A comprehensive study on the applications of denoising diffusion models for wireless systems is provided. The article highlights the capabilities of diffusion models in learning complicated signal distributions, modeling wireless channels, and denoising and reconstructing distorted signals. First, fundamental working mechanism of diffusion models is introduced. Then the recent advances in applying diffusion models to wireless systems are reviewed. Next, two case studies are provided, where conditional diffusion models (CDiff) are proposed for data reconstruction enhancement, covering both the conventional digital communication systems, as well as the semantic communication (SemCom) setups. The first case study highlights about 10 dB improvement in data reconstruction under low-SNR regimes, while mitigating the need to transmit redundant bits for error correction codes in digital systems. The second study further extends the case to a Sem- Com setup, where diffusion autoencoders showcase superior performance compared to legacy autoencoders and variational autoencoder (VAE) architectures. Finally, future directions and existing challenges are discussed.

We study a multicarrier integrated sensing and communication (ISAC) system where a base station (BS) equipped with a dynamic metasurface antenna (DMA) simultaneously supports multiple communication users and (radar) sensing targets. The subcarriers are divided into two disjoint sets: one for user communications and the other for radar sensing. We consider the frequency response of DMA elements under a Lorentzian-constrained model. The objective is to maximize the minimum signal-to-interference-plus-noise ratio (SINR) among all users, while satisfying radar SINR constraints, the BS transmit power budget, and DMA weight constraints. To this end, we jointly optimize the communication and radar beamforming vectors and the DMA weights. To handle the non-convexity of the resulting problem, we employ an alternating optimization framework with fractional programming to iteratively solve two decoupled subproblems. Simulation results show that the Lorentzian-constrained DMA design provides about a 80% gain over random DMA weights.

Existing semantic communication (SemCom) frameworks typically emphasize on channel-adaptive neural encoding-decoding schemes, lacking full exploration of signal distribution. Moreover, having autoencoder architectures as backbone, their encoding is tightly coupled to a matched decoder, causing scalability issues in practice. To solve these issues, diffusion autoencoder models are proposed. A neural encoder extracts the high-level semantics, and then a conditional denoising diffusion model probabilistic model (C-DDPM) at the decoder learns the source distribution through signal-space denoising, while the noisy semantic latents are incorporated as the conditioning input, to "steer" the decoding process towards the semantics intended by the transmitter. Simulations over CIFAR-10 highlights 50% and 33% improvement in terms of the learned perceptual image patch similarity (LPIPS) metric compared to autoencoders with matched decoder architecture, and variational autoencoders (VAE) benchmarks, respectively.

Dynamic metasurface antennas (DMAs) have emerged as a promising solution for reducing cost and power consumption in next-generation wireless systems employing large antenna arrays. Existing studies on DMA-based architectures typically adopt a frequency-flat approximation to model the frequency response of individual elements, in which only the phase of each element is adjustable, while the amplitude tuning remains uninvolved. Additionally, this approximation is valid only for narrowband signaling and becomes increasingly inaccurate in wideband and ultra-wideband scenarios. In contrast, by adopting the accurate nonlinear frequency-selective Lorentzian-constrained model and jointly optimizing the resonance frequency and quality factor of each element, both the phase and magnitude responses are directly incorporated in the resource management process, thereby providing a more faithful representation of the frequency-dependent behavior of DMA elements in wideband and ultra-wideband systems. Motivated by this, we propose a rate-splitting multiple access (RSMA) design for DMA-based architectures using the frequency-selective Lorentzian model in wideband transmissions, aiming to maximize the minimum achievable sum rate across all subcarriers. To this end, we jointly optimize the transmit beamforming vectors across all subcarriers as well as the resonance frequency and the quality factor of all DMA elements. Simulation results indicate that the nonlinear frequency-selective Lorentzian model achieves an approximately 45% performance gain over the frequency-flat approximation in DMA architectures with RSMA for wideband transmissions.

