High-power pulsed generators have become indispensable in high-power microwave generation (HPM) [1–5]. However, they are also significant sources of electromagnetic (EM) noises due to the rapid switching and high voltage/current characteristic inherent in their operation. The generation of transient EM fields can interfere with surrounding electronic systems, creating challenges in ensuring system reliability and electromagnetic compatibility [6–8]. Effective shielding techniques, grounding, and filtering are essential to mitigate EM noise, and numerous studies have addressed these issues through numerical simulations [6–14], experimental approaches [15, 16], and both [17, 18].

HPM systems commonly utilize gigawatt (GW) level pulsed generators [4, 5, 19, 20], which give rise to substantial electromagnetic interference. To ensure stable operation, it is crucial to analyze the intensity and frequency characteristics of EM noise generated by these generators, as such noise may disrupt sensitive electronic components. Despite its importance, research on EM noise generated by GW-level pulsed generators remains insufficient.

Our previous work investigated HPM system with GW-level pulse-forming-network (PFN) Marx generator using the particle-in-cell solver of CST Studio Suite [5]. Based on this, we analyzed EM noise generated by a PFN-Marx generator in an HPM system. Due to the generator’s complex structure and operating mechanisms, analytical methods alone are insufficient to study noise behavior. Thus, experimental measurements and numerical simulations are required. CST simulations, which have been widely employed in prior studies [6, 7, 11–14, 18], provide a powerful tool for analyzing EM noise.

In this study, CST simulations were conducted to investigate the EM noise generated by a PFN-Marx generator, and experimental measurements were performed to validate the simulation results. Section II introduces the simulation model of the HPM system. Section III presents the simulated EM noise spectra. Section IV describes the experimental setup and results. In Section V, the reduction of EM noise through simple shielding was evaluated using CST. The conclusions are presented in Section VI. Unlike previous studies for electromagnetic interference that mainly rely on simplified structures or low-voltage systems, this work provides an experimentally validated, geometry-based analysis of electromagnetic noise generated by a GW-level PFN-Marx generator, with a focus on identifying structure-driven emission behavior and assessing practical noise mitigation potential.

Fig. 1 shows the CST model of the HPM system consisting of a PFN-Marx generator, a transmission line, and an HPM source. To better reflect the actual physical geometry, the present model follows the 3D drawings intended for actual fabrication, except for a few auxiliary components; it differs from the model used in the previous study only by minor modifications in several supplementary components [5]. Nylon insulators are inserted between the components to separate the vacuum and pressurized regions (the yellow parts in Fig. 1). The PFN-Marx generator is composed of 16 PFN modules with spark-gap switches. When the switches are triggered, the voltage stored in the PFN modules is stepped up, generating a high-voltage pulse on the order of several hundred kilovolts.

The time-domain solver of the CST microwave studio was employed to simulate the PFN-Marx generator. The simulation was conducted using an open boundary condition to minimize artificial reflections at the domain boundaries. Given the asymmetry of the PFN-Marx generator’s internal structure, no symmetry plane was applied, ensuring that the simulation fully captures the device’s inherent geometric complexity. The computational domain was defined to extend 1,500 mm in the Y and Z directions from the outermost surfaces of the device, while the X direction was limited to 100 mm to reduce computational cost. A hexahedral mesh comprising 13,873,419 elements was employed. To accurately resolve the critical physical process occurring within the device such as switch operation and the Marx erection process, the mesh was refined in the interior region, while a coarser mesh was adopted in the exterior.

Its configuration follows the previous research approach, which omits the charging process to reduce simulation time. Discrete ports were defined between the switch gaps, and the Marx erection process was initiated by applying a voltage pulse to the ports [5]. The voltage applied to the ports corresponds to the charging voltage of the PFN module, which is amplified sixteen-fold through the Marx erection process. The generated high-voltage pulse is delivered to the HPM source via the transmission line. Since the focus of this study is on EM noise emitted from the PFN-Marx generator, the HPM source was simplified as a matched lumped resistor, thereby excluding RF noise generated by HPM source itself. The resistor was connected to a metallic chamber that served as the return path for the pulse.

