Skip to content

First bench session

The linear, single-session walkthrough for putting the populated boards on the bench for the first time and getting all three modulators (STAIR, STAIR_ALT, PSC) verified end-to-end.

This page renders firmware/stm32-f303re/FIRST_BENCH_SESSION.md verbatim — the file ships in the firmware repository and is the authoritative copy.

Headline safety goal: do not burn a TLP250, do not burn a MOSFET. Every step has explicit pass/fail criteria; if anything is unexpected, STOP and consult the bring-up reference troubleshooting before continuing.


  • USB cable (Nucleo ST-LINK end to PC)
  • Two bench DC supplies, 0–60 V, with current limiting
  • One 15 V bench supply for gate drive (or use the B0515S after Phase 3)
  • Dummy load: 100 Ω, 10 W+ resistor (a few in parallel is fine)
  • Oscilloscope, 2+ channels, 50 MHz+ bandwidth, with two probes
  • Multimeter
  • Heatsinks attached to all 8 MOSFETs — no exceptions
  • Safety glasses, hands clear of the breadboard during operation
  • A way to instantly cut DC bus power (bench-supply output enable button)
  • A second person if possible

Session pass/fail checklist (track top-to-bottom)

  • A. Code + dashboard ready
  • B. First boot verified, defaults loaded, lock=OK
  • C. Sensors read known voltage correctly (5 V → vdc ≈ 5 V)
  • D. All 8 gates switch cleanly under STAIR with no DC bus
  • E. STAIR cascaded at 10 V — 5-level staircase confirmed
  • F. STAIR_ALT cascaded at 10 V — same staircase, bridges balance
  • G. PSC single bridge at 5 V — 3-level PWM, other bridge silent
  • H. PSC cascaded at 5 V — 5-level PWM visible ← the project deliverable
  • I. PSC ramped to 50 V step-by-step without trouble
  • J. PSC sustained run, bridges thermally matched

Section A — Code and dashboard ready

A.1 — On the dev PC: commit and push the branch

If you've already pushed since the last code change, skip to A.2.

cd C:\Users\furka\Projects\5-Level-Cascaded-H-bridge-Inverter-with-STM32-Nucleo-F303RE
git status
git branch --show-current     # must print: pwm-rewrite-configurable

git add -A
git commit -m "PSC-PWM rewrite: configurable modulator, STAIR_ALT fallback, auto-cancel on dashboard connect"
git push -u origin pwm-rewrite-configurable

A.2 — On the bench PC: clone the repo (if empty folder)

The bench PC starts with an empty folder, no git history yet. Open PowerShell:

# Make a Projects folder if you don't already have one
mkdir C:\Projects -ErrorAction SilentlyContinue
cd C:\Projects

# Clone your fork (the one we pushed to in A.1)
git clone https://github.com/feaksel/5-Level-Cascaded-H-bridge-Inverter-with-STM32-Nucleo-F303RE.git

# Enter the freshly-cloned directory
cd 5-Level-Cascaded-H-bridge-Inverter-with-STM32-Nucleo-F303RE

# Switch to the working branch (git clone defaults to main)
git checkout pwm-rewrite-configurable

# Confirm the newest commit matches what you pushed in A.1
git log --oneline -3

First-time HTTPS clone: GitHub will prompt for credentials. Password auth is deprecated — use a Personal Access Token instead. Create one at GitHub → Settings → Developer settings → Personal access tokens → Tokens (classic) → Generate new. Scope: repo. Paste the token when prompted for the password. Windows Credential Manager will remember it for subsequent operations.

If git isn't installed on the bench PC: download from git-scm.com/download/win, accept all defaults during install, restart PowerShell, then run the clone.

If the folder already has the repo from a previous session (skip this if you just cloned fresh):

cd C:\Projects\5-Level-Cascaded-H-bridge-Inverter-with-STM32-Nucleo-F303RE
git fetch
git checkout pwm-rewrite-configurable
git pull
git log --oneline -3

A.3 — Set up the dashboard (first time only)

cd <repo>
py -3 -m venv dashboard\.venv
dashboard\.venv\Scripts\python -m pip install --upgrade pip
dashboard\.venv\Scripts\python -m pip install -r dashboard\requirements.txt

A.4 — Verify dashboard launches

dashboard\.venv\Scripts\python dashboard\run_dashboard.py

The dashboard window opens. Close it for now.

