One of the main events of SIC’25 is the poster session on the afternoon Tuesday 01/7. 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.
CALL FOR POSTERS NOW OPEN!!!!
If you are willing to present a poster with recent research results, please register before the 23rd of June – REGISTRATION IS NOW CLOSED
Poster Format
Posters are recommended to be prepared in A0 size and landscape format.
Poster List
Alejandro Almodóvar, Universidad Politécnica de Madrid (UPM), Adrián Javaloy, University of Edinburgh, UK, Juan Parras, UPM, Santiago Zazo, UPM, Isabel Valera, Saarland University, Germany and Max Planck Institute for Software Systems, Germany
DeCaFlow: A deconfounding causal generative modelWe present DeCaFlow, a new Deep Learning method that can learn from purely observational data (i.e., data collected without conducting experiments) and a known causal diagram to answer questions about cause and effect—even when there are hidden factors influencing the system. In practice, researchers often cannot measure every factor that affects their variables of interest. DeCaFlow overcomes this by using available “proxy” measurements (indirect indicators of the unmeasured influences) to correct for bias and recover true causal relationships. Once DeCaFlow is trained on a dataset and its causal structure, it can accurately predict how intervening on one variable (for example, changing a treatment or policy) will affect others. Furthermore, it can answer “what if” questions (counterfactuals), such as “What would have happened if we had made a different decision?” whenever it can answer the corresponding intervention question. In experiments on several benchmarks—including a complex biological dataset with multiple hidden factors—DeCaFlow consistently outperforms existing methods, demonstrating that it can be applied directly to any problem where the causal diagram is known.
Patricia A. Apellániz, Juan Parras, Santiago Zazo, Universidad Politécnica de Madrid
An Improved Tabular Data Generator with VAE-GMM Integration
The increasing use of machine learning in various fields necessitates robust methods for creating synthetic tabular data that preserve key characteristics while mitigating data scarcity challenges. State-of-the-art approaches, such as CTGAN and TVAE, encounter challenges with the intricate structures inherent in tabular data, which often comprise both continuous and discrete features with non-Gaussian distributions. To address these limitations, we propose a novel approach based on Variational Autoencoders (VAEs) enhanced with a Bayesian Gaussian Mixture (BGM) model. Unlike other methods that alter the Gaussian prior of the VAE, our approach trains the VAE conventionally and then applies the BGM model to the learned latent space. This allows for a more accurate representation of the underlying data distribution during data generation. Moreover, our model offers enhanced flexibility by accommodating various differentiable distributions for individual features, enabling the handling of continuous and discrete data types. Thorough validation on three real-world datasets, including medically relevant ones, demonstrates significant outperformance compared to CTGAN and TVAE. Our model demonstrates promise as a valuable tool for generating synthetic tabular data across diverse domains, particularly in healthcare.
Jesús Gutiérrez-Gutiérrez, Xabier Insausti, Íñigo Barasoain-Echepare, Marta Zárraga-Rodríguez, University of Navarra, Tecnun School of Engineering
Gradient-Based Recursive Optimization of Residual Stream
We present an algorithm for computing the gradient of residual stream flow resulting from agri-food industries, with applications in optimizing the benefit obtained by the different by-products available through the corresponding value chain. By applying a change of variables, we reformulate the original constrained maximization problem as an unconstrained one. This transformation allows us to compute the partial derivatives of the objective function more efficiently, resulting in a recursive formulation for the gradient.
Álvaro Callejas Ramos, Universidad Carlos III de Madrid (UC3M), Matilde Sánchez Fernández, UC3M, Antonia Tulino, Universita di Napoli Federico II, Italy
Pilot Optimization for Sparse Multidimensional Parametrical Channel
In this work we focus on the optimization of the pilot matrix in the context of the multidimensional parametric channel estimation in the millimeter wave (mmWave) band problem. Due to the sparse structure of this band, a gridless multidimensional spectral estimation formulation is proposed, which is solved by minimizing the atomic norm. The main objective is to design pilot sequences that maximize the recoverability of the channel and the frequencies containing the parameters of interest with the minimum possible length, guaranteeing uniqueness conditions in the estimation. For this purpose, a greedy algorithm is introduced to explore all possible sets of pilots from a discrete alphabet, optimizing its range and reducing mutual coherence. The results of the simulations show that this design, based on maximizing the k-rank of the pilot matrix, significantly improves the channel recovery performance and accurate frequency estimation.
