Smart Metering Architecture Explained: DLMS/COSEM, OBIS Codes, AMI Head-End Systems and Meter Data Management (2026)

Smart Metering Architecture Explained: DLMS/COSEM, OBIS Codes, AMI Head-End Systems and Meter Data Management (2026)

Smart Metering Architecture Explained: DLMS/COSEM, OBIS Codes, AMI Head-End Systems and Meter Data Management (2026)

Most smart metering projects that fail do not fail at the meter. They fail in the seams: a head-end that reads the right register with the wrong scaler, a data concentrator that drops a day of load profile, or a meter data management system that silently estimates a week of readings nobody notices. The protocol underneath most of these systems outside North America is DLMS/COSEM, and it is far more than a wire format. It is an object model, an addressing scheme and a security framework at once. That matters now because utilities are installing meters at a pace the back office was never sized for. India alone reports 7.24 crore (72.4 million) smart meters installed as of 30 June 2026, against 20.33 crore sanctioned under its distribution reform scheme. This guide gives you the end-to-end picture: how the object model works, how OBIS codes name every value, how the head-end talks to meters, how the MDMS turns raw reads into billable data, and where it all breaks.

What this covers: the COSEM object model and OBIS codes, the AMI reference architecture from meter to billing, a sequence-level walk-through of a DLMS session, worked sizing math, security suites, comms technology trade-offs, failure modes and a Python sketch using an open-source DLMS library.

Context and Background

Advanced metering infrastructure (AMI) is the combination of interval-recording meters, two-way communication networks, a head-end system (HES) that manages those meters, and a meter data management system (MDMS) that stores, validates and distributes the data. Automated meter reading (AMR), its predecessor, was mostly one-way and walk-by or drive-by. AMI adds remote commands such as connect and disconnect, firmware updates, tariff changes and event push.

Two protocol families dominate. Outside North America, the IEC 62056 series, better known as DLMS/COSEM, is the default. DLMS stands for Device Language Message Specification; COSEM stands for Companion Specification for Energy Metering. In North America, the ANSI C12 family dominates: C12.18 for the optical port, C12.19 for the data tables, and C12.22 for transporting those tables over networks. NEMA’s summary of the C12 standards lists their IEEE equivalents as IEEE 1701 (C12.18) and IEEE 1377 (C12.19).

The DLMS User Association (DLMS UA) maintains the specifications and runs conformance testing. Its core documents are colour-coded books. The Blue Book (DLMS UA 1000-1) defines the COSEM interface classes and the OBIS identification system; edition 18 was released in two parts on 27 February 2026. The Green Book (DLMS UA 1000-2) covers the application layer, lower layers and communication profiles; edition 12 was released on 28 May 2025. The White Book is the glossary. IEC then publishes the content as parts of IEC 62056, such as 62056-5-3 (application layer), 62056-6-1 (OBIS) and 62056-6-2 (interface classes). You can check the current editions on the DLMS UA core specifications page.

Smart meters rarely live alone on a grid. They feed the same operational picture as substation automation and distributed energy resources. If you work on the substation side, our guide to IEC 61850 substation automation with GOOSE, MMS and Sampled Values covers the protocol that sits one tier up. Meter data increasingly feeds DER aggregation too, which our virtual power plant architecture with OpenADR 3 and IEEE 2030.5 piece explores.

Regulation is pushing the data rate up. In the EU, Article 20 of Directive (EU) 2019/944 requires that smart meters let customers be metered and settled “at the same time resolution as the imbalance settlement period in the national market,” and that non-validated near-real-time data be available at no extra cost. As European imbalance settlement moves to 15-minute periods, 15-minute interval data stops being a premium option and becomes the baseline. That single requirement multiplies the data volume every downstream system must handle.

DLMS COSEM Reference Architecture: From Meter Object to Bill

DLMS/COSEM separates what a meter knows from how you reach it. COSEM models every quantity as an object of a standard interface class, named by a six-part OBIS code. DLMS defines the services (GET, SET, ACTION, push) used to access those objects over any communication profile, from an optical port to G3-PLC or TCP/IP.

