An EMD-Based Filtering Algorithm for the Fiber-Optic SPR Sensor Volume 8, Number 3, June 2016 Tao Wang Tiegen Liu Kun Liu Junfeng Jiang Lin Yu Meng Xue Yunxia Meng
DOI: 10.1109/JPHOT.2016.2564439 1943-0655 Ó 2016 IEEE
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
An EMD-Based Filtering Algorithm for the Fiber-Optic SPR Sensor Tao Wang,1,2 Tiegen Liu,1,2 Kun Liu,1,2 Junfeng Jiang,1,2 Lin Yu,1,2 Meng Xue,1,2 and Yunxia Meng3 1
College of Precision Instrument and Optoelectronics Engineering, Tianjin University, Tianjin 300072, China 2 Key Laboratory of Optoelectronics Information Technology, Tianjin University, Ministry of Education, Tianjin 300072, China 3 North Automatic Control Technology Institute, Taiyuan 030006, China DOI: 10.1109/JPHOT.2016.2564439 1943-0655 Ó 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Manuscript received April 5, 2016; revised April 25, 2016; accepted May 4, 2016. Date of publication May 6, 2016; date of current version May 18, 2016. This work was supported by the National Natural Science Foundation of China under Grant 61108070, Grant 61227011, Grant 61378043, and Grant 61475114; by the National Instruments Program under Grant 2013YQ030915; by the National Basic Research Program of China under Grant 2010CB327802; by the Tianjin Natural Science Foundation under Grant 13JCYBJC16200; by the Science and Technology Key Project of Chinese Ministry of Education under Grant 313038; and by the Shenzhen Science, Technology, and Innovation Commission under Grant JCYJ20120831153904083. T. Wang and T. Liu contributed equally to this work. Corresponding authors: K. Liu and J. F. Jiang (e-mail:
[email protected];
[email protected]).
Abstract: We present a filtering algorithm based on empirical mode decomposition (EMD) for the data demodulation of the fiber-optic surface plasmon resonance (SPR) sensor. After data preprocessing by the EMD-based filtering algorithm, three methods are compared to obtain the valley wavelength of the SPR. The results show that by using the Gaussian-fitting method, the smallest standard deviation is obtained. Based on the proposed algorithm, the related temperature sensor based on fiber-optic SPR gets the sensitivity of 335.7 pm/°C at 20 °C and 626.9 pm/°C at 99 °C, resulting in temperature resolutions of 0.06 °C at 20 °C and 0.03 °C at 99 °C via the spectrometer resolution of 20 pm. Index Terms: Surface plasmon resonance (SPR), empirical mode decomposition (EMD), temperature sensor.
1. Introduction Over the past decade, surface plasmon resonance (SPR) has been used in a wide range of physical, chemical, and biological sensing applications [1]–[4]. Surface plasmons are the charge density oscillations at the metal-dielectric interface resulting in the production of surface plasmon wave which propagates along the interface [5]. For the excitation of surface plasmons, either a prism, a grating or an optical fiber is used. Since first proposed by Jorgenson and Yee [6], the optical fiber based SPR sensor has drawn much attention due to its attractive advantages of compactness, high sensitivity, and real-time remote sensing capability over the prism and the grating based SPR sensors. Diverse configurations have been implemented to make fiber optic SPR sensors, such as D-shape, cladding-off, fiber tip, and tapered fiber structures [7]. Most current optical fiber based SPR sensors employ visible light source near 650 nm for excitation of SPR. This wavelength is not suitable for optical fiber communication [8], leading to multimode transmission in fiber. The
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
mode distribution of light in the fiber is very sensitive to mechanical disturbances and the disturbances occurring close to the sensing area of the fiber may cause mode coupling and mode noise [5]. In order to obtain the desired result, spectral data with noise of the fiber optic SPR sensor should be carefully handled. On the other hand, Huang et al. [9] proposed the empirical mode decomposition (EMD) technique, which is a data driven signal method, suitable for analyzing non-stationary and nonlinear signals. The EMD method decomposes the data into a series of intrinsic mode functions, which provide high frequency components and low frequency components. For the data acquired from the SPR sensors, the frequency of the noise is quite high while the frequency of the effective signal is quite low, without frequency aliasing. Therefore, it is feasible to utilize EMD method to process the SPR sensor data. Optical fiber based SPR sensors have been widely used in the field of chemical and biological sensing applications for their high sensitivity. Water is the widely used solvent in the chemical and biological field, with thermo-optic coefficient of about 1:0 104 C1 in the visible region [10]. It is important to study the effect of temperature on the water solvent by using fiber-optic SPR sensor. In this paper, we study the water temperature sensor based on fiber-optic SPR. We propose an effective filtering algorithm, with the IMFs decomposed by EMD as filter banks for noise reduction preprocessing. Finally, three methods for extracting the resonance wavelength of the SPR spectrum are compared.
