RJ-45 Pin Convention
RS-485 on two pairs, the inverter’s signal ground on pins 3 and 6, DC input on pins 4 and 5. Every cable, board and harness we make follows this. Revised 2026-09-03 — it supersedes the 2026-08-29 convention and is not compatible with it.
Contacts facing you, latch down, pin 1 on the left · colours are T568B
The pinout
| Pin | T568B colour | Function | Notes |
|---|
Go by pin number, never by colour. These are T568B colours. Crimp T568A instead and the pins keep these functions, but the orange and green pairs swap colour — so a cable that looks wrong can be right, and one that looks right can be wrong.
Mapping to your inverter
Every brand numbers its own communication port differently, and none of them will match this socket. Go by the signal, never by the pin number — this is the single easiest thing to get wrong. Huawei is shown here as a worked example; look up your own brand in the documentation.
Note the inversion in this example: Huawei carries A+ on its pin 1, we carry it on pin 2. Wiring “pin 1 to pin 1” gets you a reversed pair — and it will be a different inversion, or none, on another brand.
Reversed A/B is silent, not destructive
The bus simply goes quiet — no damage, no error, nothing in a log. If a scan finds nothing, swapping the two is the first thing to try, and it costs nothing.
Power
Pins 4 and 5 carry 5–30 V DC, pin 5 positive. Note the polarity — positive is the higher-numbered pin. This pair is ours; the rest of the socket follows Deye’s scheme, but Deye leaves these two alone.
Feed the RS-485 transceiver from 5 V, not 3V3. The module regulates internally and 3.3 V sits right at its dropout — it will appear to work on the bench and then fail under load or temperature.
Never plug this into a network port
This connector is RJ-45 shaped and is not Ethernet. Pins 4 and 5 carry DC, and on gigabit those two pins are an active data pair — so a Svitgrid cable in a switch, router or laptop puts up to 30 V straight onto it.
Label both ends of every cable you make.
Signal ground, and the duplicated pair
Pins 3 and 6 carry the inverter’s ground, where it offers one. RS-485 is differential but not isolated: a receiver only decodes A against B while both sit inside its common-mode window, referenced to its own ground. The pair carries the signal; this carries the reference that makes it interpretable.
Plenty of communication ports expose only A+ and B−. Where there is no ground to take, pins 3 and 6 are unused and the link runs two-wire — which is not wrong so much as undefined: it works whenever both ends happen to sit close enough in common mode, and when it stops the bus simply goes quiet. Take the ground when the inverter offers it.
A+ and B− each appear twice — on pins 2 and 7, and 1 and 8 — so a run can be passed on without a splice. They are the same net, not two buses: two inverters across the two pairs are two slaves on one bus and need different slave ids.
Pin 8 used to be earth. It is now B−, so a cable made to the old convention shorts B− to earth here and the bus goes quiet with nothing in any log. Re-terminate old cables rather than adapting them.
The firmware side
RS-485 lands on GPIO6 (TX) and GPIO5 (RX), on UART2.
Never GPIO43 / GPIO44
Those are UART0’s IOMUX pads. IOMUX overrides the GPIO matrix, and the transmitter dies partway through boot while the write call returns the full byte count and reports success. The wire stays silent and nothing reports an error.
When wiring and firmware disagree, run the pad scan. It transmits from every broken-out pad with the pad number in the Modbus slave byte, so a listener on A/B names the pad the firmware is actually driving. A meter cannot do that — it confirms a net without saying which pad is being driven.