In semi-decentralized federated learning, devices primarily rely on device-to-device communication but occasionally interact with a central server. Periodically, a sampled subset of devices uploads their local models to the server, which computes an aggregate model. The server can then either (i) share this aggregate model only with the sampled clients (sampled-to-sampled, S2S) or (ii) broadcast it to all clients (sampled-to-all, S2A). Despite their practical significance, a rigorous theoretical and empirical comparison of these two strategies remains absent. We address this gap by analyzing S2S and S2A within a unified convergence framework that accounts for key system parameters: sampling rate, server aggregation frequency, and network connectivity. Our results, both analytical and experimental, reveal distinct regimes where one strategy outperforms the other, depending primarily on the degree of data heterogeneity across devices. These insights lead to concrete design guidelines for practical semi-decentralized FL deployments.

Efficient and low-complexity beamforming design is an important element of satellite communication systems with mobile receivers equipped with phased arrays. In this work, we apply the simultaneous perturbation stochastic approximation (SPSA) method with successive sub-array selection for finding the optimal antenna weights that maximize the received signal power at a uniform plane array (UPA). The proposed algorithms are based on iterative gradient approximation by injecting some carefully designed perturbations on the parameters to be estimated. Additionally, the successive sub-array selection technique enhances the performance of SPSA-based algorithms and makes them less sensitive to the initial beam direction. Simulation results show that our proposed algorithms can achieve efficient and reliable performance even when the initial beam direction is not well aligned with the satellite direction.

In the above article [1], this shows a corrected statement and proof of Theorem 1. The analysis in [1] incorrectly treated subsets of scheduled devices independently, overlooking that global iterates depend on all subsets. In addition, in the proof of Lemma 2 (now Lemma 3 in this document), the expectation over the randomness in the gradient sparsification was inaccurate.

Decentralized learning enables distributed agents to collaboratively train a shared machine learning model without a central server, through local computation and peer-to-peer communication. Although each agent retains its dataset locally, sharing local models can still expose private information about the local training datasets to adversaries. To mitigate privacy attacks, a common strategy is to inject random artificial noise at each agent before exchanging local models between neighbors. However, this often leads to utility degradation due to the negative effects of cumulated artificial noise on the learning algorithm. In this work, we introduce CorN-DSGD, a novel covariance-based framework for generating correlated privacy noise across agents, which unifies several state-of-the-art methods as special cases. By leveraging network topology and mixing weights, CorN-DSGD optimizes the noise covariance to achieve network-wide noise cancellation. Experimental results show that CorN-DSGD cancels more noise than existing pairwise correlation schemes, improving model performance under formal privacy guarantees.

Federated learning (FL) has received significant attention in recent years for its advantages in efficient training of machine learning models across distributed clients without disclosing user-sensitive data. Specifically, in federated edge learning (FEEL) systems, the time-varying nature of wireless channels introduces inevitable system dynamics in the communication process, thereby affecting training latency and energy consumption. In this work, we further consider a streaming data scenario where new training data samples are randomly generated over time at edge devices. Our goal is to develop a dynamic scheduling and resource allocation algorithm to address the inherent randomness in data arrivals and resource availability under long-term energy constraints. To achieve this, we formulate a stochastic network optimization problem and use the Lyapunov drift-plus-penalty framework to obtain a dynamic resource management design. Our proposed algorithm makes adaptive decisions on device scheduling, computational capacity adjustment, and allocation of bandwidth and transmit power in every round. We provide convergence analysis for the considered setting with heterogeneous data and time-varying objective functions, which supports the rationale behind our proposed scheduling design. The effectiveness of our scheme is verified through simulation results, demonstrating improved learning performance and energy efficiency as compared to baseline schemes.

This work focuses on the communication perspective of decentralized learning over wireless networks, using consensus-based decentralized stochastic gradient descent (D-SGD). Considering the actual communication cost or delay caused by in-network information exchange in every iteration, our goal is to achieve fast convergence of the algorithm measured by improvement per transmission slot. We propose BASS, an efficient communication framework for D-SGD over wireless networks with broadcast-based subgraph sampling. More explicitly, in every iteration, we activate multiple subsets of non-interfering nodes to broadcast model updates to their neighbors. These subsets are activated randomly over time with some probabilities under a given communication cost (e.g., number of transmission slots per iteration). During the consensus update step, only bi-directional links are effectively considered to preserve the communication symmetry. As compared to existing link-based scheduling methods, the broadcasting nature of wireless channels provides inherent advantages in speeding up convergence of decentralized learning by creating more communicated links under the same number of transmission slots.

