Solutions

Metering network commissioning: a 12-point checklist

Twelve measurable checks for a metering network: variant match, two-way link budget, poll arithmetic, and a 24-hour soak before handover.

Metering network commissioning: a 12-point checklist

A metering network rarely fails on the day it is installed. It fails three weeks later, at two in the morning, when one collector stops answering and nobody can tell whether the meter, the bus, the radio or the database is at fault. By then the installer has left, the cabinet key is with the property manager, and the only evidence is a gap in a consumption curve.

The checklist below is the one we run before a network is accepted. It has twelve points in four phases — pre-power, bus and power, radio path, data path — and every point has a measurable pass criterion, because "looks fine" is what the installer says before the first winter. Two of the checks catch most of the real failures: the protocol variant of the collector and the two-way link budget. Both are decided before anyone climbs a pole, and both are expensive to fix afterwards.

Why a metering network fails after handover

Three layers fail independently, and the symptoms look almost identical from the dashboard.

The bus layer is RS485: a collector at the top of a cabinet, meters wired in a daisy chain, one pair of wires carrying everything. A misplaced termination, a swapped A/B pair or a duplicated meter address produces a node that answers sometimes and not others — which reads like a radio problem, not a wiring problem.

The radio layer is the 470–510 MHz TPUNB link between collectors and the concentrator. Its failure mode is distance and clutter: a link that works in October on dry leaves can lose 10–15 dB in June when the trees are in leaf. If the commissioning budget had 10 dB of margin, that link was never viable; it was just lucky for one season.

The data layer is the part nobody tests: timestamps, missing values, alarms. A network can poll 100% of its meters and still be useless if the concentrator writes a zero when a read times out, or if meter clocks drift until the daily profile lands in the wrong hour.

The checklist separates the three, in this order, because a bus fault that is misdiagnosed as a radio fault will cost you a site visit and an antenna you did not need.

Phase 1 — Before anything is powered: know what you are installing

1. Match the collector variant to the meter protocol

The TP6306 collector ships in five variants, and they are not interchangeable by firmware alone. The -WL reads cold-water meters over an M-Bus segment using CJ/T 188 framing; the -WH speaks a private water-control protocol; the -EM reads electricity meters with DL/T 645; the -SE is a Modbus RTU sensor-terminal collector limited to six slaves, not thirty-two.

Pass criterion: one line per meter in the address plan, stating meter type, protocol, baud rate and collector variant, with a successful read of the first, a middle and the last meter on each bus.

What it prevents: the classic procurement error of ordering the -SE because it was in stock, then discovering that a 32-meter water riser is now a 6-meter riser. It also catches a subtler mistake — mixing CJ/T 188 water meters and DL/T 645 electricity meters on one RS485 segment, which technically works electrically and fails completely at the protocol level.

2. Audit the meter addresses before the bus is live

New meters leave the factory on address 1. After a replacement, two meters on the same segment can carry the same address, and the collector will happily read whichever one answers first. The consumption curve then shows one meter reading twice and another reading nothing — a failure that survives every radio check.

Pass criterion: a scan of the whole address range with no duplicates, and every meter physically labelled with its address on the terminal cover.

What it prevents: silent double-counting. This is the only check on the list that can be done entirely from the office, and it is the one most often skipped.

Phase 2 — Bus and power

3. Terminate and bias the RS485 segment once

Up to 32 slaves on one collector (TP6306 specifications) is the standard unit-load limit, and it assumes a proper daisy chain: 120 Ω at each physical end, bias resistors at one end, shield bonded at one end only, no star branches longer than a stub.

Pass criterion: with the bus idle and all slaves powered, the differential voltage between A and B is above 200 mV; with a 120 Ω termination check, the measured resistance across A–B is about 60 Ω.

What it prevents: the intermittent node. A missing termination does not fail at commissioning — it fails at the longest cable run, in the noisiest cabinet, on the coldest day, and only for the meter with the weakest driver.

4. Verify power and earthing under load, not on an open circuit

The TP6306 runs from 220 V AC and draws under 2 W; the TP5510 concentrator runs from DC 12 V with a 5–28 V input range and averages about 0.25 W. Both numbers are small enough that a bad joint will pass a voltmeter test and fail under a switching load.

