Age-Aware Application-Layer FEC (A³L-FEC)

• Age-Aware Application-Layer FEC is a protocol that uses Age of Information to control flow and minimize data staleness via adaptive FEC and rate adjustments.
• It employs both fixed and variable FEC schemes over UDP, combining MDS coding and dynamic rate control based on empirical AoI feedback.
• Empirical evaluations demonstrate significant reductions in age violations and lower delays compared to protocols like TCP-BBR and ACP⁺.

Age-Aware Application-Layer FEC (A³L-FEC) is an application-layer flow-control protocol designed to minimize data staleness in communication networks by leveraging Age of Information (AoI) as its primary feedback signal. Operating atop UDP, A³L-FEC employs forward error correction (FEC) and adaptive rate control with the explicit objective of reducing the frequency and duration of so-called age violations—events where the AoI surpasses an application-defined threshold. This protocol represents an advance from purely theoretical AoI models towards a deployable scheme exhibiting substantial empirical gains over established protocols such as TCP-BBR and ACP⁺, as quantified in both simulation and emulation testbeds (Baghaee et al., 2024).

1. Age of Information and Age Violations

AoI, denoted $\Delta(t)$, quantifies the staleness of data at a monitoring point as the time elapsed since the freshest received sample was generated: $\Delta(t) = t - t'$, where $t'$ is the generation time of the most recently received sample. In practical systems, AoI exhibits a saw-tooth pattern—growing linearly between receptions and dropping sharply upon fresh packet arrival (see Figure 1, (Baghaee et al., 2024)).

The protocol targets application-specified peak-age thresholds, denoted $\mathrm{AVT}$. An age violation is an event for which $\Delta(t) \geq \mathrm{AVT}$, formally indicated by $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$. Aggregated over $T$ slots, the age-violation rate is defined as:

$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$

(Equation (7), Sec. III–D, (Baghaee et al., 2024)). Minimizing this metric is the explicit goal of A³L-FEC.

2. Protocol Design: Packet Generation, Coding, and Flow Control

A³L-FEC orchestrates packet-level FEC atop UDP, adapting its coding and sending rate based on AoI feedback and observed network conditions. Two implementation variants are defined:

2.1 A³L-FEC-FSFB: Fixed Sampling, Fixed Block-Length

• Time-slotted operation: Samples $s_\tau$ of $K$ bits generated each slot are split into $\Delta(t) = t - t'$0 data chunks and encoded by an $\Delta(t) = t - t'$1 MDS code, resulting in $\Delta(t) = t - t'$2 coded chunks. Recovery of any $\Delta(t) = t - t'$3 enables decoding.
• Transmission policy: In slot $\Delta(t) = t - t'$4, the transmitter selects a subset $\Delta(t) = t - t'$5 of chunks (from the latest $\Delta(t) = t - t'$6 samples) to transmit. The average transmission rate is $\Delta(t) = t - t'$7 (codewords/slot), corresponding to $\Delta(t) = t - t'$8 chunks per slot.
• Loss modeling: The probability a chunk is lost before the bottleneck is $\Delta(t) = t - t'$9, after is $t'$0, yielding compound chunk-drop probability:

$t'$1

(Eq. (1), (Baghaee et al., 2024))

• SIS (Stationary Independent Selection) policy: For age offsets $t'$2, chunk selection in each slot is independent with probability $t'$3, optimal under fixed rate $t'$4 by

$t'$5

(Eq. (10), (Baghaee et al., 2024))

• Rate adaptation/congestion control: After each monitoring interval of $t'$6 slots, the receiver computes the empirical age-violation rate $t'$7, mean chunk delay $t'$8, and forms EMAs $t'$9, $\mathrm{AVT}$0. The transmitter adapts $\mathrm{AVT}$1 according to Algorithm 1 (see Section 2.3): multiplicative probes (by factor $\mathrm{AVT}$2), reductions, and fine-grained additive adjustments, all bounded to maintain queue stability.

