Deep learning (DL) based methods for orthogonal frequency division multiplexing (OFDM) radio receivers demonstrated higher signal detection performance compared to the traditional receivers. However, the existing DL-based models, usually adapted from computer vision, aren’t well suited for wireless communications. These models require high computational resources and memory, and have significant inference delays, limiting their use in resource-constrained settings. Additionally, reducing network size to ease resource demands often leads to notable performance degradation. This paper introduces collaborative networks (CoNet), a novel neural network (NN) architecture designed for OFDM receivers. CoNet uses multiple small ResNet or CNN subnet-works to simultaneously process signal features from different perspectives like capturing channel correlations and interference patterns. These subnetworks fuse their outputs through interaction operations (e.g., element-wise multiplication), significantly enhancing detection performance. Simulation results show CoNet significantly outperforms traditional architectures like residual networks (ResNets) in bit error rate (BER) and reduces inference delay when both nets have the same size and the same computational complexity.

Towards fast, hardware-efficient, and lowcomplexity receivers, we propose a compression-aware learning approach and examine it on free-space optical (FSO) receivers for turbulence mitigation. The learning approach jointly quantize, prune, and train a convolutional neural network (CNN). In addition, we propose to have the CNN weights of power of two values so we replace the multiplication operations bit-shifting operations in every layer that has significant lower computational cost. The compression idea in the proposed approach is that the loss function is updated and both the quantization levels and the pruning limits are optimized in every epoch of training. The compressed CNN is examined for two levels of compression (1-bit and 2-bits) over different FSO systems. The numerical results show that the compression approach provides negligible decrease in performance in case of 1-bit quantization and the same performance in case of 2-bits quantization, compared to the full-precision CNNs. In general, the proposed IM/DD FSO receivers show better bit-error rate (BER) performance (without the need for channel state information (CSI)) compared to the maximum likelihood (ML) receivers that utilize imperfect CSI when the DL model is compressed whether with 1-bit or 2-bit quantization.

Deep learning (DL) methods have emerged as promising solutions for enhancing receiver performance in wireless orthogonal frequency-division multiplexing (OFDM) systems, offering significant improvements over traditional estimation and detection techniques. However, DL-based receivers often face challenges such as poor generalization to unseen channel conditions and difficulty in effectively tracking rapid channel fluctuations. To address these limitations, this paper proposes a hybrid receiver architecture that integrates the strengths of both traditional and neural receivers. The core innovation is a discriminator neural network trained to dynamically select the optimal receiver whether it is the traditional or DL-based receiver according on the received OFDM block characteristics. This discriminator is trained using labeled pilot signals that encode the comparative performance of both receivers. By including anomalous channel scenarios in training, the proposed hybrid receiver achieves robust performance, effectively overcoming the generalization issues inherent in standalone DL approaches.

Federated learning is a promising approach to training deep learning models over distributed devices without sharing their local datasets. Low Earth orbit (LEO) satellites have the potential to use the massive amount of collected Earth imageries and sensor data to train artificial intelligence (AI) models and provide global services such as disaster detection. However, the structure of LEO networks is different from that of the terrestrial networks, which makes the traditional (star-based or hierarchical-based FL) inefficient. This paper proposes a new distributed FL approach that is customized to the LEO structure. The approach is based on parallelizing the FL operations and decentralizing the aggregations over several satellites, aiming at reducing the convergence time at a given energy constraint. Simulation results show that the proposed algorithm converges significantly faster than traditional FL-LEO approaches proposed in the literature under the same energy consumption.

Free space optical (FSO) communication has the potential to develop ultra-fast data links that can be used in a vari-ety of sixth-generation (6G) applications, including heterogeneous networks with enormous connectivity and wireless backhauls for cellular systems. This paper proposes several low-complexity deep neural networks designed to detect received symbols in FSO systems, eliminating the need for channel state information (CSI). We consider different intensity modulation (IM) schemes (e.g., on-off shift keying (OOK) and pulse amplitude modulation (PAM)) and different FSO systems (e.g., single-input single-output (SISO) and single-input multiple-output (SIMO)). We design the neural network to be able to exploit the temporal correlation of the channels and the common information received at every single photodetector (the case of SIMO). The convolutional neural network (CNN) is designed in a way that it can learn the channels without using pilots and also can jointly combine the received signals and detect the symbols for a wide range of signal-to-noise ratio (SNR) values and different turbulence levels. Numerical results show that the proposed CNN demonstrates a better performance compared to the maximum likelihood approach that utilizes imperfect channel information for symbol detection.