DLMS COSEM smart metering architecture showing meters, neighbourhood network, data concentrator, head-end system, MDMS and billing

Figure 1: The layered AMI reference architecture, from DLMS/COSEM meters through the neighbourhood network and head-end system to meter data management and billing.

Figure 1 reads bottom-up as a data path and top-down as a command path. Electricity meters act as DLMS/COSEM servers. Gas and water meters often hang off the electricity meter over wired or wireless M-Bus. Meters reach the head-end either through a neighbourhood area network (NAN) that terminates at a data concentrator unit (DCU), or directly over cellular. The HES is the DLMS client. It hands normalised reads to the MDMS, which feeds billing, the customer information system (CIS), outage management and DER management. A key management system, ideally backed by a hardware security module (HSM), supplies the per-meter keys the HES needs to open a secured session.

The object model is the contract, not the transport

The most important design idea in DLMS/COSEM is that the data model is independent of the network. A Register object holding cumulative active energy looks identical whether you read it over an infrared probe, a power line or an LTE modem. The Green Book defines communication profiles that map the same application layer onto different lower layers. IEC 62056 lists, among others, an HDLC-based profile (62056-7-6), a TCP-UDP/IP profile (62056-9-7), a G3-PLC profile (62056-8-5) and wired and wireless M-Bus profiles (62056-7-3).

This separation has practical consequences. A head-end vendor can support a new radio without rewriting its meter logic. A utility can mix PLC and cellular meters in one fleet and keep a single data model in the MDMS. It also means interoperability problems are rarely about the protocol frames. They are about which objects a meter exposes, with which access rights, under which association.

A physical meter contains one or more logical devices. Every meter has a management logical device at a reserved address, which must expose the Association object and the logical device name. Each logical device then exposes a set of COSEM objects. Each object is an instance of an interface class, identified by a class_id and a version. Figure 2 shows the hierarchy.

COSEM object model hierarchy of physical device, logical devices, interface classes and OBIS-named objects

Figure 2: The COSEM object model. A physical meter hosts logical devices, which expose objects of standard interface classes, each named by an OBIS code and carrying attributes and methods.

The figure shows why discovery matters. A client connects to a logical device through an association, then reads the Association object’s object list to learn which objects exist and what it may access. Two meters from different vendors can both be “DLMS compliant” and still expose different object lists. The interface classes most implementers meet first are listed below, with class IDs taken from the DLMS UA Blue Book excerpt.

Interface class class_id What it holds Typical use
Data 1 A single value without units Serial number, firmware version
Register 3 Value plus scaler and unit Cumulative energy, instantaneous voltage
Extended register 4 Register plus status and capture time Maximum demand with timestamp
Demand register 5 Current and last average over a period Demand billing
Profile generic 7 A table of captured rows Load profile, event logs, billing history
Clock 8 Date, time, deviation, DST Time synchronisation
Script table 9 Executable scripts Tariff switching, billing period reset
Association LN 15 Object list, access rights, authentication Session context
Image transfer 18 Block-wise firmware image Firmware update
Push setup 40 Push destination and contents Alarms, periodic push
Security setup 64 Security policy, suite, certificates Key and certificate management
Disconnect control 70 Relay state and control Remote connect and disconnect
Limiter 71 Thresholds and actions Load limiting, prepaid emergency credit

OBIS codes: a six-part address for every value

An OBIS code is the logical name of a COSEM object. IEC 62056-6-1 (edition 4.0, published December 2023) defines it as a hierarchical structure of six value groups, A to F. Each group is one byte, so codes are often written as A.B.C.D.E.F or as A-B:C.D.E*F.

  • A identifies the medium. 0 means abstract objects such as the clock, 1 means electricity, and other values cover gas, heat and water.
  • B identifies the channel, used when a device measures several inputs. 0 means no channel specified.
  • C identifies the physical quantity, such as active power import, reactive power or voltage on a given phase.
  • D identifies the processing method, such as a time integral (energy), an instantaneous value or a maximum.
  • E gives further classification, most often the tariff rate, where 0 means the total across all rates.
  • F identifies the billing period or historical value, where 255 means “current” or “not used.”

