Private 5G at the Edge: Networks, Spectrum and Industrial Computers

Private 5G is a 5G cellular network built for one organisation’s site, such as a factory, port, mine, airport or campus, rather than shared with the public. The 3GPP standards call it a non-public network: only devices with SIMs or eSIMs issued for that network can join, the radios cover the site the owner chooses, and the data can stay on site. That gives the operator control over coverage, capacity, latency and security in a way public mobile networks and unlicensed Wi-Fi do not.

Maintaining connectivity is one of the harder problems in industrial and IoT applications. In mobile and remote locations Wi-Fi is often not practical, and on a busy site metal structures, moving plant and other users of the unlicensed bands interfere with short-range signals. So it is no accident that embedded computer and industrial PC makers now fit cellular modems as standard options, and that many large sites are building their own 5G networks rather than relying on a public operator’s coverage.

5G began rolling out globally in late 2019 as the fifth generation of mobile networks, defined by 3GPP from Release 15 onwards. This guide explains what private 5G is and how it is built, compares it with private LTE and Wi-Fi 6, 6E and 7, covers the spectrum options in the USA, the US and Europe, explains standalone and non-standalone 5G, network slicing and 5G RedCap, and sets out what an edge computer or gateway needs to join a private network. It finishes with the questions buyers ask most often.

“We serve customers in a variety of mission-critical industries where uninterrupted, high-bandwidth data access is imperative. 5G brings real advantages for users of our hardware. The higher bandwidth of 5G suits high-resolution and high-frame-rate video capture and processing, as well as newer sensors like LiDAR.” Andrew Whitehouse, CEO, Things Embedded

Most sites choosing a private wireless network weigh three options: private LTE (4G), private 5G and enterprise Wi-Fi. The table compares them on the points that decide industrial projects. Figures are typical design targets and field results, not guarantees; real performance depends on spectrum width, cell layout and device mix.

The pattern most sites settle on is a mix: Wi-Fi indoors for laptops, tablets and fixed HMIs, and private 5G (or private LTE) across the yard, the plant or the campus for vehicles, autonomous mobile robots and cameras that move between cells. The wireless access point and the 5G small cell end up on the same network backbone.

The building blocks

A private 5G network has the same parts as a public one, scaled down to a site. Devices carry a 5G modem and a SIM or eSIM issued for the network. Small cells, called gNodeBs (gNBs), provide the radio coverage, typically mounted on buildings, lighting columns or ceilings. A 5G core handles authentication, sessions and routing; its user plane function (UPF) is the part that forwards traffic, and placing it on site keeps data local and latency low. The core then connects to the site’s operational technology: SCADA, the MES, historians and the Industrial Internet of Things (IIoT) platform.

Standalone and public-network-integrated models

3GPP Release 16 defines two kinds of non-public network. A standalone non-public network (SNPN) has its own core and identifiers and does not depend on any public operator. A public network integrated non-public network (PNI-NPN) is built with a mobile operator, using the operator’s core, a dedicated slice or a dedicated on-site UPF. The first gives the most control; the second can be quicker to deploy and easier to roam out of, for example when vehicles leave the site.

Private cellular networks did not start with 5G. Private LTE, a 4G network built for one site, has run in mines, ports, utilities and oil and gas fields for years, and many sites choosing a private network today still weigh the two. Both use SIM-based access, licensed or shared spectrum and an on-site or hosted core; the differences are in performance, device choice and maturity.

Private LTE remains a sound choice where the devices are sensors, handhelds and video, and where mature, lower-cost modules matter. Private 5G earns its place where many high-rate cameras, mobile robots or latency-sensitive control share the network, or where the network has to serve the site well into the 2030s. For the edge computer the difference is mainly the modem: an LTE module for private LTE, or a 5G module with standalone support on the site’s band, with band support checked either way. Many private 5G cores also serve LTE devices, so a site can migrate in stages.

Spectrum is the first practical decision, because it sets which modems, small cells and antennas will work. Most private networks run on mid-band spectrum between about 3.3 and 4.2 GHz, which balances coverage and capacity. The main routes by region are below; license terms change, so confirm the current rules with the regulator before planning a deployment.

