Contact Us

Modeling and Autotuning of AVP Control with Inductor DCR Current Sensing

Fig. 1. Proposed Model (a) Block diagram, (b) Zoc under mismatch.
The inductor DCR sensing technique is widely used due to lossless sensing and simple implementation. However, mismatch between the time constant of the DCR sensor (Cs) and the inductor (Ls) increases the output voltage spike during load transient, but there is no analytical model to understand the dynamic. Although adding more output capacitors can avoid a transient spike, component cost and the area on motherboard also increase. The prior autotuning method solves the mismatch issue by adjusting Cs close to Ls such that the output capacitors can be reduced, but the implementation is too complex for monolithically integration.

This paper first presents a new equivalent circuit model to predict the small-signal characteristic of the current-mode control with DCR current sensing accurately. The model contains an ReLsCe resonant circuit and a current source to represent the mismatch effect at high frequency, while a RsCs low-pass filter loop represents the low-frequency effect. The model shows that the closed-loop output impedance (Zoc) contains a pole-zero pair from the DCR sensor which makes Zoc away from the desired load-line resistance (RLL), so the transient response impairs.

Secondly, this paper develops a simple pole-zero compensation technique to maintain a constant output impedance and a fast transient under possible mismatch conditions, so the output capacitor can be minimized with a simpler autotuning circuit on the VR controller. The idea is adding a simple and easy-integrated compensator (HCom) along the current feed-back path to create an adjustable pole-zero pair to correct the output impedance close to RLL.

The proposed autotuning configuration in Fig. 2 contains Hcom with a high-pass filter (RC) and a variable gain amplifier (A), and a Vo slope detector. When a step-up load transient event is identified at Vi', HCom is adjusted gradually based on sensed dVo/dt, because Zoc can be estimated by sensing dVo/dt. When ZocLL, the slow response makes dVo/dt <0. When Zoc>RLL, dVo/dt>0. The SIMPLIS simulation with the state-of-art output filter model of laptop VR and the experiment results on a commercial VR12 controller confirms the effectiveness of the proposed method.

Fig. 2. Block diagram of proposed autotuning configuration.