“The screen went black at 10:14 on a Saturday. Nobody noticed until Monday.” Digital signage does not fail loudly — it fails silently, at scale, and in public. This guide turns industrial cellular router selection into an availability budget: what 99.9% actually costs across 500 stores, the four failure modes behind a black screen, and the PoE and content-delivery arithmetic nobody puts in the proposal.
Written by E-Lins Engineering Team
A screen that is dark at 10:14 on a Saturday is not a network outage. It is a campaign that did not run, a promotion nobody saw, and a brand promise broken in front of customers — and in most retail estates nobody finds out until a store manager mentions it days later. That gap between failure and knowledge of failure is where digital signage networks actually lose their uptime.
This guide is written for the engineer who has to specify a industrial cellular router for an estate of screens and wants an availability number they can defend. We work through what “three nines” means once you multiply it by store count, the four failure modes that produce a black screen and which one dominates, the switchover window you should design to, the PoE budget arithmetic that decides how many players one router can feed, and the overnight content window that decides whether a 4K campaign actually lands.
Where we quote a figure, we say where it comes from. Where our hardware is the wrong answer, we say that too — because in signage the most expensive mistake is not a slow network. It is a router that cannot power the player.
What “Three Nines” Actually Costs You
Availability percentages are abstractions until you convert them. A month has roughly 43,800 minutes, so the arithmetic is unforgiving and worth doing before anyone signs a service level.
| Availability | Down per month | Down per year | × 100 stores | × 500 stores |
|---|---|---|---|---|
| 99.0% | 438 min (7.3 h) | 87.6 h | 730 h | 3,650 h |
| 99.9% (“three nines”) | 43.8 min | 8.8 h | 73 h | 365 h |
| 99.99% | 4.4 min | 52.6 min | 7.3 h | 36.5 h |
| 99.999% | 26 s | 5.3 min | 44 min | 3.7 h |
Read the × 500 stores column, because that is the one nobody puts in the proposal. At 99.9% — a figure most wired broadband providers will happily quote — a 500-screen estate accumulates 365 hours of downtime a month, roughly fifteen solid days of screen-time lost across the network. Individually, each screen is only dark for 44 minutes. Collectively, the estate is dark for half a month.
There is a second conversion that matters more, and it is the one this article keeps returning to: digital signage network uptime is not link availability. A screen playing a locally cached playlist is serving its purpose even while the WAN is down. A screen with a healthy link and an empty local cache is black. The number that matters is content on the right screen at the right time, and that is a composite of link, power, storage and scheduling — not a router uptime figure.
The Four Failure Modes Behind a Black Screen
When a screen goes dark, the cause is almost always one of four things. They have very different signatures, very different repair times, and — this is the important part — very different detection times.
| Mode | Typical cause | How fast is it detected? | Typical MTTR | Mitigation that works |
|---|---|---|---|---|
| Link | ISP outage, carrier maintenance, SIM data cap, backhaul fibre cut | Hours to days — nobody is watching | Minutes (automatic) or days (manual) | Dual WAN with automatic failover; heartbeat to NMS |
| Power | Tripped breaker, failed PSU, PoE budget exhausted, cleaning crew unplugs it | Days | Truck roll | PoE from a protected feed; DI/DO state reporting |
| Content | Failed sync, corrupt file, expired licence, wrong scheduling window | Days — often only when someone walks past | Minutes (remote re-push) | Local cache with last-known-good fallback; proof-of-play |
| Device | Player hung, storage full, thermal shutdown in a ceiling void | Days | Truck roll | Hardware watchdog; scheduled reboot; temperature survey |
Notice that in three of the four rows the detection time dwarfs the repair time. That is the structural fact about signage: your availability number is mostly a detection problem wearing a reliability costume. A fifteen-minute carrier outage that fails over automatically costs you nothing if the screen keeps playing and nobody notices. The same outage costs you a full trading day if the player has no local cache and the alarm arrives with the Monday morning report.
