One of the main events of SIC’26 is the poster session on the afternoon Thursday January 29th. The poster session is an excellent opportunity for workshop attendees to get to know each other and to discuss the research topics in more detail. We hope that this will result in a relaxed atmosphere for networking, informal discussions, and the exchange of research ideas.
Poster chair: Jana de Wiljes
CALL FOR POSTERS NOW OPEN!!!!
If you are willing to present a poster with recent research results, please register before January 19th. REGISTRATION IS NOW CLOSED
Poster Format
Posters are recommended to be prepared in A0 size and landscape format.
Poster List
N. Abedini (VU Amsterdam ) J. de Wiljes (TU Ilmenau) S. Dubinkina (VU Amsterdam )
Filtering with Randomised Observations: Sequential Learning of1 Relevant Subspace Properties and Accuracy Analysis
State estimation that combines observational data with mathematical models is central to many applications and is commonly addressed through filtering methods, such as ensemble Kalman filters. In this article, we examine the signal-tracking performance of a continuous ensemble Kalman filtering under fixed, randomised, and adaptively varying partial observations. Rigorous bounds are established for the expected signal-tracking error relative to the randomness of the observation operator. In addition, we propose a sequential learning scheme that adaptively determines the dimension of a state subspace sufficient to ensure bounded filtering error, by balancing observation complexity with estimation accuracy. Beyond error control, the adaptive scheme provides a systematic approach to identifying the appropriate size of the filter-relevant subspace of the underlying dynamics.
Darian Pérez-Adán, Dariel Pereira-Ruisánchez, Óscar Fresnedo, Ignacio Santamaria, Luis Castedo and John S. Thompson
Curriculum Self-Supervised Learning Framework for BD-RIS Optimization in MU-MIMO Systems
Beyond-diagonal reconfigurable intelligent surfaces (BD-RISs) have emerged as a promising technology for wireless communications, enabling full coupling among surface elements. However, existing optimization approaches either assume fixed power allocation or suffer from prohibitive computational complexity. We propose a curriculum self-supervised deep learning approach for joint BD-RIS and precoder optimization in multi-user (MU) multiple-input multiple-output (MIMO) uplink systems. In particular, we introduce a curriculum learning strategy that progressively transitions from isotropic power allocation to iterative waterfilling. Simulation results using realistic 3GPP channels show that our curriculum approach achieves competitive performance with manifold optimization baselines while maintaining significantly lower computational complexity and improved user fairness.
Dariel Pereira-Ruisánchez, Department of Computer Engineering, CITIC Research Center, University of A Coruña, A Coruña, Spain; Michael Joham, School of Computation, Information and Technology, Technical University of Munich, Germany; Óscar Fresnedo, Department of Computer Engineering, CITIC Research Center, University of A Coruña, A Coruña, Spain; Darian Pérez-Adán, Department of Computer Engineering, CITIC Research Center, University of A Coruña, A Coruña, Spain; Luis Castedo, Department of Computer Engineering, CITIC Research Center, University of A Coruña, A Coruña, Spain; Wolfgang Utschick, School of Computation, Information and Technology, Technical University of Munich, Germany
Pixel-Based CF-mMIMO: Addressing the AP Cooperation Cluster Formation in Fronthaul-Limited O-RAN Architectures
This paper investigates access point (AP) cooperation cluster formation in user-centric cell-free massive MIMO (CF-mMIMO) communication systems characterized by fronthaul links with capacity restrictions. Specifically, we consider an open radio access network (O-RAN) architecture that, although it favors the deployment of ultra-dense networks, is constrained in the number of APs that can be active simultaneously. In this context, we propose an innovative framework termed pixel-based CF-mMIMO, which enables efficient control of both the AP activation and cooperation cluster formation. Recognizing the parallels with pixel-based reconfigurable antennas, the proposed framework allows dynamic reconfiguration of the network coverage map with reasonably low computational cost. The high scalability and performance of the framework are mainly supported by a learning model based on graph neural networks (GNNs) that effectively exploits the existing graph-like structures in CF-mMIMO systems. Extensive simulation experiments demonstrate that the proposed approach achieves competitive spectral efficiency (SE) in challenging scenarios with dense AP deployments and numerous user equipments (UEs).
