Color Sensor Interface Circuit Design

A color sensor measures the intensity of light in a few wavelength bands and reports the result as numbers. Turning those numbers into a colour that matches what a person sees requires an understanding of the light source, the filtering and the way the readings are scaled.

What the Sensor Measures

The sensing element is a photodiode array with a set of filters over the individual diodes: typically red, green, blue and clear, or in some devices a set of interference filters that approximate the response of the human eye.

Each channel produces a current proportional to the light in its band. The currents are integrated onto a capacitor and converted to digital values, so the output is proportional to the product of the light intensity and the integration time.

A clear channel without a filter is usually included because it gives a measure of the total light. It is used to compensate for the illumination level when the colour of the light itself is not what matters.

IR filter and Spectral Response

An IR filter is essential in most applications. Silicon responds far into the infrared, and without a filter a red reading is contaminated by infrared that the eye cannot see, which makes a warm source read wrongly.

The filter is usually a thin layer deposited on the package, and it is not perfect. Its transmission in the near infrared is a few percent rather than zero, and that residual is enough to shift a measurement under an incandescent lamp while barely mattering under a white LED.

A better approach for critical work is to modulate the source and measure only at that frequency, or to use a sensor whose filters are interference types rather than absorptive. Both cost more and are chosen from the accuracy required.

Integration Time and Gain

integration time is the period over which the photocurrent is accumulated. A long integration time collects more light and gives a better signal to noise ratio, at the cost of a slower measurement and a greater sensitivity to movement and flicker.

The automatic gain control in most sensors selects both the integration time and the internal gain so that the largest channel uses a good fraction of the full scale range. The choice has to avoid saturating any channel, because a saturated channel cannot be corrected afterwards.

Flicker from mains powered lamps is the difficulty with long integration times. A time that is an exact multiple of the mains period averages the flicker to zero, and one that is not produces a reading that varies from sample to sample.

White Balance and Scaling

white balance is the correction applied so that a white object reads as white under a given illuminant. Without it, an LED lit scene reads blue and a tungsten lit scene reads orange, because the sensor reports the light rather than the object.

The correction is a set of three multipliers, one per channel, derived from a measurement of a white reference under the expected light. The multipliers are applied in software, and they change when the illuminant changes, which is why a good implementation estimates the illuminant as well.

Where the illuminant is known and fixed, the multipliers are constants and the design is simple. Where the light varies, the sensor needs either a measurement of the source, a known reference patch in the field of view, or an algorithm that estimates the illuminant from the scene itself.

Calibration and Accuracy

Calibrate the sensor channel by channel against a reference instrument. A calibrated lamp and a white tile give the raw counts for each channel, and the multipliers follow from the ratio between the channels.

The temperature of the sensor matters, because the dark current of the photodiodes and the transmission of the filter both change with it. A temperature reading next to the sensor allows the dark level and the gain to be corrected.

The accuracy of the whole system is limited by the match between the sensor response and the response of the eye, which is a matter of filter design rather than of the circuit. Good enough for a display backlight adjustment does not mean good enough for a colour measurement instrument.

Layout, Aperture and Stray Light

Place the sensor under an aperture that defines the field of view, and make the aperture opaque and matt black on the inside. Light that reaches the sensor after reflecting from the enclosure adds to the reading and destroys the calibration.

Keep the sensor away from the edge of the board where the enclosure wall meets it, because the joint is usually the place where light leaks in. A gasket or a wall around the aperture is standard practice.

The supply and ground of the device should be quiet, and the digital lines should not run across the aperture. The photocurrents are small, and a coupling path from the interface lines into the analog front end appears as noise in the least significant bits. The general approach is described in our guide to mixed signal board design.

Interfacing and Data Handling

Most sensors present their results over a two wire serial bus, with registers for the channel data and for the configuration. The configuration register also controls the gain and the integration time, and reading it back after a write confirms the device is responding.

The raw counts need to be converted to a colour space the application can use. A conversion matrix maps the four channel readings to the tristimulus values, and the matrix is derived from the filter responses, or from a characterisation of the actual part.

The clear channel provides a useful cross check. A large discrepancy between the sum of the filtered channels and the clear channel indicates a saturated or a faulty channel, and the reading can be rejected rather than used.

Verification and Faults

Verify the sensor with a known white reference under the intended light source and record the raw counts. Comparing those counts with the datasheet expectations at the same illumination level confirms the whole optical path including the aperture and the overlay.

A reading that is unstable points to flicker from the light source. Set the integration time to a multiple of the mains period and repeat the measurement with the source on direct current if possible.

A reading that is stable but wrong in one channel suggests a blocked or a damaged filter, or a foreign object over part of the aperture. The release checks that keep these measurements consistent from unit to unit are collected in our PCB design release checklist, and the assembly points we inspect are listed in judging PCB quality.

FAQ

Why does my colour reading change with the light source? The sensor measures the light as well as the object. Apply white balance multipliers for the illuminant, or provide a known reference patch.

Do I need an IR filter? For any measurement with a warm light source, yes. Silicon responds to infrared that the eye cannot see and the colour reading shifts without one.

What integration time should I choose? As long as the measurement rate allows, and an exact multiple of the mains period to average out lamp flicker.