When to choose a CoaXPress camera: bandwidth, cable length and sync

CoaXPress asks for a PCIe frame grabber, coax cables and cameras that speak the standard, so it earns its place only when your numbers say so. The deciding numbers are on this page: what each CXP link carries, how to turn pixel count and frame rate into megabytes per second, how far coax reaches at each speed, and when a frame grabber replaces external trigger wiring. Anyone sizing a high-speed or multi-camera inspection station can work the example with their own sensor and get a one-pass decision between GigE, USB3, 10GigE and CoaXPress; the general interface rankings sit in most reliable camera interface: GigE, USB3, Camera Link or CoaXPress.

How much bandwidth does a CoaXPress link carry?

One coax link at CXP-6 carries 6.25 Gbps on the wire — about 625 MB/s of image payload after 8b/10b encoding — and a four-link camera reaches 25 Gbps; CXP-12, the CoaXPress 2.0 level, doubles each link to 12.5 Gbps, or 50 Gbps over four links.

The standard defines six levels, CXP-1 through CXP-6, from 1.25 to 6.25 Gbps per cable, and CoaXPress 2.0 added the 12.5 Gbps level the market labels CXP-12, as the CoaXPress reference tabulates. Cameras aggregate links: two connections at CXP-6 move about 1.25 GB/s of payload, four move about 2.5 GB/s. Each link is point-to-point with its own payload, so cameras on one card never contend for bandwidth the way devices behind an Ethernet switch do.

How do you work out the bandwidth your camera needs?

Multiply width by height by bytes per pixel by frame rate, then add about 15 percent for protocol overhead; if the result sits under about 118 MB/s GigE fits, under about 380 MB/s USB3 fits, and under about 1,150 MB/s a single 10GigE link fits.

Resolution and frame rate trade against each other, and bandwidth is their product — the camera resolution vs frame rate tradeoff page covers the sizing side. Bits per pixel matter as much as pixels: 8-bit mono or Bayer costs 1 byte, 10-bit about 1.25, and host pipelines that unpack to 16-bit words double it.

The payload figures are practical estimates after protocol overhead; the reach figures are the standard's, and they assume proper coax grade — the next section covers that.

How long can the coax cable run be?

At the top speeds the standard reaches 40 m — that is CXP-6 and CXP-12 — while the slower levels CXP-1 through CXP-5 stretch well past that, and transmission adapters carry any level beyond coax's own limit when a mid-run repeater point is acceptable.

Real reach depends on cable quality: attenuation and shielding decide how much of the documented length you get at 6.25 or 12.5 Gbps, so match the coax grade to the link speed. The comparison that matters against other interfaces: Camera Link tops out around 7–10 m of copper, USB3 at a few metres, GigE runs to the full Ethernet length in the table above but only at its ~118 MB/s. Coax also shrugs off the motor and drive noise on a factory floor, and one cable carries data, camera control and power. Order grade and length to match your level from the camera data cables range.

When does frame-grabber synchronization beat external triggers?

A frame grabber clocks every port from one oscillator, so the cameras on one card expose together with no external trigger wiring at all; the skew that remains between cameras is cable physics, about 5 ns per metre of coax.

Run the numbers: a 10 m length difference between two camera cables costs 50 ns, and against a 33.3 ms frame period at 30 fps that is a 0.00015% offset — far below anything a fast-moving part shows between frames. An external trigger tree, by contrast, adds a distribution stage, per-camera cable delays and jitter with every camera you hang on it; the card's per-port trigger logic replaces the whole tree. Sync only pays when the measurement needs it, and there are cases where hardware sync calibrated worse than software sync — the trigger fix page walks through that failure. The card's outputs can drive external devices too: the frame grabber trigger output page shows the wiring.

Is 10GigE good enough instead?

For a single camera whose payload stays under about 1.1 GB/s, 10GigE reaches the same Cat6a length as GigE with a plain network card and no frame grabber, so it wins on simplicity; CoaXPress wins on guaranteed per-link bandwidth, camera density per PCIe slot and hardware sync.

The card-side difference shows in two parts:

All 2 share: Bus status indicators 1; Humidity 10% RH to 95% RH (no condensation); Client software MVS. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

An octo-port CXP-6 card carries eight links in one PCIe slot — eight single-link cameras or two four-link cameras — about 5 GB/s on one card. Replicating that on Ethernet means four dual-port 10GigE cards plus switch ports, with the packet-traffic trade-offs the machine vision camera interface comparison lays out; GigE's own ceiling is worked through in how much frame rate fits on GigE Vision.

Which CoaXPress parts make up the chain?

The chain is three pieces — a camera with one or more CXP links, a frame grabber with that many free ports at a matching speed class, and coax rated for the level and run length — plus a free PCIe slot and GenTL-capable software.

A concrete pairing: the MV-CH250-90YC-M58S-NN is a 25 MP CXP-12 color camera with a global shutter, a 1.1" sensor and an M58 mount.

On the card side, the MV-GX1008 is an octo-port CXP-6 frame grabber for CoaXPress cameras.

Order the speed class deliberately: the link runs at the level both of its ends support, so a CXP-12 camera on a CXP-6 card works at 6.25 Gbps per link and gives up half the per-link rate. How images then flow through the card into host memory is covered in frame grabber image acquisition for Camera Link and CoaXPress, and the frame grabbers range lets you size ports per camera.

When should you stay on GigE or USB3?

Stay on GigE or USB3 when one camera's stream fits the simple budgets: a 2 MP camera at 40 fps moves about 83 MB/s — 1920 × 1080 pixels × 1 byte × 40 — which GigE carries over a standard-length run with PoE and no frame grabber.

• Up to about 110 MB/s, one camera: GigE Vision. PoE and the full Ethernet length with no card; tune the link for packet loss, the failure mode behind horizontal lines in live images.
• Up to about 380 MB/s, one camera, a few metres: USB3 Vision. No card, but real throughput depends on host tuning — see the USB3 low frame rate and latency fixes.
• About 0.4–1.1 GB/s, one camera, at Ethernet lengths: 10GigE with a network card, no frame grabber.
• Above that, or two or more cameras exposing together, or runs longer than coax allows without repeaters: CoaXPress with a frame grabber.

Most single-camera inspection up to 4 MP at conventional frame rates lands in the first two rows; CXP starts where those rows end.

What do you check before you order a CoaXPress chain?

Four data-sheet checks decide the order: the camera's highest CXP level and link count, the frame grabber's port count and per-port level, the coax grade and length for your run, and the host side — a free PCIe slot wide enough for the aggregate payload.

• Camera: highest CXP level and number of links — a four-link camera occupies four card ports.
• Frame grabber: free ports and per-port level, matched to or above the camera class.
• Cables: coax grade and length; grade matters more as you approach 12.5 Gbps per link.
• Host: PCIe 3.0 moves about 985 MB/s per lane, so a x4 slot carries about 3.9 GB/s — check it against your aggregate.

Start from the camera finder filtered by resolution and interface, then step through card and cables in the vision configurator; the systems topic page collects the neighbouring deep-dives.