High-brightness LEDs are being touted as the next big revolution in lighting. They are already being used in applications such as architectural, theater, and automotive lighting and they are revolutionizing other markets such as LCD television backlighting. However, there are several challenges when it comes to lighting design with high-brightness LEDs. The exact color as well as flux can vary from part to part creating color variation from one lighting fixture to the next. As temperature increases, the flux decreases and color drifts, causing inconsistent flux and color output under different operating conditions. Combining these variances with the fact that many fixtures use a mixture of color from three or more LEDs to create the desired color, designers soon realize how challenging it is to create uniform color output from fixture to fixture and across operating conditions.
The addition of a color sensor to provide a feedback loop can alleviate these challenges and ensure accurate and reliable color output. While the decision to add optical feedback is an easy one, actually implementing it unearths additional challenges, including sensor selection, calibration and placement, as well as control mechanisms. However, with the proper engineering, these challenges can be overcome and the results will be an accurate and reliable light fixture.
Why accurate color?
Not all lighting designers will need accurate color. For instance, a stoplight simply needs a red, green and yellow light. If you look closely at LED stoplights you can tell that there is a slight variation from one green light to another. Sometimes there is even a variation within the hundreds of green LEDs in one stoplight. But from a distance, all the driver sees is a green light. The same relaxed requirements can be made for an RGB color-mixing home accent light fixture where users control the levels of red, green and blue to create the color they want. In this case, users adjust the color until they achieve the desired output, and any drift in color or flux due to temperature will most likely be undetectable since there will be no side-by side comparison to another fixture. Ambient temperature will most likely be regulated as well.
In other applications, color accuracy can be very important. Take, for example, a theater set lighting designer who is using several LED flood lights to create an orange-illuminated wall. If one of those LED flood lights is a more reddish orange than the other two, the audience will be able to tell, given the side-by-side comparison. For another example, consider a car manufacturer who is developing an interior lighting scheme for its next luxury model. They have chosen a cool white for the gauges, buttons and LCD navigation display backlight. However, most likely all of these fixtures are going to come from other manufacturers, and if one has a cool white, another neutral white, and yet another a warmer white, the interior will look gaudy. In order to prevent this, the automobile manufacturer will specify stringent color accuracy requirements that the manufacturers must adhere to.
Methods of creating color
The first thing a designer must consider is what color or colors they are going to create. If it's just a simple green color, as in the case of the stoplight, then the design is as simple as selecting a green LED which has a wavelength that is close to the desired one, and a flux that is bright enough and be done with it. If, however, the designer wants to create a very specific color, often a form of color mixing (referring to the mixture of light from two or more different color LEDs) is required. Even if the desired color is red, if it is a very specific red ¨C for instance a corporation wanting to wall wash a display with their specific corporate red ¨C then color mixing should be used instead of simply using red LEDs. The reason for this is simple; due to variances in manufacturing, LED vendors cannot guarantee that it will be the same red from LED to LED, and across temperature. They have taken steps to narrow it down, by binning their LEDs in groups by dominant wavelength. However, these bins are often not exact enough, varying by up to 10 nm in wavelength per LED, which can significantly affect the color being output. When multiple LEDs are used, adjustments can be made to the mixture of the LEDs to compensate for these variations and pinpoint the exact color that is desired.
Since the best approach is to use multiple LEDs to create a specific color, the next question is how many LEDs and which colors. This all depends on what color(s) you want to create. If all you want is a cool white, selecting a neutral white and a blue may do the trick. If the output colors of these two LEDs are placed on the International Commission on Illumination (CIE) color chart, then the line between the two would be the plot of all the possible colors that can be created with these two LEDs; an illustration of this is shown in Figure 1. If that line goes through the desired color, with acceptable tolerance, then these LEDs will do. Remember to take into account the variation of color output from LEDs in the same color bin, and do a worst-case analysis with both LEDs. If it no longer is capable of producing the desired color, it may be necessary to move to a three-LED system. Alternatively, if the design is for a stage lighting fixture which is configurable by the user to create hundreds or even thousands of colors, then at least three LED colors will be required, most commonly consisting of red, green, blue, and sometimes amber LEDs. This creates the largest achievable color space (or gamut). This is shown in Figure 2.
Once the LEDs are selected it becomes an exercise in determining what brightness 
percentage mixture from each LED will create the desired color. For instance an equal brightness of red, green and blue will create white. Chances are, however, that if you turn all three LEDs on to maximum brightness, the color will not be an exact white. This is due to some LEDs being brighter than the others and thus more dominating, as well as the variations in dominant wavelengths. So, use the dominant wavelength and maximum flux stated in the datasheet to create a calculation which says what percentage of each is needed. This calculation will only be as exact as the datasheet values are, and remember, these LEDs are binned in ranges which introduce error. The only way to eliminate this error is to either a) calibrate every system to the LEDs on the board by creating unique algorithm variables or using look-up tables for each board or b) adding a color sensor to measure what color is being output and adjusting for any inaccuracies.
