1. INTRODUCTION

Developments in spectroscopy have been constantly driven by the intent to use the available light more effectively, together with the ambitious aims to permanently reduce the size of the used devices. Today filter-based spectrometers [1–14] play a leading role in miniaturization [1,2], while maintaining a high resolution. The following resolution values have been achieved in the visible spectral range: 7 nm for a single nanowire [3], 2–3 nm for quantum dots [4], 1.7–5 nm for Fabry–Perot filter arrays [5], and 0.7 nm for a linear variable filter (LVF) [6], to name a few. To work in the near-infrared wavelength region, which is of great importance for biological and chemical sensing [7], a variety of filter types and on-chip systems have been reported, for example, plasmonic filter arrays [8], Fabry–Perot filter arrays [9,10], LVFs [11,12], and organic photodetectors [13,14]. The developments of such systems are strongly motivated by a growing demand for handheld devices providing point-of-use measurements of, e.g., textiles, foods, and pharmaceuticals [15]. Several filter-based handheld spectrometers are already commercially available. For example, both of the battery-powered modules MicroNIR OnSite-W (VIAVI Solutions) [15] and trinamiX (trinamiX GmbH) [16] employ an LVF as a dispersive element, which covers spectral ranges of 950–1650 and 1400–2500 nm, respectively. A different type of portable spectrometer called SCiO is introduced by Consumer Physics [15]. It is based on bandpass filters connected with a photodiode array (12 pixels) and detects in a wavelength region of 740–1070 nm. Another compact module employing 16 narrowband organic photodetectors is commercialized by Senorics (wavelength range of 1200–1700 nm) [14]. Despite the benefits of filter-based spectrometers, one drawback remains: their low detection efficiency. A fundamental reason for this is that the narrowband transmission of each filter element implies strong broadband reflection. Consequently, when the incident broadband light hits the narrowband filters, only a small fraction of the total light passes through them, while almost all the light is reflected and hence not used for detection.

To improve the efficiency of filter-based spectrometers, the spectral preselection concept has been proposed [17]. The principle is as follows: to achieve efficiency enhancement, the incoming broadband light can be divided spectrally and spatially into partial ray bundles before reaching the related filters or filter areas. The previous studies [17,18] demonstrated the implementation of this concept using dichroic beam splitters as additional spectral preselection optics. However, such a method has significant limitations. First, the integration of additional dichroic beam splitters is costly and complex; thus, the advantage of such an improvement is reduced. Secondly, there are difficulties in further miniaturizing the system and reaching a significantly larger efficiency increase factor. Moreover, efficiency-enhanced acquisition of the continuous spectrum [18] is not possible for filters such as LVFs in combination with a single line detector.

In this paper, we present a completely different approach to increase the efficiency of filter-based spectrometers, overcoming the main limitations of previous studies and omitting beam splitters. The demonstrated spectroscopic modules possess a minimum number of mechanical parts and consist only of a solid assembly that simultaneously plays the role of a filter, spectral preselection mechanism, and light guide. The implementation of the module can be achieved with a simple technique using multiple reflections between the filters and the parallel mirror separated by air. Moreover, such a design does not change the conventional detection procedure used for filter spectrometers, that is, a straightforward reading of the intensity distribution behind discrete filters (or LVFs). Thus, using the proposed concept, a line detector or a row of standalone photodiodes can be used instead of an area detector. In comparison with other similar filter systems that utilize standard parallel illumination of the entire filter array (or LVF dispersive direction), this study experimentally demonstrated the possibility of reaching an efficiency increase factor of up to approximately 100.

2. BASIC CONCEPT AND IMPLEMENTED MODULES

A. Detection Principle Based on Multiple Reflections

The introduced concept enables efficient detection of an entire broadband light bundle that is significantly smaller than the dispersive area or line of a given filter system. The filter system can effectively record all addressed spectral information of this narrow and intensive light bundle by applying an appropriate pattern of successive transmissions and reflections. In the following, the basic methodology and two fabricated optical modules are presented.

