GRU-Attention Model for Linearizing Millimeter-Wave Transmitters in Vehicle to Satellite Communication Systems

Recently, with the increasing deployment of vehicle-to-everything (V2X) communication systems, wireless transmitters have been extensively deployed in vehicles. And to enable high-capacity and high-speed communications everywhere in the world, the communication from vehicle to satellite (V2S) has emerged, particularly utilizing high throughput satellite (HTS) technology. However, the nonlinearity of the millimeter wave transmitter (mmWT) in a vehicle ends affects the quality of the V2S uplink transmitted signal gravely. Digital predistortion (DPD) linearization has shown promise in mitigating the nonlinearity of mmWTs. In addition, due to the powerful and complex feature fitting ability, neural networks have been used to nonlinear behavioral model and DPD linearize for broadband wireless communication systems. In this work, a gated recurrent unit (GRU) with attention mechanism (GRU-attention) model is proposed for modeling and linearizing mmWTs in V2S communication systems. The attention mechanism in the GRU-attention model calculates a distribution matrix during the training process, which is utilized to weigh the GRU-attention model parameters, thereby enhancing the modeling accuracy. The linearization performance of the GRU-attention is experimentally evaluated by using a Ka-band 29 GHz mmWT with a 100 MHz bandwidth and 64 quadrature amplitude modulation (64-QAM) 5G NR signal. The experimental results demonstrate that the GRU-attention improves the adjacent channel power ratio of the mmWT by 18 dB, which is 4 dB better than that of the conventional GRU model.