Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
451385 | Computer Networks | 2010 | 16 Pages |
We propose deriving wireless simulation models from experimental traces of radio signal strength. Because experimental traces have holes due to packet losses, we explore two algorithms for filling the gaps in lossy experimental traces. Using completed traces, we apply the closest-fit pattern matching (CPM) algorithm, originally designed for modeling external interference, to model signal strength.We compare the observed link behavior using our models with that of the experimental packet trace. Our approach results in more accurate packet reception ratios (PRR) than current simulation methods, reducing the absolute error in PRR by up to about 0.3 in the experiments we present. We also find that using CPM for signal strength improves simulation of packet burstiness, reducing the Kantorovich–Wasserstein (KW) distance of conditional packet delivery functions (CPDFs) by a factor of about three for intermediate links. Our model reduces the factor of error in the number of parent changes in the standard TinyOS collection protocol (CTP) by an order of magnitude or more as compared to a real signal power trace in two simple test scenarios. We show that our methods are robust to the sampling frequency of the learning deployment and are thus generally applicable for simulating arbitrary applications without a pre-determined packet transmission frequency.These improvements give low-power wireless network simulators a better capability to capture real-world dynamics and edge conditions that protocol designers typically must wait until deployment to detect.