Microwave structure behavior prediction enables the estimation of circuit response over a frequency range, playing a crucial role in the design of radio frequency (RF) structures. Deep neural network (DNN) approaches have demonstrated their capability to simulate microwave structure behaviors. Nonetheless, the quality and utility of the model are constrained by the availability of data and computational capabilities. These inherent disadvantages hinder the extensive application of DNN in microwave structure behavior prediction. Transfer learning has recently been produced as a method offering improved accuracy and speed for predicting microwave circuit behavior. This paper proposes a novel transfer learning-based model to expedite the prediction process for a sequence of frequency samples. Through experimental validation, it is illustrated that the proposed methodology outperforms the conventional DNN techniques for microwave structure behavior prediction by effectively reducing the required data and shortening the training time. The proposed model also facilitates the fine-tuning of hyperparameters and reduces the simulator computing load.

In this work, we present a digital power amplifier (DPA) and a signal-optimized control technique suitable for the amplification of orthogonal frequency division multiplexing (OFDM). OFDM is a high peak-to-average power ratio (PAPR) signal that is naturally arising from high spectrum efficiency modulations. A 3-bit DPA is implemented with three power-scaled transistors which are turned on and off based on the signal amplitude, while phase modulation is restored using the RF input signal. The unavoidable non-linearities at the DPA output due to the PA switching are minimized by accounting for the OFDM signal probability density function (PDF). This PDF is a priori knowledge to design an optimal quantizer that minimizes distortion by distributing the DPA power levels where the signal amplitude is more similar to the original one. Back-off efficiency within 2^3=8 possible states is then optimized by implementing a load-modulating power combiner. Theory and an example design of the combiner network are provided and demonstrated for this DPA. The reported DPA prototype operates at 1.5 GHz with a 3-bit control and achieves a maximum power-added efficiency (PAE) of 64.3% and maintains a drain efficiency greater than 47% over the output power range from 36.6 to 45.2dBm (8.6dB range).

Microwave structures behavior prediction is an important research topic in radio frequency (RF) design. In recent years, deep-learning-based techniques have been widely implemented to study microwaves, and they are envisaged to revolutionize this arduous and time-consuming work. However, empirical data collection and neural network training are two significant challenges of applying deep learning techniques to practical RF modeling and design problems. To this end, this letter investigates a transfer-learning-based approach to improve the accuracy and efficiency of predicting microwave structure behaviors. Through experimental comparisons, we validate that the proposed approach can reduce the amount of data required for training while shortening the neural network training time for the behavior prediction of microwave structures.