Long-range and low-power communications are suitable technologies for the Internet of things networks. The long-range implies a very low signal-to-noise ratio at the receiver. In addition, low power consumption requires reduced signaling, hence the use of less complex protocols, such as ALOHA, so reduced communication coordination. Therefore, the increase of objects using this technology will automatically lead to an increase in interference. In this letter, we propose a detector for Long Range (LoRa) networks based on an autoencoder for denoising and dealing with the interference, followed by a convolutional neural network for symbol detection. Simulation results demonstrate that the proposed approach outperforms both the convolutional neural network-based detector and the classical LoRa detector in the presence of interference from other LoRa users. The proposed detector shows around 3 dB gain for a target Symbol Error Rate (SER) of 10−4.

Validation measurements for the Future Railway Communication System in railway environments is a very challenging task and should be reduced to a minimum for cost and time efficiency. “Zero-on-site testing” consists of using simulation/emulation tools and testing procedures to allow validation and verification activities in the laboratory to avoid complex and expensive trials with trains on real-world sites. A solution to test a communication network in a laboratory under realistic conditions consists of injecting perturbations at the IP level (such as additional delay, packet losses, or jitters). It is essential to correlate the IP impairments with the radio environment, their effects on end-to-end transmission, and how the network and its elements react. To do so, IP impairments (or the conditions that lead to them) should be generated in such a way that allows assessing their impact on the performance of transmissions. This paper presents the results of an experimental research based on an original emulation platform (Emulradio4Rail platform), able to emulate and test wireless systems taking into account both physical layer as well as IP traffic in realistic railway environments. Different types of IP traffics are injected at the application layer and go through the platform. The work aimed at extracting various statistical distributions of classical IP metrics (delay, packet loss, jitter, throughput) versus time, as a function of radio channel conditions for Wi-Fi and LTE transmissions in typical railway environments. Then, the modeled IP impairments statistical distributions can be considered at the IP level to test very easily wireless system modems. The results and methodology can be considered for the evaluation of the Future Railway Mobile Communication System modems.

Spectrum sensing based on Beamforming, like others classification problem, require feature selection to perform learning algorithms and enhance the classification task. This paper proposes a novel version of the Dingo Optimization Algorithm (DOA) to optimize feature selection for a Deep Neural Network (DNN) classifier. Two improvements are introduced to avoid the premature convergence problem and stagnation in the local optima of the original DOA. First, the chaos strategy is executed to produce a high level of diversification in the algorithm, which improves its ability to escape from potential local optimums. Second, the weight factor is introduced to boot up the search process to the global optima. Here, the aim is to improve the DOA for feature selection in the deep learning approach in order to enhance the performance of blind spectrum sensing based on Beamforming in the context of cognitive radio (CR). Through simulations results, we illustrate that our algorithm, called Chaotic Dingo Optimization Algorithm (CDOA), outperforms the original one and a set of state-of-the-art optimization algorithms (i.e., HS, BBO, PSO, and SA) for feature selection in the learning approach.

The Internet of Things faces a significant scaling issue due to the rapid growth of the number of devices and asynchronous communications. Different technologies in the license-free industrial, scientific, and medical (ISM) band have been widely deployed to fill this gap. LoRa and Sigfox are the most common. Many devices can use the ISM band if they obey the regulations and cope with internal and external interference. However, when there is massive connectivity, the effect of the inter and intra-network interference between multiple networks is significant. This study uses a deep learning-based technique to decode signals and deal with the interference in the uplink of a LoRa network. Two classification-based symbol detection methods are proposed using a deep feedforward neural net-work (DFNN) and a convolutional neural network (CNN). The proposed receivers can decode the signals of a selected user when many LoRa users transmit simultaneously using the same spreading factor over the same frequency band (intra-spreading factor interference), and multiple Sigfox users interfere (inter-network interference). Simulation results show that both receivers outperform the conventional LoRa receiver in the presence of interference. For a target symbol error rate (SER) of 10 −3 , the proposed DFNN and CNN-based receivers attain around 2 dB and 3.5 dB gain, respectively.

