This paper investigates a cell-free multiple-input multiple-output integrated sensing and communications (ISAC) system, focusing on channel estimation and coordinated beamforming while minimizing system overhead. To achieve this, we jointly optimize pilot sequences during channel estimation and coordinated beamforming during data transmission to enable more efficient target detection. Specifically, at the central processing unit, we employ a Transformer for channel estimation and a graph neural network (GNN) for coordinated beamforming. First, we propose a time-division duplexing protocol for channel estimation and coarse target detection. A supervised learning approach is adopted to approximate the optimal solution, leveraging a well-trained dataset. Second, we model the cell-free ISAC network topology as a heterogeneous graph, comprising different types of nodes and edges, and employ a GNN-based approach for coordinated beamforming. Both learning models are trained offline and deployed online. Simulation results demonstrate the effectiveness of the proposed schemes in terms of channel estimation accuracy, target detection performance, and quality of communication signals.

This paper proposes a coordinated beamforming scheme for a multi-cell integrated sensing and communication (ISAC) system. A target-centric graph is constructed, with the coordinate system centered at the detection target. Specifically, base stations (BSs) and users are represented as nodes, with their coordinates serving as node features, while channel realizations between nodes are modeled as edge features. To evaluate sensing performance, the Neyman-Pearson detector is employed to compute the detection probability for a fixed false alarm probability. The optimization problem is formulated to maximize the detection probability of the target at the origin while ensuring QoS communication requirements and satisfying the transmit power budget. This sensing-centric problem is addressed using graph neural networks (GNNs), which generate parameterized policies for coordinated beamforming. The GNNs are trained via primal-dual approach, leveraging a small duality gap for efficient convergence. Additionally, specific layers are utilized to generate user association policies for communication links, enabling efficient processing of the graph-structured data after sparsity enhancement. Simulation results validate the feasibility and effectiveness of the proposed GNN-based approach in achieving high detection probability.

Recently, over-the-air computation (AirComp) leverages the superposition property of wireless channels to enable efficient function computation over a multiple access channel (MAC). However, existing digital AirComp methods either rely on single-symbol modulation, which limits flexibility and robustness, or on multi-symbol extensions that suffer from high complexity or approximation errors. To overcome these limitations, we propose a new multi-symbol modulation framework, termed sequential modulation for AirComp (SeMAC), which encodes each input into a sequence of symbols with distinct constellation diagrams across multiple time slots. This approach increases design flexibility and robustness against channel noise. Specifically, the modulation design is formulated as a non-convex optimization problem and efficiently solved through a successive convex approximation (SCA) combined with stochastic subgradient descent (SSD). For fixed modulation formats, we further develop SeMAC with power adaptation (SeMAC-PA) to adjusts transmit power and phase while preserving the modulation structure. Notably, numerical results show that SeMAC improves computation accuracy by up to 14 dB compared to the existing methods for computing nonlinear functions such as the product function.

We investigate a novel characteristic of the conjugate function associated to a generic convex optimization problem, which can subsequently be leveraged for efficient dual decomposition methods. In particular, under mild assumptions, we show that there is a specific region in the domain of the conjugate function such that for any point in the region, there is always a ray originating from that point along which the gradients of the conjugate remain constant. We refer to this characteristic as a fixed gradient over rays (FGOR). We further show that this characteristic is inherited by the corresponding dual function. Then we provide a thorough exposition of the application of the FGOR characteristic to dual subgradient methods. More importantly, we leverage FGOR to devise a simple stepsize rule that can be prepended with state-of-the-art stepsize methods enabling them to be more efficient. Furthermore, we investigate how the FGOR characteristic is used when solving the global consensus problem, a prevalent formulation in diverse application domains. We show that FGOR can be exploited not only to expedite the convergence of the dual decomposition methods but also to reduce the communication overhead. FGOR is extended to nonconvex formulations, and its advantages in stochastic optimization are demonstrated. Numerical experiments using quadratic objectives and a regularized least squares regression with real datasets are conducted. The results show that FGOR can significantly improve the performance of existing stepsize methods and outperform the state-of-the-art splitting methods on average in terms of both convergence behavior and communication efficiency.

This work proposes a reliable leakage detection analysis for water distribution networks (WDNs) by combining efficient and emergent machine learning techniques. In this study case, we analyze pressure and flow measurements from pumps in Stockholm, Sweden, where we consider a residential district metered area of the WDN. Our solution aims at detecting leakage in WDNs using a prototype-based model (PBM) while preserving data privacy by proposing a federated learning approach. The machine learning strategies we adopt have low complexity, and the numerical experiments show the potential of using federated prototype-based techniques for leakage detection on monitored WDNs. Specifically, our experiments show that the proposed learning method can obtain higher detection rates at each pumping station than the conventional centralized approach, e.g. improvements of purity rates up to 7.6% in one of the pumping stations, which increased the minimum values from 92.13%, obtained through centralized learning, to 99.11%, obtained via federated learning.

