A merging unit (MU) converts the analog secondary signals from current transformers (CTs) and voltage transformers (VTs) into time-synchronized digital samples and publishes them as IEC 61850 Sampled Values (SV) on an Ethernet process bus. It sits in the switchyard or the switchgear, close to the instrument transformers, and replaces the long copper runs between the yard and the relay panel.
Any device on the process bus can subscribe to that stream: a protective relay, a digital fault recorder (DFR), or protection functions running as software on a substation server. Every digital substation design, and every vPAC design, starts here.
| Merging units at a glance | |
| What it does | Digitizes CT and VT secondaries and publishes IEC 61850 Sampled Values |
| Where it sits | In a yard cabinet or in the switchgear, next to the instrument transformers |
| Key standards | IEC 61850-9-2 (SV mapping), IEC 61869-9 (digital interface), IEC 61869-13 (stand-alone merging unit), UCA 9-2LE guideline |
| Protection sample rate | 4,800 samples per second (80 samples per cycle at 60 Hz) |
| Metering and power quality rate | 14,400 samples per second (IEC 61869-9) or 256 samples per cycle (9-2LE) |
| Time synchronization | ±1 µs, typically PTP per IEC/IEEE 61850-9-3 or IEEE C37.238-2017 |
| Network redundancy | PRP or HSR per IEC 62439-3 |
| Related devices | Remote I/O for breaker status and trip; a process interface unit (PIU) combines both |
| Role in vPAC | Required, because protection functions on a server have no analog inputs and no output contacts |
What is a merging unit?
Start with what feeds it. CTs and VTs are the instrument transformers that make high-voltage quantities safe to measure. A 2000:5 CT turns 2,000 A of line current into 5 A. A VT steps the line voltage down to a secondary of 115 or 120 V line-to-line. IEEE C57.13 standardizes those ratings, which is why relays, meters, and merging units in North America are built for 5 A and 115 or 120 V inputs. Many US utilities still call the VT a PT (potential transformer), and on transmission lines the voltage source is often a coupling capacitor voltage transformer (CCVT). To a merging unit, all of them are voltage inputs.
The merging unit sits at the boundary between the analog and digital parts of the substation. On its input side it takes the same CT and VT secondaries that would otherwise land on relay terminals. On its output side it publishes a digital stream of instantaneous values that any IEC 61850 device can read.
The name describes the job. A merging unit samples the phase and neutral currents and the phase and neutral voltages at the same instant and merges them into one time-coherent dataset. The term comes from IEC 60044-8, the 2002 standard for electronic current transformers, which described the merging unit as the device that combines several sensor outputs into one digital interface. IEC 61869-9 has since replaced that interface.
Merging unit, remote I/O, and PIU: which device does what
Three device names come up in every process bus project, and vendors don't use them consistently. Only the first comes from an IEC standard.
| Term | What it does | Where the term comes from | How US engineers use it |
| Merging unit (MU), stand-alone merging unit (SAMU) | Measures: digitizes CT and VT signals and publishes Sampled Values | IEC 61869-9 and IEC 61869-13 | Universal in IEEE papers, guides, and utility pilot projects |
| Remote I/O (RIO) | Switches: reads breaker status and operates trip and close outputs on GOOSE commands | IEEE PSRC Working Group K15 ("remote input/output module") | Well understood; vendors also say breaker interface unit, switchgear interface unit, or bay I/O |
| Process interface unit (PIU) | Both, in one box | IEEE PSRC Working Group K15 and vendor product lines | Common, and often used loosely for any yard device on the process bus |
This article keeps the first two apart. A merging unit only measures, and remote I/O only switches. A PIU is a merging unit with integrated remote I/O. IEC 61869-9 draws a similar line with its conformance classes: a class a merging unit publishes Sampled Values and nothing else, and GOOSE, the message type remote I/O depends on, starts at class b.
Remote I/O also shortens the breaker control cable, which carries two conductors per trip coil (transmission breakers often have two coils), the close circuit, the 52a and 52b contacts, and usually supervision circuits. With remote I/O in the yard cabinet, all of that copper ends a few feet from the breaker. The figures draw it as two bundles, the trip circuit and the status pair.
How a merging unit changes the substation
Figure 1. In a conventional substation, every secondary circuit is a point-to-point copper run.
