Multiple path optical coherence tomography using re-circulating loops has previously been presented as a means of simultaneously acquiring images from multiple depths in multiple imaging channels. The configurations reported so far present the drawback that the strength of the signal from one channel to the next, reduced as the number of circulations increased. A decay of signal of not better than 4 dB from one channel to the next was reported. We present a technique to reduce this attenuation by using polarization maintaining fiber, and modulation of the drive current of the semiconductor optical amplifiers contained in each arm. The effect of these improvements resulted in a decay less than 20 dB from the 1st channel to the 10th channel.
©2012 Optical Society of America
Optical Coherence Tomography (OCT)  has become a valuable tool in high resolution imaging, most notably in ophthalmology. Spectral domain (SD) OCT  has obvious benefits when imaging a moving object, in terms of speed and sensitivity over time domain methods. Time domain methods, however, are still applicable to many areas of interest in high resolution imaging, such as in microscopy, where high transversal resolution throughout the whole depth range is not achievable using SD-OCT methods . In addition, even when using the fastest spectral OCT system reported to date, production of an en-face image requires a significant fraction of a second. This may be too long if the object imaged is subject to movement, and therefore multiple path en-face OCT is of particular interest when a number of en-face plane OCT images is required.
The ability to measure multiple points in depth simultaneously or at least near simultaneously is one of the inherent advantages of OCT . It has previously been shown that the combination of multiple active delay lines  in both the reference and object arms of an interferometer may be used to simultaneously provide OCT signal from several depths. For each depth, a signal-processing channel can be identified, where each channel operates as a coherence gate from a unique and different depth by interfering the same number of roundtrip passes in the two interferometer arms. The axial separation between channels is adjustable by altering the difference between the round trip path lengths in the two recirculating loops. In this way, multiple depths en-face OCT images can be acquired with an image for each such channel. A major problem with the previous multiple path configuration was the decay in sensitivity from one channel to the next, larger than 40 dB for the first 10 channels.
Here we present an improved configuration which exhibits less than 20 dB decay over the first 10 channels. The configuration makes use of polarization maintaining fiber and employs the two active delay lines in a non-stationary state , where their gain is temporally modulated. To minimize losses in the re-circulating loops, isolators used in previous configurations are here eliminated. In addition, power modulation was applied to compensate for the loss increase with the number of re-circulations, achieved by synchronous modulation of the semiconductor optical amplifiers and of the frequency shifters in both the reference and object arms.
A simplified schematic diagram of the interferometer system with re-circulating delay lines in both arms is shown in Fig. 1 . The principle of using re-circulating active loops in both the reference and object arms of the interferometer has been applied to configurations driven by a broadband source5 and a swept source . We present here a configuration driven by a broadband source. Light from a super-luminescent diode (SLD, central wavelength 1050 nm, line-width ≈30 nm) is amplified by an Ytterbium doped fiber amplifier (YTT), (equipped with isolators on the input and output) and then split at directional coupler C1. Light entering the reference path is again split at directional coupler C3. Half of the power is sent to a secondary loop where it passes through an air gap, an acousto-optic frequency shifter (AOFS1, 150 MHz, Gooch and Housego, UK), and a semiconductor optical amplifier (SOA1, Superlum, Moscow). While the frequency shifter is open, light can continuously circulate through this loop, half coupling back into the same loop at C3. Half of the output power from the main loop enters a fiber optic circulator at port 1, where returned light from a reference mirror, at port 2 is directed via port 3 to a 50:50 coupler C4. The arrangement is duplicated in the object arm via AOFS2, SOA2, C2 and a circulator. All circulators and couplers used are of a polarization maintaining type (PM) and all fiber used is of PM type, with the exception of the outputs from the SLD and Ytterbium amplifier.