In wireless communication systems, in order to respond to the perception of risks related to electromagnetic field exposure and allocate radio resources, the estimation of the received power and exposure map is an essential task and a challenge. This paper proposes an algorithm for estimating electromagnetic field exposure maps using U-net architecture based on convolutional neural networks. The power map estimation is transformed into an image reconstruction task by image color mapping, where every pixel value of the image represents received power intensity. The designed model learns wireless signal propagation characteristics in a realistic indoor environment while considering various positions of the Wi-Fi access points. Results show that indoor propagation phenomena and environment models can be learned from data producing an accurate power map to measure the electromagnetic field.

In spectrum sensing issues, like in any other classification problem, the performance of the classification task is significantly impacted by the feature selection. This paper proposes a new hybrid optimization algorithm to optimize feature selection for a Deep Neural Network (DNN) classifier. To surpass the premature convergence problem and improve the exploitation ability of the original Black Widow Optimization Algorithm (BWO), we mix a modified version of BWO and Particle Swarm Optimization (PSO), called MBWPSO. The aim is to enhance the performance of a blind spectrum sensing approach in the context of cognitive radio (CR) for wireless communications. Computer simulations show that the MBWPSO algorithm outperforms the original one and a set of state-of-the-art algorithms (i.e., HS, BBO, PSO, and SA) algorithms. The MBWPSO also exhibits the best performance once applied for feature selection in the above context

The attractive properties of ergodicity, unpredictability, and initial state sensitivity have made chaotic maps the go-to tools in many applications, including cryptography and cyber–physical systems. Despite this, two challenges arise when using chaos systems in cryptography: (i) some one-dimensional (1D) chaotic maps do not satisfy the unpredictability property, and (ii) although they exhibit a complex and chaotic behavior, high-dimensional (HD) chaotic maps incur higher computational complexity. To address these issues, this paper proposes a new 1D chaotic map dubbed the improved Sine-Tangent map (IST map) that is derived from of the Sine map, a tangent function, and a Chebyshev polynomial of the first kind. Relative to chaotic maps, the proposed IST map provides better unpredictability and ergodicity, a vast chaotic range, enhanced complex behavior, and competitive computational complexity. Based on the IST map, we also introduced an encryption scheme for securing medical images in telemedicine. It consists of two diffusion phases, i.e., a bitwise XOR operation and a bitwise expanded XOR (eXOR) operation, with an intermediated confusion one, i.e., random circular-shift. An overriding step is foremost first completed before performing these cryptography phases, i.e., key generation. The secret key of the IST map is updated using the sine and cosine values of the sum of pixels of the input image. This leads to a unique secret key for each image. That is, one-time chaotic sequences are produced for each input image. The cyclic pattern of the sine and cosine values of the sum of pixels provides a prominent sensibility to small changes in the input image. Thus, the proposed algorithm is capable of resisting any chosen/known plaintext attacks. A performance analysis shows that the proposed algorithm outperforms a set of state-of-the-art comparison algorithms and its variants based on Sine, SE, and ST maps since it allows the best performance/complexity trade-off.

This paper proposes blind spectrum sensing (SS) in a narrowband context called Beamforming Fusion assisted Spectrum Sensing (BFSS). Considering a channel with angles of arrival (AoA), we jointly exploit beamforming algorithms to make decisions about the detection of users on frequency resources. The proposed method is totally blind and does not require knowledge of the noise power, the channel estimation, and the source signal. A state-of-the-art comparison of SS methods using beamforming is provided to validate our contribution in a shallow SNR region.

The European Train Control System (ETCS) relies today on the 2nd generation cellular system GSM-R. GSM-R obsolescence has triggered the evolution of the current European Railway Traffic Management System (ERTMS) to the Future Railway Mobile Communication System (FRMCS), under development. FRMCS considers the use of several radio access technologies in parallel such as Wi-Fi, LTE, Satellite and 5G. Testing new communication systems along railway tracks is time and money consuming. Consequently, it is important to develop a zero-on-site-testing approach thanks to the development of emulation platforms. These platforms, combining hardware and software, are able to reproduce in laboratory real railway infrastructure behavior and scenarios. This paper presents the EMULRADIO4RAIL platforms, developed in the framework of Shift2Rail program and gives examples of results obtained.

