PROCEEDINGS OF SPIE

SPIEDigitalLibrary.org/conference-proceedings-of-spie

Characterization of thin layered

structures using deconvolution
techniques in time-domain and
Fourier-domain optical coherence
tomography

Sébastien Vergnole, Daniel Lévesque, Guy Lamouche,

Marc Dufour, Bruno Gauthier

Sébastien Vergnole, Daniel Lévesque, Guy Lamouche, Marc Dufour, Bruno

Gauthier, "Characterization of thin layered structures using deconvolution
techniques in time-domain and Fourier-domain optical coherence
tomography," Proc. SPIE 6796, Photonics North 2007, 67961H (30 October
2007); doi: 10.1117/12.778745

Event: Photonics North 2007, 2007, Ottawa, Canada

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use

 Characterization of Thin Layered Structures using
Deconvolution Techniques in Time-Domain and
Fourier-Domain Optical Coherence Tomography
Sébastien Vergnole, Daniel Lévesque, Guy Lamouche, Marc Dufour and Bruno Gauthier
IMI-NRC, 75 bd de Mortagne, Boucherville, QC, Canada

ABSTRACT
Much of the current activity in optical coherence tomography aims at increasing the image resolution. Nowadays,
two kinds of OCT techniques are available. The first approach is the Time-Domain OCT (TD-OCT) which
usually relies on a moving part into the reference arm to probe the sample in depth. The second approach is
the Fourier-Domain OCT (FD-OCT) in which the signal is acquired as a function of the wavelength and the
depth profile of the sample is obtained by Fourier transform. Theoretically, in both techniques, the resolution is
limited by the central wavelength of the source and by its full width at half maximum. Nevertheless, it is shown
in this paper that this resolution may be improved by using deconvolution technique based on Wiener filtering
and Autoregressive Spectrum Extrapolation (ASE). In our experiment, thanks to deconvolution an improvement
of a factor up to 4 is obtained in TD-OCT and about 2 in FD-OCT. As an illustration, the approach is applied
to TD and FD-OCT measurements of the profile of a carbon-epoxy composite to evaluate the performance in
determining the thickness of the upper layer within a resolution better than that provided by the conventional
processing of the OCT envelope.
Keywords: TD-OCT, FD-OCT, Deconvolution, Resolution

1. INTRODUCTION
Optical Coherence Tomography (OCT) is now an established technique for cross-sectional imaging in biomedical
applications. But OCT can also be used in Non-Destructive Evaluation (NDE) of materials for industrial
applications. There are two families of OCT systems:

• the Time Domain OCT (TD-OCT) which contains a sample arm in which a sample is illuminated with a
broadband source and the backscattered light is collected and made to interfere with that from a reference
arm. The reference arm contains an optical delay line to continuously vary the depth at which the sample
is probed.
• the Fourier Domain OCT (FD-OCT) where the signal is acquired as a function of the wavelength. The
depth profile of the sample is then achieved by computing the Fourier transform.

One of the key issue in imaging samples is to improve the depth resolution to see more and more details in the
probed materials. Theoretically, the resolution is limited by the center wavelength of the source λ0 and by its
Full Width at Half Maximum (FWHM) ∆λ. Indeed, for a source with a gaussian shape the resolution in air is
equal to:
2 ln 2 λ20
δz = · (1)
π ∆λ
It means that in the case of a sample made of two close reflectors one cannot separate them if the space between
each other is smaller than δz. One technique commonly used in ultrasonic NDE to improve resolution is decon-
volution.1, 2 It consists in post-processing the signal taking into account the interferometric signal of one single
reflector. Deconvolution has already been demonstrated in TD-OCT. Kulkarni et al. use a linear shift invariant
Further author information: (Send correspondence to S.V.)
S.V.: E-mail: Sebastien.Vergnole@cnrc-nrc.gc.ca, Telephone: +1 450 641 5137

Photonics North 2007, edited by John Armitage,

Proc. of SPIE Vol. 6796, 67961H, (2007)
0277-786X/07/$18 · doi: 10.1117/12.778745

Proc. of SPIE Vol. 6796 67961H-1

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
 system model3 and Wang uses the Wiener filtering approach.4 Both of them limited their studies to biological
samples.

