1. INTRODUCTION

The various applications of spectroscopy are currently developing at a rapid pace and undergoing a significant expansion far beyond the previously dominated areas of scientific research and laboratory use. More specifically, there is a growing demand for spectroscopic solutions related to process control in many manufacturing industries, pharmaceutical production, a broad range of applications in the environmental industry, and particularly in the agriculture and food industries [1–6]. The instrumental requirements for the extended application fields are mostly very ambitious and often differ considerably from one another. Furthermore, partially seemingly contradictory specifications are targeted simultaneously, such as a broad spectral range, a high spectral resolution with a maximum detection efficiency, and, in addition, a compact and robust instrumental setup enabling a fast data acquisition.

Spectroscopic instruments capable of simultaneously meeting all requirements, referred to as all-purpose tools, are not yet available. Hence, many existing spectroscopic instruments have recently been adapted and improved, which has resulted in the introduction of completely new concepts. This includes a variety of classical, dispersion grating-based spectrometers, such as compact spectrometers that employ concave reflection gratings [7–9], Czerny–Turner configurations and modified versions thereof [10], and echelle spectrometers [11]. Significant progress in the development of dispersion grating-based spectrometers has been made in recent years, e.g.,  by reducing aberrations and improving spectral resolution [12–14], by introducing cross-grating for size reduction of echelle spectrometers [15,16], or by developing new fabrication technologies for the grating allowing tailored and optimized spectral efficiency [17,18], to identify only a few. Despite this progress, there are still requirements that cannot be achieved with dispersion grating-based spectrometers, such as combining compactness with high spectral resolution and wide wavelength range. For example, the smallest spectrometer [9] has the size of nearly ${0.15}\;{{\rm cm}^3}$, but suffers from weak resolution [full width at half-maximum $({\rm FWHM})\; \approx \;{17}\;{\rm nm}$]. Compact grating spectrometers with better performance in one optical property are larger [7,8] and are either limited by bandwidth or resolution. Therefore, the development of alternative spectrometer types remains important, in particular with respect to a strong miniaturization potential.

An alternative to grating spectrometers are filter array or continuously variable filter-based spectrometers. These static spectrometers are characterized by a simple concept in which multiple discrete filters, often arranged in an array, or laterally variable filters are implemented directly in front of the detection pixels of a two-dimensional photodetector. This simple concept offers a highly compact and rigid setup that allows extremely slim sensors to be used, which eliminates any moving elements and makes it possible to capture the target spectrum in a short period of time. A multitude of different approaches have been developed for the filters comprising, for example, Fabry–Perot (FP) filter arrays [19–24], plasmonic color filters [25–27], absorptive filter arrays composed of colloidal quantum dots [28], and linear variable filters (LVFs) [29,30]. In general, the simple detection principle involves an even distribution of the light to be analyzed over the entire area of the filter field. Thus, the complete spectral information is incident on each filter of the array, but only a specific and narrow bandwidth is transmitted and reaches the detector pixels. Using many filters with small transmission peaks or using an appropriate gradient for LVFs a wide wavelength range with a high spectral resolution is detectable in a single-shot measurement. The basic concept of filter-based spectrometers is quite flexible and offers in principle an almost freely configurable lateral distribution of the filter function, thereby allowing for application-specific tailoring of the spectrometer. Some filter arrays are very small and their optical characteristics are already competitive. The comparison of such spectrometers with other spectrometer types could be found in Ref. [23], and their improved detection range (${\sim}{163}\;{\rm nm}$) in Ref. [24].

As a disadvantage, filter-based spectrometers typically suffer from low detection efficiency. This can be explained by the basic detection principle, in which each individual narrowband filter of the array or filter segment receives the entire wavelength range of the light to be analyzed, but only a small fraction of this range passes the filter and can be used for recording. Thus, most of the incident light that reaches each filter element is largely reflected or absorbed.

