A small filter can work—if the inverter’s timing is tuned
An inverter prototype cut the product of its inductor and capacitor by 83.3% while keeping voltage and nominal current crossover targets comparable. That matters because these components are the physical filter that smooths a power converter’s output; shrinking them can make grid-supporting equipment more compact. But the result came with a condition: the controller and filter have to be designed together, because digital delay can turn a source of damping into a destabilizing force.
The study examines a grid-forming inverter, a power converter designed to behave more like a voltage source than a device that merely follows the grid. Such inverters are increasingly important as wind, solar and storage connect to electricity networks. They can help establish voltage and frequency, but they must also remain stable when their own controls interact with the filter and with the wider grid.
Obi and Jung show that, in one widely relevant control arrangement, simply increasing the current controller’s gain can make that stability problem worse. They then show how a modest low-pass filter in one controller pathway can recover a useful operating range—without weakening another feedback path that helps regulate voltage. On a 3-kVA laboratory prototype, the approach enabled a substantially smaller filter product, and a separate test with the same high-resonance filter ran stably at 2.68 times the unshaped controller’s nominal current crossover (Obi and Jung, 2026).
The practical promise is not that every inverter can now use an 83% smaller filter. The result is a design option: engineers may be able to trade controller settings against passive-component size, while checking stability under the particular grid and operating conditions they face.
The Science
An inverter connected to the grid needs a filter between its switching electronics and the network. In this study, that filter is an LC pair: an inductor and a capacitor. Together, they smooth the rapid voltage changes produced by switching. Their values also set a natural oscillation frequency, called the LC resonance. Raising that resonance can permit a smaller inductor-capacitor product, although the filter still has to meet requirements for harmonic suppression and component stress.
The authors study a grid-forming inverter using virtual-admittance control. In plain terms, the controller makes the inverter respond to voltage differences as though it were connected through a chosen virtual resistance and inductance. An inner current loop and an outer voltage loop work together to produce the desired voltage-source behavior. Their analysis concentrates on the inner loops, whose behavior is fast enough to be studied separately from the slower power-control loops.
A key ingredient is voltage-feedback decoupling, or VFD. It feeds measured capacitor voltage into the current controller to compensate for coupling inside the filter. The paper considers unity VFD, meaning that this feedback is not attenuated, alongside active damping from capacitor-voltage feedback. Active damping is a control-generated effect that suppresses unwanted oscillations, rather than a physical resistor that dissipates energy.
The complication is timing. A digital controller measures, computes and updates the inverter command in discrete steps. That creates delay. Feedback that would damp an oscillation in an ideal, instantaneous system can arrive late enough to push it in the opposite direction. Under the paper’s single-sampled, single-update pulse-width modulation scheme, the combined delay is 1.5 sampling periods. At a sampling frequency of 10 kHz, the delay makes inverter-current feedback’s equivalent resistance turn negative above one-sixth of the sampling frequency—about 1.67 kHz. Equivalent resistance here is a way to describe whether a feedback path adds damping (positive) or tends to feed oscillation (negative).
The paper first analyzes the feedback paths as equivalent impedances at the filter capacitor. It then checks its simplified design criteria against a more complete model that retains the virtual-admittance loop, resonant controller, damping path and digital delay. The authors also test the design on a 3-kVA prototype rated at 110 V line-to-line, with a 10-kHz sampling and switching frequency. These are laboratory results, not a demonstration across a utility-scale fleet or every kind of grid connection.
What They Found
The central result is a gain trade-off. Increasing proportional gain in the current loop usually makes that loop respond faster. But in the resonance band, stronger gain also makes delayed current feedback more influential. Above the delay-related frequency boundary, that path acts against damping supplied by unity VFD. The combined range of frequencies with positive equivalent resistance—the positive-virtual-resistance region—then contracts. In practical terms, a filter resonance that was stable at a lower current gain can become unstable when the gain is raised.
The authors address this by putting a first-order low-pass filter on the proportional part of the current controller. A low-pass filter reduces the strength of a signal’s faster-changing components. Here it softens the delayed proportional feedback near the LC resonance, while the controller’s resonant branch remains unchanged and unity VFD is retained. The shaping therefore targets the path that creates the high-frequency damping problem, rather than broadly weakening all feedback.
For the representative filter, the reduced model finds a stable effective-damping-gain window bounded by approximately 0.126 and 1.36. The lower boundary reflects too little damping support; the upper boundary reflects excessive damping gain destabilizing a resonance-band mode. The corresponding positive-gain crossings occur at about 342 Hz and 1.61 kHz. These values describe the study’s representative design, not universal settings for other inverters.