✅ Pass A: dashboard launched without errors.


Section B — Flash and first boot

⚠️ Critical ordering: dashboard FIRST, flash SECOND. The firmware auto-starts 3 s after every reset. With the dashboard already connected when CubeIDE resets the Nucleo at end-of-flash, the dashboard auto-cancels by sending STATUS the instant it sees the boot message. Without the dashboard, you race the 3 s timer.

The big USB connector at the top of the Nucleo board. Wait ~5 s for Windows to enumerate two devices: - STMicroelectronics ST-LINK (debugger) - USB Serial Device / STLink Virtual COM Port (your UART)

Note the COM number from Device Manager → Ports.

B.2 — Open dashboard, connect to COM port — BEFORE flashing

dashboard\.venv\Scripts\python dashboard\run_dashboard.py

In the dashboard: 1. Source dropdown → Live serial 2. Click Refresh → your COM port appears 3. Select it → click Connect 4. Status changes to "Connected to COMx" 5. Event log shows [TX] STATUS (auto-cancel) immediately

If the old firmware is running, you'll see $E,UNKNOWN_COMMAND in response to STATUS — that's expected; the old firmware doesn't know the new commands. Ignore it.

B.3 — Open the project in STM32CubeIDE

  1. Launch STM32CubeIDE, pick a workspace folder
  2. File → Open Projects from File System...
  3. Browse to the repo root → Finish
  4. Wait for indexing to finish (bottom-right progress bar)

B.4 — Build

Project → Build All (or Ctrl+B). Console should end with:

arm-none-eabi-size  Debug/5levelchb.elf
   text    data     bss     dec     hex filename
  36660     468    3668   40796    9f5c Debug/5levelchb.elf
Finished building target: 5levelchb.elf

Build errors? Check Console panel; usually toolchain or missing include. Don't proceed to flash if build failed.

B.5 — Flash

Run → Run As → STM32 C/C++ Application (or click green play arrow). First time: accept the default Debug Configuration. CubeIDE flashes and resets the Nucleo.

B.6 — Verify boot in dashboard

Watch the event log. Within 1 s of flash completion you should see:

[ACK]    BOOT_SELF_TEST_DONE
[TX]     STATUS (auto-cancel post-reset)
[ACK]    STATUS
[STATUS] ms=42,state=IDLE,mode=...
[CONFIG] mod=STAIR,fsw=500,bridge=BOTH,ffund=50.00,mi=0.95,cntoff=0,lock=OK
[TELEMETRY ARRIVING] $T,...                   ← 20 Hz

Confirm: - [ ] $A,BOOT_SELF_TEST_DONE arrived - [ ] STATUS (auto-cancel post-reset) event present → auto-start cancelled - [ ] $C config line shows defaults: STAIR / 500 / BOTH / mi=0.95 - [ ] lock=OK (for STAIR at FSW=500, cntoff=0 and lock=OK) - [ ] Telemetry frames arriving at ~20 Hz (dashboard plot updates)

✅ Pass B: boot clean, defaults loaded, dashboard in control.

❌ If lock=BAD here, something's wrong with the diagnostic itself (in STAIR mode, expected offset is 0, so it should always be OK). Stop and investigate.


Section C — Sensor decode verification (no power stages)

This catches the MCP3201 shift problem early. Phase 4 of the main bringup doc, condensed.

C.1 — DC inputs disconnected, 5 V bench supply standalone

  1. With both bridge DC inputs disconnected, take a separate small bench supply, set it to 5.0 V exactly, current limit 100 mA.
  2. Touch its + lead to bridge 1's DC bus input terminal, − to GND_HV for that bridge. Just touch, no permanent wiring yet — this is a one-shot test.
  3. Dashboard: send STATUS. Watch the $S,... response.
  4. The vdc1 field should read 4.7–5.3 V.
vdc1 reading Diagnosis Action
4.7–5.3 V ✅ MCP3201 decode correct Proceed to C.2
~64 V ❌ Bit-shift is >> 3 but should be >> 1 See HARDWARE_BRINGUP.md Phase 4 fix — swap and reflash before continuing
0 V / NAN ❌ Sensor not seeing the input Check wiring of voltage divider
Stuck mid-scale ❌ Sensor hardware fault Check 78L05 supply, MCP3201 power

C.2 — Repeat for bridge 2

Same procedure on bridge 2's DC input. vdc2 should read 4.7–5.3 V.

