The Attention-Based Deep Neural Network DPD for Wideband Power Amplifier

Digital pre-distortion (DPD) is one of the most common and cost-effective processes, and through its linearization capability, it is widely used to correct the nonlinearity property of power amplifiers (PA). Different methods are used in designing the DPD, however the effectiveness of the method in terms of miniaturizing the hardware and attaining the efficiency at the same time is the most crucial thing to consider. The neural network, specifically the deep neural networks (DNN) architecture, is selected as the best method to design a DPD and to reduce in complexity of the design, the attention mechanism with DNN is used. The result analysis is made based on the adjacent current power ratio (ACPR), the error vector magnitude (EVM), and the normalized mean square error (NMSE). The comparison of the proposed method with the Volterra-based models shows a good result in terms of all performance measurements. The presented approach is evaluated using the input and output data of the GaN PA working under a 15 MHz long-term evolution (LTE) signal sampled at a sampling rate of 92.16 MHz.