Radio frequency wireless power transfer (WPT) is a promising technology to charge low-power devices in future wireless systems. In this letter, energy-efficient WPT in wideband systems is studied, ensuring that each device meets its minimum energy harvesting (EH) requirement under both perfect and imperfect channel state information (CSI) scenarios. To this end, the digital beamformer, the true time delayers (TTDs), and the phase shifter-based analog beamformer are jointly optimized. Since the energy beamforming flexibility depends on the base station (BS) antenna's architecture, the BS employs two hybrid beamforming architectures, sub-connected (SC) and fully-connected (FC) TTD, which charge EH devices in the near-field region. Since minimizing the transmit power is non-convex and decision variables are highly coupled, an alternating optimization algorithm based on semi-definite programming and semi-definite relaxation is proposed to solve the original problem. Furthermore, a matrix scale reduction scheme is leveraged to decrease computational complexity. Simulation results show that the SC architecture performs better than the FC architecture in both perfect and imperfect CSI scenarios. Additionally, the SC architecture decreases convergence time relative to the FC architecture by more than ten times.

Emerging massive machine-type communications service class needs to support many devices while ensuring that scarce radio resources are utilized efficiently. Nonorthogonal multiple access is proposed to minimize the signaling overhead and optimize resource allocation. However, during the initial access, the base station (BS) is presented with the challenge of identifying sparsely active devices in the absence of knowledge about the sparsity and channel state information. The user channels in most practical scenarios have common reflection paths, making them partially correlated, which can be exploited to improve the detection performance at the BS. In this context, we formulate a novel multiuser detection (MUD) problem in spatially correlated Rician channels, which we reformulate as a multilabel classification problem utilizing deep learning techniques. We propose two diverse approaches to tackle this problem: 1) ViT-Net, a vision transformer-based architecture, and 2) FAR-Net, a fully activated deep neural network featuring residual connections. Our analysis highlights the significance of spatial correlation for MUD, which can accord around 13% higher overloading ratio compared to the noncorrelated scenario. Numerical evaluations demonstrate the effectiveness of the proposed model in addressing spatial correlation compared to the existing deep-learning models.

Glioblastoma (GBM), a common cancer of the central nervous system (CNS), is considered incurable worldwide. The treatment of GBM varies from patient to patient, as conventional medical treatments do not apply to all patients with similar symptoms. Therefore, drug efficacy recommendation systems are very useful in the treatment of various types of cancers. The Genomics of Drug Sensitivity in Cancer (GDSC) database was used as the primary data source, containing 135,242 cell line-drug interactions, inhibitory concentration of cancer drug for more than 800 cancer cell lines. Each cell line provides gene expression values, which are further normalized through Z-transformation for each gene. However, we utilized only 47 genes in our research due to limitations in computer processing speed and memory. A drug efficacy recommendation system was constructed using a deep learning method that combines gene expression, drug Simplified Molecular Input Line Entry System (SMILES), and inhibitory concentration features. A panel of 47 genes associated with GBM was processed using two deep learning models: Artificial Neural Network (ANN) and Convolutional Neural Network (CNN). This approach addresses the challenge of personalized treatment for GBM, offering the potential for improved therapeutic outcomes. The results of the recommendation system are calculated based on Half Maximal Inhibitory Concentration (IC50) values, which represent the therapeutic effectiveness in inhibiting the growth of GBM cells. CNN outperformed ANN with a significant margin, achieving a Root Mean Square Error (RMSE) of 0.9822 compared to 1.2127. These results are also consistent with other metrics, including Pearson correlation, Spearman correlation, and Mean Absolute Error (MAE). According to the study, the system can accurately predict the effectiveness of drugs on GBM cancer genes. This study has the potential to predict drug efficacy during medical procedures.

Integrating extremely large antenna arrays (ELAAs) with extremely large reconfigurable intelligent surfaces (XLRISs) in millimeter wave (mmWave) communications places devices in the near-field (NF) region, significantly boosting spectral efficiency (SE). This paper investigates beam focusing for SE maximization by determining the optimal placement of XL-RIS in a multi-input single-output (MISO) system. Specifically, to maximize SE, the transmit beamforming vector at the base station (BS) and the phase-shifting vector at the XL-RIS are jointly optimized. To explore the optimal placement of XL-RIS, we consider three scenarios where the positions of the BS and user are fixed, but the XL-RIS is placed in either the NF or farfield (FF) of both. Since the SE maximization problem is nonconvex with highly coupled variables, we propose an alternating optimization algorithm that decouples the problem into two subproblems: transmit beamforming optimization and phase shift optimization. Both sub-problems are reformulated as convex problems using the semi-definite programming (SDP) technique and are then solved with standard convex optimization tools. Simulation results show that placing the XL-RIS in the NF region of the BS achieves higher SE compared to other configurations.