Resonance-placement limits across studied gains
Reported physical LC-resonance limit endpoints across the studied current-gain range. The low and high values are ranges across conditions, not matched point-by-point comparisons.
| Label | Value |
|---|---|
| Unshaped VFD–AD, lower reported limit | 1.36 kHz |
| Unshaped VFD–AD, upper reported limit | 2.67 kHz |
| Shaped control, lower reported limit | 3.17 kHz |
| Shaped control, upper reported limit | 3.53 kHz |
Across the range of current-gain conditions examined, the unshaped VFD–AD design’s resonance-placement limit fell from 2.67 kHz to 1.36 kHz. With proportional-path shaping, the reported limits were between 3.17 and 3.53 kHz. The baseline virtual-admittance/current-control configuration without the same VFD–AD support remained limited to approximately one-sixth of the sampling frequency. At the paper’s 10-kHz sampling rate, that boundary is about 1.67 kHz. The authors find that the gain window eventually closes as physical LC resonance rises; shaping pushes that closure to a higher resonance in the cases studied.
The experimental results offer two different demonstrations of design flexibility. First, the prototype achieved an 83.3% reduction in LC product at comparable voltage and nominal current crossovers. The LC product is the product of the filter’s inductance and capacitance, and it is inversely related to the square of its resonant frequency. A reduction does not mean each component shrank by 83.3%; the paper reports the product, not identical percentage reductions in both parts.
Second, with the same high-resonance filter, the shaped controller demonstrated stable operation at 2.68 times the unshaped nominal current crossover. This is a separate comparison from the smaller-filter test: it shows that shaping can also create room to raise current-loop bandwidth without changing that filter. The paper’s abstract reports unloaded tests supporting the damping-gain limits, and grid-connected tests supporting additional constraints that depend on grid strength.
The method’s underlying idea is visible in the impedance analysis. Voltage-feedback decoupling contributes positive equivalent resistance over a wider frequency span than the delayed current-feedback path does. But the current loop and capacitor-voltage damping path can increasingly oppose that contribution as proportional gain rises. Filtering the proportional path changes its magnitude and phase, increasing the impedance of the adverse feedback contributions near resonance. It leaves the VFD path itself intact.
Why This Changes Things
The result makes filter design less of a one-way compromise. A common intuition is that a compact filter requires accepting more resonance risk, or that a faster current loop is always an unambiguous improvement. The paper shows why neither assumption is reliable in this control architecture: filter resonance, current-loop gain, active damping and sampling delay all shape the same stability margin.
That matters for grid-forming converters because their value depends partly on behaving predictably when the surrounding grid changes. Smaller passive components can support more compact, lighter or potentially lower-material equipment. They may also ease packaging in renewable-energy and storage systems. But this study does not quantify manufacturing savings, lifecycle emissions or cost. Those benefits remain possible downstream consequences, not measured outcomes of the prototype experiment.
The paper also offers a more disciplined way to choose controller settings. Rather than select a filter and then tune the control around it by trial and error, designers can estimate a two-sided range of effective damping gain. The lower edge guards against insufficient voltage-loop support; the upper edge guards against destabilizing the resonance mode. The authors derive those bounds from resonance-band crossings, then check them against the complete model. The reduced model’s lower and upper bounds differ from complete-model values by 21.1% and 0.79%, respectively, underscoring why the simplified criteria are useful for screening but not a substitute for detailed verification.
The frequency-shaping choice is also intentionally modest. It affects the proportional path but leaves the resonant branch that tracks the fundamental frequency alone. That separation is useful: it offers a way to reshape damping at higher frequencies without redesigning every part of the controller. The authors find that setting the shaping cutoff near the physical LC resonance extends the positive-resistance region; lowering it much further adds phase lag without much additional extension.
There is a broader lesson for power electronics. Stability is not only a property of a component or a controller considered in isolation. It emerges from how hardware, feedback and computation interact. In this case, a digital delay that is easy to overlook becomes a central design constraint—and a small change to one signal path can restore design options.
What’s Next
The strongest evidence comes from analysis and a 3-kVA prototype. Before treating the result as a general recipe, engineers need to test the method across different power levels, filter designs, control platforms and switching arrangements. A different sampling/update scheme changes the effective delay, so the reported one-third-of-sampling-frequency placement limit applies to the studied conditions and gain range, not to every grid-forming inverter.
The grid-connected tests point to another boundary: the grid itself changes which current gains are admissible. The paper uses output-impedance analysis to identify grid-dependent constraints and reports tests at selected weak- and strong-grid operating points. That is important because a filter and controller stable in an isolated or strong-grid setup may behave differently when connected to a network with greater impedance. The supplied results do not establish performance across the full range of grid conditions, disturbances or power levels that a field deployment would encounter.
There are also engineering trade-offs beyond stability. The filter must still attenuate switching harmonics and keep currents and voltages within component ratings. The proposed low-pass filter adds phase lag to the current loop and lowers its nominal crossover at a fixed gain; designers must check that the resulting bandwidth still meets their transient and regulation needs. The study frames its shaping cutoff between the current-loop design scale and the physical resonance, with the representative choice at 0.9 times that resonance.
The next step is not simply to shrink every filter. It is to use delay-aware analysis early enough that engineers can compare compactness, damping, bandwidth and grid robustness before hardware is fixed. If the approach holds across a wider set of converter designs, it could make grid-forming equipment easier to tailor without forcing a choice between compact hardware and stable control. That is a small but consequential piece of the larger work of building power systems that can rely on renewable generation and storage while keeping the grid steady.