Data-driven link quality prediction using link features

Abstract

As an integral part of reliable communication in wireless networks, effective link estimation is essential for routing protocols. However, due to the dynamic nature of wireless channels, accurate link quality estimation remains a challenging task. In this article, we propose 4C, a novel link estimator that applies link quality prediction along with link estimation. Our approach is data driven and consists of three steps: data collection, offline modeling, and online prediction. The data collection step involves gathering link quality data, and based on our analysis of the data, we propose a set of guidelines for the amount of data to be collected in our experimental scenarios. The modeling step includes offline prediction model training and selection. We present three prediction models that utilize different machine learning methods, namely, naive Bayes classifier, logistic regression, and artificial neural networks. Our models take a combination of PRR and the physical-layer information, that is, Received Signal Strength Indicator (RSSI), Signal-to-Noise Ratio (SNR), and Link Quality Indicator (LQI) as input, and output the success probability of delivering the next packet. From our analysis and experiments, we find that logistic regression works well among the three models with small computational cost. Finally, the third step involves the implementation of 4C, a receiver-initiated online link quality prediction module that computes the short temporal link quality. We conducted extensive experiments in the Motelab and our local indoor testbeds, as well as an outdoor deployment. Our results with single- and multiple-senders experiments show that with 4C, CTP improves the average cost of delivering a packet by 20% to 30%. In some cases, the improvement is larger than 45%.

A Novel Predictive Link Quality Metric for Mobile Ad-Hoc Networks in Urban Contexts

“…The last hardware based metric called LQI has been proposed for the 802.15.4 standard but it can only be exploited on specific radio chips. Even if LQI presents a certain correlation (better than the RSS and closer to the SNR) with the PRR [10], it does not provide a relevant estimation for intermediate link quality [13].…”

An adaptable fuzzy-based model for predicting link quality in robot networks.

“…Liu and Cerpa published two similar works that use machine learning for predicting the chances of successful packet delivery over a short-term window [24,93]. The works primarily differ in two regards.…”

“…First, the learning in [93] takes places offline, while in [24] the learning transpires online. Secondly, the prediction windows, or temporal relevance, of the estimators are slightly different: [93] is designed to output whether or not the next packet will be successful based on the binary output of a classifier, while the binary classifier output in [24] is intended to be valid for a slightly longer period by predicting whether the probability of packet delivery will be above some predefined threshold (e.g., 90%) during the next short-term window. Both algorithms are designed for short-term routing protocols that attempt to boost delivery efficiency by exploiting the correlation of packet delivery over short timeframes.…”

“…Feature engineering revealed that the attributes could be reduced to PRR combined with any one of the physical metrics with negligible differences. In [93], the authors evaluate three different classifiers (i.e., Naïve Bayes, Logistic Regression, and Neural Network) and found the logistic regression model to be the best performer. Evaluation results indicate that the model could be accurately trained using a minimum of several disparate links (i.e., roughly 5-7) with about 1000 labeled samples per link in the composite training set.…”

“…Better accuracy would likely be obtained by maintaining individual classifiers for each link and training each one using individualized training sets that were collected over its specific link. Overall, the major disadvantage of [93] is that the offline design requires on-site data collection and training to be performed prior to implementation.…”