Neural Network based Dynamic Impedance Matching for Transmitter Efficiency Enhancement in Implantable Devices

This paper presents a novel approach to improve the power-delivery efficiency of transmitters in implantable wireless transceiver integrated circuits that operate in environments of varying impedance. The proposed approach employs a built-in dynamic impedance matching strategy that is based on a deep-neural-network (DNN). It is demonstrated via simulations with a 2.4GHz CMOS differential power amplifier (DPA) designed in 22nm FDSOI technology with a power control range from −60dBm to 20dBm. Based on simulated samples from 6-ports integrated into the matching network at the output of the DPA, the DNN constructs a predictive model for the power delivered to the antenna, which is stored in the wireless device to associate future measurements at those 6 ports with actual impedance conditions at the transmitter’s output and corresponding power levels delivered to the antenna. The model is designed to support a VSWR up to 4:1 for the load, representing possible conditions within the human body. Under extreme VSWR conditions, it enhances efficiency from the low teens to the high thirties.