Figure 1 shows the working principle of the optical module. The central part of this figure demonstrates the simplest case (first option) by using discrete filters. The broadband “white” light to be analyzed, which originates, for example, from an optical fiber output or focus point, is collimated and incident on the first filter (indicated as number 1) at an angle $\alpha$ as a parallel light bundle with a diameter $a$. Light of a certain central wavelength and spectral bandwidth is transmitted through the first filter and used for detection. All residual wavelengths that are not transmitted are reflected and directed to the mirror placed parallel to the filter surface at a distance $b$. After reflection from the mirror, the ray bundle hits the next filter (number 2); again, a light bundle of a certain bandwidth, but with a different central wavelength, is transmitted and detected. The process continues until all broadband light is successively and completely detected after transmitting through all filters from 1 to ${N}$. For this described first option, the number of detectable spectral channels is equal to the number of reflections on the filter plane. Note that the number of detectable spectral channels can be greater than the number of reflections (second option). This can be achieved if all filter elements illuminated by the consecutively reflected ray bundle not only consist of one single filter but are composed, e.g., of an array of different filters. The incoming ray bundle illuminates in such a way that there are several discrete filter combinations for every reflection step. Similarly, a continuous spectrum can be detected by using an LVF instead of a discrete filter. In this case, the numbers 1 to ${N}$ indicate the area numbers of the LVF, where successive transmissions and reflections occur. To capture the addressed continuous spectral region properly, all consecutive light bundles should cover the entire dispersive direction of the LVF without any gaps or overlap, that is, when neighboring light bundles are connected with each other. The condition for such an arrangement (continuous spectrum detection) is the following geometrical relation between the angle $\alpha$, ray bundle diameter $a$, and distance $b$:

 

Fig. 1. Basic concept: a collimated broadband light bundle with a diameter $a$ hits the first filter at an incident angle $\alpha$, and a narrow spectral part of this light is transmitted and detected. The reflected light bundle hits the parallel mirror located at a distance $b$ and is directed to the next filter. This process is repeated until all filters or filter parts from 1 to ${N}$ are addressed. The number of detectable spectral channels can be either equal, as indicated by “=” (1. option, for every reflection step only one discrete filter is addressed) or greater than the number of reflections, as indicated by “${\gt}$” (2. option). The second option can be achieved if discrete filters are replaced by filter arrays (not illustrated) or LVFs.

Download Full Size | PDF

B. Simplest Experimental Realization: Module Based on Discrete Edge Filters

A row of different long-pass dichroic filters was used for the simplest demonstration of the concept (1. option shown in Fig. 1). To assemble the module, the row of dichroic filters and a parallel-placed silver mirror were separated via a spacer with a defined thickness, as shown by the photograph of the module in Fig. 2(a) and exploded view of the three-dimensional (3D) model in Fig. 2(b). As the spacer, a clearance (cut with a ${{\rm CO}_2}$ laser) in the middle of a fused silica glass substrate served as free space, where multiple reflections occurred. For the available glass spacer thickness of 1.15 mm ($\pm {0.02}\;{\rm mm}$), the width of the filters was chosen as 2.3 mm ($\pm {0.05}\;{\rm mm}$). The required filter size was obtained using a wafer saw. The angle of incidence for these dichroic long-pass filters was defined by the manufacturer (Semrock, Chroma Technology) as ${45}\;{\pm}\;{1.5}^\circ$. A comprehensive list of the 10 filters used in this study, along with their parameters, is presented in Table S1 (see Supplement 1). The used mirror was a sputtered, reflection-enhanced silver mirror (Chroma Technology) with an average reflectance of ${\ge}{98}\%$ in the wavelength range of 380–1100 nm. The size of the module was ${30.2} \times {7} \times {3.2}\;{\rm mm}$.

 

Fig. 2. (a) Photograph of the module. Ten dichroic filters are clearly visible. For size comparison, a millimeter ruler is shown. (b) 3D representation of the module. The glass spacer with a clearance sets a precise distance between the mirror and filters. (c) 3D representation of the module, line camera, and closed housing (half-section).

Download Full Size | PDF

The module was connected to a line camera (Eureca Messtechnik GmbH, USB camera board e9u-LSMD-TCD1304-STD) and mounted in a housing, as shown by the 3D representation in Fig. 2(c). The line camera had a Toshiba TCD1304 sensor with 3648 pixels and a pixel size of ${8} \times {200}\;{\unicode{x00B5}{\rm m}}$. For a first test of the module, a collimated halogen lamp (Thorlabs QTH10/M, electrical power of the bulb 10 W) was used. The light was delivered by a multimode optical fiber (NA 0.22, core diameter 50 µm) with two reflective collimators. The focal lengths of the two collimators were 50.8 and 7 mm for fiber input and output, respectively. Based on the parameters of the optical fiber and collimators the corresponding light bundle diameter delivered to the optical module was approximately 3 mm.

