On Noisy ARQ in Block-Fading Channels

Behrooz Makki, Alexandre Graell i Amat, Thomas Eriksson

This paper, published in the IEEE Transactions on Communications, explores how to make wireless data transmission more reliable when the "receipt confirmation" (feedback) is itself prone to errors. The authors develop a mathematical framework to optimize power usage across retransmission rounds, demonstrating that smart power allocation can significantly mitigate the damage caused by noisy feedback.

TL;DR

This paper provides a comprehensive framework for analyzing Hybrid ARQ protocols in wireless channels with unreliable feedback. It proves that adaptive power allocation is essential for maintaining high throughput and low outage, and reveals that while Incremental Redundancy (INR) is generally more efficient, simpler repetition schemes are surprisingly more robust to feedback noise.

What Problem Does This Paper Address?

In wireless communication, systems use Automatic Repeat Request (ARQ): if a receiver can't decode a packet, it sends a "NACK" (Negative Acknowledgement) back to the transmitter to ask for a retransmission. Most research assumes this feedback link is perfect.

In reality, feedback channels are noisy. If a "NACK" is misread as an "ACK," the transmitter stops sending a failed packet, causing a data "outage." If an "ACK" is misread as a "NACK," the transmitter wastes power and time re-sending data the receiver already has. This paper tackles the challenge of optimizing transmission power and throughput when these feedback errors are unavoidable.

Methods

The authors employ an information-theoretic approach to model the system. They focus on two main Hybrid ARQ (HARQ) protocols:

  1. Repetition Time Diversity (RTD): The transmitter sends the exact same codeword again.
  2. Incremental Redundancy (INR): The transmitter sends new parity bits to help the receiver decode.

The study covers two temporal scenarios: Long-$L_c$ (the channel stays the same during all retransmissions) and Short-$L_c$ (the channel changes every round).

The core technical contribution is a closed-form expression for a power-limited throughput optimization problem. This allows engineers to calculate the best transmission power for each round to maximize data flow while staying within a total energy budget.

Datasets

This is a theoretical and numerical study; therefore, no real-world datasets were used. Instead, the authors evaluated their models using numerical simulations based on Nakagami-$N$ fading channel models. These are standard mathematical models used to simulate different levels of wireless signal interference and "fading" (signal dropping).

Key Insights and Ideas

  • The Robustness Trade-off: INR is the "high-performance" option, achieving higher throughput than RTD under most conditions. However, RTD is more "robust"—it loses less of its performance when the feedback channel gets noisier compared to INR.
  • Power Allocation is Non-Negotiable: In noisy feedback environments, using the same power for every retransmission is highly inefficient. Optimal power allocation is "crucial" for minimizing the probability of a total system outage.
  • Low SNR Convergence: Interestingly, at very low Signal-to-Noise Ratios (SNR), the performance of the complex INR protocol and the simpler RTD protocol converge. In "bad" signal areas, the extra complexity of INR doesn't provide much benefit.
  • Variable-Length Coding: Implementing INR with codewords of different lengths (variable-length coding) significantly improves its robustness against feedback errors compared to fixed-length versions.

Key Results

  • Outage Reduction: Optimal power allocation results in "considerable" reduction in outage probability compared to uniform power allocation, even when feedback is unreliable.
  • Throughput Sensitivity: Throughput is relatively stable when feedback error rates are small, but it becomes highly sensitive as the number of allowed retransmissions increases.
  • Mapping Scenarios: The authors proved that analytical assertions for the "Long-$L_c$" scenario (static channel) are also valid for the "Short-$L_c$" scenario (rapidly changing channel), providing a unified theory for different mobility speeds.

Limitations and Future Work

  • Perfect Receiver CSI: The paper assumes the receiver has perfect Channel State Information (CSI), which is an idealized assumption. Real-world receivers often have imperfect estimates of the channel.
  • Gaussian Inputs: The results are restricted to Gaussian input distributions. While theoretically important, real systems use discrete constellations (like QAM), which might behave slightly differently.
  • Feedback Complexity: The study focuses on simple 1-bit feedback (ACK/NACK). Future work could explore multi-bit feedback or "soft" feedback where the receiver provides more detail about how badly the packet failed.

Summary updated

Abstract

Assuming noisy feedback channels, this paper investigates the data transmission efficiency and robustness of different automatic repeat request (ARQ) schemes using adaptive power allocation. Considering different block-fading channel assumptions, the long-term throughput, the delay-limited throughput, the outage probability, and the feedback load of different ARQ protocols are studied. A closed-form expression for the power-limited throughput optimization problem is obtained, which is valid for different ARQ protocols and feedback channel conditions. Furthermore, the paper presents numerical investigations on the robustness of different ARQ protocols to feedback errors. It is shown that many analytical assertions about the ARQ protocols are valid both when the channel remains fixed during all retransmission rounds and when it (in)dependently changes in each round. As demonstrated, optimal power allocation is crucial for the performance of noisy ARQ schemes when the goal is to minimize the outage probability.
BibTeX
@article{makki2014noisy,
  author = {Behrooz Makki and Alexandre Graell i Amat and Thomas Eriksson},
  title = {{On Noisy ARQ in Block-Fading Channels}},
  journal = {IEEE Trans. Vehicular Technology},
  year = {2014},
  url = {https://openreview.net/forum?id=2au3OyKiFbU},
}

Loading related content...

Figure