Limitations of ICCD detectors and optimized 2D phosphor thermometry

The use of intensified CCD (ICCD) cameras is widely spread in many research disciplines. For 2D detectors such as CCD cameras, the most common intensifiers consist of three functional units: a photocathode, a microchannel plate (MCP) and a phosphor screen, see figure 1. The photocathode converts the incident photons to photoelectrons, which are accelerated by an electrical field toward the MCP. This in turn multiplies these electrons which are converted back into photons by the phosphor screen. Lastly these photons are directed to a CCD chip by either an optical fiber bundle or a lens system, and read out as a digital image. The addition of an image intensifier brings certain benefits. Primarily the MCP serves to intensify the signal several orders of magnitude, enabling detection of very weak signals. In addition, by pulsing the control voltage to either the photocathode or the MCP, the image intensifier can serve as a very fast electronic shutter, enabling time gating for short-lived signals which would otherwise have been lost in the interfering background. Furthermore, the photocathode can be made sensitive to UV radiation. Effectively, the image intensifier can frequency shift an incoming UV signal, to which the CCD itself is not sensitive, into the visible range where the CCD has a high sensitivity. Together these features make ICCD cameras very useful in pulsed laser applications, where short laser pulses, of the order of 10 ns duration, are used for visualization of various substances, either by detecting the scattered laser light itself, e.g. Rayleigh scattering, or a signal emitted from the substance induced by the laser light, e.g. laser-induced fluorescence, laser-induced phosphorescence or laser-induced incandescence.

Zoom In Zoom Out Reset image size

However, it has been shown that ICCD cameras suffer from spatially non-uniform image gain factors and nonlinear response functions [1, 2]. The nonlinearity is mainly due to the onset of saturation within the MCP, when the camera is subjected to high light intensities in combination with excessive gain settings. This falloff in response, or gain saturation, has to do with the MCP recovery process and is caused by the MCP's inability to produce enough secondary electrons from the original photoelectron [3, 4]. In addition, bleaching of the photocathode, especially when collecting intense long-lived radiation, can also contribute to saturation. The phosphor screen can also be subject to saturation, causing a nonlinear output due to response falloff at higher intensities. Other saturations can occur after the intensifying process, i.e. in the CCD chip (filled charge wells causing blooming) and in the A/D converter, where the analog charge from the CCD is converted into digital signals. These types of saturations are usually easily avoided, since it is rather obvious when they occur.

Saturation within the intensifier is usually not a concern when performing qualitative measurements; however, quantitative measurements, and in particular the two-color ratio measurement technique, are easily affected by small nonlinearities. The presented work describes how systematic errors caused by nonlinearity due to saturation of the MCP can be avoided by operating the detectors within the proper workspace. In this context, it should be mentioned that in most practical situations it is not obvious when the MCP begins to become subject to gain saturation.

Thermographic phosphors (TPs) have been utilized in temperature measurement applications for several decades. Used for both point and two-dimensional surface measurements, TPs offer a combined temperature-sensitive range that stretches from cryogenic up to 2000 K, depending on the choice of phosphor [5–10].

The TP used in this experiment is BaMg₂Al₁₆O₂₇:Eu (BAM) [11]. When excited by a UV-laser pulse at 355 nm it emits broadband phosphorescence peaking at around 440 nm, with a lifetime τ of about 2 µs at room temperature. This peak broadens toward shorter wavelengths as the phosphor is heated; see figure 2. Thus, the phosphorescence intensity of the shorter wavelength side of the peak relative to the longer wavelength side of the peak will increase with temperature. Using two interference filters, one with 10 nm spectral width (FWHM) centered at 456 nm and one with 40 nm (FWHM) centered at 400 nm, a temperature sensitive intensity ratio of the phosphorescence can be determined. A pair of filtered images can be recorded by the use of either two cameras or an image-doubling device (stereoscope) [12]. Dividing the intensities in these images by each other, pixel-by-pixel, will result in a temperature-dependent ratio image. The ratios are then converted into temperatures using a predefined calibration curve derived from measurements performed in a controlled environment. Note that the procedure of utilizing signal ratios rather than absolute intensities brings an inherent advantage with this approach. The resulting ratio becomes independent of the overall phosphorescence intensity and laser excitation intensity. Hence, the technique does not require absolute homogeneous exciting laser profile, phosphor coating nor a linear relation between laser excitation and phosphorescence intensity. Despite the appearance of the spectra in figure 2, it is not recommended to use BAM for temperature measurements higher than 800 K due to degradation mechanisms [13].