I need a sensor, now what?
Deciding to use a color sensor for accuracy in a system is easy; actually designing it in is much more difficult. The biggest challenge is the sensor placement and mechanical construction of the system. Additionally, selecting a sensor and developing an optical feedback algorithm around it takes a fair amount of effort as well. There are also common pitfalls which should be avoided to keep the headaches to a minimum.
When looking at sensor placement there are a few things to keep in mind. The first is that color sensors cannot handle direct light from the LEDs. The light output from high-brightness LEDs is very bright, and at the highest brightness levels, the sensor will saturate, creating inaccurate measurements. The inaccuracy depends on sensor, but at the time of this article, I have yet to find a color sensor in the sub $10 range which can handle direct light from the LEDs. So, since pointing the sensor at the LEDs is not an option, the best thing is to have the sensor receive reflected light, which is at a much lower intensity. Additionally, the goal should be for the sensor to view an accurate representation of the color which is viewed by the observer of the light, meaning the sensor in a theater lighting fixture should see the light just as the audience does, but at a lesser intensity. The color the sensor sees should be fully diffused, with no individual LED artifacts, and should not have its color altered by reflecting it off of a colored surface.
In the case of a flood light, here are a couple options for mechanical construction and board layout. In both cases the sensor would be placed in the center of the LED array, with a collimator around it as seen in Figure 3c. The collimator prevents the strong LED rays from shining directly into the sensor. Remember that the goal is for the sensor to measure diffused light at a reduced intensity. However, there needs to be a method of reflecting a portion of the light back to the sensor. This is where performance and form-factor requirements come into play. One method would be to add a partially reflecting lens to the cap of the light fixture, as seen in Figure 3a. This would deflect only a small portion of the light back towards the sensor and allow the designer to select a lens which would control the width of the light beam. A minor disadvantage to this would be that the full potential brightness of the LEDs would be diminished by the lens, but more importantly the cost of this system would go up because a good lens which has very little loss will be expensive. If the cost point of the system can handle a lens, this would be the most ideal design. If not, another approach would be to remove the lens and have an exposed opening where the light escapes, but curve the case in a manner which would reflect the stray light back towards the sensor as shown in Figure 3b. This would be less expensive than the first design, but would loose more light and is more susceptible to ambient light interference due to the exposed opening. The choice is determined by the requirements of the system for cost, performance and form factor.
When it comes to sensor selection there are many vendors offering inexpensive color sensors (sub-$3 range in volume) that would serve well in most color sensing applications. They all have slight variations on functionality but essentially output a red, green and blue intensity measurement which can then be used to calculate the exact CIE color coordinates, given calibration. The way this is done is based on the assumption that the LED brightness and the sensor measurements for each color have a linear relationship. Thus, during a one-time calibration procedure this relationship is calculated and used to translate sensor measurements to LED dimmer values. Additionally, there needs to be a relationship between the sensor readings and CIE tri-stimulus values X, Y, and Z. To accomplish this, another calibration is then needed which each LED is individually set to the highest brightness setting and the tri-stimulus values recorded with a color meter along with the output from the sensor. This is repeated with all LEDs and with this data a relationship between the sensor values and the CIE color space can be created.
If all of this sounds tricky, that's because it is, especially when developing firmware to implement this functionality. This tends to be the largest barrier-to-entry for many LED designs. It would take much more than this article to explain how to develop firmware to perform the optical feedback. There is a short cut though, as many microcontroller companies which sell into the high-brightness-LED market have LED driver firmware developed already which provides full optical feedback. A one-time calibration procedure adapts the firmware to the individual design.
The last thing to bear in mind when designing an optical feedback system is that heat is a killer to accuracy. Optical feedback can help to correct for temperature variation of the LEDs to a certain extent. However, the sensors themselves are also susceptible to accuracy drift at extreme temperatures and can have problems of their own when expected to perform accurate measurements at high temperatures. All steps should be made to keep the temperature of the system down, if accuracy is of great concern. This is especially true during calibration. If calibration is performed while either the LEDs or sensor (or both) are hot, it can throw off the calibration to the extent that even when the system is operating at room temperature the accuracy will be minimal.
Incorporating optical feedback into a high-brightness-LED design has its challenges and added cost, and it is not for every design, but it can pay for itself quickly by adding accuracy and robustness.