The canonical example is 1.0.1.8.0.255. Group A is 1, so electricity. B is 0, no specific channel. C is 1, active power import (the sum over phases, quadrants I and IV). D is 8, time integral 1, which turns power into cumulative energy. E is 0, the total across all tariff rates, and F is 255, the current value. In the meter’s display shorthand this is “1.8.0,” the line every meter reader knows as total import kWh. Swap E to 1 and you get 1.0.1.8.1.255, the import register for tariff rate 1. Change C to 2 and you get export, the register a rooftop-solar customer cares about. The clock lives at 0.0.1.0.0.255, where A is 0 because time is not a medium-specific quantity.

There is a subtlety that catches every new integrator. The OBIS code tells you what a value means, but not how to scale it. A Register’s attribute 2 holds the raw value and attribute 3 holds scaler_unit, a pair of a power-of-ten exponent and a unit enumeration. A raw value of 123456 with a scaler of -3 and unit Wh means 123.456 Wh, not 123,456 kWh. Treating OBIS as a complete semantic contract, while ignoring attribute 3, is the single most common source of bills that are off by a factor of 1,000.

Associations and security are part of the model

DLMS/COSEM does not bolt security on at the transport layer. An application association is negotiated before any data moves. The client sends an AARQ (association request), the meter replies with an AARE (association response), and each association has its own object list and access rights. A public client may read the clock and serial number with no authentication. A meter-reading client may read registers and profiles. A management client, with stronger authentication, may write tariffs or operate the disconnect relay.

Authentication comes in three tiers. No security is the lowest. Low-level security (LLS) uses a password sent in the association request. High-level security (HLS) uses a four-pass challenge-and-response, with mechanisms that open-source libraries expose as MD5, SHA-1, GMAC, SHA-256 and ECDSA. MD5 and SHA-1 remain in the specification for legacy reasons and should not be used in new deployments.

Data protection is governed by security suites, configured in the Security setup object (class 64). Suite 0 is symmetric: AES-GCM-128 for authenticated encryption, with AES-128 key wrap for key transport. Suite 1, named ECDH-ECDSA-AES-GCM-128-SHA-256, adds public-key cryptography for signatures and key agreement. Suite 2, ECDH-ECDSA-AES-GCM-256-SHA-384, raises the symmetric and hash strength. A 2023 IoTBDS paper on an HSM-based suite 1 meter notes that most smart meters deployed so far support suite 0 mechanisms. In practice that means most fleets depend on symmetric keys held both in every meter and in the head-end, which puts key management at the centre of the security architecture.

Every ciphered frame carries an invocation counter that increments with each message. The receiver rejects counters it has already seen, which defeats replay. That protection has an operational cost, covered later in the failure modes.

Deeper Analysis: A DLMS Session, Sizing Math and the MDMS Pipeline

Architecture diagrams hide the exchanges that actually consume time on a constrained network. This section walks through a single read, then scales it to a fleet, then follows the data into the MDMS.

Walking through one DLMS/COSEM read

Figure 3 shows a typical secured read from a head-end through a concentrator to a meter using the HDLC profile and HLS authentication.

DLMS COSEM session sequence with HDLC connect, AARQ and AARE association, HLS challenge, GET requests and release

Figure 3: A DLMS/COSEM read session. The link is opened with SNRM and UA, the association with AARQ and AARE plus an HLS challenge exchange, then the client reads scaler, value and load profile before releasing.

On HDLC-based profiles, the client first opens the data link with SNRM (set normal response mode) and the meter answers UA (unnumbered acknowledgement). On the wrapper-based TCP/UDP profile this step does not exist; IANA registers port 4059 for the dlms-cosem service. The client then sends an AARQ carrying its challenge. The meter returns an AARE with its own challenge. With HLS, the client proves knowledge of the key by invoking the Association object’s reply_to_HLS_authentication method, and the meter verifies it. Only then does the association open.

