Skip to content

Model architecture

Final Simulink circuit — the as-modeled topology
The final Simulink circuit — two cascaded H-bridges with the TLP250-isolated gate drive (per chb-isolation), the LS-PWM/PSC modulator, and the LC filter + load stage. This is the v3 model (chb-5level-rl-nospike.slx) after the snubber-tuning pass that removed VDS ringing.

Three Simulink models live in simulation/simulink/. They share a common block structure; the differences are in the gate-driver model, the modulator, and the load.

Common structure

flowchart LR
    Vdc1[V_DC₁] --> HB1[H-bridge 1]
    Vdc2[V_DC₂] --> HB2[H-bridge 2]
    Ref[50 Hz sine reference] --> Mod[LS-PWM / PSC
modulator] Mod --> HB1 Mod --> HB2 HB1 --> Sum[Σ cascade] HB2 --> Sum Sum -->|optional| LC[LC filter] LC --> Load[Resistive / RL load] Sum -->|no filter| Load Load --> Meas[Scope, FFT, THD]

Per-model details

chb-5level-v1.slx — IPD LS-PWM baseline

The first-pass model. Ideal complementary switches, no parasitics, no dead time. Used to:

  • Confirm the 5-level cascade arithmetic (output level ∈ {−2, −1, 0, +1, +2}·VDC).
  • Produce the headline THD = 4.9 % figure with IPD level-shifted PWM at 5 kHz.
  • Validate the IEEE 519-2022 voltage-distortion compliance argument.

Inputs: two independent 50 V DC sources, 50 Hz reference, MI = 0.95. Output measurement: the cascade output is scoped + FFT'd to extract THD.

chb-5level-v2.slx — gate-driver sweep

Adds gate-driver behavioural models:

  • IR2110 (bootstrap): the bootstrap capacitor follows VS of the high-side MOSFET. Because the upper bridge's VS floats at the cascade voltage, the bootstrap path never returns to true ground, and the simulated gate drive collapses to < 5 V on the upper bridge — MOSFETs fail to turn on. Output waveform: severely distorted, irrecoverable.
  • TLP250 (optical): independent isolated 15 V supply per driver. All eight gates see 15 V regardless of floating potential. Output waveform: matches the ideal 5-level cascade.

This model is the evidence that killed the IR2110 path before any silicon was committed. The TLP250 → 6N137 → MCP3201 sensing chain on the actual hardware traces directly to this simulation.

chb-5level-rl-nospike.slx — RL load + LC filter sweep

Adds:

  • LC output filter (parametrised — see the THD analysis for the values tried).
  • RL load (10 Ω + 5 mH) to surface the wavy-current artefact under inductive load.
  • No spike in the title refers to the snubber-tuning pass that removed the VDS ringing the earlier model showed.

Used to size the bench LC filter (planned roadmap item; not built).

Simulink — output waveform without snubbers (showing V_DS spikes)
The "before" view — Simulink output when the snubber network is removed. The VDS ringing visible here is exactly what the 22 Ω 2 W + 2.2 nF / 630 V snubber on every MOSFET damps in the as-built hardware.

How to run

  1. Install MATLAB R2023b or later with Simscape Electrical.
  2. Open the .slx file.
  3. Press Run. Default simulation time is 100 ms (≈ 5 fundamental periods).
  4. The output scopes show: cascade voltage, per-bridge contributions, load current.
  5. To extract THD, run thd(simout.OutputVoltage.Data, simout.OutputVoltage.Time(2) - simout.OutputVoltage.Time(1), 50).

A reusable thd-analysis.m helper is not yet committed — the team ran analysis directly in the MATLAB console. Adding the helper to simulation/matlab/ is tracked as a polish item.