eess.SPNov 1, 2026

FADE: A Flow-Based Approach for Multichannel Tail Estimation

Authors: Parmida Valiahdi, Niloofar Mehrnia, Mustafa Furkan Beker, Ubeydullah Erdemir, Ibrahim Hokelek, Sinem Coleri

Organizations: Department of Electrical and Electronics Engineering, Koc University, Istanbul, Turkey · Department of Electrical Engineering and Computer Science, KTH Royal Institute of Technology, Stockholm, Sweden · Communications and Signal Processing Research (HISAR) Lab., TUBITAK-BILGEM, Kocaeli, Turkey · Department of Electronics and Communication Engineering, Istanbul Technical University, Istanbul, Turkey

Abstract

Ultra-Reliable Low-Latency Communication (URLLC) is vulnerable to rare low-received-power events, and the spatial-diversity benefits of multiple receivers can be significantly reduced when they experience simultaneous deep fading. Multivariate Extreme Value Theory (MEVT) provides a principled framework for characterizing such joint tail behavior. However, conventional MEVT methods often rely on restrictive parametric assumptions or discrete representations of the dependence structure, which may not fully capture complex multivariate tail behavior. This paper proposes Flow-based Angular Density Estimation (FADE), a flow-based approach for characterizing the dependence between extreme events across receivers within the MEVT framework. FADE transforms multivariate extremes into a standard extreme-value representation and uses a flow-based density estimator to learn the dependence structure from the transformed data without imposing a predefined parametric form. The proposed framework is evaluated using experimentally collected data from a UAV communication system, in which the UAV serves as the transmitter and communicates with two ground receivers. FADE provides closer agreement with the empirical joint CDF, achieving reductions of up to 22.4% in MAE compared with the state-of-the-art Logistic and Maximum Empirical Likelihood (MEL) methods, while also improving distributional agreement and reducing absolute bias.

Figures & tables

Explore similar work

CardsList
  1. EVT-Based Generative AI for Tail-Aware Channel Estimation

    Apr 27, 2026Parmida Valiahdi, Niloofar Mehrnia, Walid Saad +1Synthetic Data AugmentationWireless Communications

  2. Physics-informed VAE-EVT for Tail Aware Radio Map Prediction

    Aug 15, 2026Amanda Sheron Gamage, Niloofar Mehrnia, James GrossRadio Propagation ModelingWireless Communications

  3. Low-Altitude Fluid Antenna Network with Multi-Agent Reinforcement Learning

    Aug 28, 2026Tong Zhang, Yanfei Su, Shuai Wang +3Wireless CommunicationsWireless Resource Allocation