Pass criterion: measured voltage inside the declared range at the terminal with the device transmitting, PE continuity to the cabinet earth bar below 1 Ω, and no shared breaker with a variable-frequency drive or a contactor coil.

What it prevents: the "works until the lift starts" fault. A collector sharing a circuit with a VFD sees its supply sag on every motor start; the radio resets, the poll cycle restarts, and the log fills with timeouts that look like coverage gaps.

Phase 3 — The radio path, calculated before it is climbed

5. Budget the link in both directions

This is the check that decides whether a site is viable. Using only datasheet numbers, the downlink from the TP5510 to the TP6306 is 26 dBm into a -111 dBm receiver, and the uplink is 20 dBm into a -116 dBm receiver:

Path

Transmit

Receiver sensitivity

Gross budget

Net after 3 dB of connectors and body loss

TP5510 → TP6306 (downlink)

+26 dBm

-111 dBm

137 dB

134 dB

TP6306 → TP5510 (uplink)

+20 dBm

-116 dBm

136 dB

133 dB

TP5510 → TP5510 (DTU pair)

+26 dBm

-116 dBm

142 dB

139 dB

The two directions of a collector link are balanced within 1 dB, which is a useful diagnostic: if a site works one way and not the other, the budget is not the explanation, and you should be looking at the antenna, the connector or an interferer rather than at distance.

Distance then follows from free-space loss, 85.9 dB at 1 km and 470 MHz by ITU-R P.525:

Fade margin allowed

TP5510 → TP6306

TP6306 → TP5510

TP5510 → TP5510

10 dB

80 km

72 km

143 km

20 dB

25 km

23 km

45 km

30 dB

8.1 km

7.2 km

14 km

Three curves of free-space distance against allowed fade margin at 470 MHz, for the downlink, the uplink and a DTU pair, with the 20-30 dB planning band shaded

Pass criterion: the planned distance is inside the 20–30 dB margin column, and the margin is written down together with the antenna gains and feeder losses it assumes.

What it prevents: commissioning a link at 10 dB margin in autumn. Free-space numbers are an upper bound; ITU-R P.1546 exists precisely because terrain and clutter remove tens of dB that the inverse-square law knows nothing about. In a suburban metering network, planning at 20 dB and accepting nothing below 15 dB is the difference between a network that survives a wet spring and one that generates tickets every June.

6. Measure the noise floor on the carrier you intend to use

The band is shared. Before the first collector is mounted, put a receiver on the planned carrier and record the floor for ten minutes, at the same time of day as the planned collection window.

Pass criterion: measured noise floor at least 10 dB below receiver sensitivity, i.e. -126 dBm or lower in a 200 kHz carrier, with no burst longer than the frame length you plan to send.

What it prevents: blaming the link budget for an interferer. A floor at -118 dBm does not fail a sensitivity test by itself, but it eats 8 dB of the margin you just calculated, and it will be there every night at the same hour because it is a neighbouring network's collection window.

7. Treat the antenna and feeder as part of the budget

The radio connectors are SMA female on both units, and the feeder is where budgets quietly disappear. At 470 MHz, thin RG-58-class cable loses on the order of 0.5 dB per metre — an assumption worth checking against the cable datasheet, because it means a 10 m run costs 5 dB, nearly a fifth of the entire 30 dB planning margin. A 5D-FB-class cable is closer to 0.15 dB per metre.

Pass criterion: VSWR below 2:1 at the radio port, feeder loss included in the written budget, mast height and polarisation recorded, and an external surge arrestor on any antenna that leaves the cabinet.

What it prevents: the antenna that was installed for mechanical convenience. A 3 dB loss is invisible on an RSSI bar and halves the usable distance; an antenna mounted at 1.5 m instead of 6 m loses more than that to the first row of buildings.

