+ All Categories
Home > Documents > Periodic reference subtraction method for efficient background fixed pattern noise removal in...

Periodic reference subtraction method for efficient background fixed pattern noise removal in...

Date post: 10-Sep-2016
Category:
Upload: ji-hyun-kim
View: 214 times
Download: 0 times
Share this document with a friend
5
Periodic reference subtraction method for efcient background xed pattern noise removal in Fourier domain optical coherence tomography Ji-hyun Kim, Jae-Ho Han, Jichai Jeong Department of Brain and Cognitive Engineering, Korea University, 1, 5Ka, Anam-dong, Sungbuk-ku, Seoul 136-701, Republic of Korea abstract article info Article history: Received 8 July 2011 Received in revised form 6 December 2011 Accepted 9 December 2011 Available online 23 December 2011 OCIS codes: (110.4500) Optical coherence tomography (100.2980) Image enhancement (060.2350) Fiber optics imaging (120.3890) Medical optics instrumentation Keywords: Optical coherence tomography Fixed pattern noise FD-OCT Reference spectrum subtraction We have demonstrated an efcient scheme of automatically removing xed pattern noise in Fourier domain optical coherence tomography by using a periodic reference spectrum subtraction method. By periodically acquiring the reference spectra using a separate light absorber placed to the right of the scan lens, we were able to adaptively compensate the background xed pattern due to the spectral intensity variation of the source. The adaptive removal of xed pattern noise was effectively performed by controlling the refer- ence spectrum acquisition rate (R). A seawater pearl was used for a test sample under an intentional abrupt source power change to validate the proposed method. Based on this method, it is possible to perform imme- diate cold start scanning because it is not necessary for a stabilization period of the light source, as well as a manual process of reference spectrum acquisition for obtaining clear image under unstable environment. © 2011 Elsevier B.V. All rights reserved. 1. Introduction Optical coherence tomography (OCT) is an optical imaging tech- nique which has received considerable attention because OCT can provide a high-resolution cross-sectional image of samples [13]. OCT commonly uses near-infrared broadband light as a low coher- ence source that can penetrate into the sample in comparison with a sonogram. OCT, however, has been limited to subsurface area imag- ing due to the inherent scattering properties of near-infrared light in the sample so that most applications have been focused on demon- strating imaging areas in ophthalmology [47], dermatology [811], and cardiology [12,13]. Recently, Fourier domain OCT (FD-OCT) has become a mainstream OCT scheme because of higher imaging speed and better sensitivity than time domain OCT (TD-OCT) by skipping the practical depth-scanning mechanism and by directly extracting the depth information from the modulated envelope of the spectrum by signal processing [1417]. In spite of providing various attractive features, the FD-OCT scheme has the following artifacts: complex conjugate, xed- pattern, auto-correlation, and cross-correlation. From the so-called xed patter artifact, multi-line horizontal xed-patterns signicantly deteriorate acquired image results. Basically there are two kinds of methods that have been typically used in the FD-OCT system to alle- viate the xed-pattern noise: by simply subtracting the real reference spectrum captured in the absence of a specimen, and a signal proces- sing approach using the acquired overall data [18,19]. In the former case, additional signal processing is not required for resolving the xed pattern noise resulting in a fast and effective performance in the FD-OCT system with use of a highly stable light source. However, in this case based on a reference spectrum, subtraction is neither al- ways convenient nor effective if the light source shows moderate changes in the source spectral intensity prole. For an endoscopic FD-OCT [20,21], ber bending can cause a signicant optical power uctuation which gives rise to a severe xed pattern noise in the produced cross-sectional image. In the latter method based on signal processing using the obtained B-scan data matrices, a mean- and median-line can be used for sub- traction purposes instead of using the really measured spectrum in the former case mentioned above [19]. This way of compensation can also be applied to a source showing a gradual power variation. However, it requires an additional processing for the mean and/or median line calculation followed by a sequential subtraction in a whole two-dimensional B-scan image, which results in a much slower performance than a simple Fourier transform process. There- fore, the processing speed and accommodation of the source showing a gradual or moderate change in the source emitting intensity spectrum should be overcome in order to efciently cancel out the xed-pattern noise. Optics Communications 285 (2012) 20122016 Corresponding author. Tel.: + 82 2 3290 3233. E-mail address: [email protected] (J. Jeong). 0030-4018/$ see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.optcom.2011.12.045 Contents lists available at SciVerse ScienceDirect Optics Communications journal homepage: www.elsevier.com/locate/optcom
Transcript
Page 1: Periodic reference subtraction method for efficient background fixed pattern noise removal in Fourier domain optical coherence tomography