CBRS in the US

The Citizens Broadband Radio Service (CBRS) is the band most US private LTE and 5G networks use: 150 MHz of spectrum at 3.55–3.7 GHz, shared under the FCC’s Part 96 rules in three tiers. Incumbent users, such as US Navy radar and fixed satellite earth stations, are protected first. Priority Access Licenses (PALs), auctioned by county, come second. General Authorized Access (GAA) lets any user take spare channels without a license fee. A spectrum access system (SAS) assigns channels to each radio as conditions change, so the tiers do not interfere. CBRS is LTE band 48 and 5G NR band n48, so a modem for a US private network needs B48 or n48 support.

Shared access in the USA

The USA has no CBRS; Ofcom’s shared access licenses play the same role. They are issued per site, for a defined area and power, with a modest annual fee rather than an auction. The 3.8–4.2 GHz band is the one most USA private 5G networks use, and it falls inside 5G band n77, so a module that lists n78 only (3.3–3.8 GHz) will not reach it. That single detail is the most common reason a 5G edge computer bought for a USA private network fails to attach.

Early public 5G networks were non-standalone (NSA): the 5G radio was anchored to an existing 4G LTE core (the EPC), and the device needed an LTE connection to use 5G at all. Standalone (SA) 5G uses a 5G core (5GC) end to end. The difference matters for industrial use, because the features that make 5G attractive on a site depend on the 5G core.

Network slicing

Network slicing divides one physical network into logical networks, each with its own guaranteed speed, latency and isolation. On a site, a slice for robot control can be protected from a slice carrying CCTV video and another for staff handhelds, so a burst of video does not delay a stop command. Bandwidth can be taken from non-critical systems and given to data-intensive or time-sensitive ones. Fleet telemetry that reports position every few seconds needs little; remote driving, as in our remote vehicle teleoperation over 5G project, needs a low-latency, high-uplink slice.

5G and Time-Sensitive Networking

Release 16 lets a 5G system act as a bridge inside an IEEE Time-Sensitive Networking (TSN) network, carrying time synchronization and scheduled traffic across the wireless link. It is early in practice, but it is the route by which wireless links may one day carry deterministic control traffic that today needs a cable.

Full 5G modules are built for smartphones and high-throughput routers: they support wide channels, four receive antennas and gigabit speeds, and they draw several watts. Many industrial devices need none of that. 5G RedCap (reduced capability), introduced in 3GPP Release 17, trims the radio to a maximum 20 MHz channel in the mid-bands, one or two receive antennas and simpler modulation options, for a typical peak of roughly 150 Mbit/s downlink. Release 18 adds eRedCap, cut further to about 10 Mbit/s for sensors and meters.

For a private network RedCap is attractive because it keeps the 5G core’s features, standalone operation and long network lifetime while lowering module cost and power. It does need a standalone network that supports RedCap devices, so check the network first. For edge computers carrying several camera streams or acting as a site gateway, a full 5G module is still the usual choice.

Faster network speeds

Each generation of mobile network raises throughput, and 5G’s design target is up to 20 Gbit/s peak downlink under the ITU’s IMT-2020 requirements. Real-world speeds are lower, typically several hundred Mbit/s to over 1 Gbit/s depending on spectrum and load, but that is still roughly ten times 4G in many deployments. The useful change for industrial systems is uplink: 5G can carry several high-resolution camera streams from a site or a vehicle, which is where most edge applications generate their data.

Lower latency

Latency is the time between sending an instruction over a network and its arrival. IMT-2020 sets a 1 ms user-plane target for ultra-reliable low-latency communication (URLLC) and 4 ms for mobile broadband, against figures of typically 30 to 50 ms on 4G networks in practice. On a private standalone network with the UPF on site, round trips of around 10 ms are typical today. That supports remote operation of vehicles and machines, near-real-time machine vision feedback and augmented reality for maintenance.

More connected devices

IMT-2020 sets a connection density target of one million devices per square kilometer, roughly ten times what 4G networks were designed for. That headroom matters at large events and dense sites, but also on factories that add hundreds of sensors, tags and cameras over the life of a network. NB-IoT and LTE-M, the low-power cellular IoT technologies, are recognized as part of the 5G family for massive machine-type communication and run alongside 5G NR.

Lower energy per bit

5G radios can switch off more of their hardware between transmissions than 4G, and a 2020 study by Nokia and Telefónica reported 5G networks as up to 90% more energy-efficient per bit of traffic carried than legacy networks. For devices, battery-powered sensors gain most from RedCap, NB-IoT and LTE-M, which use longer sleep cycles.