This is why the two highest-leverage items in most signage designs are not the fastest modem or the biggest screen. They are a heartbeat that someone actually monitors and a local cache that keeps playing when everything upstream is gone. Both are cheap. Both are routinely skipped because neither appears in a screen’s spec sheet.
| 43.8 min |
| 99.9% per screen, per month |
| 365 h |
| Same figure × 500 screens |
| < 100 kbps |
| CMS management traffic per player |
| 5–10 Mbps |
| Typical sync bandwidth per player |
The Link: Store Broadband Is Not What the Brochure Says
Retail connectivity is usually asymmetric, contended, and installed by whoever was cheapest in the fit-out. Upstream — the direction that carries proof-of-play, heartbeats and screen health — is the direction nobody provisioned.
The practical answer is a second path that does not share fate with the first. A wired drop and a cellular link fail independently: a fibre cut in the street does not touch the tower, and a carrier outage does not touch the building’s leased line. The H900 series, the H820 and the H820QO all implement three-way redundancy across cellular, the RJ45 WAN and a Wi-Fi client, with failover, failback and load balancing between them — so a store with usable neighbouring Wi-Fi has a third option rather than a second.
| Path | Role in a signage estate | Watch out for |
|---|---|---|
| Primary wired ( fibre / DSL / cable ) | Bulk content sync, overnight push | Asymmetric upstream; shared building circuit; long ISP repair windows |
| Cellular (4G / 5G) | Failover, heartbeat, remote management, low-rate content delta | Data caps; indoor penetration in metal-ceiling retail units |
| Wi-Fi client | Third path where neighbouring or mall Wi-Fi is available | Not a substitute for a managed link; treat as opportunistic |
Design the cellular path — your 4G backup for digital signage — for management and continuity, not for bulk content. A campaign file is measured in hundreds of megabytes; a heartbeat, a proof-of-play record and a remote command are measured in kilobytes. Published industry guidance puts CMS management traffic at under 100 kbps per player. If your failover link only has to carry that, a modest data plan covers a very long outage — and you stop treating cellular as an expensive spare fibre.
| The one configuration that prevents most “the backup never kicked in” tickets |
| Point the health check at something that proves the path you care about, not at a public DNS resolver. A resolver proves the carrier is attached; your CMS or NMS endpoint proves the service is reachable. Most “dual SIM did not work” reports turn out to be a health check that was passing happily while the actual service was unreachable. |
The Switchover Window (and Why the Screen Should Never Notice)
Dual SIM failover in digital signage is a bounded interruption, not continuity, and the bound is worth knowing before you promise anything. On single-modem hardware — the architecture in the H700, H720, H750 and H820Q series — the router detects the failure, tears down the session and re-dials on the standby SIM. Field measurement puts the switch itself at 15–45 seconds, but the health-check policy dominates the total.
In acceptance testing on H750 units at three sites, measured time from path failure to confirmed connectivity on the backup SIM was:
| Health-check policy | Strong backup signal | Moderate | Marginal | Side effects |
|---|---|---|---|---|
| Conservative — 10 s timeout, 30 s interval, 3 retries | 147 s | 163 s | 211 s | No false switches in 72 h |
| Aggressive — 5 s timeout, 10 s interval, 2 retries | 68 s | 92 s | 112 s | 2 false switches in 72 h |
Now the important reframe: with a sensible digital signage failover switchover time, none of those numbers should be visible to a customer. A 147-second WAN gap is irrelevant if the player keeps playing from local storage. It is catastrophic if the player streams. This is the single design decision that separates estates that survive carrier events from estates that report them.
| Player architecture | During a 147 s WAN gap | During a 6 h ISP outage | Verdict |
|---|---|---|---|
| Streaming only | Black within seconds | Black for the full outage | Never for retail estates |
| Cached, no fallback logic | Plays last campaign, may freeze at end | Depends where the playlist was | Partial |
| Cached with last-known-good fallback | Uninterrupted | Plays approved fallback loop | Specify this |
Add one more safeguard that costs nothing: a scheduled reboot during closed hours. Players and routers that never restart accumulate state — full logs, wedged processes, leaked memory — and a weekly 03:00 restart removes a whole class of “truck roll for a reboot” incidents before they happen.