This paper describes a simple method that significantly increases the efficiency of filter-based spectrometers. The basic principle for efficiency enhancement involves a spatial, wavelength-dependent redistribution of the light to be analyzed so that each filter element or filter segment receives only a pre-selected and adjusted spectral partition of the entire incident light. With this approach, the reflection and absorption losses on each filter element are significantly reduced. To prove the basic concept, from the different options used to achieve a spatial, spectral redistribution of the incoming light field, we chose and employed a simple concept based on the application of a set of dichroic filters and mirrors. For proof-of-concept, the efficiency enhancement module was combined with FP filter arrays [21–23]. For quantitative analysis, the setup comprising the dichroic filters and mirrors was compared with a reference setup, which dispenses with the wavelength-dependent spatial redistribution of the light field to be analyzed. It has to be mentioned that the aim of this contribution is not to introduce a competitive, commercial spectrometer but to confirm the proof-of-concept of wavelength separation for improved detection efficiency. The demonstrated experimental verification shows this clearly.

 

Fig. 1. Schematic comparison of the spatial light distribution in filter array-based spectral sensors. (a) Conventional setup; (b) basic concept for efficiency enhancement; left side: detection principle in overview, right side: local conditions at a selected single filter element. (a) Each filter pixel receives the whole spectrum but only a narrow spectral line is detected and most of the incident light is lost. (b) Incident light is spatially and spectrally redistributed across the filter array by a primary optical module. Each filter element receives an accumulated, pre-selected spectral partition of the entire incident light.

Download Full Size | PDF

2. BASIC CONCEPT TO INCREASE DETECTION EFFICIENCY OF FILTER ARRAY-BASED SPECTRAL SENSORS

The fundamental efficiency problem associated with a conventional filter array-based spectral sensor is schematically depicted in Fig. 1(a), and it is compared with the basic concept of the proposed efficiency enhancement in Fig. 1(b). The left side of both sub-figures presents an overview of the detection principle by visualizing the illumination of the complete area of the filter array by the incoming light. On the right side of both sub-figures, a magnified view of the local situation at a selected single filter element is displayed. In a common setup, the incoming polychromatic (“white”) light to be analyzed is uniformly distributed across the spatial extent of the entire array [Fig. 1(a) left]. Each of the specific filter pixels receives the entire spectrum, though only a narrow spectral line is transmitted and used for detection. Most of the incident light is reflected or absorbed and lost for detection, respectively. On the other hand, for efficiency enhancement, the entering “white” light is spatially and spectrally redistributed across the filter array by a primary optical module, while the total intensity collected by the detector remains unchanged. In this case, each specific filter element receives a pre-selected spectral partition of the incident light with spatially accumulated intensity. A specific adjustment of the spectral partition and filter element involves a significant increase in the intensity of the desired wavelength on the filter element compared to the standard detection setup. As a result, the transmitted intensity drastically increases for the specific wavelength of the individual filter element, while reflection or absorption losses are significantly reduced.

It is important to note that the working principle of the enhancement approach requires a certain correlation between the distribution of the spectral characteristics of the filter elements across the area and the spatially spectral partition of the incident light. In particular, the wavelength selectivity of the filter elements must not be distributed in a completely random manner, but the filter elements must be allocated in “spectral groups” of the same wavelength range so that spectral allocation is possible. This seems to be a limitation, but practically many filter-based systems are designed in this way, e.g., using LVFs [29,30], or combining FP filter arrays with different stop bands in one group [24].

To employ the spectral pre-selecting optical module, a variety of different approaches is possible, which can be distinguished on one hand by the lateral extension, the angle dependency, and the spectral composition (e.g., continuous versus stepwise) of the outgoing light distribution, and on the other hand, by the complexity, size, and necessary alignment efforts of the layout. In particular, simple dichroic filter-based approaches are advantageous in terms of reduced complexity and increased efficiency.

