3. Research

1. Channel-gain cartography

Channel-gain (CG) cartography relies on sensor measurements to construct maps providing the attenuation profile between arbitrary transmitter-receiver locations. Existing methods for channel-gain cartography build upon the intuitive principle that spatially close radio links exhibit similar shadowing.

State-of-the-art on this subject includes tomography-based approaches, where shadowing effects are modeled by the weighted integral of a spatial loss field (SLF) that captures the propagation environment. Conventionally, the SLF is learned via regularized least-squares (LS) methods provided that the the SLF is statistically homogeneous and e.g., modeled as a zero-mean Gaussian random field. However, these approaches are less effective when the propagation environment is spatially heterogeneous due to a combination of free space and objects in different sizes and materials (e.g., as easily seen in urban areas), which subsequently induces statistical heterogeneity in the SLF.

To account for environmental heterogeneity, the novel methods here leverage the Bayesian framework to learn the piecewise homogeneous SLF through a hidden Markov random field (MRF) model, which captures spatial correlations of neighboring regions exhibiting related statistical behavior. Efficient field estimators are derived by using approximate Bayesian inference methods such as Markov chain Monte Carlo (MCMC) or variational Bayes (VB), which are powerful tools for Bayesian inference when analytical solutions of the minimum mean-square error (MMSE) or the maximum a posteriori (MAP) estimators are not available.

To get the idea of channel-gain cartography, Figs. 2 and 3 display the SLF and channel gain map, respectively, constructed by the variational Bayes algorithms [C8] with the real dataset from [1]. Since shadowing is caused by obstructions, it is interesting to see that the underlying SLF in Fig. 2 clearly reveals structural patterns of the testbed in Fig. 1. The channel gain map in Fig. 3 is estimated given the receiver location marked with the black cross. Channel gain between every location in the area of interest and the receiver is color-coded in Fig. 3. Stronger attenuation is observed when signals propagate through either more building materials (bottom-right side of the CG map), or the concrete wall (left side of the CG map), while less attenuation along the entrance of the structure (top-right side of the CG map). This stems from the fact that free space and objects are more distinctively delineated in the SLF by the proposed method. All in all, the simulation results confirm that our approach could provide more site-specific channel-state information of the propagation medium, and thus endows the operation of cognitive radios or IoT with more accurate interference management. Note that our work can be readily applicable to radio tomography, which is imaging by sectioning through the use of penetrating wave in RF, since the SLF exactly serves such purpose.

More interesting research has been conducted along with CG cartography. For example, data-driven sensor selection methods were proposed in [J6],[C7],[C8] by cross-fertilizing the idea from the fields of experimental design and active learning. Please check following papers for details:

Fig. 1 Testbed configuration

Fig. 2 Spatial loss field

Fig. 3 CG map

Reference

[1] B. R. Hamilton, X. Ma, R. J. Baxley, and S. M. Matechik, “Propagation modeling for radio frequency tomography in wireless networks,” IEEE J. Sel. Topics Sig. Proc., vol. 8, no. 1, pp. 55–65, Feb. 2014.