Exploring Frequency Domain OCT Systems

John Grasel
Summer 2011

Frequency domain optical coherence tomography allows for in-vivo non-

destructive imaging at speeds two orders of magnitude faster than time domain
methods while increasing the signal to noise ratio. These benefits are the re-
sult of acquiring information on all scatterers in a depth scan all the time. We
explore methods of maintaining depth resolution through dispersion compen-
sation and lateral resolution and signal to noise ratio through slab-based OCT
and Bessel beams, and we examine methods of resolving the complex conjugate
ambiguity. We investigate ISAM and its ability to dynamically refocus data
in software, and look at ways of maintaining the phase stability necessary for
motion sensitivity. We calculate what specifications a hypothetical instrument
would have compared to our current designs.

1
Contents
1 Introduction 3

2 Design Specifications 3
2.1 Current Instrumentation . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Recent Research . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Slab-Based FD-OCT . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Theory 4
3.1 FD-OCT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3.2 Rayleigh Length . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3 Slab-Based FD-OCT . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.4 Signal to Noise Ratio . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.4.1 Frequency-Domain SNR . . . . . . . . . . . . . . . . . . . 9
3.4.2 Time-Domain SNR . . . . . . . . . . . . . . . . . . . . . . 11
3.5 Non-Gaussian Spectra . . . . . . . . . . . . . . . . . . . . . . . . 12

4 Specific Design Challenges and Solutions 12

4.1 Complex-Conjugate Ambiguity . . . . . . . . . . . . . . . . . . . 12
4.2 Bessel Beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4.3 Gabor-Domain OCT . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.4 ISAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.5 Dispersion Compensation . . . . . . . . . . . . . . . . . . . . . . 16
4.6 Phase Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

5 Overall Instrument Design 17

5.1 Hardware and Data Acquisition . . . . . . . . . . . . . . . . . . . 17
5.2 Signal Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

6 Slab-Based OCT Performance 19

7 References 21

2
1 Introduction
In 2011, it was decided that a new microscope was necessary to meet the HMC
OCM lab’s biological imaging requirements. The current optical coherence mi-
croscopes (OCM), Michelson and Fabry, have wavelengths centered at 1300 nm
and 850 nm, respectively. The largest issue with both instruments is their imag-
ing speed. A large 10 million voxel image takes over 10 minutes on the 1300 nm
instrument and over 15 minutes on the 850 nm instrument. This is an inconve-
nience for imaging static objects like cell monolayers, and it is detrimental for
images of dynamic objects like frog embryos, where cell motion and cell division
during the acquisition blurs the image. While the lateral resolution in the X
and Y dimensions is a respectable 5 µm, the axial resolution in the Z direction
is 15 µm in air, and 11 µm in water. Improving both the acquisition time and
the axial resolution could result in the visualization of cells.
Frequency domain optical coherence tomography (FD-OCT) is a new tech-
nology that has the potential to let us achieve the speedup we desire. FD-OCT
differs from time domain OCT (TD-OCT) in how the sample image is con-
structed. In the time domain, coherence gating extracts a voxel at a particular
depth in the sample. In the frequency domain method, the interferometric sig-
nal created by mixing the sample and reference light is sampled as a function
of wavenumber and yields an entire depth scan (A-scan). There are two ap-
proaches to FD-OCT: spectral domain OCT (SD-OCT) and swept source OCT
(SS-OCT). SD-OCT uses a broad SLD source and a spectrometer in conjunction
with a linear camera array as the detector. SS-OCT uses a standard detector,
but uses a frequency-swept laser.

2 Design Specifications
2.1 Current Instrumentation
Our current OCM instruments with wavelengths centered at 850nm and 1300nm
have lateral resolutions of 5µm, but poor depth resolutions of 15µm in air. Addi-
tionally, they take upwards of 5 minutes to image a 106 -voxel cube (100x100x100)
with motion sensitivity enabled. For imaging frog embryos, the movement of
the sample during image acquisition and the coarse depth resolution limit our
ability to see individual cells. The primary goal of a new instrument is to in-
crease our voxel resolution to 5µm in each direction and acquire a significantly
larger 8 · 106 -voxel cube (200x200x200) in under a minute.

2.2 Recent Research

FD-OCT is appealing for a new instrument because of its ability to rapidly
acquire images. Instead of imaging a single voxel at a time like TD-OCM, FD-
OCT images an entire A-scan (axial scan along the depth axis) at once. This
enables the microscope to simply scan across the X-Y plane to acquire a three
dimensional image. SS-OCT and SD-OCT have been reporting 100 kHz A-scan

3
rates routinely. A new type of swept source, the Fourier-domain mode-locked
(FDML) laser, allows for megaHertz A-scan rates, allowing billions of voxels
to be imaged every second. However, because of the tradeoff between speed
and image quality,1 these increases have resulted in decreases in the signal to
noise ratio. While many instruments tout their remarkable lateral resolution
when imaging the surface of a sample, they ignore the fact that their lateral
resolution is reduced by an order of magnitude at the bottom of their image.2
This image quality may work for some applications, but our group desires an
OCT microscope that can achieve uniform, high resolution images throughout
a sample, like our current TD-OCM instruments achieve.

2.3 Slab-Based FD-OCT

We have identified a novel design that enjoys the faster acquisition time and
SNR benefits of imaging FD-OCT systems while maintaining the uniform high
resolution of OCM. Instead of imaging the entire depth of an image with an
A-scan, we limit our scan depth, imaging a three dimensional slab by scanning
over the X-Y plane. We then move our focus plane down by the depth thickness
of our slab and image another slab, repeating until our stacked slabs combine
to form a complete image. While the procedure of acquiring multiple slabs
increases the image acquisition time, it also increases the SNR and resolution
throughout the image. A Slab-Based FD-OCT system can be implemented with
either SS-OCT or SD-OCT.