Diego Cuevas, Mikel Gutiérrez, Jesús Ibáñez, Ignacio Santamaria, Universidad de Cantabria
Low Probability of Detection Communication Using NoncoherentThis paper proposes a noncoherent low probability of detection (LPD) communication system based on direct sequence spread spectrum (DSSS) and Grassmannian signaling. Grassmannian constellations enhance covertness because they tend to follow a noise-like distribution. Simulations showed that Grassmannian signaling provides competitive bit error rates (BER) at low signal-to-noise ratio (SNR) regimes with low probability of detection at the unintended receiver compared to coherent schemes that use QPSK or QAM modulation formats and need pilots to perform channel estimation. The results suggest the practicality and security benefits of noncoherent Grassmannian signaling for LPD communications due to their improved covertness and performance.
Elena Berardini, Reza Dastbasteh, Josu Etxezarreta Martinez, Shreyas Jain, and Olatz Sanz Larrarte, University of Bordeaux, Tecnun – University of Navarra, Indian Institute of Science Education and Research
Asymptotically good CSS-T codes and a new construction of triorthogonal codes
We proposed a new systematic construction of CSS-T codes, a family of quantum codes with a desirable symmetry toward the T-gate. We used this construction to prove the existence of asymptotically good binary CSS-T codes, resolving a previously open problem in the literature. We also discussed the application of these codes in dealing with coherent noise. Finally, we developed a new construction of triorthogonal codes, which broadens the range of codes available for magic state distillation—an extremely costly step in realizing universal quantum computation.
Tomás Domínguez-Bolaño, Valentín Barral, Carlos J. Escudero, and José A. García-Naya; CITIC Research Center, University of A Coruña, Spain
Large-Scale Gantry Robot System for High-Repeatability Measurements and Ground Truth Dataset Generation
This work presents a large-scale 3-axis gantry robot system designed for high-precision positioning across a 6m × 6m × 2m working envelope. The system enables high-repeatability measurements and ground truth dataset generation for indoor localization technologies, wireless channel measurements, and other research requiring precise spatial reference. The robot uses igus self-lubricating linear units for maintenance-free operation, 2 brushless motors (X-axis), 1 brushless motor (Y-axis), and a stepper motor (Z-axis), all with encoders. LinuxCNC serves as the control platform, coordinating the 3 axes through MESA Electronics 7I96S and 7I77 interface cards that communicate with igus dryve D1 motor controllers, implementing closed-loop control for precise real-time operation.To calibrate and validate the robot’s positioning, an OptiTrack motion capture system installed within the same structure as the robot has been used to measure positioning errors within the robot’s working volume. This revealed small errors due to minor structural misalignments. To compensate for these small errors a custom LinuxCNC kinematics module was developed, enabling the gantry robot positioning to achieve sub-centimeter precision.
Santiago Díaz, Adam Podhorski, Tecnun, University of Navarra, Spain.
Environmental noise mapping with drone swarms in urban and industrial environments
Currently, sound measurement techniques are limited by systems that are inflexible, expensive, and difficult to deploy in real-world environments. Traditional acoustic cameras require rigid infrastructure, such as tripods, cranes, or fixed mounts, making them difficult to use in large areas, dynamic environments, or at high altitudes. The objective of the Activity A6.2 consists in developing a distributed acoustic signal acquisition and processing system based on swarms of unmanned aerial vehicles (UAVs) equipped with microphone arrays for the decentralized estimation of the sound field in open environments.
