Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SINR Estimation under Limited Feedback via Online Convex Optimization

Lorenzo Maggi, Boris Bonev, Reinhard Wiesmayr, Sebastian Cammerer, Alexander Keller

TL;DR

Numerical experiments show that the proposed online convex optimization framework consistently yields more accurate SINR estimates than state-of-the-art schemes and exhibits continual learning capabilities, which are essential for adapting to time-varying SINR regimes.

Abstract

Accurate signal-to-interference-plus-noise ratio (SINR) estimation is essential for resource allocation in wireless systems, yet it is often hindered by limited and intermittent feedback. We propose an online convex optimization framework to estimate the SINR from ACK/NACK feedback, channel quality indicator (CQI) reports, and previously selected modulation and coding scheme (MCS) values. Specifically, we minimize a regularized binary cross entropy loss using a mirror descent method enhanced by Nesterov momentum for accelerated SINR tracking. The hyperparameters can be automatically tuned online via an expert-advice algorithm. Numerical experiments across multiple ray-traced scenarios show that our method consistently yields more accurate SINR estimates than state-of-the-art schemes and exhibits continual learning capabilities, which are essential for adapting to time-varying SINR regimes.

SINR Estimation under Limited Feedback via Online Convex Optimization

TL;DR

Numerical experiments show that the proposed online convex optimization framework consistently yields more accurate SINR estimates than state-of-the-art schemes and exhibits continual learning capabilities, which are essential for adapting to time-varying SINR regimes.

Abstract

Accurate signal-to-interference-plus-noise ratio (SINR) estimation is essential for resource allocation in wireless systems, yet it is often hindered by limited and intermittent feedback. We propose an online convex optimization framework to estimate the SINR from ACK/NACK feedback, channel quality indicator (CQI) reports, and previously selected modulation and coding scheme (MCS) values. Specifically, we minimize a regularized binary cross entropy loss using a mirror descent method enhanced by Nesterov momentum for accelerated SINR tracking. The hyperparameters can be automatically tuned online via an expert-advice algorithm. Numerical experiments across multiple ray-traced scenarios show that our method consistently yields more accurate SINR estimates than state-of-the-art schemes and exhibits continual learning capabilities, which are essential for adapting to time-varying SINR regimes.
Paper Structure (9 sections, 12 equations, 5 figures, 3 algorithms)

This paper contains 9 sections, 12 equations, 5 figures, 3 algorithms.

Table of Contents

  1. Introduction
  2. Problem statement
  3. estimation method
  4. estimation from ACK/NACK feedback
  5. Integrating the feedback
  6. Hyperparameter self-tuning
  7. Theoretical guarantees
  8. Numerical experiments
  9. Conclusions

Figures (5)

  • Figure 1: estimation with momentum. A comparison among , and momentum methods for estimation from ACK/NACK. offers a good compromise between tracking speed and stability.
  • Figure 2: Impact of hyperparameters on estimation accuracy. Root mean square error (RMSE) of the prediction evaluated in two scenarios across different values of the reliance $\alpha$, momentum $\beta$, and stepsize $\eta$. The dataset is produced using algorithm pedersen2007frequency. The optimal $\alpha,\beta,\eta$ depend on the scenario, motivating the need for online tuning.
  • Figure 3: Self-tuning. Cumulative density function (CDF) of the percentage of experts estimating more accurately than expert-advice algorithms ( and ), across all scenarios.
  • Figure 4: Comparison with . estimation accuracy improvement with respect to SALAD wiesmayr2025salad and for different values of stepsize $\Delta$ with 20 experts.
  • Figure 5: Spectral efficiency versus estimation accuracy trade-off arising when exploring random values with probability $p_e$.