In a conventional substation, every CT and VT secondary runs as control cable through the cable trench to test switches and a relay panel in the control house. Breaker status and the 125 Vdc trip circuit take the same route. A CT circuit loops in series through every device that needs the current, each device adds burden, and adding a meter or a DFR later means rewiring a CT circuit.
Figure 2. The merging unit moves the analog-to-digital boundary out to the yard. Remote I/O in the same cabinet handles the breaker.
A merging unit moves the analog-to-digital boundary out to the yard. It sits in a yard cabinet next to the breaker, so the copper runs are a few feet instead of a few hundred. Fiber carries the Sampled Values to the process bus switches in the control house, where the relay subscribes. A DFR or a PMU can join later without touching the CT circuit.
The trip path changes as well. The relay sends a GOOSE trip message over the process bus. Remote I/O in the same yard cabinet receives it and energizes the breaker trip coil through a trip-rated output. It also reads the 52a and 52b contacts and publishes breaker position as GOOSE.
There is a safety benefit too. An open CT secondary, one of the more dangerous hazards in a relay panel, can now only happen on a short run inside the yard cabinet. And because the link to the control house is fiber, ground potential rise in the yard during a fault stays in the yard.
How a merging unit works
Figure 3. The analog signal chain inside a merging unit. Remote I/O is a separate function and not shown.
Inside, a merging unit follows a fixed sequence:
-
Input transformers isolate the CT and VT secondaries and scale them to electronic signal levels.
-
An anti-aliasing filter removes frequency content that the sampling rate cannot represent.
-
An analog-to-digital converter digitizes each channel. Digital filtering and calibration correct magnitude and phase errors from the analog stages, and the MU resamples to the published rate where needed.
-
The MU aligns each sample set to the station time reference and tags it with a sample counter (SmpCnt) and a synchronization flag (SmpSynch).
-
The SV publisher packs the samples into an Ethernet frame and sends it out on both network ports.
The sample counter is how relays line up data from different merging units. A synchronized MU resets SmpCnt to zero at the top of each second and counts to 4,799 at 4,800 samples per second. Under 9-2LE the frame carries no timestamp at all; the counter does that job.
Converter resolution and filter design vary by vendor. What belongs in your specification is the published output and its accuracy class.
Sampled Values: IEC 61850-9-2LE vs. IEC 61869-9
IEC 61850-9-2 defines how sampled values map onto Ethernet frames but leaves many parameters open, so two companion documents define what a merging unit actually publishes. This is where many procurement specifications go wrong.
One term first. An ASDU (application service data unit) is one sample set: the value of every channel in the dataset, taken at one sampling instant, together with the sample counter and the synchronization flag that describe it. An Ethernet frame can carry one ASDU or several, and that choice is the main difference between the two profiles.
Figure 4. Same sample rate, different framing.
IEC 61850-9-2LE is an implementation guideline from the UCA International Users Group, and most early process bus projects used it. It ties the sample rate to system frequency (80 samples per cycle) and fixes the dataset at four currents and four voltages. IEC 61869-9 is the IEC standard that followed. Its preferred protection rate of 4,800 samples per second applies at 50 and 60 Hz alike, and datasets are configurable through SCL. The notation F4800S2I4U4 reads as 4,800 samples per second, two ASDUs per frame, four currents, and four voltages.
| IEC 61850-9-2LE | IEC 61869-9 | |
| Document type | UCA users group guideline | IEC standard (2016) |
| Protection sample rate | 80 per cycle (4,800/s at 60 Hz) | 4,800/s at 50 or 60 Hz |
| ASDUs per frame | 1 | 2 |
| Frames per second | 4,800 at 60 Hz | 2,400 |
| Dataset | Fixed, 4 currents and 4 voltages | Configurable, I4U4 preferred |
| Metering variant | 256 per cycle, 8 ASDUs per frame | 14,400/s, 6 ASDUs per frame |
Two things follow from the framing. With two samples per frame, the first sample waits for the second before the frame goes out, so half of all samples pick up one extra sample interval of 208.3 µs. No protection algorithm will care, but it belongs in the latency budget.
The bigger effect is on packet rate. Bandwidth drops only a little, since two samples share one frame header (a 9-2LE protection stream runs at roughly 5 to 6 Mbps on the wire), but the packet rate is cut in half. A 12-bay station with duplicated protection has 24 streams: 115,200 frames per second under 9-2LE, 57,600 under IEC 61869-9. A hardware relay rarely notices. A server running protection functions does, because every frame costs interrupt and processing time, and that sets how many bays one host can protect.
IEC 61869-9 also makes SmpSynch more than a yes-or-no flag: 0 means not synchronized, 1 a local clock, 2 a global clock traceable to UTC, and 5 to 254 identify specific local clocks, so a relay can tell whether two streams share a time source.
If your specification only says "4,800 samples per second," a 9-2LE merging unit and an IEC 61869-9 merging unit can both comply and still put different traffic on your network. Name the profile.
Metering needs its own stream. The protection stream serves relays, DFRs, and PMUs well. Revenue metering and power quality analysis want a higher rate and, more importantly, a more accurate signal, and that accuracy is set before the merging unit. A relaying-class CT core (C class under IEEE C57.13) is built to stay accurate at fault current. A metering-class core (0.3 class, for example) is built to be accurate at load current. A merging unit fed from a relaying core will not produce revenue-grade data, however fast it samples. So the metering stream comes from inputs wired to a metering core and goes out separately, at 14,400 samples per second under IEC 61869-9 or 256 samples per cycle under 9-2LE. Many merging units publish both streams at once.
Types of merging units
Figure 5. Where the digitization happens determines the type.
A stand-alone merging unit (SAMU) connects to conventional CTs and VTs. IEC 61869-13 covers it and explicitly includes instrument transformers built to IEEE C57.13, so 5 A CTs and 115 or 120 V VTs are in scope. Its output follows IEC 61869-9. The SAMU is the natural fit for retrofits, because the existing instrument transformers stay in place.
Non-conventional instrument transformers (NCITs), such as Rogowski coils, optical current sensors, and low-power voltage dividers, produce a low-power or optical signal (the IEC 61869-6 family covers the low-power types) and usually come with the merging unit built in. Air-core sensors do not saturate, which helps protection during high-current faults with DC offset. NCITs fit greenfield stations and gas-insulated switchgear (GIS) best, because they replace the instrument transformers themselves.
Some protective relays publish SV from their own analog inputs, a practical way to share an existing CT circuit with more subscribers without rewiring it.
Switchgear-integrated merging units are installed at the factory inside metal-clad switchgear or a GIS bay. The wiring is finished and tested before the gear ships.
A process interface unit (PIU) is a merging unit with integrated remote I/O. It saves a device per bay and fits wherever one yard device should handle both measurement and switching.
| Type | Typical fit | Reuses existing CTs and VTs | Watch for |
| Stand-alone MU (SAMU) | Retrofits and brownfield stations | Yes | CT saturation still applies; yard cabinet environment |
| NCIT with integrated MU | Greenfield stations and GIS | No | Sensor and MU share one lifecycle |
| MU function in a relay | Adding subscribers to one circuit | Yes | One relay now serves several applications |
| Switchgear-integrated MU | New metal-clad switchgear and GIS | Supplied with the gear | Interoperability with other vendors' subscribers |
| Process interface unit (PIU) | Bays where one device should measure and switch | Yes | Trip-output ratings and test-mode behavior |
Time synchronization: why microseconds matter
Figure 6. Relays pair samples by counter value. A 1 ms skew between merging units shows up as a differential current that is not there.
A relay that combines currents from two merging units, a bus differential (87B) for example, assumes that samples with the same counter value were taken at the same instant. That holds only if both merging units started counting at the same instant, and nothing in the SV frame lets the relay detect a skew.
The tolerances are tight. At 60 Hz, a time error of 1 ms equals 21.6 electrical degrees, which is plenty to produce a false differential current. One microsecond equals 0.0216 degrees. That is why IEC 61850 time performance class T5, ±1 µs, is the target for process bus applications.
The 9-2LE guideline originally assumed a one-pulse-per-second (1PPS) signal over dedicated fiber. Modern designs distribute the Precision Time Protocol (PTP, IEEE 1588) over the process bus itself, using the power utility profile in IEC/IEEE 61850-9-3. In North American specifications you will also see IEEE C37.238-2017. It is compliant with IEC/IEEE 61850-9-3 and adds continuous monitoring of time inaccuracy along the path. Both profiles use Layer 2 multicast and the peer delay mechanism. To hold ±1 µs, every switch in the timing path should act as a boundary clock or a transparent clock.
Subscribers need the same time. A relay stamps its event records with it and checks SmpSynch before it trusts a stream, and a PMU cannot produce synchrophasors without time traceable to UTC. In practice every device on the process bus, remote I/O included, takes PTP time from the same grandmaster.
Plan for loss of time. Use two grandmasters with independent GNSS antennas. Under IEC 61869-9, a merging unit that loses its reference enters holdover and signals its state through SmpSynch. What happens next is a relay setting decision: a function that takes all its inputs from one merging unit can keep working, and a function that combines streams from different merging units has to block. Decide that for each protection function before commissioning.
The process bus network: PRP, HSR, and traffic
Figure 7. PRP sends every frame on two independent networks. When a link on LAN A fails, copy A stops at the break and every receiver beyond it simply keeps using copy B.
On a process bus, the Ethernet network is part of the protection system. If a switch drops SV frames, a relay loses its currents, so the network deserves the same redundancy thinking as the station battery or the trip circuit.
Parallel Redundancy Protocol (PRP, IEC 62439-3) is the most common answer. A merging unit with two ports, a doubly attached node (DAN), sends every frame on two completely separate networks, LAN A and LAN B. The receiver keeps the first copy and discards the duplicate, so a failure on one LAN causes no interruption and needs no reconvergence. Devices with a single port, called singly attached nodes (SANs), connect through a redundancy box (RedBox). LAN A and LAN B are never interconnected.
High-availability Seamless Redundancy (HSR) applies the same duplicate-discard principle to a ring, sending each frame in both directions. It saves switches, but every node sits in the forwarding path.
A few traffic rules apply regardless of topology:
-
SV (EtherType 0x88BA) and GOOSE (EtherType 0x88B8) are raw Ethernet frames, not IP packets, so IGMP snooping has no effect on them. Contain the multicast traffic with VLANs and static multicast filtering.
-
Tag SV and GOOSE with a high IEEE 802.1Q priority so they never wait behind bulk traffic.
-
Budget bandwidth per link. Dozens of SV streams fit comfortably on Gigabit Ethernet, while a 100 Mbps port into a single relay fills up faster than many expect.
-
Keep station bus traffic such as MMS, engineering access, and file transfers off the process bus or strictly separated from it.
Advantages and engineering trade-offs
The benefits are well known. The trade-offs get less attention and belong in the design from day one.
| Advantage | Why it matters |
| Less copper | Fewer control cable runs and fewer terminations to check |
| Safer CT circuits | Open-circuit hazards are confined to short runs in the yard cabinet |
| Galvanic isolation | Fiber keeps ground potential rise out of the control house |
| No added CT burden | New subscribers read the stream instead of joining the CT circuit |
| Faster changes | Adding a subscriber is a configuration change in SCL |
| Easier retrofits | New protection can be commissioned in parallel with the old system |
| Trade-off | What to do about it |
| Time synchronization becomes critical | Redundant grandmasters and PTP-aware switches, plus holdover behavior defined per function |
| The network becomes protection-critical | PRP or HSR, and clear ownership between P&C and networking teams |
| Test switches have no direct equivalent | Train relay techs on IEC 61850 test modes and the simulation flag |
| Interoperability is not automatic | Test the exact merging unit and relay pairing, including the SV profile |
| Electronics move into the yard | Specify IEC 61850-3 and IEEE 1613 ratings and a suitable yard cabinet |
| Maintenance scope may change | Agree with your compliance team how merging units fit your PRC-005 program |
| Cybersecurity | IEC 62351-6 authentication for GOOSE and SV exists but is optional; most sites rely on segmentation |
Two of these deserve a closer look for US utilities.
Start with testing. A relay tech isolates a conventional relay by opening FT-style test switches in the CT, VT, and trip circuits. On a process bus there is nothing to open. Isolation becomes logical: IEC 61850 Edition 2 devices support test and blocked modes, and SV and GOOSE frames carry a simulation flag, so a test set can inject signals that only a relay in test mode accepts. It works, and relay techs will want to see it on the bench before they trust it at cutover.
Then there is compliance. The NERC definition of a Protection System includes the voltage and current sensing devices that provide inputs to protective relays, and a merging unit sits in that signal path, so expect it to fall within PRC-005 scope. We are not aware of NERC guidance written specifically for merging units; agree on the approach with your compliance team early. Whether process bus devices fall under NERC CIP depends on your BES Cyber System categorization and connectivity.
Merging units and vPAC
Figure 8. Three generations of the same line bay with the same three circuits: trip, breaker status, and CT. In the two digital rows the SAMU publishes SV and remote I/O receives the GOOSE trip. Only the place where 21 and 50BF run changes.