Zoom In Zoom Out Reset image size

The accuracy of these experiments, ideally, depends only on the accuracy of the reference thermocouples used during calibration. The precision on the other hand depends on several factors, such as the S/N ratio, flat field camera correction and the perfection of the image superposition of the two images that are to be divided. As this paper will show, both the accuracy and the precision depend on properties of the camera, which from the beginning are not obvious, but the focus will be kept on the precision.

In order to make quantitative two-color ratio measurements, it is important that the detector response is linear. The presented work shows that when performing quantitative two-color temperature measurements with TPs using a Princeton Instruments PI-MAX2 ICCD camera, it is necessary to stay below a certain count level in order to avoid nonlinear effects which would otherwise cause false absolute values and reduced precision. For the experiments performed, the trade-off between useable irradiance and gain corresponds to a number of counts less than half the numerical range of the A/D converter.

The results presented herein are valid at room temperature for the particular phosphor and individual camera used; tests must be performed with any new system in order to secure correct readouts. However, since the phosphorescence generally decreases with higher temperature, it is assumed that nonlinear effects due to too high irradiance will not occur once detector saturation is avoided at room temperature.

Figure 3 shows the experimental setup. The laser used is a 10 Hz pulsed Nd:YAG operating at its third harmonic, producing a wavelength of 355 nm with an energy of ∼0.5 mJ/pulse. The laser beam is directed and expanded to illuminate a glass plate surface coated with BAM. The phosphor is applied using an airbrush pen (Rich AB 300) containing a solution of the phosphor, binder (HPC) and ethanol. The thickness of the layer is approximately 10–20 µm. On the camera side of the plate, a filter wheel with appropriate filters is placed. The phosphorescence, seen through the filters, is recorded by a PI-MAX2 ICCD camera. All the measurements are performed at room temperature; thus, it is assumed that the temperature distribution across the coated phosphor surface is homogeneous.

Zoom In Zoom Out Reset image size

Since this experiment focuses on the measurement error introduced by the ICCD camera rather than evaluating the overall thermometry, it is advisable to suppress extraneous error contributions. Two such errors would be incorrect image superposition and flat field correction, which would be necessary in the case of using two cameras or a stereoscope. In order to eliminate these, only one camera is used for this experiment. By switching the interference filters in front of the camera, the two images would be identical with exception to the wavelength detected and also the pixel-to-pixel variations on the CCD chip would cancel out by the image division. While this approach eliminates the source of errors caused by image superposition and flat field correction, it assumes stability in time. To reduce the influence of shot-to-shot variations in the laser profile, accumulations of ten images were made for each measurement.

As described in the setup, the filter wheel is positioned as close to the phosphor-coated surface as possible, rather than in front of the camera lens, as might be expected. The reason for this is an image distortion introduced by the design of the filters which caused displacement of the focal plane and translation of the image. Since the distortion differs between the two filters, this effect would obstruct straightforward superposition of the two images. This distortion effect would probably be eliminated by the use of higher image quality filters, which were unavailable at the time of the experiment. With the filters positioned close to the coated surface, it is impossible to illuminate the phosphor from the side facing the camera. Hence, the surface is illuminated from the backside, which requires a transparent phosphor substrate (the glass plate). The phosphor is coated on the surface facing the laser; as a result, the glass plate acts as a filter, rejecting spurious UV laser contribution to the recorded images while being transparent for the phosphorescence emission. This arrangement is limited in a realistic application due to both setup issues regarding the interference filters and restriction to measurements with more temporally stable temperatures, since the measurements cannot be performed in a single shot.

In order to investigate the linearity of the camera, different areas of the BAM-coated surface are illuminated with different laser fluences. This is achieved by using a filter stack, consisting of five identical neutral-density filters, successively partwise overlapping each other. By this approach, six different areas of the BAM-coated surface are illuminated, each with different laser fluence: one without any laser attenuation, and five with gradually increasing optical densities. The transmission of the five gradually increasing optical densities was measured to be 9%, 14%, 22%, 39% and 67%.

Figure 4 shows an example of two phosphorescence images (a and b) and the resulting ratio image (c) for one particular gain setting on the image intensifier (150 of 255). At this gain setting the different areas, despite illumination with different laser fluence, result in the same ratio, as can be seen in figure 4(c), and thus would read the same temperature as expected. For each area, the mean value and the standard deviation of the ratio is investigated. This is done for 26 different gain settings, spanning 0–250 (max value 255) in increments of 10. With six areas of different illumination intensities, this results in 156 different gain and laser fluence combinations investigated. Prior to the experiment, the laser energy is adjusted so that, in the measurement with a gain of 250, the maximum number of counts in the area with highest intensity is just below the saturation level of the 16 bit A/D converter, i.e. 65 535 counts.

