Online machine learning algorithms to predict link quality in community wireless mesh networks

Bote‐Lorenzo, Miguel L.; Gómez‐Sánchez, Eduardo; Mediavilla-Pastor, Carlos; Asensio-Pérez, Juan I.

doi:10.1016/j.comnet.2018.01.005

Cited by 29 publications

(59 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, the work also presents an optimized prediction method that considers the knowledge about the expected LQ values (saturation). Bote-Lorenzo et al [23] identified some limitations and problems of predictive models generated using batch machine learning algorithms in [48]: they cannot be updated once their training has been completed; they impose a predeployment effort (a set of quality samples must be collected in each link before building the corresponding prediction models); and predictions are not available until the observation period is over (which might take a long time). To address these problems, they propose a new hybrid online algorithm for LQ prediction.…”

Section: Link and End-to-end Quality Prediction In Heterogeneous Netwmentioning

confidence: 99%

“…The evaluation of the proposed algorithm shows that it can achieve a high accuracy while generating a low computational load. The analysis performed by Bote-Lorenzo et al [23] includes a comparison with a baseline (ETX last value). The best overall performance is achieved by both the hybrid online algorithm and the SVM batch algorithm, which outperforms the baseline performance.…”

Section: Link and End-to-end Quality Prediction In Heterogeneous Netwmentioning

confidence: 99%

See 1 more Smart Citation

Time Series Analysis to Predict End-to-End Quality of Wireless Community Networks

et al. 2019

View full text Add to dashboard Cite

Community Networks have been around us for decades being initially deployed in the USA and Europe. They were designed by individuals to provide open and free “do it yourself” Internet access to other individuals in the same community and geographic area. In recent years, they have evolved as a viable solution to provide Internet access in developing countries and rural areas. Their social impact is measurable, as the community is provided with the right and opportunity of communication. Community networks combine wired and wireless links, and the nature of the wireless medium is unreliable. This poses several challenges to the routing protocol. For instance, Link-State routing protocols deal with End-to-End Quality tracking to select paths that maximize the delivery rate and minimize traffic congestion. In this work, we focused on End-to-End Quality prediction by means of time-series analysis to foresee which paths are more likely to change their quality. We show that it is possible to accurately predict End-to-End Quality with a small Mean Absolute Error in the routing layer of large-scale, distributed, and decentralized networks. In particular, we analyzed the path ETX behavior and properties to better identify the best prediction algorithm. We also analyzed the End-to-End Quality prediction accuracy some steps ahead in the future, as well as its dependency on the hour of the day. Besides, we quantified the computational cost of the prediction. Finally, we evaluated the impact of the usage for routing of our approach versus a simplified OLSR (ETX + Dijkstra) on an overloaded network.

show abstract

Section: Link and End-to-end Quality Prediction In Heterogeneous Netwmentioning

confidence: 99%

Section: Link and End-to-end Quality Prediction In Heterogeneous Netwmentioning

confidence: 99%

Time Series Analysis to Predict End-to-End Quality of Wireless Community Networks

et al. 2019

View full text Add to dashboard Cite

show abstract

“…This contributes to maximize the message delivery rate and minimize traffic congestion at both levels (i.e., LQ and EtEQ) with a small average mean absolute error. Additionally, Bote-Lorenzo et al [23] identified several problems and limitations of the predictive models used in [20], through the use of batch machine learning algorithms. Similarly, Lowrance and Lauf [24] provide an exhaustive study of the existing prediction methods to estimate link quality, considering online learning algorithms that increasingly update the network model.…”

Section: Introductionmentioning

confidence: 99%

“…However, the use of predictors increases the complexity of the routing protocols, because of the additional hardware and software required to make and validate predictions. Moreover, penalty mechanisms are usually introduced to the system when there is a high rate of mispredictions, which negatively affect the performance of these protocols.Prediction mechanisms have been embedded in routing protocols to foresee several aspects of a network, such as nodes mobility [13], reliability of its topology [14,15] and quality of links and end-to-end paths [16][17][18][19][20][21][22][23][24]. These mechanisms have also been used to reach particular communication…”

mentioning

confidence: 99%

Predicting Topology Propagation Messages in Mobile Ad Hoc Networks: The Value of History