Cell free (CF) massive multiple-input multiple-output (mMIMO) is a promising technology in realizing beyond fifth generation (B5G) networks. Massive deployment of access points (APs) in a CF-mMIMO system increases the spectral efficiency, however it also increases the energy consumption and fronthaul requirements. Since recently reconfigurable intelligent surface (RIS) is shown to be a cost-effective solution to improve performance of wireless networks, RIS can be a promising technology to enhance the performance of CF-mMIMO systems. In this work, we study the downlink (DL) performance of CF-mMIMO system aided by multiple RISs, while considering correlated Rician fading channels, discrete RIS phase-shifts and low-complexity channel estimation (CE) protocol. Given the obtained imperfect channel state information (CSI), we derive lower bounds on the rates achieved using conjugate beamforming (CB) and zero-forcing (ZF) precoders. The obtained bounds depend on channel statistics, RIS phase-shifts and number of RIS elements. To optimize the performance with respect to RIS phase-shifts, we formulate a maximization problem and propose a sub-optimal genetic algorithm (GA)-based solution. Through simulations, we demonstrate that distributed RIS deployment outperforms centralized RIS deployment in terms of DL throughput. Interestingly, we demonstrate that the DL throughput improves as number of RISs increases until an optimal number of distributed RISs over which the DL performance of the system starts to drop. We discuss this effect under varying the number of APs and RISs. We extend the analysis by considering different precoders, CE and RIS optimization schemes, and verify the accuracy of our derived analytical results by Monte-Carlo simulations.

In large scale dynamic wireless networks, the amount of overhead caused by channel estimation (CE) is becoming one of the main performance bottlenecks. This is due to large number of users whose channels should be estimated and the user mobility. This work proposes a new hybrid channel estimation/prediction (CEP) scheme to reduce overhead in time-division duplex (TDD) wireless cell-free massive multiple-input-multiple-output (mMIMO) systems. The scheme proposes sending a pilot signal from each user only once in a given number (window) of coherence intervals (CIs). Then minimum mean-square error (MMSE) estimation is used to estimate the channel of this CI, while a deep neural network (DNN) is used to predict the channels of the remaining CIs in the window, exploiting the temporal correlation between the consecutive CIs. By doing so, CE overhead is reduced by at least 50 percent at the expense of negligible CE error for practical user mobility settings. Consequently, the proposed CEP scheme improves the spectral efficiency compared to the conventional MMSE CE approach, which is demonstrated numerically.

Federated learning (FL) has emerged as a distributed machine learning (ML) technique to train models without sharing users’ private data. In this paper, we introduce a decentralized FL scheme that is called federated learning empowered overlapped clustering for decentralized aggregation (FL-EOCD). The introduced FL-EOCD leverages device-to-device (D2D) communications and overlapped clustering to enable decentralized aggregation, where a cluster is defined as a coverage zone of a typical device. The devices located on the overlapped clusters are called bridge devices (BDs). In the proposed FL-EOCD scheme, a clustering topology is envisioned where clusters are connected through BDs, so as the aggregated models of each cluster is disseminated to the other clusters in a decentralized manner without the need for a global aggregator or an additional hop of transmission. To evaluate our proposed FL-EOCD scheme as opposed to baseline FL schemes, we consider minimizing the overall energy-consumption of devices while maintaining the convergence rate of FL subject to its time constraint. To this end, a joint optimization problem, considering scheduling the local devices/BDs to the CHs and computation frequency allocation, is formulated, where an iterative solution to this joint problem is devised. Extensive simulations are conducted to verify the effectiveness of the proposed FL-EOCD algorithm over FL conventional schemes in terms of energy consumption, latency, and convergence rate.

