The unit at the bottom of a control cabinet rarely has any paperwork behind it. A small-form computer, a multi-port network card, an open-source firmware image: it runs, the data reaches the platform, and the bill comes to a fraction of a finished device. At the prototype stage this is an efficient way to work, and the same box can carry a data-collection script, saving a gateway. The trouble starts when twenty of them have to ship, sit inside a customer's plant, and pass that customer's IT acceptance. The question then moves: not whether it runs, but who owns it for the following few years.
Between an industrial 4G router and a self-assembled unit, the comparison that matters is not compute. It is who carries the device when it fails. Sorting it by stage makes that clearer.
A collection scheme needs a script added, an upload format changed, an unfamiliar meter connected. The self-assembled box can be altered the same evening; a vendor's firmware waits for the next release. Self-assembly wins this stretch, provided the unit lives one or two weeks, the person who built it stands next to it, and it serves a single workstation. A bench hides the rest: a socket within reach, someone watching, twenty-something degrees.
A cabinet does not relax because the device inside was built in-house. The same conditions apply: whether there is a free length of DIN rail, how far the 24 V rail sags when a motor starts, how much hotter the cabinet runs than the room in summer, and where the antenna can leave a steel enclosure.
These four items sit in the least conspicuous corner of a finished product's datasheet: a 104×102×28 mm housing, DIN-rail or wall mounting, wide-voltage input with reverse-polarity protection, terminal block plus DC socket for redundant supply, -20 to 70 °C, and hardware and software watchdog. The USR-G806w lists them on the same page as its three alarm types: device offline, weak signal, traffic overrun. The self-assembled equivalent is usually a 12 V adapter, a printed bracket, and a temperature curve nobody measured.
Antenna and Ethernet port are the two items skipped most often. A cellular module with its antenna inside the box loses signal the moment the door closes, and the only fix is to bring the antenna outside; a finished device puts the connector on the panel, where the rubber antenna can be swapped for a feeder cable. Interference from drives and contactors also travels down the network cable into the NIC, and the port on a finished device is specified at 1.5 kV isolation, a layer most onboard NICs do not have.
Ports go beyond Ethernet. Cabinets often hold meters and temperature controllers that offer nothing but RS485, which means an extra serial server in a self-assembled build; a built-in serial terminal removes that device, its power supply and one round of acceptance. The terminal on the USR-G816 covers 1200 to 460800 baud, the top rate on RS485.
The first unit is brought up from one person's memory; the tenth needs a template. Rule tables, address ranges, APN settings and tunnel parameters end up spread across scripts, memory cards and chat logs, and the next engineer starts from nothing. What a finished device offers here is configuration that can be copied and recovered: export as a template and import it in batches, open the built-in web page remotely to change parameters, reboot or upgrade firmware, and every change keeps a date. The USR-G816 class of device imports an OpenVPN profile and certificate in one click, connects as a client to three servers, and can act as a server itself.
Firmware versions belong to the same argument. A self-assembled unit follows a community release, and when one plugin is abandoned the whole collection chain needs a replacement; a finished device lets the firmware version be pinned, offers a keep-configuration option during an upgrade, and rolls back to an archived configuration.
Multi-site rollouts carry one more rework trap: units of the same model tend to leave the factory on the same subnet, and several of them joined together collide on addresses. Assigning addresses by site number before shipment avoids it.
During site acceptance the customer's IT team rarely asks how fast the link runs. What comes up is which standards the unit-level certification covers, what EMC level was tested, whether serial and Ethernet ports are isolated, and which document revision records the firmware upgrade path. This is the hardest step for a self-assembled build: each board carries its own certification, and assembled together they are not one certified device. No parameter makes up for that, and checking the target model against the certificate list before ordering is cheaper than producing the paperwork afterwards.
An interface and rule list is usually requested as well: which device holds which address, and which rule permits traffic to where. A device that exports its configuration produces this as a by-product; a self-assembled build has to write it from scratch.
