y_{n+1p})}{Act_P(t)}~~[m;m] \end{aligned}$$
(5)

where the X parameter represents the position of each mobile node at the X-axis, and Y represents the position of each mobile node at the Y-axis. Act\(_p\)(t) represents the actual position of the mobile node on the X and Y-axis.

Within the network parameters needed to determine a resilient communication path, the PDR must be evaluated, PDR\(_{sig}\) Eq. (6) where:

$$\begin{aligned} PDR_{sig} = \frac{\sum {Number \; of \; signalling \; packets}}{\sum {Number \; of \; sent \; signalling \; packets}} \end{aligned}$$
(6)

number of signalling packets describes the number of signalling packets received within a specific node. The number of sent signalling packets describes the number of all signalling packets sent from the source node. Equation (7) describes the packet delivery ratio of data packets, where:

$$\begin{aligned} PDR_{dat} = \frac{\sum {Number \; of \; data \; packets}}{\sum {Number \; of \; sent \; data \; packets}} \end{aligned}$$
(7)

number of data packets describes the number of data packets received within a specific node. The number of sent data packets describes the number of all data packets sent from the source node to the destination node.

To determine the most suitable node for routing and data transmission, it is necessary to determine the PRR parameter, which tells the rate of signalling packets that have not been modified and arrived from the source to the destination node during the routing process. The calculation of PRR Eq. (8) is shown below, where:

$$\begin{aligned} PRR = PRR_{RREQ} + PRR_{RREP} + PRR_{RREE} + PRR_{RACK} \end{aligned}$$
(8)
$$\begin{aligned} PRR_{RREQ} = WC_{RREQ} * \frac{RREQ_{Delivered}}{RREQ_{Send}} * TC_{RREQ} \end{aligned}$$
(9)
$$\begin{aligned} PRR_{RREP} = WC_{RREP} * \frac{RREP_{Delivered}}{RREP_{Send}} * TC_{RREP} \end{aligned}$$
(10)
$$\begin{aligned} PRR_{RREE} = WC_{RREE} * \frac{RREE_{Delivered}}{RREE_{Send}} * TC_{RREE} \end{aligned}$$
(11)
$$\begin{aligned} PRR_{RACK} = WC_{RACK} * \frac{RACK_{Delivered}}{RACK_{Send}} * TC_{RACK} \end{aligned}$$
(12)

WC\(_{RREQ}\) is the weighting factor of the RREQ signalling packet. WC\(_{RREP}\) is the weighting coefficient of the RREP signalling packet, WC\(_{RREE}\) is the weighting coefficient of the Route Error Packet (RREE) signalling packet and WC\(_{RACK}\) is the weighting coefficient of the Reliable Acknowledgement Packet (RACK) signalling packet. The value of these weighting coefficients was 0.25 for each of these coefficients. TC\(_{RREQ}\) is the timing coefficient of the RREQ. TC\(_{RREP}\) is the time coefficient of the RREP. TC\(_{RREE}\) is the time coefficient of the RREE and TC\(_{RACK}\) is the time coefficient of the RACK signalling packet.

Fig. 10
figure 10

Flow chart of inserting and verification data in decentralized blockchain technology

Full size image

These parameters were entered into a decentralized blockchain database (Fig. 10), after the execution of the consensus algorithm, the block of data was distributed along the communication path. In the case of multi-path communication, it was necessary to define rules for inserting blocks into the blockchain database. If communication via multiple communication paths (CP) was used, the entity that inserted the data block into the blockchain database was determined based on the previous communication and the record of the block insertion. If the sum of successful data verification operations (SO) of communication path SO\(_{CP1}\) > SO\(_{CP2}\) holds, then communication path CP1 is responsible for the block insertion process. The principle of decentralized blockchain technology implies that the data thus verified cannot be changed, added or removed from the database, and therefore a given network is more resilient in further routing processes.

A summary of the proposed algorithm (Algorithm 1, Fig. 11) designed for resilient routing has been presented below. The proposed resilient routing algorithm has been implemented in the reactive and proactive routing protocols R-AODV and R-OLSR. In this case, a situation where blockchain technology was distributed in a MANET was considered.

Algorithm 1
figure f

Pseudocode of proposed resilient routing algorithm

Full size image
  1. 1.

