Skip to content

Iteration 4 — as-built (two single-bridge modules, IRFB4110, PSC)

Five distinct cascade levels at 100 V — the iteration-4 demonstration result
The iteration-4 deliverable on the scope — five distinct cascade output levels at 100 V under sustained PSC-PWM, no filter. Both bridges thermally matched, no protection trips during the run.

Status: as-built. Bench-validated and demonstrated. KiCad project at hardware/single-bridge-v4/kicad/, gerbers at hardware/single-bridge-v4/gerbers/, photos at hardware/single-bridge-v4/photos/ and docs/assets/images/ (demo day).

What was attempted

Iteration 4 was a rebuild from the iteration-3 lessons rather than a tweak on top of iteration 3. Six structural changes:

Change From To Rationale
Board topology Single dual-bridge PCB Two identical single-bridge PCB modules Modularity; each cell can be bench-tested in isolation; bridges become interchangeable.
Stack-up 2-layer 4-layer, 1.6 mm FR-4, JLCPCB Dedicated ground planes solve the 5V_GND ↔ 50V_GND coupling from iteration 3.
Power MOSFET IRFZ44N (55 V) IRFB4110 (100 V, 4.5 mΩ) VDSS headroom; resolves the TVS clamp / MOSFET breakdown mismatch from iteration 1.
Modulator IPD LS-PWM at 500 Hz PSC-PWM at 5 kHz Bridge-thermal symmetry; better filter behavior; better textbook THD.
Dead time 2 µs (IRFZ44N) 3 µs (BDTR.DTG = 0xA0) Compensates for the IRFB4110's ~2× gate charge.
MISO topology 3-independent assumption 2 MISO lines (lower-island dedicated, upper-island shared DC2 + CUR) Reflects the actual as-fabricated routing. Firmware rewritten to sequential one-CS-at-a-time reads.

The firmware that drives this hardware lives on the pwm-rewrite-configurable branch — imported into this repo via git subtree with full history.

What worked

The bench validation confirmed every headline goal:

Goal Result
5 distinct cascade output levels visible on scope ✅ — see the scope-pwm-cascade-output.jpeg capture in populated photos.
PSC carrier phase lock ✅ — $C,...,lock=OK reported on every config line and across runs.
Bridges thermally matched ✅ — within ~3 °C under sustained PSC load.
Sensing clean throughout the run ✅ — no false SENSOR_LOST events, STATUS line clean over multi-minute sessions.
Dashboard auto-cancels firmware auto-start ✅ — STATUS echoed on connect and on every BOOT_SELF_TEST_DONE; operator always in control when the dashboard is open.
Protection chain protects ✅ — TVS clamps below MOSFET VDSS; UV/OV/OC/IMBAL with N-of-M debounce; FAULT_OUT pin pulled LOW on trip.

What's still soft (carries over to the roadmap)

  • No LC output filter. The bench demo ran into a resistive load. For non-trivial loads, the LC filter is the next item.
  • Open-loop control. Modulation index is set by the operator. Closed-loop control is the natural follow-on.
  • No enclosure. Open-bench operation only. See thermal enclosure.
  • Bench-only verification. No compliance testing, no grid-side integration. See grid tie and product path.

What to remember about iteration 4

  • It is not an incremental fix to iteration 3 — it's a re-architecture motivated by what the earlier iterations made painful.
  • The modular two-PCB-module decision is the most important structural choice; everything else hangs off it (interchangeable boards, single fab order, simpler bring-up procedure).
  • The IRFZ44N → IRFB4110 swap is the kind of mid-project component change that needs the firmware (dead time) and the protection chain (TVS clamp vs. MOSFET VDSS) to be updated together. Both were.
  • The PSC carrier-shift lock diagnostic was added defensively before the bench session — it caught a real issue on day one and saved an afternoon of scope time.