3 Theory
3.1 FD-OCT
Spectral domain and swept source OCT systems share many governing equa-
tions. Consider an SD-OCT instrument source with a Gaussian profile with a
center wavelength λ0 and a full-width half-maximum (FWHM) spectral width
∆λFWHM . The comparable swept source instrument has a center wavelength
λ0 and a FWHM sweep range ∆λFWHM . In a swept source, we actually sweep
over a larger range than ∆λFWHM to map out the Gaussian spectra. This sweep
range will be called ∆λsweep . The spectral domain equivalent is the bandwidth
of light that the diffraction grating spreads out over the linescan camera ele-
ments. In wavenumber space, the center wavenumber becomes
2π
k0 = (1)
λ0
1 from Equation (39)
2 from Equation (20)

4
and the wavenumber change becomes
 
1 1
∆kFWHM = 2π −
λ0 − ∆λFWHM /2 λ0 + ∆λFWHM /2
8π∆λFWHM
=
4λ20 − ∆λ2FWHM
2π∆λFWHM
≈ (2)
λ20
2
where in the last step we neglect (∆λFWHM ) with respect to λ20 . Similarly,
2π∆λsweep
∆ksweep ≈ (3)
λ20

The laser intensity profile S(k), which we’ll assume to be a Gaussian, is generally
described by its variance σk2 in the equation
1 2 2
S(k) = p 2
e−(k−k0 ) /2σk (4)
2πσk

We now wish to take the Fourier transform of this intensity spectrum to find
the coherence function γ(z).
Z ∞
γ(z) = S(k)eikz dk
−∞
Z ∞
1 2 2 2
= p 2
e−(k −2kk0 +k0 )/2σk eikz dk
2πσk −∞
Z ∞
1 2 2 2 2
= p 2
e−(k −2kk0 +k0 −2iσk kz)/2σk dk
2πσk −∞
Z ∞
1 2 2 2 2 4 2 2
= p 2
e−(k−(k0 +iσk z)) /2σk e(2ik0 σk z−σk z )/2σk dk
2πσk −∞
2 2
= e−σk z /2 ik0 z
e (5)

where the second term describes the fringes. We notice that we can relate the
variance of the coherence function σz2 to the variance of the spectrum σk2 by
1
σz2 = (6)
σk2

To relate the FWHM spectral width of the source to σk2 ,

S(k0 + ∆kFWHM /2) = S(k0 )/2

2 2 2 2 1
e−(∆kFWHM /2) /2σk
= e−∆kFWHM /8σk =
2
2
∆kFWHM = 8 ln 2 · σk2

5
and similarly,
2
∆zFWHM = 8 ln 2 · σz2
We plug these into Equation (6) to find

1 8 ln 2 4 ln 2
∆zFWHM = = (7)
2 ∆kFWHM ∆kFWHM
where the extra factor of one half comes from the transition from optical path
length in the interferometer arm, where light passes through the sample twice,
to physical depth in the sample. Plugging in Equation (2), we find

2 ln 2 λ20
∆zFWHM = (8)
π ∆λFWHM
These calculations assumed that the light source was propagating through a
material with an index of refraction of 1, like air. In tissue, which we assume
due to its water composition has an index of refraction of 4/3, we can say that
λ0 and ∆λFWHM are reduced by 3/4, so the axial resolution becomes

3 3 ln 2 λ20
∆zFWHM, tissue = ∆zFWHM, air = (9)
4 2π ∆λFWHM
We notice that the axial resolution of the microscope is totally dependent on
the spectrum of the source used.
The other parameters of the microscope are governed by the constraints
of confocal geometrical optics. The Gaussian beam is focused by a lens with
a certain numerical aperture (NA). Numerical aperture is typically defined in
optics as
D
NA = n (10)
2f
where n is the index of refraction of the material surrounding the lens (1 in air),
D is the lens diameter, and f is the lens focal length. However, in laser physics,
lasers have a Gaussian profile and don’t use the full diameter of the lens. Thus,
we only care about 86% spot size of the incident beam w1 :
w1
NA = (11)
f

where 2w1 is the diameter of the beam, measured between the 1/e2 intensity
points3 and we assume that n = 1. From Gaussian optics, we can relate the
spot size w1 of a laser of wavelength λ0 at the input of the lens to the minimum
(focussed) spot size w0 at the output:

f λ0
w0 = (12)
πw1
3 This could be described as “Full width at e2 maximum of the intensity”

6
Combining Equation (13) and Equation (12), we see that
λ0
NA = (13)
πw0
Solving for the spot size w0 , we see that
λ0
w0 = (14)
π · NA
which leads to a lateral FWHM (see Equation (18)) of
√
2 ln 2 λ0 λ0
∆xFWHM = ≈ 0.375 (15)
π NA NA
which we define as the lateral resolution of our system [11]. If we let ∆xFWHM
become a fundamental quantity of our setup (i.e. a quantity we define from
which all others are derived) along with λ0 and ∆λFWHM , then Equation (15)
determines the numerical aperture of our lens.

3.2 Rayleigh Length

Ideally, the lateral resolution would remain constant throughout the sample.
However, the beam is only focused in one depth plane, and it spreads at other
depths, dictated by the propagation of intensity of a Gaussian beam. The time-
averaged intensity distribution of a Gaussian beam with r as the radial distance
from the center axis of the beam and z is the axial distance from the beam’s
narrowest point (the waist) is
2
−2r2
  
w0
I(r, z) = I0 exp (16)
w(z) w2 (z)
where w(z) is the spot size, the radius where the intensity drops off to 1/e2 of
its axial value, and w0 is w(0). Looking only at the waist,
−2r2
 
I(r) = I0 exp (17)
w02
The radius where the intensity drops off to 1/2 its axial value, r1/2 is
I0 p
I(r1/2 ) = ⇒ r1/2 = w0 ln 2/2
2
Thus the lateral full-width-at-half-maximum ∆xFWHM = 2r1/2 is given by
√
∆xFWHM = w0 2 ln 2 (18)
The Rayleigh length (or Rayleigh range) in air, defined as the z coordinate
where the beam area has doubled, is given by
πw02 π ∆x2FWHM
zr = = (19)
λ0 2 ln 2 λ0

7
In addition, we can then see that, like spot size, our lateral resolution increases
as s  2
z
∆xFWHM (z) = ∆xFWHM 1 + (20)
zr
Note that in tissue, the effective central wavelength is a factor of 4/3 smaller,
so our Rayleigh length becomes 4/3 larger because of Equation (18) and Equa-
tion (12).

