A Novel CNN-based Architecture for Over-the-Air 5G OFDM Channel Estimation

In this work, a novel deep learning (DL) architecture is introduced to deal with the channel estimation in over-the-air (OTA) fifth-generation (5G) orthogonal frequency-division multiplexing (OFDM) scenarios. Considering the scarcity of practical DL-based channel estimation works, a realistic indoor non-line-of-sight (nLoS) scenario is explored through the adoption of an OTA wireless setup. Furthermore, a novel convolutional neural network (CNN) architecture is proposed combining parallel connections of convolutional layers to capture richer variety of channel features and patterns. The proposed architecture has a decrease of $5 \times$ trainable parameters when compared to state-of-the-art DL-based models, achieving a reasonable signal-to-noise ratio (SNR) gain of up to 0.1 dBs in practical scenarios.