Abstract
We present a miniaturized particle sensor collecting scattered light in two solid angle intervals by Fresnel ring lenses. The particle size is determined from the ratio of both scattering amplitudes (intensity ratio) in addition to a linear diversity combining technique, generating a 3D particle size matrix that reduces the ambiguity by the index of refraction on the particle size identification. A signal-to-noise ratio of 30.3 was achieved for 147 nm sized polystyrene latex particles. Measurements of polydisperse particle size distribution show good agreement with the results by a scanning mobility particle sizer.
© 2014 Optical Society of America
1. Introduction
Laser-based detection of particles in air has made significant progress since its first commercial applications decades ago [1]. In recent years, new sensor techniques have been developed, e.g., whispering gallery mode resonators and ring resonators, pushing the detection limit to particles with sizes below 10 nm [2,3]. To monitor the particle concentration in everyday environments more efficiently, miniaturization and mobility of the sensors are obligatory. Complex setups are less suitable for this application, whereas most available systems are capable of delivering particle size sensitivity only above 0.25 μm, e.g., the Particle Measuring Systems Handilaz and the Grimm EDM 107, or are quite complex, e.g., photoacoustic sensor systems [4–6].
Particles with diameters below 2.5 μm (PM2.5) or even below 0.1 μm are the most interesting, as smaller particles are expected to exhibit a higher health risk [7,8].
In this paper we describe an optical particle sensor (OPS) that uses laser light scattering and whose structural parts are completely composed of plastics. It has high accuracy and sensitivity down to particles with diameters of 147 nm and an expected particle size limit near 90 nm. The measurement technique is based on an intensity ratio approach, which allows continuous operation of the system without repeated calibration. To further enhance the reliability, linear diversity combining (LDC) is applied, which uses the mean-free sum of both scattering signals analog to the intensity ratio approach. This combined detection scheme additionally allows a further reduction of the well-known uncertainty of the particle size identification by the index of refraction.
Section 2 explains this approach and the setup in more detail before Section 3 shows the results of measurements.
2. Approach
A. Optical Setup: Fresnel Ring Lenses and Laser Beam Shaping
The light scattering intensity ratio is a well-established technique to reliably sense particles when crossing a laser beam and determine its particle size [1]. In most applications, single detectors or detector arrays are used to collect the scattered light at two separate solid angle sections. The particle size can then be derived from the ratio of both amplitudes using either calibration results or scattering theory, e.g., Mie or Rayleigh theory.
The intensity of the collected scattered light can in general be described by
where denotes the number of scattering particles in the laser beam, is the distance between the scattering particle and the detector, is the incident laser light intensity, is the collected scattered light power, is the area of the detector, and is a scattering function, e.g., covered by Mie or Rayleigh theory, depending on the scattering angle , the particle diameter , the laser wavelength , and the index of refraction of the particle . Hence, the amount of scattered light strongly depends on the incident laser light as well as the detector area and the distance between the scattering particle and detectors.To increase the sensitivity, one can collect a higher amount of scattered light using mirrors or lenses [9]. As spherical mirrors show the problem of image distortion, and elliptical mirrors only allow collection of scattered light from a small spatial fraction of the laser beam without image distortion, we use Fresnel lenses. In comparison to aspherical glass lenses, this kind of lens has the great advantage of low -numbers and can thus be positioned close to the source of the light scattering while collecting a major fraction of the scattered light.
In our setup we place the Fresnel lenses in the forward scattering direction () as larger particles scatter the major fraction of light at low angles.
We use three Fresnel lenses on a single lens mount to collect two solid angle intervals: the first is positioned toward the laser beam source and spans around the laser beam trap, collecting and collimating the scattered light (see Figs. 1 and 2). Behind this lens, two Fresnel ring lenses are combined. The smaller ring spans around the back part of the laser beam light trap, and the second ring spans around the smaller ring as its inner diameter equals the outer diameter of the smaller ring lens. Transparent adhesive is used to fix the position of the lenses and seal the lenses and the lens mount gas-tight. To support two separate angle intervals, the collected and collimated scattered light needs to be imaged onto two separate avalanche photodiodes (APDs). For this purpose, the Fresnel ring lenses are cut eccentrically, which means an off-center cut (see Fig. 1). The effective detector area of the Fresnel lenses covering a solid angle interval can be described in general by
where denotes the longitudinal distance between the scattering particle and the first Fresnel lens. The detection area of the solid angle intervals by both Fresnel ring lenses should be as large as possible and equal.Figure 2 shows a scheme of the sensor setup. The sensor system comprises three mounts, manufactured using rapid prototyping: the cover including the inlet nozzle, the measurement chamber with a mount for the laser optics and the APDs, and the mount for the Fresnel lenses. The laser optics consists of two aspherical lenses forming a well-defined laser beam for particle scattering with a small beam waist. The inlet nozzle directs the aerosol stream through the laser beam, and the position of the laser beam waist is aligned with the center of the aerosol stream.
