The recent article on airglow (“The Night Glows Brighter in the Near-IR,” April 2012
) contains misleading statements and a generally sloppy treatment of radiometric units of measure. Some behind-the-scenes math by the authors and more scrupulous editing by Photonics Spectra
would have gone a long way to making this article more correct and meaningful. Instead, it reads like a marketing “puff piece.” Ironically, the article tends to downplay the signal-to-noise advantage of InGaAs and visible-enhanced InGaAs sensors for airglow-only imaging scenarios as compared with silicon sensors used in the same conditions. InGaAs and visible-enhanced InGaAs cameras are clearly superior under certain conditions, but the article simply asserts this with minimal supporting data.
The article should have related the observed airglow sterance values for airglow to signal-to-noise ratios for the Xenics InGaAs camera with a typical short-wavelength infrared (SWIR) lens f
/stop setting, the appropriate pixel active area, read noise and dark current, and a practical frame rate (30 Hz comes to mind). The same calculation also should have been performed for a low-light silicon sensor, such as an sCMOS or electron-multiplying CCD camera, and, possibly, Gen 3 night-vision goggles based on a microchannel plate image intensifier. Without this exposition, the reader is left wondering just how useful the airglow is for night vision and which technology has the advantage.
The answer is that InGaAs cameras are very useful for this application, provided they have low enough noise-equivalent irradiance (NEI). This NEI requirement generally translates to a need for a noise level of less than 50e—
in hopes of imaging with airglow at a reasonable signal-to-noise ratio with low f
number optics at 30-Hz frame rates. By comparison, experience tells us that the noise floor for silicon cameras can be down around 1 to 2 e—
per pixel per frame at 30-Hz frame rates, but that isn’t enough to give good 30-Hz imagery from airglow alone – there is simply not enough no-moon airglow signal in the visible band and near-IR band out to the silicon cutoff around 1.1 µm.
It is simply untrue that the spectral irradiance is “several times stronger” in the 900- to 1700-nm band relative to the visible band, as the second sentence of the article asserts. The spectral radiant sterance is much stronger in the InGaAs band relative to the visible band. By careful examination of curve 1 in Figure 1 in the article, one can see that the spectral radiant sterance at the peak of the curve is actually about 40 times greater than the spectral radiant sterance at 0.510 µm, where the scotopic (low light) response of the eye peaks. No one would consider the factor 40 equivalent to “several.” Even at the peak of the no-moon airglow curve in the visible band at 0.577 µm (emission resulting from monoatomic oxygen), the peak SWIR signal at 1.6 µm is still 18 times greater. It appears that the graph in Figure 1 was misinterpreted as having a linear Y-axis, rather than a log base 10 axis.
I am not even sure why the authors chose the term “spectral irradiance” to describe the airglow radiation in a given waveband, since spectral irradiance is a function of wavelength, and irradiance is the radiant sterance integrated over the effective solid angle. The irradiance on a surface illuminated by airglow will vary with the angle between the illuminated surface and the sky zenith. Radiometric units are being bandied about here with no clarification or exposition.
Using the data from Figure 1, curve 1, my calculations show that the effective photon sterance for an InGaAs camera is 180 times the effective photon sterance for low-light (scotopic) human vision. Again, this ratio is quite a bit more than “several.” That is the point of using SWIR imagers to exploit the airglow radiation – there is 180 times more total usable signal available to an InGaAs camera relative to the total signal usable by the unaided human eye and 22 times more signal available relative to a monochrome silicon sensor. My calculations also involved convolving both the photon spectral response curve of a typical InGaAs sensor, a typical silicon sensor and the low-light human eye luminous response curve with the spectral radiant sterance curve in Figure 1 converted to photon units. These convolutions were then integrated over the appropriate passbands. Flir’s VisGaAs detectors give similar results to standard InGaAs because almost all the airglow signal is at 0.9 µm and above.
The values I calculated are based on curve 1, but the airglow intensity is variable with cloud cover, solar activity and latitude, and that all should have been noted in the article. A survey of other airglow measurements yields various values of photon sterance for an InGaAs camera on the order of 1010
photons per second per square centimeter per steradian. This is comparable to results using the Vatsia curve 1 data. It would be useful to have given a range of sterance values one can expect based on airglow measurements in the literature and, as mentioned earlier, to have made some attempt to relate these observed sterance values to signal-to-noise ratios for the Xenics NV camera and other competing low-light electro-optics technologies.
