A machine learning approach for predicting SINR

Ekaterina Bobrikova, Anna Platonova, Ekaterina Medvedeva, Yuliya Gaidamaka, Sergey Shorgin
The article proposes a method for assigning a modulation-code scheme by a base station scheduler, based on predicting the value of the signal-to-interference ratio on the equipment of a mobile user at the next time slot from a sequence of known values of this ratio in the past. Prediction is performed using machine learning; for this, a single-layer neural network was built and applied to solve a multi-parameter optimization problem using the stochastic gradient method. The trained neural network for the predicted value of the signal / interference ratio allows the scheduler to select the modulation-code scheme correctly, thereby ensuring the level of quality of data transmission in the radio channel required for the provision of the service.