Computational offloading is an efficient way to help constrained IoT devices by performing heavy tasks on Edge servers, especially tasks related to Machine Learning. Moreover, due to their limited learning capacity and memory size, such devices can only store a limited amount of data as a training set for their learning. Consequently, learning prediction is bound to be smeared with relatively high error. To mend that issue, IoT devices can federate the learning process with their pairs via an Edge server. However, offloading repeatedly the learning model through a wireless access network is time consuming. Hence, although learning collectively can reduce the learned model variance, it inflicts a communication cost depending on the selected Edge server. Therefore, in this paper, we model the Edge Selection problem as a non-cooperative game where devices autonomously and efficiently select an Edge server to reduce both their learning error and their communication cost. Depending on the characteristics of the dataset, we discern two different types of games. For each game type, we implemented and compared a semi-distributed algorithm based on Best Response dynamics. We compared the obtained results with the optimal centralized approach and with a less computationally intensive meta-heuristics, to assess the price of anarchy. Our numerical analysis shows that the Best Response algorithm strikes a good balance between efficiency and swift convergence.

Computational offloading is a pivotal solution to several Internet of Things (IoT) issues as it helps subdue the constrained nature of IoT devices. By harnessing the large capacity at the Edge, IoT devices with limited battery and storage can delegate certain tasks, especially those related to Machine Learning. Because of their restricted capacity, such devices can only store a limited amount of data as a training set for their learning, leading to a faulty prediction with high error rate. To tackle that issue, IoT devices can federate the learning process with other devices while the Edge server acts as an aggregator. However, selecting the appropriate Edge is a significant challenge. In fact, although learning collectively can reduce the prediction error, it also brings about a communication cost that depends on the selected Edge. Thus, in this paper, we propose a Non-Cooperative game where devices autonomously and efficiently select an Edge server in order to reduce both their learning error and communication cost.

In this paper, we aim to tackle the challenges of resource block scheduling and power allocation in the context of multi-cell full-duplex wireless networks. This is a more realistic setting than the single cell scenario, and it better envisions how full-duplex wireless communications could eventually be implemented. We propose an optimal queue-aware joint scheduling and power allocation algorithm for full-duplex wireless networks in a multi-cell scenario. Because of its mathematical intractability, we decouple the problem and solve it for scheduling first, and for power allocation second. We consider both indoor and outdoor scenarios and show that the gains of multi-cell full-duplex wireless networks, with respect to their half-duplex counterparts, are not always prevalent. Furthermore, we highlight the importance of inter-cell cooperation when it comes to scheduling resources and show that depending on the scenario at hand, interference mitigation from inter-cell cooperation can improve the performance of user equipment in terms of throughput and waiting delay. Finally, we show that power allocation can improve user equipment throughput with its efficiency being tied to the deployment scenario at hand.

This paper explores the use of multi-agent deep learning as well as learning to cooperate principles to meet strict service level agreements, in terms of throughput and end-to-end delay, for a set of classified network flows. We consider agents built on top of a weighted fair queuing algorithm that continuously set weights for three flow groups: gold, silver, and bronze. We rely on a novel graph-convolution based, multi-agent reinforcement learning approach known as DGN. As benchmarks, we propose centralized and distributed deep Q-network algorithms and evaluate their performances in different network, traffic, and routing scenarios, highlighting both the effectiveness of our proposals and the importance of agent cooperation. We show that our DGN-based approach meets stringent throughput and delay requirements across different scenarios, decreasing silver and bronze flow median waiting delays by more than 50 % and reducing the SLA violations of the latter by nearly 60 %, with respect to a classic priority queuing approach.

Faced with the limitations of the Aloha random access scheme and spread spectrum techniques, LoRaWAN is yet to realize its potential as the flagship technology for large-scale Internet-of-Things applications. LoRaWAN allows for low power and long range communications. Nonetheless, concurrent transmissions on the same spreading factors (SFs), increased with the inevitable densification of device deployment, will lead to collisions and degradation in performance. The problem is further amplified due to the shortage in radio resources, with multiple operators utilizing the same unlicensed frequency bands. In this article, we investigate different interoperator cooperation schemes and devise multiple algorithms for SF assignment in a multioperator LoRaWAN deployment scenario. We start by proposing a proportional fair optimal formulation for the assignment with the objective of maximizing the logarithmic sum of the normalized throughput per SF. Under the assumption of partial operator cooperation, we propose a gradient ascent-based iterative algorithm for solving the SF assignment problem, and a game theory-based approach, wherein each network operator seeks to maximize its own normalized throughput. Finally, and with cooperation between different operators bound to be limited, we use recurrent neural networks to enable the prediction of the success rate per SF. This prediction allows the different operators to assign SFs with minimum cooperation. We simulate our proposals and compare them to the legacy LoRaWAN approach as well as others in the state of the art, highlighting the gains they produce in terms of total normalized throughput and packet delivery ratios.

The benefits of full-duplex wireless communications, specifically with respect to their half-duplex counterparts, are now well under examination. As a result, research into the corresponding scheduling and power allocation algorithms has thrived. In this paper, we propose a non-cooperative game theoretic algorithm for power allocation in full-duplex orthogonal frequency division multiple access networks. The game is played between user equipment on the uplink, and the base station on the downlink. The objective of the game is two-fold: maximizing the signal-to-noise-plus-interference ratio, while hindering the harmful interferences resulting from full-duplex operation. We prove that our game is super-modular. For such a game, a best response algorithm is capable of attaining a Nash equilibrium. We simulate our proposal along with a fairness based scheduling algorithm and show that it improves user equipment throughput and reduces the waiting delay.

In Half-Duplex (HD) systems, at a given time instant, a radio resource is exclusively assigned to one User Equipment (UE) either for transmission or for reception. Full-Duplex (FD) networks promise to increase the system's capacity by allocating resources simultaneously to two UEs: one uplink UE and one downlink UE. In order to enhance the network's spectral efficiency, the system needs to deal with two major types of interference: self-interference and co-channel interference. This is one of the main challenges of scheduling in FD systems. In this article, we propose two algorithms for scheduling in FD Orthogonal Frequency Division Multiple Access Systems (FD-OFDMA). First, we propose an FD Maximum Signal-to-Interference-plus-Noise ratio (FD Max-SINR) algorithm, which allocates resources in FD depending on the SINR values of the UEs. Second, we propose a Hybrid Max-SINR scheduling algorithm. This hybrid algorithm chooses astutely between allocating the resources either in HD or FD, in a manner that maximizes the SINR. We evaluate these algorithms and compare them to HD Max-SINR in terms of UE throughput and average waiting delay. According to the simulation results, FD Max-SINR provides, in comparison with its HD counterpart, increased throughput for uplink UEs. It also almost doubles the throughput for the downlink ones. FD Max-SINR reduces the waiting delay that UEs undergo. Furthermore, for relatively low values of Self-Interference Cancellation (SIC), our Hybrid Max-SINR algorithm can still provide higher network throughput compared to HD Max-SINR.