Tomotake Sasaki

Digital dead-time control for two phase double-ended forward converter

Two phase double-ended forward converter is suitable for more than 2kW high efficiency power supply by the feature of low surge voltage. One issue of the converter is dead-time loss of synchronous rectifier FETs. In this paper, we propose a precise digital dead-time control method for a 2.5 kW (12 V / 208 A) output power supply. Proposed method is confirmed by experiment and theoretical analysis that the method reduces the 6 W loss at half load.

Rapid control prototyping for server power supply with high-resolution PWM

In this paper, we report the first rapid control prototyping (RCP) system with a sub-nanosecond (ns) high-resolution Pulse Width Modulation (PWM) for server power supply circuits. Conventional RCP systems are designed mainly for engine and motor control. Based on the relevant response time and control accuracy requirements, their PWM update periods and resolutions are over 100 microseconds (μs) and 100 ns, respectively. The shortest update time reported thus far is 12.5 μs. However, for server power supply, the required PWM update period and resolution are less than 10 μs and sub-ns, respectively. Therefore, we developed a new interface board with high-speed and high-resolution PWM output, which is connected between an under-development power supply board and a commercially available RCP machine to increase the RCP systems speed and resolution. Moreover, we have developed new software that improves data transfer latency and maintains a constant PWM update period. The combination of the new interface board, high-performance RCP machine, and new software realizes the required short PWM update period and high-resolution PWM of 6.5 μs and 0.15 ns, respectively. This is the shortest PWM update period reported yet, and it is adequate for operating server power supply. We applied the new RCP system to a 500 W full bridge converter with 100 kHz feedback control. Frequency response and soft start functions are easily confirmed by the new RCP system.

Policy gradient reinforcement learning method for discrete-time linear quadratic regulation problem using estimated state value function

In this paper, we propose a policy gradient reinforcement learning method which directly estimates the gradient of the state value function (V-function) with respect to a feedback coefficient matrix using measurable data and uses it for policy improvement. The proposed method can be applicable to the case where the state-action value function (Q-function) is difficult to estimate, and can update the policy in an effective direction for reducing the accumulated cost.

Gain controlled high efficiency power factor correction circuit

The purpose of this study is to improve the efficiency of a boost-type power factor correction (PFC) circuit with a digital controller. One of the biggest losses in PFC circuits is the switching loss. Switching loss can be reduced by low boost ratio operation, but these results in a greater voltage drop at transient response. The voltage drop can be suppressed by using a high gain setting for the voltage control loop, but this lowers the power factor, which creates a harmonic distortion of the AC power line. To improve this trade-off, we developed a gain control method that adjusts the gain to high only at the moment of transient response. This method improves both the efficiency and the power factor of PFC circuits. We implemented the proposed method using a digital controller and demonstrated a power factor improvement of 0.08 points and an efficiency improvement of 0.4 points for a 2.5 kW PFC circuit.

Derivation of integrated state equation for combined outputs-inputs vector of discrete-time linear time-invariant system and its application to reinforcement learning

For a discrete-time linear time-invariant partially observable system that satisfies the well-known algebraic condition of observability, we derive a fully observable system equation using an augmented state vector combining output and input sequences. The derived representation enables application of reinforcement learning methods for fully observable linear quadratic regulation problems to partially observable ones avoiding the problems invoked in existing methods.

State-dependent virtual hierarchization of batteries for realizing a glocal control in energy network systems and its application to peak power reduction in office

We discuss a model predictive, hierarchical control for energy network systems including batteries. We propose a three-layered, state-dependent virtual hierarchical structure for the batteries in a system and an algorithm for controlling batteries to realize a glocal control in the field of energy network with energy storage devices. Computational experiments are contained to show the effectiveness of our algorithm.