Radiated EMI Characterization of the AMK FS-Kit Inverter System
RIT Racing's F33 utilizes the AMK Racing Kit consisting of four DD5-14 permanent magnet synchronous hub motors, each driven by a KW26-S inverter operating at an 8 kHz PWM switching frequency on a 600 VDC bus. During bench testing, significant electromagnetic interference was observed across multiple vehicle subsystems. Symptoms included CAN bus message dropouts between the vehicle control unit (VCU) and inverter nodes, repeated STM32 microcontroller resets on the CAN line, and interference with nearby laptop touchpads.
Preliminary bench sweeps using an oscilloscope probe indicated a broadband peak at approximately 8 kHz corresponding to the inverter PWM fundamental, with a second harmonic visible at 16 kHz. The existing harness employed twisted shielded pairs with single-point shield grounding for CAN and low-voltage (LV) signal lines. Despite this shielding approach, the interference persisted, indicating that the coupling path was not adequately addressed.
Measurements were performed using a Keysight EDUX1052G digital storage oscilloscope, and a 10x passive voltage probe with the ground lead removed served as a near-field electric field probe. The FFT was configured with center frequency at 8 kHz and span of 10 kHz. Probe positioning was controlled using non-conductive rulers, calipers, and marking tape to establish repeatable standoff distances from each emission source.
The vehicle was configured for static bench testing, high voltage enabled, wheels elevated off ground, inverters operating at idle and light-load setpoints. The coordinate origin was defined at the center of the inverter cooling plate. All cable sets were installed in their as-run configuration, including DC bus cables, three-phase motor leads, LV harness, and Ethernet interconnects. On the vehicle, the inverter enclosure sits in the rear of the chassis. E-shelf, an aluminum honeycomb panel above and forward of the inverters, houses the complete LV electronics stack (VC, C70, IMU, PBX, shutdown circuit, SSDB, e-HVD, MSD, energy meter, LV battery). The HV DC bus cables are routed above E-shelf between the accumulator and inverters, while the three-phase motor leads route along the bottom face of the car to the hub motors, physically separated from E-shelf.
The measurements were organized around the principal cable and enclosure interfaces of the inverters. The inverter enclosure was measured with the mu-metal lined lid installed to characterize shielded performance at varying standoff distances, and again with the lid removed to establish an unshielded baseline under otherwise identical conditions. The HV three-phase cables between the inverter and motor were probed at increasing radial distances from the conductor surfaces, and the DC bus cables feeding the inverter were measured in a similar fashion along their length.
On the low-voltage side, E-shelf was measured to assess the EMI environment at the vehicle’s LV electronics. Additional measurements were taken on the LV signal lines, inverter Ethernet link, and GNSS antenna cable.
The original test plan included a common-mode current calibration step, driving a known sinusoidal current at 8 kHz through a calibration cable via an audio amplifier and series resistor to establish a transfer impedance relationship between probe voltage and true cable CM current. This calibration was not completed during the testing due to time constraints, so all measurements in this report are expressed as raw probe voltage magnitudes (FFT units) rather than calibrated CM current in amperes.
At each measurement point, the probe was held stationary for 5–10 seconds and a single-shot FFT acquisition was captured. Per-point metadata included timestamp, shield state (ON/OFF), operating point (idle/load), Cartesian coordinates (x, y, z) referenced to the cooling plate center, probe orientation (X+ on vehicle), and FFT magnitude at 8 kHz and 16 kHz. Each capture was saved as both CSV and PNG using the naming convention YYYYMMDD_state_idle/throttle_point_x###_y###_z###_8k.csv. Background noise measurements were recorded with inverters off for subsequent baseline subtraction.
The table below the measured 8 kHz peak magnitude, band RMS, and relative level in dB for each measurement location. The unshielded inverter (lid removed) is used as the 0 dB reference baseline. Negative dB values indicate emission levels exceeding the open-inverter baseline; positive values indicate attenuation below it.
The phase cables on the vehicle exhibited the highest 8 kHz emission levels at −29.9 dB relative to baseline (magnitude 9,239 at the 8 kHz bin), followed by the phase cables measured outside the vehicle at −24.3 dB (4,851). The phase cable connector (−21.8 dB, 3,636) and electronics shelf (−21.5 dB, 3,489) also exceeded the open-inverter baseline by over 20 dB. The DC bus lines measured −15.7 dB (1,804), and the Ethernet cable −11.0 dB (1,052). Below the unshielded baseline, the VC and C70 tethers measured +6.4 dB (142) and +6.7 dB (137) respectively, indicating lower emission levels than the open inverter at those locations. The shielded enclosure measured +9.6 dB below baseline (98), confirming effective radiated attenuation by the mu-metal lined housing. The LV charger and LV lines, despite high overall spectral content, measured only 71 and 44 respectively at the 8 kHz bin (+12.4 and +16.6 dB below baseline), indicating that their dominant emissions are at other frequencies.
