Deep Learning-aided Channel Estimation For Universal Filtered Multi-carrier Systems

In this paper, a novel channel estimation technique is proposed for Universal Filtered Multi-Carrier (UFMC) systems. The proposed technique employs Deep Learning (DL) models which utilize the, commonly discarded, odd-indexed samples of the received signal in order to enhance the channel estimation. Three DL models with different sets of input features are proposed. The three proposed DL models were trained and then deployed into a UFMC system to evaluate their performance. The performance metric for the training stage is the Normalized Mean Squared Error (NMSE) between the estimated channel and actual channel coefficients. For deployment stage, both NMSE and Bit Error Rate (BER) are chosen as performance metrics. The proposed models are compared versus conventional Least Square (LS) channel estimator. The results show that the proposed DL-models outperform the LS channel estimator for various Signal to Noise Ratio (SNR) even for channel models which are different from the one used for training. The SNR gains of utilizing the proposed models are 5-6dBs and 2-3dBs on average for NMSE and BER, respectively.