Power management strategy for the XIAO nRF52840 Dreamcast BLE adapter running on a single-cell LiPo (tested with 500mAh, recommended 1000mAh).
Scope: every figure on this page was measured on a XIAO nRF52840 build. Retail Pulsar V1 hardware carries a different power topology — a separate always-on regulator and an IP5306 fuel gauge in place of the XIAO's voltage-curve estimate — so its standby figures are not the ones below and are characterized separately.
| State | Current Draw | Source | Notes |
|---|---|---|---|
| Active gaming | ~57-67 mA | Measured (FNB58) | Boost converter + controller + BLE radio |
| BLE advertising (slow) | ~0.5-2 mA | Estimated | 500ms interval, boost off |
| System Off | ~5-8 µA | Datasheet | QSPI flash in DPD, pins disconnected |
Tested with FNB58 USB power meter, active BLE connection with occasional controller input:
- Idle connected draw: ~57 mA
- Active input draw: ~67 mA
- Sleep drain: ~2% overnight, at the resolution limit of the gauge — consistent with self-discharge, but the measurement is not precise enough to separate the two.
Measured draw figures yield ~7-8 hours of active gaming on 500mAh, scaling to ~14-16 hours on 1000mAh.
| Battery | Active Gaming | Sleep Standby |
|---|---|---|
| 500 mAh | ~7-8 hours | weeks+ (calculated, not measured) |
| 1000 mAh | ~14-16 hours (estimated) | weeks+ (calculated, not measured) |
On the standby column: active-gaming hours are measured; standby is arithmetic from the ~5-8 µA System Off datasheet figure, which implies far longer than any test has run. Overnight testing cannot distinguish a µA-class sleep current from self-discharge, so treat standby as a lower bound on a correctly-configured board, not a verified result.
Note: The LiPo discharge curve is nonlinear. The battery spends a long time in the 3.7-3.9V plateau then drops quickly below 3.5V. Actual runtime may vary with controller usage intensity.
Voltage-based percentage using a 10-point LiPo discharge curve lookup table with linear interpolation between entries. This is the industry standard approach for embedded devices — Xbox controllers only report 4 discrete levels, and projects like Meshtastic use similar tables.
| Voltage | Percentage |
|---|---|
| 4100 mV | 100% |
| 4000 mV | 80% |
| 3900 mV | 60% |
| 3800 mV | 40% |
| 3700 mV | 30% |
| 3600 mV | 20% |
| 3500 mV | 10% |
| 3400 mV | 5% |
| 3300 mV | 0% |
- 100% = 4100mV — the BQ25101 charger often terminates before reaching 4200mV. Reporting 100% at 4.1V matches user expectations (same approach as phones/laptops)
- 0% = 3300mV — the battery protection circuit shuts down at this voltage under load (measured empirically: device dies at ~3.3V)
- 8x SAADC oversampling — hardware-averaged ADC reads for noise reduction
- Monotonic decrease — reported percentage never increases unless USB charging is detected, eliminating confusing voltage-recovery bounces after sleep or load changes
- 60-second read interval — uses
Instantcomparison for drift-free timing
Realistic accuracy is +/-10-15% in the flat middle region (3.7-3.9V), better at extremes. This is consistent with other voltage-based approaches and adequate for a battery indicator. A fuel gauge IC (e.g., MAX17048, around $2.80) would improve to +/-5% but only makes sense on a custom PCB.
The 5V boost converter + Dreamcast controller draws ~45 mA and serves no purpose when there's no BLE host connected. The boot flow now:
- Boot → BLE advertise (boost OFF, no controller polling)
- BLE connects → enable boost → detect controller → start polling
- BLE disconnects → disable boost → stop polling → advertise again
- Timeout → System Off
The XIAO's P25Q16H QSPI flash draws several mA in standby. We don't use it. At startup, a GPIO bit-bang SPI sequence sends the DPD command (0xB9), then disconnects the QSPI pins. CS (P0.25) stays driven HIGH to prevent accidental wake from bus noise.
This was the root cause of unexpectedly high System Off current (milliamps instead of microamps).
The XIAO has the inductor for REG1 DCDC (confirmed via Zephyr devicetree). Enabling it (config.dcdc.reg1 = true) saves ~2-3 mA during radio TX/RX. REG0 is NOT enabled (VDDH tied to VDD, no inductor).
GPIO state survives System Off on the nRF52840. Every unused pin is explicitly disconnected (input, no pull, Hi-Z) before entering System Off to prevent current leakage. Eight pins are disconnected; three are kept driven (QSPI CS HIGH, boost SHDN LOW, charge ISET LOW).
During BLE advertising (boost off), Maple Bus pins are also disconnected via MapleBus::set_low_power(). The external 10kΩ pull-ups hold both lines at 3.3V with zero current.
