We apply the self-supervised learning technique of vision transformers masked autoencoder (ViT MAE) models with the goal of to producing a feature extractor vision transformer (ViT) backbone for neural networks that receive seismic data as input. We then evaluate the quality of these backbones by coupling them to a simple linear prediction head and fine-tuning these models in a seismic semantic segmentation task. We compare domain-specific ViT MAE against cross-domain pretrained and randomly initialized ViTs, and show that it yields superior performance in low-data regimes. Furthermore, we also demonstrate that pretraining loss correlates with downstream performance, supporting its use as a proxy for feature quality.

The Internet of Things (IoT) increasingly relies on edge computing nodes to decentralize computation and enhance processing power near IoT devices. However, IoT edge computing nodes are generally not designed for highly intensive machine learning (ML) training. In current IoT architectures, multiple edge computing nodes are strategically positioned near IoT devices, each accessing only a portion of the data generated by the entire IoT network. In this paper, we bring the concept of Federated Learning (FL) to this scenario, by enabling each IoT edge computing node to run lightweight ML models on local datasets cooperatively. Our primary goal is to design a decision tree-based solution for cooperative IoT edge computing, termed Federated Decision Trees (FeDT). To achieve this, we propose four FL strategies based on decision trees, which aggregate the learning contributions of multiple FL clients while adhering to FL principles. Our results demonstrate that the proposed strategies can achieve approximately $80 \%$ of the performance of a centralized ML model in terms of Pearson correlation. Furthermore, compared to FedAVG, a classical FL solution, FeDT requires approximately four times fewer training rounds to converge.

The 5 G networks and their Open Radio Access Networks (O-RAN) architecture face significant challenges in resource management due to their extended flexibility and technological diversity. O-RAN’s open, disaggregated architecture creates a heterogeneous environment that requires effective integration and analysis of data from various components. In this context, accurately predicting the computational utilization of virtual base stations (vBS) emerges as a critical challenge, essential for optimizing resource allocation and addressing the dynamic demands of 5 G and 6 G O-RAN networks. Traditional forecasting techniques often struggle with the complexity and variability of data in this scenario, necessitating advanced AI and ML approaches. We propose an Adaptive Neuro-Fuzzy Inference System (ANFIS) model for multi-target regression to predict CPU utilization in vBS. By combining neural networks and fuzzy logic, ANFIS enhances both prediction accuracy and interpretability, making it ideal for the complexities of O-RAN 5G/6G networks. Our model, tested on publicly available O-RAN datasets, outperforms traditional ML methods. These results position ANFIS as an effective tool for optimizing resource management and enabling transparent decision-making in 5G/6G infrastructures, providing valuable support for network operators seeking efficient and scalable solutions in the evolving ORAN landscape.

Patched inference is a widely used technique in machine learning (ML) that enables fixed-shape models to process arbitrarily large or variably sized inputs by dividing them into smaller, compatible patches. This approach is particularly useful in domains such as seismic processing, medical imaging, and electron microscopy, where data samples often exceed the memory capacity of individual computing nodes. While patched inference is effective for leveraging pre-trained models and operating on resource-constrained hardware, there remains a lack of tools supporting its efficient, distributed execution at scale.To address this gap, we introduce SPINN (Scalable Parallel INference Network), a Python library designed to streamline and accelerate patched inference on high-performance computing (HPC) systems. SPINN supports data partitioning, patch-wise processing using user-defined ML models, and result aggregation, all while leveraging distributed computing frameworks such as Dask and Ray.We validate SPINN on two seismic interpretation tasks, fault detection and facies segmentation, using both public and large-scale private data (up to 272 GB). Experiments demonstrate that SPINN enables smoother prediction outputs via overlapping patches and achieves superlinear scalability with Dask in HPC environments, significantly outperforming conventional solutions such as the NVIDIA Triton Inference Server in large-scale scenarios. SPINN thus emerges as a robust and scalable solution for applying deep learning inference to massive data samples in memory-constrained or compute-intensive settings.

Optimal code-expanded random access (OptCeRA) is a promising random access (RA) technique for supporting massive Internet of Things (MIoT) in mobile networks. It allows devices to employ, as contention resources, codewords formed by the transmission of preambles in consecutive RA slots and selected from a maximum average distance code. This paper introduces the adaptive version of the OptCeRA scheme to cope with substantial random access channel (RACH) demands, which is particularly relevant for supporting MIoT scenarios in such networks. A practical algorithm to dynamically adjust the OptCeRA scheme parameters to the RA channel load based on optimizing the RA success probability is proposed. Numerical results show that the proposal significantly increases the capacity of the networks in terms of the number of simultaneous random access attempts compared to the state-of-the-art RA schemes.