The acceptance steps can be fixed at the same time: pull the primary link once, time the recovery, then confirm whether the business layer retried or stopped in that gap. Both have to pass before acceptance closes.
The cost curve of a self-assembled build bends upward. The person who wrote the firmware is reassigned and nobody can read the scripts; a major system upgrade breaks the dependencies; there is one spare, and production waits while a replacement is sourced. What a finished device offers is replaceability: two or three interchangeable tiers in the same series, images and configurations archived with site number and date, and a spare flashed with the same image on acceptance day and actually swapped in once, because an image nobody has tried does not count.
Counting spares by unit alone also fails: one antenna or power terminal takes a site dark.
Two situations hold. One is volume large enough to absorb certification and firmware maintenance. The other is a device that is part of the company's own product, where the form factor has to fit a proprietary housing; at that point the work is not DIY but custom development, which is what a supplier's ODM and EMS lines exist for. The test is concrete: name who maintains the firmware for the next three years, who issues the certificate, and who stocks the spare. A pilot of three to five sites cannot absorb either cost, and the saving is usually spent on the first service visit.
Both routes move data out of the plant. The difference shows on the day the device goes in and in year three. Buying an industrial 4G router buys the four small items, configuration that can be copied, documents that can be produced, and a name to call; self-assembly buys the cheapest few days, on the assumption that someone can carry it afterwards.
1. What is the real difference between an industrial 4G router and a self-assembled soft router?
Not compute. A self-assembled unit is often faster and more flexible to modify; what it lacks is a finished form factor and ownership: wide-voltage and wide-temperature ratings, DIN-rail mounting, antenna routing, watchdog, remote management, unit-level certification and spares. None of these matter at the prototype stage; every one of them shows up in a customer plant.
2. Is building one cheaper? How should the bill be counted?
The hardware gap shows up on the quotation only. The rest of the bill is site visits and downtime, the customer's IT re-acceptance and admission documents, and who maintains configuration and firmware for the next three years. The more units and the more remote the site, the larger those three items grow.
3. Why do industrial routers keep emphasising wide-voltage input?
The cabinet is nominally 24 V, and motor starts plus voltage drop over long cable runs pull it down to the teens for a moment. The range matters more than the nominal figure: a DC 9-36 V input with reverse-polarity protection survives a miswired terminal and a transient sag, and two supply inputs on separate circuits remove one cause of downtime.
4. Whose temperature does a wide-temperature rating describe?
The air around the device, not the outdoor temperature. Inside a cabinet that adds the device's own heat and the heat from a nearby drive, and a sealed outdoor enclosure adds solar gain. The practical method is to measure intake air and housing surface temperature during the hottest week of the year and stay twenty degrees or more below the rated limit.
5. With no public IP address, how is a device in the cabinet maintained remotely?
A cellular link usually hands out a carrier private address, commonly in the 10.x or 100.64.x range, so inbound connections from the internet never arrive and port forwarding is not a route. The device opens a tunnel outward instead, or the remote management platform opens the built-in web page to change parameters, reboot or upgrade firmware. IoT and private-network SIMs also need the APN set at activation; without it the modem never registers.
6. How many devices can an industrial 4G router support?
Three numbers need separating: associated wireless clients, concurrent sessions, and physical ports. The Wi-Fi on a USR-G806w carries up to 20 clients, which suits a cabinet with a few devices; a dozen devices in one station that also need wireless coverage is where a model rated at 256 clients, such as the USR-G809s, comes in. Sessions follow the actual traffic: one camera holds one or two streams, one supervisory PC holds a dozen long connections.
Uplink demand decides it, not downlink peak. Dozens of telemetry and status points are easier to size on 4G; multiple camera streams, or a batch of data that has to clear within one window, are what bring 5G into the picture. Local coverage needs checking first as well, whether SA, NSA or LTE only, since a 5G unit falls back when coverage is absent.
8. Is data lost when the link drops?
The first question is whether retransmission sits on the device or on the platform. A working failover is only step one; whether the data actually catches up after recovery is the acceptance item, and sampling intervals set too tight multiply the backlog. Confirm failover takes