As an improvement to the configuration previously demonstrated , light is routed via polarization maintaining (PM) fiber. The AOFS1, AOFS2, SOA1 and SOA2 are also equipped with PM fiber and all couplers, C1 to C4, incorporate PM fiber with the fast axis being blocked in the couplers. In addition, polarizers and circulators are included at the outputs of the object and reference beams. The closed loops containing SOAs and AOFSs operate like swept sources , the difference being that the bandwidth of the signal re-circulated is much larger than that of an equivalent swept laser source, in order to ensure low coherence gating.
The AOFSs devices are placed entirely within the secondary loops for both reference and object arms, and are not shared with the main loop of the interferometer. This is one of several differences to the configuration previously reported , and allows for the use of a single coupler for each loop, which diminishes losses. Isolators are not used in the loops of the system, again in order to avoid losses. Another difference is that a circulator is used in each reference and object paths, in order to achieve symmetry in optical design, with a further change being that a SLD light source with an Ytterbium amplifier is employed as a source to provide controllable power to the interferometer. In this respect the Ytterbium amplifier is used here solely as a power booster.
A main optical path difference (OPD)main can be defined, as the difference of optical paths measured along the main paths of the interferometer, from C1 to C4 and incorporates the path up to different depths in the object. This can be adjusted by moving the reference mirror. A differential optical path difference (OPD diff) can also be defined as the difference between the lengths along the two secondary loops. This can be adjusted by modifying the optical path lengths of AP1 and AP2. A different depth in the object is selected by:
By eliminating several sources of losses and using PM fiber, the new design reduces the optical gain required at the optical amplifiers. This is useful because of the positive effect of lower gain on noise properties when using SOAs .
Further, to address the signal decay from one channel to the next of the previously communicated configuration , here we propose a progressive periodic modulation of the SOAs currents to compensate for such effects. The principle of modulation requires that we limit the number of re-circulations for the time interval required to periodically modulate the gain. We illustrate this principle by altering the gain over 10 channels. This is achieved by switching the frequency shifters simultaneously using a 90% duty cycle at 1 MHz frequency. This determines an active time of 900 ns, which for a loop roundtrip time of 40 ns, limits the number of roundtrips to 22 circulations. To have at least two cycles of the carrier frequency during each active time interval of 900 ns, the minimum carrier frequency is 2 MHz. A beat frequency Δf = 5 MHz was chosen, by driving the frequency shifters at 150 MHz and 155 MHz. Signals from different depth positions from inside the object are encoded on multiples of the beat frequency so that if the difference of frequencies is 5 MHz, then this confers 1 loop, 10 MHz two loops and so on.
The multiple paths system was tested in three regimes: (i) AOFSs devices open all the time and SOAs operated with DC only drive current (no modulation), (ii) with digital modulation of the AOFSs (on/off modulation) and SOAs operated with DC only drive currents (square wave modulation), and finally (iii) with digital modulation of the AOFSs and ramp modulation of the SOAs drive currents.
Idealized opening and closing of loops
If we wish to work under CW excitation of the loops, and intend to make use of the higher order circulations (not attenuated significantly), the higher order circulations would need to make up a useful proportion of the light that is directed into the main loop. To illustrate how modulation of the loop may help to evenly distribute the powers in this mode of operation it is useful to consider the idealized loop shown in Fig. 2 .