In Fig. 2, two possible concept variants of a dichroic filter-based optical module for spectral redistribution are shown schematically. The first example in Fig. 2(a) presents a setup based on cascaded Köster prisms. The incoming light is incident perpendicularly on the facet of a first prism, so that a wavelength-dependent change of the propagation direction inside the prism is prevented. The opposing vertical facet of the first prism is covered with a dichroic layer, which separated the transmitted and reflected light by wavelength. The transmitted partition enters a symmetrical prism. The ray bundles leaving both primary prisms at the outgoing facets enter secondary prisms, again in perpendicular orientation to the respective facets. The primary and secondary prisms are separated by a small gap. Inside the secondary prisms, the respective partial spectra are incident on further dichroic filters, which induce refined wavelength separation. Finally, all partial ray bundles are totally internally reflected and leave the bottom facets of the prisms as a wavelength separated and parallel set, which enters the filter array.

 

Fig. 2. Some possible concepts for the spectral redistribution based on the use of dichroic filters. (a) Setup based on cascaded Köster prisms and dichroic filters on specific facets. (b) Configuration employing a series of consecutively arranged tilted dichroic mirrors.

Download Full Size | PDF

As a second example, Fig. 2(b) shows a configuration that employs a series of consecutively arranged tilted dichroic and standard mirrors. The light entering the first dichroic mirror reflects a selected wavelength range (e.g., blue-green short wavelengths) to a selected area of the filter array. The remaining longer wavelengths are transmitted and interact in analogy with the next dichroic filter. For a fine spectral separation, multiple dichroic filters can be subsequently applied. Finally, a standard mirror directs the remaining spectral portion to the respective area of the filter array. In this approach, all spectrally separated ray bundles are oriented in parallel and target the filter array in a perpendicular direction. By comparison, the previously described setup allows for simple adjustment and high stability, whereas the second concept requires a smaller volume and fewer optical elements. The second approach has been selected to demonstrate the efficiency enhancement in our proof-of-concept.

3. PROOF-OF-CONCEPT FOR EFFICIENCY-ENHANCED FILTER ARRAY-BASED SPECTRAL SENSOR

The basic working principle of our efficiency enhancement concept has been experimentally demonstrated in combination with specific FP filter arrays. Each filter element (pitch) of the array consists of two parallel highly reflecting mirrors [distributed Bragg reflectors (DBRs)] separated by a resonance cavity between them. Dependent on the individual cavity height, each filter element transmits a narrow spectral band with a specific wavelength. The manufacturing process of the FP filter arrays involves in essential plasma-enhanced chemical vapor deposition to create the layers for the DBRs, while nanoimprint technology is used to achieve multiple cavities with different heights in a single step. To obtain a high lateral homogeneity of the residual layer, the filter wavelengths are arranged in a specific manner following the volume equalized design. A detailed description of the fabrication process and the characteristics of the FP filter arrays can be found elsewhere [24].

For our experiments three columnar substrates, each comprising three filter arrays were available. Figure 3(a) shows a photograph of a single columnar element with the three filter arrays clearly visible. Each array comprises ${12} \times {12}$ individual pitches, whereas only the central ${8} \times {8}$ pitches are elaborated as specific filter elements, which allow for the transmission of 64 individual wavelengths. The three different arrays in one column have been designed equivalently, but some minor deviations during the imprint process of each array cause small differences in the spectral characteristics of the corresponding filter pitches. The set of 64 transmission curves of each filter array covers a limited spectral range of about 50 nm. To address an extended wavelength range, three columns have been arranged in parallel, with each being responsible for a different spectral region. Figure 3(b) presents a schematic of the three-column arrangement comprising the individual filter arrays. The functional active ${8} \times {8}$ individual filter pitches of each array are marked in bold. Each filter pitch covers an area of ${40}\;{\unicode{x00B5}{\rm m}} \times {40}\;{\unicode{x00B5}{\rm m}}$. The distance between the pitches in an array measures 11 µm. In the schematics presented in Fig. 3(b), the distance between the columns is not drawn to scale, but for implementation needs it measures 0.4 mm. As an example, the ${8}\times {8}$ matrix table in Fig. 3(c) is indicating the peak maximum of the transmission curves of the individual filter pitches for the central filter array. The filter peaks are ranging from a minimum of 576 nm to a maximum of 630 nm. The associated measured transmissions curves for the individual filter pitches are presented in Fig. 3(d), respectively. Due to manufacturing characteristics, some wavelength peaks appear multiple times (e.g., the 579 nm peak in the 8th row of the matrix occurs twice) and some discrete peaks are missing (e.g., the 577 nm peak is not present).

