Akademik Veri Yönetim Sistemi
Engineering Science and Technology, an International Journal, cilt.70, 2025 (SCI-Expanded, Scopus)
This paper presents a detailed performance evaluation on the use of various deep neural networks for channel estimation in orthogonal frequency division multiplexing with index modulation (OFDM-IM) systems. The channel estimation performances of the models previously employed for OFDM in the literature are analysed in OFDM-IM. In the context of OFDM systems, the placement of pilot symbols within the process of channel estimation demonstrates a degree of flexibility. Conversely, within the framework of OFDM-IM, the placement of pilot symbols is constrained to the position subsequent to each sub-block, given that not all sub-carriers are active. Consequently, the process of channel estimation in OFDM-IM can be more arduous than in OFDM. In this work, a new channel estimation model is also introduced, which combines Long Short Term Memory (LSTM) and residual network (ReEsNet) for OFDM-IM systems, and which is termed LSTM-ReEsNet. A comparative analysis is conducted between available models and the introduced LSTM-ReEsNet model, in terms of training time, computation time, computational complexity, mean squared error (MSE) and bit error rate (BER). Furthermore, the paper presents BER analyses for different power delay profiles (PDP) in order to measure the generalisation capabilities of the employed models. The findings demonstrate that the proposed LSTM-ReEsNet model exhibits improved in performance when compared to traditional methods and comparable performance with other deep learning (DL) models. Notably, training time is significantly reduced with the proposed LSTM-ReEsNet model.