Dynamic metasurface antenna (DMA) technology is a promising solution for future wireless communication systems, offering substantial gains in spectral efficiency (SE). This paper investigates the potential of DMA-based architectures in maximizing the SE of a near-field multi-user communication system. The impact of radio frequency (RF) impairments and spatial correlation effects among the DMA elements on the SE are also analyzed. Specifically, we jointly optimize the transmit beamforming vectors and the configurable weights of the DMA’s metamaterial elements while satisfying the maximum power budget. The stated SE optimization problem is non-convex, with highly coupled decision variables, making it challenging to solve. To address this, we propose an algorithm based on alternating optimization and semi-definite programming. Unlike existing approaches that simplify the power budget constraint by applying it only to the digital component of hybrid beamforming—an unrealistic assumption—this work enforces the constraint on the entire transmitted signal (actual radiated electromagnetic power) for a more accurate solution. Simulation results demonstrate that incorporating the power budget constraint without relaxation leads to a significant SE improvement over relaxed methods.

In this paper, conditional denoising diffusion probabilistic models (CDiffs) are proposed to enhance the data transmission and reconstruction over wireless channels. The underlying mechanism of diffusion models is to decompose the data generation process over the so-called “denoising” steps. Inspired by this, the key idea is to leverage the generative prior of diffusion models in learning a “noisy-to-clean” transformation of the information signal to help enhance data reconstruction. The proposed scheme could be beneficial for communication scenarios in which a prior knowledge of the information content is available, e.g., in multimedia transmission. Hence, instead of employing complicated channel codes that reduce the information rate, one can exploit diffusion priors for reliable data reconstruction, especially under extreme channel conditions due to low signal-to-noise ratio (SNR), or hardware-impaired communications. The proposed CDiff-assisted receiver is tailored for the scenario of wireless image transmission using MNIST dataset. Our numerical results highlight the reconstruction performance of our scheme compared to the conventional digital communication, as well as the deep neural network (DNN)-based benchmark. It is also shown that more than 10 dB improvement in the reconstruction could be achieved in low SNR regimes, without the need to reduce the information rate for error correction.

Semantic communication (SemCom) systems aim to learn the mapping from low-dimensional semantics to high-dimensional ground-truth. While this is more akin to a "domain translation" problem, existing frameworks typically emphasize on channel-adaptive neural encoding-decoding schemes, lacking full exploration of signal distribution. Moreover, such methods so far have employed autoencoder-based architectures, where the encoding is tightly coupled to a matched decoder, causing scalability issues in practice. To address these gaps, diffusion autoencoder models are proposed for wireless SemCom. The goal is to learn a "semantic-to-clean" mapping, from the semantic space to the ground-truth probability distribution. A neural encoder at semantic transmitter extracts the high-level semantics, and a conditional diffusion model (CDiff) at the semantic receiver exploits the source distribution for signal-space denoising, while the received semantic latents are incorporated as the conditioning input to "steer" the decoding process towards the semantics intended by the transmitter. It is analytically proved that the proposed decoder model is a consistent estimator of the ground-truth data. Furthermore, extensive simulations over CIFAR-10 and MNIST datasets are provided along with design insights, highlighting the performance compared to legacy autoencoders and variational autoencoders (VAE). Simulations are further extended to the multi-user SemCom, identifying the dominating factors in a more realistic setup.

A significant challenge in the wireless-communication field revolves around the growing demand for data usage, all while dealing with the limitations of available resources. One potential solution lies in leveraging a local area network (LAN) within the same frequency band as a service provider (SP) especially when the SP’s bandwidth is underutilized. This approach aims to minimize the resource-demand mismatch. This paper focuses on addressing the issue of interference detection when two such networks coexist within the same frequency spectrum. Our study introduces an innovative methodology that harnesses machine-learning techniques to tackle this challenge. We have delved into various ML methods used in the physical layer of wireless communication for similar purposes. As a result, we have developed a deep-learning model designed to identify the presence of interference. This, in turn, enhances the quality of service (QoS) for both networks by effectively mitigating any identified interference. Specifically, we employ a binary classifier utilizing a convolutional neural network (CNN) architecture to detect interference between two networks operating at the same frequency. To evaluate the effectiveness of this binary classifier in identifying interference, we conducted a series of experiments. Our results have demonstrated an accuracy exceeding 90% when the interferer has been introduced at a 500 m radius from the local base station, but it has done so by adding only an inference latency of 0.126 ms.