We study information freshness in a cache updating system with randomly located caches whose distribution follows a Poisson point process. A finite set of content items/files are maintained at a central server, updated randomly over time, and requested randomly by users in the network. To provide timely service for content requests, the central server proactively delivers new file versions to the distributed caches, subject to some constraint on the updating costs per unit area. When a user requests a file, it retrieves the most up-to-date version from the set of caches located within its searching range. Considering the randomness in content requests, the number of caches and cache updating decisions, we derive the distribution of user-perceived version age of files and propose a spatial information freshness metric. Our results illustrate the interplay between cache density and file updating probabilities in terms of their impact on information freshness in large random network.

Federated Learning (FL) is a collaborative machine learning (ML) framework that combines on-device training and server-based aggregation to train a common ML model among distributed agents. In this work, we propose an asynchronous FL design with periodic aggregation to tackle the straggler issue in FL systems. Considering limited wireless communication resources, we investigate the effect of different scheduling policies and aggregation designs on the convergence performance. Driven by the importance of reducing the bias and variance of the aggregated model updates, we propose a scheduling policy that jointly considers the channel quality and training data representation of user devices. The effectiveness of our channel-aware data-importance-based scheduling policy, compared with state-of-the-art methods proposed for synchronous FL, is validated through simulations. Moreover, we show that an “age-aware” aggregation weighting design can significantly improve the learning performance in an asynchronous FL setting.

We consider distributed average consensus in a wireless network with partial communication to reduce the number of transmissions in every iteration/round. Considering the broadcast nature of wireless channels, we propose a probabilistic approach that schedules a subset of nodes for broadcasting information to their neighbors in every round. We compare several heuristic methods for assigning the node broadcast probabilities under a fixed number of transmissions per round. Furthermore, we introduce a pre-compensation method to correct the bias between the consensus value and the average of the initial values, and suggest possible extensions for our design. Our results are particularly relevant for developing communication-efficient consensus protocols in a wireless environment with limited frequency/time resources.

We consider a resource-constrained IoT network, where users make on-demand requests to a cache-enabled edge node to send status updates about various random processes, each monitored by an energy harvesting sensor. The edge node serves users’ requests by either commanding the corresponding sensor to send a fresh status update or retrieving the most recently received measurement from the cache. We aim to find a control policy at the edge node to minimize the average age of information (AoI) of the received measurements upon requests, i.e., average on-demand AoI, subject to per-slot transmission and energy constraints. We develop a low-complexity algorithm – termed relax-then-truncate – and prove that it is asymptotically optimal as the number of sensors goes to infinity. Numerical results assess the performance of the proposed method.

Intelligent reflecting surface (IRS) and device-to-device (D2D) communication are two promising technologies for improving transmission reliability between transceivers in communication systems. In this paper, we consider the design of reliable communication between the access point (AP) and actuators for a downlink multiuser multiple-input single-output (MISO) system in the industrial IoT (IIoT) scenario. We propose a two-stage protocol combining IRS with D2D communication so that all actuators can successfully receive the message from AP within a given delay. The superiority of the protocol is that the communication reliability between AP and actuators is doubly augmented by the IRS-aided first-stage transmission and the second-stage D2D transmission. A joint optimization problem of active and passive beamforming is formulated, which aims to maximize the number of actuators with successful decoding. We study the joint beamforming problem for cases where the channel state information (CSI) is perfect and imperfect. For each case, we develop efficient algorithms that include convergence and complexity analysis. Simulation results demonstrate the necessity and role of IRS with a well-optimized reflection matrix, and the D2D network in promoting reliable communication. Moreover, the proposed protocol can enable reliable communication even in the presence of stringent latency requirements and CSI estimation errors.