The increasing number of devices together with uncoordinated transmissions result in a major challenge of scalability in the Internet of things. This paper deals with signal detection in the uplink of a LoRa network through a deep learning-based approach. Two strategies are proposed: regression for bit detection based on a deep feedforward neural network and classification for symbol detection based on a convolutional neural network. These receivers can decode a selected user's signals when multiple users simultaneously transmit over the same frequency band with the same spreading factor. Simulation results show that both receivers outperform the classical LoRa one in the presence of interference. The results show that the introduced approach is relevant to deal with the scalability issue.

In modulation identification issues, like in any other classification problem, the performance of the classification task is significantly impacted by the feature characteristics. Feature weighting boosts the performance of machine learning algorithms, particularly the class of instance-based learning algorithms such as the Minimum Distance (MD) classifier, in which the distance measure is highly sensitive to the magnitude of features. In this paper, we propose an improved version of the Salp Swarm optimization Algorithm (SSA), called ISSA, that will be applied to optimize feature weights for an MD classifier. The aim is to improve the performance of a blind digital modulation detection approach in the context of multiple-antenna systems. The improvements introduced to SSA mainly rely on the opposition-based learning technique. Computer simulations show that the ISSA outperforms the SSA as well as the algorithms that derive from it. The ISSA also exhibits the best performance once it is applied for feature weighting in the above context.

A machine learning algorithm aims at designing a mathematical model based on a given training data set. Generally, the built model has a set of parameters that need to be adjusted. Since the performance of a given model depends on its settings, the parameters have to be carefully chosen through a fine-tuning step. A good model selection not only boosts performance but also allows a well-generalized model, i.e., a model that works sound on unseen data. In this paper, we assess the effectiveness of some metaheuristic optimization algorithms for support-vector machines (SVM) model selection. Computer simulations show that optimization algorithms that overall outperforms other algorithms using benchmark functions can be, further, definitely used for an efficient SVM model selection for classification. Thus, we show that Teaching–Learning-Based Optimization algorithm is faster and also enables the most accurate classification, even against other proposed methods in the literature for SVM model selection.

The complexity of modern communication systems is remarkable, and the efforts needed to put into service a new one are substantial as well. In some industrial sectors, circumstances are even harder. For example, in railways, the tests to be done are costly due to the integration in the rolling stock plus the need to have physical access to the railway tracks. Therefore, it is worth having a suitable emulator that considers many different radio-access technologies (RAT) in several railway scenarios (viaducts, tunnels, rural, hilly, etc.). Moreover, it should be able to do an end-to-end emulation, absolutely transparent for the application layer (this is, considering not only the physical layer but the network one as well). In this paper, we highlight the physical layer aspects considered in the construction of this emulator. Integration with the network layer is briefly mentioned, as well as the whole architecture.

This paper presents a new blind spectrum sensing (SS) algorithm based on a machine learning model: the radial basis function support-vector machines (RBF-SVM). As features, the introduced approach uses statistical tests that are based on the eigenvalues of the received signals covariance matrix. Since the decision on the frequency resource occupancy is in fact an issue of labeling binary data, SVM is intended as a potential technique for SS paradigm. The flexibility of SVM for linearly non-separable and high dimensional data makes it a good candidate for our issue, particularly that we consider low signal to noise ratios (SNR). Computer simulations shows that the proposal outperforms classical non-cooperative SS algorithms.