✅ Pass C: both sensors track applied voltage linearly.


Section D — Gate drive verification, DC bus DISCONNECTED

⚠️ DC bus inputs MUST be physically disconnected during this section. Bench supply OFF is not enough — pull the wires from the bridge DC input terminals so a slipped finger on the supply button can't energize anything.

⚠️ TLP250 protection points: - 15 V supply must be ≤ 16 V. Don't apply 20 V "for headroom." TLP250 absolute max VCC is 20 V; the 100 nF bypass cap is rated 25 V. - 220 Ω input limit resistor must be in place — without it, MCU pin sees direct LED forward voltage and overcurrent. - 100 nF bypass cap must be directly across Pin 5 / Pin 8, leads bent to touch the IC. Breadboard hop of >2 cm causes oscillation at 5 kHz and the TLP250 will heat fast.

D.1 — Gate-drive supply on

External 15 V bench supply (or B0515S if you've validated those in Phase 3 of the main doc) connected to +15V_Drive and Drive_GND rails on both bridges. Current limit 200 mA per bridge.

D.2 — STAIR baseline at 500 Hz (default mode)

Dashboard:

STOP                  (in case auto-start has fired — it should not have)
CONFIG                (verify STAIR/500/BOTH defaults)
MI 0.5                (moderate amplitude so all 5 levels are exercised)
START

Look for $A,RUN in event log within ~10 ms.

D.3 — Scope each gate

With STAIR running, scope Vgs (gate-to-source) of each of the 8 MOSFETs one at a time. Expected at 500 Hz with MI 0.5:

  • HS gates (Q1/Q3 of each bridge): 0 V ↔ ~14 V transitions, but with STAIR these happen slowly (level transitions every 2 ms or so).
  • LS gates (Q2/Q4): complementary to their HS partners.
  • Dead-time gap visible (~3 µs) on every transition.

Touch-test each TLP250 package. Should be cool to the touch (room temperature) — STAIR is gentle on the drivers.

D.4 — Bridge isolation at the gate level

STOP
BRIDGE B1
START

Now ONLY bridge 1's gates should switch. Bridge 2's 4 gates should be silent (Q5/Q7 HS at 0 V; Q6/Q8 LS at ~14 V — freewheel state).

Scope to confirm. Then:

STOP
BRIDGE B2
START

Now bridge 2's gates switch and bridge 1's are silent. Confirm.

STOP
BRIDGE BOTH

D.5 — Quick PSC gate check (still no DC bus)

MOD PSC
FSW 5000
MI 0.3
CONFIG                (confirm cntoff ~ 3199, lock=OK)
START

⚠️ Stop in 30 seconds maximum. This is just a sanity check — we don't want to thermal-soak TLP250s at 5 kHz without bypass-cap verification yet.

Scope PWM_1H (PA8) and PWM_3H (PB6) simultaneously: - Both should switch every 200 µs (5 kHz) - Rising edges offset by ~50 µs (90°) - Vgs amplitude 0–14 V, clean transitions

Touch-test TLP250s. They should still be cool. If any TLP250 is already warm after 30 s, STOP — that driver's bypass cap is inadequate and will fail under load.

STOP

D.6 — Verdict on TLP250 health

After Section D, all 8 TLP250s should be: - [ ] At room temperature - [ ] Producing clean Vgs waveforms (no ringing, no slow edges) - [ ] Maintaining dead-time on every transition - [ ] Responding to bridge-isolation commands correctly

✅ Pass D: gate drive is solid. Safe to add DC bus voltage.

❌ If any TLP250 is warm, oscillating, or has malformed Vgs — STOP. Apply fixes from HARDWARE_BRINGUP.md Phase 7b troubleshooting (bypass cap closer, shorter gate wires, etc.) and re-run Section D. Do NOT add DC bus voltage until D passes cleanly.


Section E — STAIR cascaded at 10 V (baseline / regression)

The OLD bench-validated PWM. If this doesn't work, the new branch has broken something — fix that before going further.

