فصلنامه پدافند الکترونیکی و سایبری
سال دهم شماره 4 (پیاپی 40، زمستان 1401)

The presence of misbehaving users in Cognitive Radio Networks (CRN) can disrupt the process of spectrum sensing and detecting the status of the Primary User (PU). In order to reduce the destructive effect of this group of users in CRNs, in this paper, a new mechanism based on reinforcement learning for cooperative spectrum sensing is presented. The proposed method is a cooperative spectrum sensing mechanism based on user weighting, according to which users receive a weight commensurate with how they behave in spectrum sensing. The reinforcement learning model used in the proposed method is a learning automata which, using reward and penalty processes, allocates more weight to users with normal behavior in sensing the spectrum and less to misbehaving users. In this way, the learning automata updates the users' weight vector based on the response received from the environment, after performing a sensing operation in each repetition. After repeating the sensing operation several times, the learner will be able to optimize the user's weight vector. In order to evaluate the proposed method, its performance in the simulation environment has been tested and the results have been compared with the existing method for cooperative spectrum sensing. The results show that using the proposed method in the presence of misbehaving users will significantly improve network performance.

Directional modulation(DM) is an emerging technology for securing wireless communication at the physical layer and is mostly used in the line of sight propagation channels such as millimeter wave communications, next-generation cellular, satellite, and radar networks. This promising technology, unlike key-based cryptographic methods and the key-based physical layer security approaches, locks information signals without any requirement of keys. The locked information can only be fully recovered by the legitimate receiver(s) priory known by DM transmitters. This technology can be implemented in different ways. In the phased array method, modulation is implemented in the antenna. In this method by changing the phase of each element, the angle-dependent modulation is constructed. for this reason, the modulation constellation points in the undesired directions are distorted and deviate from the standard mode.In this article, DM is implemented in the baseband by using the orthogonal vector method and separating the information radiation pattern from the interference radiation pattern (random artificial noise). This is a new method for simultaneously sending several signals with DM and random artificial noise. In this method, the bit error rate (BER) probability in the direction of the legitimate receiver(s) is improved from 10^(-3) to 10^(-5) and at least the signal secrecy rate is also increased one bit per second per hertz (bandwidth unit). Also, the results of the simulations show that as the number of antennas increases, the secrecy rate increases, the amount of power allocated to artificial noise decreases, and the power efficiency of the transmitter increases.

The spoofing attack is one of the most serious interferences in the Global Positioning System (GPS). By propagating a signal structurally similar to the original GPS signal, the spoofers try to influence the function of different parts of the receiver and force it to make a wrong positioning. This study focus on the acquisition stage. During the acquisition process, GPS receivers estimate the values of Doppler frequency and Pseudo Random Noise (PRN) code phase of the received signal, which are necessary for tracking the GPS satellite signals. One of the effects of the spoofing signal in the acquisition unit of the receiver is to increase the interactions in the Quadrate correlation taps (Q-correlation tap). In 2018, adding a denoising unit on the Q-correlation tap in the acquisition stage to reduce the interactions mentioned above was presented as a spoofing mitigation method. In this paper, the mentioned method is placed as the primary basis of the work. Here, by using powerful methods of evolutionary computing, the denoising unit added in the Q-correlation tap is tried to be optimally adjusted to mitigate the spoofing attack. Specifically, to achieve a more efficient denoising method for spoofing mitigation, the Particle Swarm Optimization (PSO) algorithm is proposed to determine the critical parameters of the Discrete Wavelet Transform (DWT) based on the Haar wavelet. In order to evaluate the proposed method, first, the noise reduction performance of the algorithm is measured on four benchmark signals, namely Blocks, Bumps, Heavy Sine, and Doppler. Then, compared to four traditional methods, namely, Rigrsure, Heursure, Sqtwolog, and Minimaxi, the developed de-nosing method outperformed the former methods by 47.3%, 38.4%, 47,3%, and 30%, respectively. Finally, the proposed algorithm was placed in the Q-correlation tap of the GPS receiver acquisition stage, and its performance in reducing the spoof effects was investigated. The results show that the proposed algorithm is 37.74% more efficient compared to the method that was considered the primary method.

