Laser Frequency Noise Cancelation in a Phase-Shifted ... - IEEE Xplore

Laser Frequency Noise Cancelation in a Phase-Shifted Fiber Bragg Grating Ultrasonic Sensor System Using a Reference Grating Channel Volume 8, Number 1, February 2016 Lingling Hu Guigen Liu Yupeng Zhu Xiangyu Luo Ming Han

DOI: 10.1109/JPHOT.2016.2527018 1943-0655 Ó 2016 IEEE

IEEE Photonics Journal

Noise Cancelation in an FBG Sensor System

Laser Frequency Noise Cancelation in a Phase-Shifted Fiber Bragg Grating Ultrasonic Sensor System Using a Reference Grating Channel Lingling Hu, Guigen Liu, Yupeng Zhu, Xiangyu Luo, and Ming Han Department of Electrical and Computer Engineering, University of Nebraska-Lincoln, Lincoln, NE 68588 USA DOI: 10.1109/JPHOT.2016.2527018 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 November 26, 2015; revised January 27, 2016; accepted February 3, 2016. Date of publication February 8, 2016; date of current version February 18, 2016. This work was supported by the U.S. Office of Naval Research (ONR) under Grant N000141310159, Grant N000141410139, and Grant N000141410456. Corresponding author: M. Han (e-mail: [email protected]).

Abstract: In an ultrasonic sensor system based on phase-shifted fiber Bragg gratings ðFBGsÞ, a cost-effective way for sensor demodulation is to set the laser wavelength at the center of the steep spectral slope of the FBG refection spectrum. Frequency noise of the laser source ultimately limits the signal-to-noise ratio (SNR) of the sensor system. Here, we demonstrate a noise cancelation method based on a pair of FBGs, which is capable of improving the SNR of system beyond the limitation set by the laser frequency noise. In this method, one of the gratings is used as the sensor to detect the acoustic signal, and the other grating, isolated from the signal, is used as a reference that gives the noise information. Employing a data postprocessing method, the noise is subtracted from the original detected signal. We show that the SNR of the sensor system can be improved by up to 20 dB using this method. Index Terms: Sensors, fiber gratings, diode lasers.

Ultrasonic sensors are being used in many applications such as non-destructive evaluation, structure health monitoring, and biomedical imaging [1], [2]. In recent years, fiber-optic ultrasonic sensors, as an alternative to traditional piezoelectric sensors, have come into sight due to their many advantages such as small size, light weight, and immunity to electromagnetic interference. In particular, ultrasonic sensors based on fiber Bragg gratings (FBGs) have additional advantages, including in-line fiber structure, ease for multiplexing, and single-ended operation, that make them more attractive in many applications [3]–[6]. In these FBG systems, ultrasonic detection is achieved by detecting the ultrasound-induced shift of the FBG reflection spectrum. One of the most cost-effective ways to detect the spectrum shift is to use a coherent light source whose wavelength is set on the spectral slope of the reflection spectrum of the FBG. The spectral shift caused by ultrasonic signals is converted to intensity variations of the reflected laser light that can be conveniently detected by a photodetector. A concern for an FBG ultrasonic sensor system in practical applications is the limitation of the signal-to-noise ratio (SNR). The signal strength, which is proportional to the slope of the reflection spectrum, can be increased by the use of -phase shift fiber Bragg gratings ðFBGsÞ as the sensing element [7]–[10]. A

Vol. 8, No. 1, February 2016

6801108



Fig. 1. (a) FBG ultrasonic sensor system. Cir: circulator; Amp.: amplifier. (b) Laser noises on the slope.

FBG features a phase shift in the middle of an otherwise periodic grating structure, which leads to a strong optical resonance and a narrow notch in the reflection spectrum with much larger spectral slope than a regular FBG of the same length. However, accompanying with the enhanced sensitivity is the equally increased noise originating from the laser frequency noise because the laser frequency fluctuations are indistinguishable from the grating spectral shifts. Although a high performance laser source with small frequency noise can be used to improve the SNR of the sensor system, it inevitably induces budget burdens in practical applications. Pound-Drever-Hall technique [11], [12] can also be used to suppress the laser frequency noise by locking the laser to a wavelength reference. In addition to the increased complexity of the system, it requires a wavelength tuning capability of the laser source over the ultrasonic frequency range of interest, which is difficult to achieve. Therefore, it is of great interest to suppress the frequency noise while using a low-cost laser source. Laser intensity noise can be removed by simultaneously monitoring the difference between reflection and transmission of the phase shifted FBG using a balanced photodetector [9]. The balanced detection requires a loop operation of the fiber-optic sensor system, which may be less attractive in practice. Moreover, balanced detection cannot improve the SNR related to laser frequency noise. As shown later, at the sharp slope of FBG, the frequency noise becomes the dominant noise source of the sensor system. Therefore, novel cost-effective methods for frequency noise reduction are of great interest for FBG ultrasonic sensor system. In this paper, we propose a noise cancelation method based on a pair of FBGs that can efficiently reduce the intensity noise and the frequency noise from the laser source. Our results show that the SNR of a FBG sensor system based on a low-cost distributed feedback (DFB) laser is improved by up to 20 dB by using this method. We start with a brief SNR analysis of the FBG sensor acoustic detection system [Fig. 1(a)] [13]. We only consider the noise contributions of the laser source and ignore the thermal noise and shot noise of the photodetector. The signal and noise hence refer to the current from the photodetector. Assuming an ultrasonic wave of frequency causes a sinusoidal frequency shift of the grating, the corresponding optical response can be written as Pref ¼ P0 þ P0 k v cost