 

Fig. 3. Fabry–Perot (FP) filter arrays and their experimental spectral characteristics [24]. (a) Photograph of a single columnar element with three filter arrays. (b) Schematic representation of three neighboring columns comprising individual filter arrays. Each column is tailored for a specific spectral bandwidth. (c) Matrix representation of the peak maximum wavelength of the transmission curves for the filter pitches of the central filter array. (d) Measured transmissions curves for the filter pitches of the central filter array. The full width of half-maximum (FWHM) of the individual transmission filter lines has an average of 3 nm.

Download Full Size | PDF

The FWHM of the individual transmission curves ranges from 1.7 to 5 nm with an average of 3 nm. The overlap of the transmission curves allows for accessing to the complete wavelength interval without significant gaps. In addition to the specific transmission peak, the spectral curves for each filter pitch are characterized by a limited extended stop band. At wavelengths shorter than the lower stop band limit and longer than the upper limit, the filters are again transparent. For the spectral curves displayed in Fig. 3(d), which corresponds to the filters of the central array, the lower stop band limit is found at approximately 540 nm and the upper limit at 645 nm, respectively. To suppress these disturbing contributions, additional cut-off filters must be applied.

For our purpose to demonstrate the proof-of-concept we have selected only the central filter array of each column. In addition to the array of the central column, which includes a wavelength range of 576–630 nm, the adjacent left filter array addresses a spectral interval of 521–571 nm, while the right filter array’s interval ranges from 628 nm to 685 nm for the array on the right. To record the light transmitted by the filter pitches of the different arrays, the three filter columns have been placed in front of a 1/1.8" CCD sensor chip with ${1.600} \times {1.200}\;{\rm pixels}$ and lateral pixel dimensions of ${4.4}\;{\unicode{x00B5}{\rm m}} \times {4.4}\;{\unicode{x00B5}{\rm m}}$ (Sony ICX274AL). The distance of the filters to the active CCD region measures 1.3 mm.

To demonstrate the approach for efficiency enhancement, a setup comprising successively arranged dichroic filters has been selected for wavelength-dependent separation and spatial redistribution of an incident ray bundle. A 3D representation of the setup is shown in Fig. 4(a) (CAD model). Using the side view of the CAD model the schematic concept is depicted in Fig. 4(b). First, the incoming light hits a prism element covered with a dichroic layer stack, which transmits light wavelengths shorter than 491 nm, while reflecting all longer wavelengths. This wavelength separation is necessary to exclude wavelength contributions from the detector, which are lower than the minimum spectral stop band limit of the filter arrays. The prism as a substrate for the filter (e.g., instead of a plano-parallel plate) is used only for integration and compactness purposes of the mechanical setup. The ray bundle reflected by the first dichroic element approaches the next dichroic mirror, which reflects wavelength shorter than 565 nm. The reflected bundle is directed to the first filter array with corresponding transmission characteristics. The remaining bundle passes the dichroic filter and interacts with the subsequent dichroic mirror. This element reflects wavelengths smaller than 625 nm, which are directed to the second filter array. The longer wavelengths propagate toward a third dichroic mirror. Here, light with wavelengths shorter than 702 nm is reflected and larger wavelengths are transmitted. Analogous to the first dichroite, the last wavelength separation is necessary to eliminate wavelengths longer than the maximum stop band limit.

 

Fig. 4. Model for the (a) setup, (b) implemented efficiency enhancement module, and (c) reference system. (a) 3D representation showing the integrated deflection mirror, dichroic mirrors for wavelength separation, and the filter arrays. (b) Cross section of the efficiency enhancement module. The given values are indicating the transmission and reflection characteristics of the individual filters. (c) Cross section of the reference module providing a uniform distribution of the incoming light across the filter arrays.