The PoE Budget: The Spec Line That Decides Your Bill of Materials
Here is the failure mode we see most often in signage, and it is entirely avoidable: the network design assumes the router will power the media player, and the router ordered is a PoE PD model — it can receive power over Ethernet but cannot supply it. The result is a PoE injector or a small switch added back into a ceiling void that was supposed to have one cable and one device.
| PD is not PSE. Check this line before you order. |
| PD (powered device) means the router takes power from the cable. PSE (power sourcing equipment) means it puts power onto the cable. In our range, PSE is available on the H900, H900pf, H900t6, H820 and H750. The H900uf, H685f, H685, H685w and H820QO are PoE PD only — they are powered by PoE, they do not power anything. |
Once you have a PSE model — the right PoE router for digital signage — the question is how much power is actually available at the far end. The standards numbers are worst-case over 100 m of cable:
| Standard | PSE output | Guaranteed at PD | Realistic for signage |
|---|---|---|---|
| 802.3af (PoE) | 15.4 W | 12.95 W | Compact Android player, small 10–15″ screen |
| 802.3at (PoE+) | 30 W | 25.5 W | Mainstream 4K player, single display |
| 802.3bt Type 3 (PoE++) | 60 W | 51 W | High-brightness player, small outdoor-rated display |
| 802.3bt Type 4 (PoE++) | 90 W | 71.3 W | Player plus peripheral, or a large-format display |
Then subtract what the router itself consumes. On these platforms that is small: roughly 1.2 W idle (100 mA at 12 V), 3.1 W while transmitting (260 mA at 12 V), and 9.6 W worst case (800 mA at 12 V). The PSE supply itself is offered in three ratings — 48 V at 1 A, 1.5 A or 2 A, i.e. 48 W, 72 W or 96 W of total budget — and that rating, not the per-port standard, is what decides how many players one router can feed.
A realistic worked example: a mainstream 4K media player draws 10–18 W. On a 48 V 1.5 A (72 W) PSE supply, allowing 10 W for the router and roughly 15% for cable and conversion losses, you have about 53 W of usable output — enough for two or three players, or one player with comfortable headroom. Ask for the 2 A supply if you intend four ports at PoE+.
One myth worth killing: the display itself rarely runs on PoE. Industry electrical guidance puts display draw at 50–100 W below 32 inches, 100–200 W from 32 to 55 inches, and 150–400 W above 55 inches — well beyond PoE for anything but the smallest units. PoE in signage powers the in-ceiling media player PoE device, the compact screen, or the router itself. Budget mains power for the glass.
The Overnight Push: Whether the Campaign Actually Fits
Content is where digital signage content delivery bandwidth is actually consumed, and it is the part most often left to “the network will handle it”. It will not, if nobody did the arithmetic.
| Content profile | Peak bandwidth | Daily volume per player |
|---|---|---|
| Images only | 5 Mbps | 50–200 MB |
| Mixed images and video | 10–20 Mbps | 200–500 MB |
| Video-heavy (1080p) | 25–50 Mbps | 500 MB – 2 GB |
| 4K video | 50–100 Mbps | 1–5 GB |
The sizing formula is simple and worth writing into the design document — especially if you are specifying a 5G router for digital signage for 4K campaigns:
| Content delivery arithmetic |
| Required bandwidth = (content size × update frequency) ÷ sync window, plus 50% overhead. Ten players, 500 MB daily update, two-hour window: 500 MB ÷ 7,200 s = 0.55 Mbps per player; × 10 players = 5.5 Mbps minimum; add 50% overhead → 8.25 Mbps recommended. |
Scale that to a real estate and the shape of the problem changes. Two thousand screens at 300 MB per day is 600 GB of content. Spread evenly over a six-hour overnight window that is an average of about 28 MB/s — roughly 220 Mbps of aggregate egress, which is perfectly manageable. The failure mode is not the average, it is the peak: if every player begins its sync at 02:00 because that is when the CMS fires, the instantaneous demand is two thousand simultaneous transfers against an upstream that was provisioned for the average. That is a thundering herd, and it is why the three scheduling strategies matter more than raw bandwidth:
- Scheduled sync — confine transfers to off-peak hours. The baseline for any estate; also the cheapest.
