[Paper Review] Statistical Age-of-Information Optimization for Status Update over Multi-State Fading Channels

Yuquan Xiao, Qinghe Du|arXiv (Cornell University)|Mar 20, 2023

Age of Information Optimization16 references4 citations

TL;DR

This paper proposes a near-optimal sampling interval scheme for status update systems over multi-state fading channels to minimize statistical age of information (AoI) under age violation probability constraints. By approximating the problem via fractional programming for small AoI exponents and convex optimization for large exponents, the scheme adapts sampling intervals based on channel state information (CSI), with the counterintuitive result that for extremely stringent AoI requirements, the optimal interval converges to a constant regardless of CSI variation.

ABSTRACT

Age of information (AoI) is a powerful metric to evaluate the freshness of information, where minimization of average statistics, such as the average AoI and average peak AoI, currently prevails in guiding freshness optimization for related applications. Although minimizing the statistics does improve the received information's freshness for status update systems in the sense of average, the time-varying fading characteristics of wireless channels often cause uncertain yet frequent age violations. The recently-proposed statistical AoI metric can better characterize more features of AoI dynamics, which evaluates the achievable minimum peak AoI under the certain constraint on age violation probability. In this paper, we study the statistical AoI minimization problem for status update systems over multi-state fading channels, which can effectively upper-bound the AoI violation probability but introduce the prohibitively-high computing complexity. To resolve this issue, we tackle the problem with a two-fold approach. For a small AoI exponent, the problem is approximated via a fractional programming problem. For a large AoI exponent, the problem is converted to a convex problem. Solving the two problems respectively, we derive the near-optimal sampling interval for diverse status update systems. Insightful observations are obtained on how sampling interval shall be tuned as a decreasing function of channel state information (CSI). Surprisingly, for the extremely stringent AoI requirement, the sampling interval converges to a constant regardless of CSI's variation. Numerical results verify effectiveness as well as superiority of our proposed scheme.

Motivation & Objective

• Address the limitation of average AoI and peak AoI metrics in capturing tail behavior of age violations in dynamic wireless channels.
• Tackle the high computational complexity of statistical AoI minimization under age violation probability constraints in multi-state fading channels.
• Develop an efficient, near-optimal sampling interval policy that adapts to channel state information (CSI) while ensuring statistical AoI performance.
• Investigate the impact of AoI exponent on sampling interval adaptation and identify regime-specific optimization strategies.

Proposed method

• Approximate the statistical AoI minimization problem for small AoI exponents using fractional programming to simplify the non-convex objective.
• Transform the problem for large AoI exponents into a convex optimization problem to enable efficient solution via bisection search.
• Introduce a variable $ k^* $ to partition channel states, assigning maximum sampling intervals $ n_{\text{max}} $ to poor states and reduced intervals to good states.
• Use bisection search to jointly optimize the Lagrange multiplier $ \alpha $ and sampling intervals $ n_k $, ensuring constraint satisfaction.
• Convert continuous optimal intervals to integers via rounding (up or down), then validate feasibility to maintain constraint compliance.
• Select the better-performing scheme between small- and large-exponent approximations for each given AoI exponent to achieve near-optimal performance.

Experimental results

Research questions

• RQ1 How does the optimal sampling interval vary with channel state information (CSI) under statistical AoI constraints?
• RQ2 What is the impact of the AoI exponent on the structure of the optimal sampling policy?
• RQ3 Can the prohibitive computational complexity of statistical AoI minimization be effectively reduced while maintaining near-optimality?
• RQ4 Under what conditions does the sampling interval become independent of CSI, and why does this occur?
• RQ5 How does the proposed scheme compare to conventional average-peak-AoI and maximum-peak-AoI oriented policies in terms of statistical AoI performance?

Key findings

• For small AoI exponents, the optimal sampling interval decreases with better channel state information (CSI), indicating adaptive sampling based on channel quality.
• For large (infinite) AoI exponents, the optimal sampling interval converges to a constant value independent of CSI, indicating a saturation effect under stringent age requirements.
• The proposed scheme achieves statistical AoI performance very close to the optimal solution obtained via exhaustive search, with significantly reduced computational complexity.
• The computational complexity of the proposed method is $ O(2^K) $, far lower than the $ O(n_{\text{max}}^K) $ complexity of exhaustive search, enabling practical deployment.
• Numerical results confirm that only in extreme cases (small or infinite AoI exponent) do average-peak-AoI or maximum-peak-AoI oriented schemes achieve minimum statistical AoI.
• The scheme effectively upper-bounds the age violation probability while minimizing the achievable minimum peak AoI, making it suitable for real-time applications with strict freshness constraints.