At the bottom of the device, ports for the trigger input and PFN charging were included (blue boxes in Fig. 1). However, since the triggering and charging processes were not considered in this simulation, these ports remain open. Since the real system includes precharged capacitors and connected charging lines, it may introduce discrepancies between the simulation and the actual system. Consequently, this model only accounts for EM noise generated during the Marx erection and high-voltage pulse transmission. The EM noise can be radiated outside the chamber through unshielded parts, such as the nylon insulators and insulation gas inlet. However, as the auxiliary components are not fully represented in this model, the nylon insulators are considered the primary emission path for EM noise in this model. These aspects will cause differences between the actual device operation and the simulation.

The voltage pulses obtained from the CST simulations are shown in Fig. 2. The corresponding electric fields (E-fields) were analyzed to characterized EM noise. An example of the time-evolution of E-field measured by the E-field probe is also shown in Fig. 2. The E-field was measured at a distance of 1,000 mm for the 400 kV case. When the switch is triggered at 0 ns, the voltage rises within approximately 30 ns, a timescale determined by the capacitance and inductance of the PFN circuit. The subsequent flat-top levels of 240 kV, 320 kV, and 400 kV reflect the different voltage applied to the switch gap, i.e., the charging voltage of the capacitor. Small oscillations in the flat-top plateau arise from impedance mismatch between the PFN-Marx generator and the transmission line. The EM noise begins to appear during the voltage rise and gradually decreases as the pulse enters the flat-top plateau. The rapid voltage rise is likely associated with the observed transient EM emission, although the exact mechanisms may involve multiple transient effects within the device.

Figs. 3 and 4 illustrate the distribution of the E-field intensity for the 400 kV case. During the initial voltage rise, strong E-fields appeared around the spark-gap switches, leading to uneven emissions. The E-field was subsequently emitted to the outside via adjacent insulators, as shown in Figs. 3(a) and 4(a). After 5 ns, although the high-voltage pulse is still rising in amplitude, it has reached the HPM source, and the axial E-field distribution at the insulator becomes uniform, as shown in Figs. 3(b) and 4(b). As observed in Fig. 2, the EM noise emitted into space reaches its peak around 20 ns and gradually decreases thereafter, which is also evident in the 2D field distribution, as shown in Fig. 3(c)–3(e).

The E-field is strongly emitted through the areas where the insulator is exposed, with an intensity of several thousand V/m; the strong E-field was observed at the side of the insulator plates covering the PFN-Marx generator (referred to as the top and bottom plate) and in the vicinity of the charging port (referred to as bottom). EM noises from the insulator plates escaped through the gaps between metal screws and are relatively uniform azimuthally. The bottom port, however, exhibited anomalously strong emissions, which are considered artifacts of the simulation with omitted charging and primary-power stages. Therefore, this observation is likely different from the EM noise of interest.

For quantitative analysis, E-field probes were placed at four positions (top, bottom, top plate, bottom plate) as shown in Fig. 3(a). Probes were placed at distances of 100 mm, 250 mm, 500 mm, 750 mm, and 1,000 mm from these positions to measure the E-field in the X, Y, Z directions (E_X, E_Y, E_Z). The X direction is perpendicular to the 2D map in Fig. 3, which corresponds to the vertical direction relative to the ground in the experiment of Section IV. The Y direction is horizontal relative to the device, and the Z direction is perpendicular to the device. Therefore, the coordinate systems of the top and bottom (Yellow), are rotated 90° with respect to ones of the top and bottom plates (grey), as shown in Fig. 3(a).

Fig. 5 shows the maximum E-field strength measured at 1,000 mm for the three voltages. As the voltage increases, the E-field also increases linearly. The lowest E-field was measured in the Z direction, and the highest one was measured in the Y direction. At the top, E-field strengths were lower than at other positions, which is attributed to the distance from the exposed insulators and the PFN-Marx generator. In contrast, the bottom recorded the highest E-field, with a maximum of 784 V/m in the Y direction. The E-field at the top and bottom plates were similar, with the highest ones measured at 650 V/m and 605 V/m, respectively, in the Y direction.