Optics Communications 285 (2012) 2012–2016

Contents lists available at SciVerse ScienceDirect

Optics Communications

j ourna l homepage: www.e lsev ie r .com/ locate /optcom

Periodic reference subtraction method for efficient background fixed pattern noiseremoval in Fourier domain optical coherence tomography

Ji-hyun Kim, Jae-Ho Han, Jichai Jeong ⁎Department of Brain and Cognitive Engineering, Korea University, 1, 5Ka, Anam-dong, Sungbuk-ku, Seoul 136-701, Republic of Korea

⁎ Corresponding author. Tel.: +82 2 3290 3233.E-mail address: [email protected] (J. Jeong).

0030-4018/$ – see front matter © 2011 Elsevier B.V. Alldoi:10.1016/j.optcom.2011.12.045

a b s t r a c t

a r t i c l e i n f o

Article history:Received 8 July 2011Received in revised form 6 December 2011Accepted 9 December 2011Available online 23 December 2011

OCIS codes:(110.4500) Optical coherence tomography(100.2980) Image enhancement(060.2350) Fiber optics imaging(120.3890) Medical optics instrumentation

Keywords:Optical coherence tomographyFixed pattern noiseFD-OCTReference spectrum subtraction

We have demonstrated an efficient scheme of automatically removing fixed pattern noise in Fourier domainoptical coherence tomography by using a periodic reference spectrum subtraction method. By periodicallyacquiring the reference spectra using a separate light absorber placed to the right of the scan lens, wewere able to adaptively compensate the background fixed pattern due to the spectral intensity variation ofthe source. The adaptive removal of fixed pattern noise was effectively performed by controlling the refer-ence spectrum acquisition rate (R). A seawater pearl was used for a test sample under an intentional abruptsource power change to validate the proposed method. Based on this method, it is possible to perform imme-diate cold start scanning because it is not necessary for a stabilization period of the light source, as well as amanual process of reference spectrum acquisition for obtaining clear image under unstable environment.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

Optical coherence tomography (OCT) is an optical imaging tech-nique which has received considerable attention because OCT canprovide a high-resolution cross-sectional image of samples [1–3].OCT commonly uses near-infrared broadband light as a low coher-ence source that can penetrate into the sample in comparison witha sonogram. OCT, however, has been limited to subsurface area imag-ing due to the inherent scattering properties of near-infrared light inthe sample so that most applications have been focused on demon-strating imaging areas in ophthalmology [4–7], dermatology [8–11],and cardiology [12,13]. Recently, Fourier domain OCT (FD-OCT) hasbecome a mainstream OCT scheme because of higher imaging speedand better sensitivity than time domain OCT (TD-OCT) by skippingthe practical depth-scanning mechanism and by directly extractingthe depth information from the modulated envelope of the spectrumby signal processing [14–17].

In spite of providing various attractive features, the FD-OCTscheme has the following artifacts: complex conjugate, fixed-pattern, auto-correlation, and cross-correlation. From the so-calledfixed patter artifact, multi-line horizontal fixed-patterns significantlydeteriorate acquired image results. Basically there are two kinds of

rights reserved.

methods that have been typically used in the FD-OCT system to alle-viate the fixed-pattern noise: by simply subtracting the real referencespectrum captured in the absence of a specimen, and a signal proces-sing approach using the acquired overall data [18,19]. In the formercase, additional signal processing is not required for resolving thefixed pattern noise resulting in a fast and effective performance inthe FD-OCT system with use of a highly stable light source. However,in this case based on a reference spectrum, subtraction is neither al-ways convenient nor effective if the light source shows moderatechanges in the source spectral intensity profile. For an endoscopicFD-OCT [20,21], fiber bending can cause a significant optical powerfluctuation which gives rise to a severe fixed pattern noise in theproduced cross-sectional image.