The figures below compare the FFT spectra measured at the inverter surface with the enclosure lid removed (unshielded baseline) and with the mu-metal lined lid installed. The 8 kHz peak magnitude decreased from 295 to 98 arbitrary units with the shield in place, corresponding to 9.6 dB of attenuation. This confirms that the enclosure provides meaningful radiated emission reduction from the inverter modules.
At E-shelf, the measured 8 kHz peak magnitude was 3,489, approximately 11.8× the unshielded baseline (295) and 35.7× the shielded enclosure level (98). It is not at the inverter enclosure boundary itself, yet it sees emission levels far exceeding the open inverter, confirming that common-mode currents on the DC bus cables are radiating into the LV electronics environment as they pass through this area.
EMI magnitude was measured as a function of radial distance from the cable surface to characterize the near-field spatial decay. The figure on the left shows the 8 kHz band magnitude versus distance with both linear and power-law regression fits. The power-law model (R² = 0.968) yields a decay rate of approximately −8.55 dB/inch, consistent with near-field dipole radiation behavior.
The figures below show smoothed contour maps of the near-field emission along the phase cable, computed from FFT analysis of 11 spatial measurements at 0.275-inch increments. The figure on the left maps ΔdB versus distance at each harmonic of the 8 kHz fundamental (1× through 6×). All harmonics exhibit strong spatial decay with increasing distance, the 1× (8 kHz) fundamental reaches −30 dB at 3 inches, while the 3× (24 kHz) harmonic decays even more steeply, exceeding −35 dB. Even the weaker higher-order harmonics (4×, 5×, 6×) show −20 to −30 dB attenuation at the outer measurement boundary.
The figure on the right reorganizes the same data into 1/3-octave bands. The 8 kHz band exhibits the steepest spatial gradient, reaching −30 dB at 3 inches, consistent with conducted near-field decay from the PWM fundamental. The 3× harmonic at 24 kHz (captured in the 25 kHz 1/3-octave band) shows a similarly steep decay. However, several bands between the harmonics, notably at 10–12.5 kHz and ~20 kHz, appear to show zero distance dependence in the ΔdB contour, remaining near 0 dB across the full measurement range. Examination of the absolute magnitudes reveals that these inter-harmonic bands are approximately 900× weaker than the 8 kHz fundamental at every distance measured, including the nearest point (e.g., the 10 kHz band RMS is ~0.002 V at all positions, compared to 1.85 V for the 8 kHz band at contact). Because these bands sit at or near the noise floor of the Keysight EDUX1052G for the probe configuration used, the ΔdB plot renders them as 0 dB change, not because they carry significant energy, but because the noise floor is spatially uniform. The cable-emitted energy is overwhelmingly concentrated at the PWM fundamental and its odd harmonics, with negligible real content between them.
The measurement data establish that cable-conducted common-mode currents at the enclosure boundary are the dominant EMI coupling mechanism in the AMK FS-Kit inverter installation. All measurement locations associated with HV cable runs (phase leads, DC bus, Ethernet) exceeded the open-inverter baseline by 10 to 30 dB. E-shelf recorded 11.8× the baseline despite being physically separated from the inverter enclosure, attributable to the HV DC bus cables routed directly above it. The mu-metal lined enclosure provides approximately 9.6 dB of radiated attenuation (295 → 98 at 8 kHz), and subsequent testing of the shielded enclosure in isolation confirmed that its residual radiated emission level is comparable in magnitude to mains-frequency pickup from a standard wall outlet at 60 Hz. In other words, the enclosure shield is performing adequately, radiated leakage through the housing is not the limiting factor in system EMI performance.
The phase cables represent the single largest emission source, with peak magnitudes at 8 kHz approximately 31× the unshielded inverter baseline. The cable distance sweep confirmed that this energy is tightly concentrated at the PWM fundamental and its odd harmonics, with inter-harmonic bands falling at or below the oscilloscope noise floor at all measurement distances. Spatial decay follows a power-law relationship at approximately −8.5 dB/inch, providing a quantitative basis for harness routing separation requirements.
