IEEE 802.11 (Wi-Fi) is one of the technologies that provides high performance with a high density of connected devices to support emerging demanding services, such as virtual and augmented reality. However, in highly dense deployments, Wi-Fi performance is severely affected by interference. This problem is even worse in new standards, such as 802.11n/ac, where new features such as Channel Bonding (CB) are introduced to increase network capacity but at the cost of using wider spectrum channels. Finding the best channel assignment in dense deployments under dynamic environments with CB is challenging, given its combinatorial nature. Therefore, the use of analytical or system models to predict Wi-Fi performance after potential changes (e.g., dynamic channel selection with CB, and the deployment of new devices) are not suitable, due to either low accuracy or high computational cost. This paper presents a novel, data-driven approach to speed up this process, using a Graph Neural Network (GNN) model that exploits the information carried in the deployment’s topology and the intricate wireless interactions to predict Wi-Fi performance with high accuracy. The evaluation results show that preserving the graph structure in the learning process obtains a 64% increase versus a naive approach, and around 55% compared to other Machine Learning (ML) approaches when using all training features.

Traffic Classification (TC) systems allow inferring the application that is generating the traffic being analyzed. State-of-the-art TC algorithms are based on Deep Learning (DL) and have outperformed traditional methods in complex and modern scenarios, even if traffic is encrypted. Most of the works on TC assume the traffic flows on a wired network under the same network management domain. This assumption limits the capabilities of TC systems in wireless networks since users’ traffic on one network domain can be negatively impacted by undetected users’ traffic from other network domains or detected ones but with no traffic context in a shared spectrum. To solve this problem, we introduce a novel framework to achieve TC at any layer on the radio network stack. We propose a spectrum-based procedure that uses a DL-based classifier to realize this framework. We design two DL-based classifiers, a novel Convolutional Neural Network (CNN) spectrum-based TC and a Recurrent Neural Networks (RNN) as baseline architecture, and benchmark their performance on three TC tasks at different radio stack layers. The datasets were generated by combining packet traces from real transmissions with an 802.11 standardcompliant waveform generator. Performance evaluations show that the best model can achieve an accuracy above 92% in the most demanding TC task, a drop of only 4.37% in accuracy compared to a byte-based DL approach, with micro-second perpacket prediction time, which is very promising for delivering real-time spectrum-based traffic analyzers.

Autonomous network management is crucial for Fifth Generation (5G) and Beyond 5G (B5G) networks, where a constantly changing environment is expected and network configuration must adapt accordingly. Modeling tools are required to predict the impact on performance (packet and delay loss) when new traffic demands arrives and when changes in routing paths are applied in the network. Mathematical analysis and network simulators are techniques for modeling networks but both have limitations, as the former provides low accuracy and the latter requires high execution times. To overcome these limitations, machine learning (ML) algorithms, and more specifically, graph neural networks (GNNs), are proposed for network modeling due to their ability to capture complex relationships from graph-like data while predicting network properties with high accuracy and low computational requirements. However, one of the main issues when using GNNs is their lack of generalization capability to larger networks, i.e., when trained in small networks (in number of nodes, paths length, links capacity), the accuracy of predictions on larger networks is poor. This paper addresses the GNN generalization problem by the use of fundamental networking concepts. Our solution is built from a baseline GNN model called RouteNet (developed by Barcelona Neural Networking Center-Universitat Politècnica de Catalunya (BNN-UPC)) that predicts the average delay in network paths, and through a number of simple additions significantly improves the prediction accuracy in larger networks. The improvement ratio compared to the baseline model is 101, from a 187.28% to a 1.828%, measured by the Mean Average Percentage Error (MAPE). In addition, we propose a closed-loop control context where the resulting GNN model could be potentially used in different use cases.

Next-generation communication systems will face new challenges related to efficiently managing the available resources, such as the radio spectrum. DL is one of the optimization approaches to address and solve these challenges. However, there is a gap between research and industry. Most AI models that solve communication problems cannot be implemented in current communication devices due to their high computational capacity requirements. New approaches seek to reduce the size of DL models through quantization techniques, changing the traditional method of operations from a 32 (or 64) floating-point representation to a fixed point (usually small) one. However, there is no analytical method to determine the level of quantification that can be used to obtain the best trade-off between the reduction of computational costs and an acceptable accuracy in a specific problem. In this work, we propose an analysis methodology to determine the degree of quantization in a DNN model to solve the problem of AMR in a radio system. We use the Brevitas framework to build and analyze different quantized variants of the DL architecture VGG10 adapted to the AMR problem. The evaluation of the computational cost is performed with the FINN framework of Xilinx Research Labs to obtain the computational inference cost. The proposed design methodology allows us to obtain the combination of quantization bits per layer that provides an optimal trade-off between the model performance (i.e., accuracy) and the model complexity (i.e., size) according to a set of weights associated with each optimization objective. For example, using the proposed methodology, we found a model architecture that reduced 75.8% of the model size compared to the non-quantized baseline model, with a performance degradation of only 0.06%.

Optical Network Virtualization has gained attention in recent years mainly from industry due to its direct application to Optical Data Centers. One of the main challenges involved in optical virtualization deals with the efficient mapping of virtual resources onto physical resources, specially when considering the intrinsic constraints of the elastic optical layer based on O-OFDM such as optical spectrum continuity and contiguity. In this paper, we propose a greedy randomized heuristic algorithm based on Path-Ranking on the link mapping stage, aiming to improve the acceptance ratio in terms of allocated Virtual Optical Networks. In order to evaluate our proposal, we implement a baseline algorithm frequently used in virtual network embedding problems. Results shown that our mapping approach outperforms the baseline algorithm in terms of blocking probability and load balancing.