(1)

where P0 (W) is the dc optical power determined by the operating point of the laser on the spectral slope, v (Hz) is the amplitude of the grating frequency shift induced by the ultrasonic wave, k (Hz−1) is the slope of the normalized grating spectrum, and t denotes time. The sinusoidal term in Eq. (1) contains the information of the ultrasonic wave; then the power of the signal is expressed as 1 {2S ¼ R2 P02 k 2 v 2 2

(2)

where R (A/W) is the responsivity of the photodetector. The noise contribution from the laser includes the laser relative-intensity-noise (RIN) and the laser frequency noise. Assuming the RIN is white in the detection bandwidth f , the noise


6801108



Fig. 2. (a) Proposed ultrasonic sensor system with noise cancelation. Cir: circulator; Amp.: amplifier. (b) Schematic diagram of the noise cancelation method.

power related to RIN is given by {2RIN ¼ R2 P02 SRIN f

(3)

where SRIN (1/Hz) is the single-sided power spectral density (PSD) of the RIN. When the laser is locked on the spectral slope that functions as a frequency discriminator, the fluctuations in the laser frequency result in fluctuations of the optical power, as shown in Fig. 1(b). The noise amplitude related to this laser frequency noise is proportional to the frequency slope k , and its power is given by {2frq ¼ R2 P02 k 2 Sfrq f

(4)

where Sfrq (Hz2/Hz) is the single-sided PSD of the laser frequency noise. It is assumed that Sfrq is white over f . Combing (2)–(4), the SNR of the system is given by SNR ¼

iS2 2 þ i2 iRIN frq

¼

k 2 v 2 : 2ðSRIN þ k 2 Sfrq Þf

(5)

In the limit of very large frequency slope, the laser frequency noise contribution dominates, and the maximum SNR is given by SNRmax ¼

v 2 : 2Sfrq f

(6)

This maximum SNR is fundamentally limited by the laser frequency noise and is independent on the frequency slope of the sensor [14], [15]. The proposed noise cancelation method can improve the SNR of a FBG ultrasonic sensor system beyond the limitation set by the laser frequency noise. This significant SNR improvement is achieved by actively measuring the laser frequency noise using a reference FBG channel and subsequently subtracting the noise from the signal of the sensing channel. More specifically, as shown in Fig. 2(b), in the sensing channel, a DFB laser is locked to the spectral slope of a FBG sensor and in the reference channel, a matching FBG is locked to the laser. When ultrasonic signal impinges on the sensor FBG, the optical signal carries information about the acoustic wave as well as the laser RIN and laser frequency noise. The reference FBG,


6801108



Fig. 3. (a) System setup. Osc.: oscilloscope; Cir.: circulator; LPF: low-pass filter; BPF: bandpass filter; PD: photodetector; PC: polarization controller; Attn.: attenuator; Amp.: amplifier. (b) Resonance dips of the two identical FBGs.