Generative AI has received significant attention among a spectrum of diverse industrial and academic domains, thanks to the magnificent results achieved from deep generative models such as generative pre-trained transformers (GPT) and diffusion models. In this paper, we explore the applications of denoising diffusion probabilistic models (DDPMs) in wireless communication systems under practical assumptions such as hardware impairments (HWI), low-SNR regime, and quantization error. Diffusion models are a new class of state-of-the-art generative models that have already showcased notable success with some of the popular examples by OpenAI1 and Google Brain2. The intuition behind DDPM is to decompose the data generation process over small ``denoising'' steps. Inspired by this, we propose using denoising diffusion model-based receiver for a practical wireless communication scheme, while providing network resilience in low-SNR regimes, non-Gaussian noise, different HWI levels, and quantization error. We evaluate the reconstruction performance of our scheme in terms of mean-squared error (MSE) metric. Our results show that more than 25 dB improvement in MSE is achieved compared to deep neural network (DNN)-based receivers. We also highlight robust out-of-distribution performance under non-Gaussian noise.

Multi-access Edge Computing (MEC) can be implemented together with Open Radio Access Network (O-RAN) to offer low-cost deployment and bring services closer to end-users. In this paper, the joint orchestration of O-RAN and MEC using a Bayesian deep reinforcement learning (RL) framework is proposed. The framework jointly controls the O-RAN functional splits, O-RAN/MEC computing resource allocation, hosting locations, and data flow routing across geo-distributed platforms. The goal is to minimize the long-term total network operation cost and maximize MEC performance criterion while adapting to varying demands and resource availability. This orchestration problem is formulated as a Markov decision process (MDP). However, finding the exact model of the underlying O-RAN/MEC system is impractical since the system shares the same resources, serves heterogeneous demands, and its parameters have non-trivial relationships. Moreover, the formulated MDP results in a large state space with multidimensional discrete actions. To address these challenges, a model-free RL agent based on a combination of Double Deep Q-network (DDQN) with action branching is proposed. Furthermore, an efficient exploration-exploitation strategy under a Bayesian learning framework is leveraged to improve learning performance and expedite convergence. Trace-driven simulations are performed using an O-RAN-compliant model. The results show that our approach is data-efficient (i.e., converges significantly faster), increases the reward by 32% compared to its non-Bayesian version, and outperforms Deep Deterministic Policy Gradient by up to 41%.

In this letter, denoising diffusion implicit models (DDIM), a computation-efficient class of probabilistic diffusion models, are proposed for improving the reconstruction performance of end-users in cell-free massive MIMO (mMIMO) downlink. The idea is to leverage the “denoising” characteristic of diffusion models to remove the hardware and channel imperfections, as well as the interference signals, and finally reconstruct the downlink signals. First, it is shown that the data transmission in cell-free mMIMO downlink can be modeled as a forward diffusion process, assuming the aggregated effect of residual impairments and multi-user interference as Gaussian-distributed signals. Then the data reconstruction is carried out via a reverse diffusion process within the DDIM framework. Numerical results in terms of both wireless-specific and learning-specific hyperparameters are provided to highlight the improvement in the reconstruction performance and post-processed SINR. We also trade-off computation complexity against data reconstruction quality by adjusting the hyperparameters of our denoising model without the need for re-training.

Wireless networks are inherently graph-structured in which graph representation learning can be utilized to solve complex network optimization problems. In graph representation learning, feature vectors for each entity in the network are calculated such that they could capture spatial and temporal dependencies in their local and global neighborhoods. Specifically, graph neural networks (GNNs) are powerful tools to solve these complex problems because of their expressive representation and reasoning power. In this article, the potential of graph representation learning and GNNs in wireless networks is presented. An overview of graph representation learning is provided which covers the fundamentals and concepts, such as feature design over graphs, GNNs, and their design principles. The potential of graph representation learning in wireless networks is presented via a few exemplary use cases and some initial results on the GNN-based access point selection for cell-free massive Multiple-Input Multiple-Output (MIMO) systems.

