Gradient Descent for Fast Perturbation and Observation in Low-Power Energy Harvesting

Energy harvesters can be used to power IoT sensors. Maximum power tracking of energy harvesters allows more energy to be extracted to meet higher power requirements. However, both the open-circuit voltage method and the perturbation and observation method, which were commonly adopted for energy harvesters with ac output, have disadvantages. The former method cannot achieve conjugate impedance matching, and the latter method is slow and inefficient. Therefore, it is quite challenging to achieve conjugate impedance matching with high speed, high matching efficiency, and low power consumption for milliwatt applications. In this article, a fast perturbation and observation method based on gradient descent is proposed, which optimizes the perturbation directions and perturbation steps to find the maximum power point quickly and with low perturbation oscillation. Low-power implementation using intermittent analog–digital hybrid control is also designed. The proposed approach can find the optimal impedance within 4 s and the power matching efficiency is 97.71%, measured at resonant frequency. Compared to resistive matching, the proposed method can increase the output power of the selected electromagnetic vibration energy harvester by up to 41.2% when deviating from the resonant frequency. The overall efficiency is 90.08% at an output voltage of 2.3 V and 73.80% at an output voltage of 5.3 V.