Josu Etxezarreta Martinez(1,2), Paul Schnabl(1,3), Javier Oliva del Moral(1,4), Reza Dastbasteh (1), Pedro M Crespo (1), Ruben M Otxoa (5)
(1) Department of Basic Sciences, Tecnun – University of Navarra, 20018 San Sebastian, Spain.
(2) Cavendish Laboratory, Department of Physics, University of Cambridge, Cambridge CB3 0HE, UK.
(3) Institute for Theoretical Physics, University of Innsbruck, A-6020 Innsbruck, Austria.
(4) Donostia International Physics Center, 20018 San Sebastian, Spain.
(5) Hitachi Cambridge Laboratory, J. J. Thomson Avenue, Cambridge, CB3 0HE, United Kingdom.
Leveraging biased noise for more efficient quantum error correction at the circuit-level with two-level qubitsTailoring quantum error correction codes (QECC) to biased noise has demonstrated significant benefits. However, most of the prior research on this topic has focused on code capacity noise models. Furthermore, a no-go theorem prevents the construction of CNOT gates for two-level qubits in a bias preserving manner which may, in principle, imply that noise bias cannot be leveraged in such systems. In this work, we show that a residual bias up to η∼5 can be maintained in CNOT gates under certain conditions. Moreover, we employ controlled-phase (CZ) gates in syndrome extraction circuits and show how to natively implement these in a bias-preserving manner for a broad class of qubit platforms. This motivates the introduction of what we call a hybrid biased-depolarizing (HBD) circuit-level noise model which captures these features. We numerically study the performance of the XZZX surface code and observe that bias-preserving CZ gates are critical for leveraging biased noise. Accounting for the residual bias present in the CNOT gates, we observe an increase in the code threshold up to a 1.27% physical error rate, representing a 90% improvement. Additionally, we find that the required qubit footprint can be reduced by up to a 75% at relevant physical error rates.
O. Fabián González, Víctor Elvira, Joaquin Miguez, Universidad Carlos III de Madrid
On the role of dimension in nested importance samplers
Many Bayesian problems have high-dimensional models where only some variables are estimation targets; the rest are nuisance parameters we ideally want to integrate out. When analytical integration is impossible, computational methods like nested importance sampling (including SMC2, IS2, etc.) use numerical integration. This paper explores how the nuisance parameter’s dimension affects the error bounds of these methods. Specifically, we show that under certain assumptions, approximation error bounds can remain uniform even as the nuisance parameter’s dimension grows indefinitely.
Juan F. Martín, Javier R. Fonollosa, Giuseppe Cocco, Universitat Politècnica de Catalunya (UPC)
Quantum Error Mitigation with Tensor NetworksThe realization of fault-tolerant quantum computing is still far on the horizon, forcing state-of-the-art quantum hardware to rely heavily on noise mitigation techniques. While most existing methods focus on post-processing strategies, we explore an alternative pre-processing approach: addressing the noise problem before measurements are taken on the quantum device. In this work, we revisit a fundamental question: can we find an observable Y such that its expectation value on a noisy quantum state E(ρ) matches the expectation value of a target observable X on the noiseless quantum state ρ? Inspired by the introduction of Tensor Error Mitigation (TEM), we address this problem by leveraging Tensor Networks techniques, providing an affirmative answer in practical scenarios. Our results demonstrate that our proposal performs effectively at slightly greater circuit depths compared to TEM, with a classical computational complexity ∼ 10^6 times more efficient. Notably, our approach eliminates the need for additional quantum operations, such as the set of informationally complete positive operator-valued measurement (IC-POVM) implemented in TEM, so there is no a priori assumption on the quantum state, and no tomographic strategy is required. Instead, our method requires only a single noisy quantum circuit execution.
Andrea Guamo-Morocho, University of Vigo, Jordi Borras, Technical University of Catalonia, Nuria Gonzalez-Prelcic, University of Carlifornia, San Diego, Roberto López-Valcarce, University of Vigo.
