RC (surface) racing transponder with STM32C011, featuring:

• OpenStint protocol reference implentation (this is not an RC3 clone!)
• 5 MHz BPSK using SPI MOSI
• MOSFET driver repurposed as a high-current source for the antenna coil
• Tuned antenna, signal level is ca. 50% of the commercial deal when measured with a smallish near-field H-probe.
• Reverse polarity protection
• Power line filtering (TODO: LISN measurements....)
• LCSC product codes added.

Related projects: OpenStint decoder with HackRF One | Loop Amplifier

JLCPCB manifactures and assembles 5 panels of 2x4s, grand total of 40 pcs, for $130, including taxes and shipping.

The firmware compiles with cmake (and downloads the necessary STM32 libraries meanwhile). Make sure to use Release build: creating the BPSK signal with the right timing is critical, and -O0 optimalization level does not guarantee that. If you see signal coming out from the transponder, but it can not be decoded, this should be the first thing to check.

cmake -DCMAKE_BUILD_TYPE=Release .
make

st-flash --connect-under-reset --format ihex write transponder.hex

The programming points are separated by 2 mm. I use a pinheader of the same pitch with pogo pins attached to them as a makeshift programming adapter.

This project use a FET driver from TI to produce the necessary antenna current. This IC needs 4.5 V minimum to operate, meaning we can not use this transponder with 1s batteries (1/12 pancars). I could not find any push-pull driver which can operate at 3.3 V and 5+ MHz. Some MCUs exists which can source/sink 50 mA directly from GPIO, but as of now, I don't know how to make this transponder work with such a low current. This design requires ±170 mA, and still produces only 1/2 of the signal level as an RC4 hybrid.

Having a design that works with 50 mA would have benefits, like:

• 1s battery support
• LDO integrated with reverse battery protection (reduced component count)
• No need of UCC27517 driver (reduced component count)
• Fewer components perpendicular to the magnetic field (reduced eddys)

Unfortunately, I do not have the required analog brain to figure this out. I feel we should move to a 6-layer board: the docs/antenna2.asc ltspice sim suggests that increasing the turn count reduces the peak current less than the extra turns contribute to the magnetic moment (~N*I*A). Unfortunately, this is just a guess, and I'm not really enjoying the development based on trial-and-error.

UPDATE 2025-12-11: I kept reading on NFC and alike topics. I think there are two ways forward:

1. Keep the current design, it's not that bad after all. I see 3 areas for improvement: signal conditioniong (not sinusoidal enough), signal strength (minor problem) and board space (also minor). We can push the gate driver a bit further, up to the 4-500 mA range (I've done that before), solving the signal strength problem. Most NFC designs I've seen adds some LC-network between the push-pull stage and the antenna's LC-tank (improving both on power transfer and spectral purity). By removing the antenna loop from the top layer we can gain some board space (layer 2..4 should have 3-3-3 loops, 9 in total). We can move the components on top layer above the antenna wires, where the magnetic field is less perpendicular to the components (less eddy, more board space). No need for 6 layers. On the negative side, it can't support 1s batteries :(.
2. Experiment with a totally different topology. The current limitation after all is the current which the FET driver can safely push through the antenna loop. Any phase difference between the driving signal and the LC-tank forces the driving current to go extreme. This is less of a problem with a single discrete transistor. If we add (+) a positive feedback from the LC-tank to the control signal of the MCU, we can approximate zero-voltage switching. By tuning the feedback*amplification to below-1, only the control signal could bring the circuit into oscillation. I'm looking at you, Meissner oscillator...

See relases for gerber, pos and BOM files. These are directly uploadable to JLCPCB. As of now (2025-12-10), manufacturing 5 panels (40 transponders) with assembly costs $160, while 20 panels (160 pcs of transponders) cost $300 (prices with shipping and taxes to EU).

Manufacture with 1 mm PCB width (so coil tuning matches mine).

kikit panelize \
    --layout 'hspace: 3mm; vspace: 3mm; rows: 4; cols: 2' \
    --tabs 'type: fixed; vwidth: 4mm; hwidth: 4mm; vcount: 0' \
    --cuts 'type: mousebites; offset: -0.25mm' \
    --framing 'type: frame; hspace: 3mm; vspace: 3mm; width: 4mm' \
    --post 'millradius: 1.5mm' \
    ...

kikit fab jlcpcb --assembly --schematic ./openstint-transponder.kicad_sch ./panel/panel-8.kicad_pcb panel/