A 5G-rated industrial LTE router goes into a cabinet, a dozen devices come online, and every tap on the HMI panel takes two seconds while camera feeds stutter. A more expensive model is swapped in and nothing changes. The pattern is common on shop floors, and the device is rarely the cause. The order of judgement is what goes wrong: bandwidth is assumed to be short, a bigger box is bought, and one question is skipped, which segment is actually queuing.
Slow can be split into three segments: what the air interface delivers, what the device forwards, and what the configuration allows. Three measurements, one for each.
Count the uplink, not the total bandwidth. Downlink peaks are marketing figures, and a site that stalls almost always stalls on the uplink.
The inventory is best written in four columns: device name, interface, address, and the destinations it is allowed to reach. Once the list is complete, the number of devices sending data outward is known, along with where that data goes.
Serial telemetry and IO states are negligible: tens to hundreds of bytes every few minutes, tens to hundreds of megabytes in a month. Video is the bulk. The formula is bitrate times seconds, divided by eight and by one thousand and twenty-four. Continuous 1080p uplink at 2 Mbps runs about 21 GB a day, roughly 600 GB a month. Keeping recordings local and sending only alarm clips is therefore the cheapest single step. Batch program downloads and history playback are large per event but happen a few times a day, and a night shift window absorbs them.
The result usually sits far below the rating of any model on the list. What needs checking then is not the size of the purchase but the segment where the queue forms.
A single public speed test measures the downlink and does not match what the site feels. Two figures are needed for comparison: the latency between two devices inside the site, and the latency from this device out to the platform. Both slow means sustained heavy traffic inside. Inside normal with outbound slow puts the problem on the egress segment.
The same position is then measured in three states: cabinet door closed, door open, and antenna moved outside the cabinet on a feeder. The difference is what the metal enclosure takes away, and that figure belongs in the handover document as a reference for next year's retest. Take one reading in the morning, one at midday and one in the evening, recording the steady value rather than the peak, since base station load shifts with the hour.
When one particular device needs watching, the built-in ping, traceroute, name resolution and packet capture tools are enough; no laptop has to be carried on site.
One confusion is worth clearing up: turning a borrowed wireless link into a single Ethernet uplink is a sound approach, and the adapter costs little. What decides the experience is how much that borrowed link can give and how long it holds.
The rated peak is a ceiling, not a promise. 5G SA figures of 2.4 Gbps downlink and 900 Mbps uplink are theoretical and lose ground in a closed workshop, among dense metal, and when the base station is busy.
Port speed is not necessarily the bottleneck. 4G uplink sits in the 50 Mbps range in theory and lower in practice, so a 100 Mbps port is no constraint at that level. Gigabit ports serve traffic inside the site and lateral movement, not the egress.
Device capacity is not session count. 256 refers to associated wireless clients; sessions are a separate figure. A camera holds one or two streams, a host computer a dozen long connections. Twenty devices may open forty sessions, or four hundred.
There is one way to judge: compare against the largest number on site. Peak outbound bandwidth is one, the number of device pairs exchanging traffic internally is another, and the larger wins. Headroom follows how many devices are likely to be added within a year, not a doubling of today's peak.
Rate limiting invites retransmission and the queue gets tighter. Set priority for critical traffic first instead of capping everything at the start.
Rule sets grow linearly with device count: allow rules added one by one with the catch-all deny last, so twenty devices mean twenty-odd rules, and a reboot is needed before they take effect. Verify on the spot afterwards, reachable for those on the list and blocked for those off it, because that round of testing is itself the handover record.
When a screen is checked remotely, tunnel reconnection is often slower than link switching, and a live link does not mean a live tunnel.
With many wireless clients, look at the live session table on the device rather than at the signal bars.
Where many devices exchange traffic internally, the gigabit edge model fits: two ports configurable as WAN or LAN, six LAN plus two SFP slots with one of them shared with a copper port, auto-sensing gigabit with network isolation transformers, Wi-Fi 6 dual band rated 2976 Mbps and 256 clients, USB and SD expansion that keeps recordings local, wide-range power with reverse polarity protection, DIN rail mounting.