Zoom In Zoom Out Reset image size

Figure 5 displays the measured mean value, in counts, for all 156 different gain and laser fluence combinations for each filter applied (left: the 456 nm interference filter, henceforth IF456, and right: the 400 nm interference filter, henceforth IF400). Because of the exponential behavior of the camera gain, the signal intensity is displayed using a logarithmic scale. The maximum mean value from the area illuminated with highest laser fluence and amplified with the highest gain, illustrated by the upper-right corner in the IF456 chart, is 45 000 counts or ∼10^4.6.

Zoom In Zoom Out Reset image size

Figure 6 illustrates the camera's ability to reproduce the different intensities induced by the filter stack. The bars represent the measured transmission (67%, 39%, 22%, 14% and 9%) through the five different parts of the filter stack, in relation to 100% transmission, indicated by the first bar. The blue filled circles show the normalized average signal counts through the IF456 at gain 80, obtained from figure 5 (left). They show good agreement with the average filter transmission, indicating that the overall camera response (including the MCP) operates in its linear regime.

Zoom In Zoom Out Reset image size

The red circles, on the other hand, show the normalized average signal counts at the highest gain value of 250. Here, the signal counts measured for 0, 1 and 2 filters do not reach the expected values such that the 'camera's response' is lowered to 35% only, reflecting that the ICCD camera has left its linear operating workspace. Since only the gain has changed between the two cases displayed in figure 6, while the photon flux remained constant, the saturation can be attributed to the MCP. The red circles start to differ from the expected values when less than three filters are present, where the signal level exceeded 29 000 counts or 10^4.464 (corresponds to the left chart in figure 5, third column, top row). At that point, the initial amount of phosphorescence light hitting the photocathode was reduced to 22% by the filter stack.

Figure 7 utilizes all data information displayed in both charts in figure 5 to extract the camera's response function for the two investigated wavelengths: 400 and 456 nm. This is done by simulating the expected count behavior for the two charts, starting with the first column for each chart (representing five filters' transmission), assuming that each decrease in transmission through the steps of the filter stack increases the counts with respect to the measured transmission. Since the highest count level, reached behind five filters (at a gain of 250), corresponds to approximately 10 000 counts, the whole first column in both charts in figure 5 represents data accumulated in the linear response regime.

Zoom In Zoom Out Reset image size

The data displayed in figure 7 represent the overall camera response for a photocathode that operates in the linear regime. The counts through IF400 (blue circles) show a slightly steeper slope compared to those through IF456 (red triangles). This could be explained by the wavelength-dependent photocathode sensitivity which possibly emits slightly more primary electrons at 400 nm. For both wavelengths, the initial linear behavior tends to tip between 20 000 and 30 000 counts, indicating the beginning of MCP saturation. Another interesting aspect from figure 7 is that assuming a linear detector response, a measured output signal of 45 000 counts corresponds to a simulated signal level of 130 000 counts. It is remarkable that this is more than twice the numerical range of the A/D converter.

Each of the 156 count values from the IF400 image set is divided by the value from the corresponding gain and laser illumination from the IF456 image set. The result is a ratio chart illustrated by figure 8 (left), which basically could be thought of as the right chart divided by the left chart in figure 5 (if the charts in figure 5 showed the true count values and not the logarithm). The right chart in figure 8 is the same as the left, only with the ratio values translated into temperatures, using a ratio to temperature relation resulting from reference measurements done at temperatures ranging from room temperature to 500 K.

Zoom In Zoom Out Reset image size

As can be seen in figure 8, the mean ratio value and hence the mean corresponding evaluated temperature are rather constant within the major part of the chart, with the exception of the upper-right corner. Despite the measurements done at constant room temperature, it is clearly seen that the evaluated temperature gives a false result and too high values are acquired for the high laser illumination and high intensifier gain. As can be seen on the right chart in figure 8, this results in a false readout of the absolute temperature, with a deviation of almost 200 K from the true value.

In terms of precision, figure 9 shows the most interesting results. It shows the spatial standard deviation, i.e. the pixel-to-pixel standard deviation, within each gain and laser illumination combination area, translated into corresponding temperature standard deviation. For BAM, the relation between temperature and ratio is rather linear within the considered temperature range, with a dT/dR relation equal to 0.0026 K⁻¹, where T is the temperature and R the ratio. To illustrate, a standard deviation in the ratio of 0.009 corresponds to a standard deviation in temperature of 0.009/0.0026 = 3.5 K.

Zoom In Zoom Out Reset image size

As can be seen in figure 9, a diagonal-shaped area, marked in between the two dashed lines, could be identified as a proper working space for higher precision measurements. Within this area, the camera exhibits a linear response independent of MCP gain or laser illumination. Below the lower line, the standard deviation becomes high simply because the S/N level is too low. The signal is still high enough to give reasonable values for the absolute temperature, as can be seen in figure 8, but the noise contribution is too high for the pixel-to-pixel precision to be sufficiently high. The number of counts at which this occurs is very low, about 30 counts per pixel.