Full-Duplex Beamforming Design for mmWave Multi-Target ISAC SystemsThis paper introduces a novel full-duplex (FD) integrated sensing and communication (ISAC) system operating in the millimeter-wave (mmWave) band. The proposed system adopts a fully connected hybrid beamforming architecture and employs orthogonal frequency-division multiplexing (OFDM) to efficiently support both downlink communication and monostatic sensing. While prior FD ISAC studies often assume a narrowband SI channel, this work addresses the more practical scenario of a frequency-dependent SI channel. The design problem aims to maximize system performance while ensuring radar gain constraints across multiple target directions, mitigating SI, and satisfying unit-modulus constraints. To address the complexity of this non-convex problem, we first develop fully digital beamformers and then derive their hybrid approximations. Although conventional Euclidean distance minimization is effective for the UE combiner, we develop an alternative hybrid precoder optimization strategy that enhances robustness against SI. Additionally, the BS combiner is refined to enhance the reception of target reflections. We employ a joint angle-range-Doppler estimation technique that integrates multiple signal classification (MUSIC) for angle estimation and the periodogram method for estimating range and velocity. Simulation results demonstrate the effectiveness of our proposed framework in achieving reliable communication and accurate multi-target sensing.
Jesús Ibáñez, Diego Cuevas, Ignacio Santamaría, Universidad de Cantabria, Jesús Gutiérrez, IHP – Leibniz-Institut für Innovative Mikroelektronik, Germany.
Reciprocity calibration OTA experiments in multicarrier cell-free massive MIMOCF-mMIMO is an emerging technology that promises to increase the capacity and robustness of wireless mobile communication, while achieving more uniform coverage. However, the non-reciprocity of the transceivers’ hardware transmission and reception chains prevents the use of the channel reciprocity principle in time division duplex schemes. In this work, it is verified experimentally that the phase difference between the transmitting and receiving chains of commercial transceivers has a linear relationship with frequency, which allows the efficient extension of two popular reciprocity calibration algorithms, Argos, and the recently proposed BeamSync to the multicarrier case. Moreover, an over-the-air (OTA) measurement campaign has been carried out using a USRP-based test-bed composed of 3 access points (AP) with two antennas each and a single-antenna user equipment (UE). In the experiments, up-link and down-link 5G-NR OFDM signals are transmitted, allowing the comparison of down-link SNR performance when the UE is served by all APs without reciprocity calibration or calibrated using the multicarrier versions of Argos or BeamSync. Results show the practical feasibility of applying the proposed multicarrier reciprocity calibration techniques and the enormous gain they provide in terms of capacity and diversity.
Yudong Li, Universidad Carlos III de Madrid (UC3M), Matilde Sánchez Fernández, UC3M, Antonia Tulino, Universita di Napoli Federico II, Italy
Goal-oriented multicast source encoding with partial information
This study 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.
Roberto Maneiro-Catoira, Julio Brégains, José A. García-Naya, and Luis Castedo, CITIC Research Center & Department of Computer Engineering, Universidade da Coruña
Multibeam TMAs for Space-Based Applications
We propose a model of satellite-borne TMA (Time-Modulated Array) capable of transmitting four directional beams with fully independent control. We evaluate the SWaP-C (Size Weight Power and Cost) concept and compare it with an equivalent conventional VPS(Variable Phase Shifter)-based multibeam phased array, showing the advantages of our TMA. Additionally, other value-added advantages of the proposed TMA for space-based applications are an excellent angular resolution as well as the inherent ability to compensate the Doppler effect.
Aniol Martí, Universitat Politècnica de Catalunya (UPC), Luca Sanguinetti, UniPi, Jaume Riba, UPC, Meritxell Lamarca, UPC
Coherent and Noncoherent Detection in Dense Arrays: Can We Ignore Mutual Coupling?This paper investigates the impact of mutual coupling on MIMO systems with densely deployed antennas. Leveraging multiport communication theory, we analyze both coherent and noncoherent detection approaches in a single-user uplink scenario where the receiver ignores mutual coupling effects. Simulation results indicate that while coherent detection is generally more accurate, it is highly sensitive to mismatches in the coupling model, leading to severe performance degradation when antennas are closely spaced, to the point of becoming unusable. Noncoherent detection, on the other hand, exhibits a higher error probability but is more robust to coupling model mismatches.