2. Theory The refractive index of water varies with temperature, thus we can monitor the water temperature by detecting its refractive index. The relationship of refractive index and temperature for water is given by the paper [11]. The schematic of fiber-optic SPR sensor is shown in Fig. 1. It can be used to measure the temperature of water, based on a transmission structure. The SPR sensor can be modeled based on the principle of attenuated total reflection (ATR) with Kretschmann’s configuration [12], [13]. In the proposed optical fiber based SPR sensor, the sensing part consisting of a fiber core-gold layer-sensing medium is shown in Fig. 1. When the incident light enters the fiber core and undergoes total internal reflection at the interface between fiber core and the metal film, a part of the light continues to propagate in the form of an evanescent wave and excites oscillation of the electrons confined in the gold film to form a surface plasmon wave (SPW). Considering a system of three superposed layers, layer 1 corresponds to a dielectric medium (optical fiber) with refractive index n0 (n 20 ¼ "0 , where "0 is the dielectric constant of fiber core), layer 2 is a metallic layer (gold layer) characterized by a dielectric function "m , and layer 3 responds to a dielectric medium (sensing medium) with a dielectric constant "s . The classical SPR coupling condition is satisfied by equaling the tangential component of the wave-vector kx of the incident light propagated in layer 1 with the propagation constant sp of the SPW, presenting at the layer 2/layer 3 interface. If the incident beam is propagated in layer 1 with an incident angle with respect to the normal of the interface and is the vacuum wavelength ( ¼ 2c=!, where c and ! are light speed in vacuum and angle frequency, respectively), then the coupling condition can be written as ffi !rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffi ! "m "s sp ¼ kx ) (1) ¼ "0 sin: c c "m þ "s The value of the dielectric function "0 , "m and "s were defined previously. All of them are related to the incident wavelength . Equation (1) will be satisfied when the reaches a special value, thus the surface plasmon resonance will occur. The wavelength is the resonance wavelength, which is very sensitive to the refractive index of the sensing medium on metal surface. In the optical fiber based SPR sensing setup, the relationship between the effective transmitted power and several other parameters can be written as [14] R 2 Nð;l;dÞ PðÞd R Ptrans ¼ cr R (2) 2 PðÞd cr
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
Fig. 1. Schematic of optical-fiber-based SPR sensor.
where R is the function of ; d ; l; "0 ðÞ; "s ðÞ, and "m ðÞ, which is the light reflectivity after once reflection on the fiber-gold interface and can be calculated using multilayer thin film reflectance theory [14], [15]. Nð; l; d Þ is the number of reflections at the gold surface inside the fiber, which is the function of the angle , the interaction length l and the core diameter d . PðÞd ¼ ½"s sincos=ð1 "s cos2 Þ2 d is the power arriving at the fiber-end between the angle and pffiffiffiffiffi pffiffiffiffiffi þ d . cr ¼ arcsinð "s = "0 Þ is the critical angle of the fiber. The refractive index of medium varies with the change of the temperature. After the temperature of the sensing medium changes, the resonance wavelength of the SPR sensor will shift. Since the thermo-optic coefficient of silica and gold is much smaller than most sensing medium, we can ignore the refractive index change of silica and gold when the temperature changes, then the resonance wavelength changes only caused by the temperature change of the sensing medium. Thus, we will obtain the transmitted power based on the temperature Ptrans ¼ f ð; d ; l; "0 ðÞ; "m ðÞ; "s ð; T ÞÞ:
(3)
In this paper, because water is the most important solvent in chemical and biological sensing applications areas, we choose water as the sensing medium to fabricate the fiber-optic SPR sensor.
