Solutions

RS485 meter-room wiring: A/B, shield and common mode

A meter-room RS485 bus usually fails in month three, not on day one. Termination, bias math, common-mode budget, and the LED codes that name the broken wire.

Most RS485 buses in a meter room work perfectly on the day they are commissioned. They fail in month three, after a hot summer, when one meter on the far end stops answering and the building manager says the radio "went bad". The radio is usually fine. What changed is a few hundred millivolts of common-mode voltage, or a termination resistor that was never fitted, or a shield that got bonded at both ends and is now carrying the difference between two earth bars.

This is the part of a wireless meter-reading project that has nothing to do with radio: the copper between the collector and the meters. It is also the part that the device datasheet does not cover. Our TP6306 collector spec gives you a pin table, a wire colour for each terminal and an IP rating. It does not give you termination, biasing, or an isolation figure. This walkthrough fills those in, with the arithmetic shown so you can redo it for your own bus.

Meter-room RS485 bus with a TP6306 collector, daisy-chained DL/T 645 meters, 120 Ω terminators at both ends and a shield bonded at one point only

The bus is a transmission line, not a wire

RS485 is defined by a handful of numbers, and every field problem traces back to one of them. The driver is specified to develop at least 1.5 V across the pair. The receiver only needs 200 mV to decide a state. That leaves 1.3 V of differential budget, which is 17.5 dB of margin — comfortable, and the reason RS485 survives in electrically filthy rooms. The number installers forget is the common-mode window: the standard defines a range of −7 V to +12 V, and TI's AN-1057 on bulletproofing RS-485 interfaces is blunt about what happens when two nodes sit at different earth potentials. Differential margin does not help you if the whole pair has drifted outside that window, because the receiver stops being defined at all.

Texas Instruments' SLLA070D application report states all of them directly: up to 32 unit loads on one bus with a single driver active at a time, about 1200 m of cable as a practical maximum, 1.5 V minimum differential drive, 200 mV receiver threshold, and the −7 V to +12 V common-mode window. Those are the numbers to design against; everything else on a meter-room drawing is a consequence of them.

RS-485 common-mode window: receivers stay defined between −7 V and +12 V, while a differential margin of 1.3 V is a separate budget

The distinction matters because the two budgets are spent by different things. Differential margin is eaten by cable attenuation and reflections: at 9600 baud over 180 m of twisted pair you will not notice it. Common-mode budget is spent by the building's electrical installation — earth potential differences, a variable-frequency drive on the same tray, a neutral that is not bonded where you assumed it was.

Wiring table: what connects to what

The TP6306 is deliberately boring to wire. The vendor spec lists each terminal with its factory wire colour, which is worth keeping in the drawing because the colours are the only thing an electrician on a ladder can check without a laptop.

Terminal

Function

Factory wire

Connects to

3.1

RS485 B

Green 20 AWG VW-1

Meter B (often labelled 485−)

4.1

RS485 A

Yellow 20 AWG VW-1

Meter A (often labelled 485+)

5

220 V AC −

Black 20 AWG VW-1

Neutral, same circuit as the panel

6

220 V AC +

Red 20 AWG VW-1

Live, via the panel breaker

3.2

12 V DC +

White, 0.35 mm² (TP6306-WL only)

Cold-water meter bus supply

4.2

12 V DC −

Brown, 0.35 mm² (TP6306-WL only)

Cold-water meter bus return

Two details in that table are easy to get wrong. First, A and B are not universal: some meter vendors print A/B, others print D+/D−, and the mapping between the two is not consistent across manufacturers. If a whole bus reads nothing, swap A and B at the collector before you touch anything else — it costs ten seconds and it is the single most common cause of a "dead" RS485 port. Second, the collector's 12 V output exists only on the -WL variant, which converts RS485 to M-Bus for cold-water meters. Do not plan a 12 V sensor supply around a -EM unit; it does not have one.