In a time domain OCT system, the signal to noise ratio is theoretically limited by the shot noise and the optical power directed at the sample . In a conventional design, the power that would contribute towards the interference signal would be the total power incident on the sample, but in our design it is the fraction of power that has circulated n times around the loop that contributes to channel ‘n’ signal. Additionally it is the total power in all orders that contributes to the shot noise. We define PT as the total optical power exiting the recirculating loop and Pn as the optical power exiting the loops that has circulated n times, such that:
In our idealized shot noise limited time domain OCT system, the signal to noise for each channel can be defined as:
If the gain were independent of input power and was kept at unity (the loop gain is one), the output will rise on each roundtrip by the equivalent of the input power level to the coupler. An ideal steady increase in the output power is shown in Fig. 3 . For simplicity, let us ignore the circulating amplified spontaneous emission (ASE) and consider the loop gain always equal to 1. Let us define PT in this case as the total power in a complete cycle, where the duration of a complete cycle is given by the roundtrip time tr, multiplied by the number of loops, m allowed, i.e. the number of loop transits that can take place while the AOFS is open. For instance, for m = 22 roundtrips with tr around 40 ns, the loop is open for a complete cycle of T = m tr = 880 ns. In this cycle, for an input power Pin, the total power is:
The total power, Pn, that has travelled round n times during the cycle T is:
Consequently, the average fraction of power, Pn, out of the total power transferred to the output, PT is:
Using m = 22, the fraction of power contained that has travelled once (n = 1) around the loop, would be: and the fraction of power that has travelled for n = 10 times is:
As the SOA gain is not independent of the input power, to maintain a unity gain, one would need to adjust drive current to the SOA as the input power changes. We investigate the operation of the system in three different configurations as described above.
Wave modulation is used with two objectives. A first objective is to restrict the number of roundtrips to a value m = 10 to stop accumulation of ASE. The number of m roundtrips is controlled by the period of the modulation, T = mτr, where τr is the round trip time. A second objective is to modulate the gain within the time interval T to enhance the higher order of roundtrips. For the first goal, a rectangular signal is suited while for the second goal, a ramp signal, both of T duration.
In the diagram in Fig. 3, when no power is applied, the first step is made of ASE only. Starting from an initial value of the ASE, this increases with each trip through the loop. Not only that this creates noise, but eats up from the population level. In this way, even without any input to the recirculating loop, the loop will saturate from its initial input of ASE, therefore, there would still be a need to stop the loop and start again.
Several measurements were performed to quantify the behavior of the secondary loops alone and contributing to interference, before applying the configuration in imaging.
Characterization of the individual loops
When the optical amplifier is driven in the steady state (DC drive current as configuration in the regime (i) described above, circulating light from each loop will settle to a fixed output power. This output will be dependent not only on the SOA gain controlled via its drive current, but also on the level of input optical power. As the gain of the SOA device depends on the optical input power, increasing circulating power acts as a negative feedback on the overall loop gain, i.e. increasing the input power reduces gain. This ultimately limits the fraction of overall power exiting the loop that has transited many times. This phenomenon is illustrated in Fig. 4 , which displays the photo-detected signal at the output of C4 when SOA2 is not driven and the path up to the reference mirror is blocked. The behavior of the SOA in SOA1 is tested under different excitation powers from the optical source (the SLD alone was used, no booster), for two different driving currents. The modulation depth of pulsations in the photo-detected signal is due in part to the proportion of light circulating in the loop as ASE light rather than entirely due to the SLD light. In Fig. 4 left, a clear periodic oscillation is seen. These are free relaxation oscillations in the closed loop and their period is determined by the transit time of the loop, tr = 40 ns. It should be noted that the loop does not contain isolators, which means that amplified ASE light can propagate in both directions.
By increasing the SLD power from 85 to 500 µW (Fig. 4 right), the relaxation oscillation is disturbed and loses the regular behavior. An increased optical input power to the loop has increased the level of circulating power in the loop, which in turn has reduced the effective roundtrip gain of the loop (via the ‘Saturation Power’ parameter of the optical amplifier). When the roundtrip gain is below 1 for any light in the amplifier, including circulating ASE light, no oscillation will be seen. When using the system without modulation of the amplifiers, the roundtrip gain must be kept below 1 or the light sent to the reference and object will contain a large component of amplified ASE light. The distribution of power to specific roundtrip numbers (Pn) relative to total power of light sent to the object (PT) is limited in part by how the round trip gain is set relative to 1.