Here, we present a deconvolution algorithm based on Wiener filtering and Autoregressive Spectral Extrapo-
lation (ASE) that is applied in both TD-OCT and FD-OCT. First, the resolution enhancement is evaluated by
measurements on a glass wedge, and then it is applied to the thickness measurement of the upper layer of epoxy
in a carbon-epoxy composite.

2. DECONVOLUTION: THEORETICAL BACKGROUND

2.1 Principle
The input pulse propagating in a material is bandwidth-limited. The relation between the input x(t) and the
output y(t) is the convolution:
y(t) = h(t) ∗ x(t) (2)
where t is the time, h(t) is the impulse response, having a shorter pulse duration and which fully characterizes
the reflector (boundary, defect). It must be noted that in TD-OCT, the optical path is proportional to the time
whereas in our FD-OCT setup, the optical frequency is proportional to the time.

Without noise, h(t) could be estimated using Direct Fourier Transform (DFT) as:

Y (f )
H(f ) = (3)
R(f )

where f is the frequency, R(f ) is the DFT of a reference pulse r(t) representative of the input x(t). By computing
an inverse DFT, h(t) is obtained and used for imaging.

2.2 Algorithms
Two algorithms are used in this study for real signal in the presence of noise. Let us consider a signal with N
points and acquired with a sampling frequency fs . The first algorithm is based on Wiener filtering and can be
expressed as follow:

R(k)

H(k) = Y (k) (4)
|R(k)|2 + χ2 |R(k)|2max

where k = N · f /fs , R(k)
is the conjugate complex of R(k), and χ2 is a tunable factor, typically 0.01.

The second one is a wideband version of the Wiener filtering described above using Autoregressive Spectral
Extrapolation (ASE). Formally, it is written as:
L

Ĥ (k) = − afj Ĥ (k − j) if f > fmax
j=1
L

Ĥ (k) = − abj Ĥ (k + j) if f < fmin
j=1

Ĥ (k) = H(k) if fmin < f < fmax (5)

where the coefficients afj and abj are the Lth order forward and backward prediction coefficients of Ĥ (k). These
coefficients can be computed by using Burg’s technique.5 The order L is typically between 5 and 10.

3. EXPERIMENTAL SETUP
A TD-OCT system and a FD-OCT sytem are available in our institute. They are described below. Both of them
are fiber-based OCT setups.

Proc. of SPIE Vol. 6796 67961H-2

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
 3.1 Time Domain OCT setup
The TD-OCT setup is shown in figure 1 and is based on a Michelson configuration. The source used in our
experiment is a superluminescent diode (SLD) with a 1300 nm center wavelength and a 35 nm FWHM which
leads to a theoretical depth resolution of δz = 21.2 µm. The light is split with a 2×2 coupler into the reference
arm and the sample arm. The reference arm is fitted with a delay line which allows to vary the optical path.
The delay line is a galvanometer-mounted rhombric prism with a repetition rate of few kHz which has been used
for many years in our institute. The backreflected lights from both arms are then combined into the coupler.
The resulting interferometric signal is, at last, sent to a balanced detection.
Oscillating
rhombric prism
Circulator

SLED 1 2 coupler
3
50/50

Balanced
detection Sample

Figure 1. Time Domain OCT setup

3.2 Fourier Domain OCT setup

Fourier Domain OCT has concentrated a lot of researchers’ efforts in recent years. The two main advantages
of this technique is that there is no moving part in the reference arm and that all the points of an A-scan are
acquired simultaneously. It has also be shown that the signal to noise ratio was better than in TD-OCT.7 In the
case of FD-OCT, as we collect the signal as a function of the wavelength, two kinds of setup can be used. The
first one is called Spectral Domain OCT (SD-OCT) which consists of using a broadband source (SLED mainly)
and, then, acquiring the spectral interferogram by means of a spectrometer.8, 9 The second one is Optical Fourier
Domain Imaging (OFDI)10 also known as Swept Source OCT (SS-OCT). Here a wavelength-swept source is used
and the spectral interferogram can be acquired with a monopixel detector. This is the latter setup that is used
in our institute (see fig. 2).