Download Full Size | PDF

In our setup, the partial spectral ranges of the three filter arrays adjoin nearly directly each other, so that an almost continuous overall wavelength range is accessible. To prevent or reduce the occurrence of spectral gaps with reduced detection efficiency in the setup, the edge wavelengths of the dichroic filters should be adjusted to the spectral bands of the filter arrays. In particular, the edge wavelengths of the dichroic filters should correlate to the spectral transition regions of the filter arrays. In our proof-of-concept, this demand has been nearly achieved, with the exception of smaller regions which were due to restrictions concerning the simultaneous availability of adapted dichroic filter and spectral bands of the arrays. In an optimized setup, both quantities can be precisely adjusted with respect to each other. Hence, it would be advantageous to tailor a spectral overlap for adjacent filter arrays to ensure coverage of regions that correspond to the transition regions of the dichroic filters. The dichroic filters used in our experiments show transition regions of approximately 5–7 nm measured as the 10%–90% slope of the transmittance curve.

 

Fig. 5. Photographs of the manufactured efficiency enhancement module. (a) Efficiency module integrated in a mounting for the adaption to a detector. The deflection mirror and the dichroic mirrors are clearly visible. For size comparison a Euro-cent coin is shown. (b) Closed module with entrance aperture. (c) Back side of the module, showing the filter arrays.

Download Full Size | PDF

In addition, a reference system has been prepared to simulate a conventional approach of the filter spectrometer without spectral redistribution, but which provides a uniform distribution of the incoming light across the three filter arrays. That means the reference system simulates the behavior of a conventional filter system, in which the incoming light bundle hits directly the whole area of the three filters. For a suitable comparison, we have applied an equal design for the reference system as for the proof-of-concept that differs only in minor details, and distinguishes mainly in the spectral distribution across the filter arrays. In detail, Fig. 4(c) depicts the schematic setup of the reference system. The dichroic mirror responsible for the 565 nm separation of the initial setup is replaced by a simple wavelength-independent beam splitter with a reflection/transmission ratio of 1/2 (one third is reflected, two thirds are transmitted). Furthermore, the dichroic mirror responsible for the 625 nm reflection is replaced by a wavelength-independent beam splitter with a reflection/transmission ratio of 1/1. A minor difference in both setups concerns the first dichroic element on the prism and the last, which are replaced in the reference setup for simplification reasons by metallic mirrors. To compensate for the different conditions in the reference system, the input wavelength range is limited to 490–700 nm by using appropriate filter elements outside the sensor.

To connect to a camera and to obtain an exact position with respect to the CCD chip, a mechanical adapter has been fabricated. The module is very compact, measuring only ${17.5}\;{\rm mm} \times {17.5}\;{\rm mm} \times {7.8}\;{\rm mm}$. The photo in Fig. 5(a) shows the front side of the adapter with the redistribution module, and a 1-Euro-cent is also displayed for size comparison. To avoid disturbing stray-light effects during the measurement, the adapter is optically sealed by a cover that is screwed into the housing of the adapter, as shown in Fig. 5(b). The cover is equipped with a small aperture, thus allowing the input light to be guided onto the initial dichroic prism. Finally, the adapter with the redistribution module is connected to the camera system, which could be read out directly. Figure 5(c) shows the back side of the adapter with the substrates carrying the filter arrays.

4. EXPERIMENTAL VERIFICATION OF THE WORKING PRINCIPLE

To experimentally verify the efficiency enhancement method for filter-based spectrometers, a simple optical setup has been used. Figure 6(a) shows schematically the basic components of the setup. The light of a mercury-cadmium (HgCd) arc lamp is coupled into an optical fiber and guided to the spectral sensor. At the fiber output, light is collimated by a lens and directed to the entrance aperture of the spectral sensor comprising the described spectral pre-selecting module, the filter arrays, and the CCD camera. It has to be mentioned, that in general the collimation of the input light is required for filter-based spectrometers, because the wavelength selectivity of FP filters is strongly dependent on the incidence angle. For example, a 6° change in the incidence angle is responsible for a wavelength shift of nearly 1 nm [23,24]. To select a specific wavelength of the HgCd lamp, different filters can be integrated into the setup between the collimating lens and entrance aperture.

