# Temporal starvation in multi-channel CSMA networks: an analytical framework

arXiv: https://arxiv.org/abs/1808.03602 · HTML: https://mlsanity.com/paper/1808.03602/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 9 Aug 2018.

## Built on by (most cited first, 4 of 4)

- [Dimension-Free Rates for Natural Policy Gradient in Multi-Agent Reinforcement Learning](https://mlsanity.com/paper/2109.11692/) (2021)
- [Knowing What Not to Do: Leverage Language Model Insights for Action Space Pruning in Multi-agent Reinforcement Learning](https://mlsanity.com/paper/2405.16854/) (2024)
- [Scalable spectral representations for multi-agent reinforcement learning in network MDPs](https://mlsanity.com/paper/2410.17221/) (2024)
- [Scalable Multi-Agent Reinforcement Learning with General Utilities](https://mlsanity.com/paper/2302.07938/) (2023)

## Abstract (arXiv)

In this paper we consider a stochastic model for a frequency-agile CSMA protocol for wireless networks where multiple orthogonal frequency channels are available. Even when the possible interference on the different channels is described by different conflict graphs, we show that the network dynamics can be equivalently described as that of a single-channel CSMA algorithm on an appropriate virtual network. Our focus is on the asymptotic regime in which the network nodes try to activate aggressively in order to achieve maximum throughput. Of particular interest is the scenario where the number of available channels is not sufficient for all nodes of the network to be simultaneously active and the well-studied temporal starvation issues of the single-channel CSMA dynamics persist. For most networks we expect that a larger number of available channels should alleviate these temporal starvation issues. However, we prove that the aggregate throughput is a non-increasing function of the number of available channels. To investigate this trade-off that emerges between aggregate throughput and temporal starvation phenomena, we propose an analytical framework to study the transient dynamics of multi-channel CSMA networks by means of first hitting times. Our analysis further reveals that the mixing time of the activity process does not always correctly characterize the temporal starvation in the multi-channel scenario and often leads to pessimistic performance estimates.