    In the case where the blockchain database was distributed among the nodes in the network, the routing itself was provided by a DNN whose input parameters consisted of QoS and technical parameters determined from previous communications.

  2. 2.

    For MANETs, it is important to determine whether communication can take place exclusively via single-path or multi-path links.

    1. (a)

      In the case of a unidirectional link, it was necessary to determine whether the node with which communication can be established is the end node, i.e. the destination node. In this case, the network parameters PDR\(_{sig}\), PDR\(_{dat}\) and PRR were calculated.

    2. (b)

      In the case of multi-path routing, it was necessary to determine whether all the communication paths found met the condition of resilient transmission, if this condition was not met, the extracted data was inserted into the data block. If the given communication paths satisfied the resilient transmission condition and the destination node was found, the same network parameters were determined for single-path communication.

  3. 3.

    After determining the type of communication and calculating the QoS parameters, the technical parameters V\(_M\)(t), V\(_B\)(t), V\(_{CPU}\)(t) and V\(_P\)(t) were determined.

  4. 4.

    The parameters determined in the previous steps were the input parameters of the DNN (Fig. 9). The output of the DNN was a specific node that satisfied the conditions of resilient data transmission and was suitable as a resilient routing node.

  5. 5.

    The information of the selected resilient node was stored in a block in the blockchain database (Fig. 5), which was distributed to the MANET.

  6. 6.

    Based on these parameters, a resilient communication path was selected in terms of the success rate of delivery of the sent packets and the maintenance of the communication path.

Fig. 11
figure 11

Flow chart of a proposed resilient routing algorithm

Full size image

4 Performance evaluation

4.1 Simulation model and parameters

To evaluate and compare the proposed resilient routing algorithm, simulations with different numbers of nodes and the movement speed of each node have been performed as shown in Table 1. In each simulation scenario, the success of the proposed resilient routing algorithm will be compared with the baseline type of reactive and proactive routing algorithms AODV and OLSR. From Table 1, it can be seen that the achieved results of the original routing protocols (AODV and OLSR) and the modified resilient routing protocols (R-AODV and R-OLSR) have been analyzed in terms of different numbers of nodes in the network, where the simulation scenarios ran from 20 to 100 nodes in the network, and in terms of the movement speed of the nodes, where 5 simulation scenarios were tested with the speeds of 0-2, 2-4, 4-6, 6-8 and 8-10 m/s. The Monte Carlo method was chosen as the model to evaluate the results, which ensures the simulation of random movements of mobile nodes as well as random obstacles and signal interference by which the proposed resilient routing algorithm can be analyzed more accurately. The actual implementation of the proposed model was carried out in Network Simulator 3 (NS3) and through the Python programming language into which all the necessary libraries were integrated.

Table 1 Simulation parameters
Full size table

4.2 Analyzed metrics

The parameters PDR\(_{sig}\), PDR\(_{dat}\), Rate of signalling and data packets, Throughput and E2E delay were used to evaluate the proposed resilient routing algorithm for all simulation scenarios.

The parameter Throughput speaks about the amount of data that has been transmitted in a certain time interval within the network. Throughput Eq. (13) is calculated as:

$$\begin{aligned} Throughput = \frac{\sum D_R}{t}~~[Mb/s] \end{aligned}$$
(13)

where D\(_R\) is the amount of data received by the destination node and t is the simulation time.

The parameter PDR\(_{sig}\) Eq. (6) tells the amount of signalling packets that were received from the source node to the destination node compared to the signalling packets that were sent. Thus, this parameter tells about the success rate of packet delivery. Similarly, the parameter PDR\(_{dat}\) Eq. (5) tells about the delivery success rate of the sent data packets.

The ratio between signalling and data packets describes the transmission success rate and the sustainability of the communication path, thus in the simulation, the routing protocol should achieve a decreasing tendency for signalling packets and an increasing tendency for data packets. The fewer signalling packets needed to maintain and find the communication path due to more data packets transmitted will be obtained. The ratio between signalling packets and data packets is calculated as Eq. (14):

$$\begin{aligned} SOR = \frac{S_P}{P} * 100~~[\%] \end{aligned}$$
(14)

where SOR is the Signalling Overhead Ratio, \(S_p\) is the number of signalling packets and P is the total number of packets.

