OBJECTIVE: Cardiac diseases are a leading cause of global mortality. Electrocardiograms (ECGs) are essential for detecting abnormal cardiac rhythms. Smartwatches can record ECGs, similar to lead I ECGs recorded by a patient vitals monitor in a hospital, potentially helping clinicians in early diagnosis and improved management of cardiovascular diseases. While AI models have classified arrhythmias with human-level accuracy, their potential for broad screening remains underutilized. METHODS: We propose a deep learning based framework for diagnosing various cardiac arrhythmias using 10-second lead I ECG recordings, demonstrating lead I's utility in remote monitoring. Robustness was tested by introducing noise to simulate real-world conditions. Additionally, a novel data similarity assessment metric was developed to enhance transfer learning and external dataset validation. RESULTS: Using over 60,000 ECGs from the PhysioNet Challenge 2021, the trained model classified clean lead I ECGs in one dataset with a test-fold area under receiver operating characteristic curve (AUC), sensitivity, and specificity of 0.915, 0.867 and 0.858 respectively. For signals with 0 dB signal-to-noise ratio from the same dataset, the respective performance metrics dropped slightly to 0.899, 0.862 and 0.818. External validation across three separate datasets showed a minimum AUC of 0.807. The data similarity metric outperformed an existing method in improving classification, particularly with limited target dataset samples, i.e. 50. CONCLUSION: The proposed Cardiac Arrhythmia Risk Evaluation from Lead-I ECG (CARE-I) framework enables accurate arrhythmia detection across diverse populations in real-world noisy environments, thus enhancing model generalisation and early diagnosis.

Despite significant efforts, AI powered automation in wireless networks remains hindered by several challenges, including lack of interpretability of traditional ML models, lack of resiliency against training data size and training data errors, and the resource-intensive process of determining suitable neural network architecture and hyper-parameters. To address these challenges, we pioneer a novel method by incorporating domain knowledge into the neural network design even before exposure to training data, akin to the innate intelligence in a newborn’s connectome design. We test our approach on both real data and simulated data and benchmark it against two common alternatives for system modeling, analytical modeling and conventional neural networks. Traditional neural networks perform well with abundant training data but struggle significantly when faced with realistically limited data, with the proposed domain-inspired approach achieving a 47.6% reduction in test MSE compared to traditional neural networks, while the pure analytical model-based approach performs the worst.

The evolution of wireless cellular networks to support Digital Twins (DTs) requires robust propagation models. Traditional propagation modeling methods, though fundamental, lack the realism, completeness, and computational efficiency required for effective DT synchronization. This inadequacy underscores the need for models that can seamlessly integrate with real-world network dynamics. Therefore, this work critically examines the resilience of conventional machine learning based models and highlights their vulnerabilities to data scarcity and the inherent dynamism of wireless networks. To address these challenges, we propose an innovative approach that leverages a multi-stage GAN for the generation and augmentation of tabular synthetic data, coupled with an Attention through Segmentation training strategy. This strategy is based on partitioning the data distribution based on the histogram of important features and replacing the single complex model with multiple simpler models focused on specific parts of the histogram. This dual approach significantly improves the resilience of the model, and our evaluations in realistic scenarios show an impressive recovery of more than $90 \%$ performance loss compared to traditional models and this improvement is achieved with a notable reduction in model complexity. Our research marks a significant advancement in the development of resilient and efficient propagation models for the next generation of wireless networks.

Data-driven Machine Learning (ML) based propagation models are essential for modern wireless network planning and optimization. However, their effectiveness is limited by scarse data conditions. Generative Adversarial Networks (GANs) often considered as a viable approach for data augmentation, struggle in these conditions because they also require large datasets for effective training. To address this challenge, we propose a novel approach that incorporates domain knowledge directly into GAN training. Using an analytical propagation equation based on 3GPP recommendations, we generate pseudo-random data to train a neural network, which then initializes the GAN generator network. This initialization improves the GAN's learning ability in extreme data scarcity. The framework enhances data generation quality by up to 52% and machine learning applicability by 60%, providing a robust solution to the scarse data problem in wireless network modeling with demonstrating the potential of integrating domain knowledge within ML methodologies.

In the rapidly evolving landscape of wireless networks, accurate and resilient propagation models are essential to achieve optimal performance and reliability. This paper presents a novel domain-aware framework for interpretable and resilient propagation models. The proposed approach represents an innovative architecture framework that is not only interpretable but can also deal with training data size scarcity. Bridges domain knowledge with machine learning. The proposed approach leverages a combination of domain expertise, analytical modeling, and customized neural networks to construct interpretable models that excel in both identical distribution and non-identical distribution test-train dataset scenarios. Through a comprehensive analysis, we demonstrate the proposed approach's ability to adapt and refine models in response to real-world variations, ensuring consistent, high-quality performance. The proposed framework not only enhances our understanding of complex systems but also paves the way for the creation of digital twins for wireless networks. Furthermore, the root mean square error of the performance metric for the proposed approach is reported as 6.97 dB, further confirming its effectiveness in accurately predicting the results of wireless propagation.