- Staggered sync — randomise or shard the start time across the fleet. This is the single fix for the thundering herd, and it costs one configuration line.
- Differential sync — transfer only what changed. On a campaign where one spot in twelve is swapped, this reduces the overnight volume by an order of magnitude.
Pair all three with edge caching for digital signage: the player should hold a complete, playable playlist at all times, so a failed sync degrades freshness rather than causing darkness. And remember which direction is scarce — a store’s upstream is what carries proof-of-play, screen health and remote commands. Size the plan for management traffic first, content second.
Keep Playing When Everything Upstream Is Gone
Everything so far has been about the network. This section is about making the network’s failures invisible — which is where the actual uptime lives.
Three mechanisms do the work. Local cache with a last-known-good fallback means the player always has something approved to show; when a sync fails, it plays the previous playlist rather than a black frame or a browser error page. A heartbeat with proof-of-play means the NMS knows a screen is dark before a customer does — and, critically, knows whether the screen is dark or merely unreachable, which are different faults with different fixes. Digital I/O gives the router a way to see physical state without any IP: a contact from the display’s power supply, a door switch on the enclosure, a relay driving a local indicator.
Most of the range provides 4 DI/DO; the H685t provides 2. That is not a lot, and it is not a replacement for a controller — it is enough for the three or four signals that must survive a total network loss. In a signage context the highest-value use is boring and effective: a digital input wired to the display’s mains contact, so “the screen has no power” and “the screen has no network” become distinguishable alarms instead of one ambiguous “offline”.
| FField pattern · retail estate |
| “We stopped measuring link uptime and started measuring minutes of dark screen. That single change moved the conversation from ‘which modem’ to ‘why did nobody know’ — and the fix was a cached fallback loop and a heartbeat, not new hardware.” |
| Recurring finding across multi-site retail signage rollouts · detection delay, not failure rate, dominated the loss |
Ceiling Voids, Window Glass and the Outdoors
Signage is installed in the three least friendly places in a building: above the ceiling, behind glass in direct sun, and outside. Each imposes a different constraint on the router.
Ceiling voids are the default, and they are hot. A suspended ceiling traps the heat that displays and lighting dump into it, and a router in a sealed void with no airflow sits well above room temperature — every 10 K roughly halves electrolytic capacitor life. The compensating factor is that this is exactly where a single-cable PoE installation pays for itself: no mains electrician, no socket within six feet, no extra circuit.
Window and atrium locations add solar gain and glare. The rule is the same as for any enclosure: put the router where the air is, not where the view is.
Outdoor and semi-outdoor screens — forecourts, drive-throughs, shopfront displays exposed to weather — are where the enclosure rating for an outdoor digital signage router becomes the first question, not the last. Every router in this range except the H820QO is IP30: protected against solid objects larger than 2.5 mm, with no water protection at all. An IP30 router belongs inside an enclosure. For an exposed location the correct answer is either the IP68 H820QO or a properly rated cabinet — not an IP30 router and a hope.
- Survey the void temperature before selecting hardware, not after the first summer.
- Confirm PD or PSE on the order line — PSE on H900 / H900pf / H900t6 / H820 / H750 only.
- Size the PSE supply (48 V 1 A / 1.5 A / 2 A) against the player count, with headroom.
- Rate the enclosure for the location — IP30 inside; H820QO or a rated cabinet outside.
- Route the antenna out of the metal — a ceiling grid and a metal back box make an effective shield.
- Plan the mains for the glass — PoE powers the player, not a 55-inch display.
Three Networks in One Store
A retail site runs three logical networks that have no business sharing a broadcast domain: signage, payment, and guest Wi-Fi. Segmentation is a security control, but on a store network it is also a performance one — a guest device streaming video competes with the overnight content push for the same upstream.
- VLAN 10 — Signage. Players and displays. Outbound HTTPS to the CMS and CDN, inbound denied by default.
- VLAN 20 — Payment and POS. Where card transactions are in scope, PCI DSS expects the cardholder data environment to be segmented from everything else — this VLAN is a compliance boundary, not a convenience.
