The operation is very simple: the track current is rectified, then feeds the LEDs wired in series on the chosen end thanks to the DIP switch. An anti-flashing capacitor circuit is provided.
The actual lamps have a diameter of about 300 mm, so 3.4 to the scale. The car’s holes are only 2.5 mm. They are then carefully drilled to 3 mm to receive LEDs of this same diameter.
These LEDs have their hemispherical end levelled with a file in a template consisting of a 3 mm thick plate. They are connected by a wire which goes around the gangway door, for a maximum of discretion.
The resistor value is to be determined by a test, either on a breadboard or with components soldered in the air. For the test, the used resistor is a 47 or 100 kΩ trimmer, taking care not to set it too close to zero! To avoid any risk of LED breakdown, put a fixed resistor of about 1 kΩ in series with the trimmer.
The current pickup is done by my usual technique of conductive bearings. As I had sliced the bogies at the time in order to refine them, I only had to force a bit to separate the parts and thus facilitate the boring of the axle boxes to place brass bearings. The axles are probably of LS Models origin, and are isolated in their middle.
In this image, between the two boxes, you can see the double DIP switch used to switch on the lamps. Its positioning required, as already said, milling the chassis at this location.
This circuit, consisting of a rectifier, a diode, a resistor, and a capacitor, so very simple, did not require a specific printed circuit. It is mounted on a perfboard.
Here is the complete wiring of the car, seen from above. The lamps are plugged for testing, but they are normally attached to the body. Two connectors based on tulip contacts are provided to easily separate the body from the chassis.
Note that the wire colours meet the NMRA requirements for decoder wiring.
In this picture, there are small steel plates, 16 mm wide, weight 6 g, which are glued close to the pivots. This ballast is just enough to reach the NEM 302 recommended value.