Achieving high data rate is always a goal that could be met by developing new technologies and investigating potential ones. Recently, the concept of cell-free massive MIMO systems (CF-mMIMO) has been considered to enhance the performance of systems that operate merely with Radio Frequency (RF) or visible light communication (VLC) technologies. In this paper, a hybrid VLC/RF cell-free massive MIMO system is proposed where an RF cell-free network is used to support a VLC cell-free network. The idea is to utilize the benefits of each network and balance the load aiming at maximizing the system’s sum-rate. The system is evaluated using zero-forcing (ZF) precoding scheme. A user association algorithm is proposed to assign users to either VLC or RF networks. Results show that the proposed algorithm outperforms a random network association of users. Results also show great potential for the proposed system compared to standalone cell-free networks.

Reconfigurable intelligent surfaces (RISs) have drawn considerable attention due to their ability to direct electromagnetic waves into desirable directions. Although RISs share some similarities with relays, the two have fundamental differences impacting their performance. To harness the benefits of both, we propose a downlink system wherein a relay and an RIS improve performance in terms of energy-efficiency. Using singular value decomposition (SVD), semidefinite programming (SDP), and function approximations, we propose different solutions for optimizing the beamforming matrices at the base-station (BS), the relay, and the phase shifts at the RIS to minimize the total power under quality-of-service (QoS) constraints. The problem is solved when the relay operates in half-duplex and full-duplex modes and when the reflecting elements have continuous and discrete phase shifts. Simulation results compare the performance of the system with and without the RIS or the relay, under different optimization solutions. The results show that the system with full-duplex relay and RIS outperforms the other scenarios, and the contribution of full-duplex relay is higher than that of the RIS. However, an RIS outperforms a half-duplex relay when the required QoS is high. The results also show that increasing the number of reflecting elements improves the performance better in the presence of a relay than in its absence.

Despite the promising high-data rate features of visible light communications (VLC), they still suffer from unbalanced services due to blockages and channel fluctuation among users. This paper introduces and evaluates a new transmission scheme which adopts cooperative non-orthogonal multiple access (Co-NOMA) in hybrid VLC/radio-frequency (RF) systems, so as to improve both system sum-rate and fairness. Consider a network consisting of one VLC access point (AP) and multiple strong and weak users, where each weak user is paired with a strong user. Each weak user can be served either directly by the VLC AP, or via the strong user which converts light information received through the VLC link, and forwards the information to the weak user via the RF link. The paper then maximizes a network-wide weighted sum-rate, so as to jointly determine the strong-weak user-pairs, the serving link of each weak user (i.e., either direct VLC or hybrid VLC/RF), and the power of each user message, subject to user connectivity and transmit power constraints. The paper tackles such a mixed-integer non-convex optimization problem using an iterative approach. Simulations show that the proposed scheme significantly improves the VLC network performance (i.e., sum-rate and fairness) as compared to the conventional NOMA scheme.

In next-generation wireless networks, providing secure transmission and delay guarantees are two critical goals. However, either of them requires a concession on the transmission rate. In this article, we consider a visible light network consisting of multiple access points and multiple users. Our first objective is to mathematically evaluate the achievable rate under constraints on delay and security. The second objective is to provide a cell formation with customized statistical delay and security guarantees for each user. First, we propose a user-centric design called secure cell formation, in which artificial noise is considered, and flexible user scheduling is determined. Then, based on the effective capacity theory, we derive the statistical-delay-constrained secrecy rate and formulate the cell formation problem as a stochastic optimization problem (OP). Further, based on the Lyapunov optimization theory, we transform the stochastic OP into a series of evolutionary per-slot drift-plus-penalty OPs. Finally, a modified particle swarm optimization algorithm and an interference graph-based user-centric scheduling algorithm are proposed to solve the OPs. We obtain a dynamic independent set of scheduled users as well as secure cell formation parameters. Simulation results show that the proposed algorithm can achieve a better delay-constrained secrecy rate than the existing cell formation approaches.

Visible light communications (VLC) have emerged as a strong candidate for meeting the escalating demand for high data rates. In this paper, we consider a VLC network, where multiple access points (APs) serve both energy-harvesting users (EHUs), i.e., users who harvest energy from light emitted by diodes and information users (IUs), i.e., users who gather data information. In order to jointly balance the achievable sum rate at the IUs and the energy harvested by the EHUs, the paper considers maximizing a network-wide utility, which consists of a weighted sum of the IUs sum rate and the EHUs harvested energy, subject to individual IU rate constraint, individual EHU harvested-energy constraint, and AP power constraints, so as to jointly determine the direct current (DC) bias value at each AP, and the power of the alternating-current (AC) signals of the users. A difficult non-convex optimization problem is solved using an iterative approach which relies on inner convex approximations, and compensates for the used approximations using proper outer-loop updates. The paper further considers solving the special cases of the problem, i.e., maximizing the sum rate, and maximizing the total harvested-energy, both subject to the same constraints. Numerical results highlight the significant performance improvement of the proposed algorithms, and illustrate the impacts of the network parameters on the performance trade-off between the sum rate and harvested-energy.

