Prediction of the Received Power of Low-Power Networks Using Inertial Sensors

TL;DR

The paper addresses unstable wireless links for buoy-based low-power IoT sensors on rough water, proposing to predict the received power $P_{rx}$ from 3D motion using an MMSE framework. To fit resource-constrained devices, it replaces matrix inversion with gradient-descent parameter estimation, achieving about 91% prediction accuracy on real deployments. The approach is evaluated against exact MMSE and Kalman-filter baselines, showing competitive performance with practical gains for adaptive transmission power. This motion-informed RSSI prediction advances link-quality estimation in dynamic aquatic environments, with synchronization between motion data and RSSI identified as a key area for refinement.

Abstract

Low-power and cost-effective IoT sensing nodes enable scalable monitoring of different environments. Some of these environments impose rough and extreme operating conditions, requiring continuous adaptation and reconfiguration of physical and link layer parameters. In this paper, we closely investigate the stability of the wireless links established between nodes deployed on the surface of different water bodies and propose a model to predict the received power. Our model is based on Minimum Mean Square Estimation (MMSE) and relies on the statistics of received power and the motion the nodes experience during communication. One of the drawbacks of MMSE is its reliance on matrix inversion, which is at once computationally expensive and difficult to implement with resource constrained devices. We forgo this stage by estimating model parameters using the gradient-descent approach, which is much simpler to implement. The model achieves a prediction accuracy of 91% even with a small number of iterations.

Figures (12)

• Figure 1: Illustration of dynamic power adaptation. A transmitter aims to transmit a packet with a power such that when received, the packet has a received power above a set threshold.
• Figure 2: The deployment of waterproof buoys on the surface of different water bodies. From left to right: A prototype low-power IoT node; deployments on: a lake on FIU's Modesto A. Maidique Campus, Miami South Beach, and Crandon Beach.
• Figure 3: Relationship between two random variables.
• Figure 4: Histograms depicting the change in the RSSI of received packets and the change in the linear acceleration of the transmitting node. The nodes were using the CC1200 radio and the channel had a center frequency of 869.3 MHz and a bandwidth of 2 MHz.
• Figure 5: Histograms depicting the change in the RSSI of received packets and the change in the linear acceleration of the transmitting node. The nodes were using the CC2538 radio (2.4 GHz).