Jordi Borras (Universitat Politècnica de Catalunya) and Roberto López-Valcarce (atlanTTic Research Center, Universidade de Vigo)
Multibeam analog beamformer design for monostatic ISAC under Self-Interference
Multibeam technology constitutes a key enabler for integrated sensing and communications (ISAC). Previous designs for millimeter-wave monostatic ISAC often overlook the self-interference (SI) induced on the co-located radar receiver and/or relax the hardware-imposed constant-modulus (CM) constraints. We address the multibeam optimization problem in monostatic ISAC with CM analog arrays, explicitly accounting for SI. First, we relax the CM constraints and provide a semi-analytic solution that illustrates the impact of SI in the communication-sensing tradeoff. The design is then adapted to CM analog beamformers, substantially reducing the performance loss incurred if CM constraints are naively imposed.
Inés P. Mariño (Universidad Rey Juan Carlos), Harold Molina-Bulla (Universidad Carlos III), Joaquín Míguez (Universidad Carlos III)
Master-slave coupling scheme for synchronization and parameter estimation in the generalized Kuramoto-Sivashinsky equation
The problem of estimating the constant parameters of the Kuramoto-Sivashinsky (KS) equation from observed data has received attention from researchers in physics, applied mathematics, and statistics. This is motivated by the various physical applications of the equation and also because it often serves as a test model for the study of space-time pattern formation. Remarkably, most existing inference techniques rely on statistical tools, which are computationally very costly yet do not exploit the dynamical features of the system. We introduce a simple, online parameter estimation method that relies on the synchronization properties of the KS equation. In particular, we describe a master-slave setup where the slave model is driven by observations from the master system. The slave dynamics are data-driven and designed to continuously adapt the model parameters until identical synchronization with the master system is achieved. We provide a simple analysis that supports the proposed approach and also present and discuss the results of an extensive set of computer simulations. Our numerical study shows that the proposed method is computationally fast and also robust to initialization errors, observational noise, and variations in the spatial resolution of the numerical scheme used to integrate the KS equation.
Yannick Sztamfater-García (Universidad Carlos III de Madrid)
Multifidelity Monte Carlo for the estimation of re-entry time windows
In the absence of robust and standardized end-of-life disposal strategies, many spacecraft undergo uncontrolled atmospheric re-entry, resulting in significant uncertainty in the prediction of decay times. Existing re-entry prediction techniques commonly rely on orbital propagation models that assume Gaussian uncertainty. While computationally convenient, this assumption often fails to capture the strongly non-Gaussian evolution of the state probability density, particularly when nonlinear orbital dynamics interact with stochastic and highly uncertain atmospheric drag. Consequently, obtaining reliable estimates of re-entry time windows remains a challenging problem.
Crude Monte Carlo simulations can, in principle, provide accurate characterizations of the decay-time distribution, but their application with high-fidelity dynamical models is typically computationally prohibitive. To address this limitation, this work proposes a multifidelity Monte Carlo framework for efficient re-entry time estimation. By strategically allocating computational effort between low- and high-fidelity dynamical models, the proposed approach significantly reduces computational cost while preserving estimation accuracy. The framework enables the reliable computation of decay-time intervals and provides a practical tool for uncertainty-aware re-entry prediction under realistic modeling constraints.
Saúl Villaescusa, Eric Meneses-Albalá, Carlos Castorena, José Manuel Badia, Germán León, Carmen Botella-Mascarell, Sandra Roger
Robust and Energy-Efficient AoA/AoD Estimation in mmWave MIMO: An Adaptive Edge-AI Approach
The deployment of 6G requires signal processing solutions that are both adaptive and energy-efficient. While U-Net architectures provide accurate AoA/AoD estimation, their performance often degrades due to the mismatch between ideal simulations and real hardware behavior. This work presents an on-device adaptation strategy to mitigate these effects. By fine-tuning the model with
impairment-augmented data, our proposal maintains system robustness against hardware non-idealities over time. We validate this approach on an NVIDIA Jetson Orin Nano device, using a co-execution framework that decouples real-time inference on the CPU from the intensive learning process on the GPU. This architecture ensures continuous operation during model updates. A key contribution of this research is a quantitative energy consumption analysis: results show that while training time is reduced by over 11x operating at maximum CPU/GPU frequencies, the power cost increases fivefold. Our study identifies the optimal configuration for sustainable Edge AI, achieving high estimation accuracy with a minimized energy footprint, essential for future autonomous and green 6G infrastructures.
Mario Refoyo, David Luengo (Universidad Politécnica de Madrid)
FastPACE: Fast PlAnning of Counterfactual Explanations for Time Series Classification
Counterfactual explanations (CFEs) have emerged as a key tool in eXplainable Artificial Intelligence (XAI) for interpreting complex machine learning and deep learning models. However, most CFE methods neglect the high computational cost of generating explanations. This limitation can be particularly severe for high-dimensional data such as time series. Moreover, many time-series CFE approaches treat validity, the requirement that the counterfactual actually changes the model prediction, as an objective to be optimized rather than as a strict constraint, often leading to invalid explanations and limiting practical applicability. In this work, we propose FastPACE, an efficient method tailored to the generation of CFEs for time-series classification that guarantees valid counterfactuals by design, substantially reducing the runtime of current state-of-the-art methods without compromising the quality of the explanations. Extensive experiments on datasets from the UCR and UEA repositories show that FastPACE matches, and in several cases improves, the explanation quality of existing approaches while being significantly faster.