To prove the working principle of the module, several theoretical aspects and experimental data should be discussed. Figure 3(a) shows the transmittance data of the dichroic long-pass filters provided by the manufacturer, indicated by their edge wavelength (transmittance ${T} = {50}\%$). The 10 long-pass filters had the following edge wavelengths: 520, 560, 580, 605, 620, 648, 685, 720, 757, and 776 nm. The folded beam path (that is, the arrangement of multiple reflections) began with the long-pass filter characterized by the longest edge wavelength followed by consecutive filters with lower edge wavelengths to capture an entire broad wavelength region in one shot. A bandpass filter with a transmittance region of 500–800 nm was used to set the working region of this module, as shown in Fig. 3(a) by the dashed line. (The filter was placed into the incoming beam path.) The portion of the initial ray bundle which transmits the first long-pass filter has a spectral curve determined by the edge wavelength of the first long-pass filter (776 nm) and the long-wavelength limit of the bandpass filter at 798 nm. All further spectral transmittance curves for the following filters can be obtained in the same manner. By simply using the transmittance or reflectance data of the dichroic filters, mirror, and bandpass filter, as well as the spectrum of the used halogen lamp and detector response, the spectral characteristics of all 10 transmitted light bundles can be calculated [see Fig. 3(b)]. The resulting spectra of the ray bundles have bandpass characteristics with a full width at half-maximum (FWHM) in the range of 15 (at 790 nm) to 36 nm (at 736 nm). The spectral characteristics of these partial spectra are listed in Table 1.

 

Fig. 3. (a) Measured transmittance curves of the 10 dichroic long-pass filters (colored curves) and one bandpass filter (black dashed curve) from representative production lots. (b) Calculated spectral curves of the 10 ray bundles (colored curves) created by the folded beam path architecture (see also Table 1), detector response curve (blue dashed curve), and spectrum of the halogen lamp (red dotted curve). (c) Experimental intensity values measured by the module. The module illuminated by the halogen lamp is represented by the black curve, and the measured reflectance spectrum of the natural green leaf is presented by the green curve.

Download Full Size | PDF

Table 1. Characteristics of the Calculated Spectral Curves of the 10 Ray Bundles

View Table

The experimental spectrum of the halogen lamp (acquisition time 0.68 ms), appearing as 10 intensity peaks of light bundles, is presented in Fig. 3(c). For this measurement, the slit width (variable slit Thorlabs VA100C/M) was adjusted to approximately 1 mm ($\pm {0.05}\;{\rm mm}$). The slit width limits the entrance beam diameter and thus determines the peak width on the detector. In its turn, the maximum gray value of each peak is determined by the transmitted light intensity and the spectral width of each bandpass. Thus, by comparing the maximum gray values of the peaks shown in Fig. 3(c) with the integrated intensity values of the theoretical curves shown in Fig. 3(b), a good correlation between the values was found. For example, the theoretical curves number 3, 4, and 5 presented in Fig. 3(b) have similar heights, but clearly different FWHMs, which correspond with the alternating maximum gray values of the experimental curves with the same numbers [see Fig. 3(c)].

To demonstrate a simple application related measurement using the module, the green curve in Fig. 3(c) shows the detected reflectance spectrum of a natural green leaf. The typical leaf spectrum was recognized. (One peak in the green region and a strong increase at approximately 700 nm can be found.) With chemometric methods, such a simple optical system even can be successfully used as a spectrometer because of its high efficiency. Assuming neglectable scattering and absorption effects, all the light that is not transmitted by every consecutive filter is further directed to the next filter. The input bundle for such a system can be narrow and have high light intensity.