E.1 — Hook up DC buses and load

  1. Bench supplies OFF, output disabled
  2. Set both supplies: 10.0 V, current limit 500 mA
  3. Connect supply 1 to bridge 1 DC input (+ to + terminal, − to bridge 1 GND_HV)
  4. Connect supply 2 to bridge 2 DC input
  5. Connect 100 Ω dummy load across the cascade AC output through the ACS712 (Node_X1 → ACS712 → 100Ω → Node_Y2)

E.2 — Run STAIR

⚠️ Set VNOM to match your test bus voltage FIRST. The default protection thresholds are sized for a 50 V bus (UV trips below 40 V). At a 10 V test bus the firmware would trip undervoltage instantly and never leave PRECHARGE. VNOM 10 rescales UV/OV/IMBAL to 8 / 11.6 / 2 V — appropriate protection for a 10 V test.

Dashboard:

STOP
VNOM 10               (match the 10 V test bus — derives UV=8, OV=11.6, IMBAL=2)
MOD STAIR
FSW 500
BRIDGE BOTH
MI 0.95
CONFIG                (verify settings — check the $P line shows uv=8.00)
START

Wait for $A,RUN. If you get $F,0x01,UV instead, you forgot VNOM or set it wrong — STOP, fix VNOM, CLEAR, retry.

E.3 — Bring up bus voltage

  1. Bench supply 1: output ON. Watch ammeter — should be < 50 mA.
  2. Bench supply 2: output ON. Same.

Dashboard vdc1 and vdc2 should both read ~10 V.

E.4 — Scope the cascade output

Set scope: - Vertical: 5 V/div (cascade should swing ±20 V) - Horizontal: 2 ms/div - Vsource: differential probe or A−B math

Expect a 5-level staircase: clear voltage steps at −20, −10, 0, +10, +20 V, each held for various integer multiples of 2 ms. Total waveform period 20 ms (50 Hz).

E.5 — Touch test after 1 minute

After 1 minute at 10 V: - Bridge 1 MOSFETs: noticeably warm - Bridge 2 MOSFETs: cooler than bridge 1 - TLP250s: still cool to touch - DC bus caps: ambient

This thermal asymmetry between bridges IS the concern #8 problem that PSC fixes. Confirming it here proves the imbalance is real and your hardware is producing it.

E.6 — Stop

STOP

Both bench supplies OFF.

✅ Pass E: STAIR works as expected. Baseline established.


Section F — STAIR_ALT cascaded at 10 V (verify the easy fallback)

Same setup as E. Confirms STAIR_ALT compiles and runs and that the bridge alternation logic is doing what it should. VNOM persists from Section E (it is not reset by STOP or a modulator change), so the protection is still scaled for the 10 V bus — but if you jumped straight to Section F, send VNOM 10 first.

F.1 — Switch to STAIR_ALT

MOD STAIR_ALT
CONFIG                (verify mod=STAIR_ALT; $P line still shows uv=8.00)
START

F.2 — Bring up bus voltage

Both supplies ON at 10 V, 500 mA limit.

F.3 — Scope

Same as E.4 — should be identical 5-level staircase output. STAIR_ALT doesn't change the output shape, only which bridge handles each ±1 step.

F.4 — Thermal check after 2–3 minutes

This is where STAIR_ALT differs from STAIR: - Bridge 1 MOSFETs: warm - Bridge 2 MOSFETs: warm (more than under STAIR) - The two bridges should converge toward similar temperature

You're alternating ±1-step ownership ~12 times per fundamental cycle, so over 2–3 fundamental cycles each bridge has done the same work.

F.5 — Stop

STOP

Both bench supplies OFF.

✅ Pass F: STAIR_ALT works, bridges balance. This is your no-PSC-needed fallback for the demo if PSC misbehaves.


Section G — PSC single bridge at 5 V (low-risk PSC entry)

⚠️ Drop the voltage to 5 V here. PSC at 5 kHz is more demanding on every component than STAIR at 500 Hz. We're starting low so any first-time PSC failure is recoverable.

G.1 — Set supplies to 5 V

Both supplies: 5.0 V, current limit 200 mA, OFF.