Clara Benlloch-Coscollà, David García-Roger, Carmen Botella-Mascarell, Sandra Roger, Departament d’Informàtica, Universitat de València
Wireless Power Transfer in Indoor Environments with Multiantenna Communications
Wireless Power Transfer (WPT) and radio-frequency energy harvesting are promising technologies for ensuring the autonomy of 6G-enabled IoT ecosystems. This work investigates the integration of WPT into communication systems, focusing on the impact of transmitter spatial configurations on harvesting efficiency. We evaluate the system performance in an indoor environment following standardized 3GPP propagation models including path loss and shadowing effects.
The study compares several multiantenna geometries including grid, cross, linear, circular, and hexagonal arrangements, to optimize the energy distribution toward the energy-harvesting receivers. Our findings show that distributed antenna topologies provide a more balanced power coverage across the service area, significantly increasing the energy harvesting potential compared to centralized
solutions. This research is part of the AIGUA5G project, funded under the Transmisiones initiative, which targets the digital monitoring of the urban water cycle, where some of the sensors are located in hard-to-reach areas without access to the electrical grid. The proposed framework provides an initial baseline for the design of sustainable, battery-less wireless networks in industrial environments.
Mohammad Solki (University of Naples Federico II, Naples, Italy); Vincenzo Norman Vitale (University of Naples Federico II, Naples, Italy); Antonia Maria Tulino (University of Naples Federico II, Naples, Italy; New York University, New York, USA); Andreas F. Molisch (University of Southern California, Los Angeles, USA); Jaime Llorca (University of Trento, Trento, Italy; New York University, New York, USA; Centre Tecnològic de Telecomunicacions de Catalunya (CTTC), Castelldefels, Spain).
Sample-Efficient Urgency-Aware Routing with Graph Neural Network Deep Reinforcement Learning
Timely delivery of deadline-sensitive traffic over dynamic and heterogeneous networks is essential for applications such as industrial automation, autonomous systems, and extended reality. We present a multi-agent solution to the Delay-Constrained Maximum-Throughput (DCMT) problem that combines (i) a topology-aware routing agent built on a Generalized Graph Neural Network (GeneralConv) with (ii) Lowest Effective Lifetime First (LELF) schedulers. We evaluate the proposed approach against a Minimum Weight Path (MWP) baseline and learning-based counterparts using multi-layer perceptrons (MLPs) on a network topology under traffic loads ranging from 30% to 120% of the network’s max-flow region. The GeneralConv router significantly improves performance over MWP and matches the best MLP-based results while requiring about 6× fewer training episodes, enabling faster attainment of high reliability under deadline constraints. Finally, we report an ablation study spanning nine design dimensions to identify which architectural and message-passing choices most effectively propagate urgency information and support robust deadline-aware routing decisions.
Fabián González and Joaquin Miguez (Universidad Carlos III de Madrid), Deniz Akyildiz and Dan Crisan (Imperial College London)
Bayesian filters with nudging for misspecified state space models
Nudging is a popular algorithmic strategy in numerical filtering to deal with the problem of inference in high-dimensional dynamical systems. We demonstrate that general nudging techniques can also tackle another crucial statistical problem in filtering, namely the misspecification of the transition kernel. Specifically, we rely on the formulation of nudging as a general operation increasing the likelihood and prove analytically that, when applied carefully, nudging techniques implicitly define state-space models that have higher marginal likelihoods for a given (fixed) sequence of observations. This provides a theoretical justification of nudging techniques as data-informed algorithmic modifications of state-space models to obtain robust models under misspecified dynamics. To demonstrate the use of nudging, we provide numerical experiments on linear Gaussian state-space models and a stochastic Lorenz 63 model with misspecified dynamics and show that nudging offers a robust filtering strategy for these cases.
Yudong Li, Matilde Sánchez-Fernández, Antonia Maria Tulino
Goal-oriented multicast source encoding with partial information
Our work introduces a novel framework for goal-oriented multicast encoding in 6G networks, focusing on semantic source coding where users request functions (or tasks) over a shared library of files, rather than the files themselves. The model is built upon a multicast scenario where a single server delivers content to multiple users, each with local side information in the form of cached file segments and a specific demand structure describing the tasks to be applied over selected files. The encoding is driven not only by data availability and demand but also by task equivalence, which is rigorously defined in terms of functional equivalence over file outputs.