Fig. 6 shows the maximum E-field strengths as a function of distance for the 400 kV case. As the distance increases, the E-field decreases rapidly. At the bottom, where the E-field is strongest, the E_X, E_Y, and E_Z at 100 mm were 7,774 V/m, 8,493 V/m, and 8,574 V/m, respectively; at 1,000 mm, they were 728 V/m, 784 V/m, and 394 V/m, respectively, indicating a reduction in intensity of more than 90%. The high rate of E-field reduction at the bottom may be due to the rapid decrease in the high potential of the charging port with increasing distance. At the top, E_X and E_Y exhibited a relatively low reduction rate, decreasing by only about 30% as the distance increased from 100 to 1,000 mm. In contrast, the EZ showed a significantly higher reduction rate of about 92%. The E-field strengths and reduction rates at the top and bottom plates are similar. The E-field strengths are between those of the top and bottom, and the reduction rates also falls within a range of approximately 50%–80%.

The time-domain E-field at 1,000 mm was proceeded using Fourier transform to obtain the frequency spectrum; the results are shown in Fig. 7. It should be noted that the Y-axis value of graphs is higher only in the case of the bottom position when compared to other measured positions. The dominant noise occurs at frequencies below 1.5 GHz, with peaks reaching tens of V/m. In the spectrum of E_X, three peaks are observed below 0.5 GHz at 0.2 GHz, 0.3 GHz, and 0.36 GHz. These peaks appear consistently at all positions, with the highest peak of 80 V/m occurring at the bottom. In the spectrum of E_Y, the same peaks were observed, but additional smaller peaks were also present below 0.5 GHz. In the spectrum of E_Z, low-frequency noise was measured below 0.1 GHz at all measured positions. This is likely due to the high potential of the charging port, as confirmed earlier. Above 0.5 GHz, several peaks of tens of V/m were grouped, particularly forming significant peaks between 0.5 and 1.0 GHz. At the bottom, the largest peaks of E_X and E_Y were measured in this frequency range; they are 87 and 78 V/m, respectively. At the top, top plate, and bottom plate, the peaks of E_Y were relatively larger than that of E_X and E_Z, whereas at the bottom, the peaks of E_X were comparatively larger. Among the positions, the peaks measured at the bottom were generally the larger than the other measured positions. These results indicate that EM noise from the PFN-Marx generator is dominated by low-frequency resonances associated with the charging port and insulator structure.

An experimental setup (Fig. 8) was constructed to measure EM noise from a device identical to the CST model. It is a schematic representation, and the relative scale of the device’s dimension and others is not accurate. The measurement frequency range was set to 0.1–1.5 GHz, corresponding to the dominant frequency range in simulations. The experiment was conducted outdoors, with the device placed on a non-conductive surface. A log-periodic antenna (LPA; model VULB 9163; Schwarzbeck, Schönau, Germany) was employed to measure EM noise. It is capable of measuring frequencies in the range of 0.03-3 GHz. The LPA was positioned 1 m from the device and mounted at a height of 1.6 m above the ground. The antenna factor and gain of the LPA are presented in Fig. 9. The data measured by the LPA was collected using a spectrum analyzer (N9030B; Keysight, Santa Rosa, CA, USA) and an oscilloscope (DSO-X92504Q; Keysight), which are capable of measuring up to 26.5 and 25 GHz, respectively. The antenna signal was transmitted to them via cables. Different cables were used for the spectrum analyzer and oscilloscope, and their transmission coefficient is shown in Fig. 10. Additionally, since the signal amplitude from the LPA was too high to be directly measured by the spectrum analyzer and oscilloscope, it was attenuated before being delivered to them. A 10-dB attenuator was used for the spectrum analyzer, while a 60-dB attenuator was used for the oscilloscope. Finally, the measured EM noise was calculated by incorporating the antenna factor, cable loss, and attenuation level.