In the latter method based on signal processing using the obtainedB-scan data matrices, a mean- and median-line can be used for sub-traction purposes instead of using the really measured spectrum inthe former case mentioned above [19]. This way of compensationcan also be applied to a source showing a gradual power variation.However, it requires an additional processing for the mean and/ormedian line calculation followed by a sequential subtraction in awhole two-dimensional B-scan image, which results in a muchslower performance than a simple Fourier transform process. There-fore, the processing speed and accommodation of the source showinga gradual or moderate change in the source emitting intensityspectrum should be overcome in order to efficiently cancel out thefixed-pattern noise.

Page 2: Periodic reference subtraction method for efficient background fixed pattern noise removal in Fourier domain optical coherence tomography

Fig. 1. SLD spectrum changes for 1-hour (a) after power on, and (b) after a 1-hourstabilization.

Fig. 2. Received SLD spectrum from the spectrometer (solid line) and optical powerdifferences (dotted line) when a silver-coated mirror is used as a sample.

2013J. Kim et al. / Optics Communications 285 (2012) 2012–2016

In this work, we proposed a periodic reference spectrum subtrac-tion in the FD-OCT system in which the reference signal was period-ically renewed to effectively cancel the fixed pattern noise owing tothe variation in the source spectrum profile. A light absorber hasbeen introduced to reacquire the reference spectrum during the B-scanning process without additional signal processing, such as meanor median calculation to remove the background artifact in the over-all image. The effect of the reference acquisition rate has been inves-tigated under an intentional reference spectrum fluctuation duringlateral scanning for validation of the effectiveness of our proposedmethod.

2. Reference spectrum fluctuation in the FD-OCT

The FD-OCT system uses low coherence interferometry that con-sists of an interferometer (Michelson interferometer), broadbandlight source, spectrometer, reference mirror, and sample stage. Inthe reference arm, a silver-coated mirror is placed for reflecting thereference beam to an interferometer, and a two-dimensional scanneris used for B-scan imaging in a sample arm. Back-reflected light fromthe sample interferes with the reference light in the interferometerand the depth information is encoded by optical frequency in time.The signal in frequency (k)-domain received from the spectrometer,ISI(k), is given below and includes DC light intensity back-reflectedfrom the reference mirror, IR(k), and sample, IS(k) [16].

ISI kð Þ ¼ IR kð Þ þ IS kð Þ þ 2ffiffiffiffiffiffiffiffiffiffiIR kð Þp ffiffiffiffiffiffiffiffiffiffi

IS kð Þp

� cos ϕS kð Þ−ϕR kð Þ−2πkτ½ �ð1Þ

The last term in Eq. (1) is an interference term that contains depthinformation, and is decoded by taking the inverse Fourier transformof ISI(k). However, because of the DC intensity terms, IR(k) and IS(k),there are stationary horizontal high-intensity lines in the correspond-ing result image near the zero delay line, the so-called fixed patternnoise. We have neglected the IS(k) term because this DC light intensi-ty back-reflected from the sample is relatively small in general com-pared to the DC reference term, IR(k).

Accordingly, a reference spectrum subtraction method has beenapplied for rejecting the fixed pattern noise by measuring the spec-trum of back-reflected light from the reference mirror without anysample before the experiments [22]. Thus, this method can removethe fixed pattern noise in an effective manner. However, opticalpower can be changed by temperature variation of light source con-trolled by constant current driving mode, and also spectrum shapecan be changed by super-luminescent diode (SLD) pumping wave-length variation and phase instability. To observe the profile changes,we measured the spectrum of the source for an hour. SLD has a1310 nm center wavelength with a 14 mW optical power at800 mA, and we used a commercial spectrometer (details for the ex-periment will be provided in the next chapter). Fig. 1(a) indicates theSLD spectrum changes for a 1-hour measurement, and the spectrumhas changed by 1.18%. Fig. 1(b) shows the same experimental resultafter having a 1-hour SLD temperature stabilization period and thespectrum experienced a change of 0.47%. These changes seem to bemeaningless, but the spectrum can be modified by approximately6% when we used a silver-coated mirror as a sample with our FD-OCT setting. Thus, these become significant (8%–20%) for spectrumfluctuations in the image processing standpoint. In Fig. 2, the mea-sured spectrum (solid line) and optical power difference (dottedline) are shown in the case of silver-coated mirror sample at a300 μm axial distance apart from the zero delay line. The maximumoptical difference was 4.7%, which was b6%, because of the signalloss by depth and tilted angle of the sample surface to the beamdirection.

