APEX: Automated Parameter Exploration for Low-Power Wireless Protocols

TL;DR

APEX tackles the challenge of efficient LPW protocol parametrization under costly real-world testing by learning a data-efficient surrogate of performance from limited trials using Gaussian processes. It casts parameter exploration as Bayesian optimization, using acquisition strategies such as GP-LCB and EI to select informative next trials while accounting for noisy measurements. Validation on two IEEE 802.15.4 protocols (Crystal and RPL) shows substantial reductions in required testbed trials compared with exhaustive search, greedy methods, and reinforcement learning, and provides robust and optimality confidences for the returned parameter sets. The open-source implementation and testbed integration make it practical for non-experts to tailor LPW stacks to diverse application requirements, such as minimizing energy $E_c$ subject to $PRR \,\ge\, 65\%$.

Abstract

Careful parametrization of networking protocols is crucial to maximize the performance of low-power wireless systems and ensure that stringent application requirements can be met. This is a non-trivial task involving thorough characterization on testbeds and requiring expert knowledge. Unfortunately, the community still lacks a tool to facilitate parameter exploration while minimizing the necessary experimentation time on testbeds. Such a tool would be invaluable, as exhaustive parameter searches can be time-prohibitive or unfeasible given the limited availability of testbeds, whereas non-exhaustive unguided searches rarely deliver satisfactory results. In this paper, we present APEX, a framework enabling an automated and informed parameter exploration for low-power wireless protocols and allowing to converge to an optimal parameter set within a limited number of testbed trials. We design APEX using Gaussian processes to effectively handle noisy experimental data and estimate the optimality of a certain parameter combination. After developing a prototype of APEX, we demonstrate its effectiveness by parametrizing two IEEE 802.15.4 protocols for a wide range of application requirements. Our results show that APEX can return the best parameter set with up to 10.6x, 4.5x and 3.25x less testbed trials than traditional solutions based on exhaustive search, greedy approaches, and reinforcement learning, respectively.

Figures (5)

• Figure 1: Example of parameter exploration and modeling of protocol performance using Crystal. Energy consumption experimentally measured through testbed trials using six initial random parameter sets (a); polynomial regression model fitted to these initial experimental results (b); energy consumption and respective fit after an exhaustive search of all sixteen parameter sets (c).
• Figure 2: Crystal's performance when executing ten experiments per parameter set. The best parameter set (brown arrow) is different from that observed in Fig. \ref{['challenge_all']}(c).
• Figure 3: High-level overview of the APEX framework.
• Figure 4: Top: GP fit and regression after six testbed trials ($n=6$) for a single parameter ($b=1$), where $x$ represents the values of that parameter. Bottom: updated GP fit and regression after a new trial ($n=7$).
• Figure 5: Illustration of how APEX derives optimality. We compute an exponential fit of the cumulative sub-optimality $T(n)$, and derive the angle $\theta_n$ of the fit's tangent after the $n$-th testbed trial. A higher $\theta_n$ (e.g., for n=36) indicates a low optimality. After additional testbed trials (e.g., n=66), a low $\theta_n$ indicates closeness to optimality.