Achievable Second-Order Asymptotics for MAC and RAC with Additive Non-Gaussian Noise

Yiming Wang, Lin Bai, Zhuangfei Wu, Lin Zhou

TL;DR

This paper provides a rigorous theoretical analysis of multi-user communication systems (MAC and RAC) operating under arbitrary, non-Gaussian noise in the finite blocklength regime. The authors prove that Joint Nearest Neighbor (JNN) decoding consistently outperforms the more computationally efficient Successive Interference Cancellation (SIC) in terms of achievable second-order rates, particularly in random access scenarios where user activity is unknown.

What Problem Does This Paper Address?

In classical information theory, "capacity" is defined as the maximum transmission rate possible when the message length (blocklength) approaches infinity. However, modern applications like 5G/6G and the Internet of Things (IoT) require ultra-reliable low-latency communication (URLLC), where blocklengths are short. In this "finite blocklength" regime, there is a significant gap between the theoretical capacity and what is actually achievable.

Furthermore, most existing research assumes that channel noise follows a Gaussian distribution. In practice, interference can be non-Gaussian, and receivers often use "mismatched" decoders (decoders optimized for Gaussian noise but applied to other noise types). This paper addresses the challenge of characterizing the second-order asymptotics—essentially the "speed" at which a system approaches its capacity as blocklength increases—for Multiple Access Channels (MAC) and Random Access Channels (RAC) under these realistic, non-Gaussian conditions.

Methods

The authors employ several sophisticated mathematical tools to derive new achievability bounds:

  • Spherical Codebooks: Codewords are generated uniformly on the surface of a high-dimensional sphere, which is known to be robust for mismatched decoding.
  • Decoding Strategies: They compare Joint Nearest Neighbor (JNN) decoding, which looks at all users' signals simultaneously, against Successive Interference Cancellation (SIC), which decodes users one by one.
  • Refined Bounds: They utilize a modified Random Coding Union (RCU) bound and the Function Version of the Berry-Esseen Theorem to approximate the error probability for multi-dimensional rate regions.
  • Rateless Coding for RAC: For the Random Access Channel, where the number of active users is unknown, they implement a rateless scheme with one-bit feedback to allow the receiver to terminate transmission once it estimates the user activity pattern.

Datasets

As a theoretical information theory paper, it does not use empirical datasets. Instead, the authors validate their mathematical derivations through:

  • Numerical Simulations: They plot the derived second-order rate regions (using parameters like $P_1=5, P_2=2, \epsilon=0.2$) to visualize the performance gap between JNN and SIC.
  • Theoretical Comparisons: They show that when their results are specialized to Gaussian noise, they perfectly recover existing benchmarks in the literature (e.g., results from MolavianJazi & Laneman, 2015).

Key Insights and Ideas

  • JNN vs. SIC in MAC: While JNN and SIC share the same first-order rate region (they have the same capacity), JNN is strictly superior in the second-order. This means for short, real-world messages, JNN can support higher data rates for the same reliability level.
  • The RAC Penalty: In Random Access Channels (where the receiver doesn't know who is talking), JNN is superior even in the first-order. SIC performs poorly here because all users must share the same codebook, and the decoding of the first user is heavily penalized by interference from others.
  • Robustness of Mismatched Decoding: The paper demonstrates that a decoder designed for Gaussian noise is surprisingly effective against arbitrary noise, provided the noise has finite moments (specifically 2nd, 4th, and 6th moments). The performance depends on the noise's fourth moment ($\xi$), which characterizes its "peakedness" or "flatness" compared to a Gaussian.
  • Unified Formula: For JNN decoding, the authors provide a unified second-order asymptotic formula that covers all boundary cases of the rate region, simplifying previous piecewise definitions.

Key Results

  • MAC Performance: Numerical results show that for a finite blocklength, the JNN achievable rate region is visibly larger than the SIC region. For example, in a two-user scenario, JNN allows for a significantly smaller "penalty" term as blocklength $n$ increases.
  • RAC Performance: In random access scenarios, JNN decoding achieves a strictly larger first-order asymptotic rate than SIC. As the number of active users $k$ increases, the achievable rate for both drops, but SIC drops much faster (see Fig 7a).
  • Asymptotic Recovery: The derived bounds are shown to be "tight" in the sense that they recover the state-of-the-art third-order results for Gaussian noise up to the second-order term.

Limitations and Future Work

  • Achievability Only: The paper focuses on achievability (proving what is possible). It does not provide a matching converse (proving what is impossible), which is necessary to claim the results are the absolute "capacity" of the finite blocklength regime.
  • Static Channels: The model assumes additive noise but does not account for channel fading or multi-antenna (MIMO) complexities, which are common in wireless environments.
  • Third-Order Terms: While the paper masters second-order asymptotics, deriving the even more precise "third-order" terms for non-Gaussian noise remains an open challenge due to the difficulty of bounding specific error events ($E_{1,2}$) in non-Gaussian settings.

Summary updated

Abstract

We first study the two-user additive noise multiple access channel (MAC) where the noise distribution is arbitrary. For such a MAC, we use spherical codebooks and either joint nearest neighbor (JNN) or successive interference cancellation (SIC) decoding. Under both decoding methods, we derive second-order achievable rate regions and compare the finite blocklength performance between JNN and SIC decoding. Our results indicate that although the first-order rate regions of JNN and SIC decoding are identical, JNN decoding has better second-order asymptotic performance. When specialized to the Gaussian noise, we provide an alternative achievability proof to the result by MolavianJazi and Laneman (T-IT, 2015). Furthermore, we generalize our results to the random access channel (RAC) where neither the transmitters nor the receiver knows the user activity pattern. We use spherical-type codebooks and a rateless transmission scheme combining JNN/SIC decoding, and derive second-order achievability bounds. Comparing second-order achievability results of JNN and SIC decoding in a RAC, we show that JNN decoding achieves strictly larger first-order asymptotic rate. When specialized to Gaussian noise, our second-order asymptotic results recover the corresponding results of Yavas, Kostina, and Effros (T-IT, 2021) up to second-order.
BibTeX
@misc{wang2024achievable,
  author = {Yiming Wang and Lin Bai and Zhuangfei Wu and Lin Zhou},
  title = {{Achievable Second-Order Asymptotics for MAC and RAC with Additive Non-Gaussian Noise}},
  year = {2024},
  eprint = {2410.10312},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2410.10312},
}
Figure