At 8 kHz, conventional ferrite chokes are ineffective because ferrite permeability peaks above 100 kHz. Nanocrystalline cores provide high inductance per turn at low frequencies due to permeability roughly 20× that of MnZn ferrite. In the future, the cable conductors are ideally threaded through a bare nanocrystalline toroid, with multiple passes providing N turns. The differential current cancels magnetically in the core so it does not saturate, regardless of load current. Only the common-mode current sees the inductance. For 5 turns on a L2063 L = 59 × 25 = 1,475 μH → Z = 74.1 Ω at 8 kHz. With Zsource = 5 Ω, attenuation = −24 dB (signal reduced to 6.3%).
The CM current path on the phase cables includes the motor winding-to-frame parasitic capacitance (Cwf). At 8 kHz this presents 4,000–10,000 Ω of impedance already in the CM loop, which limits the additional effect of a series choke. Adding 74 Ω of choke impedance to a 4,000 Ω loop provides only −0.2 dB of additional attenuation. However, the choke redistributes where the CM voltage drops, the cable segment between the choke and motor sees reduced CM voltage, while the segment between inverter and choke remains unchanged. Placing the choke as close to the inverter output as possible minimizes the radiating cable length. On the DC bus path, where source impedance is under 2 Ω, the same choke provides over 30 dB of attenuation. A choke alone provides −20 dB/decade attenuation above its corner frequency. Adding Y-capacitors (safety-rated Y2 class, 47–100 nF ceramic) from the motor frame to chassis ground creates a second-order LC filter with −40 dB/decade rolloff above resonance. With the L2063 choke at 1.48 mH and a 100 nF Y-capacitor, the LC resonance sits at approximately 13 kHz. The 8 kHz fundamental falls below this resonance and passes through the LC filter largely unattenuated, on the DC bus, the choke alone handles the fundamental due to the low source impedance. The LC filter’s contribution is steep attenuation of the second harmonic (16 kHz) and all higher harmonics at −40 dB/decade. The Y-capacitors also provide a low-impedance return path for CM current at the motor end, effectively lowering the source impedance seen by the choke and improving its performance on the phase cable paths where the choke alone has limited effect due to high parasitic capacitance impedance. The Y-capacitors in the LC filter connect the motor frame to chassis ground as part of the CM filter return path, they would not be in the encoder signal ground. The encoder requires a direct, continuous DC ground reference through the motor frame to the chassis, placing a capacitor in that path would block DC and cause the encoder ground to float, resulting in dropouts.
Testing for F33 focused on the 6–50 kHz frequency range, capturing the PWM fundamental and its first several harmonics. However, the spectral energy from PWM switching extends far beyond this range. The AMK KW26-S inverters use IGBTs with typical rise times of ~300 ns switching a 600 V DC bus. The spectral envelope of a trapezoidal switching waveform has two corner frequencies: the first at f1 = 1/(π × D × TPWM) ≈ 5.1 kHz, below which the spectrum is flat, and the knee frequency at fknee = 1/(π × trise) ≈ 1.1 MHz. Between these corners the amplitude rolls off at −20 dB/decade; above fknee the rolloff steepens to −40 dB/decade. Significant spectral content therefore persists into the low MHz range. The 1st harmonic (8 kHz) would require a 6.2 km cable to resonate which is five orders of magnitude longer than any cable on the car. Even the 125th harmonic (1 MHz) needs a 49.5 m antenna. No discrete harmonic of the 8 kHz PWM has an antenna match to the car’s cable lengths. The actual cable resonances are excited not by discrete harmonics but by broadband switching-edge content at much higher frequencies.
The current F33/F34 grounding topology grounds the phase cable shields at the inverter enclosure but does not bond them through the chassis feedthrough. The motor-side shields are grounded through the motor frame for encoder grounding, with the shield intentionally broken at the feedthrough to prevent ground loops. At 8 kHz this gap is inconsequential. At the cable’s λ/4 resonance of 49.5 MHz, the gap becomes a slot antenna radiating directly into the chassis cavity near E-shelf and LV electronics. The CAN bus shields are grounded at one end only, which is correct for low-frequency ground loop prevention. However, at the CAN cable’s λ/4 resonance of 33 MHz, the 1.5 m ungrounded shield becomes a quarter-wave monopole antenna. Adding a bypass capacitor (4.7 nF C0G/NP0 ceramic) from the shield to local ground at the floating end presents > 10 kΩ at 60 Hz, at 10 MHz it presents 3.4 Ω. This preserves the existing DC isolation while adding high-frequency shielding.