G.2 — PSC bridge 1 only

STOP
VNOM 5                (match the 5 V test bus — derives UV=4, OV=5.8, IMBAL=1)
MOD PSC
FSW 5000
BRIDGE B1
MI 0.5
CONFIG                (check: cntoff=3199, lock=OK; $P line shows uv=4.00)
START

❌ If lock=BAD, STOP. The 90° shift didn't take. Try MOD STAIR, then MOD PSC again (toggles cause a re-init). If still BAD, see HARDWARE_BRINGUP.md PSC carrier shift troubleshooting.

G.3 — Energize bridge 1 only

Bench supply 1: ON. Watch ammeter — should be < 50 mA. Do NOT turn on supply 2 yet.

G.4 — Scope cascade output

Set scope: - Vertical: 2 V/div - Horizontal: 50 µs/div - Persistence mode if available

Expect 3 distinct voltage bands: −5 V, 0 V, +5 V. PWM switching at 5 kHz between them. Because only bridge 1 is active, only 3 levels are possible.

G.5 — Verify bridge 2 silent

Scope bridge 2's MOSFETs gates briefly. Q5/Q7 (HS) should sit at 0 V. Q6/Q8 (LS) should sit at ~14 V (freewheel). No switching.

G.6 — Touch test after 30 s

  • Bridge 1 MOSFETs: warm
  • Bridge 2 MOSFETs: cool (ambient — only LS conducting freewheel current, which is ~0 with no current through the silent half)
  • TLP250s: still cool

G.7 — Stop

STOP

Supply 1 OFF.

✅ Pass G: PSC works on a single bridge at low voltage.

G.8 — Repeat for bridge 2

BRIDGE B2
START

Supply 2 ON (not supply 1).

Same expectations: 3-level output, bridge 1 silent, MOSFETs warm only on bridge 2.

STOP

Supply 2 OFF.

✅ Pass G (full): both bridges run PSC independently.


Section H — PSC cascaded at 5 V (the 5-level moment of truth)

The single most important test in the session. If this passes, the project deliverable is achieved.

H.1 — Re-arm cascade mode

VNOM 5 persists from Section G, so protection is still scaled for the 5 V bus. (If you jumped here, send VNOM 5 first.)

STOP
BRIDGE BOTH
MI 0.5
CONFIG                (sanity check: $C lock=OK, $P uv=4.00)
START

H.2 — Energize both supplies SIMULTANEOUSLY

Both supplies still at 5 V, 200 mA limit.

Either supply 1 and supply 2 both have a master "Output Enable" you can press together, or have a helper press one while you press the other. Both bridges must come up within ~100 ms of each other.

If they come up separately, one bridge will briefly drive the cascade against the other bridge's freewheel — not damaging at 5 V, but it muddies the test.

H.3 — Scope the cascade output

Same scope settings as G.4 (vertical 2 V/div, horizontal 50 µs/div, persistence on).

Now expect 5 distinct horizontal voltage bands: - −10 V (both bridges at −Vdc) - −5 V (one bridge at −Vdc, other at 0 — the new intermediate level!) - 0 V (both at 0 or balanced) - +5 V (one bridge at +Vdc, other at 0) - +10 V (both at +Vdc)

The density of points at each level varies over the fundamental cycle — more time at ±10 V near sine peaks, more time at 0 V at zero crossings.

Scope shows Diagnosis Action
5 distinct bands as described ✅ PSC working, project deliverable achieved Proceed to Section I
Only 3 bands (−10, 0, +10) 90° phase shift not effective on hardware despite lock=OK STOP supplies. Try toggling MOD STAIR → MOD PSC. If still 3-level, see HARDWARE_BRINGUP.md Fix A/D. Worst case fall back to STAIR_ALT and ship that
Asymmetric (some levels stronger than others) Bridge mismatch — DC supplies not equal, or one bridge's gate drive marginal STOP. Verify both supplies exactly equal, re-scope gates
Constantly faulting Sensors tripping under PSC switching noise STOP. EMI from PSC may be coupling into MCP3201 SPI. See troubleshooting

H.4 — DC bus current sanity

Watch both supply ammeters. They should be similar within ~20 %. If one bridge is drawing significantly more current than the other under PSC with BRIDGE BOTH, the bridges are not symmetric in hardware — investigate before raising voltage.

