Construction and performance of a 395 nm LED lamp
This page is a companion to my earlier evaluation of continuous near-UV light sources. The purpose of this investigation is finding a simple, easily built and relatively cheap continuous source of near-UV light for photography with a modified digital camera. Ideally, such a source should allow exposure times of 10 seconds or less at close range (10-20 cm), at an effective aperture around f/11-f/16 and at the minimum ISO sensitivity of the camera. The ultimate goal is a relatively powerful source of near-UV in the approximate range between 350 and 400 nm, to which the sensors of digital cameras are still sufficiently sensitive. The UV sensitivity of general-purpose digital cameras (including those enhanced by replacing the built-in UV, IR and anti-aliasing filter) is rather low, and normally does not extend to any useful degree below 300 nm. Many cameras, including modern models, are not usable even after modification. This includes, for instance, Canon models and the Nikon D300.
LEDs are solid-state semiconductor devices that produce non-coherent light, usually in a narrow range of wavelengths. Sometimes, the produced range of wavelengths is changed by coating the surface of the semiconductor with a phosphor compound, which absorbs light produced by the LED and re-emits it in a different, almost always longer, range of wavelengths. This is used, for instance, in "white-light" LEDs, which are blue or violet LEDs coated with a phosphor layer that emits a whitish light. LEDs with different composition and design are now available to produce light between roughly 200 and 1,000 nm. UV LEDs, however, tend to be expensive and low in radiation output. There are plenty of "UV" LEDs made in China availably at low prices (e.g., see here), but they produce visible light around 405 nm instead of UV. This wavelength can be useful to induce fluorescence in some materials, but is less effective than real UV light. 395 nm LEDs with a nominal power of up to 3W (actually producing about 435 mW of radiation) are now available from Winger and Prolight. These are the LEDs discussed on this page. Nichia makes 365 nm LEDs producing about 200 mW of radiation, but they are only sold by Nichia directly to other companies, and still very expensive. They will become interesting once cheaper and more widely available.
For this investigation, I attached eight Prolight 3W 395 nm star modules to a heat sink (salvaged from a discarded Intel Pentium 4 CPU) (above figure). Star modules consist of a chip carrier with lens, soldered onto a hexagonal aluminum base covered by a thin PC board. In my case, each chip carrier actually contains two LED chips under a single lens. Although star modules can be attached to a heat sink with screws, it is easier to glue them with a thermally conductive adhesive (I used Arctic Silver epoxy). These LEDs will self-destruct very quickly if operated without a suitable heat sink, or with inadequate thermal contact to one. The heat sink shown above is more than adequate for continuous operation at full power, even without a fan. Star modules can be arranged in a hexagonal pattern, but the heat sink only provided enough room for two rows of four modules, with some "wasted" space between rows. I also left a gap of less than a mm between adjacent modules to allow any thermal expansion, and connected the modules with arched bridges of naked wire for the same reason..
I subsequently wired the LEDs in series, fed by a constant-current power supply designed for this purpose. As shown above, the lamp does not use reflectors, and therefore works like a floodlight. A better efficiency can be obtained by using individual reflector "cups" around each module. You should not use lenses designed for visible LEDs, because most of them strongly absorb radiation below 400 nm. However, metal-coated reflectors work fine also in UV. Lenses specifically designed for UV LEDs are available. Using reflectors or collector lenses, however, greatly increases the risk of uneven illumination of the subject. For this reason, I decided not to use them for the time being. "Barn doors" or a beehive screen (not shown in the above picture because they would hide the emitters) can be used to restrict light to just the subject area without causing uneven illumination.
These LEDs produce a modest amount of visible light. Its effect on a subject is completely washed out by a 40 W visible lamp located at the same distance to the subject, but the LEDs themselves are quite bright when viewed directly (albeit not to the point of impairing vision, and much less than a small white LED flashlight). They are also bright enough to cause flare points if you have a filter on the lens (like in the above picture). My estimate is that less than 5% of the radiation is actually visible, so you should always wear UV protection goggles while operating a similar lamp. In order to avoid flare, you should of course avoid directly illuminating the camera lens.
I tested this lamp with a 60 mm UV Rodagon and a modified D70s. The subject is a red Poinsettia, or "Christmas star" (Euphorbia pulcherrima). Its flowers do not display any particular UV pattern, but right now I don't have much choice of other subjects. All pictures were taken at ISO 200 and f/16, with white balance for incandescent light and changing exposure time as necessary to obtain comparable pictures. No changes were made to color balance and saturation except for the last three pictures (center crops), which were converted to BW by turning the color saturation to zero. The LED lamp was placed roughly 20 cm from the subject. In addition, I took similar pictures also with the HID lamp discussed here. Most UV pictures were taken with a Schuler filter. For visible-only pictures, I used a B+W 486 filter, which cuts UV, IR and some of the deepest red, and an incandescent lamp. For IR pictures, I used filters with 720 and 820 nm cutoff wavelengths, and an incandescent lamp.
The first thing to notice is that pictures with the LED lamp (1, 2) are identical, regardless of whether a Schuler filter is used (the only difference is that exposure time more than doubles with the latter, because of absorption by the filter). This means that the LED light, as expected, is monochromatic. These pictures are blue because of the color balance used. With a daylight color balance, they would be the more typical indigo. Pictures taken with a HID and Schuler filter (3) are more indigo, because the light source contains various UV and visible wavelengths, and the filter passes some indigo as well. The LED without Schuler filter allows approximately 10-15 times shorter exposures, compared with the HID at the same distance (which forces the use of the Schuler filter). The LED lamp allows, so far, the shortest exposure among the UV sources I tested.
Pictures in visible light only (4) and with no filter (5, effectively visible + IR) are as expected. IR washes out the visible color information almost completely. The 720 nm filter (6) gives pictures that still contain some color information (the seed pods are bluish, everything else pink/indigo). All color information is lost with the 820 nm filter (7).
Pictures taken with the LED lamp (8) are very sharp, probably because of the monochromatic quality of the light. Pictures taken with the HID (10) are slightly less sharp, perhaps because of a small amount of visible and IR information. With the 820 nm filter (9), instead, resolution is lower. This is probably due to a combination of factors: IR focus shift (the lens is not designed for IR work), higher diffraction (because of the longer wavelengths) and lower overall contrast of the subject at these wavelengths.
A battery of multiple 395 nm power LEDs provides enough near-UV light to allow reasonable exposures of static subjects in macro photography. With this lamp, a UV-pass filter is necessary only to filter out ambient light and UV-induced visible fluorescence. The cost of this light source (approximately € 100 for the LEDs) is higher than for other low-cost sources, but still rather cheap, especially if you can do the assembling yourself. 395 nm may be a bit too close to visible light to allow the recording of some UV information. Brighter UV LEDs with shorter wavelengths will no doubt improve the performance of this type of lamp.