Map Based RNN Model for Proactive Prediction of Received Power Distribution in an Indoor Area

This paper presents a proactive prediction method for received power distribution using GRU (Gated Recurrent Unit), which is one type of RNN (Recurrent Neural Network), as deep learning. In addition to the 50 most recent RSSI (Received Signal Strength Indicator) data acquired approximately every 0.1 seconds as input data, the distance and LoS/NLoS between transmitter and receiver at the prediction target position were used as a map data. As an output data, the median, 5%, and 95% values of RSSI after 5 seconds were predicted. The output data was derived using 50 points (about 5 seconds) of RSSI data. We used RSSI data of 5.6GHz band wireless LAN measured in an indoor environment for training data and validation data. With the proposed method, the RMSE (Root Mean Squared Error) for the validation data is approximately 1.4 dB, improving prediction accuracy by 1.4 dB and 0.6 dB for prediction using the current observed value and prediction using only the latest RSSI, respectively.