The spectral change or the deviation from the initial referencespectrum, gd,n(w), can be calculated by the following Eq. (2) thatsubtracted the initially obtained spectrum, g1(w), from the currentspectrum gn(w).

gd;n wð Þ ¼ gn wð Þ−g1 wð Þ ð2Þ

Page 3: Periodic reference subtraction method for efficient background fixed pattern noise removal in Fourier domain optical coherence tomography

Fig. 4. Block diagram of periodic reference spectrum subtraction method in case ofR=1. Inner FOR loop indicates a B-scan, including synchronized galvanometer control,reference subtraction, and image processing steps for cross-sectional image plot. OuterFOR loop deals with reference spectrum acquisition procedure.

2014 J. Kim et al. / Optics Communications 285 (2012) 2012–2016

The SLD spectrum showed a slow and gradual modification in timeeven after having the stabilization period so that this spectral changeincreased the fixed pattern noise due to the mismatch between thecurrent and the pre-obtained reference spectra. Therefore, to weakenthis generated fixed pattern noise, the reference spectrum should berefreshed and reacquired before each experiment for complete re-moval of this noise term. In order to acquire a renewed current spec-trum for efficiently rejecting this fixed pattern noise during the lateralB-scanning process, we have introduced a periodic reference spec-trum subtraction method in the following chapter.

3. Periodic reference spectrum subtraction method

We have constructed an optical fiber-based FD-OCT system, as il-lustrated in a schematic view in Fig. 3. The SLD (SLD1021XL, Covega,USA), the broadband light source, has a 1310 nm center wavelength,a 90 nm full width half maximum (FWHM) bandwidth, and a14 mW optical output power. A 2×2 directional optical coupler actsas a fiber optic Michelson interferometer that splits SLD light intoboth the reference mirror and sample stage with a 50:50 ratio.Back-reflected light from the sample interferes with light from thereference mirror, and the fringe pattern due to interference isdetected by commercial spectrometer (NIRQuest512, Ocean Optics,USA) with a 512 pixel InGaAs CCD. The spectrometer has a 25 μmslit and 3.1 μm spectral resolution (FWHM in pixels). A two-dimensional galvanometer (GVS002, Thorlabs, USA) is attached inthe sample scanner cage for the lateral scan (B-scan) and is controlledby the galvanometer driver with a saw-tooth wave from the DAQboard (PCIe-6361, National Instrument, USA). Axial and lateral reso-lutions were measured to be approximately 13 μm in air based onour system using the SLD low coherence source and the achromaticscanning lens (LSM02, Thorlabs, USA) in the sample arm.

When lateral scanning is initiated, the galvanometer rotates thescanning mirror directed to a light absorber for reference spectrumacquisition prior to sample scanning. The absorber can be a piece ofblack paper, and is introduced to effectively block the back-reflectedlight from the sample arm for obtaining the reference spectrum.The acquired spectrum is stored in a buffer that will be used forfixed pattern noise rejection. We obtained a reference spectrumbefore each B-scan, and the overall process flow is shown in Fig. 4.The inner FOR loop indicates a B-scan, including synchronizedgalvanometer control, reference subtraction, and image processingsteps for cross sectional image plot. An outer FOR loop deals with

Fig. 3. Schematic of a fiber-based FD-OCT system setup. A light absorber is placed in thesample head cage to obtain a physical reference spectrum by blocking back-reflectedlight from the sample. (SLD: super-luminescent diode; DAQ: data acquisition board.)

reference spectrum acquisition and we define reference spectrumacquisition rate (R) as follows:

R ¼ number of reference spectrum acquisitionsnumber of lateral scans

: ð3Þ

Fig. 5 shows a sample scanning head cage structure. As can beseen, a light absorber is located beside the scan lens for rapid access,

Fig. 5. Illustration of sample scanning head cage structure. Absorber is placed to blockback-reflected light from the sample.

Page 4: Periodic reference subtraction method for efficient background fixed pattern noise removal in Fourier domain optical coherence tomography

Fig. 6. Nacre layers and nuclei of seawater pearl are shown in the image (a) with normal condition, (b) with changing SLD driving current at the indicated arrow, and (c) afterremoving fixed pattern noise using new reference spectrum.

