Estimation of signal in a 5G system for internal calibration using the adam learning algorithm

Abstract

For the purpose of the military, surveillance, and science, many devices are being digitalized recently, so the demand for 5G and radar technology is increasing. Stability is important for general RF systems such as radar. However, RF systems usually need active elements, for example, LNA (Low Noise Amplifier), PA (Power Amplifier), which has an exothermic reaction problem. These elements tend to have a thermal drift, which is performance change over temperature. If operating frequency rises, this thermal problem becomes worse. To compensate this unwanted drift phenomenon, an internal calibration network is typically installed, and the performance of the transmitter and receiver can be extracted from the signal through the calibration path. In the previous study, the gradient descent algorithm with momentum term was used. Nevertheless, this approach is likely to be inefficient when parameters oscillate in the process of learning. In addition, the previous study used CW (Continous Wave) for calibration, but there are two problems. One is the ambiguity of choosing standard peaks, and the other one is that group delay and phase difference look the same. Therefore, CW wave calibration can yield the wrong compensation. To substitute the previous study, the Adam learning algorithm was proposed to accelerate the learning process. Additionally, the proposed algorithm entails chirp signals which are original form in chirp-pulse radar. Eventually, this thesis shows the advancement in learning speed and compensation of the pure phase difference.