The support of Mobile Fronthaul (MFH) over Passive Optical Networks (PONs) poses significant challenges due to the stringent latency and bandwidth requirements of Functional Splitting (FS). This paper addresses the problem of Quality of Service (QoS) provisioning in next-generation Ethernet PON (NG-EPON) for the transport of traffic generated by multiple different FS options. We propose a PON bandwidth allocation algorithm that distributes the resources for the Optical Network Units (ONUs) serving Functional Split (FS) options based on their bandwidth and latency requirements in networks with customers renting/owning more than one ONU (multi-ONU customers). Simulation results show that our proposal significantly improves network resource utilization for multi-ONU customers, meeting the latency requirements of the different FS options while reducing the required bandwidth.

Federated Learning (FL) allows distributed devices, known as clients, to train Machine Learning (ML) models collaboratively without sharing sensitive data. A characteristic of FL for mobile and IoT environments is system heterogeneity among clients, which can vary from low-end devices with constrained communication and computing resources to powerful devices with high-speed network access and dedicated GPUs. As the server must wait for all the clients to communicate their updates, slow clients (a.k.a. stragglers) will significantly increase the training time. To tackle this problem, we propose FedPulse, a Partial Training (PT) based mechanism to mitigate the effect of stragglers in FL. The idea is to reduce the training time by dynamically allocating smaller submodels to resource-constrained clients. Experimental results on famous classification datasets show that the proposed solution outperforms other submodel allocation mechanisms and reduces the training time by up to 58% with an accuracy loss of less than 1% when compared to FedAvg.

Federated learning (FL) is a distributed approach that enables collaborative training of a shared machine learning (ML) model for a given task. FL requires bandwidth-demanding communication between devices and a central server, which is a cause of many issues such as communication bottlenecks and scaling in the network. Therefore, we introduce the CCS (Compressed Client Selection) algorithm aimed at decreasing the overall communication costs for fitting a model in the FL environment. CCS employs a biased client selection strategy that reduces the number of devices training the ML model and the number of rounds required to reach convergence. In addition, the compression method Count Sketch is implemented to reduce the overhead in client-to-server communication. A use case on the Human Activity Recognition dataset is performed to evaluate CCS and compare it with other state-of-the-art approaches. Experimental evaluations show that CCS efficiently reduces the overall communication overhead for fitting a model and its convergence in a FL environment. In particular, CCS reduces up to 90% the communication overhead compared to literature approaches while providing good convergence even in scenarios where the data are not-independently and identically distributed among client devices.

Infrastructure service providers (InPs) can employ bandwidth sharing to offer new services and business models to their customers. In this paper, we introduce a dynamic bandwidth allocation (DBA) algorithm which allows cooperation among Ethernet PON (EPON) customers so that they can share unused bandwidth among themselves without affecting their guaranteed bandwidth. Simulation results show that our proposal increases the throughput and decreases the delay for cooperative customers.

The collision of multiple MSG3transmissions due to the selection of the same preamble sequence in the Long Term Evolution (LTE) Random Access procedure is an important problem which can impact on the performance of cellular Internet of Things (IoT) networks. In this paper, we propose a standard-compatible probabilistic retransmission approach to reduce the number of collisions of MSG3messages in cellular IoT technologies. In our proposal, every Machine-Type Communications (MTC) device with an uplink grant for retransmitting an MSG3message uses a probability value to decide whether or not to transmit. Two retransmission policies were proposed to reduce the number of simultaneous MSG3messages received at the base station. To apply these policies, the estimation of the number of MTC devices trying random access in a given Random Access Opportunity is required. A novel method to estimate this value at the device side is proposed based on Random Access Response (RAR) message counting and the Access Class Barring (ACB) barring probability. Results derived via simulations show that the proposed approach decreases the number of collisions of MSG3messages, reducing the access delay and energy consumption, as well as decreasing the utilization of the Packet Uplink Shared Channel (PUSCH) when compared to conventional LTE Random Access scheme.

This paper introduces a mechanism for the support of multi-ONU service level agreements (SLAs) in dynamic bandwidth allocation (DBA) algorithms for Ethernet passive optical networks (EPON). The employment of SLAs for multiple optical network units (ONUs) instead of individual ONUs allows better utilization of the bandwidth reserved for these ONUs. The proposed DBA mechanism allows customers owning multiple ONUs to redistribute the aggregated bandwidth of the group of ONUs to better balance the bandwidth utilization. The proposed DBA can be employed in different use cases such as mobile backhauling/fronthauling, PON virtualization, and multi-site enterprise networking. Simulation results show that the proposed DBA improves the network performance.

Energy Efficiency is one of the main concerns in the design of wireless communication protocols, especially for battery-enabled devices, such as smartphones, tablets and laptops. In this paper, we focus on the impact of transmission fragmentation and resource fragmentation on the energy efficiency of SC-FDMA systems. To deal with these two problems, we introduce the Best Edge Set (BESt) algorithm for adoption in packet schedulers for Single-Carrier Frequency Division Multiple Access (SC-FDMA) LTE/LTE-Advanced systems. The BESt algorithm employs a novel Physical Resource Block (PRB) allocation strategy to avoid resource fragmentation as well as a new way to prioritize User Equipment (UE) transmissions to reduce the transmission fragmentation and the energy consumed in transmissions. Simulation results show the advantages of using the BESt algorithm and the strong correlation between transmission fragmentation and energy efficiency.