Eric Meneses, Saúl Villaescusa, Sandra Roger, Carmen Botella, Santiago Felici and Enrique Navarro. Computer Science Department, Universitat de València
Estimation of AoA and AoD in millimeter-wave MIMO channels with convolutional neural networks
Previous research has shown that estimating the Angle of Arrival (AoA) and Angle of Departure (AoD) of wireless signals is similar to finding the frequencies of sinusoidal signals observed during a pilot-based training phase. In this work, we introduce two adapted convolutional neural networks (CNNs) to solve the AoA/AoD estimation problem: a modified Residual Network (ResNet) and a U-Net. We treat this task as an image-to-image translation problem, where the input is a signal representation and the output is a map of estimated angles. To improve the accuracy of the estimates, we include a final processing step that uses blob detection techniques to identify the main peaks in the output. We conduct simulations to compare our neural network-based methods with traditional signal processing techniques and the theoretical Cramér-Rao lower bound. The results demonstrate that our deep learning approaches achieve significantly better detection performance compared to traditional signal processing methods
Lucía Pallarés-Rodríguez, Gonzalo Seco-Granados, José A. López-Salcedo, Institut d’Estudis Espacials de Catalunya (IEEC) / Signal Processing for Communications and Navigation (SPCOMNAV) research group, Universitat Autònoma de Barcelona
A Novel Blind Multipath Mitigation Approach for GNSS Multi-AntennaGlobal Navigation Satellite Systems (GNSSs) face a difficult challenge when combating multipath reflections, since, due to their high correlation with the signal of interest, they pose a significant threat for GNSS receivers. The use of array processing techniques greatly helps receivers to combat these effects, typically exploiting available information for an effective mitigation. However, relying on external and possibly out-dated references might cause the receiver to perform poorly, and thus an approach that does not require any previously obtained parameters can significantly enhance the behavior of the receiver. This poster presents a blind alternative to the reference-dependent existing techniques for multipath mitigation in GNSS receivers. Combining two well-known algorithms, SAGE and LMS, this approach allows to obtain large attenuation capacities while ensuring resilience and robustness against rapid changes in the operating environment.
Jesús Pérez, Ignacio Santamaría, Universidad de Cantabria, Alba Pagés, UPC
Blind learning of the optimal fusion rule in distributed detection systemsThis poster presents an algorithm for the fusion center to blindly estimate the sensor performance parameters in distributed detection systems over centralized wireless sensor networks (DD-WSN). The algorithm covers a wide variety of situations that may arise in DD-WSN. For example, it is applicable when the fusion center knows in advance some of the parameters of some sensors or when only a subset of sensors reports their decisions in each sensing period. Based on those estimates, decision fusion rules are derived under the minimum probability of error criterion. Simulation results show that, after sufficient sensing the sensor parameter estimates are accurate enough for the fusion rule to exhibit near-optimal detection performance.
Miguel Rivas-Costa and Carlos Mosquera, atlanTTic Research Center, Universidade de Vigo
A Size-Efficient DFT Codebook Approach for DoA Estimation with Hybrid ArraysMost existing direction-of-arrival (DoA) estimation techniques are tailored for fully digital (FD) antenna arrays. However, as modern MIMO systems scale to hundreds or thousands of antennas, FD architectures become increasingly impractical due to the prohibitive number of analog-to-digital converters (ADCs) required. Hybrid Analog-Digital (HAD) architectures offer a more hardware-efficient alternative, but conventional DoA estimation methods designed for FD arrays are not directly applicable in this setting. To overcome this challenge, we propose a novel method to reconstruct the covariance matrix of an equivalent FD array using measurements from a HAD architecture. This enables the application of classical subspace-based DoA estimation techniques with minimal adaptation. This work proposes a DFT-based codebook design which exploits the structure of Cauchy-like matrices for fast DoA estimation, making it particularly attractive for applications such as satellite-based or rapidly reconfigurable systems.