Millan

Aliagas

Molina

et al. 2019

Sensors

View full text Add to dashboard Cite

The mobile ad hoc communication in highly dynamic scenarios, like urban evacuations or search-and-rescue processes, plays a key role in coordinating the activities performed by the participants. Particularly, counting on message routing enhances the communication capability among these actors. Given the high dynamism of these networks and their low bandwidth, having mechanisms to predict the network topology offers several potential advantages; e.g., to reduce the number of topology propagation messages delivered through the network, the consumption of resources in the nodes and the amount of redundant retransmissions. Most strategies reported in the literature to perform these predictions are limited to support high mobility, consume a large amount of resources or require training. In order to contribute towards addressing that challenge, this paper presents a history-based predictor (HBP), which is a prediction strategy based on the assumption that some topological changes in these networks have happened before in the past, therefore, the predictor can take advantage of these patterns following a simple and low-cost approach. The article extends a previous proposal of the authors and evaluates its impact in highly mobile scenarios through the implementation of a real predictor for the optimized link state routing (OLSR) protocol. The use of this predictor, named OLSR-HBP, shows a reduction of 40-55% of topology propagation messages compared to the regular OLSR protocol. Moreover, the use of this predictor has a low cost in terms of CPU and memory consumption, and it can also be used with other routing protocols.Sensors 2020, 20, 24 2 of 24 important number of topology propagation (TP) messages to keep the delivery of regular messages under control. The exchange of TP messages helps reduce the probability of overflowing the network with regular messages that do not reach the destination, rendering the communication system useless at supporting the coordination of the emergency response activities. Provided that many of these communication systems involve mobile ad hoc networks, the routing protocols used to support interactions among nodes (i.e., the first responders) must be simple, efficient and reliable and have the capability of quickly adapting themselves to changes in the network topology [7-10]. Thus, these protocols intend to maximize the reachability of the target nodes, by consuming as little energy from the network as possible.Regardless of the extensive research done in the area, addressing this communication challenge, considering the previous constraints, is still an open issue. Some recent proposals try to deal with this challenge using deep learning in static or quasi-static networks [11]. Although deep learning techniques have shown positive results, these approaches are quite limited when used in mobile scenarios, like emergency responses. Moreover, they require a previous training stage to generate the network model. Novel solutions are therefore required to reduce the number of TP messages ex...

show abstract

“…Their approach would require decomposing measurements of the SINR into a time-varying component and a non-stationary random part. The authors in [48] proposed a hybrid online machine learning algorithm to estimate the quality of candidate links. Their approach combines current samples of link quality with baseline samples previously learned from past samples.…”

mentioning

confidence: 99%

Improving Route Selections in ZigBee Wireless Sensor Networks

Meka

Fonseca

2019

Sensors

View full text Add to dashboard Cite

The ZigBee wireless communication specifications forecast the use of multihop routes between nodes and define that nodes select their routes based on their costs. The specifications define how to compute a route cost from the probability of successfully transmitting on each of the routes’ links; and it is recommended that such probabilities be obtained by counting received link status messages or averaging link quality indicators from received packets. In this paper, we study the performance of these two recommended procedures, show that they can lead to degraded route selections, and propose a procedure that can improve route selections without modifications to the ZigBee protocol or frame formats. Our procedure estimates the probability of successful transmission on each link, based on information from the medium access layer during unicast packet transmissions, and includes a modification into how ZigBee nodes treat routing messages internally in order to reduce variations in the link cost estimates. Focusing on a home environment with one or two hops, our simulation results show that, in several scenarios, our procedure performs better than either of the two procedures recommended in the ZigBee specifications.

show abstract

Online machine learning algorithms to predict link quality in community wireless mesh networks

Cited by 29 publications

References 30 publications

Time Series Analysis to Predict End-to-End Quality of Wireless Community Networks

Time Series Analysis to Predict End-to-End Quality of Wireless Community Networks

Predicting Topology Propagation Messages in Mobile Ad Hoc Networks: The Value of History

Improving Route Selections in ZigBee Wireless Sensor Networks

Contact Info

Product

Resources

About