which is not subject to the ultrasonic wave being detected, is configured to produce the same noise as that from the sensor FBG. Since the detected optical intensity from the sensor is a linear superposition of acoustic signal and noise, the signal can be obtained simply by the difference between the measured intensities from the two channels. The noise cancelation method was experimentally demonstrated using a system shown in Fig. 3(a). Two 7-mm long FBGs with peak reflectivity > 99:9% were fabricated in house using a 193 nm excimer laser and a phase mask. Their reflection spectra, characterized by a narrowlinewidth scanning laser, featured extremely narrow (1.0 and 1.2 pm, FWHM) spectral notches, as shown in Fig. 3(b). A fiber-pigtailed DFB laser diode (JDS Uniphase CQF935/508) with a specified Lorentzian linewidth of 2 MHz was used as the light source. The light from the laser source was divided into two channels (sensor and reference channels) via a 1 2 coupler. In the sensor channel, the light was directed to the sensor FBG through a circulator. The reflected light from the sensor was detected by an amplified photodetector. Because the DFB laser wavelength can be tuned by its injection current, we locked the laser wavelength at the slope of the spectral notch by feeding the low-frequency (DC-200 Hz) component of the photodetector output to the laser current driver through a servo controller (LB1005, Newport). The high-frequency (25–500 kHz) component of the photodetector output, containing both signal and noise, was sampled, digitalized, and acquired by an oscilloscope and processed by a computer. The sensor FBG was glued on a 20 20 aluminum plate. A piezoelectric transducer (HD50, Physical Acoustics Corp.) powered by a function generator was used to excite ultrasonic waves on the plate. During the experiment, the piezoelectric transducer was mounted by gluing its flat surface on the plate at a distance of ∼20 cm to the sensing grating along the fiber direction. In the reference channel, an attenuator was added to adjust the optical power level so that the two channels have similar responses. The rest of the optical configuration, the photodetector and its subsequent electronic filters for data acquisition and locking, are the same as those for the sensor channel. The reference FBG was mounted on a translation stage equipped with a piezo actuator (PE4, Thorlabs), which was used to tune its Bragg wavelength by changing the voltage applied on the piezo actuator to control the strain of the grating. The electronic control capability of the FBG spectral position allowed us to lock the slope of the FBG to the laser wavelength. The reference FBG was separated from the sensor FBG and thus was free from ultrasonic signals. For both channels, a polarization controller was applied before the FBG to eliminate one of the two polarization states originated from birefringence. We then studied the noise contributions from the laser RIN and laser frequency noise for the sensor channel when the lasing wavelength was located at different positions of the spectrum.


6801108



Fig. 4. (a) Noise output when laser wavelength is at the center of spectral slope (red) and at the flat region (blue). (Insets) FBG reflection spectrum. (b) Enlarged view of the blue curve in (a). (c) Noise in the sensor channel when the laser is off.

First, the laser wavelength was locked at the center of the sharp slope of the spectral notch, the output noise (see Fig. 4(a), red curve) including contributions from both the laser RIN and frequency noise. The root-mean-square (rms) value of the noise was 38 mV. Then, the laser wavelength was set to the flat shoulder of the spectrum, the output noise [see Fig. 4(a) and (b), blue curve] contained the contribution from the laser RIN, dark current from the photodetector, as well as the electronic noises from photodetectors, their amplifiers, and the oscilloscope. The rms noise level was dramatically reduced to about 1.8 mV as shown in Fig. 4(b). Fig. 4(c) is the measurement of the dark current of photodetector with the laser diode turned off and it had an rms value of 0.8 mV. These results indicate that noise contribution from the laser frequency noise dominates by about 20 times larger than from the laser RIN and other noise sources. Next, we demonstrated the proposed noise cancelation method using the reference channel. Although a tunable attenuator was included in the reference channel to balance the responses from the two channels, small discrepancy in noise output from the two channels were inevitable due to many factors such as limited tuning resolution of the attenuator, perturbations to the fiber in the two channels, variations of the amplifier gains, and fluctuations in the laser polarization. These variations in channel response often are small and have frequencies much smaller than frequencies of the ultrasonic signals being detected. Therefore, we developed the following data processing method to overcome this problem. The output from the sensor channel contains both the signal and noise and can be expressed as C1 ðt Þ ¼ Sðt Þ þ NðtÞ

(7)

where Sðt Þ and Nðt Þ are, respectively, the signal and the noise, and they are discrete data points after sampling and digitalization by the oscilloscope and passing a digital filter (a 500 kHz low-pass digital filter was used throughout our experiments). The output from the reference channel has only noise, which differs from noise term in (7) by a coefficient to account for the different responses of the two channels: C2 ðt Þ ¼ Nðt Þ:

(8)

Again, with the assumption of slow variations in the channel responses, can be considered as a constant within the detection bandwidth of the system. In addition, it has been assumed in (8) that there is no time delay between the noise responses from the two channels. To remove the noise, the coefficient in (8) should be determined. Note that Sðt Þ and Nðt Þ are two zeromean and uncorrelated processes or Sðt Þ ¼ 0, Nðt Þ ¼ 0, and Sðt ÞNðtÞ ¼ 0. Then, the mean of the product C1 ðt ÞC2 ðt Þ is given by C1 ðtÞC2 ðt Þ ¼ N 2 ðtÞ:


(9)

6801108



Fig. 5. (a)–(c) Temporal noise output and residual noise after noise cancelation. (Insets) Enlarged view of residual noise after noise cancelation. (d) Spectra of the noises.