3.3 Slab-Based FD-OCT

Slab-Based FD-OCT intends to keep ∆xFWHM small throughout the scan, so
we limit our data acquisition to the slab one Rayleigh length above and below
the focus. Thus we want our total imaging length for one slab to be 2zr .
Assume that the detector samples data that is linear in wavenumber; this can
be done with a k-trigger (only in swept source) or computational re-sampling.
Let us define M as the number of samples taken per A-scan. For SS-OCT, the
analog to digital converter (ADC) sampling rate fs on average equal to

fs = fsweep · M (21)

where fsweep is the laser sweep frequency. However, since a swept source’s output
wavelength is linear in time, a k-trigger would sample nonlinearly in time. We
can calculate the minimum and maximum trigger frequencies to be
 
∆λsweep
fmin ≈ fsweep · M 1 − (22)
λ0
 
∆λsweep
fmax ≈ fsweep · M 1 + (23)
λ0
where we kept terms up to first order in (∆λsweep /λ0 ) 4 . For SD-OCT, M is
equal to the number of pixels we read off of the line scan camera array. Since we
have M samples linearly spaced in wavenumber, we can define the wavenumber
spacing δk:
∆ksweep 2π ∆λsweep
δk = = (24)
M M λ20
From the Nyquist sampling theorem which arises as an effect of finite sampling,
we calculate the distance between measured z data points δz:
π λ20
δz = = (25)
∆ksweep 2∆λsweep
and, completing the circle, the scanning range ∆zscan ,
π M λ20
∆zscan = = = M δz (26)
δk 2∆λsweep
4 For calculation details, see John Grasel’s Lab Notebook Pages 132 and 133

8
Now setting the scanning range ∆zscan equal to twice the Rayleigh length in air
zr , we calculate the implied M :

8π ∆λsweep ∆x2FWHM
M= (27)
3 ln 2 λ30

With the sources available today, achieving a 5µm lateral resolution results
in an M around 10-30, much lower than most OCT systems. Most SD-OCT
systems use a line-scan camera array with 1024, 2048, or 4096 pixels (M ). Thus,
implementing Slab-Based OCT with SD-OCT means wasting the majority of
the pixels on an expensive camera. However, with SS-OCT, a small M means
we can get away with a slower, cheaper analog to digital converter.
So far, we have assumed that the system has infinite spectral resolution. In
the case of SS-OCT, however, our swept source laser has a nonzero instantaneous
line-width. For SD-OCT, the spectrometer’s pixels have finite size. This finite
resolution causes a sensitivity falloff factor as we sample further away from the
plane in focus. If we define the spectral width as δλFWHM , the distance away
from the focus at which the sensitivity is half that of the plane in focus (6 dB)
z6dB is [11]
ln 2 ∆λ2FWHM
z6dB = (28)
π δλFWHM
in air.

3.4 Signal to Noise Ratio

In addition, we care about the signal to noise ratio of the system. We’ll see the
advantage of new FD-OCT systems over TD-OCT.

3.4.1 Frequency-Domain SNR

In addition, we care about the signal to noise ratio of the system. For simplicity,
we look at a single reflector located at z0 , we use a single detector, and we
only analyze a swept-source system with a rectangular spectral profile5 . If we
examine the signal current is over time, if ER and ES are the electric fields
returning from the sample and reference arms, is equal to
ηq 1 D 2
E
is (t) = |ER + ES | (29)
E 2
where the factor of two reflects the second pass of each field through the beam-
splitter and the angular brackets denote integration over the response time of
the detector; additionally, η is the quantum efficiency of the detector, q is the
electric charge, E is the energy of a single photon, such that the responsivity
5 The SD-OCT derivation is very similar to the SS-OCT derivation, and it has the same

final result. For the dual balanced detection, see Page 213 of [11]

9
of the detector is the first term of the above equation. If we define P0 as the
average laser power and k(t) is our laser sweep, then our laser’s electric field is
p
E(t) = P0 ei(k(t)z−w(t)t) (30)

Defining the reference reflector’s position as zR , and thus the sample position
at zS = zR − z0 , then we find that

E(t) p
ER (t) = √ RR ei2k(t)zR (31)
2
and
E(t) p
ES (t) = √ RS ei2k(t)zS (32)
2
where the factor of radical two is due to the beamsplitter, and RR and RS are the
power reflectivities of the reference and sample arms respectively (so their square
root is the electric field reflectivities). Combining these with Equation (30) and
Equation (29), integrating over temporal angular frequency w since it oscillates
much faster than any detector,
 
ηq 1 p p 2
is (t) = P0 /2RR e−wt + P0 /2RS e−2iz0 k(t)−wt
E 2
ηq P0  p  
= RR + RS + RR RS e2iz0 k(t) + e−2iz0 k(t)
E 4
ηq P0  p 
= RR + RS + 2 RR RS cos (2z0 · k (t)) (33)
E 4
We take the inverse discrete Fourier transform of is (t) to yield Fs (z):
M
X
Fs (z) = is (t)eik(t)z/M (34)
m=1

If we look at the case of equal paths in the sample and reference arms, where
z0 = 0, we see that
ηq P0 p
Fs (z0 ) = RR RS M (35)
E 4
as the signal power adds coherently. Any other choice of z0 will give rise to
phase factors but will still coherently sum to the same signal peak. The peak
signal power is
 ηq 2  P 2
0
Fs (z0 )2 = RR RS M 2 (36)
E 4
The noise current for shot-noise limited detection is
ηq 2 P0 ηq 2 fsweep
 
2
hin (t)i = RR BW = P 0 RR (37)
E 2 4E

10
where in the last step we chose the the detection bandwidth BW as specified by
the Nyquist frequency (= fs /2) as defined in Equation (21). Taking the Fourier
transforms of in yields Fn . By Parseval’s theorem,

ηq 2 fsweep
hFn2 i = M hi2n i = P0 RR M (38)
4E
which scales linearly with M because the noise in each spectral channel is in-
coherent. We define SNR as follows, substituting in Equation (36) and Equa-
tion (38), and setting the sample reflectivity to one:

|Fs (z0 )|2 ηP0 M ηλ0 P0 M 2π η P0 ∆λsweep ∆x2FWHM

SNR = = ≈ =
hFn2 i 4Efsweep 4hcfsweep 3 ln 2 hc λ20 fsweep
(39)
where h and c are Planck’s constant and the speed of light, respectively. We
further define sensitivity in dB as

Sensitivity = 10 log (SNR) (40)

However, with a Gaussian-shaped source spectrum clipped at 1/e2 points, the

signal to noise ratio is only 0.598 times as strong.

