Time-Average AoI Metrics

• Time-average AoI is a performance metric that measures the mean elapsed time since the latest update was generated, defining information freshness.
• Analytical methods such as renewal-reward theory and Markov chain analysis yield closed-form expressions to optimize scheduling under system constraints.
• Its applications span queueing theory, wireless networks, and cyber-physical systems, where balancing throughput, energy, and latency is critical for real-time performance.

Time-average Age of Information (AoI) is a performance metric central to the analysis and optimization of information freshness in status update systems. It captures, for each source or information stream, the long-term mean time elapsed since the most recent update at the receiver was generated. This quantity is now foundational in queuing theory, wireless networking, and cyber-physical systems, where timely information is critical. Time-average AoI exhibits complex interactions with system parameters, scheduling and sampling policies, channel reliability, and resource constraints, and has led to a broad taxonomy of analytical results and operational trade-offs in diverse environments.

1. Mathematical Definition and Fundamental Properties

For a discrete-time system, define the instantaneous AoI at the destination at time $t$ as $\Delta(t) = t - U(t)$, where $U(t)$ is the generation time of the latest received status update. The time-average AoI over a horizon $T$ is

$$\bar \Delta = \lim_{T \rightarrow \infty} \frac{1}{T} \sum_{t=1}^T \Delta(t)$$

For continuous time, the definition is analogous, using an integral over the monitoring interval. The time-average AoI reflects the sawtooth-like evolution of information age, incremented over time and reset to lower values upon successful receipt of fresher updates (Munari et al., 2020, Chen et al., 2022).

In renewal settings common to queueing systems, the area under the AoI sawtooth within each renewal cycle forms the basis for closed-form averages via the renewal-reward theorem. For instance, with i.i.d. interarrival times $X$ and associated sojourn times $T$,

$$\bar\Delta = \frac{\mathbb{E}[X T] + \frac{1}{2} \mathbb{E}[X^2]}{\mathbb{E}[X]}$$

This general form reveals the dependence of AoI on both the temporal structure of status generation and the system's service dynamics (Chen et al., 2023, Tripathi et al., 2019).

2. Closed-Form Expressions Across Models

Time-average AoI admits exact solutions in a range of queueing and network models:

• Discrete-time FCFS (Ber/G/1) queues yield (Tripathi et al., 2019):

$A_{\rm ave}^{\rm Ber/G/1} = \frac{1}{\lambda} + \frac{L_S(1-\lambda)-(1-\lambda)}{(1-\lambda)\mu} + \frac{\lambda\E[S^2]-\rho}{2(1-\rho)}$

where $L_S(\cdot)$ is the service-time PGF.

• M/M/1 queue:

$\Delta(t) = t - U(t)$0

for infinite-source, exponential arrivals and service (Chen et al., 2022).

• Energy-harvesting systems (FCFS, negligible service time):

$\Delta(t) = t - U(t)$1

with closed formulas incorporating buffer and storage constraints (Zheng et al., 2019).

• Slotted ALOHA and IRSA random access: Under memoryless activation, for IRSA,

$\Delta(t) = t - U(t)$2

with $\Delta(t) = t - U(t)$3 capturing the success probability per frame (Munari et al., 2020).

• Multi-source, multi-sensor polling: Mean AoI per source is defined over the aggregate of observed ages, optimized by intelligent scheduling over the sensor-source observation matrix (Kalør et al., 2018).
• Dual-queue updating (M/M and M/D): For equally fast parallel servers,

$\Delta(t) = t - U(t)$4

Both outperform the M/M/1 baseline for equivalent rates (Chen et al., 2022).

These closed-form derivations generalize through renewal-theoretic approaches, Markov chain stationary distributions, and moment methods, enabling explicit dependence on system parameters.

3. Methodologies for Analysis and Estimation

Rigorous evaluation of time-average AoI leverages several methodologies:

• Renewal-Reward Theory: Used to relate the average area under age curves in stochastic update systems to mean inter-update intervals and reset probabilities (Tripathi et al., 2019, Chen et al., 2022).
• Markov Chain Analysis: Geometric or higher-order Markov chains characterize AoI as a function of state transition probabilities, particularly in contention networks and systems with multiple classes of users (Farag et al., 2023).
• Moment-based Estimation: When arrival distributions are unknown but a finite set of moments is available, one can bound $\Delta(t) = t - U(t)$5 using truncated moment expansions of the arrival process PGF, yielding nonparametric upper and lower estimates (Chen et al., 2023).
• State-Flow Graph and Exact Recursion: For systems with multiple interacting servers or sensors, embedded Markov or state-flow graphs facilitate calculation of inter-refresh intervals and accumulated age (Chen et al., 2022).
• Drift-Plus-Penalty and Lyapunov Optimization: For constrained optimization (e.g., joint power and AoI constraints), Lyapunov-based techniques enforce virtual queues and stabilize time-average AoI within target bounds while optimizing secondary objectives (Moltafet et al., 2019, Fountoulakis et al., 2021).

The diversity of models requires careful adaptation of these methods to queueing discipline, arrival process, service policy, and system constraints.