- VLAN 30 — Guest Wi-Fi and store IT. Internet access, no route to either of the above.
802.1Q VLAN tagging is supported across the range, which is what makes VLAN segmentation for a retail network practical on a single-router store. On the signage VLAN, whitelist only what the player needs: 443 TCP for CMS and CDN, 80 TCP for redirects, 53 UDP/TCP for DNS and 123 UDP for NTP — time synchronisation matters more than people expect, because certificate validation and scheduled playback both depend on a correct clock. Deny inbound. That single rule removes most of the attack surface of a screen that sits in a public space.
Managing Five Hundred Screens Without Losing Your Weekend
At twenty screens you can manage devices individually. At five hundred, individual management is the failure mode — not because it is slow, but because nobody does it consistently, and inconsistent configuration is how estates drift into unreliability.
The mechanics are well established: NMS — the backbone of any NMS for a digital signage fleet — for fleet visibility and bulk operations, TR-069 for auto-provisioning, SNMP for polling state into an existing monitoring stack. The engineering question is what you do with them.
- Template everything. One golden configuration — VLANs, firewall rules, failover policy, health-check target — applied to every site, with only the site identity varying.
- Grade the alarms. A single screen offline for five minutes is a log entry. Twenty screens offline in one region is a carrier incident. If both pages the same person at 03:00, they will soon ignore both.
- Monitor the right signal. Heartbeat and proof-of-play are what make remote management of digital signage meaningful — they tell you the screen is doing its job. A router ping tells you the router is powered.
- Track data plans centrally. A SIM that hits its cap mid-campaign is a self-inflicted outage that billing could have predicted weeks earlier.
And because this is the section where estates quietly break: stagger the firmware rollout. A template pushed to five hundred devices at once is a five-hundred-site outage waiting for a bad image. Ten percent, verify, continue.
Honest Limits and Model Selection
A vendor article that does not say where the product stops is marketing. Here is where ours stops, and which model to pick for which signage job.
| Model | PoE | Wi-Fi | Cellular | Best signage fit |
|---|---|---|---|---|
| H900 / H900t6 | PD or PSE, 4 PoE LAN ports | Tri-band (t6: Wi-Fi 6) | 4G / 4G CAT6 | Store hub powering several players; high-density sites |
| H900pf | PD or PSE, 2 PoE LAN ports | Tri-band option | 5G | Flagship stores, 4K-heavy content |
| H820 | PD or PSE | 802.11 a/b/g/n/ac/ax | 4G | Single player plus Wi-Fi; three-way failover |
| H750 | PoE | 2.4 GHz n | 4G | Budget single-screen sites; MicroSD for local buffering |
| H900uf / H685f | PD only | Tri-band / dual-band | 5G | Where the player has its own power |
| H685 / H685w | PD only / adapter | b/g/n | 4G / 3G | Not for video signage — H685w is 3G WCDMA |
| H820Q | None | Dual-band Wave2 MU-MIMO | 4G | Wi-Fi-first sites where power is already local |
| H820QO | PD only | a/b/g/n/ac/ax | 4G | The only IP68 model — outdoor and exposed screens |
Two rows deserve a warning, because both are usually picked as the cheapest LTE router for retail signage and then regretted. The H685w is a 3G WCDMA router with 802.11 b/g/n — it cannot carry a video-heavy signage workload, and specifying it because it is the cheapest line item is a common and expensive mistake. And the H820Q has no PoE and no serial port: it is a Wi-Fi-first device for sites where power and content are already handled locally.
| Also worth knowing |
| Where a screen is genuinely exposed — a forecourt, a drive-through, a shopfront with weather exposure — the enclosure rating decides before anything else does. Only the H820QO is IP68. Everything else is IP30 and needs a cabinet. |
Verdict: Measure Dark Minutes, Not Link Uptime
Every signage estate we have worked on that improved its availability did so with the same three moves, and none of them were a faster modem.
First, they changed the metric from link uptime to minutes of dark screen per store per month — which immediately exposed that detection delay, not failure rate, was the dominant term. Second, they made the screen independent of the network with a cached fallback loop, turning a carrier event from an outage into a non-event. Third, they checked the boring order lines: PD versus PSE, the PSE supply rating, the enclosure rating for the actual location.