3.4.2 Time-Domain SNR

The analog of Equation (33) in the time domain is
ηq P0  p 2 2

iTDOCT (zR ) = RR + RS + 2 RR RS e−(zR −z0 ) ∆kFWHM cos (2(zR − z0 ) · k0 )
E 4
(41)
Looking at the third term where the OCT signal resides, we see the peak signal
occurs at zR = z0 , where
 ηq 2  P 2 R R
0 R S
iTDOCT (zR )2 = (42)
E 2 2
If we achieve shot-noise limited performance, which is certainly not always the
case, then the noise is given by
  
2 ηq P0 1
σTDOCT = 2qIB = 2q RR (43)
E 2 ∆t
where ∆t is the integration time suggested by Nyquist sampling of the band-
width of the detector. The signal to noise ratio is
ηP0 ∆t
SNRTDOCT = (44)
2E
Since the integration time for SS-OCT is equal to the inverse of its sweep rate,
we calculate that FD-OCT has an M/2 improvement in SNR over TD-OCT with
a rectangular spectrum. This is a result of the fact that FD-OCT instruments

11
sample all depths all the time, leading to a potential SNR improvement of
M . However, typical OCT methods give rise to complex conjugate ambiguity,
leading to the additional factor of one half. Instruments that use direct IQ
sampling would have twice the signal to noise ratio; frequency shifting methods
have an M that is twice as large.

3.5 Non-Gaussian Spectra

Many swept sources have rectangular spectra, the governing equations of which
are slightly different. Let us say that our source has a wavelength center λ0 and
a width of ∆λrect . We can represent our spectrum i(k) as the difference of two
Heaviside functions

i(k) = H(k − k0 + ∆krect /2) − H(k − k0 − ∆krect /2)

the Fourier transform of which is

r  
1 ik0 z ∆krect z
F (z) = ∆krect e sinc
2π 2

We define the resolution ∆z as the width of the main lobe of the function (i.e.
until the function reaches 0 at ±π). Thus we find that the axial resolution in
air is
4π 2λ20
∆z = = (45)
∆krect ∆λrect
which is considerably worse than the resolution given by a Gaussian spectra in
Equation (8).
Table 1 summarizes the most important formulas from this section. Three
swept sources at different wavelengths, all manufactured by Exalos, are shown
for reference.

4 Specific Design Challenges and Solutions

Besides the choice of laser in SS-OCT or source and detector array in SD-OCT,
there are other elements of an FD-OCT system that can be added.

4.1 Complex-Conjugate Ambiguity

Recall that our scanning range for a single slab is a Rayleigh length in both
directions from the focal plane, with a resulting thickness of 2zr . Additionally,
the Fourier transform of the fringe data moves it from the wavenumber domain
to the position domain with data points at

−∆zscan /2, − ∆zscan /2 + δz, ... , ∆zscan /2

However, a photodetector records a real signal, and the Fourier transform of

a real signal in k-space results in a Hermitian signal in position space. This

12
causes a single reflector located at z relative to the focal plane to appear in two
places in our slab, at −z and z, rendering the resulting image useless. This was
not a problem in traditional OCT systems, which traditionally have placed the
focal plane at the top surface of the sample and discarded the Fourier position
data from above the sample. This cuts the scanning range in half, but resolves
the ambiguity. However, because we are imaging slabs inside the sample, we
cannot set their focal planes at the surface, so we must resolve the ambiguity.
Some research has been directed at resolving the ambiguity, which allowed other
teams to place their focal planes in the middle of their samples and benefit from
increased lateral resolution and SNR. Several methods have been implemented
to resolve the ambiguity, but there are two main principles. The first is the use of
quadrature interferometry to extract the in-phase and quadrature components
of the signal. The second shifts the optical frequency and samples a real signal,
and performs hardware or software demodulation.
One of the simplest methods of quadrature interferometry is phase stepping.
Typically, five A-scans are imaged serially while incrementing the phase shift
by π/2, time-encoding the real and imaginary components of the signal. The
phase shift has been introduced with piezo-electric fiber stretchers and piezo
translaters [37, 35]. This method has the downside of requiring several mea-
surements to resolve the ambiguity, a carefully calibrated reference arm, and
total phase stability within the system 6 . Similarly, the real and imaginary
components can be encoded with different polarization, which has the bene-
fit of being instantaneous but the downside of having a complicated free-space
(fibreless) setup and suffering from polarization fading [10].
More recent methods of quadrature interferometry are based on 3x3 fiber
couplers. This method on Michelson interferometers using both SS-OCT and
SD-OCT [5, 34], a Mach-Zehnder interferometer [12], and unbalanced detection
[21, 22]. These designs are simple, but tend to only achieve a conjugate rejection
of between 20 and 30 dB due to the difficulty of producing stable quadrature
signals.
Another method shifts the peak sensitivity position away from electronic
DC, so positive and negative displacements from that position can be discerned
unambiguously. Because this technique shifts the complex conjugate, it achieves
a complete conjugate rejection (up to the noise floor). As an additional advan-
tage, the DC and autocorrelation artifacts remain centered at DC and can be
removed with a high-pass filter. Frequency shifting has been implemented with
electro-optic modulators [13] and acousto-optic modulators, but these are ex-
pensive, introduce losses into the system, and require hardware demodulation
or software post-processing. A recent instrument uses a dispersive optical delay
line, which achieves the same benefits but has no sensitivity loss, is low cost,
and is easy to implement [9]. The downside to both of these systems is that,
because they shift position information to a lower depth, a higher sampling rate
is required.
A new technique introduces a dispersion mismatch between sample and ref-
6 Error can be introduced by interferometer drift between phase-shifted acquisitions

13
erence arm that can be used to iteratively suppress complex conjugate artifacts
and thereby increase the imaging range using a fast dispersion encoded full range
(DEFR) algorithm [16, 17, 15]. It achieved a rejection of 55dB on average, but
is computationally intensive.