Support-Vector Machines (SVM) have gained momentum and have become a widely used classifier for various applications. Indeed, SVM parameters have a significant impact on the accuracy of the prediction model. The present work aims to propose an improved version of the Whale optimization Algorithm (WOA) that can choose the best model for SVM by seeking the optimal parameter values. To surpass the premature convergence problem and outperform the exploitation ability of the original WOA, two improvements are introduced. First, we increase the potential of the best solution, which redirects the search agent toward the global optimum. Second, a new search equation is added to decrease the overflow of diversity. Experimental results proved that the proposed Improved Whale optimization Algorithm (IWOA) is efficient not only in terms of quality of the final solution and convergence rate for optimization benchmarking function but also for reaching the optimal values of the SVM parameters. In this sense, the study has provided a medical diagnosis-oriented problem, classifying medical, i.e., cancer and diabetes, datasets, to demonstrate the effectiveness of the IWOA-optimized SVM.

This letter proposes a new technique that substantially improves blind digital modulation identification (DMI) algorithms that are based on higher-order statistics (HOS). The proposed technique takes advantage of noise power estimation to make an offset on higher-order moments (HOM), thus getting an estimate of noise-free HOM. When tested for multiple-antenna systems, the proposed method outperforms other DMI algorithms, in terms of identification accuracy, that are based only on cumulants or do not consider HOM denoising, even for a receiver with impairments. The improvement is achieved with the same order of complexity of the common HOS-based DMI algorithms in the same context.

In order to assess the reliability, availability, and maintainability of wireless links for the rail sector in the laboratory, it is necessary to develop measurement platforms able to emulate real railway radio environments. This type of testing platform should combine very new approaches, the so-called system in the loop (SITL) and hardware in the loop HITL). It allows connecting directly and coupling at radio frequency level (RF) the real equipment to be tested, the radio emulators (physical systems), and the simulators able to mimic the railway network and the railway channels’ behavior, including interferences. The Emulradio4rail platforms will support multiple emulation instances such as LTE, Wi-Fi, and SatCom networks. The paper will present the architecture of the platforms under development and preliminary assessments.

In this paper, we propose a new chaos-based encryption scheme for medical images. It is based on a combination of chaos and DNA computing under the scenario of two encryption rounds, preceded by a key generation layer, and follows the permutation-substitution-diffusion structure. The SHA-256 hash function alongside the initial secret keys is employed to produce the secret keys of the chaotic systems. Each round of the proposed algorithm involves six steps, i.e., block-based permutation, pixel-based substitution, DNA encoding, bit-level substitution (i.e., DNA complementing), DNA decoding, and bit-level diffusion. A thorough search of the relevant literature yielded only this time the pixel-based substitution and the bit-level substitution are used in cascade for image encryption. The key-streams in the bit-level substitution are based on the logistic-Chebyshev map, while the sine-Chebyshev map allows producing the key-streams in the bit-level diffusion. The final encrypted image is obtained by repeating once the previous steps using new secret keys. Security analyses and computer simulations both confirm that the proposed scheme is robust enough against all kinds of attacks. Its low complexity indicates its high potential for real-time and secure image applications.

This paper presents a fast and efficient cryptosystem for enciphering digital images. It employs two of the most prominent dynamical systems-chaotic maps and cellular automata. The key streams in the proposed encryption scheme are derived from the SHA-256 hash function. Hash functions produce the digest of the input plaintext, known as a hash value, which can be considered as a unique signature of the input. This makes the keys more plaintext dependent, which is a desirable property of a robust cryptosystem. These key streams are used as the secret keys (i.e., initial conditions and control parameters) of an improved one-dimensional (1-D) chaotic map, i.e., the Logistic-Sine map. As far as we know, this paper is a first that combines the well-known diffusion-confusion architecture and the fourth order 1-D memory cellular automata (MCA) for image encryption. First, a pixel-wise XOR operation is applied to the original image, followed by a pixel-wise random permutation. The resulting image is decomposed into four blocks according to the quadtree decomposition strategy. Then, a fourth order reversible MCA is applied, the blocks obtained from the quadtree decomposition are considered as the initial MCA configurations, and the transition rules are determined using the chaotic map. The performance analyses show that the proposed encryption scheme presents a high immunity against all kind of attacks while maintaining a low complexity, which outcome a notably better performance/complexity trade-off compared to some recently proposed image schemes.