Accelerator High-Voltage Power Supply Ripple Feedforward Control
In particle accelerators, synchrotron light sources, and ion implantation systems, the quality and stability of the particle beam are directly tied to the precision of the accelerating fields. For DC accelerators like Van de Graaff generators or Cockcroft-Walton multipliers, and for the injector stages of larger machines, high-voltage DC power supplies provide the fundamental accelerating potential. A key performance metric is the output voltage ripple, as any periodic fluctuation modulates the beam energy, leading to increased energy spread, reduced beam brightness, and in circular machines, potential resonance effects that can cause beam loss. Traditional feedback control alone struggles to suppress low-frequency ripple caused by line frequency harmonics or switching converter artifacts, due to finite loop gain at those frequencies. Ripple feedforward control is an advanced technique that dramatically improves suppression by actively injecting a compensatory signal derived from a direct measurement of the disturbance.
The principle is based on disturbance rejection. The ripple component on the output voltage is measured with a high-fidelity sensor, typically a precision resistive divider paired with a low-noise differential amplifier. This measured ripple signal, which contains the amplitude and phase information of the disturbance, is then processed by a feedforward controller. The controller's job is to generate a correction signal that, when added to the main control loop's command, will cause the power supply's output stage to produce an opposing voltage ripple that exactly cancels the original disturbance at the output. The critical challenge is matching the amplitude and, more importantly, the phase of the correction signal across the frequency range of interest.
The system architecture involves parallel paths. The main feedback loop operates as usual: a reference voltage is compared to a scaled-down version of the output, and the error is processed by a compensation network (PID-type) to drive the regulator. In parallel, the ripple feedforward path taps the raw output ripple. This signal is first filtered to isolate the target frequency bands—commonly the line frequency (50/60 Hz) and its first few harmonics, as well as the primary switching frequency of the supply's inverter. A tunable gain and phase adjustment circuit then processes each targeted frequency component. Analog implementations use all-pass filters and variable-gain amplifiers to manually tune the phase and amplitude. Modern digital implementations sample the ripple, perform a real-time Fourier transform or employ adaptive filter algorithms (like the Least Mean Squares algorithm) to continuously identify the ripple characteristics and synthesize the perfect anti-phase cancellation signal through a high-resolution DAC.
The correction signal is summed into the control loop at a point with sufficient bandwidth and headroom to act upon it. This is typically at the input of the PWM modulator in a switching supply or the drive input of the series-pass element in a linear regulator. The summation must be done carefully to avoid destabilizing the main feedback loop. The feedforward path is inherently an open-loop correction within a larger closed-loop system; its stability is not governed by the same criteria, but its injection point must be chosen to ensure the overall system phase margin is not degraded.
Successful implementation requires precise system identification. The transfer function from the injection point of the correction signal to the final output voltage (where ripple is measured) must be characterized. This includes the dynamics of the output stage, the output filter, and the feedback sensor. Any phase shift through this path must be accounted for in the feedforward controller's phase adjustment. This is often done during commissioning using a network analyzer or by injecting a pilot signal and observing the response. In digital systems, this identification can be automated.
The benefits are substantial. Ripple feedforward can achieve 20-40 dB of additional attenuation at specific frequencies compared to feedback alone, enabling power supplies to meet the sub-10 PPM ripple specifications required for advanced accelerators. It improves the supply's dynamic response to periodic load changes as well. However, its effectiveness is limited to predictable, periodic disturbances. Non-periodic noise or drift is still the domain of the feedback loop. Furthermore, the system must be designed to adapt to changes, such as component aging or load impedance variations, which can alter the plant transfer function. Adaptive digital controllers address this by continuously updating their filter coefficients. The integration of ripple feedforward control represents a sophisticated fusion of power electronics and signal processing, transforming a high-voltage power supply from a mere energy source into a precision instrument capable of sustaining the exquisite beam quality demanded by next-generation experimental and industrial facilities.