The last parameter required to evaluate the proposed resilient routing algorithm is the E2E delay. This parameter speaks about the time required for packets to arrive from the source node to the destination node. The E2E delay Eq. (15) can be calculated in several ways, for example, using the TraceRoute protocol, based on the packet delay, where the packet sending time and the packet receiving time are taken into account, or using timestamps. In our case, this parameter is calculated using timestamps because it includes the time required for packet processing by each node. The way this principle works is that at the time a packet is sent, the t\(_{send}\) parameter is added to the packet and at the time a packet is received by a node, the t\(_{received}\) parameter is added to the packet. This method guarantees the computation of the time required to send and receive a packet between two nodes. This procedure is repeated for each packet that is sent by the source node to the destination node. It is calculated as:

$$\begin{aligned} E2E = t_{send}^i - t_{received}^i~~[s] \end{aligned}$$
(15)

where t\(^i_{send}\) is the time of sending the i-th packet and t\(^i_{received}\) is the time of receiving the i-th packet.

4.3 Experimental results

The experimental results analyzed the parameters PDR\(_{sig}\), PDR\(_{dat}\), the ratio between data and signalling packets, throughput and E2E delay. The proposed resilient routing algorithm implemented in the reactive and proactive routing protocols R-AODV and R-OLSR was compared with the routing protocols AODV and OLSR in which this algorithm was not applied. The achieved results were analyzed based on the change in the number of nodes and the change in the mobility rate of nodes in the network. These simulations aimed to analyse the network parameters in extreme situations where it is necessary to select and communicate through resilient nodes depending on the network parameters.

The first parameter analyzed was the parameter of the number of signalling packets delivered. The achieved average results of the delivered signalling packets versus mobility and the number of individual nodes in the network showed that during experimental simulations in different scenarios with different values of node mobility, the AODV and OLSR protocols behaved in the expected way, for a network with low to no mobility, the proactive routing protocol OLSR, which was designed specifically for this type of networks, achieved better results. The signalling packet rate of the OLSR protocol varied with increasing node mobility rate from 78.11% for a network with 60 nodes at 0-2 m/s to 79.99% for a network with 100 nodes at 8-10 m/s, these results showed us that for static networks and networks with moderate mobility, this protocol was the most efficient, i.e., it needed less signalling packets to establish and maintain a communication path. The reactive routing protocol AODV exhibited the opposite behaviour and thus achieved the best results for larger mobility values. The rate of signalling packets varied from 78.71% for a network with 80 nodes at 0-2 m/s to 77.61% for a network with 60 nodes at 8-10 m/s.

The R-AODV protocol with the implemented resilient routing method achieved better average signalling packet rate results compared to the AODV protocol. The achieved results ranged from 77.55% to 75.01% with increasing node movement speed. For a network with a movement rate of 0 to 2 m/s, the smallest improvement of 0.46% was observed with 100 nodes in the network and the largest difference was achieved in a network with 80 nodes, 1.51%. For increased node movement speed of 2-4 m/s, the smallest difference of 1.71% between R-AODV and AODV protocols was achieved in networks with 60 and 100 nodes. The largest improvement was achieved in the network with 40 nodes, 2.32%. Similar results were obtained in simulations with movement speeds of 2-4 m/s and 4-6 m/s. The largest difference was obtained in the networks with 40 knots at 2.4% and the smallest difference at 1.8% in the networks with 100 knots. In the simulations with speeds of 6-8 m/s, the results were different, the largest difference of improvement of 2.36% was achieved in the networks with the smallest number of nodes and the smallest difference of improvement of 1.7% was achieved in the networks with 100 nodes. In the 8-10 m/s networks, the same improvement results were achieved for the 80 and 100 node networks at 2.74%, the network occupied by 60 nodes achieved 1.96%.

In a similar analysis of the achieved results of the signalling packet rate of the R-OLSR routing protocol with the resilient routing, method implemented compared to the OLSR protocol, an improvement of 4.12% was achieved in a network with 40 nodes and the smallest difference was analyzed in a network with 60 nodes at 3.09% at a movement rate of 0-2 m/s. In the case of the network with a node movement speed of 2-4 m/s, the average improvement results were at a lower level compared to the results for a movement speed of 0-2 m/s. The largest difference between the R-OLSR and OLSR protocols was obtained in networks with 20 nodes at 2.24% and the smallest at 1.69% in a network with 60 nodes. In networks with movement speeds at the 4-6 m/s level, an improvement of 2.44% in networks with 20 nodes and an improvement of 1.9% in networks with 80 nodes was analyzed. In networks with a movement speed of 6-8 m/s, an improvement of 1.78% to 2.82% was obtained with an increasing number of nodes in the network, with the lowest average measured improvement at 40 nodes in the network and the best improvement at 100 nodes in the network. Average improvements of 2.3% and 1.7% for networks with 20 and 80 nodes, respectively, were achieved at node movement speeds of 8-10 m/s.