The variant table matters more than it looks, because the protocol decides the polling shape:

Variant

Meter type

Protocol

Slaves

TP6306-WL

Cold water meter

RS485 to M-Bus, CJ/T 188

32

TP6306-WH

Water control unit

Private protocol

32

TP6306-EM

Electricity meter

DL/T 645

32

TP6306-SE

Sensor terminal

Modbus RTU

6

TP6306-SE-A

Sensor terminal

Modbus RTU (project variant)

6

The spec does not say why the -SE variants stop at 6 while the meter variants carry 32, but the arithmetic explains it: Modbus RTU sensors are polled one at a time, and a sensor that takes 200 ms to answer will not let you scan 32 of them inside a useful interval, whereas meter protocols are more tolerant of long gaps between transactions.

Shield: bond it once, and only once

The shield is an EMC barrier, not a signal return. Bond it to protective earth at the collector end, leave it open at the far end, and insulate the open end so it cannot touch a metal enclosure. Bonding both ends creates a loop, and in a building with more than one earth bar that loop carries current — tens of milliamps is enough to add several volts of 50 Hz common-mode voltage to the pair, which is exactly the budget you were trying to protect.

If the far end is in a different electrical zone — another building, another transformer, a rooftop plant room — a single-bonded shield is no longer sufficient, and you need galvanic isolation on the port. Be explicit about this in design review: the TP6306 datasheet does not state an isolation voltage for its RS485 port, so treat it as non-isolated and keep the bus inside one electrical zone. Our transparent RS485 module, the TP5301, does publish an ESD class (±15 kV air, ±8 kV contact) but also does not publish an isolation rating. If you need a floating port, that is a spec you must ask for, not one you can infer.

There is a cheaper mechanical fix that solves half of these cases before they start: keep the RS485 pair and the 220 V feed in separate cable glands and separate conduit. The collector is IP51 and DIN-rail mounted, so it will normally sit inside a panel with a good earth bond; the failure mode is not the collector, it is the installer who runs signal and mains in the same gland because it is one hole instead of two.

Common mode is a separate budget

Three situations account for almost every common-mode problem in a meter room.

The first is the shared neutral. A collector powered from one distribution board and a meter powered from another will sit at slightly different earth potentials even in a healthy installation. A volt or two is normal and harmless inside the −7 V to +12 V window.

The second is the drive. A variable-frequency drive or a soft starter on the same cable tray injects high-frequency common-mode noise that a meter's 9600-baud UART will not filter. The symptom is not a dead bus but an error rate that climbs with motor load. Twisted pair with a bonded shield, routed at least 300 mm from motor cables, fixes most of it.

The third is a fault, and it is the one that destroys ports. A lost neutral or a broken protective-earth bond can put mains potential on the pair. Nothing in a 20 dBm radio design protects against that; only isolation and a properly rated transient suppressor do.

Failsafe bias: the resistor nobody calculates

An idle RS485 bus with no driver enabled is undefined: A and B float to whatever the leakage currents say, and a receiver with 0 mV across its inputs can output either state. Every UART in the building then sees framing errors between frames. The fix is a pair of bias resistors at the master end that hold the idle differential above the 200 mV threshold.

The usual advice is "1 kΩ pull-up and pull-down", which sounds safe and is not: on a terminated bus that pair develops only about 145 mV of idle differential, comfortably inside the receiver's indeterminate band. TI's AN-847 on failsafe biasing works the same problem from the transceiver side. The calculation is three lines. Two 120 Ω terminators in parallel look like 60 Ω at DC; 32 unit loads at 12 kΩ each look like 375 Ω; in parallel they are 51.7 Ω. To develop 200 mV across that, the bias network has to source 3.87 mA from a 5 V rail, so the total series resistance is 1293 Ω and each leg is about 621 Ω.

Unit loads on bus

Bus resistance at bias point

Bias current

Resistor per leg

Bias network power

1

59.7 Ω

3.35 mA

716 Ω

16.8 mW

6

58.3 Ω

3.43 mA

699 Ω

17.2 mW

16

55.6 Ω

3.60 mA

667 Ω

18.0 mW

32

51.7 Ω

3.87 mA

621 Ω

19.3 mW

Two conclusions follow. The resistor value barely moves with slave count — 716 Ω to 621 Ω across the whole range — because the terminators dominate the DC load. And the power is trivial: under 20 mW. A 620 Ω resistor per leg is therefore the value to default to on a 5 V bias supply with 1-unit-load transceivers: it still develops 200 mV at 32 loads (the 680 Ω that gets quoted in most field notes only reaches 183 mV there, which is inside the receiver's indeterminate band), and it gives more margin on a shorter bus. If your transceivers are 1/8 unit load (96 kΩ), the DC load is lighter, the bus resistance rises toward the 60 Ω of the terminators, and the same 620 Ω develops about 226 mV — more margin, not less. Bias goes at one end only, the master.