Existing works on user mobility in unmanned aerial vehicle (UAV)/drone-based aerial networks consider either fixed height drone base stations (DBSs), or are limited to two-tier networks only, and do not take into account tier association biasing, which is important for developing intelligent traffic offloading and load balancing schemes to support the diverse use cases enabled by emerging networks. This paper addresses these gaps by analyzing the impact of user mobility in a multi-tier UAV heterogeneous network at varying heights serving ground users and taking into account the biased average receive power association, where each tier has an independent cell range extension (CRE) factor. We evaluate the handover probability of a mobile user by first deriving the distance distributions to the serving DBS and the probability of user association with a DBS of a specific tier. The quantification and insights from our theoretical analysis, corroborated with numerical simulations reveal that the probability of handover (dependent on the CRE factor of the tiers, height of the DBSs and velocity) must be optimized in tandem with coverage probability for optimal network performance.

The fifth generation of cellular network (5G) can facilitate in-ambulance patient monitoring, diagnosis, and treatment by a remote specialist. However, 5G coverage and link quality can vary in time and location. The ambulance route selection can help meet the communication requirements of the in-ambulance applications. In this paper, we propose an innovative ambulance route selection framework which combines the communication requirements along with the network coverage and resources. The framework leverages the minimization of drive test (MDT) data to estimate the network coverage along the ambulance routes. To address the uneven distribution of location-based user-generated MDT data, we examine the performance and trustworthiness of several interpolation techniques to enrich the global MDT map for route selection. A simulated analysis shows that the proposed framework can dynamically adapt to varying application requirements as well as rapidly changing network conditions such as outages. Results also reveal that nearest neighbor and kriging interpolation techniques help complement the proposed framework by addressing the data sparsity problem.

Fault diagnosis is turning out to be an intense challenge due to the increasing complexity of the emerging cellular networks. The root-cause analysis of coverage-related network anomalies is traditionally carried out by human experts. However, due to the vast complexity and the increasing cell density of the emerging cellular networks, it is neither practical nor financially viable. To address this, many studies are proposing artificial intelligence (AI)-based solutions using minimization of drive test (MDT) reports. Nowadays, the focus of existing studies is either on diagnosing faults in a single base station (BS) only or diagnosing a single fault in multiple BS scenarios. Moreover, they do not take into account training data sparsity (varying user equipment (UE) densities). Inspired by the emergence of convolutional neural networks (CNN), in this paper, we propose a framework combining CNN and image inpainting techniques for root-cause analysis of multiple faults in multiple base stations in the network that is robust to the sparse MDT reports, BS locations and types of faults. The results demonstrate that the proposed solution outperforms several other machine learning models on highly sparse UE density training data, which makes it a robust and scalable solution for self-healing in a real cellular network.

Minimization of Drive Test (MDT) reports are a key enabler for Machine Learning (ML)-based zero-touch automation envisioned for emerging cellular networks. However, due to numerous factors, the MDT reports are spatially sparse in nature. This sparsity undermines the performance of ML models that are built on the MDT data to estimate and optimize network KPIs. In this paper, we present and evaluate a framework to address this challenge. We leverage generative models, specifically, Gener-ative Adversarial Networks (GAN) and Variational Autoencoders (VAE) to augment the sparse multi-dimensional MDT data. Unlike image data where the quality of synthetic images produced by the generative models can be evaluated visually, establishing the authenticity of tabular synthetic data is a more complex problem. We address this problem by leveraging a tripartite approach: 1) We use several statistical measures to quantify the resemblance of synthetic data with original data. 2) We compare the performance of an ensemble learning model trained on augmented data, with that of trained on original data only 3) We benchmark the performance of the generative models with several classical ML models. This analysis is carried out for varying levels of sparsity and reveals insights about robustness of generative models against training data sparsity as well as on suitability of various methods for evaluating the quality of the generated synthetic tabular data. Results show GAN performs considerably better compared to other approaches. The presented solution thus can be used to overcome the sparsity problem in MDT reports thereby enabling ML-based automation use cases.

Providing broadband connectivity to airborne systems using ground based cellular networks is a promising solution as it offers several advantages over satellite-based solutions. However, limited range of terrestrial base stations is a key challenge in full realization of this approach. This paper addresses this problem by proposing a mathematical framework for range extension leveraging large number of antennas at the base station. In contrast to prior works where range is not considered as a design parameter, we model the signal to noise ratio as a function of both number of antennas as well as the range in line-of-sight ground-to-air systems. This allows us to derive analytical expressions to determine the number of antennas required to increase range in different frequency bands and tracking and non tracking scenarios.

This paper presents novel results for Massive MIMO time division duplex (TDD) system where the beam- forming vector in the down-link makes use of MMSE channel estimates with quasi-orthogonal pilot sequences as opposed to orthogonal pilot sequences used in the prior art. The proposed method tackles the issue of pilot contamination by adapting the length of quasi-orthogonal pilot sequences according to the channel conditions or coherence time, leading to maximization of the total throughput. We compute the optimal length of the pilot sequences, for both orthogonal and quasi- orthogonal pilot sequences, that maximizes the achievable rates. We formulate the design criteria of the proposed scheme by using a classical theorem of estimation theory for multi- variate Gaussian distribution. Then we solve the criteria using heuristic algorithms and carry out simulations for different coherence times in a multi-cell system. It is shown that our proposed scheme requires 17\% less pilot resources as compared to schemes that employ orthogonal pilots. Furthermore, an interesting trade-off between the choice of pilots and the sum rate is observed for variable length of coherence interval.