Fast Indoor Radio Propagation Prediction Using Deep-Learning

We show a dataset composed by Radio Maps Estimation (RME) and Cells Maps Estimation (CME) for the 5GHz band WIFI in indoor scenarios: it has 60 indoor constructions plans and 1000 distributions initially for a training process and 20 indoor constructions plans and 50 distributions aditionals for a test process of access points to even construction. These distributions are random and several WLAN's structures: 1 to 5 access points.

The above explain that we got a total of 61000 RME and CME, this presents that is a model without interference between channels.

Every coverage map have like maximum power delivered is Pr = Pt = 26 dBm (according to data from current commercial equipment) and like minimum power a value noise established in Pr = KTB, where K is the Boltzmann's constant, T the enviroment temperatura equal to 290°K and B the band width equal to 80MHz.

Dataset DeepFIRP is the result of a lot of simulations by a own software developed in MATLAB that work with the IEEE 802.11ax channel model.

The pictures have a depth of 8 bits and size of 256pixels X 256pixels equivalents to indoor constructions of 20 X 20 m2. These ones make reference to offices's spaces at general or classroom. We present file CSV with the data of coverage to use, too.

A application to this dataset and the codes used for generate it is found here, where we implement a U-Net model for theRME and CME in indoor enviroments. This investigation contribute in novels methods for estimate by fast way coverage and cells maps using deep-learning in comparation with the conventional phisics methods like dominath-path model or ray-tracing. Whats allows save a lot of amount time in the WLANs's designs.