The increase in standard deviation above the upper-dashed line indicates a more critical limitation of the camera. The value of counts where this limit is reached can be determined from figure 10, showing the spatial standard deviation of counts in each area for every gain and laser illumination combination through IF456 and IF400, respectively. In the measurements corresponding to a gain setting of 190 in the column for the highest level of laser illumination in figure 9, it can be seen that the standard deviation of the measured temperature, in other words precision, has degraded with a factor of 2 compared with the highest value achieved. Considering the left chart in figure 5, it can be determined that this condition corresponds to a count value of 26 000 through the IF456 (10^4.42). It can be clearly seen in figure 10 (left) that signal suppression is occurring under these conditions, i.e. the nonlinear regime has been entered. The standard deviation is reaching a maximum value at gain 190 and is then decreasing with higher gain, indicating that an upper limit for counts has been reached and that gain saturation will occur for higher gains and illuminations. It is notable that this saturation already occurs at count levels as low as 26 000 counts, less than half of the maximum counts of 65 535. The same effect is seen for the signals detected through the IF400 filter. In figure 10 (right), it can be seen that the value of the standard deviation reaches a maximum in the column for the highest laser illumination at a gain of 230. Considering the right chart in figure 5, it can be determined that this corresponds to a count level of about 29 000 counts (10^4.47).

Zoom In Zoom Out Reset image size

Figure 9 illustrates a region identified as a useful workspace for this camera and phosphor. The upper-dashed line indicates the limit for the gain/illumination trade-off. In the left chart in figure 5, the position of the upper-dashed line in figure 9 would correspond to a count value range of 25 000–30 000 counts (10^4.4–10^4.5). Higher counts should be avoided. The fact that there is a range of gain in which the absolute evaluated temperature and precision are rather constant, regardless of laser illumination (between gain 100 and 190, see figure 9), shows that the saturation occurs after the photocathode. Hence, the strength of the light source, in itself, or the photocathode are not responsible for the saturation. However, it cannot be excluded that the saturation detected, or a portion of it, could be caused by the phosphor screen in the image intensifier. It should be mentioned that the experiments are performed using one ICCD camera and one TP only, and that the saturation limit of 25 000–30 000 counts is only valid under these conditions. However, as a precaution, an additional test was performed using a nominally identical ICCD camera and the results were very similar.

Within the useful workspace, the precision in temperature measurement spans from ±1 to ±10 K. This number depends on temperature and on the choice of the spatial resolution of the detection system. This relationship will be addressed in an upcoming work by the authors.

This work has not only focused on the precision of measurements but has also mentioned accuracy-related issues. The authors would like to see reproducibility investigations on how the accuracy could be affected by long- and short-term drifts of cameras. Also similar investigations to the one described here, but with different types of ICCD cameras, and with different types of TPs (with different emission wavelengths and lifetimes) and at different temperatures, would be a suitable target for further investigations.

In this work, a useful workspace within the gain/illumination combination of an ICCD camera (Princeton Instruments PI-MAX2) has been identified. It has been shown that the camera used suffers from nonlinear effects due to gain saturation in the MCP, when exceeding a certain level of counts. The results show that there is a possibility of saturating the camera, and in so doing introducing nonlinearity, even though other components of the camera are not saturated. The level at which this occurs corresponds to as low as half the maximum counts allowed by the CCD chip, making the occurrence of saturation far from obvious. It is conspicuous that the count value read as 45 000 in fact represents an intensity that should actually correspond to a count value of 130 000. Another interesting conclusion drawn from figure 7, is that the nonlinearity of response has a slight dependence on the observed wavelength. In this study, two nominally identical ICCD cameras were used and showed similar results. However, each detector should be treated individually, and the behavior likely depends on the type of light source as well. This means that an investigation such as the one presented here should be done prior to any experiment involving quantitative measurements using a two-color ratio method.

To conclude, there is an upper limit of counts below which the ICCD cameras should be used in order to give reliable results. This limit is shown to be in the 25 000–30 000 count range out of a nominal 65 000 counts allowed by the 16-bit A/D converter. This level was shown to be valid for any gain.

It is inferred from our analysis that system nonlinearity is also dependent on the type of light source being investigated. In this study, the phosphorescence of BAM is detected at room temperature and at a certain distance between camera and phosphor. If cameras are to detect light at other wavelengths and with a different duration, the limitations of the ICCD cameras are likely to be different.

This work was supported by the Centre of Competence Combustion Processes and the HELIOS project within the seventh EU framework program.