H.5 — Touch test

After 30 s: - Both bridges' MOSFETs: warm, similar temperature to each other - TLP250s: cool to slightly warm — both bridges' TLP250s should be similar to each other. Any single TLP250 that's noticeably hotter than the others is the one that will fail first — STOP and investigate before raising voltage - Snubber resistors (22 Ω 2 W): warm (they dissipate switching energy)

H.6 — Stop

STOP

Both supplies OFF.

✅ Pass H: 5-level cascade output confirmed at low voltage. The PSC modulator works on your hardware. Ready to ramp.


Section I — PSC voltage ramp (5 V → 50 V)

Stepwise increase of bus voltage. At each step, run briefly, confirm nothing is degrading, then continue.

⚠️ At any step, if you see: - Smoke, smell, or audible noise from the breadboard → STOP IMMEDIATELY - Any MOSFET hotter than ~70 °C → STOP, let cool - Any TLP250 hotter than ~60 °C → STOP, check bypass cap on that driver (will fail at higher voltage if hot now) - Fault trip that won't CLEAR after the condition is gone → STOP, investigate - Output waveform losing one of the 5 levels → STOP, may indicate a MOSFET starting to fail - Bus voltage drooping more than 10 % under load → cap inadequate, STOP before damage

I.1 — Ramp procedure (repeat for each voltage)

For each V_target = 10, 20, 30, 40, 50 V:

  1. STOP from previous step
  2. Both supplies OFF
  3. Set both supplies to V_target
  4. VNOM <V_target> — rescale protection to the new bus voltage. This is critical at every step: at V_target = 20 V with VNOM still at 5 (from Section H) the firmware would trip overvoltage (20 V > 5×1.16 = 5.8 V) the instant you energize. Always set VNOM = the bus voltage you are about to apply. Confirm the $P line.
  5. Adjust current limits:
  6. V ≤ 20 V: 500 mA each
  7. V = 30 V: 1 A each
  8. V ≥ 40 V: 2 A each (or your load's max)
  9. START (mode still PSC, FSW 5000, BRIDGE BOTH, MI 0.5)
  10. Both supplies ON simultaneously
  11. Scope cascade output — 5 bands present at ±2·V_target
  12. Watch dashboard for ~30 s:
  13. No fault lines
  14. vdc1, vdc2 track supply settings
  15. iout peak ≈ 2·V_target·MI / 100 Ω (e.g. at 30 V → ~300 mA)
  16. Touch-test MOSFETs and TLP250s briefly
  17. STOP, both supplies OFF
  18. Wait 30 s for thermal recovery
  19. Next voltage

I.2 — Pass criteria for each step

  • 10 V: pass
  • 20 V: pass
  • 30 V: pass
  • 40 V: pass
  • 50 V: pass

✅ Pass I: PSC operates cleanly at full bus voltage.


Section J — Sustained PSC run at 50 V

Final characterization. 5–15 minutes at design conditions.

J.1 — Final config

STOP
VNOM 50            (full design bus voltage — UV=40, OV=58, IMBAL=10)
MOD PSC
FSW 5000
BRIDGE BOTH
MI 0.95            (full modulation depth)
FFUND 50.0         (50 Hz fundamental)
OC 15              (overcurrent trip at design value)
CONFIG             (final check: $C lock=OK, $P uv=40.00 ov=58.00)
START

Both supplies at 50 V, current limits set to your expected load current + margin.

J.2 — Run for 5–15 minutes

Monitor continuously. Log every 2 minutes: - Bridge 1 heatsink temperature - Bridge 2 heatsink temperature - TLP250 package temperatures (sample one per bridge) - DC bus cap temperatures - Supply currents - Any UART events

J.3 — Pass criteria for the sustained run

  • Bridge 1 and Bridge 2 heatsinks within 5 °C of each other at steady state
  • All TLP250s under 70 °C (well under 100 °C absolute max)
  • All MOSFETs under 80 °C (with TO-220 clip-on heatsinks this is easy at the project's 700 W target)
  • DC bus caps under 60 °C
  • Zero fault trips during the run
  • Scope shows clean 5-level output for the entire run, no level degradation or asymmetry developing over time

✅ Pass J: PSC is bench-validated for the demo. You're done.

J.4 — Stop and shut down

STOP

Both bench supplies OFF, then disconnect.

Disconnect gate-drive supplies.

Save the dashboard event log to disk for the project report: File → Save Log... → name it bench_session_YYYY-MM-DD.txt (or just copy-paste the event panel contents).