The noise floor needs to be considered over the total time for n circulations. Though the level of noise could be said to be related to total optical power, it would not be seen to be an increase of noise for each extra channel as all channels are simultaneously analyzed
When modulating the SOAs with current as in the (ii) and (iii) regimes explained above, the AOFSs are opened and closed in synchronism. This was done to ensure the dumping of all circulating ASE light and of any other light between modulation cycles. Figure 5 shows the photo-detector response due to light from object recirculating loop only, without any external input light. Ramp modulation of the SOA2 was used for two different base line drive currents, below (left) and above (right) the threshold. The oscillation seen in Fig. 5 (right) is owing to the gain being sufficient across the time period for lasing to take place.
Figure 6 shows the photo-detected response when light from port 2 of Circulator2 (as shown in Fig. 2) is measured while the drive current of the SOA2 is modulated with a square wave. It can be seen that the power output quickly reaches a plateau after 200 to 400 nanoseconds, corresponding to the time taken for the wave to travel 4 to 8 loops, within the time interval the AOFS is opened (900 ns).
It can be noticed that the optical input power quenches the relaxation oscillation. The SOA2 was driven with a 68 mA drive current. A 1.5 V wave was applied via a bias T to the SOA2 (90% duty cycle) and 500 µW input power incident at coupler C1. In Fig. 6 left, the dotted line shows theoretical slope of power build up if the gain was equivalent to 1. This gradient was inferred from the power increase expected for a loop time period (by an amount equal to half of power incident from the optical source to fiber coupler C1), but its intercept with the horizontal axis is arbitrary and is shown solely for purpose of the comparison of slope of the power increase with that of theoretical unity gain. It can be noticed that initially the power builds up linearly (as can be seen from the similarity of its gradient to that of the theoretical unity gain shown by a dashed line), and then the build-up rate slows down. This slowing of power build up is owing to the loop gain dropping below 1.
To achieve linear output power build up, as that shown in the idealized case in Fig. 3, we need to ramp modulate the drive currents sent to the SOA devices. If a ramp is used, then the variation of the optical power output is that shown in Fig. 6 (right). Because the gain depends on the input light level, our aim with using the ramp modulation was to keep the gain as close as possible to 1 as the light level increases. After evaluating the modulation effects on individual loops, an optimum shape was inferred for the driving signal. Then, this was the chosen shape for the driving signals applied to the SOAs in the both loops.
Interference measurements (both loops functional)
In all interference measurements, the target in the object path was a mirror (with a suitably adjusted neutral density filter for attenuation). Figure 7 shows a comparison of the photo detected interference signal using a mirror as object for the system in Fig. 1. The ramp modulation case was compared to that of no modulation. In both cases, the average power returned to the photo-detector, from the sample and from the reference arm, was kept approximately similar using slight adjustment of a variable attenuator placed in the reference arm. In Fig. 7, comparison between no modulation and ramp modulation cases only is shown. Square modulation (as shown for the power output of a single loop in Fig. 6) is not presented as it offers no benefit from redistribution of power. In fact, when using square modulation, the broadening of the signal spectrum was so large, as no benefit could be noticed. In this comparison, the OPDdiff was set to zero. By setting OPDdiff to zero, carriers from all loops are manifest all at the same time when the OPDmain is brought to zero.
Decay in the amplitude of a carrier from one loop to the next, (as seen in the frequency domain representation in the lower row in Fig. 7), is significantly reduced with respect to previous implementations. When using ramp modulation, a 10 time attenuation (20 dB) is achieved somewhere between the 13th and the 14th carrier (circulation number). This is marginally better than that achievable for no modulation, but the modulation techniques, in addition, do limit the number of circulations, here chosen to be about 10.
Figure 7 left shows attenuation from the 1st to the 11th order of 18 dB while in Fig. 7 right, less than 15 dB when applying ramp modulation. The measurements show that it is possible, by using polarization maintaining fiber and modulation, to reduce the signal attenuation from one loop to the next to less than 2 dB for at least up to ten circulations. Modulation however is not without its drawbacks. In the frequency domain, the modulation effectively blurs the signal spectral power density. This can be seen on the lower part of Fig. 7 (right) where the signal peaks are effectively broadened.