In order to meet the new escalating demand for high data rate services and applications, visible light communication (VLC) has emerged as a promising solution for the fifth-generation (5G) wireless networks and beyond. Consider a VLC network, where multiple access points (APs) serve both energy-harvesting users (EHUs), i.e., users which harvest energy from light intensity, and information-users (IUs), i.e., users which gather data information. The performance of the system becomes a function of the direct current (DC) bias values allocated to each AP. After adopting a zero-forcing (ZF) precoding approach to cancel the inter-cell interference, the paper formulates the problem of maximizing the network harvested energy subject to individual harvested energy and data rate constraints at the EHUs and IUs, respectively, so as to determine the DC bias of every AP. The paper then proposes solving such a difficult non-convex optimization problem using an iterative approach. The proposed algorithm uses well-chosen approximations of the objective and constraints functions, and compensates for the approximations using proper outer-loop updates. The paper further proposes a sub optimal heuristic which provides a feasible, yet simple, solution to the problem. Numerical results illustrate the convergence of our proposed algorithms, and highlight the significant performance improvement of the proposed algorithm as compared to the proposed baseline approach.

In this paper, we propose a joint cell formation and power allocation algorithms for energy efficiency (EE) maximization in visible light communication (VLC) networks. Unlike the previous works, we show that the cell formation and power allocation are interlinked problems and they should be solved jointly. We start by proposing a new algorithm for users clustering and then associating all the access points (APs) to the clustered users based on a proposed metric. Under the assumption that the vectored transmission is applied at each formed cell, we solve an optimization problem that aims to maximize the EE by allocating the powers for the users with quality of service (QoS) constraints. Then, we propose an algorithm that jointly allocates the power and decides which APs must participate in communication and which ones must be switched off. The numerical results demonstrate that the proposed algorithms significantly improve the EE compared to the traditional methods and algorithms.

In this paper, we propose and study a new joint load balancing (LB) and power allocation (PA) scheme for a hybrid visible light communication (VLC) and radio frequency (RF) system consisting of one RF\ access point (AP) and multiple VLC\ APs. An iterative algorithm is proposed to distribute the users on the APs and distribute the powers of these APs on their users. In PA subproblem, an optimization problem is formulated to allocate the power of each AP to the connected users for the total achievable data rates maximization. It is proved that the PA optimization problem is concave but not easy to tackle. Therefore, we provide a new algorithm to obtain the optimal dual variables after formulating them in terms of each other. Then, the users that are connected to the overloaded APs and receive less data rates start seeking for other APs that offer higher data rates. Users with lower data rates continue re-connecting from AP to other to balance the load only if this travel increases the summation of the achievable data rates and enhances the system fairness. The numerical results demonstrate that the proposed algorithms improve the system capacity and system fairness with fast convergence.

In this paper, we study the physical layer security in a two-way relay system consisting of two transceivers, one eavesdropper, and multiple relays. We study the system in case of the channel state information (CSI) of the eavesdropper is unknown, and hence artificial noise is used to degrade the signal to noise ratio at the eavesdropper. In order to reserve the maximum possible power for artificial noise, we consider the problem of minimizing the total power of the information signal transmitted by the relays, under quality of service constraints at the legitimate transceivers. This problem has been solved using semidefinite programming (SDP) and second order cone programming (SOCP) methods. Here, aiming to significantly decrease the complexity, we propose a novel approach to find the optimal solution using the generalized eigenvalue. We show that in most cases, we can provide a closed-form expression of the optimal solution. In addition, our proposed solution can be used for all quadratically constrained quadratic programs (QCQPs) with positive definite objective function and two constraints. Simulation results demonstrate the effectiveness of our algorithm in terms of optimality and low complexity compared to SDP.