Right datasheet, wrong row: what we got wrong about a cellular camera's antennas

TL;DR: Before you buy a cellular antenna for a remote camera, check whether the camera is the modem. Some of these units have a built-in radio and their own SIM, and that changes nearly every install instruction written for the category. We checked. We then wrote down "one external antenna socket" and repeated it five times across our own notes — and that number was read out of the wrong block of the datasheet's parameter table. Reading the source did not save us. A spec number has a scope, and in a datasheet the scope is drawn as a line.

There is a genre of advice about improving a cellular camera's signal, and it is mostly good advice. Aim a directional antenna. Prefer a cross-polarised panel on a fringe site. Keep the coax short. Put the router next to the antenna and run power out to the camera.

Every sentence in that paragraph assumes the camera and the radio are two devices. This article is about what happened when we checked that assumption — and then about what happened when we got the correction wrong too, which is the more useful half.

The premise nobody restated#

Our own research series opens with a line that reads like a fact and functions like an axiom: these cameras are PoE-only, so a router is mandatory at every cellular site anyway. Everything downstream inherits it. If a router is mandatory, then the router is where the SIM lives, the router is what you put next to the antenna, and the router's two antenna ports are what you are shopping for.

A day later, somebody read the datasheet.

The camera is the modem. The vendor says so on the public product page for this class of unit — "the 5G module is directly merged inside the camera" — and there is no router in the path unless you add one. Adding one is not a small upgrade to an existing box: it is a new device, a new install visit and a change to how the site is wired.

That correction is right, and it stands. Three documents dated 2026-07-21 carry the wrong premise; the document that corrects it is dated 2026-07-22. One day. We had originally written "five weeks" in this article, which was itself an unchecked number about our own documents, and the dates are printed at the top of each one.

And then we corrected it wrong#

The correcting document did not stop at "the camera is the modem". It also recorded, as a resolved fact, that the camera has a single external antenna socket — 1×50Ω SMA-Female, cited to the datasheet. It repeated that seven times in that one document — in the summary, the options table, the antenna section, the "what is wasted" list, the hardware appendix and the costed shortlist — and built a recommendation on it: a two-port MIMO panel is wasted on this camera, so buy a single-port directional instead.

That is wrong, and here is exactly how.

The datasheet's parameter table is a merged-cell grid: a label cell on the left spans several rows, and a horizontal rule separates one block from the next. The Mobile Communication Parameter block runs from Wireless Communication down through the frequency bands, the transmission speeds, and ends at SIM Card | Nano SIM. The rule falls there. The next row — Antenna | 1*50Ω SMA-Female External Antenna Connector — is the first row of the block below it, and the label on that block is LoRaWAN®.

The rows that follow it say so unmistakably once you are looking: Channel | 8 channels simultaneously, Demodulation | 1MHz, Frequency Band | CN470/IN865/EU868/…, Communication Range | Up to 15km. Those are not cellular parameters. The single SMA connector is the LoRa antenna.

The Mobile Communication Parameter block gives no antenna connector count at all. And the structure diagram two pages later settles the direction, though it is worth saying exactly what it shows rather than what it implies. The drawing has four identical tall rods and one short stub. A leader labelled LoRa Antenna terminates on the stub. A leader labelled 5G Antenna terminates on one of the four tall rods — not on all four.

That the other three are also 5G is an inference, not a caption: they are geometrically identical to the labelled one, the only separately-named antenna on the device is the stub, and no third radio appears anywhere in the specification. It is a good inference. It is still an inference, and in a piece about over-reading a document it would be absurd to launder one into a quotation.

So the number we published about the cellular radio was a number about a different radio, in the same table, one row below the line.

Why is a spec number so easy to misread?#

Because the thing that scopes it is not text. It is a rule between two rows, and everything that handles the document for you throws it away.

Copy that table out of the PDF and the label column comes away detached: extraction emits LoRaWAN® on a line of its own, sandwiched between Frequency Band and Sensitivity — because a merged label cell is drawn at the vertical centre of its block, four rows below where the block actually begins. Nothing in the output says which rows it governs. Search the whole document for "antenna" and you get three occurrences on two lines — that row, and a note about shipping dimensions. Ask anything to summarise the page and you get a clean list of specifications with the row groups flattened out of it. Every one of those paths produces "one antenna socket" from a document that does not say it.

And the page that would have corrected you is not text at all. We ran pdftotext over the structure-diagram page: it returns zero characters. The reason is blunter than a rendering subtlety — that page is a single flattened photograph of a drawing, one 2480×3509 JPEG at 300 dpi, and the page embeds no fonts whatsoever. The words 5G Antenna and LoRa Antenna are pixels. They cannot be searched, copied, indexed, or read by anything working from extracted text. The most decisive statement the document makes about this question is invisible to every tool you would reach for to check it, and obvious to a person who opens the page and looks.

Three things would have caught it, and none of them is expensive:

• Look at the rendered page, not the extracted text. The rule is visible. It takes one glance.
• Cross-check against a second representation in the same document. The structure diagram and the parameter table are two independent statements about the same hardware; here they disagree with the misreading and agree with each other.
• Ask what the neighbouring rows are about. Demodulation | 1MHz sitting under your cellular antenna row is the document telling you that you have crossed a boundary.

The first error was a premise nobody re-read. The second was different and, for a technical team, more instructive: we did go back to the source, we did cite it, and we got a wrong answer from the right document — because we treated a value as though it carried its own scope.

What survives the correction, and what does not#

The last row is the honest end state, and it is worth being blunt about: we still do not know what the cellular side of this camera exposes. What we know is that the number we had was not a fact about it. The way to get the real one is to look at the unit on the mount, not at any document — which is the same advice the first half of this article arrived at by a different route.

Three purchases that can still do nothing#

Two of these survive the correction untouched, because they never depended on the port count. They are standard radio behaviour rather than anything we measured — we never put an antenna on a site.

The booster row is the one worth pausing on, because it is the only purchase here that can leave you worse off with a better-looking dashboard. A broadband amplifier does not know which parts of the band you want. It lifts your signal and the noise beside it together, and a device reporting a bar count will read the result as an improvement. If your only instrument is that bar count, you cannot tell the difference between a booster that helped and one that hurt.

The coax row is the classic install mistake, and the standard fix — put the radio next to the antenna and run power out to the camera — is exactly the fix a camera-is-the-modem site cannot use. The choice becomes mounting the camera where the signal is, or accepting the loss.