Weighted Sum-Rate Maximization with Causal Inference for Latent Interference Estimation

Lei You

TL;DR

This paper introduces SC-WMMSE, a novel approach that uses causal inference to improve wireless network optimization. By estimating the "what-if" scenarios of interference, SC-WMMSE significantly boosts network performance and convergence speed compared to existing methods, especially in dynamic environments.

What Problem Does This Paper Address?

Wireless networks are becoming increasingly complex, with a growing number of devices and interference sources. A key challenge is optimizing network performance, specifically maximizing the weighted sum-rate (WSR), in scenarios where some interference sources are "latent" – meaning their behavior (power allocation) is unknown and uncontrollable. Traditional optimization algorithms, like the Weighted Minimum Mean Square Error (WMMSE), struggle because they rely on observed interference, ignoring the potential impact of these hidden sources. This can lead to suboptimal decisions and slower convergence, especially when the latent interference changes dynamically.

Methods

The paper's core contribution is extending the WMMSE algorithm by incorporating causal inference, specifically the Synthetic Control (SC) method.

  • WMMSE Baseline: The standard WMMSE algorithm is a well-established method for solving the WSRM problem. It reformulates the problem into a weighted sum-mean-square-error minimization.
  • The Challenge of Latent Interference: When latent interference sources (from network K⁻) exist, their unknown power allocation (denoted by 'q') makes the interference term (ηk) unpredictable. This uncertainty can lead the WMMSE algorithm to make decisions based on incomplete information, ignoring the true causal impact of latent interference.
  • Causal Inference with Synthetic Control (SC): The authors propose using SC to estimate the counterfactual interference. This means estimating what the interference would have been if the latent interference had behaved differently, or if the known network's power allocation had been different.
  • Training: The SC method is trained offline using historical data of interference observed across different links. It learns to construct a "synthetic" interference profile for a given link by creating a convex combination of interference observed on other links. This trained model acts as a proxy for the counterfactual.
  • Inference/Optimization: During the WMMSE optimization iterations, the SC-trained estimator is used to predict the counterfactual interference. The algorithm then updates its decisions by considering both the observed interference and this estimated counterfactual. This allows the algorithm to make more robust decisions, accounting for the potential impact of latent interference.
  • SC-WMMSE: This is the proposed algorithm that integrates the SC method into the WMMSE framework. It requires no additional information beyond what the original WMMSE uses.

Datasets

Not discussed in the paper. The paper focuses on a simulated wireless network environment.

Key Insights and Ideas

  • Causality Matters in Optimization: The paper highlights that simply observing interference isn't enough; understanding the causal relationship between network actions and interference is crucial for optimal performance, especially in dynamic environments.
  • Synthetic Control as a Powerful Counterfactual Estimator: SC proves to be an effective tool for estimating what would have happened in the absence of certain factors (like predictable latent interference), allowing for more informed optimization decisions.
  • Dynamic Interference is a Major Hurdle: The performance gains of SC-WMMSE are most pronounced when latent interference is dense or changes frequently, underscoring the limitations of static models in real-world wireless scenarios.
  • Convexity Constraint is Key: The paper demonstrates that enforcing a convexity constraint in the SC training phase is vital for accurate counterfactual estimation and overall algorithm performance, preventing overfitting to observed data.

Key Results

  • Superior Performance: SC-WMMSE consistently outperforms the original WMMSE and a "local" WMMSE (which ignores latent interference) in terms of both objective function maximization (sum-rate) and convergence speed.
  • Robustness to Dynamic Environments: The algorithm shows significant resilience to emerging and disappearing latent interference sources, adapting effectively to changing network conditions.
  • Impact of Latent Interference Density: The performance advantage of SC-WMMSE over baselines increases with the density of latent interference sources. When latent interference is sparse, the benefits are less pronounced, as the term ηk can be well approximated by noise.
  • Scalability: The algorithm scales well with the number of links in the known network, showing improved performance up to a certain threshold.

Limitations and Future Work

  • Theoretical Guarantees: The authors suggest further analysis and theoretical proofs of the algorithm's convergence are needed.
  • Inter-Unit Interference in Causal Models: Current causal inference research often assumes no inter-unit interference, which is not realistic for wireless communications. Future work should focus on developing causal factor models better suited for this domain.
  • Optimization of SC Group Selection: While SC is effective, optimizing the selection of the "untreated" groups for the synthetic control is noted as an area for further exploration.
  • Interplay of Intervention and Optimization: The paper points out that power allocation serves dual roles as both an intervention for causal inference and an optimization variable. Understanding how to best leverage this interplay is an open research question.

Summary updated

Abstract

The paper investigates the weighted sum-rate maximization (WSRM) problem with latent interfering sources outside the known network, whose power allocation policy is hidden from and uncontrollable to optimization. The paper extends the famous alternate optimization algorithm weighted minimum mean square error (WMMSE) [1] under a causal inference framework to tackle with WSRM. Specifically, with the possibility of power policy shifting in the hidden network, computing an iterating direction based only on the observed interference inherently implies that counterfactual is ignored in decision making. A method called synthetic control (SC) is used to estimate the counterfactual. For any link in the known network, SC constructs a convex combination of the interference on other links and uses it as an estimate for the counterfactual. Power iteration in the proposed SC-WMMSE is performed taking into account both the observed interference and its counterfactual. SC-WMMSE requires no more information than the original WMMSE in the optimization stage. To our best knowledge, this is the first paper explores the potential of SC in assisting mathematical optimization in addressing classic wireless optimization problems. Numerical results suggest the superiority of the SC-WMMSE over the original in both convergence and objective.
BibTeX
@inproceedings{you2023weighted,
  author = {Lei You},
  title = {{Weighted Sum-Rate Maximization with Causal Inference for Latent Interference Estimation}},
  booktitle = {ICC},
  year = {2023},
  url = {https://openreview.net/forum?id=OWdXOzUr3x},
}
Figure