Shannon Meets Nash on the Interference Channel

The interference channel is the simplest communication scenario where multiple autonomous users compete for shared resources. We combine game theory and information theory to define a notion of a Nash equilibrium region of the interference channel. The notion is game theoretic: it captures the selfish behavior of each user as they compete. The notion is also information theoretic: it allows each user to use arbitrary communication strategies as it optimizes its own performance. We give an exact characterization of the Nash equilibrium region of the two-user linear deterministic interference channel and an approximate characterization of the Nash equilibrium region of the two-user Gaussian interference channel to within 1 bit/s/Hz..

💡 Research Summary

The paper “Shannon Meets Nash on the Interference Channel” tackles the fundamental problem of characterizing the performance limits of a two‑user interference channel (IC) when the users act selfishly rather than cooperatively. Traditional information‑theoretic analysis assumes that all transmitters jointly optimize their coding and decoding strategies to achieve the capacity region, an assumption that is often unrealistic in wireless networks where each user cares only about its own throughput. To address this, the authors introduce a game‑theoretic framework that treats each user’s entire transmission scheme—message length, block length, codebook, encoder, and any shared randomness—as a single strategy.

In an ε‑reliable game, a strategy pair is considered valid only if the bit‑error probability for each user does not exceed ε; the payoff for a user is then simply the achieved rate. A Nash equilibrium (NE) is a strategy pair where no user can unilaterally switch to another strategy and obtain a higher payoff. Because finding exact NE in the full information‑theoretic setting is notoriously hard, the authors relax the definition to an η‑approximate NE (η‑NE): a unilateral deviation must improve a user’s payoff by more than η to be considered profitable. Since η can be made arbitrarily small, this definition captures the essence of NE while remaining mathematically tractable.

The first major contribution is a precise definition of the “information‑theoretic Nash equilibrium region” C_NE, the closure of all rate pairs that can be achieved by some (1‑ε)‑reliable η‑NE for arbitrarily small ε and η. By construction, C_NE is a subset of the classical capacity region C, but not every point in C is an equilibrium.

To obtain concrete results, the authors study the linear deterministic interference channel, a simplified model introduced by Avestimehr, Diggavi, and Tse that captures the essence of Gaussian channels in the high‑SNR regime. In this model, each transmitted signal is represented as a binary vector of “levels”; noise truncates low‑significance bits, and superposition at the receivers is performed modulo‑2 on each level. This representation eliminates inter‑level coupling and makes the analysis transparent.

For the deterministic IC the paper provides an exact characterization of C_NE. Remarkably, every point on the sum‑rate boundary of the deterministic capacity region is a Nash equilibrium. In symmetric settings (identical direct and cross gains), the symmetric capacity point is always an NE. The authors also construct explicit coding schemes for each NE point: one user may transmit its full message while the other treats interference as noise or decodes a portion of the interfering signal, depending on the channel parameters.

The second major contribution extends these insights to the real Gaussian IC. Leveraging the known fact that the deterministic model approximates the Gaussian channel to within one bit per user, the authors map the deterministic NE structure onto the Gaussian setting. They employ a Han‑Kobayashi style message splitting (common and private parts) together with a mixture of “treat interference as noise” and “decode interference” strategies. By carefully selecting the power split and decoding order, they prove that the Gaussian NE region can be bounded within 1 bit/s/Hz of the deterministic NE region, and consequently within 1 bit/s/Hz of the true Gaussian capacity region. This result parallels the celebrated 1‑bit capacity approximation for the cooperative Gaussian IC, but now applies to the non‑cooperative equilibrium scenario.

Key insights from the work include:

1. NE can be efficient – Contrary to the common belief that selfish behavior leads to highly sub‑optimal outcomes, the paper shows that in many interference regimes the Nash equilibrium coincides with the sum‑rate optimal point.
2. Full strategy freedom – By allowing arbitrary coding and decoding schemes (instead of restricting to Gaussian codebooks and fixed decoders), the equilibrium concept truly reflects the information‑theoretic limits.
3. η‑NE as a practical tool – The introduction of an arbitrarily small slack η circumvents the need for exact optimal codes, making the equilibrium definition robust and applicable to real systems.
4. Bridging deterministic and Gaussian models – The deterministic analysis provides a clean, exact picture of the equilibrium structure, which can then be transferred to the Gaussian case with a provable 1‑bit loss.

The paper concludes by suggesting extensions to multi‑user (>2) interference networks, time‑varying channels, limited channel state information, and the design of low‑complexity coding schemes that achieve the equilibrium rates in practice. Overall, the work offers a rigorous, unified framework that blends game theory and information theory to understand how selfish users can coexist efficiently on shared wireless media.