3. System and Sensor Manufacture Plastic clad silica (PCS) multimode optical fiber from Scitlion with numerical aperture of 0.37 and core diameter of 600 m, and Gold (Au) wire (99.999% pure) and chromium (Cr) (99.99% pure) purchased from ZhongNuo Advanced Material were used to manufacture the SPR sensor. To fabricate the SPR probe, a 13 cm PCS fiber was used, with 10 mm unclad in the middle. The unclad portion was first cleaned by acetone and ethanol successively in the ultrasonic cleaner. Then chromium layer and gold film with a thickness of 4 nm and 50 nm, respectively, were coated on the circumferential surface of the unclad portion by thermal evaporation method with a vacuum deposition system [16]. The coated fiber was used for studying the temperature sensing characteristic of water. The schematic configuration of the sensor is shown in Fig. 2(a). It mainly consists of the coated fiber with its coated part totally sealed in a capillary tube. The capillary tube is filled with deionized water and sealed with UV glue at both ends. The picture of the fabricated sensor is depicted in Fig. 2(b). In Fig. 3, the schematic diagram of the water temperature sensing system based on the fiberoptic SPR probe is presented. The system consists of five parts, a tungsten halogen light source, a SPR probe, a spectrophotometer (model HR 4000, Ocean Optics), an electric-heated thermostatic bath, and a platinum resistance thermometer. The electric-heated thermostatic bath, whose temperature range is from 0 °C to 99 °C (the local water boiling temperature), is used to change the water temperature in the bath. The platinum resistance thermometer is applied to monitor the real-time temperature of the water. Light from the polychromatic source (a tungsten halogen light source with spectra of 360–2000 nm and input power of 100 W) is injected into the multimode fiber and passes through the SPR probe, which is placed in the electric-heated thermostatic bath. Then, the transmitted light is measured by the spectrophotometer with a detecting range of 200–1100 nm and resolution of 20 pm. In the Fig. 3, the black and green lines are the optical fiber and electrical lines, respectively.
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
Fig. 2. Optical-fiber-based SPR sensor. (a) Schematic configuration of the sensor. (b) Photograph of the sensor.
Fig. 3. Schematic of fiber SPR temperature sensing system.
4. Experiment and Analysis In the experiment, we adjusted the electric-heated thermostatic bath to heat the water and recorded the transmitted spectrum when the temperature reached stabilization. The temperature was changed from 20 °C to 99 °C with the interval of 5 °C. The electric-heated thermostatic bath vibrates a lot when it works, leading to much noise of the acquired SPR spectrum. Therefore, it is necessary to smooth the data before locate the resonance wavelength. When the spectrum was measured, the integration time of the spectrophotometer was set to be 50 ms. One of the recorded spectra at 60 °C is shown in Fig. 4(a). Frequency domain filtering and wavelet transform are two common method for data filtering. The conventional frequency domain filtering just removes components whose frequency is above the cut-off frequency, it may remove the useful signal in the certain frequency. The wavelet analysis need the basic wavelet to process the data, so the wavelet is not an adaptive method. Empirical mode decomposition (EMD) is an adaptive and data dependent method. In this paper, EMD together with Frequency domain filtering method is used to process the SPR spectrum. EMD decomposes a nonlinear and non-stationary signal into a set of amplitude and frequency modulated (AM-FM) signal components H n , called intrinsic mode functions (IMFs), and a final residual trend r, which is a nonzero-mean low-order polynomial component [17] X ¼
n X
Hn þ r :
(4)
i¼1
The spectrum data in Fig. 4(a) is decomposed into multiple IMFs and a residue, as shown in Fig. 5. Curves from H 1 to H 7 are IMFs, and the curve r is the residue. The noise of the SPR spectrum mainly comes from light source, spectrophotometer and mode jitter. Noise from light noise and spectrophotometer shot noise mainly contribute to IMF H 1 and IMF H 2 . IMF H 3 and IMF H 4 result from mode jitter in multimode fiber, as well as IMFs from H 5 to H 7 and the residue r come from the practical resonance signal. Because IMF H 1 and IMF H 2 have no relationship with the useful resonance signal, they can be subtracted from X, and we process the IMFs H 3
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
Fig. 4. Normalized transmitted optical-fiber-based SPR spectrum at 60 °C. (a) Original spectrum. (b) Filtered result with the EMD method.
