Abstract
In decentralized wireless networks, receivers may not always possess full knowledge of the codebooks used by interfering transmitters. This paper explores the two-user Interference Channel where one receiver is "oblivious" to the interferer's codebook (IC-OR). The authors prove that this lack of knowledge does not reduce the generalized Degrees of Freedom (gDoF) and leads to a capacity gap of only a few bits compared to a fully informed system.
Introduction
Classical multi-user information theory typically assumes that every node in a network knows the encoding schemes (codebooks) of every other node. This "full codebook knowledge" allows for sophisticated joint decoding techniques to mitigate interference. However, in ad-hoc or cognitive radio environments, new devices may appear using unknown protocols. This paper addresses the fundamental limits of communication when one receiver lacks knowledge of the interferer's codebook.
The IC-OR System Model
The authors formalize this problem as the Interference Channel with one Oblivious Receiver (IC-OR). In this model, two transmitters send messages to their respective receivers. While one receiver is fully informed of both codebooks, the other—the "oblivious" receiver—only knows its own.
As shown in the system model, Decoder 1 receives both codebook indices ($F_1, F_2$), whereas Decoder 2 only receives its own ($F_2$). The central question is how this asymmetry impacts the overall capacity and the achievable rates for both users.
Methodology and Insights
The research utilizes a combination of information-theoretic outer bounds and achievable inner bounds. A key technical contribution is the use of "mixed" input distributions, combining discrete Pulse Amplitude Modulation (PAM) with continuous Gaussian noise.
A profound insight from the work is the distinction between demodulation and decoding. Even without a codebook (which maps bits to codewords), a receiver can perform symbol-by-symbol estimation (demodulation) if the interference has a sufficiently discrete structure. This allows the receiver to estimate and subtract interfering symbols, effectively mitigating noise without full decoding. Interestingly, standard Gaussian inputs—usually optimal for noisy channels—perform poorly in this scenario because their lack of structure makes them indistinguishable from thermal noise to an oblivious receiver.
Key Results
The paper provides several significant findings:
- Generalized Degrees of Freedom (gDoF): At high Signal-to-Noise Ratios (SNR), the lack of codebook knowledge results in zero loss in terms of gDoF. The "cost" of being oblivious is a constant number of bits that does not scale with power.
- Deterministic Channels: For injective fully deterministic channels, the authors established the exact capacity, showing a zero gap.
- Gaussian Channels: For real-valued Gaussian IC-OR, the capacity is shown to be within 0.5 bits per channel use of the derived outer bound. In the symmetric case, the gap between the oblivious system and a fully informed one is at most 3.34 bits per channel use per user.
Conclusion and Future Work
This work demonstrates that partial codebook knowledge is not a fundamental bottleneck for high-SNR communication in interference channels. Future research directions include extending the analysis to "two-sided" oblivious problems where both receivers lack the other's codebook—a task that involves complex additive combinatorics—and identifying the truly optimal input distributions for the IC-OR.
Summary updated