Lens testing with economy

In color cameras, detectors are filtered in a Bayer pattern to be sensitive to red, green, or blue, but not all three simultaneously.

Make everything as simple as possible," Einstein once said, "but no simpler." Excellent commercial equipment is available for evaluation and modulation transfer function testing of lenses, but at prices of $30,000 and up it is often difficult to justify, especially in the early stages of a project when the need is greatest. When we faced this dilemma earlier in our careers, we generally ended up building one-off test setups. By scrounging parts, we eventually created home-brew equipment that would do the job. We told ourselves that we were keeping costs down, as our managers watched the weeks roll by. Later, as optical consultants, we discovered we had not been alone, so we took up the challenge to build a system that we ourselves could have afforded.

The result is a tabletop setup with all the components of a classic lens bench. The commercial halogen fiber-optic source with a 3200K bulb has a wide enough spectrum for testing through the visible and near-IR spectral regions. The output of the fiber bundle is homogenized and shaped to fill the F/8 collimator lens. The light passes through a narrow-band color filter to back-illuminate a test reticle positioned at the image plane of a well-corrected collimating lens. Quick-change holders carry both filter and reticle; a three-jaw chuck holds the lens under test, and a simple mechanism allows rotation of the lens about the nodal point.

The lens under test forms an image of the reticle. A microscope relays the image to a Sony CCD camera chip. The image is captured, digitized, and transferred to a PC over a FireWire interface. If the lens under test is aberration free, then the image will be a perfect reproduction of the reticle, subject only to the unavoidable blur caused by the wave nature of light.

People often want to know the polychromatic response of a lens or other optical system. In other words, they want a single figure of merit (or plot) for white light instead of one each for red, green, and blue (RGB). This is useful, especially if the lens will ultimately be used with a monochrome camera.

Color cameras use a "Bayer pattern" of pixels in which each pixel is sensitive to either red, green, or blue, but not all three at once (see figure). The camera must interpolate to compute RGB at each pixel site. For a metrology system, this process can introduce problems. Some simple algorithms produce color artifacts when the size of a feature is very small. Camera vendors have developed adaptive algorithms that suppress artifacts, and deliver images that are pleasing to the human eye. This adds processing time, although the task can be performed by hardware within the camera.

A more serious problem is that the location of features may shift slightly. If a feature shifts laterally by a half pixel in a photograph the effect may be unnoticeable. In the case of high-resolution metrology, however, such a problem can be disastrous.

Thus, we designed our system with a monochrome camera in conjunction with a user-defined set of quick-change color filters that allow the system to measure one color at a time at precisely known wavelengths. The design is not limited to three colors—assuming filter availability, it only takes a little more time to measure at four, five, or more wavelengths, or even out into the near-IR spectral region, since the camera CCD provides sensitivity out to about 1.1 µm. The user can generate a white-light figure of merit by RGB data with suitable weighting.

We expected that most users would use the system to test lenses with an image at infinity, as with a conventional lens bench. However, we also knew that some unusual situations would arise, for example testing relay lenses at finite conjugates, or afocal instruments like beam expanders. We wanted the bench to be flexible enough to accommodate these requirements. To accomplish this, the instruments are modular and subsystems can be interchanged to create new setups. Each subsystem is keyed so that instruments are optically correct and precise factory-set collimation is preserved. For example, the collimator consists of three modules: an image source, a spacer block, and a 200-mm collimator tube. To create another setup, the 200-mm collimator tube might be replaced by one of a different focal length. The spacer block may be replaced by a beamsplitter to create an autocollimator. The image source may be replaced by an eyepiece, a CCD camera, a photodiode, a quad cell, or a fiber port.

Of course, the bench is subject to the usual caveats—the collimator must be large enough to fill the entrance pupil of the lens under test, and the image capture microscope must have high enough numerical aperture to capture the full ray bundle from the lens. However, within these limitations the system answers our challenge—a mix-and-match, component-based lens test bench that is as simple as possible—but no simpler. oe