Fig. 5. Intrinsic mode functions and residue decomposed by EMD.
and H 4 with low-pass filter, leading to effective signals H 03 and H 04 . Finally, we can reconstruct the data X’ with H 03 , H 04 , H 5 , H 6 , H 7 , and r. X 0 ¼ H30 þ H40 þ
7 X
Hi þ r :
(5)
i¼5
Using the above filtering process, we can obtain the filtered results of the original spectrum data in Fig. 4(a), and the filtered result based on EMD method is shown in Fig. 4(b).
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
Fig. 6. Conventional SPR spectrum for centroid algorithm and the inverted SPR spectrum for Gaussian fitting method. The abscissa of the geometric center under the baseline in the top curve is the resonance wavelength in the centroid method, and the symmetry axis of the Gaussian fitting curve above the threshold in the bottom curve is the resonance wavelength in the Gaussian fitting method.
After preprocessing the data with EMD method, we can easily obtain the resonance wavelength from the filtered data. Three methods are used to calculate the resonance wavelength. They are minimum method, centroid method and Gaussian fitting method. The minimum method is to locate the position corresponding to the minimum reflectivity on the filtered data. The wavelength corresponding to the minimum reflectivity is the resonance wavelength. The minimum method is the easiest method to obtain the resonance wavelength, but it is sensitive to wavelength shift or noise. The centroid method [18] for determining the resonance wavelength of an SPR-modulated spectrum is a simple one which finds the geometric center of the resonance curve. Although the geometric center is not exactly located at spectral position of the minimum reflectivity (the resonant dip is typically asymmetrical), most SPR applications involve relative measurements. Therefore, the offset of the geometric center does not affect the final measurements. A baseline value is chosen to cross the conventional SPR curve, as shown the above curve in Fig. 6. Gaussian fitting method is based on the hypothesis that the curve of the SPR spectrum below the baseline is symmetrical. We inverted the conventional SPR spectrum and fit the data above the threshold with Gaussian function, as shown in Fig. 6. The performances of the minimum method, the centroid method and Gaussian fitting method were respectively examined with 50 groups of experiment data under every temperature. The standard deviation of 50 groups of experiment data for three method is obtained in Table I. It shows that Gaussian fitting method has the smallest standard deviation. Fig. 7 shows the SPR resonance wavelength corresponding to the certain temperature, in which the results of three methods are fitted by cubic curve. It is obvious that the resonance wavelength changes a lot with the change of temperature. The trend of the fitting curves of three methods is similar and the results of three methods are nonlinear, because the thermo-optic coefficient of water for the certain wavelength in the range of 20–100 °C is not constant. We get the sensitivity of the sensor after carrying on the first derivative of the fitting curve of the Gaussian fitting method. It costs about 210 ms to filter a frame of data with the modified EMD method and to obtain the resonance wavelength with the Gaussian fitting method. The sensitivity at 20 °C can reach 0.3357 nm/°C, and the sensitivity at 99 °C can reach 0.6269 nm/°C. If the optical resolution of the spectrometer reaches 20 pm (the minimum resolution of HR 4000), the temperature
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor TABLE 1
Standard Deviation of 50 Groups of Experiment Data for Three Methods (Unit: nm)
Fig. 7. Temperature versus corresponding resonance wavelength.
resolution of the sensor can reach 0.03 °C at 99 °C and 0.06 °C at 20 °C. In addition, we carried on the directly Gaussian fitting for the experiment data, leading to temperature resolution of 0.03 °C at 99 °C and 0.09 °C at 20 °C, the standard deviation of 50 groups of experiment data for directly Gaussian fitting method is also shown in Table 1. It shows that Gaussian fitting method with EMD filtering gets better result than the directly Gaussian fitting method.
5. Conclusion This paper presents a filtering algorithm based on EMD for the spectrum data of fiber-optic SPR sensor. Besides, three methods, which are Gaussian fitting method, centroid method and minimum method, are used to acquire the resonance wavelength from the preprocessed spectrum data by EMD. The result shows that the Gaussian fitting method has the smallest standard deviation. At the same time, when the sensing medium is water, the effect of temperature on the SPR sensors is studied. In the range of 20–100 °C, the resonance wavelength changes a lot when the temperature changes. Therefore, it is necessary to control the temperature when
Vol. 8, No. 3, June 2016
6803008
IEEE Photonics Journal
EMD-Based Filtering Algorithm for SPR Sensor
the fiber optic SPR sensor is used in chemical and biological sensing application. When the optical resolution of the spectrometer is 20 pm, the temperature resolution of the sensor can reach 0.03 °C at 99 °C and 0.06 °C at 20 °C.