2.2 A³L-FEC-VSVB: Variable Sampling, Variable Block-Length

• Generate-at-will: The source may defer sampling. Within each interval of length $\mathrm{AVT}$3, it sets the sample period $\mathrm{AVT}$4 and FEC block-length $\mathrm{AVT}$5, maintaining a rate $\mathrm{AVT}$6.
• Interval operations: Each monitoring interval involves (1) exact violation computation from decode/generate times, (2) average chunk delay $\mathrm{AVT}$7 and packet-delivery ratio $\mathrm{AVT}$8 calculation, (3) evaluation of hypothetical violations under alternative $\mathrm{AVT}$9, choosing the $\Delta(t) \geq \mathrm{AVT}$0 minimizing $\Delta(t) \geq \mathrm{AVT}$1, (4) updating $\Delta(t) \geq \mathrm{AVT}$2, and (5) setting next $\Delta(t) \geq \mathrm{AVT}$3.
• Congestion control: The law employs similar EMAs, but directly tracks MinRTT and bounds $\Delta(t) \geq \mathrm{AVT}$4 between $\Delta(t) \geq \mathrm{AVT}$5, applying multiplicative changes and small probes as detailed in Algorithm 2.

2.3 Congestion-Control Pseudocode Overview

Both FSFB and VSVB implement dynamic adjustments, controlling $\Delta(t) \geq \mathrm{AVT}$6 using EMA-based feedback on age violations and chunk delays. A representative excerpt from Algorithm 1 (FSFB) is:

$\Delta(t) = t - t'$53 Algorithm 2 for VSVB is structurally analogous with block-length and interval updates.

3. Theoretical Limits and Analytical Model

The protocol’s design is informed by explicit stability and decodability analysis:

• Queue stability upper bound: For bottleneck service rate $\Delta(t) \geq \mathrm{AVT}$7 (chunks/slot) and codeword size $\Delta(t) \geq \mathrm{AVT}$8, the max stable codeword rate is

$\Delta(t) \geq \mathrm{AVT}$9

(Eq. (20), (Baghaee et al., 2024)).

• Decoding probability: The chance that sample $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$0 is decodable by time $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$1:

$\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$2

(Eqs. (21)-(22), (Baghaee et al., 2024)).

• Outage probability: For age exceeding $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$3, let $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$4 denote “age at $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$5 is $\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$6”:

$\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$7

$\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$8

(Eqs. (23)-(24), (Baghaee et al., 2024)).

4. Evaluation Methodology

4.1 MATLAB Simulations

• Topology: Single-bottleneck FCFS queue ($\mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} }$9 pkts/slot), buffer of $T$0 chunks.
• Parameters: $T$1, FEC $T$2, $T$3–$T$4, $T$5.
• Baselines: ACP⁺ protocol [26,27 in (Baghaee et al., 2024)].
• Metrics: Age-violation rate $T$6 over $T$7 slots and $T$8 runs.

4.2 ns-3 TCP AoI Simulations

• Topology: Source$T$9Router1(1000Mb/s,5ms)$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$0Router2(10Mb/s,10ms, queue=100 pkts)$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$1Sink.
• Compared TCP variants: Reno, Cubic, BIC, NewReno, Ledbat, BBR (best AoI).
• Metric: Average AoI in $$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$2 s sessions.

4.3 Mininet-WiFi Emulation

• Topology: H1$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$3S1$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$4S2$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$5S4$$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$6H2 (upstream: 1Mb/s, 30ms, 5% loss, 1000-pkt queue).
• Implementations: A³L-FEC-VSVB atop UDP (C++), TCP-BBR (Linux). BBR sample periods: $$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$7 ms; A³L-FEC-VSVB auto-adapts.
• Metrics: $$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$8, average AoI, average packet delay. Thresholds $$\mathrm{AV} = \frac{1}{T} \sum_{t=1}^T \mathbf{1}_{ \{\Delta(t)\geq \mathrm{AVT}\} },$$9 ms, $s_\tau$0 ms. $s_\tau$1 samples/run, $s_\tau$2 runs averaged.

5. Empirical Results

5.1 MATLAB: A³L-FEC-FSFB Versus ACP⁺

For $s_\tau$3, $s_\tau$4:

• $s_\tau$5: $s_\tau$6
• $s_\tau$7: $s_\tau$8
• $s_\tau$9: $K$0
• $K$1: $K$2
• $K$3: $K$4
• ACP⁺: $K$5

The optimal $K$6 configuration is within $K$7 of the stability bound $K$8. ACP⁺ suffers higher violations due to suboptimal rate utilization.

5.2 ns-3: TCP Variant Baseline

TCP-BBR achieves the lowest average AoI ($K$9 s); Cubic: $\Delta(t) = t - t'$00 s, Reno: $\Delta(t) = t - t'$01 s, Ledbat: $\Delta(t) = t - t'$02 s.