4.2 Bessel Beams

Another technique to extend the depth of focus is the use of Bessel beams instead
of Gaussian spherical beams. A conical lens called an axicon forms a Bessel beam
in the near field when illuminated with a standard Gaussian spherical beam; it
forms a ring in the far field. Unlike a Gaussian beam, whose spot size increases
(and thus lateral resolution decreases) away from the focal plane, an ideal Bessel
beam’s spot size stays constant. This results in constant lateral resolution as
a function of depth[20]. Additionally, sensitivity stays constant as well. Bessel
beams are self-reconstructing, so opaque obstacles do not create shadows on
the image[14]. Finally, to avoid losses resulting from the backscattered beam
passing through the axicon twice, a dark-field setup is used, which excludes the
unscattered beam from the image and increases image contrast[3]. Bessel beams
have been used in both SD-OCT and SS-OCT at A-Scan rates of up to 440kHz
[3]. The Bessel beam’s intensity as a function of depth z and radial coordinate
r is given by
2πk sin β 2
I(r, z) = E(Rz )2 Rz J (kr sin β) (46)
cos2 β 0
where J0 is the zero-order Bessel function of the first kind, Rz is the radius
of the incident beam that contributes to the intensity at z, E is the energy
of the incident beam at that position, and β is an axicon parameter which is
determined by the index of refraction of the lens n and the axicon’s conical angle
α: β = sin−1 (n sin α) − α.
The Bessel beam’s advantages arise from greatly extending the exponential
falloff of the incident Gaussian beam, since Rz grows much slower than z due to
the geometry of an axicon. However, there are many disadvantages that result
from the Bessel function in Equation (46). The Bessel function, and thus an
ideal Bessel beam, has an infinite number of side lobes. Each of these lobes
has a roughly equal amount of power, because the intensity falls off inversely to
the radius squared, but area grows as the radius squared [14]. Thus, an ideal
Bessel function requires infinite energy; since this is impossible, only approxi-
mations are used, which result in actual Bessel beams with finite depth of field
and and slightly-varying sensitivity as a function of depth. Even with these ap-
proximations, the central lobe of a Bessel field carries around 5% of the power
[20] (compared with 86% of a Gaussian beam within a 1/e2 radius), and the
rest of the power in the side-lobes results in a blurry axial resolution. While
the sensitivity is constant, the distribution of power throughout a depth scan
reduces the SNR around 20 dB compared to a similar Gaussian system [19]. A
dark field detection setup is complex to implement. While the Bessel beam fits
many of our design requirements, it introduces many additional problems.

14
4.3 Gabor-Domain OCT
Another group, headed by Jannick P. Rolland, had similar design considerations
and built a SD-OCT system using dynamic focusing [26]. A liquid lens, which
changes its focal length on application of an electric field, was used was used to
refocus to achieve invariant 3µm resolution throughout a 2mm cube [25]. The
microscope used a Gabor Domain OCT (GD-OCT) scheme, an algorithm was
used to extract only the in-focus portion of the acquired image, by multiplying
the image with a sliding window centered at the microscope’s focal plane with
a width as the depth of focus. Combining this technique with dynamic focusing
allows the extraction of the slabs of the cross section image around each focus
position, which are then fused to form an image with nearly invariant resolution
[33]. The lens, which only takes 100 ms to focus between slabs, is a small
improvement over our actuators. However, its focal positioning is imprecise
and must be calculated in software, and it is expensive. The GD-OCT image
stitching technique is computationally trivial and can easily be performed in
real-time.

4.4 ISAM
Interferometric synthetic aperture microscopy, or ISAM, is a post-processing
modality that achieves depth-independent lateral resolution throughout a vol-
ume imaged with a fixed focus. ISAM uses similar methods as synthetic aper-
ture radar, or SAR, which combines radar signals from a target illuminated from
many directions into a single image. Likewise, ISAM combines all of the A-scans
that compose a slab to produce a single image with invariant lateral resolution.
The model assumes a scalar model for light propagation that obeys the reduced
wave equation [32]. This model is used to propagate a focused beam into the
sample and consider the cross-correlation of singlely backscattered photons with
the reference pulse. This is used to establish the relationship between the scat-
tering potential and measured signal. ISAM is an inverse scattering procedure
to account for these effects [4].
After dispersion compensation, a spacial 2-D Fourier transform is performed
over all of the A-scans composing a slab7 . The data is shifted axially down in
wavenumber, and then shifted and diffracted up, compensating for diffraction
and moving the focus without net sample movement. A Stolt interpolation
maps the wavenumber to axial special frequency, and an inverse 3-D Fourier
transform recovers the image [27, 28, 6]. A multiplicative term may be applied
to compensate for signal loss away from focus, but there is no way of avoiding
the drop in SNR. The benefit of multiple slabs would then be to maintain SNR
throughout the image. Also, because data within a slab is combined, the system
must have complete phase stability or numerically correct phase disturbances
using a cover slip [29]. While this extra processing is complex and takes a lot of
time, it is parallelizable and feasible to perform in real-time [31]. In addition,
7 For more information, see the Projection-Slice Theorem

15
recent research suggests additional post-processing may reduce autocorrelation
artifacts [8].
For a rigorous mathematical overview, see [7]. For details on the algorithm’s
implementation, see [24]. Additionally, a slightly different adaptation of ISAM
was made for high-NA systems (OCM) [30]. The theoretical model was con-
firmed with experimental evidence [28].

4.5 Dispersion Compensation

Dispersion in the sample or any optical delay line can lead to loss of axial res-
olution. Dispersion is caused by the group velocity in a medium depending
sensitively on the wavelength of light, which is problematic when the source
bandwidth is large. Hardware dispersion compensation matches the dispersion
in the sample and reference arms by inserting water or glass in the reference
arm, but this is ineffective for 3-D imaging because the amount of dispersion in
the sample is usually depth-dependent. However, from Parseval’s Equality, dis-
persion does not degrade the power spectrum, so it can be corrected in software
without any quality consequences. The Boppart group created a method that
minimizes the entropy in the image, which enhances the contrast [23]. This al-
gorithm was extended to FD-OCT by Fujimoto’s group [38], but the algorithm
is iterative. A non-iterative algorithm uses phase information in the interfer-
ogram to calculate a generalized autoconvolution function, but the algorithm
only compensates up to second-order [2]. However, this might be sufficient for a
FD-OCT setup with a relatively small tuning range. These algorithms appear
to be parallelizable, and one has been ported to the GPU [39].