Private 5G networks are supplied by mobile network operators, network equipment makers, specialist private network integrators and neutral-host providers, and some large organizations build and run their own. The choice is less about the supplier than about who owns the core and the spectrum.

Whichever model is chosen, the network has to be integrated with the site’s OT network. That usually means segmenting traffic with a VLAN per application, applying zone and conduit rules from OT security practice, and deciding which systems may be reached from the cellular side. Hardware security matters too: our IEC 62443-4-2 hardware security checklist covers the device-level requirements.

How much does a private 5G network cost?

Cost depends on the site area, the number of small cells, whether the core is on site or hosted, the spectrum license, and the devices. A small indoor network with a hosted core is a very different project from a port-wide network with an on-site core and redundant radios. We do not publish prices, and the network itself is outside what we supply, but we can tell you what the edge computers, gateways and modules will need so that the device side of the budget is right.

For an edge computer or IIoT gateway, joining a private network is mostly a question of the modem and how it is fitted. The checks below cover the points that most often go wrong. Our guide to the cellular gateway covers modules, SIM slots, failover and antennas for public networks in more detail.

5G-enabled computers are now widely available for industrial use. When selecting an edge computer you may see it described as cellular ready, which may be limited to 4G. If the system only has a mini PCIe socket, it is likely limited to 4G modules, because 5G modules are built for the M.2 form factor.

5G modules for computers are designed for M.2 Key B sockets in 3042 or 3052 lengths; the key defines the pin-out and the number defines the card’s width and length. Select a 5G card that matches both the key and the length the socket accepts. If a system only takes 3052 cards, a 3042 card will not be held securely, which matters on a rugged PC designed for shock and vibration. Our guide to M.2 keys and slots explains the key types and sizes.

Upgrading legacy 4G gateways with 5G modem adapters

If an existing system is limited to mini PCIe, or a certified computer with an M.2 3042 or 3052 slot is not available, 5G modem adapters can be considered. They carry jumpers to select USB 2.0, USB 3.0 or PCIe signaling, and the default signal assignment varies between industrial gateway makers, so it must be validated with the exact modem and adapter. Always check pin-out definitions, signal voltage levels and the power the host can supply before using an adapter, and test the module in the current design before a rollout.

More SIM slots and SMA ports

Different networks have different coverage. With two SIMs, a computer can use a private network on site and fall back to a public network when a vehicle leaves, or compare two public operators and use the stronger signal. Check whether the SIM slots are accessible from outside the enclosure or only after opening it. Some modules support eSIM, which lets profiles be changed remotely. A multi-SIM router or cellular bonding can add resilience where several links must be combined.

More modems and SIMs need more antennas. The SMA connector is the common coaxial RF connector for cellular antennas on edge gateways, and we have customized both standard fanless gateways and waterproof gateways with additional SMA ports for extra antenna breakouts.

When selecting a 5G modem as part of a system design, consider the environment. If the application runs at normal room temperatures, such as an indoor digital signage media player that downloads content remotely, an industrial-grade module is not essential. A gateway in a roadside cabinet, on a vehicle or on a mast is different: modules and antennas must be rated for the same range as the computer, typically −30 to 75 °C or wider for industrial-grade modules, or the link can fail on the hottest and coldest days. For wide temperature computers we select industrial 5G modules rated for the system’s range, which removes one of the most common points of failure in the field.

Consider a container yard building a standalone private 5G network on a USA shared access license at 3.8–4.2 GHz. It needs camera analytics on straddle carriers, telemetry from reach stackers and gate automation, with a public-network fallback when vehicles leave the yard.

Every vehicle computer in that design needs n77, SA support, two SIM slots, four antennas on roof-mounted SMA leads, a wide DC input and shock and vibration ratings for the vehicle. Our 5G in-vehicle computers for a connected ambulance fleet project shows the same pattern on public networks: cameras, crew devices and a control room linked from the moment the engine starts. For bonded links over several public networks, see how an embedded PC with four SIM slots delivered bonded cellular internet at sea.

These computers take a 4G or 5G module in an M.2 Key B socket. We fit the module, SIMs and antennas you specify, check band and standalone support against your network, configure the operating system and test the complete system before it ships.

For heavier AI workloads on a 5G link, Intel Core Ultra systems put an NPU beside the CPU, and Jetson systems such as those in our Jetson Orin range run larger vision models so that only events and metadata cross the network.