GMSL camera power budget: verify capture card input and idle current

A four-camera GMSL rig that lists no cameras and stays cold is a power problem until proven otherwise: the coax power stage never energized, so the cameras cannot answer. Two measurements at the capture card input settle it in minutes. The supply set-point has to sit inside the card's printed input voltage range, and the input current with cameras attached has to land in the hundreds-of-milliamps class, not fractions of a milliamp. This guide gives the thresholds, the measurement method that makes the reading trustworthy, and the sizing math for the supply behind the card.

Two numbers decide whether the rig is powered

Work them in order. Voltage first: a set-point above the card's printed maximum can hold the input protection closed or damage the input stage, and a card sitting in protection shows exactly the symptom you are chasing, near-zero current and no cameras. Current second: with cameras attached and idle, a healthy four-camera coax rig draws roughly 0.09 to 1.0 A at the card input, per the arithmetic in the worked example below. Cameras that stay cold confirm the reading, because heat follows current and 0.35 mA per camera warms nothing.

That symptom pair rules out most software suspects on its own. For the link-side and cabling checks that run alongside this, see GMSL camera not detected: check capture-card power and cabling.

Check the card's printed input voltage range first

Input ranges differ card to card, and not by a little. One quad GMSL capture card tops out at 19 V; a documented eight-camera embedded GMSL platform accepts +9 V to +24 V external input and sources 12 V over coax to its cameras. Same connector class, different limits: the printed range of the card in front of you is the only one that applies.

In the worked case the supply was set to 24 V against a 19 V maximum, which is 5 V over, or 26% above the limit. Two things can follow, and both look like the same fault. The input protection can hold the rail down, which reads as almost no current. Or the input stage takes the overvoltage and degrades, which turns a reversible set-point mistake into a card replacement.

The decision: bring the rig up at a set-point printed inside the card's range. Where the platform documents a 12 V coax rail, 12 V is the natural choice. And run the host-enumeration check before the power check, because slot, lane and chipset faults look identical from the camera side; PC requirements for a frame grabber: PCIe slots, lanes and chipset covers that layer.

What a coax-powered camera should draw

Power over coax sends video downstream, control upstream and supply current over the same cable, which is why a PoC rig needs no camera power wiring at all. The capability is generational and pair-specific: early GMSL did not inherently carry power, so a given camera-and-card pair either implements it or does not. A pair built without coax power needs its own rail, and no card measurement will change that.

Two anchors set the budget, from PoC design guidance and published serializer figures for common GMSL parts:

A GMSL2 serializer draws about 260 mW in 6 Gbps mode, which on a 12 V coax rail is about 22 mA before the sensor and camera logic add their share. PoC design guidance allows up to 250 mA through one camera's network. Multiply the anchors by camera count and you get the healthy band for the rig:

The band is derived for a four-camera rig like the worked example; scale it by your camera count. A rig that powers up but browns out under streaming load is a different failure, and PoE camera illumination droop during exposure shows the same physics on the Ethernet side.

Worked example: 24 V supply, 19 V-max card, four cameras reading 1.4 mA

The case: an existing four-camera rig on an embedded ARM host that had worked before, now listing no cameras and generating no heat. One measurement session settles whether power flows, and here is the whole computation.

Inputs

• Card printed input maximum: 19 V DC, from the quad capture card's own documentation.
• Supply set-point: 24 V DC, measured at the card input.
• Cameras: four, powered over coax.
• Measured input current, cameras connected and idle: 1.4 mA total.
• Expected per-camera coax current: about 22 mA floor (serializer alone, 260 mW at a 12 V rail) up to 250 mA ceiling (PoC design guidance).

Rule in words: the set-point must sit inside the printed range, and the attached idle current must land between camera count times the floor and camera count times the ceiling.

Arithmetic

• Overvoltage: 24 V - 19 V = 5 V over the maximum; 5 / 19 = 0.26, so 26% above the printed limit.
• Per-camera measured current: 1.4 mA / 4 cameras = 0.35 mA per camera path.
• Healthy band: 4 x 22 mA = 88 mA, about 0.09 A at the floor; 4 x 250 mA = 1,000 mA = 1.0 A at the ceiling.
• Shortfall: 88 mA / 1.4 mA = about 63 times below the floor before the sensors add their share, and about 714 times below the ceiling.

Result: the supply sat 26% out of range, and the current says the cameras drew nothing, because 0.35 mA per camera is leakage, not a camera.