We consider a resource-constrained IoT network, where multiple users make on-demand requests to a cache-enabled edge node to send status updates about various random processes, each monitored by an energy harvesting sensor. The edge node serves users’ requests by deciding whether to command the corresponding sensor to send a fresh status update or retrieve the most recently received measurement from the cache. Our objective is to find the best actions of the edge node to minimize the average age of information (AoI) of the received measurements upon request, i.e., average on-demand AoI, subject to per-slot transmission and energy constraints. First, we derive a Markov decision process model and propose an iterative algorithm that obtains an optimal policy. Then, we develop an asymptotically optimal low-complexity algorithm – termed relax-then-truncate – and prove that it is optimal as the number of sensors goes to infinity. Simulation results illustrate that the proposed relax-then-truncate approach significantly reduces the average on-demand AoI compared to a request-aware greedy policy and a weighted AoI policy, and also depict that it performs close to the optimal solution even for moderate numbers of sensors.

In this paper, we consider a cellular-based Internet of things (IoT) network consisting of IoT devices that can communicate directly with each other in a device-to-device (D2D) fashion as well as send real-time status updates about some underlying physical processes observed by them. We assume that such real-time applications are supported by cellular networks where cellular base stations (BSs) collect status updates over time from a subset of the IoT devices in their vicinity. We characterize two performance metrics: i) the network throughput which quantifies the performance of D2D communications, and ii) the Age of Information which quantifies the performance of the real-time IoT-enabled applications. Concrete analytical results are derived using stochastic geometry by modeling the locations of IoT devices as a bipolar Poisson Point Process (PPP) and that of the BSs as another Independent PPP. Our results provide useful design guidelines on the efficient deployment of future IoT networks that will jointly support D2D communications and several cellular network-enabled real-time applications.

In this work, we consider a random access Internet of Things IoT wireless network assisted by two aggregators collecting information from two disjoint groups of sensors. The nodes and the aggregators are transmitting in a random access manner under slotted time, the aggregators perform network-level cooperation for the data collection. The aggregators are equipped with queues to store data packets that are transmitted by the network nodes and relaying them to the destination node. We characterize the throughput performance of the IoT network and we obtain the stability conditions for the queues at the aggregators. We apply the theory of boundary value problems to analyze the delay performance. Our results show that the presence of the aggregators provides significant gains in the IoT network performance, in addition, we provide useful insights regarding the scalability of the IoT network.

Average consensus algorithms have wide applications in distributed computing systems where all the nodes agree on the average value of their initial states by only exchanging information with their local neighbors. In this letter, we look into link-based network metrics which are polynomial functions of pair-wise node attributes defined over the links in a network. Different from node-based average consensus, such link-based metrics depend on both the distribution of node attributes and the underlying network topology. We propose a general algorithm using the weighted average consensus protocol for the distributed computation of link-based network metrics and provide the convergence conditions and convergence rate analysis.

This paper studies the interplay between device-to-device (D2D) communications and real-time monitoring systems in a cellular-based Internet of Things (IoT) network. In particular, besides the possibility that the IoT devices communicate directly with each other in a D2D fashion, we consider that they frequently send time-sensitive information/status updates (about some underlying physical processes observed by them) to their nearest cellular base stations (BSs). Specifically, we model the locations of the IoT devices as a bipolar Poisson Point Process (PPP) and that of the BSs as another independent PPP. For this setup, we characterize the performance of D2D communications using the average network throughput metric whereas the performance of the real-time applications is quantified by the Age of Information (AoI) metric. The IoT devices are considered to employ a distance-proportional fractional power control scheme while sending status updates to their serving BSs. Hence, depending upon the maximum transmission power available, the IoT devices located within a certain distance from the BSs can only send status updates. This association strategy, in turn, forms theJohnson-Mehl (JM)tessellation, such that the IoT devices located in theJM cellsare allowed to send status updates. The average network throughput is obtained by deriving the mean success probability for the D2D links. On the other hand, the temporal mean AoI of a given status update link can be treated as a random variable over space since its success delivery rate is a function of the interference field seen from its receiver. Thus, in order to capture the spatial disparity in the AoI performance, we characterize the spatial moments of the temporal mean AoI. In particular, we obtain these spatial moments by deriving the moments of both the conditional success probability and the conditional scheduling probability for status update links. Our results provide useful design guidelines on the efficient deployment of future massive IoT networks that will jointly support D2D communications and several cellular network-enabled real-time applications.