The fact that the first peak correlation amplitude shown in Fig. 7 is not of the largest magnitude in all methods is thought to be primarily the result of three factors. Firstly, the input to the loop is not thought to be a perfect linearly polarized state and therefore the first few loops will be reduced in intensity while polarization states that do not coincide with the favored polarization direction in the system are attenuated more on each roundtrip than higher order number roundtrips, that have already been filtered into more of a ‘pure’ linear state. i.e. the first circulation acts as a polarizing filter. Secondly, during the time that the loop is open, the effects of changing light level and the input of linear ramp modulation to the SOAs, means that the loop gain is never 1 from the loop being open to being closed. It is possible that the use of an arbitrary modulation waveform could correct for this. Thirdly, variation in polarization from one roundtrip to the next, beat noise between the optical source and the amplifier in the reference loop and dispersion left uncompensated contribute to further decay from one round trip to the next, larger than anticipated.
Assessment of axial resolution
Figure 8 shows the autocorrelation profiles for the first 10 channels, obtained by measuring the amplitude of the first 10 carriers versus OPDmain. The envelope of the interference signal was measured against movement of the reference mirror through coherence.
SOAs have a limited gain bandwidth and therefore broadening of the coherence profile from multiple circulations of signal light in the loop is expected. The combination of broadband input source with limited gain medium bandwidth effectively produces a narrower linewidth output on each circulation. That is, for each circulation the spectrum of the SOA signal at the loop output becomes narrower with each next roundtrip in comparison to the spectrum from the external optical source.
The AOFS devices were used to open and close the loops for this measurement in synchronism with the ramp modulation of the SOA devices (1MHz modulation at 90% duty cycle).
To evaluate the performance of the system, a simulation was conducted using parameters as specified for our components. We used the following parameters for the optical source and amplifiers; center wavelength of SOA1 1045 nm, center wavelength of SOA2 1049 nm, gain bandwidth of first amplifier 52 nm, gain bandwidth of second amplifier 42 nm and a source with center wavelength 1055 nm and line-width 30 nm (all line-widths quoted at 3 dB approximated to Gaussian shapes). Our simulations predicted that the autocorrelation function FWHM width would increase by a factor of 50% over 10 circulations. In practice, it was noticed that the absolute values of the correlation FWHM’s were larger than the theoretical width, as the FWHM values measured over the 10 channels varied in Fig. 8 between 25 to 70 μm. In addition, our modeling did not consider the polarization effects and some residual dispersion left uncompensated, which may explain the better axial resolution for channels 2 and 3 than channel 1. This behavior would suggest that it would be desirable to use much larger bandwidth SOAs than the optical source to ensure constant axial resolution over the multiple channels.
The adjustment of OPDdiff is independent on the axial resolution, determined by the bandwidth of the optical signals interfering after a given number of recirculations, as quantified by Fig. 8. The axial separation can be adjusted smaller than the depth resolution, for oversampling or larger than the depth resolution, for under sampling. The OPDdiff was set at 75 μm, larger than the largest coherence length across channels, as measured above.
Encoding the depth on frequency modulation
Figure 9 illustrates the experimental photo-detected signal in the time domain, where the OPD in the main loop is adjusted at multiple steps of OPDdiff. The horizontal scale shows time, with the AOFSs being opened just after the time moment of zero seconds. The interference signals for increasing numbers of circulations correspond to multiples of the base beat frequency (Δf = 5MHz), i.e. each depth nOPDdiff is encoded on a frequency nΔf. It can be noticed that in channels 6, 7 and 8 (corresponding to the same number of roundtrips in the secondary loops), no interference occurs in the early part of the graph as there has not been time for the light to circulate. As an example, the interference signal resulting from signals suffering eight circulations, which pulsates at 40 MHz, does not start until around 380 nanoseconds. This corresponds to just over the time taken for eight roundtrips in the secondary loops.