For a small site with dozens of telemetry points in a tight cabinet, the compact model fits: three Fast Ethernet ports, 2.4 GHz wireless at 300 Mbps for up to twenty clients, 104 x 102 x 28 mm, terminal block and DC jack for redundant supply, daily and monthly traffic bar charts with a live session view in the web page.
Where the uplink has to be sufficient and the link and its records managed together, the 5G model fits: SA and NSA, four gigabit ports with one reconfigurable as LAN, dual SIM with an optional embedded eSIM as the second, serial terminals for legacy meters, and three alarm types covering traffic overrun, offline and weak signal, pushed by email and SMS.
If the bottleneck is that judgement all sits in the cloud, aggregation can happen on site before traffic leaves: an industrial computer with 4G/5G, Ethernet and Wi-Fi as mutual backups, with routing, VPN and firewall in the same enclosure.
The order is count the demand, measure the path, check the device, and only then touch the configuration. How large an industrial LTE router should be is rarely the first question; which segment is queuing is.
Three figures are worth recording before handover: the steady uplink in the busiest hour, the session count and packet loss with every device connected, and the recovery time after the primary link is pulled. When someone asks the following year why the network is slow, those three are the starting point.
1. What separates an industrial LTE router from a home router?The SIM slot is the dividing line. Then three things follow: whether power input is wide-range with reverse polarity protection, whether the enclosure and temperature range are prepared for cabinets and outdoor sites, and whether the unit takes a DIN rail, carries a watchdog and offers a remote management entry. The numbers are concrete: 9 to 60 V DC on the edge model, operating ranges of -35 to 75 °C, -25 to 75 °C or -20 to 70 °C depending on the series, aluminium housings rated IP40, and a hardware watchdog that restarts the unit after a hang.
2. Should the choice be 4G or 5G?It follows the uplink requirement. Dozens of telemetry points with occasional images leave more margin to estimate on 4G; several video streams running together, or batch data that has to finish inside a short window, is where 5G earns its place. Before ordering, confirm whether the area has SA, NSA or only LTE, since coverage gaps fall back automatically. A 5G model carrying both SA and NSA still registers normally on a network that has not migrated yet, which keeps one hardware version usable across sites at different stages.
3. Without a fixed public IP, can equipment still be maintained remotely?Yes. Carriers mostly hand out private addresses, so connections initiated from the public side do not get in; the tunnel is instead opened outward from the device on site. A remote management platform can also open the built-in web page directly to change parameters, reboot and upgrade firmware.
4. How many terminals can one unit carry?Separate two figures first: associated clients and concurrent sessions. The former is the capacity printed in the datasheet, the latter depends on how many connections each device opens, one or two for a camera, a dozen for a host computer. The most reliable reading comes from the live session table during the busiest hour. A 256-client rating on a Wi-Fi 6 edge model is the association limit reached in open air, and office environments with obstacles cut the usable coverage to a fraction of it.
5. How is monthly data volume estimated?Reporting interval times packet size times number of points, then double it as margin. A point reporting every five minutes with a 200-byte payload comes to a few megabytes a month; the same point on a one-minute cycle multiplies that fivefold. Video is calculated separately, bitrate times seconds divided by eight and by one thousand and twenty-four. Set the threshold at seventy to eighty percent of the plan and push the overrun alarm to email and phone, rather than finding out from the bill.
6. Will data be lost while the link is down?It depends on which side holds the buffer. Caching on the device with retransmission after recovery and re-pulling from the platform side behave differently, so the question is worth settling before ordering. A device-side buffer also has a ceiling, stated as hours or megabytes, and a long outage past that ceiling loses whatever arrives after it. Acceptance is to pull the primary link, record the recovery time, and confirm whether data was retransmitted or dropped during that wait.
7. What are the two SIM slots for?Avoiding blind spots by switching carriers, and main-standby failover. The two cards are best kept on different carriers, otherwise one base station fault takes both links down together.
8. How should antennas be installed?Use the external connectors, run the feeder outside the cabinet, log the signal strength once and then adjust, since the before-and-after difference is the only evidence that something improved. Metal enclosures affect the signal a great deal, and measuring three states, door open, door closed and antenna outside, is the least effort. A short feeder to an antenna mounted outside usually beats a long cable routed through the cabinet, and the connectors are standard