4.6 Phase Stability

Phase stability within a slab is essential for Doppler imaging and ISAM. While
a standard Michelson interferometer is immune to phase jumps in the source
(with the exception of FDML lasers, which have slow phase drift [1]), it is sus-
ceptible to phase changes between the reference and sample arms caused by
thermal drift, as well as phase changes between A-scans caused by inconsistent
sampling [36]. The solution to the former problem is through a common-path
interferometer, a second reference arm that shares a majority of the sample
arm’s fiber. This adds an additional reflector to the image. If this reflector is
positioned to be above the top or below the bottom of the slab, we can slightly
increase M to image the calibration reflector. After taking the Fourier trans-
form of each A-scan to move into the position domain, the reflector’s phase can
be subtracted from each other phase to stabilize the phase between successive
A-scans. While this isolates most of the sample arm from thermal drift rela-
tive to the reference arm, thermal drift between the calibration arm and the
sample head can still cause phase instabilities. Some form of thermal insulation
could keep these to a minimum. This solution, known as inserting common path
into the instrument, adds relatively little additional complexity to the instru-
ment and the post-processing. Another solution uses a gas-cell as a spectral

16
reference to increase sensitivity and stability, but the system is complicated to
implement[18].
Even with the common path approach, phase instabilities can be introduced
from inconsistent sampling. If no k-trigger is used, there is no way of assuring
that the linear sampling starts at the same wavelength every time. Even if
DAQ is triggered to take M samples every sweep, the precision of when the
sampling clock starts cannot be more fine than the internal sample clock rate
of the converter. Thus, the phasing of the swept source output and the sample
clock drifts ofter time, causing the trigger signal to fall arbitrarily within the
sample clock cycle. This results in random shifts of the interferogram in k-
space, which by the Fourier Shift Theorem leads to a phase shift in position
space. The solution is to force the clocks of the swept source and the analog to
digital converter to be perfectly synchronous; using a k-trigger, in addition to
reducing other post-processing, realizes this solution.

5 Overall Instrument Design

5.1 Hardware and Data Acquisition
One way in which this is different from our current OCM instruments is that it
is more difficult to oversample in the Z direction. While before we specified the
Z step size, now our Z step size is determined by ∆zFWHM . However, we can
oversample by a factor of n if we acquire n slabs spaced ∆zFWHM /n from each
other.

5.2 Signal Processing

The signal processing steps depend on design decisions such as the method of re-
solving the complex conjugate ambiguity and the use of a k-trigger in sampling.
If the complex signal is directly measured, the first step is to subtract the back-
ground DC spectrum. If a k-trigger was not used, the signal must be resampled
in linear wavenumber. This is done through interpolating a signal upsampled
in the frequency domain, through the use of a forward Fourier transform, zero
padding, and an inverse Fourier transform.
Frequency shifting leaves the DC and autocorrelation artifacts centered at
DC; since the signal of interest was shifted above DC, these artifacts can be
removed with a hardware or software low-pass filter. However, the signal must
be demodulated to be centered at DC. First, the Hilbert transform is applied
to the signal to remove the conjugate artifact and eliminate aliasing when the
signal is shifted to baseband. The shifting is performed through mixing the
signal with the negative carrier (modulation) frequency in the time domain, or
through the shifting of indices in the frequency domain. Now, since the signal
occupies less than half its original bandwidth, it is oversampled by a factor
greater than two. Thus, if no k-trigger was used, we need no signal upsampling
before the linear wavenumber interpolation. However, if a k-trigger was used,

17
Figure 1: Diagram of Signal Processing Steps

18
we must reduce the signal’s bandwidth through decimation. In the frequency
domain, decimation is applied by removing high-frequency components; in the
time domain, decimation by a factor of M is a downsampling of the signal by
picking out every M th sample.
The result of these steps is a complex interferometric signal sampled at
Nyquist in linear wavenumber whose Fourier transform is the axial reflectivity
of a the sample at a particular x-y location and depth range.

6 Slab-Based OCT Performance

It is difficult to directly compare FD-OCT systems to TD-OCT systems be-
cause the Fourier transform from k-space to position space directly determines
Z spacing. Thus, we can no longer specify a 100x100x100 image; instead, we
specify that we want 100 X voxels, 100 Y voxels, and a Z scan 500µm long.
The scan depth, along with the slab depth thickness (the source’s 2zr Rayleigh
length), determine the number of slabs and the points per slab, and thus the
number of total Z voxels.
While the sweep frequency fsweep is part of what determines the speed of
a scan, there are several mechanical factors that limit how quickly data can
be acquired. The galvanometers that scan in the XY-plane take roughly 500µs
to achieve linear motion, so any data acquired during that time is bad. The
actuators take time to move and settle. A safe amount of wait time W (z) in
milliseconds where z is the distance to move in µm was empirically determined
to be: 
3z + 60 : z ≤ 40
W (z) =
2z + 100 : z ≥ 40
We’ll calculate how long each step in the process takes to image Nslabs slabs of
dimension Nx x Ny . The time taken to acquire a single Z scan tZ for a slab is
1
tZ = (47)
fsweep
so a slab Z-X scan takes
tZX = tZ Nx (48)
If we repeat each Z-X scan Xscans times, and we wait for the galvos a time tgalvo
after every Z-X scan except the last8 , then the time taken to image one slab is

tZXY = tZX Ny Xscans + tgalvo (Ny Xscans − 1) (49)

and our total image acquisition time tscan is

tscan = tpurge + tZXY Nslabs + W (z) (Nslabs − 1) (50)

where tpurge the time we wait at the beginning of every scan to flush the analog
to digital converter buffers.
8 This time sill be absorbed by the actuator wait time

19
 From here, we can generate a table of how long each instrument takes to
image various scans. Additionally, we can break down this time and see how
long is spent on what task. Table 2 shows how long it takes to acquire a
100x100xZ cube with 500µm depth, compared with a standard 100x100x100
voxel cube in Michelson and Fabry. Despite the fact that we waste over half of
the imaging time not collecting data, we still see huge speed improvements over
our current instruments. For comparison, Michelson and Fabry waste only 15%
of the voxels imaged. However, this small scan size is not making great use of
the speed increases of FD-OCT.
If we use a larger scan, acquiring a acquire a 200x200xZ cube with 1000µm
depth, we see much more speed improvement. We compare this large scan to
a 200x200x200 voxel cube on Michelson and Fabry. The results are shown in
Table 3. The 840nm instrument completes this scan in just 21 seconds, well
below our goal of one minute. We waste only 30% of our image acquisition
time, and over half of that is waiting for the galvos.