The imaging of the scattered light onto the APDs can be described using geometrical optics. The image distance is given by
where denotes the distance between the first and the second Fresnel lenses and the focal length of lenses 1 (front) and 2 (back). The lateral distance of the image in the direction to the axis then results from the magnification and the distance of the off-center cut to the center of the original Fresnel lenses (chosen as 3.3 mm; see Fig. 1). It follows thatHence, the position and size of the resulting image on the detectors directly relate to the position of the particle in the laser beam. If the eccentrical cut is large, is large too, resulting in a high imaging angle, which can possibly result in a partial blocking of the scattered light by the light trap. Therefore, should be as small as possible. Ideally, it is only limited by the minimum distance of the detectors. In conflict, the imaging distance should be as low as possible to achieve a low magnification of the image to minimize the photodetector area necessary to collect the full image. The detector noise increases with increasing photodetector area, thus reducing the achievable minimum signal-to-noise ratio (SNR).
The inner and outer diameters of the Fresnel ring lenses are constant, thus leading to certain scattering angle intervals collected by both Fresnel ring lenses. As the scattering position varies, the collected scattering angles vary as well (see Figs. 3 and 4). As a result, the detector areas vary as well [see Eq. (2) and Fig. 5]. Therefore, the scattering amplitudes can vary for the same particle sizes, depending on the scattering position. Ideally, the effective detection areas should be constant. Comparing the ratios of the actual detection areas at around the laser beam waist (), the ratios of the according detection areas differ by about 4%, inducing an equal error on the uniqueness of the measurement results. To limit these errors, the spatially accepted range around the laser beam waist should be as small as possible. Considering Eq. (3) and the relation
the collected scattering angles defined by the inner and outer diameters of the Fresnel lenses can be calculated by where and denotes the radii of the Fresnel ring lenses (, ); the focal lengths are 15.2 mm.As a result of using Fresnel ring lenses, the radial distance in Eq. (1) depends on the cosine of the scattering angle.
The laser beam is highly focused to maximize the amplitudes of the scattered light. Its beam intensity reduces with increasing distance to the laser beam waist as illustrated in Fig. 6, leading to a variation of the amplitudes of the scattered light depending on the scattering position. This dependency can be compensated by identifying the scattering position and correcting the amplitudes accordingly. This is carried out by determining the length of the scattering signals, termed time of flight (TOF) from here onwards. The diameter of the Gaussian laser beam is defined at the lateral position where the signal has dropped to , hence 13.5% of the amplitude. The intensity of a radial laser beam profile is generally given by
where denotes the laser output power, is the laser beam radius at the beam waist, is the radius at the longitudinal distance to the laser beam waist, and is the radial distance to the center (peak intensity). A transient scattering signal shows a Gaussian shape similar to Eq. (6) and can be written as where denotes the measured transient signal with the time and the specific width of the Gaussian signal where the TOF is defined as .The TOF is derived from the integral of the scattering signal. Integrating and solving Eq. (7) for a certain spatial trajectory in leads to
where denotes the scattering amplitude and erf the error function. Depending on the scattering position in , the amplitudes and TOF vary according to the laser beam intensity profile. Equation (8) cannot be solved analytically. We therefore use the regula falsi method to determine the solution numerically for each scattering signal. From measurements we found this technique works quite well with an error if the SNR is at least 10. For applications where scattering amplitudes with can be neglected in the processing step, e.g., when only large particles occur, this method works optimal. Small particles show lower SNR and thus lead to a larger error for the determination of its TOFs.However, multiple scattering positions can lead to the same TOFs although the laser beam intensity is different (see Fig. 7). This would ultimately lead to a wrong correction of the scattering amplitudes by the method described above. To identify the scattering position uniquely, the lateral parts of the laser beam were blocked by a slit-shaped aperture at the laser optics between the first and second lenses. Thus only the center part of the laser beam is left (see Fig. 7 top). This technique, however, leads to an interference pattern of the laser beam intensity (Fresnel diffraction) instead of a clear top-hat profile. In far-field measurements of the laser beam profile using a WinCamD-UCD15 we found a variation of the average peak intensity of about . In the transverse direction, the laser beam still shows a Gaussian-like profile (see Fig. 7 bottom).