From (8), we have N 2 ðt Þ ¼ C22 ðt Þ=2 , and plugging it into (9) yields ¼

C22 ðt Þ C1 ðt ÞC2 ðt Þ

:

(10)

With known, the noise from the sensor channel can be removed to extract the signal, which is given by Sðt Þ ¼ C1 ðt Þ C2 ðt Þ=:

(11)

We first investigated the performance of the proposed noise cancelation method when no ultrasonic signal was applied on the sensor. Fig. 5 shows the outputs from the sensor channel (red curve), the reference channel (blue curve), and the results after noise cancelation (green curve) for three sets of data obtained at difference times. The output from the two channels showed excellent correlation and the noise amplitude was substantially reduced by the noise cancelation method. The rms values for the signal before noise cancelation were 32.9, 37.9, and 37.6 mV for the data shown in Fig. 5(a)–(c), respectively. The average rms value of 36.1 mV corresponds to a wavelength shift of 4:5 103 pm of the FBG. After noise cancelation, they were reduced, respectively, to 3.3, 3.8, and 3.9 mV, as shown by the enlarged view of the results after noise cancelation in Fig. 5(a)–(c). The average rms value is reduced to 3.7 mV, corresponding to a wavelength shift of 4:5 104 pm. Therefore, the proposed noise reduction method was shown to suppress the noise amplitude by almost 10 times, corresponding to 20 dB improvement in the system SNR. The noise spectra before and after noise cancelation, obtained by averaging the Fourier spectra of seven data samples, are shown in Fig. 5(d). The noise level was reduced by over 20 dB at low end of the frequency range to ∼15 dB at high end of the frequency range. For comparison, the laser RIN noise spectrum obtained when the laser wavelength was set at the flat region of the FBG is also plotted in Fig. 5(d). It is seen that the noise floor from laser frequency noise after noise cancelation is about 7 dB higher than the RIN floor. These results in Fig. 5(d) are generally consistent with the results obtained from the time-domain signal. Note that a clear peak at 250 kHz was observed in the noise spectrum after noise cancelation. Further study is needed and underway to find the origin and possible ways to reduce the noise peak. To verify the SNR improvement, we then added a small ultrasonic signal to the sensor. In this experiment, a cw ultrasonic signal at a frequency of 200 kHz was applied on the aluminum plate


6801108



Fig. 6. (a) Temporal response to small continuous ultrasonic waves. (b) Spectra of output before and after noise cancelation.

Fig. 7. (a) Temporal response to ultrasonic pulses. (b) Spectra of output before and after noise cancelation.

by the PZT and detected by the sensor FBG. The temporal responses from the sensor channel (red curve), the reference channel (blue curve), and the signal after noise cancelation (green curve) are shown in Fig. 6(a). It is seen that the sinusoidal signal from the sensor channel was substantially deformed by the noise. After noise cancelation, the signal became much cleaner. The spectra of the signal before and after noise cancelation are shown in Fig. 6(b). The peak at 200 kHz corresponds to the acoustic signal and is roughly 20 dB above the noise floor for the signal from the sensor channel (red curve). After noise cancelation (green curve), the noise floor dropped by close to ∼17 dB at the low frequency end and ∼10 dB toward the high-frequency end while the signal peak at 200 kHz remained at the same level. The SNR improvement was slightly less than the 15–20 dB improvement observed in Fig. (5). Finally, we tested the effectiveness of the noise cancelation method for detection of ultrasonic pulses, which is of great interest in practical applications such as non-destructive evaluation and structural health monitoring. In this experiment, the PZT was driven by a repeating burst signal and each burst consists of five cycles of a 200 kHz sinusoidal voltage waveform. The outputs from the sensor channel (red curve) and the reference channel (blue curve) are shown in Fig. 7(a). The ultrasonic pulse signal was severely obscured by the noise. After noise cancelation, a relatively clean ultrasonic pulse emerged as shown by the green curve in Fig. 7(a). Similar to the case for cw ultrasonic waves shown in Fig. 6(b), the noise floor of the signal spectrum dropped by 10–17 dB while the signal peak remained the same after noise cancelation, as shown in Fig. 7(b). In summary, we have demonstrated a novel noise cancelation method for a fiber-optic ultrasonic sensor system based on intensity demodulation with a laser locked at the quadrature point of a FBG. The proposed concept involves two channels with matching FBGs, one as the sensor and the other as the reference channel. The sensor channel carries information on both the ultrasonic signal and the laser noise; while the reference channel is isolated from the acoustic signal and thus contains only the noise. Taking advantage of the high correlation between the noises from the two channels and low correlation between the signal and the noise, we have proposed a new data processing method to remove noise from the signal effectively. Using this methodology, a 10-time reduction in the rms value of noise amplitude has been experimentally obtained. In the presence of small acoustic signal in continuous mode, an improvement of ∼20 dB in SNR has been seen in experiments. In addition, the weak ultrasonic pulse signal otherwise immersed in the noise has been extracted effectively after noise cancelation.