Fixed
mirror
Polarization
controller
2
AOM 1
coupler 1 3
80 MHz coupler
50/50 50/50
Balanced
Circulators detection

AOM 2 1
Swept source 3
70 MHz 2

Sample

Figure 2. Optical Fourier Domain Imaging setup

More precisely, it is a Mach-Zehnder fiber based interferometer. The source is a Thorlabs swept-source with a
1325 nm center wavelength and a 85 nm FWHM. The A-scan rate is 16 kHz. The theoretical axial resolution is

Proc. of SPIE Vol. 6796 67961H-3

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
 then δz = 9.1 µm. It must be noted that the setup is fitted with acousto-optic modulators to remove the mirror
image and autocorrelation noise generated by the computation of the Fourier Transform.11, 12

4. EXPERIMENTAL RESULTS
In this section, measurements have been taken with TD-OCT and OFDI setups. First, we show an example
of resolution improvement. Then, we present the approach to determine the resolution limit of our systems.
Finally, measurements on a carbon-epoxy composite are presented.

4.1 Example of deconvolution results

Here is shown an example of signal post-processing. In this case, the sample is made of two very close reflectors.
The envelope of the OCT signal does not enable us to separate these two reflectors (figure 3, upper right). After
ASE deconvolution, the two reflectors can be clearly seen (figure 3, bottom right).
J!
!

Amr2ie

Figure 3. Example of resolution enhancement in TD-OCT using the ASE deconvolution technique for a separation of
5.9 µm between reflectors. Upper left: OCT signal, upper right: envelope of the OCT signal, bottom left: spectra of the
raw and deconvoluted signals, bottom right: envelope of the deconvoluted data.

4.2 Resolution enhancement on a glass wedge

The aim of this measurement is to determine the resolution limit of our system using deconvolution. The sample
consists of two glass slides superimposed with a slight angle as shown on the sketch in figure 4.

To interferometer

Moving Probe

Separation

Position
Figure 4. Sketch of the glass wedge.

Proc. of SPIE Vol. 6796 67961H-4

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
 The separation between the two glass slides varies linearly as a function of the probe position. The reference
signal used in the deconvolution is chosen where the separation is sufficient to capture a portion from a single
reflector. The reference signal taken from the bottom of the top slide or from the top of the bottom one gives
about the same performance.

Figure 5 shows the separation as a function of the position. Without deconvolution the minimum separation
achievable is around 21 µm which corresponds to the expected theoretical resolution. With the Wiener filtering
deconvolution, the separation decreases to 8 µm and with the ASE deconvolution, it reaches 5 µm. It means
that thanks to deconvolution, in our TD-OCT setup, the resolution is improved by a factor of at least 4.

100

..1•

AA •None
• Wiener
AASE

0.1 1.0 10.0

Position (mm)

Figure 5. Resolution Enhancement in TD-OCT. Square: without deconvolution, circle: Wiener filtering, triangle: ASE

As far as the OFDI setup is concerned (see figure 6), the experimental resolution without deconvolution is about
12 µm. There is a noticeable difference between this measurement and the theoretical value of 9 µm. It is partly
due to the fact that the swept-source doesn’t have a gaussian shape. Applying the Wiener filtering enables to
achieve a 8 µm resolution. At last, the maximum resolution obtained with the ASE deconvolution is around
6 µm. Therefore, a enhancement of a factor 2 is achieved in our OFDI setup.

100

.
• None
S Wiener
ASE
1

1.0 10.0
Position (mm)

Figure 6. Resolution Enhancement in OFDI. Square: without deconvolution, circle: Wiener filtering, triangle: ASE

Consequently, it is shown in this section that deconvolution enables to reach an enhanced resolution in TD-OCT
and in OFDI as well.

Proc. of SPIE Vol. 6796 67961H-5

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
 4.3 Surface profile of carbon-epoxy composite
In this section, we apply these deconvolution algorithms to measurements in a carbon-epoxy composite. Figure 7
shows a photo of the sample. This kind of fiber-reinforced composites are notably used to replace metallic alloys
in aircraft and spacecraft structures.