Depending on the scattering position, the image can exceed the photodetector area. Hence, the spatial range of acceptance needs to be limited by accepting only an according TOF interval. On the other hand, it is very useful to accept a large spatial range to increase the speed of the acquisition of a particle size distribution. In total, if the particle concentration is high or the measurement time only plays a minor role, a smaller interval should be chosen.
The position of the laser beam waist needs to match with the center of the radial-symmetric aerosol stream. Figure 8 shows a simulation of the velocity field using Comsol Multiphysics. From the center to a lateral position of from the laser beam waist, the velocity reduces to half of its peak value ( peak at a flow rate of ). Thus the TOFs increase proportionally with increasing distance to the laser beam waist. Therefore, the correction of the scattering amplitudes has to be derived from calibration measurements with a monodisperse particle size.
B. Signal Processing: Intensity Ratio and Linear Diversity Combining
Besides the well-known intensity ratio technique, we use a LDC technique, which is known from wireless communication to offer the maximal SNR for multiple signal channels [10]. In our sensor system, a LDC equals the mean-free sum of the scattering signals of both angle intervals and delivers a maximal SNR. The average white Gaussian noise (AWGN) of the scattering signals increases by , whereas the LDC signal equals the sum of the SNR of both APDs.
The identification of the scattering amplitudes and TOFs uses four steps. First, the cross covariance of both scattering signals is generated and only signals with amplitudes exceeding are accepted for further processing.
In a second step, the LDC signal is analyzed for symmetric Gaussian scattering events and the corresponding amplitudes and TOFs are determined. Only highly symmetric scattering events are accepted. This reduces the problem of coincident particles as these would likely lead to nonsymmetric scattering events. There are two exceptions, however: when coincident particles cross the laser beam at the very same time, and if a large and a small particle cross coincidently, where the scattering signal by the large particle dominates and the small particle only leads to a small fluctuation of the total scattering signal.
In a third step, the scattering amplitude of each event is determined in both scattering signals. To increase the SNR, the noise on the signals is reduced by a Savitzky–Golay polynomial filter of order 2 and a window length of 7, when the number of data points of an event exceeding is larger than the filter window length.
In a fourth step, the optimal scattering amplitudes of both angle intervals and the amplitude ratios are determined. The accuracy of the evaluated amplitudes directly relates to its SNR, and the LDC amplitude equals the sum of both amplitudes. Therefore, the difference between the LDC amplitude (maximal SNR) and the amplitude with the higher SNR of both scattering angles equals the remaining amplitude. Using this approach, it is possible to detect scattering amplitudes with on one solid angle interval, which would normally not be accepted.
Then, the particle size is identified using a 3D matrix.
C. Particle Size Identification: 3D Matrix
Multiple particle sizes and indices of refraction can lead to the same scattering amplitudes. This ambiguity is a major problem of the intensity ratio approach. In the past, it was treated, e.g., by using an additional laser wavelength, polarization, or additional scattering information [11–13].
In a more straightforward approach to support a unique particle size identification, we include the LDC amplitude as another axis besides the amplitude ratio. This leads to a 3D matrix with the particle size on the axis. The 3D matrix is simulated using Mie theory and the BHmie algorithm [14], supporting an arbitrary number of indices of refraction and particle sizes. For example, when measurements take place in a well-confined environment where only one certain particle type is present, the simulation is carried out for that single index of refraction. Otherwise, multiple indices of refraction and other intervals of particle sizes can be considered, depending on the purpose of the measurement and the environment. Then, for a measured pair of amplitude ratio and LDC amplitude, the next neighbor in the 3D matrix is chosen as the particle size (see Fig. 9).
Figure 9 shows an excerpt of a simulated matrix assuming five equally distant indices of refraction from 1.3 (water) to 1.7 and 98 equally distant particle sizes in the interval of 75…2500 nm. The matrix uses the LDC amplitudes as the axis, the amplitude ratio as the axis, and the corresponding particle diameters as the axis.
There is one flaw, however. Multiple combinations of particle size and index of refraction can lead to the same LDC amplitudes and amplitude ratio in the simulation. Then, two (or more) different particle sizes can have the same LDC amplitude/amplitude ratio combination in the matrix. It is not possible to determine which particle size is correct. In this case, the average of the multiple particle sizes is taken as the result.