The wireless communication systems of today rely to a large extent on the condition of the accessible channel state information (CSI) at the transmitter and receiver. Channel aging, denoting the temporal and spatial evolution of wireless communication channels, is influenced by obstructions, interference, traffic load, and user mobility. Accurate CSI estimation and prediction empower the network to proactively counteract performance degradation resulting from channel dynamics, such as channel aging, by employing network management strategies such as power allocation. Prior studies have introduced approaches aimed at preserving high-quality CSI such as temporal prediction schemes, particularly in scenarios involving high mobility and channel aging. Conventional model-based estimators and predictors have historically been considered state-of-the-art. Recently, the development of artificial intelligence (AI) has increased the interest in developing models based on AI. Previous works have shown high potential of AI-aided channel estimation and prediction, which inclines the state-of-the-art title from model-based methods to be confiscated. However, there are many aspects to consider in channel estimation and prediction employed by AI in terms of prediction quality, training complexity, and practical feasibility. To investigate these aspects, this chapter provides an overview of state-of-the-art neural networks, applicable to channel estimation and channel prediction. The principal neural networks from the overview of channel prediction are empirically compared in terms of prediction quality. An innovative comparative analysis is conducted for five prospective neural networks characterized by distinct prediction horizons. The widely acknowledged tapped delay line (TDL) channel model, as endorsed by the Third Generation Partnership Project (3GPP), is employed to ensure a standardized evaluation of the neural networks. This comparative assessment enables a comprehensive examination of the merits and demerits inherent in each neural network. Subsequent to this analysis, insights are offered to provide guidelines for the selection of the most appropriate neural network in channel prediction applications.

To meet the growing demands for connectivity and reliability in cellular networks, it is essential to ensure reliable quality of service (QoS) guarantees for end users. The integration of predictive QoS (pQoS) in cellular networks enables proactive fulfillment of QoS requirements for a diverse range of applications, including intelligent transportation systems. This study presents a pQoS framework in cellular networks, particularly for connected vehicles, that divides the road into segments, clusters them, and assigns a pQoS model to each cluster. By implementing this framework, we mitigate the concept drift of the pQoS model induced by variations in the propagation environment and interference. Each predictive cluster model is locally trained on vehicles traveling within the cluster boundaries using federated learning. A significant challenge is balancing the trade-off between the number of clusters, prediction accuracy, and communication overhead for updating local models. This trade-off suggests the novel problem of performing a joint optimization of the training and number of clusters. To address such difficult optimization, we propose an iterative approximate solution using proximal alternative minimization for which we provide convergence guarantees. Ultimately, by evaluations with real-world data, our numerical findings reveal that our proposed clustered predictive model reduces the mean absolute percentage error by 8%, and the mean absolute error by 7%, compared to conventional predictive approaches proposed by prior studies.

This paper investigates over-the-air computation (AirComp) over multiple-access time-varying channels, where devices with high mobility transmit their sensing data to a fusion center (FC) for averaging. To combat the Doppler shift induced by time-varying channels, each device adopts orthogonal time frequency space (OTFS) modulation. Our objective is minimizing the mean squared error (MSE) for the target function estimation. Due to the multipath time-varying channels, the OTFS-based AirComp not only suffers from noise but also interference. Specifically, we propose three schemes, namely S1, S2, and S3, for the target function estimation. S1 directly estimates the target function under the impacts of noise and interference. S2 mitigates the interference by introducing a zero padding-assisted OTFS. In S3, we propose an iterative algorithm to estimate the function in a matrix form. In the numerical results, we evaluate the performance of S1, S2, and S3 from the perspectives of MSE and computational complexity, and compare them with benchmarks. Specifically, compared to benchmarks, S3 outperforms them with a significantly lower MSE but incurs a higher computational complexity. In contrast, S2 demonstrates a reduction in both MSE and computational complexity. Lastly, S1 shows superior error performance at small SNR and reduced computational complexity.

We consider the problem of non-coherent over-the-air computation (AirComp), where n devices carry highdimensional data vectors x_i ∈ R^d of sparsity ||x_i||₀ ≤ k and the sum of these data vectors has to be computed at a receiver. Previous results on non-coherent AirComp require more than d channel uses to compute functions of x_i, where the extra redundancy is used to combat non-coherent signal aggregation. However, if the data vectors are sparse, sparsity can be exploited to offer significantly cheaper communication. In this paper, we propose to use random transforms to transmit lower-dimensional projections s_i ∈ R^T of the data vectors. These projected vectors are communicated to the receiver using a majority vote (MV)AirComp scheme, which estimates the bit-vector corresponding to the signs of the aggregated projections, i.e., y=sign (Σ_i s_i). By leveraging 1-bit compressed sensing (1bCS) at the receiver, the real-valued and high-dimensional aggregate Σ_i x_i can be recovered from y. We prove analytically that the proposed MVCS scheme estimates the aggregate data vector Σ_ix_i with ℓ₂-norm error ϵ in T=O (k n log (d) / ϵ²) channel uses. We consider distributed histogram estimation, a canonical building block for federated analytics, as an aplication for MVCS where the data vectors x_i are inherently 1 -sparse. Our numerical evaluations demonstrate that our scheme achieves the same order of communication cost as state-of-the-art methods while avoiding the complexity and overhead of additional cryptographic tools.