 

Fig. 6. Experimental verification of the efficiency enhancement method. (a) Schema of the experimental setup. (b) Photograph of the spectral sensor with the incorporated efficiency enhancement module and the entrance aperture at the front side adapted to the CCD camera. The highly compact module is integrated completely inside the mount of the camera.

Download Full Size | PDF

Figure 6(b) shows a photo of the spectral sensor with the incorporated efficiency enhancement module and the entrance aperture on the front side. Although the setup is only intended to serve as a proof-of-concept, the configuration offers an integrated and highly compact system that fits almost entirely within the mount of the CCD camera.

From the variety of spectral lines provided by the HgCd lamp, specific lines have been selected for experimental verification by different filters. In particular, all wavelengths shorter than 498 nm are directed to a light trap by the first dichroic filter and do not reach any of the filter arrays. The prominent Cd-line at 508 nm is reflected toward the first filter array but blocked by the array’s stop band.

Subsequently, the intense green line at 546 nm is captured by the first filter array. The following yellow double line at 577 nm and 579 nm reaches the second filter array, and the red line at 644 nm is then detected by the third filter array. For this measurement, no additional external filters have been used, and all three filter arrays have been addressed simultaneously. Figure 7(a) shows a section of the directly recorded signal from the CCD camera. In particular, the three separated filter arrays are clearly visible, and in each of the filter arrays a number of bright pixels appear with different intensities. Although each filter array has been only exposed by narrow spectral lines, the detector’s response to several related filter pitches simultaneously can be explained by the small separation between the peak wavelength of the individual filter pitches and the extension of their spectral bandwidths. For a more detailed view, Figs. 7(b)–7(d) show 3D representations of the lateral intensity distributions of the three individual filter arrays. The indicated wavelengths at the top of each diagram represent the spectral line or spectral double line incident on each of the three filter arrays. The arrows in each diagram are labeling the specific filter pitch showing the maximum transmitted intensity, and the associated numbers are quantitative measures for the detected relative intensity (gray scale ranges from 0 to 255 units). The particular peak pattern for each array results from a superposition of the incident spectral lines and the transmission curves of the filter pitches distributed across the filter array. For example, the intensity pattern shown in Fig. 7(c) is caused by the superposition of the filter pitch distribution shown in Fig. 3(c) and the incident spectral double line at 577 nm and 579 nm. In the bottom row of the intensity pattern, three filters are active, namely two 579 nm filters and a 578 nm filter, which are separated by a single, completely different filter pitch in between. A further relevant filter element is found in the row above, near the right edge of the array, which has been designed for a maximum transmittance of 580 nm. The filter array, that detects the 644 nm spectral line, responds explicitly to the specific corresponding active filter shown in Fig. 7(b). Figure 7(d) shows the final implemented filter array, which is illuminated with a 546 nm line. Here, due to the respective filter array characteristics, three filters are active (central wavelength of the filters 546, 547, 547 nm). The appearance of two identical filters (547 nm) is caused by production issues.

 

Fig. 7. Measurement results. (a) Section of the recorded CCD camera image when the efficiency module is illuminated with three spectral lines or double lines, respectively. The separated filter arrays are clearly visible. (b)–(d) Detailed view of the lateral intensity distributions of the three individual filter arrays in a 3D representation. Due to the characteristics of the filter pitches of the array several elements transmit the incident light. (b) Representation for the 644 nm line, (c) 577–579 nm double line, and (d) 546 nm line.

Download Full Size | PDF

In the intensity patterns presented in Figs. 7(b)–7(d) the displayed peaks clearly correspond to the respective filters of the array. A comparison of the maximum pixel intensity peaks of the three arrays (see corresponding arrow indications) leads to a ratio of 19/35/226 (normalized: 1/1.84/11.9), which approximately correlates to the ratio of the tabulated intensity values for these lines [i.e., 1(0.9)/2/6 for wavelengths 577(579)/644/546 nm] [31]. Any deviation may be attributed to the variation of the maximum transmission efficiencies of the individual filter pitches that ranges in maximum between 50% and 97% with an average value of 70% [24]. Residual divergence of the incoming light could also influence the results. This may also affect the measurement of the maximum intensity at the 580 nm filter element.