20
7 References
References
[1] Desmond C. Adler, Robert Huber, and James G. Fujimoto, Phase-sensitive
optical coherence tomography at up to 370,000 lines per second using
buffered fourier domain mode-locked lasers, Opt. Lett. 32 (2007), no. 6,
626–628.
[2] Konrad Banaszek, Aleksandr S. Radunsky, and Ian A. Walmsley, Blind
dispersion compensation for optical coherence tomography, Optics Commu-
nications 269 (2007), no. 1, 152 – 155.
[3] Cedric Blatter, Branislav Grajciar, Christoph M. Eigenwillig, Wolfgang
Wieser, Benjamin R. Biedermann, Robert Huber, and Rainer A. Leitgeb,
Extended focus high-speed swept source oct with self-reconstructive illumi-
nation, Opt. Express 19 (2011), no. 13, 12141–12155.
[4] P.S. Carney and S.A. Boppart, Depth of field and resolution meet in new
oct approach, BioOptics World (2011), 19–23.

[5] Michael Choma, Marinko Sarunic, Changhuei Yang, and Joseph Izatt, Sen-
sitivity advantage of swept source and fourier domain optical coherence to-
mography, Opt. Express 11 (2003), no. 18, 2183–2189.
[6] B.J. Davis, T.S. Ralston, D.L. Marks, S.A. Boppart, and P.S. Carney,
Interferometric synthetic aperture microscopy: Physics-based image recon-
struction from optical coherence tomography data, Image Processing, 2007.
ICIP 2007. IEEE International Conference on, vol. 4, 16 2007-oct. 19 2007,
pp. IV –145 –IV –148.
[7] Brynmor J. Davis, Daniel L. Marks, Tyler S. Ralston, P. Scott Carney, and
Stephen A. Boppart, Interferometric synthetic aperture microscopy: Com-
puted imaging for scanned coherent microscopy, Sensors 8 (2008), no. 6,
3903–3931.
[8] Brynmor J. Davis, Tyler S. Ralston, Daniel L. Marks, Stephen A. Bop-
part, and P. Scott Carney, Autocorrelation artifacts in optical coherence
tomography and interferometric synthetic aperture microscopy, Opt. Lett.
32 (2007), no. 11, 1441–1443.
[9] Al-Hafeez Dhalla and Joseph A. Izatt, Complete complex conjugate resolved
heterodyne swept-source optical coherence tomography using a dispersive
optical delay line, Biomed. Opt. Express 2 (2011), no. 5, 1218–1232.

[10] Zhihua Ding, Yonghua Zhao, Hongwu Ren, J. Nelson, and Zhongping Chen,
Real-time phase-resolved optical coherence tomography and optical doppler
tomography, Opt. Express 10 (2002), no. 5, 236–245.

21
[11] W. Drexler and J.G. Fujimoto, Optical coherence tomography: Technology
and applications, Biological and medical physics, biomedical engineering,
Springer, 2008.
[12] C Flueraru, H Kumazaki, S Sherif, S Chang, and Y Mao, Quadrature
machzehnder interferometer with application in optical coherence tomog-
raphy, Journal of Optics A: Pure and Applied Optics 9 (2007), no. 4, L5.
[13] Erich Götzinger, Michael Pircher, Rainer Leitgeb, and Christoph Hitzen-
berger, High speed full range complex spectral domain optical coherence
tomography, Opt. Express 13 (2005), no. 2, 583–594.
[14] R. M. Herman and T. A. Wiggins, Production and uses of diffractionless
beams, J. Opt. Soc. Am. A 8 (1991), no. 6, 932–942.
[15] Boris Hermann, Bernd Hofer, Christoph Meier, and Wolfgang Drexler,
Spectroscopic measurements with dispersion encoded full range frequency
domain optical coherence tomography in single- and multilayered non- scat-
tering phantoms, Opt. Express 17 (2009), no. 26, 24162–24174.
[16] Bernd Hofer, Boris Považay, Boris Hermann, Angelika Unterhuber, Ger-
ald Matz, and Wolfgang Drexler, Dispersion encoded full range frequency
domain optical coherence tomography, Opt. Express 17 (2009), no. 1, 7–24.
[17] Bernd Hofer, Boris Považay, Angelika Unterhuber, Ling Wang, Boris Her-
mann, Sara Rey, Gerald Matz, and Wolfgang Drexler, Fast dispersion en-
coded full range optical coherence tomography for retinal imaging at 800
nm and 1060 nm, Opt. Express 18 (2010), no. 5, 4898–4919.
[18] R.V. Kuranov, A.B. McElroy, N. Kemp, S. Baranov, J. Taber, M.D. Feld-
man, and T.E. Milner, Gas-cell referenced swept source phase sensitive op-
tical coherence tomography, Photonics Technology Letters, IEEE 22 (2010),
no. 20, 1524 –1526.
[19] Kye-Sung Lee and Jannick P. Rolland, Bessel beam spectral-domain high-
resolution optical coherence tomography with micro-optic axicon providing
extended focusing range, Opt. Lett. 33 (2008), no. 15, 1696–1698.
[20] R. A. Leitgeb, M. Villiger, A. H. Bachmann, L. Steinmann, and T. Lasser,
Extended focus depth for fourier domain optical coherence microscopy, Opt.
Lett. 31 (2006), no. 16, 2450–2452.
[21] A. Maheshwari, M. A. Choma, and J. A. Izatt, Heterodyne swept-source
optical coherence tomography for complete complex conjugate ambiguity re-
moval, 5690 (2005), 91–95.
[22] Youxin Mao, Sherif Sherif, Costel Flueraru, and Shoude Chang, 3x3 mach-
zehnder interferometer with unbalanced differential detection for full-range
swept-source optical coherence tomography, Appl. Opt. 47 (2008), no. 12,
2004–2010.

22
[23] Daniel L. Marks, Amy L. Oldenburg, J. Joshua Reynolds, and Stephen A.
Boppart, Autofocus algorithm for dispersion correction in optical coherence
tomography, Appl. Opt. 42 (2003), no. 16, 3038–3046.
[24] Daniel L. Marks, Tyler S. Ralston, Brynmor J. Davis, P. Scott Carney, and
Stephen A. Boppart, Interferometric synthetic aperture microscopy: tissue
structure inferred by computed imaging techniques, 6864 (2008), no. 1,
686407.
[25] Panomsak Meemon, Supraja Murali, Kye sung Lee, and Jannick P. Rolland,
Gabor domain optical coherence microscopy, Frontiers in Optics, Optical
Society of America, 2008, p. FWW4.

[26] S. Murali, K. S. Lee, and J. P. Rolland, Invariant resolution dynamic focus

ocm based on liquid crystal lens, Opt. Express 15 (2007), no. 24, 15854–
15862.
[27] T. S. Ralston, D. L. Marks, P. Scott Carney, and S. A. Boppart, Interfer-
ometric synthetic aperture microscopy, Nature Physics 3 (2007), 129–134.

