Predictive Estimation of Wireless Link Performance from Medium Physical Parameters Using Support Vector Regression and k-Nearest Neighbors - Traffic Monitoring and Analysis Access content directly
Conference Papers Year : 2014

Predictive Estimation of Wireless Link Performance from Medium Physical Parameters Using Support Vector Regression and k-Nearest Neighbors

Abstract

In wireless networks, the physical medium is the cause of most of the errors and performance drops. Thus, an efficient predictive estimation of wireless networks performance w.r.t. medium status by the communication peers would be a leap ahead in the improvement of wireless communication. For that purpose, we designed a measurement bench that allows us to accurately control the noise level on an unidirectional WIFI communication link in the protected environment of an anechoic room. This way, we generated different medium conditions and collected several measurements for various PHY layer parameters on that link. Using the collected data, we analyzed the ability to predictively estimate the throughput performance of a noisy wireless link from measured physical medium parameters, using machine learning (ML) algorithms. For this purpose, we chose two different classes of ML algorithms, namely SVR (Support Vector Regression) [1] and k-NN (k-Nearest Neighbors) [2] to study the tradoff between complexity and estimation accuracy. Finally, we ranked the pertinence of the most common physical parameters for estimating or predicting the throughput that can be expected by users on top of the IP layer over a WIFI link.
Fichier principal
Vignette du fichier
978-3-642-54999-1_7_Chapter.pdf (305.82 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-01396474 , version 1 (14-11-2016)

Licence

Attribution

Identifiers

Cite

Guillaume Kremer, Philippe Owezarski, Pascal Berthou, German Capdehourat. Predictive Estimation of Wireless Link Performance from Medium Physical Parameters Using Support Vector Regression and k-Nearest Neighbors. 6th International Workshop on Traffic Monitoring and Analysis (TMA), Apr 2014, London, United Kingdom. pp.78-90, ⟨10.1007/978-3-642-54999-1_7⟩. ⟨hal-01396474⟩
179 View
66 Download

Altmetric

Share

Gmail Facebook X LinkedIn More