Table 2 shows the average PDR\(_{sig}\) parameter improvement results achieved for all test scenarios. The achieved results describe the average improvement of the R-AODV and R-OLSR protocols against the AODV and OLSR protocols, and also the comparison of the resilient protocols with each other.

Table 2 Achieved results of average improvement of PDR\(_{sig}\) depending on the mobility of mobile nodes [%]
Full size table

As part of the analysis of the achieved results of the PDR\(_{sig}\) parameter, the results of the PDR\(_{dat}\) parameter of the number of delivered data packets were also analyzed. This parameter was also analyzed concerning node mobility and the number of nodes in the network.

Analysing the obtained results of the PDR\(_{dat}\) parameter for the R-AODV and AODV protocols, results were obtained showing that for a network with a movement speed of 0-2 m/s, the largest difference of 4.14% between these protocols was in networks with 20 nodes, and the smallest difference of 1.97% was in networks with 60 nodes. For a speed of 2-4 m/s, the maximum and minimum improvement results were obtained at the extremes, that is, the smallest difference of 2.18% was observed in networks with 100 nodes, and the largest of 3.64% for 20 nodes. For 4-6 m/s networks, an almost identical result of maximum improvement was recorded in 20 nodes networks at 3.44%. On the contrary, the smallest difference of 2.64% was measured in networks with 40 nodes. In simulations of networks with movement speeds of 6-8 m/s and 8-10 m/s, the maximum and minimum values were measured identically in networks with 40 and 100 nodes. For the networks with 6-8 m/s movement speed, the values were 2.34% and 3.32%, respectively; the networks with 8-10 m/s movement speed achieved average values of 2.64% and 3.52%, respectively.

From the results obtained for the PDR\(_{dat}\) parameter for the R-OLSR and OLSR routing protocols, it was found that the largest difference between these protocols was obtained in networks with 20 nodes for all movement speeds with a vignette of 8-10 m/s. The smallest amount of difference was measured in networks with 100 nodes for speeds of 2-4 m/s, 4-6 m/s and 6-8 m/s. In the network with 0-2 m/s speed, the largest difference of 3.76% and the smallest difference of 1.63% was obtained in the network with 20 nodes. The network with 2-4 m/s speed achieved almost identical results at the highest measured difference of 3.66%, the minimum difference between the analyzed protocols was 1.9% in the network with 100 nodes. The 4-6 m/s and 6-8 m/s networks achieved their maximum and minimum differences identically with 20 and 100 nodes, respectively. In the networks with 20 nodes, values of 3.46% and 3.76% were achieved, and in the networks with 100 nodes, results of 2.44% and 2.66% were achieved. In the 8-10 m/s test scenario, the maximum difference of 3.33% was obtained in the network with 80 nodes and the minimum difference of 3.11% in the network with 20 nodes.

The average results of the improvement of the PDR\(_{dat}\) parameter are shown in Table 3. Similar to the PDR\(_{sig}\) parameter, the results with unmodified routing protocols, but also across the proposed resilient algorithm are analyzed.

Table 3 Achieved results of average improvement of PDR\(_{dat}\) depending on the mobility of mobile nodes [%]
Full size table

One of the parameters to determine the efficiency of the network is the ratio between signalling and data packets. This parameter describes the number of signalling packets required to establish and maintain a communication path and the amount of data packets transmitted from the source to the destination node. Thus, the network is more efficient if the number of signalling packets is reduced and the number of data packets is increased, or if the number of signalling packets is as low as possible and the number of data packets is as high as possible. Figures 12, 13, 14, 15 and 16 show the results obtained for this parameter as a function of mobility and the number of nodes in the network. The signalling packets showed a decreasing trend and the data packets showed a stagnant to slightly increasing trend. For all the parameters we analyzed, the total time of the experimental simulations was also important. Since the distribution of the blockchain technology, DNN input parameters and routing parameters was performed using the implemented resilient algorithm, it can be concluded that with increasing simulation time, the R-AODV and R-OLSR protocols could perform better as the information about the nodes suitable for routing and data transmission would be distributed in the network.