Once open, the client issues GET requests. A well-behaved client reads attribute 3 (scaler_unit) once per object and caches it, then reads attribute 2 (value) as often as needed. Load profiles are read from a Profile generic object, filtered by a date range, and returned in blocks because a day of interval rows does not fit one frame. Finally the client releases the association with RLRQ. Meters can also push, sending unsolicited data notifications for alarms such as a tamper event or a cover opening, as configured in a Push setup object.

Count the round trips. A cold read needs at least four exchanges before the first byte of data: link setup, AARQ, and the HLS action. On a PLC network where each round trip might take seconds, those handshakes can cost more time than the data itself. That is why mature head-ends keep associations open where the meter permits, batch many attributes into one GET-with-list, and prefer push for periodic data on IP-capable meters.

Python sketch: reading the import register

The Gurux DLMS libraries are a widely used open-source implementation. The Python package gurux-dlms is licensed under GPLv2; version 1.0.203 was released on 26 August 2026. The sketch below reuses the GXDLMSReader helper class from the Gurux Python client example repository, which handles framing, block transfer and the association handshake. Adjust addresses and authentication to your meter.

# pip install gurux-dlms gurux-net
# GXDLMSReader.py is copied from Gurux.DLMS.Python/Gurux.DLMS.Client.Example.python
from gurux_common.enums import TraceLevel
from gurux_net import GXNet
from gurux_net.enums import NetworkType
from gurux_dlms import GXDLMSClient
from gurux_dlms.enums import InterfaceType, Authentication
from gurux_dlms.objects import GXDLMSRegister, GXDLMSClock
from GXDLMSReader import GXDLMSReader

# Logical name referencing, client 16 (public) to server 1, no auth.
# A meter-reading client typically uses LOW or HIGH_GMAC instead.
client = GXDLMSClient(
    True, 16, 1, Authentication.NONE, None, InterfaceType.WRAPPER
)
media = GXNet(NetworkType.TCP, "192.0.2.10", 4059)  # documentation IP
reader = GXDLMSReader(client, media, TraceLevel.INFO, None)

media.open()
try:
    reader.initializeConnection()          # AARQ/AARE (+ SNRM on HDLC)
    energy = GXDLMSRegister("1.0.1.8.0.255")  # total active energy import
    reader.read(energy, 3)                 # scaler_unit first
    reader.read(energy, 2)                 # value, scaled by the library
    print("Import energy:", energy.value, energy.unit)

    clock = GXDLMSClock("0.0.1.0.0.255")
    reader.read(clock, 2)
    print("Meter time:", clock.time)
finally:
    reader.close()                         # RLRQ and disconnect

Order matters in that sketch. In the Gurux Python library, GXDLMSRegister multiplies attribute 2 by the scaler only if attribute 3 has already been read; otherwise you get the raw integer. Production code should also read the frame counter before a ciphered association, handle Deciphering failed errors, and never hardcode keys. Treat the snippet as an illustration of the call order, not a production client.

Worked sizing math for a national fleet

The numbers below are illustrative, built from stated assumptions so you can rerun them with your own. Assume 10 million meters, 15-minute intervals and four profile channels: active import, active export, reactive import and reactive export.

A 15-minute interval gives 96 rows per meter per day. Four channels give 384 values per meter per day. Across 10 million meters that is 3.84 billion interval values per day, or about 1.4 trillion per year. That is before events, daily billing snapshots, instantaneous voltage profiles and power-quality logs.

On the wire, a profile row in A-XDR encoding carries a timestamp (a 12-byte date-time octet string plus tag and length), four 32-bit values with their type tags, and often a status byte. A rough estimate is 35 to 45 bytes per row. At 40 bytes, a meter’s day is about 3.8 KB of payload. Add DLMS, ciphering and transport overhead and 5 KB per meter per day is a reasonable planning figure. That is about 50 GB per day across the fleet. It looks modest until you remember PLC links run at kilobits per second and are shared by hundreds of meters per concentrator.

Session rate is the harder constraint. If the head-end pulls daily load profiles in a six-hour night window, it must complete 10 million sessions in 21,600 seconds, about 463 sessions per second sustained. With each session needing several round trips and some percentage of retries, the HES must keep thousands of concurrent sessions in flight. If instead every meter pushes each interval, the HES ingests 10 million messages every 900 seconds, about 11,100 messages per second, all day. Push smooths the network but turns the HES into a high-throughput streaming system.