4. AoI in Resource-Constrained and Stochastic Environments

Time-average AoI serves as an explicit constraint or optimization objective in networks facing wireless channel unreliability, limited energy/battery resources, or application-imposed age constraints:

• AoI-constrained bandit and learning: In the presence of transmission failures, optimal scheduling can be cast as a constrained Multi-Armed Bandit, with algorithms guaranteeing to respect per-source AoI guarantees and achieving sublinear regret scaling $\Delta(t) = t - U(t)$6 (Huang et al., 2021).
• Wireless powered and blockage-prone scenarios: AoI in energy harvesting (EH) systems encompasses the combined statistics of charging times, transmission successes, and renewal intervals. For example, the average AoI in a pinching antenna-assisted WPCN with probabilistic LoS blockage is derived by characterizing the negative-binomial energy harvesting process and transmission success probabilities (Hu et al., 8 Nov 2025).
• Joint AoI–resource optimization: Dynamic programs and Lyapunov drift methods balance freshness against energy or power consumption, ensuring feasibility w.r.t. average AoI requirements over long horizons (Moltafet et al., 2019).

These settings highlight a spectrum of trade-offs—between energy, throughput, and age—that must be jointly managed.

5. Operational Trade-offs and Design Guidelines

Optimizing time-average AoI requires navigating several inherent trade-offs:

• Throughput vs. Freshness: Systems such as IRSA with advanced SIC decoding achieve lower AoI at higher loads, but with frame-length and repetition trade-offs that must be tuned to system scale (Munari et al., 2020).
• Parallelism and Determinism: Multiple servers or paths (e.g., dual-queue systems) substantially reduce AoI by leveraging path diversity, with deterministic servers outperforming random for average AoI when properly matched to arrival rates (Chen et al., 2022).
• Policy Selection: Causal and nonpreemptive scheduling policies (e.g., long-wait, PAoI-threshold, postponed-plan) can be analytically optimized to minimize AoI under constraints of channel/processing times, reducing average age in edge computing and multi-hop settings (Zhu et al., 2022).
• Estimation Accuracy: In practical systems with unknown statistics, truncating to a modest number ($\Delta(t) = t - U(t)$7–7) of arrival moments suffices to bound AoI within tight margins, except in heavy-traffic or highly variable regimes (Chen et al., 2023).

Optimal operation often entails dynamic adaptation—tuning scheduling policy, probing degree distributions, assigning subchannels, or adjusting power and activation thresholds to match the environment.

6. Time-Average AoI in Multi-Traffic and Heterogeneous Networks

Modern applications frequently require simultaneous support for AoI-sensitive and deadline-oriented traffic:

• AoI–reliability dual constraints: In IIoT and control applications, time-average AoI and reliability/latency (e.g., deadline-missed probability) must be jointly characterized. The average AoI for 'generate-at-will' flows in the presence of concurrent urgent traffic is a simple inverse of the successful update probability, yielding explicit trade-off curves with system load, access probabilities, and buffer deadlines (Farag et al., 2023).
• CMDP and Lyapunov relaxations: Frameworks using constrained Markov decision processes, virtual queues, and per-frame or per-slot minimization enable AoI minimization under additional performance constraints (timely throughput, energy budgets) (Fountoulakis et al., 2021).

These frameworks provide both analytical and algorithmic recipes for allocating resources among heterogeneous flows according to operational priorities.

7. Discrete-Time Effects and Generalizations

Discrete-time analysis reveals integer corrections and PGF substitutions compared to continuous-time results, but proof techniques (renewal-reward, residual life, Markov chains) and key insights largely translate. For example, in the Ber/G/1 queue, the time-average AoI contains both PGF-based sojourn corrections and variance terms mirroring those in their continuous-time analogs (Tripathi et al., 2019).

Extensions cover:

• Vacation models: Additional service interruptions require explicit correction terms, increasing average age.
• Preemption and infinite-server models: Preemptive LCFS and $\Delta(t) = t - U(t)$8 setups yield formulas incorporating the minimum of interarrival and service statistics, and redundancy times, respectively (Tripathi et al., 2019).
• Common observations and sensor fusion: Scheduling across networks of sensors with overlapping coverage admits policies that exploit redundancy, leveraging waiting intervals for greater “information gain,” often substantially reducing time-average AoI (Kalør et al., 2018).

The general toolkit for time-average AoI applies seamlessly across stochastic processes, queueing disciplines, and network architectures.

---

References

• (Munari et al., 2020) Average Age of Information of Irregular Repetition Slotted ALOHA
• (Chen et al., 2022) Analysis of Age of Information in Dual Updating Systems
• (Hu et al., 8 Nov 2025) Average AoI in Pinching Antenna-assisted WPCNs with Probabilistic LoS Blockage
• (Kalør et al., 2018) Minimizing the Age of Information from Sensors with Common Observations
• (Chen et al., 2023) Estimating Age of Information Using Finite Order Moments
• (Moltafet et al., 2019) Power Minimization in Wireless Sensor Networks With Constrained AoI Using Stochastic Optimization
• (Farag et al., 2023) On the Analysis of AoI-Reliability Tradeoff in Heterogeneous IIoT Networks
• (Zheng et al., 2019) Closed-Form Analysis of Non-Linear Age-of-Information in Status Updates with an Energy Harvesting Transmitter
• (Zhu et al., 2022) Online Scheduling of Transmission and Processing for AoI Minimization with Edge Computing
• (Fountoulakis et al., 2021) Scheduling Policies for AoI Minimization with Timely Throughput Constraints
• (Tripathi et al., 2019) Age of Information for Discrete Time Queues
• (Huang et al., 2021) AoI-Constrained Bandit: Information Gathering over Unreliable Channels with Age Guarantees