Nowadays, detecting unusual events in the network has been the subject of many researches. Network traffic is huge and very large, and this leads to high data size and increased noise, which makes it very difficult to extract meaningful information to detect abnormal events. Early detection of attacks improves the stability of a system. Each attack is a type of specific behavior; But some attacks may behave similarly and differ only in some features. This article presents a new way to detect malware and attacks in the cloud computing environment. In this method, data clustering separates the data from each other to provide better conditions for model construction by balancing the data in different classes. This research uses a combination of Adabost, Random Forest and Bosted Gradient Tree algorithms as ensemble learning to improve malware detection in cloud computing. In order to combine boosted learners and build a higher level model, the voting mechanism is used. In the proposed model, ensemble learning, using the strengths of various algorithms, creates a useful, high-performance system for detecting malware in cloud computing. By applying the proposed method on real data, it was observed that the accuracy of the proposed method is equal to 99.96%, its accuracy is equal to 99.97% and its recall is equal to 99.95% which compared to previous methods, it has a noticeable advantage, but its computational complexity has not changed much.

In today's digital era, security issues and cyber attacks have become a serious and attention-needed concern as they hamper secured and vital information relating to organizations or individuals. Accordingly, timely detection of these vulnerabilities made by intruders is essential, wherein the cornerstone of security ensures the user's data privacy as an intrusion detection system (IDS). On the other hand, with the rapid development of machine learning (ML) and deep learning (DL) methods in the data world, one of their significant applications is dedicated to IDS using state-of-the-art classification algorithms, which has been the subject of numerous research to enhance accuracy and reliability in recent years. As a consequence, this paper presents a hybrid model integrating feature selection, classification, and hyper-parameters optimization. First, the initial massive features are subjected separately to the modified mutual information (MMI), genetic algorithm (GA), and Anova F-value approaches, followed by extracting the common outputs as optimal and reduced final features. Subsequently, a compound CNN and LSTM classifier (CNN-LSTM) is employed, where its hyper-parameters will be determined through a random switch grey wolf-whale optimization algorithm (RS-GWO-WOA) instead of a time-consuming trial and error manual process. Ultimately, to analyze the suggested scheme, a comparison with other strategies in terms of accuracy, precision, recall, F1 score, and periods of time on the NSL-KDD dataset has been accomplished, confirming the superiority of the developed approach.

In recent years, cloud computing has attracted much attention as a new computing model for providing infrastructure, platform, and software as a service. There is an important challenge in trust management between cloud providers, service providers, and service applicants due to the industry's rapid adaptation of cloud computing. Trust management has become very challenging in cloud computing since cloud service applicants need to choose effective, reliable, and low-risk services. One of the most important factors, which can be considered in the applicant's trust or distrust of service, is the various parameters of service. Therefore, it is necessary to use approaches to evaluate the trust of the cloud services considering their service quality parameters and their identified performance requirements. In this paper, a model is introduced to evaluate the trust for the cloud services using Bayesian network. Since the trust actually deals with probabilities, and the Bayesian network also uses probabilities to solve the problems, the Bayesian network can be used to assess the trust. The proposed method is compared with various data mining techniques to assess trust in cloud services. The results show that the accuracy, absolute error, root mean square error, and square error in the Bayesian network are 94/53, 0/037, 0/137, and 0/038, respectively. The proposed method is more efficient than different data mining techniques for trust assessment in cloud services.

In this paper, a new algorithm called Motion Coding Gravitational Search Algorithm (MGSA) is proposed to find a moving target using a unmanned aerial vehicles (UAVs). Using the laws of physics and the properties of the earth, each dimension has its own equation of motion based on the type of variable. Many traditional exploratory methods can not achieve the desired solution in high-dimensional spaces to search for a moving target. The optimization process of the gravitational search algorithm, which is based on the gravitational interaction between particles, the dependence on the distance and the relationship between mass values, and the fit calculation, make this algorithm unique. In this paper, the proposed MGSA algorithm is proposed to solve the path complexity challenge problem in order to find the moving target through motion coding using UAVs. A set of particles in the path of search for the target will reach a near-optimal solution through the gravity constant, weight factor, force and distance, which evolved with many search scenarios in a GSA algorithm. This coded method of motion makes it possible to preserve important particle properties, including the optimum global motion. The results of the existing simulation show that the proposed MGSA improves the detection performance by 12% and the time performance by 1.71 times compared to APSO. It works better.