Storage is the third constraint. If the MDMS stores each interval value as a timestamped record of about 16 bytes before compression, 3.84 billion values cost about 61 GB per day, or around 22 TB per year of raw data. Versioning for VEE adds more, because estimated and edited values must coexist with the originals. Columnar time-series compression typically shrinks interval data substantially, but plan capacity on uncompressed figures and treat compression as headroom.

Scale that to India’s sanctioned programme of 19.79 crore consumer meters, or about 198 million. At the same assumptions, that is roughly 76 billion interval values a day. No single MDMS instance handles that; Indian deployments are split across many distribution utilities and service providers, which is itself an architectural choice driven by data volume.

Inside the meter data management system

The MDMS is where raw meter reads become data a utility can bill on. Its core job is VEE: validation, estimation and editing. Figure 4 shows the pipeline.

Meter data management system VEE pipeline from raw reads through validation, estimation and editing to billing determinants

Figure 4: The MDMS validation, estimation and editing (VEE) pipeline. Every read is validated; failures are estimated and optionally hand-edited, while a versioned store keeps the original alongside every correction.

Validation applies rules to each read. Typical checks are gaps in the interval sequence, spikes above a plausible maximum for the service, negative consumption on an import-only register, timestamps outside the expected window, and a sum check. The sum check compares the total of interval consumption with the change in the cumulative register over the same period. If they disagree beyond a tolerance, either the profile or the register is wrong.

Estimation fills the gaps. Simple methods interpolate linearly between good reads. Better methods use the customer’s historical profile for the same weekday and season, scaled to match the register delta. Editing lets an analyst override values, with every change recorded in an audit trail. The MDMS must keep the original read, each estimate, and each edit as separate versions, because billing disputes and regulatory audits need to show what was known and when.

After VEE, the MDMS computes billing determinants, such as consumption per time-of-use bucket or maximum demand, and passes them to the CIS. It also feeds settlement exports, network analytics, and increasingly near-real-time feeds for demand response. If you have built a usage-based billing pipeline in software, the same idempotency and late-event problems appear here; our guide to usage-based billing and metering architecture covers those patterns from the SaaS side.

Integration between HES and MDMS is where standards help most. IEC 61968-9 defines Common Information Model (CIM) message profiles for meter reading and control. IEC 62056-6-9 maps those CIM profiles onto DLMS/COSEM data models, so a MeterReadings message in the back office can be traced to specific OBIS-named objects in the field. Utilities that insist on these interfaces can replace the HES or MDMS without rewriting the other.

Choosing the neighbourhood network

The last-mile network shapes everything above it: latency, session rate, firmware-update time and cost. The matrix below compares the common options. Qualitative ratings are the author’s assessment; numeric figures come from the cited sources.

Technology Medium Standards base Strengths Weaknesses Best fit
G3-PLC Power line, narrowband OFDM below 500 kHz ITU-T G.9903; DLMS profile IEC 62056-8-5 No radio licence or SIM; follows the grid topology; IP-based Noise, impedance changes and phase coupling; low throughput Dense urban and European grids with many meters per transformer
PRIME Power line, narrowband OFDM ITU-T G.9904; PRIME Alliance v1.4 Open alliance specification; DCU at transformer Similar PLC limits; regional ecosystem Utilities already on PRIME
Wi-SUN FAN Sub-GHz RF mesh Wi-SUN FAN 1.1 IPv6 mesh; FSK at 50 kbps up to OFDM at 2.4 Mbps per Cisco; multi-vendor certification Needs mesh density and collector siting North America, Japan, India, smart-city co-use
Proprietary RF mesh Sub-GHz RF Vendor specific Mature, tuned for meters Vendor lock-in Legacy fleets
NB-IoT and LTE-M Licensed cellular 3GPP No private network to build; per-meter IP Recurring SIM cost; indoor coverage; operator dependency Sparse areas and quick starts
4G Cat-1 Licensed cellular 3GPP Higher throughput for firmware and power quality Higher power and module cost Commercial and industrial meters

For a deeper look at cellular trade-offs, read our LoRaWAN vs NB-IoT vs LTE-M comparison for industrial IoT and our explainer on PSM, eDRX and RAI power saving on LTE-M and NB-IoT, which matters for battery-powered gas and water meters.