C. LVF Module for Efficient Continuous Spectrum Acquisition

To demonstrate continuous spectrum acquisition, the setup was modified by simply replacing the row of dichroic filters [see Fig. 2(b)] with an LVF (LF103252, Delta Optical Thin Film A/S [19]). The LVF had the following parameters: a substrate thickness of 2 mm, detectable spectral range between 450 and 850 nm within a substrate length of 24.7 mm, average transmission values between 70 % (450 nm) and 90% (850 nm), FWHM of 4% at the central wavelength, and blocking level OD4 (wavelength range of 200–1100 nm). The size of the module was ${30.1} \times {7} \times {6}\;{\rm mm}$, and the following module parameters were chosen based on Eq. (1): a necessary input beam diameter of approximately 2 mm ($\pm {0.05}\;{\rm mm}$), a distance of 3.0 mm ($\pm {0.02}\;{\rm mm}$) between the LVF and the mirror (that is, the thickness of the glass spacer with a clearance), and an angle of incidence of 18°. Figure 4 shows a photograph of the module (without detector) illuminated with a halogen lamp where, under an angle of incidence of 18°, the visible part of a continuous spectrum can be recognized. The folded beam path began at the long-wavelength side of the LVF. For the next step, an illumination of separate narrowband spectral lines was applied to calibrate the LVF-based module by using metal interference filters. The optical parameters of several of these filters are listed in Table S2 (see Supplement 1). The different filters were placed one by one in the light beam path of the halogen lamp between the module and mirror collimator. The transmittance spectra of all 16 filters used (central lines from 450 to 825 nm in 25 nm steps) are shown in Fig. S1 in Supplement 1. The central wavelengths were defined using Gaussian fitting functions. A fitted curve revealing the dependence between the wavelengths and pixel values of the LVF is shown in Fig. S2 (see Supplement 1). Subsequently, several continuous spectra were recorded.

 

Fig. 4. Photograph of the LVF-based module illuminated with the halogen lamp. The visible part of the continuous spectrum of the halogen lamp can be recognized. The size of the module is ${30.1} \times {7} \times {6}\;{\rm mm}$.

Download Full Size | PDF

Figure 5(a) shows a diffuse raw spectrum of the sun (dark orange curve) and a reflectance spectrum of the natural green leaf (green curve). The spectrum of the sun was not measured directly but the back-reflected light off white paper was detected for several reasons. First, direct sunlight was too intense, even with a minimal line camera exposure time of 0.01 ms. Secondly, the reflectance of the white paper is nearly constant over the entire addressed spectral region (450–825 nm); consequently, the scattered sunlight is considered as a reference spectrum. Thirdly, the subsequently analyzed specimen, for example, a natural green leaf, could be placed at the identical distance from the mirror collimator as the white paper. Thus, diffuse spectra of the sun and different reflective specimens could be recorded consecutively under the same conditions. Raw spectra presented in Fig. 5(a) are results of the combination of the initial spectra with a detector response [see the blue dashed curve in Fig. 3(b)] and transmittance of the LVF at different longitudinal positions. These raw spectral curves possess a continuous course without reaching zero intensity for the entire addressed wavelength region (450–825 nm). There are only a few narrow and steep decreased positions visible in the raw spectrum [see in Fig. 5(a)] when two neighboring ray bundles are supposed to be connected with each other. These decreases can be explained most probably by misalignment or diffraction effects at the apertures. The decreased positions are clearly visible at the beginning of the folded beam path at the long-wavelength side of the spectral curve shown in Fig. 5(a) and almost disappear by the end, because of the continuous broadening of the light bundle owing to its residual divergence.

 

Fig. 5. (a) Reference raw spectrum of the sun scattered off white paper and raw reflectance spectrum of the natural green leaf measured using the LVF-based module. (b) Calculated reflectance spectra of different natural leaves and a reflectance spectrum of human skin (palm). Photographs of the measured leaves and skin (palm) are presented at the top left of the figure. The smoothed part of the palm reflectance spectrum (moving average) with two indicating arrows shows a characteristic absorption of oxy-hemoglobin (two spectral dips).

Download Full Size | PDF

In the following, several additional application related measurements were successively performed and recorded. Figure 5(b) shows the calculated reflectance spectra of different natural leaves (green, yellow, and dark) and human skin (palm). The spectra were obtained by dividing the raw spectrum of the reflective specimen (leaf, palm) by that of the light source (sunlight). For example, the calculated curve of the natural green leaf [see Fig. 5(b)] was obtained by dividing the two raw spectra shown in Fig. 5(a). By comparing the obtained spectra with the results of other studies, a good correlation can be found (for example, see Fig. 1 of [20]). The resulting spectra of natural leaves, especially healthy green and dark leaves, were characterized by a strong increase (called red-edge) between $\sim{680}$ and 710 nm with a further high reflectance value in the near-infrared region explained by the high chlorophyll content. The spectral peak of the natural green leaf defining its green color was found at approximately 550 nm. The spectrum of the human palm, as the place with high blood flow, revealed an absorption characteristic of oxy-hemoglobin, found as two dips between 525 and 580 nm, as shown in separate parts of the spectral curve (moving average) with two indicating arrows in Fig. 5(b). These measurements (leaves and palm) were obtained with an acquisition time of 100 ms. To additionally demonstrate the ability of the system to measure narrowband light sources, a measurement of a Mercury–Argon spectral lamp was performed (see Fig. S3 in Supplement 1).