6801108



The SNR improvement of this method is ultimately limited by the non-correlated part of the noises in the sensor channel and the reference channel. These non-correlated noises may come from many independent sources such as environmental perturbations to the fibers in these two channels and the electronic noises of the detectors and the signal acquisition system. The investigated noise cancelation method relieves the requirements on high-performance laser source and, thus, exhibits the great potential for achieving highly sensitive ultrasonic detection by a cost-effective fiber-optic sensor system.

References [1] A. Rosenthal et al., “Sensitive interferometric detection of ultrasound for minimally invasive clinical imaging applications,” Laser Photon. Rev., vol. 8, no. 3, pp. 450–457, 2014. [2] G. Rousseau, A. Blouin, and J.-P. Monchalin, “Non-contact photoacoustic tomography and ultrasonography for tissue imaging,” Biomed. Opt. Exp., vol. 3, no. 1, pp. 16–25, Jan. 2012. [3] I. M. Perez, H. L. Cui, and E. Udd, “Acoustic emission detection using fiber Bragg gratings,” in Proc. SPIE, 2001, vol. 4328, pp. 209–215. [4] D. C. Betz, G. Thursby, B. Culshaw, and W. J. Staszewski, “Acousto-ultrasonic sensing using fiber Bragg gratings,” Smart Mater. Struct., vol. 12, no. 1, pp. 122–128, 2003. [5] G. Wild and S. Hinckley, “Acousto-ultrasonic optical fiber sensors: Overview and state-of-the-art,” IEEE Sens. J., vol. 8, no. 7, pp. 1184–1193, Jul. 2008. [6] A. Minardo, A. Cusano, R. Bernini, L. Zeni, and M. Giordano, “Response of fiber Bragg gratings to longitudinal ultrasonic waves,” IEEE Trans. Ultrason., Ferroelectr., Freq. Control, vol. 52, no. 2, pp. 304–312, Feb. 2005. [7] T. Q. Liu and M. Han, Analysis of pi-phase-shifted fiber Bragg gratings for ultrasonic detection. IEEE Sens. J., vol. 12, no. 7, pp. 2368–2373, Jul. 2012. [8] A. Rosenthal, R. Razansky, and V. Ntziachristos, “High-sensitivity compact ultrasonic detector based on a pi-phase shifted fiber Bragg grating,” Opt. Lett., vol. 36, no. 10, pp. 1833–1835, May 2011. [9] Q. Wu and Y. Okabe, “High-sensitivity ultrasonic phase-shifted fiber Bragg grating balanced sensing system,” Opt. Exp., vol. 20, no. 27, pp. 28353–28362, 2012. [10] J. J. Guo and C. X. Yang, “Highly stabilized phase-shifted fiber Bragg grating sensing system for ultrasonic detection,” IEEE Photon. Technol. Lett., vol. 27, no. 8, pp. 848–851, Apr. 2015. [11] R. Drever et al., “Laser phase and frequency stabilization using an optical resonator,” Appl. Phys. B, Photophys. Laser Chem., vol. 31, no. 2, pp. 97–105, 1983. [12] R. Drever et al., “A gravity-wave detector using optical cavity sensing,” in Proc. 9th Int. Conf. Gen. Relativity Gravitation 1980. Cambridge, U.K., 1983, pp. 265–267. [13] T. Fink, Q. Zhang, F. W. Guo, W. Arhens, and M. Han, “Highly-sensitive -phase-shifted fiber-Bragg-grating ultrasonic sensors for structural health monitoring,” in Proc. Int. SAMPE Tech. Conf., 2012. [14] G. Di Domenico, S. Schilt, and P. Thomann, “Simple approach to the relation between laser frequency noise and laser line shape,” Appl. Opt., vol. 49, no. 25, pp. 4801–4807, Sep. 2010. [15] A. Rosenthal, D. Razansky, and V. Ntziachristos, “Wideband optical sensing using pulse interferometry,” Opt. Exp., vol. 20, no. 17, pp. 19016–19029, 2012.


6801108