5.3 Mininet-WiFi: A³L-FEC-VSVB vs TCP-BBR

• $\Delta(t) = t - t'$03 ms: A³L-FEC-VSVB attains $\Delta(t) = t - t'$04; BBR at $\Delta(t) = t - t'$05 ms: $\Delta(t) = t - t'$06, $\Delta(t) = t - t'$07 ms: $\Delta(t) = t - t'$08, $\Delta(t) = t - t'$09 ms: $\Delta(t) = t - t'$10, $\Delta(t) = t - t'$11 ms: $\Delta(t) = t - t'$12.
• $\Delta(t) = t - t'$13 ms: A³L-FEC-VSVB $\Delta(t) = t - t'$14; BBR at $\Delta(t) = t - t'$15 ms: $\Delta(t) = t - t'$16, $\Delta(t) = t - t'$17 ms: $\Delta(t) = t - t'$18, $\Delta(t) = t - t'$19 ms: $\Delta(t) = t - t'$20, $\Delta(t) = t - t'$21 ms: $\Delta(t) = t - t'$22.
• AoI: A³L-FEC-VSVB: $\Delta(t) = t - t'$23–$\Delta(t) = t - t'$24 ms; BBR: $\Delta(t) = t - t'$25 ms.
• Delay: A³L-FEC-VSVB: $\Delta(t) = t - t'$26 ms; BBR: $\Delta(t) = t - t'$27 ms.

5.4 Coding-Rate Tuning in VSVB

As $\Delta(t) = t - t'$28 increases, $\Delta(t) = t - t'$29 grows due to higher queuing. For $\Delta(t) = t - t'$30 ms, $\Delta(t) = t - t'$31 remains low; at $\Delta(t) = t - t'$32 ms, $\Delta(t) = t - t'$33. The network RTT $\Delta(t) = t - t'$34 ms is a hard lower bound for viable $\Delta(t) = t - t'$35 (Baghaee et al., 2024).

6. Configuration Guidelines and Operational Recommendations

• Set $\Delta(t) = t - t'$36 above the one-way RTT by a safe margin, e.g., $\Delta(t) = t - t'$37.
• Choose $\Delta(t) = t - t'$38 for sufficient statistical stability in PDR estimation.
• Initialize $\Delta(t) = t - t'$39 to $\Delta(t) = t - t'$40 (VSVB) or $\Delta(t) = t - t'$41 (FSFB).
• Use $\Delta(t) = t - t'$42 for multiplicative probes; $\Delta(t) = t - t'$43 for EMAs.
• Bound $\Delta(t) = t - t'$44 to $\Delta(t) = t - t'$45 (FSFB) or $\Delta(t) = t - t'$46 (VSVB).
• Increase $\Delta(t) = t - t'$47 above $\Delta(t) = t - t'$48 in high-loss settings; minimize redundancy in low-loss to limit queue buildup.

7. Limitations and Prospective Research

Several avenues and constraints are identified:

• Overhead/redundancy tradeoff: FEC imposes $\Delta(t) = t - t'$49 redundant chunks, increasing link utilization; in high-loss/tight-delay regimes, fine-tuning redundancy is critical.
• FSFB slow-start: Initialization with $\Delta(t) = t - t'$50 can temporarily exceed bottleneck capacity. VSVB mitigates this using MinRTT-aware startup.
• Multi-flow/multi-server: The scheme generalizes to multiple transmitters $\Delta(t) = t - t'$51 via

$\Delta(t) = t - t'$52

(Eq. (25)). Practical scaling to large IoT scenarios is an open challenge.

• Real-world transport: Further validation beyond Mininet-WiFi—including cellular, LoRa, NB-IoT, DTN, deep-space—remains outstanding.
• Value-aware extensions: Adapting FEC strength to semantic or content “value” of updates.
• Energy-age tradeoff: Joint optimization in energy-harvesting sources is a prospective direction.

In summary, A³L-FEC constitutes a practical, feedback-driven approach to AoI optimization, combining MDS-based FEC with fine-grained age-aware rate adaptation. Its two principal variants are tailored to fixed-rate and adaptive operational regimes. Empirical studies demonstrate 10×–100× reductions in age violations versus ACP⁺ and TCP-BBR, halving or bettering average AoI, while retaining minimal queueing delay relative to network propagation delay. Theoretical bounds, algorithmic procedures, and practical guidelines outlined here inform deployment and extension in diverse, loss- and delay-prone environments (Baghaee et al., 2024).