[28] T.S. Ralston, G.L. Charvat, S.G. Adie, B.J. Davis, P.S. Carney, and S.A.
Boppart, Interferometric synthetic aperture microscopy: Microscopic laser
radar, Opt. Photon. News 21 (2010), no. 6, 32–38.
[29] T.S. Ralston, D.L. Marks, P.S. Carney, and S.A. Boppart, Phase stability
technique for inverse scattering in optical coherence tomography, Biomedi-
cal Imaging: Nano to Macro, 2006. 3rd IEEE International Symposium on,
april 2006, pp. 578 –581.
[30] Tyler S. Ralston, Daniel L. Marks, Stephen A. Boppart, and P. Scott Car-
ney, Inverse scattering for high-resolution interferometric microscopy, Opt.
Lett. 31 (2006), no. 24, 3585–3587.
[31] Tyler S. Ralston, Daniel L. Marks, P. S. Carney, and Stephen A. Boppart,
Real-time interferometric synthetic aperture microscopy, Opt. Express 16
(2008), no. 4, 2555–2569.
[32] Tyler S. Ralston, Daniel L. Marks, P. Scott Carney, and Stephen A. Bop-
part, Inverse scattering for optical coherence tomography, J. Opt. Soc. Am.
A 23 (2006), no. 5, 1027–1037.
[33] Jannick P. Rolland, Panomsak Meemon, Supraja Murali, Kevin P. Thomp-
son, and Kye sung Lee, Gabor-based fusion technique for optical coherence
microscopy, Opt. Express 18 (2010), no. 4, 3632–3642.

[34] Marinko Sarunic, Michael A. Choma, Changhuei Yang, and Joseph A. Izatt,
Instantaneous complex conjugate resolved spectral domain and swept-source
oct using 3x3 fiber couplers, Opt. Express 13 (2005), no. 3, 957–967.

23
[35] Piotr Targowski, Maciej Wojtkowski, Andrzej Kowalczyk, Tomasz Ba-
jraszewski, Maciej Szkulmowski, and Iwona Gorczynska, Complex spectral
oct in human eye imaging in vivo, Optics Communications 229 (2004),
no. 1-6, 79 – 84.
[36] B. Vakoc, S. Yun, J. de Boer, G. Tearney, and B. Bouma, Phase-resolved
optical frequency domain imaging, Opt. Express 13 (2005), no. 14, 5483–
5493.
[37] M. Wojtkowski, A. Kowalczyk, R. Leitgeb, and A. F. Fercher, Full range
complex spectral optical coherence tomography technique in eye imaging,
Opt. Lett. 27 (2002), no. 16, 1415–1417.

[38] Maciej Wojtkowski, Vivek Srinivasan, Tony Ko, James Fujimoto, Andrzej
Kowalczyk, and Jay Duker, Ultrahigh-resolution, high-speed, fourier do-
main optical coherence tomography and methods for dispersion compensa-
tion, Opt. Express 12 (2004), no. 11, 2404–2422.
[39] K. Zhang and J.U. Kang, Real-time numerical dispersion compensation
using graphics processing unit for fourier-domain optical coherence tomog-
raphy, Electronics Letters 47 (2011), no. 5, 309 –310.

24
Table 1: Comparison of Exalos Swept Sources - The specifications for
three swept sources produced by Exalos are shown in the first four rows. The
lateral resolution is also defined. From these five parameters, all other aspects
of the instrument are determined. Note that the instrument at 1310nm has the
highest SNR, the 1050nm has the best resolution, and the 840nm instrument
has the longest Rayleigh range. Trade-offs such as these have to be made in the
design of an FD-OCT system.
Sources Formula 1 2 3

Wavelength (nm) λ0 1310 1050 840

Wavelength Sweep Range (nm) ∆λsweep 100 100 45

Wavelength Sweep FWHM (nm) ∆λFWHM 100 100 45

Sweep Rate (kHz) fsweep 100 100 100

Source Power (mW) P0 10 10 5

Lateral Resolution (µm) ∆xFWHM 5 5 5

2
2 ln 2 λ0
Axial Resolution (air, µm) ∆zFWHM = π ∆λFWHM 7.55 4.85 6.90
3
Axial Resolution (tissue, µm) 4 ∆zFWHM 5.66 3.64 5.17
λ0
Objective NA NA = 0.37 ∆xFWHM 0.097 0.078 0.062
2
4π ∆xFWHM
2x Rayleigh Len (tissue, µm) 2zr = 3 ln 2 λ0 115.3 143.9 179.9
2
8π ∆λsweep ∆xFWHM
Samples per A-Scan M= 3 ln 2 λ30
14 27 23
η λ0 P0 M
SNR (dB) SNR = 4hc fsweep 122.1 124.0 119.3

Avg ADC Sample Rate (MHz) fs = M fsweep 1.40 2.70 2.30

∆λsweep
Max ADC Sample Rate (MHz) fmax = M fsweep (1 + λ0 ) 1.51 2.96 2.42

25
Table 2: Time Comparison of Exalos Swept Sources - Time taken to
acquire a 100x100xZ voxel cube with a 500µm scan depth and 6 X-Scans for
motion sensitivity.
Exalos Instrument 1310 1050 840
Z Voxels 70 108 69
Scan Time (s) 6.52 5.46 4.32
Speedup Over Michelson 49.7 59.3 75.0
Speedup Over Fabry 72.0 86.0 108.7
% Waste to Purge 10.7% 12.8% 16.2%
% Waste to Galvo 23.0% 21.9% 20.8%
% Waste to Actuator 20.3% 21.3% 21.3%
% Not Wasted 46.0% 43.9% 41.7%

Table 3: Time Comparison of Exalos Swept Sources - Time taken to

acquire a 200x200xZ voxel cube with a 1000µm scan depth and 6 X-Scans for
motion sensitivity.
Exalos Instrument 1310 1050 840
Z Voxels 126 189 138
Scan Time (s) 30.3 24.0 21.0
Speedup Over Michelson 78.6 99.3 113.6
Speedup Over Fabry 116.6 147.3 168.6
% Waste to Purge 2.3% 2.9% 3.3%
% Waste to Galvo 17.8% 17.5% 17.1%
% Waste to Actuator 8.7% 9.7% 10.9%
% Not Wasted 71.2% 69.9% 68.6%

26