Giuseppe Cocco and Javier Rodríguez Fonollosa
On the Error Exponent Distribution of Code Ensembles over Classical-Quantum Channels
We show that the probability distribution of the error exponent in i.i.d. code ensembles over classical-quantum (CQ) channels with arbitrary output states accumulates above a threshold that is strictly larger than the CQ random coding exponent (RCE) at low rates, while coinciding with it at rates close to the mutual information of the channel. This, combined with recent results on the CQ RCE and the CQ sphere packing bound, implies that the ensemble distribution of error exponents concentrates around the CQ RCE in the high rate regime. Moreover, in the same rate regime the threshold we derive coincides with the ensemble-average of the exponent, that is, the typical random coding (TRC) exponent.
Álvaro Callejas Ramos (UC3M), Matilde Sánchez Fernández (UC3M), Antonia Tulino (Universitá Federico II di Napoli), Juan José Murillo Fuentes (Universidad de Sevilla)
Recovery of Real-Valued Sparse Models from Noisy Measurements in the Multi-Dimensional Spectral Domain
This work addresses the exact recovery of real-valued sparse models from noisy measurements in the multi-dimensional spectral domain. An atomic norm-based approach is proposed to enable gridless estimation of the model frequencies and atoms under partial sampling. Sufficient conditions for unique identifiability of real-valued signals are derived by exploiting their conjugate symmetry and multi-level Toeplitz structure. Recovery guarantees are established through theoretical results linking sparsity, dimensionality, and number of measurements. The effectiveness of the method is demonstrated through simulations and an application to spectral analysis of textile patterns in artistic images.
Vahid Vahidpour, Roberto López Valcarce, atlanTTic Research Center, Universidade de Vigo
OFDM Sidelobe Suppression: CSIT-aware Orthogonal Precoding
We present a channel-aware spectral precoding framework for multicarrier modulation minimizing weighted leakage in prescribed frequency bands via a semi-unitary precoder. Its right-unitary invariance is exploited to embed channel state information at the transmitter without altering the power spectral density (PSD). By right-unitary rotations through singular value and geometric mean decompositions, we respectively derive low-complexity zero-forcing and successive-interference cancellation based decoders. A closed-form PSD characterization enables rigorous out-of-band radiation analysis and comparison with former pre-equalized and spectrally precoded OFDM (PSP-OFDM) and null-subcarrier schemes. Simulations over block-fading multipath channels demonstrate substantially stronger sidelobe suppression and improved error rates at medium/high SNR. The results establish channel-aware spectral precoding as a principled, efficient solution to joint spectrum and reliability control.
Santiago Diaz, Adam Podhorski
ENVIRONMENTAL NOISE MAPPING WITH DRONE SWARMS IN URBAN AND INDUSTRIAL ENVIRONMENTS
Developing a distributed acoustic signal acquisition and processing system based on swarms of unmanned aerial vehicles (UAV’s) equipped with microphone arrays
for the decentralized estimation of the sound field in open environments: design of distributed algorithms for acoustic signal measurement and processing, capture discrete measurements of acoustic signals in space and time using sensors embedded in UAV’s, including UAV’s ego-noise and integration on-board sensors and computing in the UAV to read its flight data such as position, speed, rotor speed, and synchronize them with acoustic measurements.
Olatz Sanz, Reza Dastbsateh, Mikel Hernaez, Sara Capponi, Pedro Crespo, Josu Etxezarreta
Quantum Synthetic Data Generation for in silico clinical trials
Clinical trials are essential for biomedical progress but remain inefficient for rare and heterogeneous diseases such as myelodysplastic syndrome (MDS). In silico trials, based on computationally generated synthetic data, offer a promising alternative for improving statistical power and reproducibility. However, classical generative models often rely on rigid assumptions and require large datasets, limiting their ability to capture complex biological patterns in small and high-dimensional cohorts. In this work, we propose a quantum generative model based on a mapped Tensor Network (TN) with incomplete patient data. The model encodes joint probability distributions into quantum states and represents them as Matrix Product States (MPS), allowing efficient sampling and circuit implementation. Our results demonstrate that the quantum implementation achieves lower mean absolute error (MAE) and Jensen–Shannon divergence (JSD) compared to classical baselines such as CTGAN, CopulaGAN, and TVAE, particularly in low-sample regimes (n<300). This approach establishes a pathway toward practical quantum-enhanced data generation in computational medicine.