The decision is rarely purely technical. PLC has no recurring per-meter fee but costs more in commissioning and troubleshooting. Cellular is quick to deploy but adds a recurring connectivity fee to every meter for its whole life. Multiply even a small monthly tariff by 120 months and it becomes a line item comparable to hardware, so run the lifetime number before choosing. RF mesh needs up-front network design. Many large programmes end up hybrid, with RF mesh or PLC as the default and cellular for hard-to-reach meters.

Head-end responsibilities beyond reading

Reading is the easy part of a head-end. The harder responsibilities are the ones that change meter state, because each is a remote action on customer premises.

Remote disconnect and reconnect use the Disconnect control object (class 70). The HES invokes a method, and the meter operates its relay. That single capability is what makes prepaid metering practical, and it is also the most dangerous command in the system. A bug or a compromised HES that issues a mass disconnect can drop thousands of homes at once and create a load swing large enough to matter to the distribution network. Good head-ends rate-limit disconnects, require a second approval above a threshold, and log every command with the operator and the reason.

Firmware updates use the Image transfer object (class 18). The HES transfers the image in blocks, asks the meter to verify it, and then activates it, often at a scheduled time. Over PLC or mesh, transferring an image of a few hundred kilobytes to every meter can take days or weeks. Some networks support multicast distribution to cut that time, an idea familiar from LoRaWAN FUOTA with multicast and fragmentation. Metrology firmware is often legally separated from communications firmware, because changing anything that affects measurement may require recertification.

Time synchronisation keeps every interval aligned. The HES writes or adjusts the Clock object (class 8). A clock that drifts by a few minutes misaligns interval boundaries, which distorts time-of-use billing and settlement. Tariff changes are delivered through activity calendars, special-day tables and scripts. Event and alarm handling ingests pushed alarms, such as cover open, magnetic tamper, power failure and restoration, and routes them to outage management or revenue protection.

Trade-offs, Gotchas, and What Goes Wrong

The protocol is mature. Most failures come from configuration, scale and operations, not from bugs in DLMS itself.

Conformance does not equal interoperability. DLMS UA conformance testing checks that a meter implements the protocol and the objects it claims correctly. It does not guarantee that two vendors expose the same objects under the same OBIS codes, with the same access rights, in the same association. A utility that buys from three vendors without a companion specification ends up writing three sets of adapters. The fix is a utility- or country-level companion specification that fixes the object list, associations, security policy and push contents. Several countries publish such companion standards; India’s is commonly cited as the IS 15959 series, though you should confirm the current parts with the Bureau of Indian Standards (not verified for this article).

Scaler errors are silent. A wrong scaler does not throw an exception. It produces plausible-looking numbers that are 10, 100 or 1,000 times off. Validate a sample of reads against the meter’s local display during commissioning, and alert on per-meter consumption that jumps by orders of magnitude after a firmware change.

Invocation counters can lock you out. Ciphered frames must carry an increasing counter, and the meter rejects any value it considers stale. If the HES loses its counter state after a failover or a database restore, every ciphered request may be rejected. Libraries handle this by reading a public frame-counter object before associating, but the HES must persist counters per meter and per key, with the same durability as the keys themselves.

Key management is the real attack surface. With suite 0, the head-end holds a symmetric key for every meter. A compromise of the HES key store is a compromise of the fleet. Keep keys in an HSM, use unique keys per meter, rotate them through the Security setup object, and separate the keys for reading from those for management. Suites 1 and 2 reduce the blast radius by moving to public-key signatures, at the cost of a PKI, certificate lifecycle and heavier meter cryptography.

Physical attacks still work. The best-known smart meter fraud case shows that protocols are only one layer. In 2012, Krebs on Security reported an FBI alert describing insiders who reprogrammed meters through the optical port using a laptop and an infrared converter, and customers who placed magnets on meters to stop measurement. The FBI estimated the affected utility’s losses at up to $400 million a year. Lock down the optical port with the same authentication as the network, log local sessions, and use tamper events with analytics that compare a meter with its transformer.