In this paper, we consider the problem of constructing the region-fault tolerant geometric spanners when the problem is restricted to a subclass of convex regions. Let S be a set of n points in the plane. In particular, in this paper, a greedy algorithm for constructing the region-fault tolerant geometric spanner of S where the region faults are a set of at most k half-planes with parallel boundaries is presented. We show that the proposed algorithm has the time complexity O(kn^3 log⁡n ), and the generated graph contains O(kn) edges. To the best of our knowledge, the best-known algorithm to construct the region-fault tolerant geometric spanner of S takes O(n log^2⁡n) time and the generated graph has O(n log⁡n) edges.

Detection of dangerous objects in images obtained by X-ray scanners in security inspections has played an important role in protecting the public space from security threats such as terrorism and the occurrence of dangerous crimes. Perform diagnostic operations by an expert despite the remarkable features of the human sensory and visual systems; Due to being exhausting, non-stop, excessive dependence on human error, etc., it has low operational value. One suitable solution for similar situations is to use car vision systems. In this study, we intend to first examine the hazardous object in the x-ray images in the SIX-ray database in a training phase with hard segmentation, and by extracting the properties of these objects by the SURF algorithm, which is capable of extracting properties even in complex conditions. It is confusing to create a database of properties of objects in different dimensions and angles. Then, in the detection phase, the experimental image first goes through a soft segmentation step, and then the image properties are extracted by the SURF algorithm. The extracted properties are matched with the properties of the object in the training database, and then the probability of the object being present, which is the ratio of the number of matching properties of the object to the total number of properties in the object, is calculated for each case. be. After finding the most likely valid matches, the M-estimator sample consensus algorithm (MSAC) removes the incorrect matching properties that originated from the image background. Finally, a two-dimensional transfer (Affine transformation) is obtained between the pairs of matching points of each valid state with the input image, and with the help of this transfer and dimensionality, a square is drawn around the object and the location of the object is identified. The following is a complete description of the training and diagnosis phase and the results of SIX-ray data.

In recent years, due to the profitability of the stock market in Iran, small and large investments were attracted to this market, but unfortunately, due to their lack of knowledge of the stock market and price forecasting, a large number of Iranians suffered great losses. In this study, we decided to use our previous research, which used a two-layer LSTM neural network, to strengthen its work and use a combination of convolution and lstm neural networks to predict stock prices on the Web Nation data set from the stock market. Use Tehran and its three databases, including ASP, car and construction. Finally, in order to evaluate the proposed method and the other two methods, three error functions, mean square error function (MSE), mean absolute error function (MAE) and root mean square function (RMSE) were evaluated. The results showed that it works much better in large datasets with high stock data and leads to fewer errors.

Quantum computers have much more computing power than classical computers and this has created a challenge in the field of public-key cryptography algorithms, which is predicted quantum computers will reach the computational power to break existing public-key cryptography algorithms by 2030. To solve this problem, NIST published a call for post-quantum cryptography algorithms. Implementing these algorithms faces challenges such as execution time and resources. One of the algorithms that made it to the third round is the CRYSTALS-KYBER algorithm. In this algorithm, by optimizing the NTT module, the execution time is reduced. Usually, the implementation of NTT is created with radix-2, but in the proposed method, radix-4 is used, and this reduces the execution time. Changes to NTT are required to implement radix-4 NTT. DIT is used to implement NTT and DIF is used to implement INTT. In NTT and INTT formulas changes are made to the twiddle factors and the values of the twiddle factors stored to the ROM. In the following, we compared radix-4 butterfly unit with radix-2 butterfly unit. By reusing results in CT and GS butterfly units, we need four multiplications, additions, and subtractions, and the structure of radix-4 butterfly unit is mentioned. The memory unit uses eight RAMs to increase read and write speeds, four of which are for writing and the remaining four are for reading. It is also necessary to make corrections to the NTT parameters which are suitable for implementation on Kyber. Next, we implemented the proposed method on two FPGA Artix-7 and Virtex-7 using Vivado software. In the implementation on Artix-7 and Virtex-7 in exchange for a slight increase in the resources, the execution time in Artix-7 is reduced by 28.74% and 12.34% compared to similar implementations.