In this paper, we study the joint power control and scheduling in uplink massive multiple-input multiple-output (MIMO) systems with random data arrivals. The data is generated at each user according to an individual stochastic process. Using Lyapunov optimization techniques, we develop a dynamic scheduling algorithm (DSA), which decides at each time slot the amount of data to admit to the transmission queues and the transmission rates over the wireless channel. The proposed algorithm achieves nearly optimal performance on the long-term user throughput under various fairness policies. Simulation results show that the DSA can improve the time-average delay performance compared to the state-of-the-art power control schemes developed for Massive MIMO with infinite backlogs.

We investigate the effect of bursty traffic in an LTE and Wi-Fi aggregation (LWA)-enabled network, where part of the LTE traffic is offloaded to Wi-Fi access points (APs) to boost the performance of LTE networks. A Wi-Fi AP maintains two queues containing data intended for the LWA-mode user and the native Wi-Fi user, and it is allowed to serve them simultaneously by using superposition coding (SC). With respect to the existing works on LWA, the novelty of our study consists of a random access protocol allowing the Wi-Fi AP to serve the native WiFi user with probabilities that depend on the queue size of the LWA-mode data. We analyze the throughput of the native Wi-Fi network, accounting for different transmitting probabilities of the queues, the traffic flow splitting between LTE and Wi-Fi, and the operating mode of the LWA user with both LTE and Wi-Fi interfaces. Our results provide fundamental insights in the throughput behavior of such aggregated systems, which are essential for further investigation in larger topologies.

Cell-free (CF) massive multiple-input multiple-output (MIMO) is an alternative topology for future wireless networks, where a large number of single-antenna access points (APs) are distributed over the coverage area. There are no cells but all users are jointly served by the APs using network MIMO methods. Prior works have claimed that the CF massive MIMO inherits the basic properties of cellular massive MIMO, namely, channel hardening and favorable propagation. In this paper, we evaluate if one can rely on these properties when having a realistic stochastic AP deployment. Our results show that channel hardening only appears in special cases, for example, when the pathloss exponent is small. However, by using 5-10 antennas per AP, instead of one, we can substantially improve the hardening. Only spatially well-separated users will exhibit favorable propagation, but when adding more antennas and/or reducing the pathloss exponent, it becomes more likely for favorable propagation to occur. The conclusion is that we cannot rely on the channel hardening and the favorable propagation when analyzing and designing the CF massive MIMO networks, but we need to use achievable rate expressions and resource allocation schemes that work well also in the absence of these properties. Some options are reviewed in this paper.

In this paper, we propose a decentralized access control scheme for interference management in device-to-device (D2D) underlaid cellular networks. Our method combines signal-to-interference ratio (SIR)-aware link activation with cellular guard zones in a system, where D2D links opportunistically access the licensed cellular spectrum when the activation conditions are satisfied. Analytical expressions for the success/coverage probability of both cellular and D2D links are derived. We characterize the impact of the guard zone radius and the SIR threshold on the D2D potential throughput and cellular coverage. A tractable approach is proposed to find the SIR threshold and guard zone radius that maximize the potential throughput of the D2D communication while ensuring sufficient coverage probability for the cellular uplink users. Simulations validate the accuracy of our analytical results and show the performance gain of the proposed scheme compared to prior state-of-the-art solutions.

In this paper, we analyze a shared access network with a fixed primary node and randomly distributed secondary nodes whose spatial distribution follows a poisson point process. The secondary nodes use a random access protocol allowing them to access the channel with probabilities that depend on the queue size of the primary node. Assuming a system with multipacket reception receivers, having bursty packet arrivals at the primary and saturated traffic at the secondary nodes, our protocol can be tuned to alleviate congestion at the primary. We analyze the throughput of the secondary network and the primary average delay, as well as the impact of the secondary node access probability and transmit power. We formulate an optimization problem to maximize the throughput of the secondary network under delay constraints for the primary node; in the case of no congestion control, the optimal access probability can be provided in closed form. Our numerical results illustrate the effect of network operating parameters on the performance of the proposed priority-based shared access protocol.