Network capacity collapses under retries. A PLC or mesh network running near capacity does not degrade gracefully. Failed reads trigger retries, retries add load, and success rates fall further. Size the collection window with margin, stagger schedules, and prefer push or long-lived associations to repeated handshakes.

VEE can hide problems. Estimation is meant to fill rare gaps. When a concentrator silently fails, the MDMS can estimate a whole feeder for weeks, and bills keep going out. Track the estimated share of reads per meter, per DCU and per feeder, and alarm when it rises.

Privacy is a design input. Fifteen-minute load profiles can reveal occupancy patterns and daily routines. EU law requires smart metering to follow data protection rules and to give customers access to their own data. Collect at the resolution the use case requires, restrict access by role, and keep the audit trail.

Practical Recommendations

Start with the data contract, not the radio. Write a companion specification that fixes the object list, OBIS codes, scalers, associations, security suite, push setups and event codes for every meter type you will buy. Make conformance to it, not just DLMS UA certification, a procurement condition, and test every vendor’s meter against your HES in a lab before field pilots.

Design the head-end as a distributed system. It is a high-concurrency session manager, a scheduler, a key consumer and a command gateway. Size it for the worst case: a full fleet re-read after an outage, or a firmware campaign running alongside daily collection. Use the session and message rates from your own sizing math, not vendor brochure figures.

Treat key management as a first-class subsystem. Put keys in an HSM, use unique per-meter keys, persist invocation counters durably, and run a key-rotation drill before go-live. If you choose suite 0 for cost reasons, document the compensating controls.

Integrate through standard interfaces. Use IEC 61968-9 CIM messages between HES and MDMS so you can replace either side. Keep MDMS reads versioned, and expose VEE statistics as operational metrics.

Choose the network per area, not per programme. Use PLC or RF mesh where meter density is high, and cellular where it is low, and compare lifetime cost including recurring fees.

A short checklist before go-live:

  • [ ] Companion specification approved, with OBIS list, scalers and associations.
  • [ ] Every meter model lab-tested against the production HES.
  • [ ] HSM-backed per-meter keys; counters persisted and failover-tested.
  • [ ] Disconnect commands rate-limited, dual-approved above a threshold, and audited.
  • [ ] Collection window sized with retry margin; push used where supported.
  • [ ] VEE rules tuned; estimated-read share monitored per DCU and feeder.
  • [ ] Optical port secured and local sessions logged.
  • [ ] Firmware campaign duration tested on the real network.

Frequently Asked Questions

What is DLMS/COSEM used for?

DLMS/COSEM is the international standard, published as the IEC 62056 series, for exchanging data with utility meters. It defines an object model (COSEM) that describes every value a meter holds, and an application protocol (DLMS) that reads, writes and invokes those objects over many networks. Utilities use it to collect energy readings and load profiles, synchronise clocks, change tariffs, operate disconnect relays, receive tamper alarms and update firmware. Because the data model is independent of the transport, the same meter logic works over an optical port, power line, RF mesh or cellular.

What does the OBIS code 1.0.1.8.0.255 mean?

OBIS code 1.0.1.8.0.255 identifies total active energy import on an electricity meter, the familiar “1.8.0” register. Value group A=1 means electricity, B=0 means no specific channel, C=1 means active power import, D=8 means time integral 1 (cumulative energy), E=0 means the total across all tariff rates, and F=255 means the current value rather than a stored billing period. Changing E to 1 selects tariff rate 1, and changing C to 2 selects active energy export. Always read the scaler and unit attribute too.

What is the difference between a head-end system and an MDMS?

A head-end system (HES) talks to meters. It manages communication, schedules reads, opens DLMS associations, holds or uses keys, sends commands such as disconnect and firmware updates, and receives alarms. A meter data management system (MDMS) stores and processes the data the HES collects. It performs validation, estimation and editing, keeps versioned reads, computes billing determinants and feeds billing, settlement and analytics. The HES is operational and network-facing; the MDMS is the system of record for meter data. They usually integrate through IEC 61968-9 CIM messages.