The careless treatment of radiometric units of measure manifests itself in several other places in the article. The Y-axis in Figure 1 in the article has incorrect units, as it did in the original paper by Vatsia et al. The units should be spectral radiant sterance, not radiant sterance, because the Vatsia group measured the radiation from the sky with a Fourier transform spectroradiometer as a function of wavelength. The four curves are thus plots of the measured spectral radiant sterance for four different moon illumination levels. The radiant sterance (also called power radiance) is the integral of the spectral radiant sterance curve over a particular passband. It is the in-band radiation available to be collected at the optical aperture of a camera. The number of photoelectrons generated by this radiation depends on the absolute spectral response or quantum efficiency curve of the detector/lens combination, which is then convolved with the spectral radiant sterance curve as well as the pixel active area, the lens f
number and optics transmission.
Those of us in the infrared camera field who are doing modeling of airglow imaging routinely convert historical measurements such as Figure 1 into “photons per second” units instead of their original “watts” power units. We thus end up with spectral photon sterance. Photon units are a more logical set of units to use for these airglow calculations because the detectors cited in Figure 2 in the article are all photon detectors, not power detectors. Figure 2 in the article has the spectral responses of the detectors in power units. If I had written the article, I would have first converted the curves in Figure 1 to photon units and then presented the spectral responses in photon units in Figure 2. The InGaAs spectral response curve is nearly flat in photon units, making it reasonable to describe it with a “square band” approximation; i.e., integrating the spectral photon sterance from 900 to 1700 nm.
Another quibble with Figure 2 is that it has an airglow curve, but the second Y-axis is not labeled with radiometric units. I’m assuming the correct Y-axis units for that curve are W/cm2
/sr/µm because the peak value at 1.6 µm is 100 times greater than the peak value in the Vatsia data in Figure 1, curve 1. The factor of 100 would come from the difference between using 10 nm and 1 µm in the unit denominator. Labeling the second axis of Figure 3 with the right units would have been a good idea to avoid confusion with the Y-axis units in Figure 1.
Figure 1. Spectral photon sterance: Vatsia’s 1972 no-moon nightglow,
The article asserts that the radiation densities of moonlight and airglow are comparable. What is meant by radiation density? Do the authors mean the spectral radiant sterance? Moonlight and airglow are indeed comparable in power units (spectral radiant sterance), but not in photon units, which, again, are more appropriate for calculations involving photon detectors. The spectral photon sterance is about 2.5 times greater at 1.6 µm than at 0.51 µm for curve 4 data in Figure 1 of the article, which was generated by data taken at 89 percent moon illumination. But, again, what is important here is the spectral photon sterance integrated over wavelength in the appropriate passband, weighted by detector spectral response in photon units.
With an 89 percent moon, InGaAs and VisGaAs detectors will have about five times the usable signal that the human eye has available. Under those same conditions, a typical monochrome silicon sensor will have about the same usable signal as an InGaAs camera. That should have been explained in more depth. Another advantage of airglow is that it comes from all directions and reduces shadows in the scene, making it harder for an observer to miss a person or other item of interest. Moonlight casts shadows that get very long at a low elevation angle, which increases scene clutter and tends to reduce the ability to find items of interest in a scene. Of course, if there is enough cloud cover, both moonlight and airglow can be reduced in intensity to the point where the only viable imaging method is thermal.
Below is a plot of the data from curve 1 in photon units, with a linear scale on both the X- and Y-axes. The bands for the three types of sensors extend out to their respective 50 percent points. A graph like this would have better served to illustrate the huge advantage that an InGaAs or visible-enhanced InGaAs sensor has over both the unaided eye and low-light silicon sensors in no-moon conditions because the areas under the curves in each band are a measure of the total available signal for these three types of sensors (assuming square-band response and the appropriate 50 percent cut points). The original Vatsia graph is hard to interpret in this way because of the semilog axis and the power units of measure. My graph also illustrates that an extended InGaAs sensor that responds to radiation out to 1.9 µm would make use of SWIR airglow that conventional InGaAs won’t detect.
Dr. Austin Richards
Flir, Commercial Systems
We would like to thank Dr. Richards for the detailed comments on our article. In general, his letter is a good complement and clarifies our text with more in-depth information. It never was our intention to fully elaborate on the different topics Dr. Richards mentions in his letter. With our introductory article, we intended to give a broad audience an idea of what can be achieved in the wavelength band outside of the visible region in the field of night imaging. This explains why most of the text gives qualitative arguments to come to the (correct) conclusion, and not so much a quantitative reasoning that would not be compatible with the limited length or scope of the article.
We believed that as Photonics Spectra
is not peer-reviewed, it was not immediately the proper medium in which to publish a complete, in-depth analysis of nightglow and low-light-level imaging, but it was perfect to show the possibilities and potential of the InGaAs-based detector systems Xenics currently produces. It is worth mentioning that we have received several other positive comments on this article from readers, who found it a nice, refreshing introduction to the topic.
We agree with Dr. Richards that it would be interesting to write a more detailed article on the subject, one that would allow the interested reader to become more acquainted with the underlying physical mechanisms on different system levels.
Jan Vermeiren, Xenics NV