Required bias resistor per leg and idle current versus the number of unit loads, computed for 200 mV idle differential on a 5 V supply

Two bias networks fight each other and double the idle current for no benefit. And bias is not a substitute for termination: on a 180 m bus at 9600 baud the terminators are still needed to stop reflections, even though the cable is nowhere near the 1200 m limit.

A worked example: 24 meters, 180 m, one collector

Take a small commercial building: 24 DL/T 645 electricity meters on three floors, 180 m of shielded twisted pair in a daisy chain, one TP6306-EM at the panel, 9600 8-N-1 on the downlink, and a 5-minute polling interval.

The Modbus RTU arithmetic (DL/T 645 frames are similar in size) is straightforward. A read request is 8 bytes; a two-register response is 9 bytes; 8-N-1 costs 10 bits per byte; Modbus RTU requires a 3.5-character gap between frames. Assume a 30 ms meter turnaround — conservative for a meter that is not doing anything else.

Line rate

Request

Response

Gap

Turnaround

Per meter

24 meters

2400 bps

33.3 ms

37.5 ms

14.6 ms

30 ms

115.4 ms

2.77 s

9600 bps

8.3 ms

9.4 ms

3.6 ms

30 ms

51.4 ms

1.23 s

19200 bps

4.2 ms

4.7 ms

1.8 ms

30 ms

40.7 ms

0.98 s

At 9600 the bus scan takes 1.23 s, and the turnaround time — not the line rate — is what dominates: doubling to 19200 saves only 250 ms because 30 ms of the 51 ms is the meter thinking. This is why raising the baud rate is rarely the answer to a slow meter room.

The radio side is where the airtime goes. The TP6306 uplinks at 2400 bps with 19200 bps downlink. Twenty-four meters at four bytes each, plus roughly twelve bytes of framing, is 108 bytes, or 450 ms of airtime. A full round is therefore about 1.7 s of bus plus air. At a 5-minute interval that is a 0.56 % duty cycle, which leaves plenty of headroom for retries and for the collector to answer an on-demand read from the gateway without dropping a scheduled cycle.

Cable loss at this length is not the problem people expect. 20 AWG copper is about 33 Ω per kilometre, so 180 m out and back is 12 Ω; the bias network draws under 4 mA, so the DC drop along the pair is about 47 mV — negligible against 1.5 V of drive. What a long bus costs you is high-frequency attenuation and reflections, and both are handled by termination and topology rather than by thicker copper.

Topology is worth stating as a rule: daisy chain, no star, no stubs longer than a few metres. A 20 m stub off the middle of the bus is a quarter-wave antenna at the wrong frequency and a guaranteed reflection. If the physical layout forces a star, put a repeater or a second collector at the centre rather than pretending the bus is a star.

Commissioning: read the LEDs before the laptop

The TP6306 has two LEDs, and the vendor spec defines them precisely enough that they replace a laptop for the first ten minutes of troubleshooting. The two-colour status LED reports both links at once:

Status LED

Meaning

Green, steady

UNB link OK, RS485 OK

Red, 1 s blink

UNB OK, RS485 fault

Red, 3 s blink

UNB fault, RS485 OK

Red, steady

Both links faulted

The green run LED handles power and joining: steady means power is good, a 1 s blink means it is joining the network, a 3 s blink means joining failed, and a 0.25 s burst for one or two seconds followed by steady means it joined successfully.

A red 1 s blink is the code you will see most often, and it almost always means A/B swapped, a broken pair, or a meter that is not powered. Work the bus from the collector outwards: disconnect all but the first meter, check the LED, then add meters one at a time. A single meter with a shorted A/B pair takes down the whole bus, and this is the only reliable way to find it.