What are DLMS security suites 0, 1 and 2?

Security suites define the cryptographic algorithms a DLMS/COSEM meter uses. Suite 0 is symmetric, using AES-GCM-128 for authenticated encryption and AES key wrap for key transport. Suite 1, ECDH-ECDSA-AES-GCM-128-SHA-256, adds elliptic-curve key agreement and digital signatures. Suite 2, ECDH-ECDSA-AES-GCM-256-SHA-384, uses stronger symmetric and hash algorithms. Most deployed meters support suite 0, which makes per-meter key management in the head-end critical. Suites 1 and 2 limit the impact of a key compromise but require a public key infrastructure.

Is DLMS/COSEM used in the United States?

North American utilities mostly use the ANSI C12 family instead. ANSI C12.19 defines the meter’s data tables, C12.18 defines the optical port protocol, and C12.22 carries those tables over networks. Many North American AMI systems also rely on proprietary RF mesh or Wi-SUN networks. DLMS/COSEM dominates in Europe, India, the Middle East, Africa and much of Asia. Global meter vendors often build both protocol stacks, so the choice is usually driven by the utility’s market and regulator rather than the hardware.

How much data does a smart meter generate?

It depends on interval length and channel count. With 15-minute intervals, a meter produces 96 rows per channel per day. Four energy channels give 384 values a day. As an illustrative estimate, that is roughly 4 to 5 KB of encoded payload per meter per day including overhead. For 10 million meters, that is around 3.84 billion interval values and roughly 50 GB of network payload per day, before events, power-quality logs and retries. Storage with VEE versioning grows faster than the raw stream.

Further Reading

References

  1. DLMS User Association, Core Specifications (Blue Book Ed. 18, Green Book Ed. 12, White Book Ed. 2): https://www.dlms.com/core-specifications/
  2. DLMS UA, Excerpts from Blue Book Ed. 17 Part 2 (interface class IDs): https://www.dlms.com/wp-content/uploads/2025/06/Excerpts-DLMS-Blue-Book-Ed-17-part-2-V1.0.pdf
  3. IEC 62056-6-1 Ed. 4.0 (2023-12) preview, VDE Verlag: https://www.vde-verlag.de/iec-normen/preview-pdf/info_iec62056-6-1{ed4.0.RLV}en.pdf
  4. IANA Service Name and Port Registry, port 4059 dlms-cosem: https://www.iana.org/assignments/service-names-port-numbers?search=405
  5. NEMA, Summary of ANSI C12 Standards: https://www.nema.org/docs/default-source/technical-document-library/c12-summary.pdf
  6. Directive (EU) 2019/944, Article 20: https://legislation.gov.uk/eudr/2019/944/article/20/2020-01-31
  7. Gurux, Gurux.DLMS.Secure (security suites, invocation counter): https://www.gurux.fi/Gurux.DLMS.Secure
  8. gurux-dlms on PyPI: https://pypi.org/project/gurux-dlms
  9. Huang et al., “The Implementation of an HSM-Based Smart Meter for Supporting DLMS/COSEM Security Suite 1,” IoTBDS 2023: https://scitepress.org/publishedPapers/2023/118253/pdf/index.html
  10. ITU-T G.9903 (G3-PLC): https://www.itu.int/t/aap/recdetails/7888
  11. PRIME Alliance specification: https://prime-alliance.org/alliance/specification/
  12. Cisco, Wi-SUN FAN 1.1 capabilities (March 2025): https://blogs.cisco.com/industrial-iot/cisco-ir8100-among-first-products-certified-for-wi-sun-fan-1-1
  13. T&D India, “India’s smart meter population at 7.24 crore: Parliament” (August 2026): https://www.tndindia.com/indias-smart-meter-population-at-7-24-crore-parliament/amp/
  14. Krebs on Security, “FBI: Smart Meter Hacks Likely to Spread” (April 2012): https://krebsonsecurity.com/2012/04/fbi-smart-meter-hacks-likely-to-spread/

By Riju — about

Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *