URLLC, short for ultra-reliable low-latency communication, is the 5G service category designed to deliver small data packets with user-plane latency of about 1 millisecond and a success probability of at least 99.999 percent. It is one of the three usage scenarios that shaped 5G networks, alongside eMBB for high throughput and mMTC for massive numbers of low-power devices.
In telecommunications, URLLC targets traffic in which a late packet is as harmful as a lost one: industrial motion control, protection systems in power grids, remote operation of machines and cooperative vehicle functions. Meeting both targets at once requires changes across the radio interface, the 5G Core and the placement of computing resources.
How URLLC achieves low latency and high reliability
URLLC reduces delay by shortening transmission time intervals and by removing scheduling handshakes, and it raises reliability through redundancy and conservative coding. These measures cost spectral efficiency, which is why URLLC capacity is far lower than the capacity of the same carrier used for broadband.
On the radio side, 5G NR supports mini-slot transmission of 2, 4 or 7 OFDM symbols instead of a full 14-symbol slot, and wider subcarrier spacing such as 30, 60 or 120 kHz that shortens each symbol. Configured grant uplink lets a device transmit on pre-allocated resources without first sending a scheduling request, saving a full request and grant cycle.
Reliability comes from low-rate modulation and coding tables tuned for a block error rate near 10 to the power of minus 5, PDCP duplication over two carriers or two cells, and repetitions of the same transport block. Downlink preemption allows the gNodeB to interrupt an ongoing eMBB transmission so that an urgent URLLC packet goes out immediately.
The table below maps the main techniques to the target they serve.
| Technique | Primary target | Effect | Cost |
|---|---|---|---|
| Mini-slots | Latency | Transmission can start within a fraction of a slot | Higher control overhead |
| Wider subcarrier spacing | Latency | Shorter symbols and slots | Larger guard and cyclic prefix share |
| Configured grant uplink | Latency | No scheduling request round trip | Reserved resources may sit idle |
| PDCP duplication | Reliability | Two independent copies over different paths | Doubles radio resource use |
| Low-rate MCS tables | Reliability | Robust decoding at low signal quality | Lower spectral efficiency |
| Downlink preemption | Latency | Urgent data overrides broadband traffic | Degrades the preempted eMBB block |
URLLC performance targets and 3GPP releases
The baseline URLLC requirement set by ITU-R for IMT-2020 is a 32-byte packet delivered with 99.999 percent success within 1 ms of user-plane latency. 3GPP introduced the first URLLC features in Release 15 and strengthened them substantially in later releases.
The following table summarizes how the specifications evolved.
| Release | Main additions | Typical target |
|---|---|---|
| Release 15 | Mini-slots, low-rate MCS, configured grant, duplication | 1 ms, 99.999 percent |
| Release 16 | Enhanced scheduling, multi-TRP, TSN integration, intra-UE prioritization | 0.5 to 1 ms, up to 99.9999 percent |
| Release 17 | Unlicensed band operation, improved feedback, positioning support | Industrial IoT profiles |
| Release 18 and later | 5G-Advanced refinements, XR and deterministic traffic | Use-case specific |
Latency figures in the standards refer to the radio access segment. End-to-end delay also includes transport, core functions and the application server, which is why practical deployments place the user plane function and the application at the network edge.
URLLC compared with eMBB and mMTC
The three 5G service classes optimize different and partly conflicting goals. eMBB maximizes data rate, mMTC maximizes connection density and battery life, and URLLC maximizes timeliness and certainty of delivery for small payloads.
The comparison below shows why one radio configuration cannot serve all three equally well.
| Attribute | eMBB | URLLC | mMTC |
|---|---|---|---|
| Main goal | Peak and average throughput | Latency and reliability | Device density and energy |
| Typical payload | Large, continuous | Small, periodic or event-driven | Very small, infrequent |
| Latency target | About 4 ms user plane | About 1 ms or lower | Seconds acceptable |
| Reliability target | Best effort with retransmission | 99.999 percent or higher | Moderate |
| Spectral efficiency | Very high | Low | Low |
Operators separate these classes with network slicing. A URLLC slice, identified in standards by the slice/service type value 2, receives dedicated scheduling priority, edge user plane functions and strict quality-of-service flows, while broadband traffic shares the remaining resources.
Where URLLC is deployed and how it is configured
Most production URLLC today runs in private 5G networks at factories, ports, mines and utilities rather than in public consumer networks. These sites control their spectrum, coverage and interference, which makes the stringent targets achievable and verifiable.
A practical URLLC deployment usually combines the following elements.
- A 5G SA architecture with a 5G Core, because the full feature set is not available in 5G NSA mode.
- Local user plane functions and edge computing to keep the end-to-end path short.
- Time-sensitive networking integration so that 5G behaves like a bridge in an industrial Ethernet network.
- Dense cell planning and multi-TRP coverage to avoid single points of radio failure.
- Strict traffic engineering, keeping URLLC load well below cell capacity.
Typical application requirements vary widely, as the next table shows.
| Application | Latency budget | Reliability need |
|---|---|---|
| Motion control in machine tools | 0.5 to 2 ms cycle time | 99.9999 percent or higher |
| Mobile robots and AGVs | 5 to 10 ms | 99.999 percent |
| Power grid differential protection | Below 10 ms end to end | 99.999 percent |
| Remote crane operation | 10 to 20 ms | 99.99 to 99.999 percent |
Limitations and common misconceptions about URLLC
The most common misconception is that every 5G connection delivers 1 ms latency. In public networks, typical round-trip times are often 10 to 30 ms or more, because URLLC features are rarely activated for consumer traffic and the path to internet servers is long.
- Reliability is statistical: 99.999 percent still allows one failed packet in 100,000, so applications need safe fallback behavior.
- Device support is limited, since many modems implement only part of the URLLC feature set.
- Capacity is expensive, because robust coding and duplication consume several times the resources of broadband traffic.
- Interference and blockage in metal-rich environments can break guarantees unless coverage is redundant.
- Measurement is difficult, as proving six-nines reliability requires very long test campaigns.
Put simply
Most mobile data is like parcel delivery: a few seconds of delay rarely matters. URLLC is the express courier of 5G, built for tiny messages that must arrive almost instantly and almost never get lost.
Think of a robot arm that must stop the moment a sensor detects a person nearby. If the stop command is late, the result can be dangerous. URLLC reserves fast lanes on the network, sends important messages twice when needed and keeps the processing close by, so that such commands arrive in about one thousandth of a second.
In practice
An automotive parts plant replaces cabled connections to 40 automated guided vehicles with a private 5G SA network running a dedicated URLLC slice. The vehicles exchange position and safety messages with a central controller every 10 milliseconds, and an emergency stop must reach all of them within 20 milliseconds.
Engineers deploy six indoor radio units with overlapping coverage, place the user plane function and the fleet controller in an on-site edge server, and enable configured grant uplink with PDCP duplication. Over a four-week test the network delivers 99.999 percent of safety messages on time, with typical radio latency under 5 milliseconds. The plant removes trailing cables and floor rails, and reconfiguring a production line now takes days instead of weeks.