Above the bus, the radio side needs a signal check. If you replace the collector with a transparent DTU — the TP5500 is the IP65 outdoor unit we use for that — commissioning moves to a serial terminal, because the DTU family exposes an AT command set that the collector does not:

AT+VER?                 -> firmware version
AT+UART=9600,8,N,1      -> match the meter bus exactly
AT+UART?                -> read back: UART:9600,8,N,1
AT+FREQ?                -> current channel index
AT+RSSI?                -> rssi:-60
AT+REBOOT               -> apply and restart
AT+DEF                  -> factory defaults (do this before you ship it back)

The one command that saves a site visit is AT+UART?. Nine out of ten "the radio is dead" reports on a transparent link are a baud rate mismatch between the meter and the DTU, and reading it back takes five seconds.

Failure modes, in the order they appear

Symptom

Most likely cause

Check

Whole bus dead from day one

A/B swapped, or no 220 V

Swap A/B at collector; measure 220 V

One meter missing, rest fine

Meter address, or its pair

Read that address alone with a USB-RS485 adapter

Errors climb in the afternoon

Common-mode from a drive or a shared neutral

Move the pair away from motor cables; check earth bonds

Works cold, fails hot

Termination missing, reflections

Fit 120 Ω at both physical ends

Random framing errors when idle

No failsafe bias

Add 620 Ω pull-up/pull-down at the master

Collector reboots when a meter is connected

Mains fault on the pair

Isolate; check for lost neutral

What the datasheet does not tell you

Three things are genuinely absent from the TP6306 specification, and you should design around them rather than assume them away. There is no stated isolation voltage on the RS485 port. There are no bias resistors and no stated termination. And the ESD class is published for the TP5301 module but not for the collector. None of these are defects — they are the normal boundary of a device datasheet — but a design review that assumes "RS485 is RS485" will discover all three in the field.

The practical version of that advice: keep one electrical zone per bus, fit 120 Ω at both ends, fit 620 Ω of bias at the master, bond the shield once, and write the wire colours on the panel door. The radio link will then behave exactly as the link budget predicted, and the only thing you will be debugging in month three is the meter that was always going to fail on its own.

Can I use ordinary 2-core alarm cable instead of twisted pair?

For a 20 m run at 9600 baud, usually yes. For 180 m with 24 meters, no: untwisted pair has no defined characteristic impedance, so the 120 Ω terminators no longer match the line and reflections show up as errors that appear only when several meters answer close together. Shielded twisted pair with a drain wire costs very little more per metre than alarm cable.

Do I need a repeater for 300 m with 24 meters?

No. 300 m at 9600 baud is well inside the 1200 m guidance, and the load is 24 unit loads against a limit of 32. You need a repeater when you exceed 32 unit loads, when the bus has to cross into another electrical zone, or when the topology forces a star. Length alone, at meter-room baud rates, is rarely the trigger.

What if the meters are on different 220 V phases?

That is normal and usually fine: the RS485 common-mode window of −7 V to +12 V is wide enough to cover the small earth potential difference between phases on the same transformer. It stops being fine when the zones come from different transformers, because then the earth potential difference can be tens of volts. In that case split the bus into two collectors rather than adding isolation to one long bus — the TP6306 is cheap enough that a second unit is the simpler engineering answer.

Is the shield the same as a ground wire?

No. The shield is a barrier against radiated interference; the ground wire is a fault-current path. Bonding the shield at both ends turns it into a parallel conductor for earth currents, which is why the open end must be insulated. If the installation needs an equipotential bond between the two ends, that is a separate conductor sized for fault current, not the drain wire of a signal cable.

If you are planning a meter-room rollout and want the wiring schedule checked against the meter protocol you are using, the full collector range lists the variant, slave count and protocol for each model, and our engineers will review a bus drawing before it goes to site.

More in Solutions

  • Solutions

    How do drones transmit images?

    The real-time transmission of images and video from an unmanned aerial vehicle (UAV) to a ground station is a cornerstone technology that enables applications…