Do that, and the availability table at the top of this article stops being a threat. 99.9% with a playing screen is fine. 99.9% with a black one is not, and the difference was never the network.
| Key takeaways |
|---|
| 99.9% is 43.8 minutes per screen per month — and 365 hours across 500 screens. Multiply before you commit to a number. |
| In three of the four failure modes, detection delay exceeds repair time. A monitored heartbeat and a cached fallback loop beat any modem upgrade. |
| PoE PSE is available only on H900, H900pf, H900t6, H820 and H750. H900uf, H685f, H685, H685w and H820QO are PD only — they consume PoE, they do not supply it. |
| Size power from the supply rating, not the port standard: 48 V at 1 A / 1.5 A / 2 A = 48 / 72 / 96 W, less ~10 W for the router and ~15% for cable and conversion loss. |
| Content: 4K runs 1–5 GB per player per day. Required bandwidth = (size × frequency) ÷ window, +50%. Stagger the starts — the thundering herd, not the average, is what breaks the window. |
| Dual SIM failover is a bounded interruption: 15–45 s to switch, 68–211 s end-to-end by policy. With local caching, none of it is visible on the glass. |
Frequently Asked Questions
Q1:How much bandwidth does one digital signage player need?
Published industry guidance puts it at 5 Mbps for images only (50–200 MB/day), 10–20 Mbps for mixed content (200–500 MB/day), 25–50 Mbps for video-heavy 1080p (500 MB–2 GB/day) and 50–100 Mbps for 4K (1–5 GB/day). Cloud CMS management traffic is separate and small — under 100 kbps per player. Size the plan for management first, content second, and remember the scarce direction is upstream.
Q2:Can the router power the media player over PoE?
Only on models with PSE (power sourcing equipment): the H900, H900pf, H900t6, H820 and H750. The H900uf, H685f, H685, H685w and H820QO are PD only — they are powered over Ethernet but cannot power anything downstream. Check this line on the order; it is the most common signage specification error we see. Then size the supply: 48 V at 1 A / 1.5 A / 2 A gives 48 / 72 / 96 W of total budget.
Q3:Can PoE run the display itself?
Usually no. Display draw is 50–100 W below 32 inches, 100–200 W from 32 to 55 inches and 150–400 W above 55 inches — beyond 802.3bt for all but the smallest units. In practice PoE powers the media player, a compact screen, or the router itself; plan mains power for the glass.
Q4:How long does cellular failover take, and will the screen go black?
On single-modem hardware the switch completes in 15–45 seconds, but the health-check policy sets the total: measured end-to-end recovery was 68–112 s with an aggressive policy and 147–211 s with a conservative one. It should not matter — a player with a cached fallback loop keeps playing through all of it. If the screen goes black during failover, the architecture is wrong, not the tuning.
Q5:Can I install the router above a suspended ceiling?
Yes, and it is the most common location — but survey the temperature first. Ceiling voids trap heat from displays and lighting, and every 10 K roughly halves electrolytic capacitor life. Give the router airflow, keep it away from the display’s own heat, and make sure the antenna lead leaves the void, because a ceiling grid and a metal back box make an effective RF shield.
Q6:Which model for an outdoor or shopfront screen?
The H820QO. It is the only IP68 model in the range — every other router here is IP30, protected against solid objects over 2.5 mm with no water protection, and belongs inside an enclosure. If an IP30 model is the only option, it needs a properly rated cabinet, not a sheltered corner.
Q7:Do I need to separate the signage network from POS and guest Wi-Fi?
Yes, on both security and performance grounds. 802.1Q VLAN tagging is supported across the range: put players on their own VLAN with outbound-only HTTPS (443), HTTP (80), DNS (53) and NTP (123), deny inbound, and keep payment and guest traffic off it. Where card transactions are in scope, PCI DSS expects the cardholder data environment to be segmented — that VLAN is a compliance boundary.
Send us your estate numbers
Screen count, content profile, player power draw and where the screens physically live. We will return an availability budget, a PoE calculation and the right model for each installation position.
