arXiv subjects

Federico Penna

Publications and source records attributed to Federico Penna.

4 recordsLinked to original sources

arXiv2026

LightTune: Lightweight Forward-Only Online Fine-Tuning with Applications to Link Adaptation

Deploying machine learning (ML) algorithms on mobile phones is bottlenecked by performance degradation under dynamic, real-world conditions that differ from the offline training conditions. While continual learning and adaptation are essential to mitigate this distributional shift, conventional online learning methods are often computationally prohibitive for resource-constrained devices. In this paper, we propose LightTune, a lightweight, backpropagation-free online fine-tuning framework with provable convergence guarantees. LightTune opportunistically refines ML models using live test-time data only when performance falls below a predefined threshold, ensuring minimal computational overhead and highly efficient responsiveness. As a practical demonstration, we integrate LightTune into a block error rate (BLER) prediction algorithm for 6G mobile systems. This integration enables the ML BLER prediction model to dynamically adapt to previously unseen channel conditions in real-time. Our extensive results show a substantial reduction in the average BLER prediction error of up to 48.8% with online fine-tuning. Furthermore, we leverage this BLER prediction algorithm for link adaptation and demonstrate average throughput improvements of up to 15.5% compared to a conventional table-based outer loop link adaptation (OLLA) algorithm.

cs.NI↗

arXiv2013

Decentralized Eigenvalue Algorithms for Distributed Signal Detection in Cognitive Networks

In this paper we derive and analyze two algorithms -- referred to as decentralized power method (DPM) and decentralized Lanczos algorithm (DLA) -- for distributed computation of one (the largest) or multiple eigenvalues of a sample covariance matrix over a wireless network. The proposed algorithms, based on sequential average consensus steps for computations of matrix-vector products and inner vector products, are first shown to be equivalent to their centralized counterparts in the case of exact distributed consensus. Then, closed-form expressions of the error introduced by non-ideal consensus are derived for both algorithms. The error of the DPM is shown to vanish asymptotically under given conditions on the sequence of consensus errors. Finally, we consider applications to spectrum sensing in cognitive radio networks, and we show that virtually all eigenvalue-based tests proposed in the literature can be implemented in a distributed setting using either the DPM or the DLA. Simulation results are presented that validate the effectiveness of the proposed algorithms in conditions of practical interest (large-scale networks, small number of samples, and limited number of iterations).

cs.DC↗

arXiv2009

Theoretical Performance Analysis of Eigenvalue-based Detection

In this paper we develop a complete analytical framework based on Random Matrix Theory for the performance evaluation of Eigenvalue-based Detection. While, up to now, analysis was limited to false-alarm probability, we have obtained an analytical expression also for the probability of missed detection, by using the theory of spiked population models. A general scenario with multiple signals present at the same time is considered. The theoretical results of this paper allow to predict the error probabilities, and to set the decision threshold accordingly, by means of a few mathematical formulae. In this way the design of an eigenvalue-based detector is made conceptually identical to that of a traditional energy detector. As additional results, the paper discusses the conditions of signal identifiability for single and multiple sources. All the analytical results are validated through numerical simulations, covering also convergence, identifiabilty and non-Gaussian practical modulations.

cs.IT↗

arXiv2009

Cooperative Spectrum Sensing based on the Limiting Eigenvalue Ratio Distribution in Wishart Matrices

Recent advances in random matrix theory have spurred the adoption of eigenvalue-based detection techniques for cooperative spectrum sensing in cognitive radio. Most of such techniques use the ratio between the largest and the smallest eigenvalues of the received signal covariance matrix to infer the presence or absence of the primary signal. The results derived so far in this field are based on asymptotical assumptions, due to the difficulties in characterizing the exact distribution of the eigenvalues ratio. By exploiting a recent result on the limiting distribution of the smallest eigenvalue in complex Wishart matrices, in this paper we derive an expression for the limiting eigenvalue ratio distribution, which turns out to be much more accurate than the previous approximations also in the non-asymptotical region. This result is then straightforwardly applied to calculate the decision threshold as a function of a target probability of false alarm. Numerical simulations show that the proposed detection rule provides a substantial performance improvement compared to the other eigenvalue-based algorithms.

cs.IT↗