Measurements were performed at three positions (top, top plate, and bottom) as illustrated in Fig. 8. Both the vertical and horizontal polarizations of the EM noise were measured, corresponding to the E_X, and E_Y in the CST simulations, respectively. To minimize interference from microwaves generated by HPM source, measurements were not conducted on the side where microwave radiation occurs. Additionally, microwave absorbers were used to shield the antenna. Measurements at each position were conducted in separate trials.

Before measuring the EM noise, the background noises from the environmental and auxiliary devices were measured using the LPA and spectrum analyzer. Firstly, the environmental noises were measured with no electrical devices in operation, followed by noise measurement after activating the auxiliary devices. In this study, only the noise measurement results in the ready state, with all auxiliary devices operating, is presented in Fig. 11. The background noise occurred in the frequency range below 1 GHz, with a maximum intensity of 0.024 V/m. Since the magnitudes of the noise peaks are less than 1% of the magnitudes of the E-field peaks predicted by the CST simulation, they can be considered negligible.

Since EM noise is generated during the short operation time of the PFN-Marx generator, which lasts only a few hundred nanoseconds, an oscilloscope was used to measure the time-domain signal. As in the results of CST simulation, the data was transformed into a frequency spectrum using a Fourier transformation. The results are shown in Figs. 12 and 13. For comparison, the CST simulation results corresponding to the measurement positions and polarizations were reused.

The spectra for vertical component of EM noise (black line) correspond to those of E_X (red line) and shown in Fig. 12. At the top, multiple noise peaks smaller than 20 V/m were observed at frequencies below 1 GHz. At the top plate, a noise peak of 31 V/m was measured at 0.27 GHz, and several peaks of approximately 20 V/m were observed at frequencies above 0.5 GHz. At the bottom, the strongest noise appeared, with a 63 V/m peak at 0.93 GHz. The noise at the bottom position was stronger than at any other position, which aligns with the CST results. To quantitatively compare the measured spectrum with the simulated spectrum, the ten largest peaks in each spectrum were identified as the primary peaks (black and red circles in Fig. 12). Peaks that were too closely spaced to be clearly distinguished were excluded to ensure the reliability of the selection. First, to compare the overall amplitude level, the mean peak intensities of the selected peaks of the measured and simulated spectra were calculated, as shown in Table 1. Although some deviations exist, their mean intensities are generally comparable. Additionally, to compare the spectral distribution, the mean frequency deviation between the corresponding primary peaks were computed. For each measured peak, the frequency deviation to the nearest simulated peak was determined. To mitigate the influence of isolated outlier peaks, the two largest frequency deviations were excluded, and a trimmed mean was calculated, as shown in Table 1. The analysis revealed that the frequency deviations were within 0.1 GHz, indicating that the measured spectra and the simulated spectra exhibit a comparable spectral distribution.

The spectra of the horizontal component of EM noise (black line) corresponds to those of E_Y (red line) and is shown in Fig. 13. The spectrum at the top is similar to that of the vertical component. However, its magnitude is approximately half of the noise peak magnitude predicted by the CST simulations. At the top plate, multiple peaks of about 30 V/m were observed which are larger than the peaks of the vertical polarized noise; it also consistent with the CST results. At the bottom, the strongest noise peaks are measured, as in the CST results. Similarly, the ten largest peaks were selected, and both the mean peak intensity and the trimmed mean frequency deviation were calculated, as shown in Table 1. Consistent with the previous case, the mean intensities of the spectra were also found to be at comparable levels, and the frequency deviations were within 0.1 GHz.

Overall, the experimental spectra showed good agreement with the CST results in terms of magnitude, distribution, and relative noise levels across positions, although some discrepancies were observed. This confirms that CST-based modeling can reliably reproduce EM noise behavior of the PFN-Marx generator in the measured frequency range.