2015J. Kim et al. / Optics Communications 285 (2012) 2012–2016

and the surface is not perpendicular to the beam direction to mini-mize the residual reflection. However, if some conditions are satisfiedso that the cage is not a non-highly scattering material and the lightincident angle is tilted to the cage surface, then the absorber can beomitted from the cage because there is negligible back-reflectionfrom the cage surface in practice. Hence, it is possible to obtain a ref-erence spectrum without an absorber after placing the beam positionout of the scan lens.

4. Results and discussion

A seawater pearl was used as the specimen for the experiments.One hour of SLD stabilization was presented before scanning withthermoelectric cooling under an 800 mA constant current SLD drivingmode and 14 mW optical power output. The reference spectrum ac-quisition rate was set to 1 (R=1), which corresponds to a referencespectrum obtained for every lateral scan. There was no observable in-crease in fixed pattern noise for the first 20 minute scanning becauseof using periodic reference spectrum acquisition procedures. We as-sumed more severe conditions than a long-time scan with an SLD in-tensity change during scanning for the purpose of generating a fixedpattern noise in the resulting image. Fig. 6(a) shows the nacre layersand nuclei of the seawater pearl in normal conditions without an SLDintensity change (256×256 pixels within a 5 mm lateral scanningdistance and an 800 μm penetration depth) [23]. During the lateralscanning (B-mode), we suddenly changed the SLD driving currentfrom 800 mA (P=14 mW) to 400 mA (P=10.2 mW) at the arrowposition presented in Fig. 6(b). As a result, the fixed pattern noise be-came prominent and ruined the image. Conventionally, the manualreference spectrum acquisition step should be performed in this con-dition to reduce the noise.

In comparison, the proposed method, which automatically re-freshes the reference spectrum after each lateral scan, the fixed pat-tern noise was effectively removed for the next lateral scan asshown in Fig. 6(c), without taking a manual reference spectrum ac-quisition step. The fixed pattern noise shown in Fig. 6(b) can be di-minished in real-time if we use a larger reference spectrumacquisition rate (R), such as R=10 (10-fold reference spectrum re-freshing for a single lateral scan). Adaptability for an optical power-varying environment will be decreased if R is relatively small, andthe system gets a fast image acquisition speed. In contrast, for alarge R, the system will have high adaptability with a low-imagingspeed. However, the large-R OCT system can deteriorate the obtainedimage by washing out the fringe owing to the extremely fast galva-nometer movement with large angle rotation and vibration. There-fore, we recommend using a small value of R (b1) for stable andefficient OCT imaging.

This periodic reference subtraction method can eliminate themanual reference spectrum acquisition steps before scanning andcompensate the optical power variation of the source without

additional digital signal processing. Each noise reduction techniquewas simulated using a MATLAB 7.10.1. with Intel(R) Core(TM) i7-950 CPU for comparison of the processing speed. The periodic refer-ence subtraction method has 37% and 147% faster image processingspeed than the conventional mean and median line subtractionmethods, respectively, when the reference spectrum acquisition rate(R) is fixed by 1 with 256 A-lines in a B-scan image. The mean linesubtraction method requires horizontally averaging the real andimaginary parts of images, and then subtraction of the mean. It tookmore time, compared to measuring periodic reference spectrum.The median line subtraction method employs a sorting process thatrequires more computing load than averaging. In the consequence,the proposed method could have faster processing speed than theconventional methods.

5. Conclusion

We proposed and demonstrated a periodic reference subtractionmethod for a stable FD-OCT image by periodically refreshing the ref-erence signal under an extreme case of reference spectrum change tovalidate the effectiveness of our method. To acquire reference renew-al during the lateral B-scanning process, a light absorber was placedin the sample head cage to block back reflection from the samplearm for imitating the absence of the specimen. The degree of systemstability is controlled by the reference spectrum acquisition rate (R),and we used R=1 that will obtain reference spectrum for each B-scan. As a result, the FD-OCT system can effectively alleviate the influ-ence of a slowly-varying fixed pattern noise, even for an opticalpower variation of the light source. In addition, we can eliminatethe separate manual reference spectrum acquisition steps and the te-dious source stabilization period before initiating the scanning pro-cess. Also, based on the demonstrated method, reference spectrummeasurement can be efficiently achieved to automatically reducethe fixed pattern noise for the endoscopic OCT by placing a tinylight absorber in a probe.

Acknowledgements

This research was supported by World Class University programfunded by the Ministry of Education, Science and Technology throughthe National Research Foundation of Korea (R31-10008).

