Power and Noise

Generic engineering material, applicable to any module from any supplier.

A separate team-process reference can be found this resource.

An image sensor is an analogue device measuring small charges, sitting on a board full of switching digital electronics — which makes its supply an imaging specification rather than a power one.

Where supply noise appears in the image

Horizontal banding, where the disturbance is periodic and interacts with the line readout.

A general noise rise, where it is broadband.

For standards, components, or implementation context beyond this page, consult Nature.

Column or row patterns, where a specific internal rail is affected.

And banding that changes with machine state — a motor starting, a heater switching, a radio transmitting — which is the signature that distinguishes a supply problem from a thermal one.

The rails that matter

A module typically needs more than one supply: an analogue rail for the sensor, a digital core rail, and an interface rail.

They have different requirements. The analogue rail is the sensitive one, and noise on it appears in the image directly.

Which is why sharing a rail with a switching load is the common mistake — a supply adequate for logic is not automatically adequate for a sensor.

Read the module's requirements per rail rather than treating the supply as one number, and note the ripple specification separately from the voltage tolerance.

What helps

Separate regulation for the analogue rail, or at minimum filtering between it and everything else.

Decoupling at the module, close to the pins, of the values the supplier specifies — and where the module is on a flex, the decoupling belongs at the module end rather than at the board end.

A low-noise regulator where a switching one would otherwise inject at a frequency the sensor sees.

And a return path that is not shared with high-current switching, since a common return turns a current into a voltage the sensor measures.

Voltage at the module, not at the board

A longer cable has resistance, and the module sees less than the regulator produces.

Which on a marginal supply produces intermittent behaviour that reads as a data problem — frames dropping, initialisation failing, the sensor behaving differently when the illuminator fires.

Measure at the module during bring-up, under load, with everything else in the system running.

And check it at the extremes: cold start, full frame rate, and whatever else in the product draws current simultaneously.

Sequencing

Multiple rails frequently have a required order for power-up and power-down.

Violating it can leave the sensor in an undefined state, which presents as a module that sometimes fails to initialise and works after a power cycle.

It is in the datasheet and it is the item most often skipped when a supply is designed by somebody who did not read it.

Check it against your actual regulator behaviour, including what happens during a brownout — where rails collapse in whatever order their capacitance dictates rather than in the order intended.

Isolating a supply problem

Three tests, in order of speed.

Run from a bench supply rather than the product's own. If the banding disappears, it is the supply and not the module.

Then run the product with its other loads disabled — the motors, the illuminator, the heater — adding them back one at a time.

And photograph a uniform surface with the noise measured rather than judged, so that the effect of each change is a number.

An afternoon, and it identifies the offending load, which is the information every subsequent decision needs.

Current during capture

A sensor draws more while reading out than while idle, and the transition is fast.

Which means an average current figure understates the demand and a supply sized on it sags at exactly the moment the image is being captured.

Ask for the peak figure and the duration, and where a module strobes an illuminator, the two peaks may coincide — which is a design detail worth checking rather than discovering.

Grounding across a flex

A shared ground between board and module runs through the cable, and it carries the return for everything the module does.

A thin or long ground path turns module currents into voltage differences, which the analogue circuitry sees.

Which is why cable pin assignments allocate several conductors to ground and why using a cable with fewer is not a saving.

Where a shield exists, it is not the ground, and treating it as one is a common and consequential error.

The measurement people cannot make

Noise on a rail is small and fast, and an ordinary meter reads an average that tells you nothing.

An oscilloscope with a short ground connection — not the long lead, which forms a loop and reads its own artefacts — is what shows ripple and switching spikes.

Where no scope is available, the image itself is the instrument: photograph a uniform surface and measure the noise, with and without the suspect load running.

Which is slower and answers the question that matters, since the image is what you are protecting.

Where the illuminator shares the supply

A common arrangement and a common cause.

An illuminator drawing a large current, switched at the moment of capture, pulls the supply down exactly when the sensor is most sensitive to it.

Which produces an image that is worse with the light on — the opposite of what anybody expects, and it gets attributed to the illumination geometry rather than to the supply.

Separate the rails, or feed the illuminator from a reservoir capacitor sized for the pulse so that the draw does not reach the shared supply.

And test with the illuminator running, since a bench measurement without it misses the entire effect.

In one line

Measure at the module under load, because every number that matters here differs from the one at the regulator.

The short version

• An image sensor is an analogue device on a digital board, which makes its supply an imaging specification
• Supply noise appears as horizontal banding, a general noise rise, column patterns, or banding correlated with machine state
• Modules need several rails with different requirements, and the analogue one is where noise reaches the image directly
• Separate or filter the analogue rail, decouple at the module end of a flex, and do not share a return with switching current
• Measure the voltage at the module under load, not at the regulator, since cable resistance produces intermittent faults
• Power-up sequencing is specified and frequently skipped, which presents as a module that sometimes fails to initialise