Figure 7. Photo of the carbon-epoxy composite

One illustration of the use of OCT is to visualize the upper layer of epoxy in this material. Once again, the
measurements have been carried out in both TD-OCT and OFDI. The results achieved with the ASE algorithm
are presented in figure 8. A moving average filter has been applied in these profiles to remove some noise. We
see in each case that the epoxy layer is very well resolved. The presence of the individual carbon fiber is clearly
seen as the epoxy layer is following more or less the same periodicity. It also provides the basic surface roughness
of this material. Some difference between the two surfaces are lower than 10 µm in spite of the fact that the
theoretical resolution of our sources is larger than that. The deconvolution is therefore an efficient technique to
improve resolution in TD-OCT and OFDI as well.

'00 200 300 400 500 600 700 600 900 1000

-'0
-zo
E

-40
r

0 50 100 tso 200 250 300 350 400 450 500 550 600 650 700 750 600 650 900 950
Poitibn (urn)

Figure 8. Surface profile of the carbon-epoxy composite in TD-OCT (top) and in OFDI (bottom). The probed areas are
not the same in the two cases.

Proc. of SPIE Vol. 6796 67961H-6

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
 5. CONCLUSION
In this paper, a resolution enhancement is reported using deconvolution techniques in TD-OCT and OFDI as well.
Two algorithms have been used based on Wiener filtering and Autoregressive Spectral Extrapolation. Actually,
a resolution improvement of a factor 4 has been achieved in TD-OCT and of a factor 2 in OFDI. Therefore,
the upper layer of a carbon-epoxy composite is well resolved. OCT appears to be a promising nondestructive
evaluation technique for inspecting thin layered structures.

REFERENCES
1. K. I. McRae, “Deconvolution techniques for ultrasonic imaging of adhesive joints,” Materials Evaluation ,
pp. 1380–1384, 1990.
2. S.-K. Sin and C.-H. Chen, “A comparison of deconvolution techniques for the ultrasonic nondestructive
evaluation of materials,” IEEE Transactions on Image Processing 1(1), pp. 3–10, 1992.
3. M. D. Kulkarni, C. W. Thomas, and J. A. Izatt, “Image enhancement in optical coherence tomography
using deconvolution,” Electronics Letters 33(16), pp. 1365–7, 1997.
4. R. K. Wang, “Resolution improved optical coherence-gated tomography for imaging through biological
tissues,” Journal of Modern Optics 46(13), pp. 1905–1912, 1999.
5. S. M. Kay, Modern Spectral Estimation: Theory and Application, Prentice Hall, 1988.
6. G. Lamouche, M. Dufour, B. Gauthier, V. Bartulovic, M. Hewko, and J. P. Monchalin, “Optical delay line
using rotating rhombic prisms,” 6429, pp. 64292G–1 – 64292G–6, 2007.
7. M. A. Choma, M. V. Sarunic, C. Yang, and J. A. Izatt, “Sensitivity advantage of swept source and fourier
domain optical coherence tomography,” Optics Express 11(18), pp. 2183 – 2189, 2003.
8. A. F. Fercher, C. K. Hitzenberger, G. Kamp, and S. Y. El Zaiat, “Measurement of intraocular distances by
backscattering spectral interferometry,” Optics Communications 117(1-2), pp. 43–48, 1995.
9. G. Hausler and M. W. Linduer, “”coherence radar” and ”spectral radar”-new tools for dermatological
diagnosis,” Journal of Biomedical Optics 3(1), pp. 21–31, 1998.
10. S. H. Yun, G. J. Tearney, J. F. de Boer, N. Iftimia, and B. E. Bouma, “High-speed optical frequency-domain
imaging,” Optics Express 11(22), pp. 2953 – 2963, 2003.
11. S. H. Yun, G. J. Tearney, J. F. de Boer, and B. E. Bouma, “Removing the depth-degeneracy in optical
frequency domain imaging with frequency shifting,” Optics Express 12(20), pp. 4822 – 4828, 2004.
12. A. M. Davis, M. A. Choma, and J. A. Izatt, “Heterodyne swept-source optical coherence tomography
for complete complex conjugate ambiguity removal,” Journal of Biomedical Optics 10(6), pp. 064005–1 –
064005–6, 2005.

Proc. of SPIE Vol. 6796 67961H-7

Downloaded From: https://www.spiedigitallibrary.org/conference-proceedings-of-spie on 19 Jan 2024

Terms of Use: https://www.spiedigitallibrary.org/terms-of-use