Virtual protection, automation, and control (vPAC) moves protection functions from relay hardware into software on substation servers. Centralized protection and control (CPC) is the broader term for protecting many bays from one platform, virtualized or not.
Protection functions on a server have no analog inputs and no output contacts. Everything they know about the power system arrives as Sampled Values, and every trip leaves as a GOOSE message. Two yard functions are therefore a prerequisite for vPAC: a merging unit for the measurement side and remote I/O for the switching side. They can be separate boxes, as in Figures 8 and 9, or one PIU. Either way, count one set per bay, or two where protection is duplicated. The protection functions themselves, 21 and 50BF in Figure 8, run together in one virtual machine, the same way they share one hardware relay today. One VM per function is neither necessary nor typical.
This changes how a substation grows. Adding a bay adds protection function instances on servers that are already installed, plus a merging unit and remote I/O in the yard. Those yard devices are the part of the protection system that grows with every bay, so their number and quality set much of the cost and reliability of the design.
Figure 9. A simplified vPAC architecture. Each bay keeps a SAMU and remote I/O in the yard. Bay protection functions, bus protection, and the SCADA/EMS gateway run as three VMs on each of two redundant servers.
The server needs:
-
a real-time kernel or hypervisor with deterministic scheduling and CPU isolation for protection workloads,
-
network interfaces with hardware PTP timestamping,
-
PRP support, either on the network card or through a RedBox,
-
fast packet handling, for example SR-IOV or PCI passthrough, for tens of thousands of SV frames per second,
-
IEC 61850-3 and IEEE 1613 ratings for the control house environment.
Platforms in use today include KVM on real-time Linux (LF Energy SEAPATH and RHEL RT are examples) and VMware ESXi. For the full software stack, see our technical explainer on vPAC .
The lifecycles also come apart. A server refresh does not touch the yard, and replacing a merging unit does not touch the protection software. That boundary holds only if the SV and GOOSE interfaces are standard and nailed down in the spec.
Where Welotec and Westermo fit
Welotec and Westermo, both part of the Westermo group, supply the network and computing layers between the merging unit and the protection application.
For process bus switching, the Westermo RedFox-5728 offers 28 Gigabit Ethernet ports, up to 24 of them SFP, and is certified to IEC 61850-3 and IEEE 1613 Class 2. It supports IEEE 1588v2 PTP and forwards PRP, GOOSE, SV, and MMS traffic transparently, so it can serve as a switch inside LAN A or LAN B. The duplication and the duplicate discard happen at the end devices or at a RedBox.
For substation computing, the Welotec RSAPC Mk2 is a fanless 19-inch 2U substation computer certified to IEC 61850-3 and IEEE 1613. It is rated for -40 to +70 °C (-40 to +158 °F), uses 11th Gen Intel Xeon processors, and includes TPM 2.0. Operating system options include RHEL RT and VMware ESXi.
For PRP attachment of servers, the Welotec HSR/PRP Time-aware RedBox-DAN PCIe network card connects a substation server to LAN A and LAN B.
Merging unit specification checklist
Before a merging unit specification goes out for bid, check that it covers these points:
-
SV profile: IEC 61850-9-2LE, IEC 61869-9, or both, with the required sample rates and datasets
-
Inputs: 5 A or 1 A CTs, VT inputs matching your secondary voltage (typically 115 or 120 V line-to-line), or low-power sensors
-
Accuracy class for protection, and a separate metering stream fed from a metering-class CT core where needed
-
Time synchronization: PTP profile (IEC/IEEE 61850-9-3 or IEEE C37.238-2017), holdover duration, and SmpSynch behavior
-
Redundancy: PRP, HSR, or both, and the port types
-
Remote I/O, if the device is a PIU: number of status inputs and trip-rated outputs, with their ratings at 125 Vdc
-
Environment: IEC 61850-3 and IEEE 1613 ratings and a temperature range that suits the yard cabinet
-
Testing: IEC 61850 Edition 2 test modes and the simulation flag
-
Cybersecurity: IEC 62351-6 support, role-based access, and signed firmware updates
-
Interoperability: test evidence with the specific relays or vPAC platform you plan to use
-
Engineering: complete SCL files (ICD and CID) and a usable configuration tool
From copper to code
A merging unit looks like a small box in a yard cabinet. Its specification determines how precise your measurements are and how far your protection can move into software. Get that layer right and everything above it, from the network to the protection software, gets easier.