Phase 4 — The data path

8. Do the poll arithmetic before promising a collection interval

A complete sweep of the bus is pure arithmetic. Assume an 8-byte request, a 64-byte response, 8 bytes of framing and CRC, and 100 ms of turnaround per meter — assumptions, but ones that match a Modbus or DL/T 645 exchange:

t_meter = (8 + 64 + 8) * 8 / bitrate + 0.1 s
cycle   = N * t_meter * (1 + retry fraction)

At 2.4 kbps, the uplink rate the TP6306 uses for its 19200/2400 split, one meter costs 0.37 s, so 32 meters take 11.7 s per sweep. At 19.2 kbps the same sweep takes 4.3 s; at 76.8 kbps, 3.5 s.

Full read-cycle time against the number of meters on one collector, at 2.4, 19.2 and 76.8 kbps, with one-minute, five-minute and one-hour reference lines

Pass criterion: the calculated sweep, times the expected retry rate, fits inside the collection interval with at least 50% headroom.

What it prevents: the five-minute interval that quietly becomes a fifteen-minute interval. At 2.4 kbps, hourly collection of 32 meters occupies 0.33% of the channel; five-minute collection occupies 3.9%, still comfortable, but a 1-minute interval at 32 meters uses 19.5% of the airtime and leaves no room for retries when a single meter starts flapping. If the requirement is one minute, the answer is a faster symbol rate or a second collector, not a longer timeout.

9. Check every clock against one reference

Meter clocks are not the network's clocks. A 20 ppm crystal drifts 1.73 s per day, which is 12 s per week — past any reasonable gate after the first week of operation.

Pass criterion: every meter and collector timestamp within ±5 s of the concentrator after 24 hours, and the time-sync command present in the poll plan rather than run by hand.

What it prevents: consumption profiles that land in the wrong hour. This is invisible in a total, because the total is still right; it only shows up when someone reconciles a daily curve with a billing period and finds a step that no meter actually recorded.

10. Fire a real alarm, not a simulated one

Alarms are the part of a metering network that is tested least and trusted most. A low-battery or tamper event must travel the same path as a normal read, and it must not be collapsed into the daily batch.

Pass criterion: force one alarm at a meter, and confirm it arrives at the concentrator within one collection cycle, carries the meter address and the event code, and is not overwritten by the next scheduled read.

What it prevents: the alarm queue that fills with generic "device offline" entries and no meter identity. When the first real leak alarm arrives, the operator needs an address, not a timestamp.

11. Prove that a failed read is not a zero

The most expensive data bug in metering is the substitution of zero for missing. A read that times out and is stored as 0 kWh is indistinguishable, in a monthly total, from a meter that genuinely consumed nothing — until the customer disputes the bill.

Pass criterion: disconnect one meter for one cycle and confirm the database records a null or a missing flag with a retry count, that the consumption total does not change, and that a deliberately corrupted frame is retried rather than stored.

What it prevents: silent data loss across a whole network. The check takes ten minutes and one screwdriver.

12. Soak for 24 hours, then hand over the evidence

A network that has run for an hour has proved nothing. Run it for a full day, across the temperature swing and the load profile, with per-meter accounting rather than a network average.

The 24-hour run is also the only honest test of the thermal envelope. A TP6306 rated to +75 °C inside a sealed steel cabinet in direct sun can see 80 °C in July; an outdoor TP5510 at -40 °C is fine, but its antenna connector is not if the assembly was made with the wrong gland. Put the device where the numbers fit, measure the cabinet rather than the air, and log the internal temperature through the hottest hour of the soak.

Horizontal bars: declared operating ranges for the TP6306 collector (-25 to +75 °C), the TP5510 concentrator (-40 to +70 °C) and a typical water meter pit (0 to 55 °C)

Pass criterion: at least 99% poll success across the network, no single meter below 95%, received signal spread within 6 dB of the commissioning measurement, and no consumption counter moving backwards. The handover pack contains the address map, per-meter RSSI, per-meter success rate, the link budget with its assumptions, photographs of every termination, and the collector variant and serial numbers.

What it prevents: the argument in month three about whether the network was ever commissioned. A 24-hour record with per-meter numbers settles it in a minute; a network average of 98% hides the two meters that have been at 60% since day one.

