On discrete alphabets for the two-user Gaussian interference channel with one receiver lacking knowledge of the interfering codebook

Alex Dytso, Daniela Tuninetti, Natasha DevroyeIta 2014 · DOI

TL;DR

This paper explores the impact of partial codebook knowledge in a two-user Gaussian interference channel. They show that even when one receiver doesn't know the interfering user's codebook, discrete input signals can achieve near-optimal performance, especially in strong interference regimes, challenging the assumption that full codebook knowledge is always necessary for high rates.

What Problem Does This Paper Address?

In many information theory scenarios, it's assumed that every node in a network knows the codebooks used by all other nodes. This is often impractical in real-world distributed or cognitive radio networks where nodes might join or leave dynamically, or where new systems need to coexist with older ones without sharing all their internal workings. This paper tackles the question of how much communication performance (specifically, channel capacity) degrades when one receiver in a two-user Gaussian interference channel (G-IC) is "oblivious" – meaning it doesn't know the codebook of the interfering transmitter. They investigate whether this lack of knowledge fundamentally limits the achievable rates compared to the classical G-IC where all codebooks are known.

Methods

The core of the paper lies in developing new theoretical tools and applying them to the G-IC-OR (Gaussian Interference Channel with Oblivious Receiver).

  • Novel Lower Bound for Discrete Inputs: A key contribution is a new, computable lower bound on the mutual information achievable by a discrete input on a point-to-point Gaussian channel. This bound is crucial for analyzing the performance of discrete signals.
  • Discrete Input Design: They show how to choose the cardinality (number of distinct symbols) of a discrete input alphabet as a function of the signal-to-noise ratio (SNR) to achieve performance close to the theoretical capacity of a point-to-point Gaussian channel.
  • Achievable Rate Region for G-IC-OR: Using these tools, they derive an achievable rate region for the G-IC-OR, specifically considering a scenario where one transmitter uses a discrete input and the other uses a Gaussian input.
  • Generalized Degrees of Freedom (gDoF) Analysis: The paper analyzes the performance in terms of gDoF, a measure of how capacity scales with SNR in the high-SNR regime. This allows for a comparison of the G-IC-OR with the classical G-IC.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and deriving achievable rate regions and performance bounds, rather than empirical evaluation on specific datasets.

Key Insights and Ideas

  • Discrete Inputs Can Be "Good" for Both Signal and Interference: Surprisingly, discrete inputs, when carefully chosen, can be effective not only for transmitting the desired signal but also for acting as "good" interference (i.e., interference that doesn't overly degrade the desired signal's reception). This contrasts with the common notion that Gaussian inputs are always optimal for channels.
  • Obliviousness Doesn't Always Hurt in Strong Interference: In strong and very strong interference regimes, the lack of codebook knowledge at one receiver does not reduce the generalized degrees of freedom (gDoF) compared to the classical G-IC. This is a significant finding, suggesting that practical network designs might not need to share all codebooks in such scenarios.
  • Near-Capacity Performance with Discrete Inputs: By appropriately selecting the number of discrete symbols based on SNR, the achievable rate can be made to approach the channel capacity within an additive gap of $O(\log(\log(\text{SNR})))$. This demonstrates that discrete signaling can be remarkably efficient.
  • Sum-Capacity Gap is Small: The sum-capacity of the G-IC-OR is shown to be within $O(\log(\log(\text{SNR})))$ of the classical G-IC's sum-capacity, especially in strong interference. This refines the gDoF results and highlights the resilience of communication systems to partial codebook knowledge.

Key Results

  • gDoF Achievability: The paper demonstrates that the G-IC-OR can achieve gDoF that approach the classical G-IC's gDoF to any desired accuracy in strong and very strong interference.
  • Sum-Capacity Bound: The sum-capacity of the G-IC-OR is shown to be within $O(\log(\log(\text{SNR})))$ of the classical G-IC's sum-capacity in strong interference.
  • Specific Rate Bounds: Achievable rate regions are derived, showing specific bounds for $R_1$ and $R_2$ (rates for user 1 and user 2) and their sum, $R_1 + R_2$. For instance, in very strong interference, achievable rates are given by expressions involving $N$ (number of discrete points) and SNR.

Limitations and Future Work

  • Symmetric Channel Assumption: While the paper mentions results can be extended to asymmetric cases, the primary analysis often focuses on symmetric G-IC-OR for simplicity.
  • Discrete Input Optimization: The paper focuses on specific types of discrete inputs (e.g., PAM). Further work could explore a wider range of discrete constellations.
  • Generalization to More Users/Complex Networks: The current work is for a two-user G-IC. Extending these findings to more complex network topologies with multiple users and different types of partial codebook knowledge would be a natural next step.
  • Practical Implementation: While theoretically sound, the practical implementation of these discrete signaling schemes and the precise selection of $N$ as a function of SNR would require further investigation.

Summary updated

BibTeX
@inproceedings{dytso2014discrete,
  author = {Alex Dytso and Daniela Tuninetti and Natasha Devroye},
  title = {{On discrete alphabets for the two-user Gaussian interference channel with one receiver lacking knowledge of the interfering codebook}},
  booktitle = {Ita},
  year = {2014},
  doi = {10.1109/ita.2014.6804258},
}
Figure