Olatz Sanz (1), Mikel Hernaez (2,3,4,5), Reza Dastbasteh (1), Pedro Crespo (1), Sara Capponi (6,7) and Josu Etxezarreta (1).
(1) Tecnun, University of Navarra
(2) CIMA University of Navarra, Cancer Center Clinica Universidad de Navarra (CCUN), Pamplona, Spain
(3) Navarra Institute for Health Research (IdiSNA), Navarra, Spain
(4) Instituto de Ciencia de los Datos e Inteligencia Artificial (DATAI), University of Navarra, Navarra, Spain)
(5) Centro de Investigacion Biomedica en Red de Cancer (CIBERONC), Madrid, Spain
(6) Functional Genomics and Cellular Engineering, AI and Cognitive Software, IBM Almaden Research Center, San Jose, CA, 95120, USA
(7) Center for Cellular Construction, San Francisco, CA, 94158, USA
In silico patient generation via Quantum Computers
Recent advances in clinical research have been driven by access to large-scale medical data, enabling population-level insights and novel therapies. However, traditional approaches face limitations in cost, time, and applicability to rare or vulnerable populations. In silico methods, powered by computational modeling and artificial intelligence (AI), have emerged as a complementary solution, simulating biological systems and optimizing personalized therapies. Despite their potential, these models struggle to capture the full complexity of clinical data distributions.
To address this, we propose tensor networks as an advanced framework for modeling clinical and biological data. Tensor networks efficiently encode high-order correlations between variables, even with limited samples, enhancing predictive accuracy in AI-driven clinical outcomes. Moreover, their inherent compatibility with quantum circuits unlocks exponential computational advantages—leveraging quantum superposition and entanglement to generate synthetic data at unprecedented scales. This hybrid approach enables faster, more precise synthetic data generation, accelerating AI training and improving in silico clinical trial design.By integrating tensor networks with quantum computing, we present a transformative pathway to overcome classical limitations, reduce costs, and expedite the development of personalized medicine. This methodology bridges computational efficiency with biological fidelity, paving the way for next-generation clinical research.
Víctor M. Tenorio, Universidad Rey Juan Carlos, Elvin Isufi, Delft University of Technology, Geert Leus, Delft University of Technology, Antonio G. Marques, Universidad Rey Juan Carlos
Tracking Network Dynamics using Probabilistic State-Space ModelsWe introduce a probabilistic approach for tracking the dynamics of unweighted and directed graphs using state-space models (SSMs). Unlike conventional topology inference methods that assume static graphs and generate point-wise estimates, our method accounts for dynamic changes in the network structure over time. We model the network at each timestep as the state of the SSM, and use observations to update beliefs that quantify the probability of the network being in a particular state. Then, by considering the dynamics of transition and observation models through the update and prediction steps, respectively, the proposed method can incorporate the information of real-time graph signals into the beliefs. These beliefs provide a probability distribution of the network at each timestep, being able to provide both an estimate for the network and the uncertainty it entails. Our approach is evaluated through experiments with synthetic and real-world networks. The results demonstrate that our method effectively estimates network states and accounts for the uncertainty in the data, outperforming traditional techniques such as recursive least squares.
Hamza Touati, Sandra Roger, Carmen Botella-Mascarell, Computer Science Department, Universitat de València
A Study on Range Estimation in a Monostatic ISAC ContextWith the evolution of wireless systems toward higher frequencies and wider bandwidths, Integrated Sensing and Communication (ISAC) has emerged as a promising paradigm that enables sensing and communication to share the same waveform and hardware resources. In this work, we investigate the sensing performance of two waveform types: OFDM, a communication-centric waveform widely used in current mobile systems, and PMCW, a radar-centric waveform based on pseudo-random binary sequences. We evaluate their ability to estimate the range of a moving target by analyzing reflected echoes, using two classical signal processing techniques—FFT and MUSIC. Results show that PMCW provides robust range estimation even at longer distances, while OFDM suffers from performance degradation. Additionally, MUSIC demonstrates consistent accuracy across waveforms in terms of RMSE.