The whole checklist on one page

#

Check

Pass criterion

Failure it prevents

1

Collector variant vs meter protocol

First, middle and last meter read on every bus

-SE ordered for a 32-meter riser; mixed CJ/T 188 and DL/T 645

2

Meter address audit

No duplicates in the address range

One meter counted twice, one never read

3

RS485 termination

200 mV idle differential, ~60 Ω across A–B

Intermittent node at the end of the run

4

Power and earth under load

Voltage in range while transmitting, PE below 1 Ω

Resets whenever a VFD starts

5

Two-way link budget

Distance inside the 20–30 dB margin column

A link that only worked in autumn

6

Noise floor on the carrier

≤ -126 dBm in 200 kHz, no long bursts

Interference blamed on distance

7

Antenna and feeder

VSWR < 2:1, feeder loss in the budget

5 dB lost in 10 m of thin cable

8

Poll arithmetic

Sweep with retries fits the interval with 50% headroom

A five-minute promise that becomes fifteen

9

Clock check

All nodes within ±5 s after 24 h

Profiles in the wrong hour

10

Live alarm test

Alarm arrives in one cycle with an address

Offline queue with no meter identity

11

Missing versus zero

Timeout stores null, total unchanged

Zero-filled registers in a bill

12

24-hour soak and pack

≥99% overall, no meter <95%, counters monotonic

An argument in month three

What this looks like on one real bus

Take a water riser with 32 meters, one TP6306-WL in the cabinet, and a TP5510 concentrator 3 km away on the roof of the block. The bus is 180 m of twisted pair, the meters are CJ/T 188, and the collection interval is hourly.

The sweep is 11.7 s at the 2.4 kbps uplink rate, so the channel is busy for 0.33% of the hour and there is room for two full retries of every meter before the interval is at risk. The link budget at 3 km needs 95.4 dB of free-space loss; against the 133 dB uplink budget that leaves 37.6 dB before any allowance for clutter, so the site sits comfortably in the 30 dB column even after 10 m of feeder and a 6 dB foliage allowance. The noise floor measured at -129 dBm leaves 13 dB below the -116 dBm sensitivity gate. The cabinet reaches 61 °C in August, inside the TP6306's +75 °C limit but outside the meter pit's typical 0–55 °C range, which is why the collector lives in the cabinet and not in the pit.

The only item that fails is the clock check on day one: two meters are 40 s behind. They are replaced under warranty, but the failure would have been invisible for a week without the check — 20 ppm is 1.7 s per day, and 40 s is a meter that has been sitting in a warehouse since last year.

FAQ

Does a 32-slave RS485 bus need a repeater?

Not if the wiring is a single daisy chain with proper termination and the cable stays within the baud-rate limit. It needs a repeater when the bus is star-wired, when more than 32 unit loads are attached, or when the run exceeds the length the chosen baud rate supports. The TP6306's limit of 32 slaves is a unit-load limit, not a suggestion; the -SE variant drops to six because its Modbus RTU segment is intended for sensor terminals rather than a full meter riser.

Can I run everything at 76.8 kbps to shorten the sweep?

You can, and the sweep drops from 11.7 s to 3.5 s for 32 meters, but the sensitivity figure in the datasheet is quoted at 19.2 kbps. Faster symbol rates buy time and cost range, so the trade is only worth making when the link budget has margin to spare. If a site is already in the 20 dB column at 19.2 kbps, moving to 76.8 kbps to fix a collection-interval problem will move the problem to the radio layer.

How many collectors can one concentrator serve?

The TP5510 is specified for up to 100 terminals, in both broadcast and directed modes. The practical limit is usually the poll arithmetic rather than the terminal count: 100 meters at 19.2 kbps is a 13.3 s sweep before retries, which is fine for hourly collection and tight for anything faster. Plan the concentrator around the interval you promised, not around the maximum in the datasheet.

What is the single most common commissioning miss?

The protocol variant. It is decided at purchase, it is invisible on the invoice, and it produces a network that powers up, links perfectly, and reads nothing. The link budget is second, mostly because it is done once in the office and then contradicted on site by an antenna installed where the ladder was shortest.

If you are commissioning a metering network and want the variant, budget and poll arithmetic checked against your own meter list before the cabinets are sealed, the contact form reaches the engineering team, and the product catalogue lists the collector and concentrator variants with their full specifications.

More in Solutions