Fig. 12
figure 12

Average ratio of signalling and data packets depending on the mobility of mobile nodes with a movement speed of 0 - 2 m/s

Full size image
Fig. 13
figure 13

Average ratio of signalling and data packets depending on the mobility of mobile nodes with a movement speed of 2 - 4 m/s

Full size image
Fig. 14
figure 14

Average ratio of signalling and data packets depending on the mobility of mobile nodes with a movement speed of 4 - 6 m/s

Full size image
Fig. 15
figure 15

Average ratio of signalling and data packets depending on the mobility of mobile nodes with a movement speed of 6 - 8 m/s

Full size image
Fig. 16
figure 16

Average ratio of signalling and data packets depending on the mobility of mobile nodes with a movement speed of 8 - 10 m/s

Full size image

The throughput parameter describes the amount of data that can be transferred to the network in a certain amount of time. The results obtained for this parameter, shown in Figs. 17, 18 and 19, were analyzed as well as the parameters PDR\(_{sig}\), PDR\(_{dat}\), and the ratio between signalling/data packets, as a function of the number and mobility of nodes in the network. On average, the R-AODV and R-OLSR protocols performed better compared to the AODV and OLSR protocols.

Based on the results obtained, it was possible to observe a clear trend in the behaviour of the different routing protocols. In Fig. 17, which shows the achieved throughput results as a function of the number of nodes in the network at 0-2 m/s, it can be seen that the R-OLSR routing protocol achieved higher throughput values compared to R-AODV, a similar trend could be observed for the OLSR and AODV protocols. In this test scenario, a maximum difference of 24.38% and 21.49% was achieved between the R-OLSR and OLSR protocols and between the R-AODV and AODV protocols, respectively. The minimum improvement between the R-OLSR and OLSR protocols was achieved at 15.29%. The R-AODV protocol achieved a minimum network improvement rate of 12.17%. From these results, it could be concluded that the R-AODV protocol generally performed worse in networks with little or no node movement compared to the R-OLSR protocol, which performed better in these networks.

A similar trend could be observed in Fig. 18, where the permeability parameter was analyzed at a moving speed of 2-4 m/s. In general, as the movement speed of nodes in the network increases, the value of throughput decreases because it is more difficult to establish and maintain the established communication path. Nevertheless, the R-OLSR and Fig. 19 decreased as the number of nodes in the network increased. In networks with this node movement rate, the difference between these protocols was 2.72% in favour of the R-OLSR protocol. Based on these results, it can be said that with increasing node mobility rate, the R-AODV protocol performed better than the R-OLSR protocol, symmetric results and trends could also be observed for the OLSR and AODV protocols.

Fig. 17
figure 17

Dependence of throughput on the mobility of mobile nodes with movement speed 0 - 2 m/s

Full size image
Fig. 18
figure 18

Dependence of throughput on the mobility of mobile nodes with movement speed 2 - 4 m/s

Full size image

In Figs. 20 and 21, it could be observed that the throughput decreased for all the routing protocols tested, but for increasing movement speeds of 6-8 m/s and 8-10 m/s, the R-AODV protocol performed the best among all the protocols we tested. In these scenarios, the R-AODV routing protocol achieved an average improvement of 17.51% over the AODV protocol and the R-OLSR protocol achieved 24.38% over the OLSR protocol.

Fig. 19
figure 19

Dependence of throughput on the mobility of mobile nodes with movement speed 4 - 6 m/s

Full size image
Fig. 20
figure 20

Dependence of throughput on the mobility of mobile nodes with movement speed 6 - 8 m/s

Full size image
Fig. 21
figure 21

Dependence of throughput on the mobility of mobile nodes with movement speed 8 - 10 m/s

Full size image

On average (Table 4), the R-AODV and R-OLSR protocols performed better compared to the AODV and OLSR protocols. Similar to the parameters analyzed above, R-AODV performed better with higher mobility and a higher number of nodes in the network. R-OLSR achieved better results in fewer mobile networks and networks with fewer nodes.