Section K — Making PSC the default

If Sections G–J all passed, you may want PSC to be the auto-start default for the demo (so the inverter comes up in PSC mode on a fresh boot, no UART needed).

Edit Core/Inc/pwm_config.h:

#define PWM_DEFAULT_MODULATOR       MODULATOR_PSC       /* was MODULATOR_STAIR */
#define PWM_DEFAULT_SWITCHING_HZ    5000u               /* was 500u */

Rebuild and re-flash. Boot now starts in PSC mode and auto-start (if no UART) fires PSC at 5 kHz with cascade mode and MI 0.95.

Verify by resetting and watching the $C line after boot — should show mod=PSC,fsw=5000.


"What's burning?" — quick triage if something goes wrong

If you see, smell, or hear something bad STOP IMMEDIATELY (bench supply Output Off button). Then diagnose by symptom:

Symptom Likely cause Action
TLP250 hot (>70 °C) Bypass cap inadequate, gate ringing at 5 kHz Add 100 nF directly on Pin ⅝. Verify with scope at MI 0.3 in Step A
TLP250 instantly hot on power-up Shorted output / MOSFET gate-source short Replace MOSFET, check 10 kΩ gate-pulldown installed
MOSFET hot fast / smoking Shoot-through or stuck-on Verify dead-time on scope (3 µs default for IRFB4110). If breadboard gate drive is slow, bump PWM_DEAD_TIME_DTG to TIM_DTG_4US_AT_64MHZ and reflash
MOSFET hot evenly over time High switching loss at 5 kHz Better heatsink, lower FSW (FSW 2000), or lower MI
One bridge much hotter than other Bridge mismatch in hardware (different MOSFET, missing snubber, etc.) Inspect both bridges component-by-component
DC bus cap warm High ripple current — PSC stresses bus caps more than STAIR Add another 1000 µF in parallel right at the bridge DC input
78L05 hot Sensing island drawing too much Check for short on +5V_Island rail
Smell of burning insulation Wire too thin for current, or insulation contact with hot part Inspect every wire, replace if discolored
Pop + smoke Capacitor reversed or overvoltage Definite component replacement, inspect what caused overvoltage

Never-do list

  • ❌ Never apply > 80 V to a single bridge DC bus. The IRFB4110 MOSFETs are 100 V rated and the 1.5KE62A TVS clamps spikes at 84.5 V, so the design target of 50 V per bridge has comfortable margin — but don't push the steady-state bus near the TVS clamp voltage
  • ❌ Never increase MI above 0.95 (firmware enforces — but don't try)
  • ❌ Never run at > 100 % HS duty for any leg (firmware enforces 95 % clamp — but verify by scoping CCR if you change anything)
  • ❌ Never start with both supplies at significantly different voltages
  • ❌ Never short the AC output (no resistor, just a wire) under any modulator
  • ❌ Never START if $C shows lock=BAD and the modulator is PSC — that's broken
  • ❌ Never assume a passing Phase E (STAIR) means PSC will work — different physics, different stress
  • ❌ Never walk away from the bench mid-run

Always-do list

  • ✅ Always have the dashboard connected before flashing (auto-start protection)
  • ✅ Always start at low voltage and ramp gradually
  • ✅ Always check lock=OK before starting PSC
  • ✅ Always have both supplies' Output buttons reachable
  • ✅ Always touch-test components frequently (back of hand only — never grip a hot part)
  • ✅ Always STOP between voltage steps to let things cool briefly
  • ✅ Always save the event log after a session

Post-session

  • Disconnect everything in reverse order (load → DC bus → gate drive → USB)
  • Save dashboard log
  • Note observations (especially: did PSC produce 5 levels? bridge thermals matched?)
  • Decide which modulator goes into the demo:
  • PSC if Sections G–J all passed → most impressive (real PWM, 5 levels)
  • STAIR_ALT if PSC didn't deliver 5 levels → safest fallback (5 levels visible, bridges balanced, not real PWM)
  • STAIR if you had problems with both → original behavior, has the bridge imbalance but proven reliable

Update CHANGELOG.md with the session outcome.

If PSC works, you've solved concern #8 and met the project's 5-level PWM-controlled deliverable. That's the headline result.