The shielding effect is evaluated in this section using the validated CST model to estimate the resulting reduction in EM noise. As identified in Section III, EM noise propagates outward primarily through exposed insulators. To mitigate this, simple metal shields were added around the insulators in the CST model. The resulting spectra of the EM noise for E_Y at 1,000 mm are shown in Fig. 14; the metal shields are shown in Fig. 15(a). With the implementation of shielding, low-frequency noise was significantly suppressed. As the frequency increased, however, the shielding effectiveness progressively diminished, and the residual emissions reached their maximum near 12.8 GHz. Even at this peak, the field magnitude remained on the order of tens of mV/m, representing a reduction of more than three order of magnitude relative to the unshielded case. Similar behavior was observed for the E_X and E_Z components; therefore, their spectra are not presented here.

To quantitatively assess the shielding performance, the EM noise ratio between the shielded and unshielded cases was evaluated, as summarized in Table 2. The EM noise ratio was defined as the ratio of the integrated EM noise strength, obtained by integrating the respective noise spectra over the 0.1–18 GHz frequency range. Across all measurement positions and field orientations, the EM noise ratio remained below 1%, demonstrating that even a simple metallic shield can provide substantial attenuation of EM noise.

To compare the spatial distribution of EM noise, the two-dimensional noise distributions for the shielded and non-shielded cases are shown in Fig. 15(b) and 15(c). A time of 50 ns was selected as a representative snapshot at which the radiated field has sufficiently propagated, enabling a clear comparison of the spatial EM noise distributions between the two cases. As confirmed earlier, the EM noise amplitudes differ significantly between the two cases; therefore, note that the color bar limits are set to 10 V/m and 10,000 V/m for the shielded and non-shielded cases, respectively. In the non-shielded case, strong EM noise is emitted from the insulator parts, whereas in the shielded case, localized noise was observed near the transmission line where the outer conductor diameter changes. This may be linked to current density variations and impedance mismatches, though the exact mechanism requires further study. It also remains unclear whether such factors can generate noise at frequencies more than ten times higher than that of the voltage pulse. Importantly, since the simulation has only been experimentally validated in the frequency range around 1 GHz, further investigation is required for its applicability at higher frequency range.

This study presented an experimentally validated analysis of EM noise generated by a GW-level PFN-Marx generator, combining CST simulation and measurement to investigate structure-driven emission behavior and mitigation potential. The CST model, which included the transmission line and HPM source, focused on the generator operation and high-voltage pulse transmission. While simplifications were made (omission of detailed spark-gap models and auxiliary power supplies), the simulation results showed good agreement with experimental measurements, confirming that the primary source of EM noise is the operation of the Marx generator and pulse transmission.

Spectral analysis revealed distinct peaks below 1.5 GHz, with E-field strengths on the order of tens of V/m at a distance of 1 m from the device. These peaks are expected to originate from phenomena such as the switch turn-on process and resonances between parasitic inductances and capacitances. However, since the EM noises persist throughout the voltage pulse, whereas the switch turns on rapidly and operates consistently during this period, parasitic LC resonance inherent to the structure of PFN-Marx generator is likely the dominant contributor. This also explains why both the calculated and measured spectra are not continuous but exhibit distinct peaks at specific frequencies.

They were primarily generated within the device and leaked to the exterior through the insulator. Shielding of the insulators effectively suppressed low-frequency noise, though residual high-frequency components (approximately 12.8 GHz) with magnitudes of tens of mV/m remained. Since complete shielding is difficult in practice due to operational requirements (e.g., insulating gas tubes, control cables), low-frequency noise below 1.5 GHz should be a key consideration in HPM system design.

This work provides one of the few experimental validations of CST-based EM noise modeling for PFN-Marx generators and demonstrates that such simulations can serve as a reliable and practical tool for HPM system design. The validated model enables the analysis of noise intensity and spectral characteristics at critical component locations prior to system construction, providing valuable input for electromagnetic interference assessment and shielding evaluation. In particular, this approach allows potential electromagnetic interference issues to be identified and addressed at an early design stage, supporting more informed decisions on structural layout and shielding strategies. As a result, the proposed methodology can help reduce design iteration, development cost, and overall implementation time in high-power pulsed systems where extensive experimental testing is difficult to perform.