Abstract. Empirical mode decomposition (EMD) based hybrid prediction methods can be an efficient way to allocate resources for ultra reliable low latency communication (URLLC). In this thesis, we have considered efficient resource allocation for the downlink channel at the presence of several interferers. Initially, we have generated desired signal that we need to transmit in downlink and total interference signal that will affect the desired signal transmission. Then, we used EMD to decompose the total interference signal power into intrinsic mode functions (IMFs) and residual. Due to the properties of EMD, decomposed IMFs become less random as IMF number increases. As a result of that property, prediction model training process become less complex and prediction accuracy also increases as randomness of signal decreases. Long short term memory (LSTM) deep neural network method and auto regressive integrated moving average (ARIMA) time series method are deployed to predict future interference power values based on historical values. For each decomposed component (IMFs and residual), two prediction models have been trained using LSTM and ARIMA methods. Finally, predicted components of IMFs and residual are added together to form total predicted interference power.

According to the predicted interference power, resources are allocated for downlink transmission of the signal and evaluated it with the baseline estimation techniques. The research demonstrates that the suggested method achieves near optimal resource allocation for URLLC.