Comparative Analysis of LSTM and GRU for Uplink Data Rate Prediction in 5G Networks

This paper presents investigations on uplink data rate prediction in 5G Non-Standalone (NSA) networks using 4G and 5G parameters that are available from the modem, with a use case in Intelligent Transportation Systems (ITS). Prediction of uplink performance for these systems is challenging as it is affected by many aspects such as location, travel speed, weather conditions, and the traffic load in the cell during the time of data collection and prediction. Here we have used variants of recurrent neural networks (RNN) called long short-term memory (LSTM) and gated recurrent unit (GRU) which frequently outperform other algorithms in this area. The experimentation is done using various lags to predict the different future values, and the performance is measured using root mean squared error (RMSE) and mean squared logarithmic error (MSLE). Three real-time datasets are used for training, validation, and testing; the data is trained on various epochs 5, 10, 20, and lags 5, 10 by doing different combinations of them to predict 1 and 3 future values. Both algorithms have performed well using the RSRQ (4G) and NR-RSRQ (5G) metrics. Predicting the uplink data rate will help the ITS devices anticipate the strength of the signal and switch to different networks preemptively to prevent disconnection.