References [1] Y. Zhao, Z.-Q. Deng, and Q. Wang, “Fiber optic SPR sensor for liquid concentration measurement,” Sens. Actuators B, Chem., vol. 192, pp. 229–233, Mar. 2014. [2] R. Verma and B. D. Gupta, “Fiber optic SPR sensor for the detection of 3-pyridinecarboxamide (vitamin B3) using molecularly imprinted hydrogel,” Sens. Actuators B, Chem., vol. 177, pp. 279–285, Feb. 2013. [3] D. Li et al., “A method to detect the mixed petrol interface by refractive index measurement with a fiber-optic SPR sensor,” IEEE Sensors J., vol. 14, no. 10, pp. 3701–3707, Oct. 2014. [4] J. Homola, “Present and future of surface plasmon resonance biosensors,” Anal. Bioanal. Chem., vol. 377, no. 3, pp. 528–539, Oct. 2003. [5] J. Homola, S. S. Yee, and G. Gauglitz, “Surface plasmon resonance sensors: Review,” Sens. Actuators B, Chem., vol. 54, no. 1, pp. 3–15, Jan. 1999. [6] R. Jorgenson and S. Yee, “A fiber-optic chemical sensor based on surface plasmon resonance,” Sens. Actuators B, Chem., vol. 12, no. 3, pp. 213–220, Apr. 1993. [7] B. Lee, S. Roh, and J. Park, “Current status of micro-and nano-structured optical fiber sensors,” Opt. Fiber Technol., vol. 15, no. 3, pp. 209–221, Jun. 2009. [8] X. Zhao et al., “Improvement of the sensitivity of the surface plasmon resonance sensors based on multi-layer modulation techniques,” Opt. Commun., vol. 335, pp. 32–36, Jan. 2015. [9] N. E. Huang et al., “The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis,” Proc. Roy. Soc. London A, Math., Phys. Eng. Sci., vol. 454, no. 1971, pp. 903–995, Mar. 1998. [10] D. Lide, CRC Handbook of Chemistry and Physics 2004–2005: A Ready-Reference Book of Chemical and Physical Data. Boca Raton, FL, USA: CRC, 2004. [11] P. Schiebener et al., “Refractive index of water and steam as function of wavelength, temperature and density,” J. Phys. Chem. Ref. Data, vol. 19, no. 3, pp. 677–717, 1990. [12] R. KretschmannE, Radiative Decay of Non Radiative Surface Plasmons Excited by Light, vol. 23A. Berlin, Germany: Z. Naturforsch, 1968, pp. 2135–2136. [13] B. Gupta and R. Verma, “Surface plasmon resonance-based fiber optic sensors: Principle, probe designs, and some applications,” J. Sens., vol. 2009, pp. 1–12, 2009. [14] A. K. Sharma and B. Gupta, “Absorption-based fiber optic surface plasmon resonance sensor: A theoretical evaluation,” Sens. Actuators B, Chem., vol. 100, no. 3, pp. 423–431, May 2004. [15] M. Niggemann et al., “Intrinsic fiber optical gas sensor based on surface plasmon resonance spectroscopy,” in Proc. Int. Soc. Opt. Photon. Eur. Symp. Opt. Environ. Public Safety, 1995, vol. 2508, pp. 303–311. [16] Y. Zhao, Z.-Q. Deng, and H.-F. Hu, “Fiber-optic SPR sensor for temperature measurement,” IEEE Trans. Instrum. Meas., vol. 64, no. 11, pp. 3099–3104, Nov. 2015. [17] M. K. Saxena et al., “Empirical mode decomposition based dynamic error correction in SS covered 62.5/125 m optical fiber based distributed temperature sensor,” Opt. Laser Technol., vol. 67, pp. 107–118, 2015. [18] K. Kukanskis et al., “Detection of DNA hybridization using the TISPR-1 surface plasmon resonance biosensor,” Anal. Biochem., vol. 274, no. 1, pp. 7–17, 1999.
Vol. 8, No. 3, June 2016
6803008