A pair of 6215 Cambridge Technology galvo-scanners and a telescope with two lenses of 7.5 cm in a 4f configuration were inserted into the path from output 2 of Circulator 2. A galvo-scanner (line) was driven with triangular ramps of 500 Hz and the other (frame) with a saw-tooth signal at 2 Hz and in synchronism, trigger signals were sent to the PCI 5124 card to build en-face OCT images according to procedures described in earlier reports . A simple scattering object was chosen, in the form of a tilted piece of white paper, to illustrate the system’s ability to acquire en-face OCT images from multiple depths simultaneously. The system was adjusted as in Fig. 9 (OPDdiff set to 75 μm). The AOFSs were used to open and close the loops, while the SOA drive currents were synchronously ramped. An en-face image of 500 lines of 500 pixels each is acquired in 0.5 s, which means that each transversal pixel is captured in 2 µs. Using a digitizer operating at 200 MSamples/sec, 400 samples are used for each transversal pixel. A program was written to perform FFT and select post-processing a band of 250 kHz around each carrier, followed by a rectifier. The images in Fig. 10 prove that the system allows differentiation of at least eight planes that were previously measured to be separated by 75 μm. Some blurring of the signal is seen, due to effective crosstalk between carriers which is most evident in panes one to four. We found that this could be mitigated by not using the AOFSs to block the loops, which means that the cross talk is chiefly due to the modulation. This may therefore mean that an optimum trade-off must be sought between the on–off switching period of AOFSs and the detrimental final outcome of crosstalk. The shorter the duration the AOFSs are on, lower ASE is manifested with less decay from one channel to the next. However, the shorter the time, the larger the bandwidth of the disturbing electrical signal with impact on the noise on each channel.
Previous set-ups  using the principle of secondary loops with amplification were not able to provide sufficient sensitivity for multiple channels from the retina in-vivo. In Fig. 11 we demonstrate first en-face images collected using such a configuration, from the optic nerve of a volunteer (AP). The 10 en-face images have been acquired in the time required to generate a single en-face frame, ie 0.5 s. This is also the first time that such images are truly acquired and generated in the same time. In , although the signals from multiple depths were present simultaneously in the photo-detected signal, an RF tunable filter was used to provide the images sequentially, by tuning the RF filter on each channel carrier. Here, the NI 5124 card collects whole the spectrum at once and this is then split in channels for each carrier and in this way, all 10 channels are acquired in the same time. A 2 s delay was required for the Labview to process the photo-detected signal and display the images. The last image in Fig. 11 represents the compound image obtaining by summing all the individual channel images. The AOTFs were opened for 2 μs which determines an overall bandwidth of 500 kHz on each channel. In these circumstances, a sensitivity of 88 dB was measured in the first channel for 2.2 mW sent to the eye.
A polarization maintaining interferometer configuration using active re-circulating delay lines based on multiple round trips in each arm is presented, together with a method of synchronous control of power distribution in the two arms. Using such a configuration, interrogation of multiple paths simultaneously to serve the purpose of generating several en-face OCT images from different depths is demonstrated, with less attenuation from one depth to the next than previously reported. The control of power is achieved by modulating the drive currents of the SOAs to compensate for losses and in this way, gain approaching unity along the recirculation loops for more than 10 roundtrips is made possible. More complex modulation signals applied to the SOAs could theoretically be used to create a more favorable power distribution across circulation numbers.
The previously reported multiple path configurations were constructed with single mode fiber, the amplifiers were placed in between isolators, and the loops operated in stationary state. Such configurations exhibited more than 4 dB attenuation from one recirculation to the next. Using modulation in the configuration reported here, this attenuation was significantly reduced to less than 2 dB (as measured and shown in Fig. 7, lower left).
