Predicting coverage in wireless local area networks with obstacles using kriging and neural networks

Abstract

In this paper, a new approach based on ordinary kriging is proposed to predict network coverage in wireless local area networks. The proposed approach aims to reduce the cost of active site surveys by estimating path loss at points where no measurement data is available using samples taken at other points. To include the effect of obstacles on the covariance among points, a distance measure is developed based on an empirical path loss model. The proposed approach is tested in a simulated wireless local area network. The results show that ordinary kriging is able to estimate path loss with acceptable error levels.

Hybrid Model and Data Driven Algorithm for Online Learning of Any-to-Any Path Loss Maps

“…The learning of radio maps has been a major topic of interest both in academia and the industry for years [2]- [12]. In recent years, the framework of tomographic projection technique (TPT) has gained a great deal of attention as a model that characterizes the long-term shadowing of links caused by objects such as buildings or trees [7], [13], [14], and in turn this shadowing is used as a proxy to characterize the path loss. In TPT, a spatial loss field (SLF) captures the absorption generated by objects in a field, while a window function models the influence of each location on the attenuation that every link experiences [13].…”

“…The SLF in [13] is assumed to be a zero-mean Gaussian random field, and consequently, the shadowing loss experienced on arbitrary links can also be seen as a Gaussian random field. The treatment of shadow fading as a Gaussian random field has led several authors [7], [10], [16] to use Kriging interpolation for the estimation of coverage maps. In [11], [16], a statespace extension of the general path loss model is adopted in order to track coverage maps using the Kriged Kalman filter.…”

“…Problem (7) can be viewed as the minimization of the empirical cost of a model-constrained least squares regression regularized with the elastic net. Note however that ( 7) is not jointly convex in f , α 1 , ..., α t , but it is convex in f if α 1 , ..., α t are fixed, and vice versa [20].…”

“…Note however that ( 7) is not jointly convex in f , α 1 , ..., α t , but it is convex in f if α 1 , ..., α t are fixed, and vice versa [20]. As a result, we consider an alternating minimization strategy to address (7) where, at time t of arrival of new measurements, one set of variables is updated while the remaining ones are kept constant. This process is carried out until a stopping criterion is met.…”

“…Some approaches for radio maps estimation are pure datadriven methods [2]- [4] in the sense that no physical model for the propagation of radio signals is considered, but instead they exploit the expected spatial correlation of the channel characteristics. Other approaches either rely on fixed mathematical models to describe the propagation of radio signals [5]- [7], or they attempt at learning such models without context information [8], [9]. Model-based methods have the advantage that they can generate reliable estimations with low computational complexity and little to no side information.…”