References

[1] D. Huang, E.A. Swanson, C.P. Lin, J.S. Schuman, W.G. Stinson, W. Chang, M.R. Hee,T. Flotte, K. Gregory, C.A. Puliafito, J.G. Fujimoto, Science 254 (1991) 1178.

[2] M. Szkulmowski, A. Wojtkowski, T. Bajraszewski, I. Gorczynska, P. Targowski, W.Wasilewski, A. Kowalczyk, C. Radzewicz, Optics Communications 246 (2005) 569.

[3] H. Wang, M.W. Jenkins, A.M. Rollins, Optics Communications 281 (2008) 1896.[4] J.G. Fujimoto, W. Drexler, J.S. Schuman, C.K. Hitzenberger, Optics Express 17

(2009) 3978.[5] I. Svorenova, P. Strmen, Z. Olah, Bratislava Medical Journal 111 (2010) 306.

Page 5: Periodic reference subtraction method for efficient background fixed pattern noise removal in Fourier domain optical coherence tomography

2016 J. Kim et al. / Optics Communications 285 (2012) 2012–2016

[6] G. Savini, M. Carbonelli, P. Barboni, Current Opinion in Ophthalmology 22 (2011)115.

[7] M. Gora, K. Karnowski, M. Szkulmowski, B.J. Kaluzny, R. Huber, A. Kowalczyk, M.Wojtkowski, Optics Express 17 (2009) 14880.

[8] T. Gambichler, G. Moussa, M. Sand, D. Sand, P. Altmeyer, K. Hoffmann, Journal ofDermatological Science 40 (2005) 85.

[9] Y. Hori, Y. Yasuno, S. Sakai, M. Matsumoto, T. Sugawara, V.D. Madjarova, M.Yamanari, S. Makita, T. Araki, M. Itoh, T. Yatagai, Optics Express 14 (2006) 1862.

[10] N. Krstajic, L.E. Smith, S.J. Matcher, D.T.D. Childs, M. Bonesi, P.D.L. Greenwood,M. Hugues, K. Kennedy, M. Hopkinson, K.M. Groom, S. MacNeil, R.A. Hogg, R.Smallwood, IEEE Journal of Selected Topics in Quantum 16 (2010) 748.

[11] A. Alex, B. Povazay, B. Hofer, S. Popov, C. Glittenberg, S. Binder, W. Drexler, Journalof Biomedical Optics 15 (2010) 026025.

[12] M.A. Choma, M.J. Suter, B.J. Vakoc, B.E. Bouma, G.J. Tearney, Journal of BiomedicalOptics 15 (2010) 056020.

[13] M.W. Jenkins, D.C. Adler, M. Gargesha, R. Huber, F. Rothenberg, J. Belding, M.Watanabe, D.L. Wilson, J.G. Fujimoto, A.M. Rollins, Optics Express 15 (2007) 6251.

[14] S.H. Yun, G.J. Tearney, J.F. de Boer, N. Iftimia, B.E. Bouma, Optics Express 11 (2003)2953.

[15] A.F. Fercher, C.K. Hitzenberger, G. Kamp, S.Y. Elzaiat, Optics Communications 117(1995) 43.

[16] M.E. Brezinski, Optical Coherence Tomography: Principles and Applications,Academic Press, Amsterdam; Boston, 2006.

[17] R. Leitgeb, C.K. Hitzenberger, A.F. Fercher, Optics Express 11 (2003) 889.[18] N.A. Nassif, B. Cense, B.H. Park, M.C. Pierce, S.H. Yun, B.E. Bouma, G.J. Tearney, T.C.

Chen, J.F. de Boer, Optics Express 12 (2004) 367.[19] S. Moon, S.W. Lee, Z.P. Chen, Optics Express 18 (2010) 24395.[20] A.R. Tumlinson, J.K. Barton, B. Povazay, H. Sattman, A. Unterhuber, R.A. Leitgeb, W.

Drexler, Optics Express 14 (2006) 1878.[21] J.U. Kang, J.H. Han, X.A. Liu, K. Zhang, C.G. Song, P. Gehlbach, IEEE Journal of

Selected Topics in Quantum 16 (2010) 781.[22] M. Wojtkowski, Applied Optics 49 (2010) D30.[23] M.J. Ju, S.J. Lee, E.J. Min, Y. Kim, H.Y. Kim, B.H. Lee, Optics Express 18 (2010)

13468.


Recommended