Virtualized Radio Access Networks (vRANs) are fully configurable and can be implemented at a low cost over commodity platforms to enable network management flexibility. In this paper, a novel vRAN reconfiguration problem is formulated to jointly reconfigure the functional splits of the base stations (BSs), locations of the virtualized central units (vCUs) and distributed units (vDUs), their resources, and the routing for each BS data flow. The objective is to minimize the long-term total network operation cost while adapting to the varying traffic demands and resource availability. In the first step, testbed measurements are performed to study the relationship between the traffic demands and computing resources, which reveals high variance and depends on the platform and its load. Consequently, finding the perfect model of the underlying system is non-trivial. Therefore, to solve the proposed problem, a deep reinforcement learning (RL)-based framework is proposed and developed using model-free RL approaches. Moreover, the problem consists of multiple BSs sharing the same resources, which results in a multi-dimensional discrete action space and leads to a combinatorial number of possible actions. To overcome this curse of dimensionality, action branching architecture, which is an action decomposition method with a shared decision module followed by neural network is combined with Dueling Double Deep Q-network (D3QN) algorithm. Simulations are carried out using an O-RAN compliant model and real traces of the testbed. Our numerical results show that the proposed framework successfully learns the optimal policy that adaptively selects the vRAN configurations, where its learning convergence can be further expedited through transfer learning even in different vRAN systems. It also offers significant cost savings by up to 59% of a static benchmark, 35% of Deep Deterministic Policy Gradient with discretization, and 76% of non-branching D3QN.

Thanks to the outstanding achievements from state-of-the-art generative models like ChatGPT and diffusion models, generative AI has gained substantial attention across various industrial and academic domains. In this paper, denoising diffusion probabilistic models (DDPMs) are proposed for a practical finite-precision wireless communication system with hardware-impaired transceivers. The intuition behind DDPM is to decompose the data generation process over the so-called "denoising" steps. Inspired by this, a DDPM-based receiver is proposed for a practical wireless communication scheme that faces realistic non-idealities, including hardware impairments (HWI), channel distortions, and quantization errors. It is shown that our approach provides network resilience under low-SNR regimes, near-invariant reconstruction performance with respect to different HWI levels and quantization errors, and robust out-of-distribution performance against non-Gaussian noise. Moreover, the reconstruction performance of our scheme is evaluated in terms of cosine similarity and mean-squared error (MSE), highlighting more than 25 dB improvement compared to the conventional deep neural network (DNN)-based receivers.

We study the joint active/passive beamforming and channel blocklength (CBL) allocation in a non-ideal reconfigurable intelligent surface (RIS)-aided ultra-reliable and low-latency communication (URLLC) system. The considered scenario is a finite blocklength (FBL) regime and the problem is solved by leveraging a novel deep reinforcement learning (DRL) algorithm named twin-delayed deep deterministic policy gradient (TD3). First, assuming an industrial automation system with multiple actuators, the signal-to-interference-plus-noise ratio and achievable rate in the FBL regime are identified for each actuator in terms of the phase shift configuration matrix at the RIS. Next, the joint active/passive beamforming and CBL optimization problem is formulated where the objective is to maximize the total achievable FBL rate in all actuators, subject to non-linear amplitude response at the RIS elements, BS transmit power budget, and total available CBL. Since the amplitude response equality constraint is highly non-convex and non-linear, we resort to employing an actor-critic policy gradient DRL algorithm based on TD3. The considered method relies on interacting RIS with the industrial automation environment by taking actions which are the phase shifts at the RIS elements, CBL variables, and BS beamforming to maximize the expected observed reward, i.e., the total FBL rate. We assess the performance loss of the system when the RIS is non-ideal, i.e., with non-linear amplitude response, and compare it with ideal RIS without impairments. The numerical results show that optimizing the RIS phase shifts, BS beamforming, and CBL variables via the proposed TD3 method is highly beneficial to improving the network total FBL rate as the proposed method with deterministic policy outperforms conventional methods.