A stereo depth module and its processor board are sold separately, and the first question at the bench is whether joining them simply works: does the viewer show a depth map with no calibration step? It does, when the two are the matched pair the depth module datasheet names. This guide is for engineers building depth sensing into a machine rather than buying a cased camera. It explains what the factory calibration carries, how to match the processor revision, when a flex cable beats a rigid interposer, and exactly what a lens swap costs in depth error, with the arithmetic shown.
One calibrated pair, not two generic boards
A depth module is the sensor side of a stereo depth camera: two synchronized cameras with the baseline fixed at manufacture and, on many modules, an inertial measurement unit. The processor board is the compute side: it takes both images, runs the stereo matching and emits the depth stream.
The module carries the calibration taken on the production line: the intrinsics of each camera (focal length, principal point, distortion), the extrinsics between the two cameras, and the IMU-to-camera alignment where an IMU is present. The processor reads that calibration at start-up and applies it to every frame. A matched pair therefore behaves like the equivalent cased camera with the housing left off: depth appears in the viewer on first power-up, with no calibration step of your own.
That single fact drives the purchase. Order the module and its processor as one listed pair, not as two independent finds - the pairing is a specification, not an assembly accident.
Match the processor revision first
Processor boards ship in hardware revisions, and each revision supports a defined subset of modules. Short-range modules, IMU-equipped modules and long-range families typically sit behind different revisions, because each adds the support its module group needs. The mapping is printed as a pairing table in the depth module datasheet, and it is the first table to read - before resolution and range.
A revision mismatch fails early and loudly. The assembled rig enumerates no depth device at all, or shows the colour stream with no depth beside it. That signature appears before any image-quality question, so read the pairing table before suspecting optics, cables or drivers. When depth is listed but drops out intermittently, the fault usually sits further down the chain; the walkthrough of depth-only failures on a stereo depth camera covers that path.
The decision is short: find your module in the pairing table and order the revision printed next to it.
Flex cable or rigid interposer: a mechanical choice
The interposer is the small adapter that sits between the module's board-to-board connector and the processor's, passing the signals straight through. Flex cable and rigid board are electrically interchangeable: both are pass-throughs for the same connector family, which is shared across the module range. Choose between them on mechanics alone.
| Flex interposer cable | Rigid interposer board | |
|---|---|---|
| Stack geometry | Any offset or angle within the cable's reach | Fixed height and orientation, as designed |
| Vibration | Bend must be routed and anchored | Stiffest option; joints stay load-free |
| Enclosure fit | Folds into tight or awkward housings | Needs a straight, planned stack |
| Re-work | Suits mounts you reposition | Suits mounts set once |
Flex cable is made for the bend: its coverlay stays flexible after curing, unlike the solder resist on a rigid board. Treat it as a static route all the same. Bend it once at assembly, keep the radius at roughly ten times the flex thickness per the bend rules in IPC-2223, and anchor it so machine vibration never re-works the bend or strains the connectors.
The rigid board wins when the module-to-processor stack is fixed and the housing is known: a short, stiff connection that cannot drift or flutter.
Decision: fixed stack in a known housing, take the rigid board; offset, adjustable or cramped geometry, take the flex cable with the bend anchored.
What the factory calibration carries, and when it is yours again
The inherited calibration answers the questions your first calibration would: focal length and distortion per camera, the exact pose between the two cameras, and the IMU's orientation relative to the stereo pair. The processor applies these to every frame, which is why depth is trustworthy immediately.
Three additions reopen the question:
- Glass in the path. A protective window in front of the lenses shifts apparent depth by about one third of its thickness (glass, refractive index about 1.5). Keep it normal to the optical axis and subtract the offset, or recalibrate through it.
- Extra sensors. Putting the IMU to work, or adding a second module, brings inter-sensor poses into play. The trade-off between factory defaults and your own measurements is worked through in camera-IMU extrinsics: factory defaults versus your own calibration, and multi-module rigs need their inter-camera poses measured as in multi-camera point cloud fusion with depth camera extrinsics.
- Resolution changes. Binning or cropping rescales the intrinsics, and depth reconstructed with unscaled values wobbles; camera intrinsics after binning: scale them correctly shows the fix.
Swapping the module's lenses: what breaks, by how much
Stereo depth rests on one equation, with Z the distance, f the focal length in pixels, B the baseline and d the measured disparity:
Change the lens and every line shifts, because the processor keeps applying the factory focal length. Worked example, with illustrative inputs: baseline B = 25 mm, pixel size 3 μm, factory lens f = 2.0 mm, and a part standing at Z = 200 mm.
- Factory focal length in pixels: 2.0 mm / 0.003 mm = 667 px. Disparity: d = 667 × 25 / 200 = 83 px.
- Swap in a 2.5 mm lens: f = 2.5 / 0.003 = 833 px, so the true disparity is 833 × 25 / 200 = 104 px.
- The processor still divides by 667: reported Z = 667 × 25 / 104 = 160 mm. The 200 mm part reads 160 mm - a 40 mm error, exactly the lens ratio 2.0 / 2.5.
- Depth resolution per disparity pixel improves from ΔZ = 200² / (667 × 25) = 2.4 mm to 200² / (833 × 25) = 1.9 mm.
- The nearest measurable distance rises: with a 120-pixel disparity window, Z_min = 667 × 25 / 120 = 139 mm before, 833 × 25 / 120 = 174 mm after.
- The field of view narrows by the same 20 % on the same sensor.