3. ESTIMATION OF ACHIEVED EFFICIENCY ENHANCEMENT

The obtained efficiency enhancement of the presented concept can be estimated as the ratio between the illumination intensities of light using folded beam path architecture and a common design. Typically, the common case composes the illumination of the entire area of the filter system with a circular beam of light to be analyzed. A typical example of this traditional (inefficient) illumination principle of the filter system is presented in Fig. 1 of study [12] where, in order to illuminate the entire LVF for measuring liquid absorption and emission spectrum, a large expanding optics was applied. In contrast, the proposed method based on folded beam path architecture works significantly more efficient. It enables the use of a narrower input aperture in comparison to the dispersive area to be illuminated; consequently, the entire light to be analyzed becomes much more concentrated to each filter or filter area. Because the intensity of the light is inversely proportional to the illuminated area, an efficiency increase factor of up to ${{N}^2}$ can be achieved when using, for example, a fiber collimator with a collimated beam diameter that is ${N}$ times smaller than is necessary for the common case. The presented LVF-based module using an input light bundle of approximately 2 mm for the illumination of the almost entire LVF (nearly 24 mm of the 24.7 mm was used, ${N} = {12}$) demonstrated a theoretical efficiency increase factor of ${{N}^2} = {144}$. In this study, we used a collimated light bundle of 3 mm cut by a slit down to 2 mm, which decreased the factor to 64 for this particular case. The reflection losses caused during the multiple-reflection process (10 reflections for the first module and 12 for the second) by the filter and mirror did not exceed 20%. By simply using the reflection data of the filters and mirror, the losses of the last reflected light bundles can be easily calculated (not shown). The fact that the presented LVF module would also work with, for example, a 20° angle of incidence and a 1.5 mm light bundle diameter (see Eq. 1), in such a way increasing the theoretical factor up to ${{N}^2} = {256}$, makes discussions on the efficiency-enhanced factor relatively flexible. With suitable collimation optics, an efficiency enhancement factor of up to approximately 100 compared to the mentioned common approach appears to be achievable.

4. CONCLUSION

Simple and novel efficiency-enhanced filter-based spectral modules based on the principle of multiple reflections were demonstrated. Having no additional spectral preselection optics, these two optical modules operated considerably more efficiently than systems with a conventional approach for filter illumination. The folded beam path design offers an excellent compromise between substantial efficiency increase and effort to change the established illumination principle of optical filters. Moreover, when using such a design, the overall size of the filter-based spectrometer cannot be moderately increased (see studies [17,18]), but even can be reduced, because the proposed architecture enables the use of a short-focus small-diameter collimation optics. Consequently, such systems can be extremely flat, an appreciated benefit for most spectroscopic applications, including airborne monitoring. As was shown with the first module, a straightforward but efficient spectrometer can be realized, even with affordable edge filters. The second module based on a typical LVF was able to capture continuous spectra with an efficiency enhancement factor of up to approximately 100, compared to a conventional system with an upright illumination principle. To prove the working principle, diffuse sunlight measurements and reflectance spectra of natural leaves and human skin were acquired. Furthermore, these measurements obtained a continuous spectrum across the entire spectral region (450–825 nm) using only an affordable line detector, instead of a previously necessary area detector combined with rotational arrangement of preselection optics [18].

These modules served as a proof-of-concept and were accomplished using only commercially available filters. Other wavelength regions, specific resolution and spectral range, quality of the blocking level, and dimensions of the filters can be freely designed and optimized for application-tailored tasks. Other filter types, such as Fabry–Perot filter arrays, would also benefit from the folded beam path approach if an appropriate arrangement is implemented. Future work aims to develop customized filter arrays and LVFs with a high spectral resolution and wavelength range, improved arrangement of the folded beam path design, and collimation optics.