On the Two-User Interference Channel With Lack of Knowledge of the Interference Codebook at One Receiver

Alex Dytso, Daniela Tuninetti, Natasha Devroye

TL;DR

This paper investigates the capacity of a two-user interference channel where one receiver has incomplete knowledge of the interfering transmitter's codebook. The authors show that even with this partial knowledge, the achievable rates can be remarkably close to the full codebook scenario, particularly for Gaussian channels, and in some cases, the performance loss is negligible. This work has implications for designing more practical and robust wireless communication systems.

What Problem Does This Paper Address?

In traditional information theory, it's often assumed that all nodes in a network have complete knowledge of all codebooks used. This assumption, however, is impractical in real-world scenarios like cognitive radio or large ad-hoc networks where nodes might not have access to all necessary information. This paper focuses on a specific scenario: a two-user interference channel (IC) where one receiver is "oblivious" to the interfering transmitter's codebook. The core question is: how much does this lack of codebook knowledge reduce the channel's capacity, and can we still achieve near-optimal performance?

Methods

The authors develop novel outer bounds for the interference channel with an oblivious receiver (IC-OR). They introduce the concept of "mixed inputs" – a combination of continuous (like Gaussian) and discrete distributions – which prove to be more effective than purely Gaussian inputs in certain scenarios. They analyze various types of IC-ORs, including linear deterministic approximations (LDA) and the practically relevant Gaussian noise IC-OR. Key techniques include deriving achievable rate regions using modified Han-Kobayashi schemes and analyzing the generalized degrees of freedom (gDoF) to understand performance at high signal-to-noise ratios (SNR).

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Partial codebook knowledge has a surprisingly small impact: Even with one receiver being oblivious to the interfering codebook, the achievable rates can be very close to the ideal case where both receivers have full knowledge.
  • Mixed inputs outperform Gaussian inputs: For certain interference channel models, especially at high SNRs, using inputs that combine discrete and continuous distributions (mixed inputs) yields better performance than purely Gaussian inputs. This is a significant departure from standard assumptions in Gaussian channel analysis.
  • The "oblivious" nature is manageable: The inability of the oblivious receiver to jointly decode messages is mitigated by its ability to estimate and remove the interfering signal, suggesting that the mapping from codewords to messages is less critical than the ability to suppress interference.
  • gDoF analysis reveals no loss: In the high SNR regime, the generalized degrees of freedom (gDoF) for the IC-OR are identical to those of the classical IC, implying no loss in fundamental information transmission capability.

Key Results

  • An outer bound for the IC-OR is derived, which is shown to be tight to within a constant gap for various regimes.
  • For the Gaussian noise IC-OR, the capacity is shown to be within approximately 3.34 bits per channel use of the classical IC capacity.
  • For the LDA-IC-OR, non-i.i.d. Bernoulli inputs are shown to outperform i.i.d. Bernoulli inputs in achieving optimal rates.
  • In the very strong interference regime for the Gaussian IC-OR, the gDoF region is identical to that of the classical G-IC.

Limitations and Future Work

The paper acknowledges that finding the exact optimal input distributions for the derived outer bounds remains an open problem, particularly for the Gaussian IC-OR. A significant future direction is to generalize the problem to scenarios where both receivers lack knowledge of the interfering codebook. This generalization is noted as highly non-trivial, especially when both users employ discrete inputs, due to the complexity of analyzing sums of discrete sets. The authors also point to ongoing work on discrete inputs for Gaussian channels.

Summary updated

BibTeX
@article{dytso2015two,
  author = {Alex Dytso and Daniela Tuninetti and Natasha Devroye},
  title = {{On the Two-User Interference Channel With Lack of Knowledge of the Interference Codebook at One Receiver}},
  journal = {IEEE Trans. Information Theory},
  year = {2015},
  doi = {10.1109/tit.2015.2388481},
}
Figure