Daniel Pereira-Ruisánchez, Óscar Fresnedo, Darian Pérez-Adán, Luis Castedo, Department of Computer Engineering & CITIC Research Center, University of A Coruña, Spain
C-Footprints: A Statistic-Based Clustering for Pilot Allocation in Cell-Free Massive MIMOCF-mMIMO communications rely on accurately knowing the CSI to perform coherent signal processing. However, achieving pilot contamination-free channel estimation is usually challenging due to the limited length of the coherence blocks. In this work, we introduce a novel method for user equipment clustering in CF-mMIMO, which allows us to transform the complex pilot allocation task into a computationally affordable problem. The proposed strategy, which has been termed C-footprints, exploits the spatial information encoded in the channel correlation matrices to gather potentially contaminating user equipments. The simulation results show that by implementing a simple heuristic over the resulting clusterings, we can achieve a competitive network performance in terms of sum-NMSE.
Samuel Rey, Universidad Rey Juan Carlos, Seyed Saman Saboksayr, University of Rochester, Gonzalo Mateos, University of Rochester
Non-negative Weighted DAG Structure LearningWe address the problem of learning the topology of directed acyclic graphs (DAGs) from nodal observations, which adhere to a linear structural equation model. Recent advances framed the combinatorial DAG structure learning task as a continuous optimization problem, yet existing methods must contend with the complexities of non-convex optimization. To overcome this limitation, we assume that the latent DAG contains only non-negative edge weights. Leveraging this additional structure, we argue that cycles can be effectively characterized (and prevented) using a convex acyclicity function based on the log-determinant of the adjacency matrix. This convexity allows us to relax the task of learning the non-negative weighted DAG as an abstract convex optimization problem. We propose a DAG recovery algorithm based on the method of multipliers, that is guaranteed to return a global minimizer. Furthermore, we prove that in the infinite sample size regime, the convexity of our approach ensures the recovery of the true DAG structure. We empirically validate the performance of our algorithm in several reproducible synthetic-data test cases, showing that it outperforms state-of-the-art alternatives.
Vahid Vahidpour, Roberto López-Valcarce, atlanTTic Research Center, Universidade de Vigo, Vigo, Spain
Spectrally precoded OTFS: Delay-Doppler detectionSimilarly to other multicarrier-based schemes, orthogonal time-frequency space (OTFS) modulation suffers from high out-of-band radiation (OBR) causing adjacent channel interference. Although it is possible to apply well-known techniques such as spectral precoding in the time-frequency domain, which is effective toward OBR reduction, for OTFS the impact in terms of bit error rate (BER) degradation at the receiver can be significant. We enhance maximum ratio combining detection for spectrally precoded OTFS by introducing an iterative delay-Doppler domain scheme specifically designed for this setting. Simulation results show that the proposed approach significantly reduces OBR and improves BER performance compared to state-of-the-art detectors across various precoding methods, modulation orders, and frame sizes.
Marc Vilà-Insa, Aniol Martí, Jaume Riba, Meritxell Lamarca, Universitat Politècnica de Catalunya (UPC)
Low-Complexity Detection of Permutational Index Modulation for Noncoherent Communications
This work presents a massive SIMO scheme for wireless communications with one-shot noncoherent detection. It is based on permutational index modulation over OFDM. Its core principle is to convey information on the ordering in which a fixed collection of values is mapped onto a set of OFDM subcarriers. A spherical code is obtained which provides improved robustness against channel impairments. A simple detector based on the sorting of quadratic metrics of data is proposed. By exploiting statistical channel state information and hardening, it reaches near-ML error performance with a low-complexity implementation.