There are three matrix positions in the above-mentioned matrix where multiple (two) particle sizes were found as possible solutions each. These particle sizes differ by . Hence, the total error induced on the particle size determination remains small.
Variations of the laser beam output power, the detector sensitivities, and pollution of the Fresnel lenses directly influence the scattering amplitudes. This problem can be reduced by exploiting the signal offset generated by the background light. The variations can then be compensated according to changes of the signal offset.
The reliability of the matrix method strongly depends on the certainty of the LDC amplitudes. As the laser beam profile shows a significant variation and a steady slope at its edges (see Fig. 7 top), an error remains. As the amplitude ratio is not affected by this variation and is therefore more reliable, it should be weighed significantly stronger when determining a particle size. We found a weighing factor of 4 to be optimal.
3. Measurements
A scheme of the measurement setup is shown in Fig. 10. The particles were generated from a liquid suspension (ultra pure ) by dry and the aerosol stream dried by a diffusion dryer before entering the SMPS+C reference measurement system (Grimm 5.403) and the OPS. The system uses a laser diode emitting at 450 nm with 33 mW output power (Osram 450B), APDs of Hamamatsu (S8664–20K), a transimpedance amplifier with by Femto (DHPCA-100), an analog–digital converter (ADC) with 16 bit resolution in and (ADwin-light-16), and aspherical Fresnel lenses. The laser beam with a divergence half angle of 8° generated a laser beam waist of (calculated). It was shaped by a slit-shaped aperture with 0.3 mm slit width. A flow rate of was used for measurements, and only TOFs with were accepted. 37 μs was chosen to guarantee at least three sampling points. Comsol simulations showed a velocity of (center/lateral); hence the spatial range of acceptance was about around the laser beam waist where the laser beam diameter varied between and . An optical loss of 20% was assumed in the simulations considering the Fresnel lenses, which have an opaqueness of 92%. The distance between the laser beam waist and the first Fresnel lens was 11.5 mm, leading to average scattering angle intervals of about 29.5‥37.0° and 15.6‥29.4°, and effective detector areas of 95 and after a partial blockage by the lens mount.
Figure 11 shows simulated LDC amplitudes for particles with an index of refraction of 1.3 using multiple particle sizes and laser beam diameters. A minimum of three sampling points are needed for processing; hence the minimum TOF is 37 μs at or 10 μs at , respectively, which equal certain minimal measurable laser beam diameters, depending on the aerosol velocity. The smallest particle can be detected at the laser beam waist (highest laser beam intensity), for which a minimum sampling rate of and a flow rate is needed, 55 μm laser beam diameter equals the minimum at flow rate and , and 90 μm equals the minimum at flow rate and . Hence, the minimum detectable polystyrene latex (PSL) particle size varies accordingly between near 90 nm ( sampling, flow rate) and about 200 nm ( sampling, flow rate) for a minimum and LDC noise of 1.4 mVrms. However, this is spatially limited, hence providing only a very small efficiency for small particle sizes. Larger particles or particles with higher indices of refraction scatter high amounts of light distant to the beam waist as well and are thus more efficiently detected. Figure 12 shows the measured SNR of PSL and silica particles. In measurements with PSL particles with 147 nm diameter and a flow rate of we found a SNR of 30.3.
The amplitude ratios using Mie theory are shown in Fig. 13 for PSL particles and compared with measurement results using monodisperse particle suspensions with a flow rate of . There is a high agreement between simulation and measurement results with an average relative deviation of 8%.
Ahead of operation, the particle size dependent detection efficiency of the sensor system was calibrated using the SMPS reference system and PSL particles. The result can be seen in Fig. 14. The minimal detectable particle concentration decreases with increasing particle size. For example, the minimal particle concentration for 600 nm sized particles needs to be at least , whereas particles with 1800 nm in diameter only require a concentration of about .
The correction factors for the scattering amplitudes using the TOFs were determined from measurements using monodisperse PSL with 296 nm. We found an accuracy of the particle size determination of 10%–20%.
The performance of the sensor system including the LDC and matrix method at flow rate was compared with SMPS measurements using a polydisperse solution containing PSL () with sizes of 147, 296, 496, 707, and 903 nm and silica () with sizes 490, 730, and 990 nm (see Fig. 15). About 10,000 scattering events were detected on average per minute, and only scattering amplitudes with were accepted for evaluation. In total, the results of the OPS and the SMPS are in good agreement and the size accuracy of the OPS shows to be , depending on the particle size. Concentration drops, e.g., at 400 nm, were not as well resolved as by the SMPS system and show room for improvement. Particles with diameters of and show the largest deviation, which can be addressed to the deviation between the realistic and the assumed indices of refraction.