Decision, with thresholds: correct the voltage first. Set the supply inside the printed range, power up on a current limit of about 1.5 A (1.5 times the 1.0 A ceiling) and re-measure. A reading that steps into the 0.09-1.0 A band means the rig is powered and the fault was the set-point. A reading that stays at the 1.4 mA level with the voltage in range moves the fault to the coax power stage, its enable or the cable path; power down and work the path as in the section below.

Measure the idle current so the number means something

A wall adapter hides its current; a bench supply or a multimeter in series at the card input shows it. Set the current limit to about 1.5 times the budget before the first power-up, so a fault folds the supply back instead of feeding it. Take three readings: cameras detached, which shows card overhead only; cameras attached and idle; cameras attached and streaming. The step from detached to attached is the cameras' real share, and a camera that adds nothing to the step owns the fault with its port or cable.

The 0.35 mA per camera from the worked example deserves its own note: a camera cannot draw even standby current through a rail that never energized. Near-zero current with cameras attached is not a quirk of idle cameras; it is the rail reporting that it delivered nothing.

The power-first habit transfers across buses. The USB version lives in camera has power but not detected: USB link checks after a cable swap, per-camera budgets for many cameras on one host in multiple USB cameras on one host: bandwidth and power budget, and the I/O-side wiring faults in camera GPIO wiring: prevent damage and triage power faults.

Size the input supply from the power budget

The supply behind the card follows from the same anchors:

For the worked rig, 4 x 0.25 A x 12 V = 12 W of camera budget at the ceiling. Add the card's own draw, which is printed in the same electrical table as its input range, then multiply by 1.5 for margin: 18 W minimum before the card's share. A supply sized like that rides through startup inrush and streaming transients without folding back.

Cabinet layouts often distribute at a higher rail and regulate at each station; lower current on a long run means less drop in the wire, and the branch math stays the same. The sizing discipline is the same one behind RGB, UV and IR test bench: controller and power sizing for a lighting bench.

For a cabinet front-end, an enclosed industrial unit rated 48 V, 76.8 W covers the station class described here; the LRS-75-48 is that unit. The power supplies category lists the enclosed 12, 24 and 48 V options.

At a 48 V rail, the 12 W camera branch draws 12 W / 48 V = 0.25 A, so the run to the station carries a quarter of an amp and drop stays small; the remaining capacity carries the card and the rest of the cabinet. Regulate down to the card's printed input range at the station.

When voltage is in range and current still reads zero

Then the fault sits in the delivery path, and the measurement localizes it. Reseat both ends of every coax run and re-measure. Then move cameras between ports one at a time: each moved camera should change the reading by its own share, tens to hundreds of milliamps. A camera that adds nothing on any port, or a port that never adds current for any camera, splits the fault between that cable and that port's power stage. A known-good camera on a known-good cable that still reads near zero leaves the card's coax power stage as the suspect, and the fix is a card, not software. Damaged power and I/O wiring elsewhere in the rig produces the same silence; the cables category covers power and I/O leads and frame-grabber I/O cables for the rest of the rig.

If the rebuild goes past one card, standard interfaces keep power in the link too. A dual-port PoE NIC such as the MV-GE2002P powers two cameras over their Ethernet runs straight from the host, with no separate camera rail to wire or fuse.

For eight cameras on a single PCIe host, an octo CXP-6 frame grabber such as the MV-GX1008 concentrates the whole rig, and many CoaXPress camera-and-card pairs also deliver camera power over the coax, on limits the pair sets.

Compare the frame grabbers category for the interfaces the shop's cards cover.

Bring-up checklist

• Read the card's printed input range and set the supply inside it: the set-point is the first suspect in every dead-rig case.
• Set the current limit to about 1.5 times the budget before the first power-up: a fault then folds back instead of feeding.
• Record three currents: cameras detached, attached and idle, attached and streaming: the detached-to-attached step is the cameras' real share.
• For a four-camera rig, expect about 0.09-1.0 A attached and idle: milliamp readings mean the coax rail never energized.
• Correct an out-of-range set-point before any other change: 26% over was enough to make a working rig look dead.
• Swap cameras across ports one at a time and watch the step: the port or cable that never adds current owns the fault.
• Size the supply at 1.5 times cameras plus card: 18 W minimum for the worked rig before the card's share.

To spec the capture side end to end, start at the accessories finder, step through the build configurator at the accessories stage, or send a multi-camera bill of materials through the quote page.