Long Term Evolution (LTE)-Wireless Local Area Network (WLAN) Path Aggregation (LWPA) based on Multipath Transmission Control Protocol (MPTCP) has been under standardization procedure as a promising and cost-efficient solution to boost Downlink (DL) data rate and handle the rapidly increasing data traffic. This paper aims at providing tractable analysis for the DL performance evaluation of large-scale LWPA networks with the help of tools from stochastic geometry. We consider a simple yet practical model to determine under which conditions a native WLAN Access Point (AP) will work under LWPA mode to help increasing the received data rate. Using stochastic spatial models for the distribution of WLAN APs and LTE Base Stations (BSs), we analyze the density of active LWPA-mode WiFi APs in the considered network model, which further leads to closed-form expressions on the DL data rate and area spectral efficiency (ASE) improvement. Our numerical results illustrate the impact of different network parameters on the performance of LWPA networks, which can be useful for further performance optimization.

Wireless content caching in small cell networks (SCNs) has recently been considered as an efficient way to reduce the data traffic and the energy consumption of the backhaul in emerging heterogeneous cellular networks. In this paper, we consider a cluster-centric SCN with combined design of cooperative caching and transmission policy. Small base stations (SBSs) are grouped into disjoint clusters, in which in-cluster cache space is utilized as an entity. We propose a combined caching scheme, where part of the cache space in each cluster is reserved for caching the most popular content in every SBS, while the remaining is used for cooperatively caching different partitions of the less popular content in different SBSs, as a means to increase local content diversity. Depending on the availability and placement of the requested content, coordinated multi-point technique with either joint transmission or parallel transmission is used to deliver content to the served user. Using Poisson point process for the SBS location distribution and a hexagonal grid model for the clusters, we provide analytical results on the successful content delivery probability of both transmission schemes for a user located at the cluster center. Our analysis shows an inherent tradeoff between transmission diversity and content diversity in our cooperation design. We also study the optimal cache space assignment for two objective functions: maximization of the cache service performance and the energy efficiency. Simulation results show that the proposed scheme achieves performance gain by leveraging cache-level and signal-level cooperation and adapting to the network environment and user quality-of-service requirements.

In this paper, we propose an adaptive cluster-centric small cell network with cooperative caching and transmission design, in which small base stations (SBSs) are grouped into disjoint clusters and in-cluster cache space is utilized as an entity. We design a combined caching scheme where part of the available cache space is reserved for caching the most popular content in every SBS, while the remaining is used to increase the content diversity. Depending on the availability of the requested content, either joint transmission (JT) or parallel transmission (PT) is used to deliver the content to the served user. We provide analytical results on the successful content delivery probability of both schemes for a user located at the cluster center. The optimal cache utilization strategy to maximize the cache service probability is determined as a function of network parameters, revealing an interesting tradeoff between transmission diversity and content diversity. Simulation results show that our proposed cooperative design finely combines the benefits of cache-level and signal-level cooperation schemes.

In this paper, we analyze a shared access network with one primary device and randomly distributed smart objects with secondary priority. Assuming random traffic at the primary device and saturated queues at the smart objects with secondary priority, an access protocol is employed to adjust the random access probabilities of the smart objects depending on the congestion level of the primary. We characterize the maximum throughput of the secondary network with respect to delay constraints on the primary. Our results highlight the impact of system design parameters on the delay and throughput behavior of the shared access network with massive number of connected objects.

In this paper, we propose a distributed interference and channel-aware opportunistic access control technique for D2D underlaid cellular networks, in which each potential D2D link is active whenever its estimated signal-to-interference ratio (SIR) is above a predetermined threshold so as to maximize the D2D area spectral efficiency. The objective of our SIR-aware opportunistic access scheme is to provide sufficient coverage probability and to increase the aggregate rate of D2D links by harnessing interference caused by dense underlaid D2D users using an adaptive decision activation threshold. We determine the optimum D2D activation probability and threshold, building on analytical expressions for the coverage probabilities and area spectral efficiency of D2D links derived using stochastic geometry. Specifically, we provide two expressions for the optimal SIR threshold, which can be applied in a decentralized way on each D2D link, so as to maximize the D2D area spectral efficiency derived using the unconditional and conditional D2D success probability respectively. Simulation results in different network settings show the performance gains of both SIR-aware threshold scheduling methods in terms of D2D link coverage probability, area spectral efficiency, and average sum rate compared to existing channel-aware access schemes.