Stochastic Resource Optimization of Random Access for Transmitters With Correlated Activation

Ce Zheng, Malcolm Egan, Laurent Clavier, Anders E. Kalør, Petar PopovskiIEEE Communications Letters 2021 · DOI

TL;DR

This paper tackles the challenge of optimizing resource allocation in event-triggered wireless networks where device activity is correlated. They introduce novel stochastic optimization algorithms that efficiently find near-optimal strategies for maximizing network throughput and sum-rate, outperforming existing methods by up to 30% in simulations.

What Problem Does This Paper Address?

In modern wireless systems, particularly for IoT devices and edge computing, communication is often "event-triggered." This means devices transmit only when something significant happens (e.g., a sensor detects an event), rather than at fixed intervals. A key challenge arises when the activation of these devices is correlated. For instance, if a sensor network detects an earthquake, devices near the epicenter are more likely to be activated simultaneously. This correlation makes traditional random access schemes, which assume independent device activity, inefficient. Optimizing resource allocation (like assigning time slots for transmission) in such correlated scenarios is complex because the underlying optimization problem becomes non-convex and requires estimating a large number of parameters, often exponentially many in the number of devices. Existing methods either rely on approximations or require extensive parameter estimation, limiting their practicality.

Methods

The authors propose a novel approach using stochastic optimization algorithms. Instead of trying to solve the complex, non-convex optimization problem directly, they frame it as a stochastic optimization problem. This allows them to use stochastic approximation methods, which are well-suited for situations where the exact objective function or its gradient is unknown or difficult to compute.

Specifically, they develop two main algorithms:

  1. SGA Algorithm for Throughput Maximization: This algorithm aims to maximize the expected fraction of time slots where exactly one device transmits (maximizing "throughput"). It uses an unbiased estimate of the gradient of the throughput function to iteratively update the allocation matrix.
  2. SGA Algorithm for Sum-Rate Maximization: This algorithm focuses on maximizing the expected sum-rate, which is a more comprehensive metric that accounts for the ability to decode multiple signals using successive interference cancellation. Similar to the throughput algorithm, it employs stochastic approximation to update the allocation matrix.

A key advantage of their approach is that it avoids estimating a large number of parameters related to the device activity distribution. Instead, the algorithms only require realizations of device activity and signal power, making them more scalable and practical.

Datasets

Not discussed in the paper. The paper uses simulated data based on a model described in reference [4] for evaluation.

Key Insights and Ideas

  • Correlated activity is a significant challenge: The paper clearly demonstrates that ignoring device correlation leads to suboptimal performance in random access systems.
  • Stochastic optimization is a powerful tool: By framing the problem as a stochastic optimization task, the authors bypass the need for complex analytical solutions or extensive parameter estimation, enabling more practical algorithms.
  • Simplicity in parameter estimation: The proposed algorithms require significantly fewer estimated parameters compared to prior methods, making them more scalable for networks with many devices.
  • Initial allocation matters: The simulations show that the performance of the stochastic algorithms can be sensitive to the initial allocation matrix, highlighting the presence of local optima and the importance of a good starting point.
  • Sum-rate optimization is more robust: Maximizing the expected sum-rate appears to be a more effective objective for achieving significant performance gains, especially in scenarios with high correlation.

Key Results

  • The proposed SGA algorithm for throughput maximization can achieve approximately double the throughput compared to existing algorithms in [4] for a moderate number of slots.
  • The SGA algorithm for sum-rate maximization can yield an improvement of over 30% in the expected sum-rate, particularly in scenarios with a small number of available time slots relative to the number of devices and high device activity correlation.
  • The algorithms are shown to converge with probability one to a local maximum of the respective optimization problems.

Limitations and Future Work

  • Local Optima: The algorithms are guaranteed to converge to local optima, not necessarily the global optimum. The paper acknowledges that the performance can be sensitive to the initial allocation, suggesting that further research into initialization strategies or global optimization techniques might be beneficial.
  • Assumptions: The paper assumes a specific transmission protocol (Algorithm 1) with error-free multiuser detection and rate quantization. Real-world scenarios might deviate from these assumptions.
  • Future Work: The authors suggest that their algorithms can be extended to more complex scenarios, such as those involving multiple subcarriers or more sophisticated interference management techniques. Further investigation into robust initialization strategies for the stochastic algorithms is also implied.

Summary updated

BibTeX
@article{zheng2021stochastic,
  author = {Ce Zheng and Malcolm Egan and Laurent Clavier and Anders E. Kalør and Petar Popovski},
  title = {{Stochastic Resource Optimization of Random Access for Transmitters With Correlated Activation}},
  journal = {IEEE Communications Letters},
  year = {2021},
  doi = {10.1109/lcomm.2021.3090110},
}
Figure