 

Fig. 8. Measured data for all three spectral lines or double lines and for both spectral detection configurations. Displayed is the measured intensity as a function of the exposure time for all three wavelengths. The efficiency enhancement method shows an average efficiency increase of a factor of $ {\gt} {4}$ compared to the reference system.

Download Full Size | PDF

To quantitatively compare the detection efficiency of the efficiency enhancement method with that of the reference system, the same measurement conditions have been employed for both configurations. More specifically, for each measurement, only a single spectral line has been selected from the HgCd lamp by appropriate bandpass filters. The used filters are metal interference filters from Carl Zeiss Jena (GDR), with center wavelengths of 650, 575, and 550 nm, respectively, and a FHWM of 7–11 nm. This restriction to single lines minimizes probable disturbing effects such as cross talk and also allows optimization of the signal parameters of the selected spectral line to CCD electronics without a potential overexposure at different lines. The measurements for each spectral line started with the efficiency-enhanced setup. The initial exposure time was adjusted to a level that provided an electronic signal close to the maximum of the detectable scale (gray-scale value 255) for the most active filter pixel. Five additional measurements with a step-like decay of the exposure times were obtained for better statistics and to prove the working principle also for shorter exposure times. Thereafter, all measurements using the same parameters were repeated with the reference setup. This procedure was completed for the three selected spectral lines or double lines of the HgCd spectrum. Due to the spectral bandwidth of each single filter element, the 577 nm and 579 nm double line contributes simultaneously to several filter elements. For the evaluated intensity we took the filter element showing the maximum intensity (most active filter element). That means the detected signal is a mixture of both wavelengths.

Figure 8 shows the measured data for all three spectral lines or double lines and for both spectral detection configurations. Each of the three diagrams displays the data comparison for one spectral line that is correlated to a specific filter array. As expected, each of the curves reveals a linear dependency of the measured intensity from the exposure time. For each filter array, respectively spectral line, pair-by-pair comparisons of the measured intensity values for the efficiency-enhanced setup and reference system have been conducted. As a result, average intensity ratios (= intensity measured with the efficiency-enhanced setup divided by intensity of the reference system) of 4.1 for the 546 nm line, 4.3 for the 577/579 nm double line, and 4.5 for the 644 nm line have been found. An overall average ratio of 4.3 clearly demonstrates the enhanced efficiency characteristics of the proposed approach compared to the reference system. A ratio of 4.3 is somewhat higher than the value of 3.0, which is expected as a theoretical efficiency difference between the efficiency-enhanced setup and the reference system. The deviation may be attributed to several factors in both configurations, such as the differences in the adjustment of the optical components, the transmission properties of the filter arrays, and/or the reflection properties of the optics (e.g., properties of the first deflection mirror).

5. CONCLUSIONS

We present the basic concept and the experimental implementation of a simple method to significantly increase the detection efficiency of filter array-based spectral sensors. Essentially, the concept is based on a wavelength-dependent spatial redistribution of the incident light before it reaches the filter elements. Explicitly, we demonstrate the working principle by developing a compact opto-mechanical setup comprising several dichroic elements to create the separated partial spectra. In the proof-of-concept, a significant efficiency increase has been achieved compared to a reference system without wavelength-selective redistribution. Although the demonstrated efficiency enhancement is remarkable, there is still a substantial potential for further improvement. For example, an increased number of dichroic filters could be used in the filter cascade, which would offer a finer subdivision of the partial spectra. A precise adaption of these larger number of partial spectra to tailored filter arrays with adjusted spectral characteristics to the dichroic filters will certainly lead to a further increase of the detection efficiency. Following the successful proof-of-concept, we will target in future work on the detection of broadband spectra in different applications. This will also allow assessment of the limitations, e.g., the boundaries of the partial spectra, which are responsible for wavelength separation quality. Furthermore, the concept is not limited to the FP filter arrays demonstrated in our experiments but can easily be transferred to any type of filter array, such as plasmonic or simple absorptive color filters.