While ‘perfect’ distribution of power between channels (each channel having equivalent optical power) would seem a target of the modulation technique, even distribution does not improve the signal to noise per channel overall. With even distribution, the optical power per channel, Pn would be inversely proportional to the number of channels. As signal to noise is proportional to Pn (times its ratio to the total power Pn/PT) the modulation technique may be better suited to dynamically adjusting the optical power per channel to that most appropriate for the object under test. i.e. concentrating optical power sent to the object that consists of a lower number of channels (higher signal to noise per channel) or a more even distribution when more channels or sampling positions are needed.
Use of this method for multiplexing multiple en-face OCT channels may be of interest for niche applications, but extension of this method to the Fourier domain would be of greater interest to the wider OCT field. Therefore we intend to develop this work for multiplexing multiple Fourier domain channels. As demonstrated in , a multiple path configuration like the one presented here can be driven by a swept source. In this case, the differential distance OPDdiff is adjusted larger than the maximum axial range determined by the line-width of the swept source. There are many OCT applications where an axial range only twice more and at the most, three times longer axial range than that given by the swept source line-width would be acceptable. Not much more, due to the operation under fixed focus of any spectral OCT methods. In other words, two or three re-circulations would suffice. Such application imposes fewer demands on the configuration than the application of the time domain OCT configuration presented here which would require many more channels, for the configuration to be deemed useful.
Although the value of the reference path difference, OPDdiff, employed in this paper was adjusted to be small, this could have been much larger, allowing sampling points arranged over a larger depth range. If the fiber length in the two secondary loops is perfectly matched and a slab of the same material to create OPDdiff is employed, then no dispersion is produced even at larger OPDs. In this way, the multiple path configuration with secondary loops applies a dynamic dispersion compensation method, where progressing in depth, dispersions is also compensated for. Such a behavior cannot be implemented in spectral OCT configurations being developed for long axial range interrogation.
However, such a configuration presents the same disadvantage as the spectral OCT method in terms of inability to implement dynamic focus, as all depths are collected at the same time.
The method described my find application in other areas, such as sensing, by providing results of several measurements of parameters from sensors at different distances.
The authors acknowledge the support of the Engineering Physical Sciences Research Council of the UK, grant EP/H004963/1 and of the European Research Council grant 249889. A. Podoleanu is also supported by the NIHR Biomedical Research Centre at Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology.
References and links
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, and J. G. Fujimoto, “Optical coherence tomography,” Science 254(5035), 1178–1181 (1991). [CrossRef] [PubMed]
2. A. F. Fercher, R. Leitgeb, C. K. Hitzenberger, H. Sattmann, and M. Wojtkowski, “Complex spectral interferometry OCT,” Proc. SPIE 3564, 173–178 (1998). [CrossRef]
3. K. Wiesauer, M. Pircher, E. Götzinger, S. Bauer, R. Engelke, G. Ahrens, G. Grützner, C. Hitzenberger, and D. Stifter, “En-face scanning optical coherence tomography with ultra-high resolution for material investigation,” Opt. Express 13(3), 1015–1024 (2005). [CrossRef] [PubMed]
5. L. Neagu, A. Bradu, L. Ma, J. W. Bloor, and A. G. Podoleanu, “Multiple-depth en face optical coherence tomography using active recirculation loops,” Opt. Lett. 35(13), 2296–2298 (2010). [CrossRef] [PubMed]
6. J. T. Kringlebotn and K. Bløtekjær, “Noise analysis of an amplified fiber-optic recirculating-ring delay line,” J. Lightwave Technol. 12(3), 573–582 (1994). [CrossRef]
9. S. Shimada and H. Ishio, Optical Amplifiers and their Applications (John Wiley & Sons Tokyo, 1992).
10. W. Drexler and J. G. Fujimoto, Optical Coherence Tomography: Technology and Applications (Springer, 2008).
11. R. B. Rosen, M. Hathaway, J. Rogers, J. Pedro, P. Garcia, P. Laissue, G. M. Dobre, and A. G. Podoleanu, “Multidimensional en-face OCT imaging of the retina,” Opt. Express 17(5), 4112–4133 (2009). [CrossRef] [PubMed]