The result of measurements using DEHS particles (Di-2-Ethylhexyl-Sebacat, ) are shown in Fig. 16. The results also agree well with the SMPS reference measurement. However, a concentration drop can be observed in the OPS result at very small particle sizes. This can be addressed to the required minimum SNR of 10 in the processing, as described in Section 2.A.
4. Conclusion and Outlook
We present a miniaturized sensor system with an enhanced intensity ratio approach using a LDC technique. The sensor system is capable of detecting particles with sizes of at least 147 nm with a SNR of 30.3. It offers a reduced dependence on the index of refraction for the correct particle size identification using a 3D matrix method. To magnify the sensitivity, we use Fresnel ring lenses to increase the effective detection area. As the structural components consist of plastic material, it seems feasible to produce the basic system using injection molding, thus minimizing the costs of the system. To further improve the performance of the present proof-of-concept OPS, an ADC with a higher sampling rate is highly beneficial, allowing a more precise identification of the TOFs and improved signal filtering, thus offering a lower particle size detection limit.
If the size of the sensor system is not as important as for this work, additional methods can be used to minimize the interference effects on the laser beam intensity by the laser beam aperture, e.g., using an inverse Gaussian filter.
References
1. D. Black, M. McQuay, and M. Bonin, “Laser-based techniques for particle-size measurement: a review of sizing methods and their industrial applications,” Progr. Energy Combust. Sci. 22, 267–306 (1996). [CrossRef]
2. L. He, S. K. Özdemir, J. Zhu, W. Kim, and L. Yang, “Detecting single viruses and nanoparticles using whispering gallery microlasers,” Nat. Nanotechnol. 6, 428–432 (2011). [CrossRef]
3. J. Zhu, S. K. Ozdemir, Y.-F. Xiao, L. Li, L. He, D.-R. Chen, and L. Yang, “On-chip single nanoparticle detection and sizing by mode splitting in an ultrahigh-Q microresonator,” Nat. Photonics 4, 46–49 (2009). [CrossRef]
4. A. D. Clarke, N. C. Ahlquist, S. Howell, and K. Moore, “A miniature optical particle counter for in situ aircraft aerosol research,” J. Atmos. Ocean. Technol. 19, 1557–1566 (2002). [CrossRef]
5. H. A. Beck, R. Niessner, and C. Haisch, “Development and characterization of a mobile photoacoustic sensor for on-line soot emission monitoring in diesel exhaust gas,” Anal. Bioanal. Chem. 375, 1136–1143 (2003). [CrossRef]
6. C. Liang, H. Huang, B. Ren, and Y. Zhao, “A miniaturized laser-diode-based optical sensor for laser particle counter,” Proc. SPIE 5634, 375–380 (2005). [CrossRef]
7. R. Burnett, J. Brook, T. Dann, C. Delocla, O. Philips, S. Cakmak, R. Vincent, M. Goldberg, and D. Krewski, “Association between particulate- and gas-phase components of urban air pollution and daily mortality in eight Canadian cities,” Inhalation Toxicol. 12, 15–39 (2000).
8. R. Harrison and J. Yin, “Particulate matter in the atmosphere: which particle properties are important for its effects on health?” Sci. Total Environ. 249, 85–101 (2000). [CrossRef]
9. K. Tsutsui, K. Koya, and T. Kato, “An investigation of continuous-angle laser light scattering,” Rev. Sci. Instrum. 69, 3482–3786 (1998). [CrossRef]
10. D. G. Brennan, “Linear diversity combining techniques,” Proc. IEEE 91, 331–356 (2003). [CrossRef]
11. F. Zhao, Z. Gong, H. Hu, M. Tanaka, and T. Hayasaka, “Simultaneous determination of the aerosol complex index of refraction and size distribution from scattering measurements of polarized light,” Appl. Opt. 36, 7992–8001 (1997). [CrossRef]
12. V. V. Berdnik and V. A. Loiko, “Retrieval of size and refractive index of spherical particles by multiangle light scattering: neural network method application,” Appl. Opt. 48, 6178–6187 (2009). [CrossRef]
13. M. Kocifaj and M. Držík, “Retrieving the size distribution of microparticles by scanning the diffraction halo with a mobile ring-gap detector,” J. Aerosol Sci. 28, 797–804 (1997). [CrossRef]
14. C. F. Bohren and D. R. Huffman, Absorption and Scattering of Light by Small Particles (Wiley, 1983).