Recalibration removes the scaling error: re-measure intrinsics and extrinsics, then run the disparity-to-depth math with your own values, in your own matcher or wherever the interface accepts supplied calibration. Where the interface exposes only the processor's depth output, no setting rescues a swapped lens - pick a module whose optics already match the range. Once the optics are settled, the 3D camera working distance and positioning guide sets the mounting numbers.
One honest caveat: the factory calibration also anchors the depth scale. After recalibrating, accuracy rests on your session quality - at the example geometry, 0.1 px of disparity precision is about ±0.24 mm at 200 mm (ΔZ = 200² × 0.1 / 16,675).
Pair the parts or buy the calibrated camera
Four builds cover nearly every request, each with its own decision:
- Depth inside your own housing, standard optics. Buy the module and its matched processor revision as one pair, with the interposer your mount dictates. No calibration work, depth on first power-up.
- Different range or field of view. Change the module, not the lenses: choose a module whose optics fit the task and keep its matched revision. The pairing table governs, not the shared connector.
- Depth is the goal, not the build. Skip pairing entirely and buy a complete 3D camera - the 3D sensors category lists factory-integrated profilers and stereo units, filtered further in the camera finder. The MV-DPS200P-02, for instance, is a complete projected structured-light 3D camera: depth from one unit, nothing to pair on the bench.
- Type
- Stereo / structured light
- You will run the stereo matching yourself. Then the module half is a binocular board module and the processor is your host. The board-level cameras carry the synchronized-sensor boards; the LS-BCM-0043 pairs two 1/2" global-shutter sensors at 1280 × 1024 over USB3, and the checklist in board level camera selection for embedded vision machines covers the rest. For frame rate, read the module datasheet against your matching budget before you commit. You own the calibration from day one.
- Sensor format
- 1/2"
- Lens mount
- M12
- Interface
- USB3
- Resolution
- Up to 3 MP
Whoever hosts the rig, the processor-to-host link needs the right cable. Keep passive USB3 runs to about 3 m and move to GigE beyond that; on a moving axis, specify drag-chain-rated cable. The LS-CBL-0219 is the 3 m high-flex USB3.0 camera cable, the LS-CBL-0199 the 3 m high-flex GigE cable with screw locks, and the LS-CBL-0300 carries trigger and power on an M8 8-pin lead rated for drag-chain use. Both high-flex families sit in the camera cables list, filtered by length and type in the accessories finder.
- Type
- Power and I/O cables
- Length
- 6-8 m
Checks before you order
- Pairing table entry. Module model to processor revision, from the datasheet; the mismatch is the most common dead-on-arrival cause and it fails before any image appears.
- IMU support. A module with an IMU needs the revision that reads it; the alignment itself is factory-set.
- Interposer form. Rigid for a fixed stack, flex for offsets and tight housings; bend radius of ten times the flex thickness, anchored at assembly.
- Glass. Any window normal to the optical axis, with about one third of its thickness expected as a depth offset.
- Host side. Interface and bandwidth support on the receiving end; the checks in will your camera work with a frame grabber? Compatibility checks apply to any capture host.
- Calibration budget. If optics, crops or added sensors are in scope, plan the calibration session with the project, not after it.
The decision, per build
- Matched pair, standard optics, your housing: assemble with the matched revision and the interposer your geometry wants. Depth on first power-up; no calibration session.
- Optics wrong for the task: change the module, keep the revision matched; a lens swap buys a rebuild and a recalibration to reach what a different module offers out of the box.
- Depth without the build: a complete 3D camera such as the MV-DPS200P-02 removes the pairing question; start the selection in the camera configurator.
- Your own matcher: a binocular board module and your host, with calibration on your terms.
For the wider depth-rig toolbox - working distance, calibration and point-cloud checks - the 3D topic guides collect the rest. If a pairing is unclear, send the datasheet's pairing table through the contact page and we will confirm revision, interposer and cable before you order.
Questions
Do I need to calibrate a depth module after fitting it to its processor board?
No, when the pair matches. The module stores the calibration taken on the production line - per-camera intrinsics, stereo extrinsics and, where fitted, the IMU alignment - and the processor reads it at start-up, so the assembled rig streams depth straight away. You only calibrate yourself after a change the factory could not have seen: a lens swap, added glass, a crop or binning change, or a second module. A quick sanity check is a flat plate at a known distance.
Is a flex interposer cable interchangeable with the rigid interposer board?
Electrically yes. Both pass the same board-to-board signals between the module connector and the processor connector, and both mate the connector family shared by the depth modules. Choose by mechanics: the rigid board holds a fixed stack and shrugs off vibration; the flex cable routes around offsets and tight enclosures and suits adjustable mounts. Bend the flex once at assembly, radius at least ten times its thickness, and anchor it. What is never interchangeable is the processor revision: that follows the module, not the interposer.
What happens if the processor board revision does not support my depth module?
Usually nothing subtle: the rig fails at enumeration. The viewer lists no depth device, or lists the colour stream only, before any image-quality question arises. That is the signature of a revision mismatch rather than dirty optics. The fix is the revision number the module datasheet names - processors ship in revisions precisely because short-range, IMU-equipped and long-range modules need different support. Check the table before buying the board, and before blaming cables, lenses or drivers afterwards.
Is there a general rule for the depth error after swapping a stereo module's lens?
Yes: until recalibration, the reported distance scales with the lens ratio. A part at true distance Z reads Z x (f_old / f_new), so a 2.0 mm to 2.5 mm swap reads every distance at 80 percent of the true value - a 200 mm part at 160 mm. Swap the other way and the error flips sign, reading long instead of short. Depth resolution and nearest measurable distance move by the same ratio, which is why lens choice and disparity window belong to one decision.




