Optimizing Training Signals for Millimeter Wave Power Amplifier Modeling
ECE Professor Miriam Leeser’s research on “Training Signal Optimization for Behavioral Modeling and Digital Predistortion of RF Power Amplifiers” was published in IEEE Microwave Magazine.
Abstract:
Power amplifier (PA) behavioral modeling and digital predistortion (DPD) are well-established and widely accepted processes. These processes involve selecting a model or DPD structure defined by a mathematical function and then using a combination of input and output signals along with a training algorithm to calculate the parameters of the function. The goal in both cases is to achieve accurate performance with the fewest possible parameters. An important consideration in these processes is the computational overhead and the associated time required to perform training. Besides the training algorithm itself, the size of the input and output signals used in training significantly influences the computational effort needed. This article begins with an introduction to two popular mathematical structures for PA models and predistorters. It includes a brief summary of different types of training signals and highlights in more detail one of the shortest training signals presented in the literature for RF PAs. The article provides guidance on constructing this extremely short training signal. The signal is similar to orthogonal frequency-division multiplexing-based signal standards, facilitating its use in training during the regular operation of the wireless transmitter. Experimental validation of this method is also provided in hardware, with a PA using a 100-MHz bandwidth millimeter wave 5G signal.