In this paper, we introduce the Preamble Priority-Aware (PPA) Packet Downlink Control Channel (PDCCH) resources allocation algorithm to provide Quality of Service (QoS) differentiation in the Random Access (RA) procedure of the LTE-Advanced technology. The PPA algorithm uses the preamble priority defined by the RA procedure and Radio Access Network (RAN) overload control schemes to make scheduling decisions. Results derived via simulation show that the proposed PDCCH algorithm significantly increases the chance of accessing the network as well as reducing random-access delays for user equipment employing prioritized preamble sequences. Thus, the proposed algorithm provides enhanced QoS support to prioritized users during intense RA attempts.

The concept of preamble-priority awareness in downlink control signaling scheduling was recently proposed to provide Quality of Service (QoS) differentiation in the Random Access (RA) procedure of the LTE/LTE-Advanced technology. This approach employs the information about preamble-priority levels used in the the initial phase of the RA procedure to schedule random access response messages. In this paper, we extend the application of this concept to the scheduling of other control messages and analyze its impact when the RACH Resource Separation (RRS) scheme and the traditional RA scheme with both the contention-free and the contention-based modes are used under heavily-loaded, highly-synchronized Machine-Type Communications (MTC) scenarios. Our results derived via extensive simulations show that the preamble-priority-aware concept provides QoS differentiation to users utilizing contention-free preambles. Furthermore, this concept helps to achieve the goal of isolation between traditional LTE users and MTC devices when the RRS scheme is used.

The Internet of Things (IoT) paradigm stands for virtually interconnected objects that are identifiable and equipped with sensing, computing, and communication capabilities. Services and applications over the IoT architecture can take benefit of the long-term evolution (LTE)/LTE-Advanced (LTE-A), cellular networks to support machine-type communication (MTC). Moreover, it is paramount that MTC do not affect the services provided for traditional human-type communication (HTC). Although previous studies have evaluated the impact of the number of MTC devices on the quality of service (QoS) provided to HTC users, none have considered the joint effect of allocation of control resources and the LTE random-access (RA) procedure. In this paper, a novel scheme for resource allocation on the packet downlink (DL) control channel (PDCCH) is introduced. This scheme allows PDCCH scheduling algorithms to consider the resources consumed by the random-access procedure on both control and data channels when prioritizing control messages. Three PDCCH scheduling algorithms considering RA-related control messages are proposed. Moreover, the impact of MTC devices on QoS provisioning to HTC traffic is evaluated. Results derived via simulation show that the proposed PDCCH scheduling algorithms can improve the QoS provisioning and that MTC can strongly impact on QoS provisioning for real-time traffic.

The Long Term Evolution (LTE) and LTE-Advanced technologies aim at providing improved users' experience by increasing data rate, enhancing coverage and supporting Quality of Service (QoS) to different service classes. However, a large number of User Equipment (UE) devices trying to access the network in a short period can overload the Radio Access Network (RAN). In this situation, more access attempts to the system are made than it can handle, resulting in low access probabilities and poor network performance. In this paper, we introduce the QoS-Aware Self-Adaptive RAN Overload Control (QoS-Dracon) mechanism to reduce the RAN overload problem, taking into account users' QoS requirements. This is achieved by employing a QoS Class Identifier-dependent backoff scheme and an Access Class Barring-based RAN overload control mechanism. QoS-Dracon prioritizes delay-sensitive UE devices over delay-tolerant ones when performing Random Access (RA) procedure. Results derived via simulation show that the proposed mechanism yields satisfactory access delays for delay-sensitive users regardless of the UE devices type attempting to access the channel.

This paper introduces a novel LTE uplink scheduler called Hybrid Z-Based QoS Scheduler (HZBQoS), a fully standard-compliant LTE scheduler designed to operate in ONU-eNB devices of integrated LTE/TDM-EPON networks. The HZBQoS scheduler provides delay bound and guaranteed rate even when the backhaul and mobile network are heavily loaded. We evaluated the proposed scheduler under heterogeneous traffic and compared its performance to that of another LTE uplink scheduler, called Z-Based QoS Scheduler (ZBQoS), which does not take into account the variability of the backhaul link capacity. Simulation results show that HZBQoS is able to provide QoS requirements in the integrated network and outperforms the ZBQoS scheduler.

This paper introduces a novel time-domain (TD) LTE uplink scheduler called Z-Based QoS Scheduler (ZBQoS) which is fully standard-compliant. The ZBQoS scheduler provides Quality of Service (QoS) requirements, supporting delay bound and guaranteed rate even when the network is heavily loaded. We evaluate the proposed scheduler under heterogeneous traffic and compare its performance to that of another TD scheduler, called Bandwidth and QoS Aware (BQA), recently proposed. Simulation results show that ZBQoS scheduler reduces significantly delay of real-time traffic, while it is able to maintain lower packet loss ratio (PLR), when compared with the performance of the BQA scheduler which greatly surpasses the recommended PLR value under heavily loaded scenarios.