Table 4 Achieved results of average improvement of total throughput depending on the mobility of nodes [%]
Full size table

The E2E delay describes the total time required to deliver data from the source node to the destination node. This parameter, like all the parameters analyzed, was evaluated based on the change in the number of nodes in the network and the movement speed of each node. Figures 22 and 23 show the average results obtained for the simulation scenarios with movement speeds of 0-2 m/s and 2-4 m/s, respectively. At these speeds, we can observe a trend where the R-OLSR and OLSR routing protocols achieved lower average delay values compared to the R-AODV and AODV protocols, however, even in these simulation scenarios, it is shown that the R-AODV protocol achieved lower delay values than the OLSR protocol, even though in this case it was a static to moderately dynamic network. The maximum delay improvement of the R-AODV protocol over the AODV protocol was 12.39% and the improvement of the R-OLSR protocol over the OLSR protocol was 15.83%.

Fig. 22
figure 22

Dependence of total E2E delay on the mobility of mobile nodes with movement speed 0 - 2 m/s

Full size image
Fig. 23
figure 23

Dependence of total E2E delay on the mobility of mobile nodes with movement speed 2 - 4 m/s

Full size image

At a movement speed of 4-6 m/s (Fig. 24), the results showed that the AODV and OLSR protocols achieved higher delay values, of which the AODV protocol achieved the highest values as the number of nodes in the network increased. The R-AODV and R-OLSR protocols achieved similar values, and a gradual change in trend could be observed when the R-OLSR protocol started to achieve higher average delay values compared to the R-AODV protocol.

Fig. 24
figure 24

Dependence of total E2E delay on the mobility of mobile nodes with movement speed 4 - 6 m/s

Full size image

Figures 25 and 26 show the average values at knot movement speeds of 6-8 m/s and 8-10 m/s, respectively. These graphs show the opposite trend, with the R-AODV protocol achieving lower average values than the R-OLSR and OLSR protocols. The maximum improvement of the R-AODV protocol over the AODV protocol was 14.16% at 6-8 m/s. The R-OLSR routing protocol achieved a maximum improvement of 13.44% in the 8-10 m/s network.

Fig. 25
figure 25

Dependence of total E2E delay on the mobility of mobile nodes with movement speed 6 - 8 m/s

Full size image
Fig. 26
figure 26

Dependence of total E2E delay on the mobility of mobile nodes with movement speed 8 - 10 m/s

Full size image

Based on the analyzed E2E delay results, it could be concluded that the resilient routing protocols R-AODV and R-OLSR achieved improvements in all simulation scenarios. In networks with lower mobility speeds, the R-OLSR and OLSR protocols achieved more acceptable results, but we could observe the beginning of a trend change at mobility speeds of 4 - 6 m/s. As the movement speed of nodes in the network increased and the number of nodes in the network increased, the R-AODV and AODV protocols performed better. The complete results of the improvement in average delay are shown in Table 5.

Table 5 Achieved results of average improvement of total E2E delay depending on the mobility of mobile nodes [%]
Full size table

5 Conclusion and future work

Due to the dynamic nature of MANETs, which feature highly mobile nodes and fluctuating network sizes, challenges arise in routing, maintaining communication paths, and processing data. The unstable network topology results in frequent path changes, reducing network efficiency and necessitating re-routing. This leads to increased signaling packets and energy consumption, making the network less stable. To address these issues, a novel resilient routing algorithm is proposed in this study. It uses decentralized blockchain and DNN technologies in conjunction with routing protocols. The decentralized blockchain technology serves as a repository for the data obtained from the simulation processes, which are immutable precisely due to the tensor nature of the individual data blocks and the PoW consensus algorithm that provides authentication of the data embedded in these blocks. The DNN analyzes historical behavior and current technical and QoS parameters to select nodes for routing and data transmission. Experimental simulations implementing this algorithm in R-AODV and R-OLSR routing protocols show improved QoS parameters compared to conventional AODV and OLSR protocols.

Future research will focus on optimizing the proposed solution and extending it. Since (Table 6) our proposed resilient routing algorithm is based on NNs, this fact gives us a scope to optimize the proposed algorithm at the level of determining the number of hidden layers and neurons using different methods, for example, a heuristic or statistical method. Furthermore, the proposed algorithm will be applied to C-MANET environments to enhance communication resilience within and between multiple C-MANETs, allowing extensive comparison in different extreme situations.

Table 6 Notation and explanations
Full size table