OutputDrive::HighDrive gives ~121ns rise time vs ~363ns standard drive. No power cost — purely signal quality improvement that enables reliable communication and potentially allows higher-resistance pull-ups in the future.
When USB power is present, the controller runs directly from USB 5V via Schottky diode OR circuit (2x 1N5817). The boost converter stays off and the battery charges at 100mA with nothing drawing from it. Tethered play is effectively free.
| Phase | Interval | Duration | Purpose |
|---|---|---|---|
| Fast reconnect | 20 ms | 5 seconds | Instant reconnect to bonded host |
| Slow reconnect | 500 ms | 55 seconds | Low-power reconnect window |
| Sync mode | 20 ms | 60 seconds | Active pairing (user-initiated) |
| Timeout | — | — | System Off after all phases expire |
If BLE connects but no controller is found within 60 seconds, the device enters System Off. This prevents the overnight drain scenario where a host auto-reconnects via BLE with no controller plugged in, leaving the boost converter running indefinitely.
After 10 minutes with no BLE connection (controller disconnected, host gone), the device enters System Off. Wake via sync button press (GPIO SENSE).
RTT debug logging is gated behind an rtt Cargo feature. DK builds always include it. XIAO production builds omit it — all log!() calls compile to nothing, reducing binary size and eliminating string formatting overhead. Development builds opt in with --features board-xiao,rtt.
On panic, the firmware writes the panic message to a dedicated flash page (0xF1000) using raw NVMC register writes (no SoftDevice dependency), then resets. On the next boot with RTT enabled, the stored panic is printed and the page is cleared. This replaces the silent panic-reset behavior with something debuggable.
Two 1N5817 Schottky diodes in an OR configuration route either USB VBUS or boost output to the controller 5V rail. When USB is present, the higher voltage wins (~4.7V after diode drop) and the boost shuts down. Firmware detects USB via the nRF52840 POWER.USBREGSTATUS register.
| Controller | Battery | Life | Avg Draw |
|---|---|---|---|
| Xbox One S | 2x AA (~2400 mAh) | 40 hr | ~60 mA |
| DualSense | 1560 mAh | 6-12 hr | ~100 mA |
| Switch Pro | 1300 mAh | 40 hr | ~32 mA |
| 8BitDo Pro 2 | 1000 mAh | 20 hr | ~50 mA |
| This project | 500 mAh | ~7-8 hr | ~60 mA |
| This project | 1000 mAh | ~14-16 hr | ~60 mA |
Our draw is dominated by the 5V boost converter + Dreamcast controller (~45 mA). The nRF52840 + BLE radio is only ~15 mA of the total.
- Slave latency — Setting BLE slave_latency to 2-4 could save ~200-300 µA during idle connected periods. Deferred due to BLE host-compatibility risk (e.g. with the iBlueControlMod Dreamcast BLE receiver adapter). Only effective if combined with skipping notifications when state is unchanged, which may break Xbox HID compatibility.
- Dedicated PCB — Eliminates perfboard losses and enables SMD components. Validated part choices (corrected from earlier notes):
- Boost: TPS61099DRVR (TI TPS61099x family, WSON-6/DRV adjustable boost configured for 5.0V, 300mA class, ~1µA Iq, ~90% eff). NOTE: there is no real orderable "TPS61099x50" PN; the TPS61099x suffix matters. TI's fixed 5.0V member is TPS610997, but the datasheet orderable list shows that member as DSBGA/YFF, not WSON/DRV. Use TPS61099DRVR for the WSON package and set 5.0V with the feedback divider from TI's 5V example (~1MΩ VOUT-to-FB, 249kΩ FB-to-GND). Critical: the firmware's ~45mA boost-gating saving depends on true output disconnect, so the selected SMD boost must have verified shutdown disconnect behavior or the carrier must add a load switch (high-side P-FET or load-switch IC, e.g. TPS22919) on the 5V rail driven by the enable line. A load switch remains useful for controlled slew, output discharge, and protection even if the boost has true shutdown.
- OR diodes: power Schottky (SS14, 1A, JLC Basic — or low-Vf PMEG2010), NOT BAT54. BAT54 is a small-signal Schottky (Vf ≈ 0.8V @ 100mA); at the ~50mA rail load it drops more than the 1N5817 it would replace. Earlier "~0.23V" figure only holds at low current.
- Fuel gauge IC — MAX17048 (Analog/Maxim, I2C, DFN-2x2, JLCPCB
C2682616, ~$2.80) for +/-5% battery accuracy vs current +/-10-15%. Only practical on a custom PCB. - TX power reduction — Lowering BLE TX from 0dBm to -4dBm would save ~1mA with no perceptible range impact at gamepad distances. Deferred pending range testing through plastic enclosure.