1 Introduction

Optical coherence tomography (OCT) generates cross sectional imaging of tissues in situ with micron scale resolution by measuring the echo time delay of backscattered or backreflected light.[1] OCT imaging using frequency domain reflectometry with frequency swept laser sources was demonstrated several years ago.[2, 3] Recent studies have shown that Fourier domain detection using frequency swept lasers or spectrometers can dramatically improve imaging speed or detection sensitivity compared to standard time domain detection techniques.[4–6] Frequency swept light sources are especially important for imaging in the 1300 nm wavelength range, where low cost CCDs are not available. OCT with a frequency swept light source also has the advantage of enabling dual balanced detection and avoiding the need for high performance spectrometers and CCDs which are used in Fourier/spectral domain detection. For this reason there is considerable interest in the development of frequency swept light sources for OCT.

Early studies demonstrated the utility of frequency domain reflectometry for precise measurement of distances and reflections in profilometry and fiber sensing. Different types of frequency swept lasers have been demonstrated for these applications.[7–12] However, most of these early lasers were optimized for extremely narrow instantaneous linewidth to achieve a long spatial detection range, rather than for the rate and spectral width of the frequency sweep. Sweep rates of only several Hz achieved with single mode, frequency swept lasers were too slow for OCT applications where sweep rates of several 100 Hz to several 10 kHz are desired. For this reason scaling of the sweep rates to higher speeds while maintaining a large spectral sweep range is of fundamental interest for OCT imaging.

Different laser designs for swept source OCT have been demonstrated. Early studies were performed with bulk optic designs using a Cr:Forsterite laser operating at ~1250 nm and an external cavity diode laser operating at ~800 nm at 10 Hz and 2 kHz sweep rate respectively.[2, 3] An all fiber ring laser employing a fiber Fabry-Perot tunable filter was demonstrated at sweep rates of 200 Hz.[5] Hybrid bulk optic and fiber cavity designs using a fiber coupled semiconductor amplifier as a gain medium and a diffraction grating for wavelength selection have been demonstrated. Frequency sweep rates of 500 Hz (1000 Hz effective sweep rate) and 15.7 kHz have been achieved using a galvanometer scanning mirror and a polygon rotating mirror, respectively.[13,14]. Techniques for calibrating the instantaneous wavelength of the swept source have been demonstrated.[15–19]

There are two main approaches used to build frequency swept, narrow linewidth light sources: The approach of “post-filtering” uses a broadband light source such as a superluminescent diode or a short pulse laser to generate a broad spectrum, then uses a narrowband tunable band-pass filter with a transmission window narrow enough for the desired instantaneous coherence length. The approach of post filtering has the advantage that the tuning speed is limited only by the maximum tuning speed of the filter. However, the power of the light source is usually low, due to the high loss of the filtering process. A complementary approach is “cavity-tuning”, where the spectral filtering element is used inside a laser cavity. Lasing can only occur at wavelengths within the transmission window of the filter. The advantage of this approach is that it usually generates much higher energy, since the energy loss from filtering is compensated by the repetitive laser gain. Furthermore, the mode competition in the laser can lead to a narrowing of the effective linewidth for slow frequency sweep speeds. The limitation of laser cavity tuning is that the maximum frequency tuning speed is not only limited by the maximum tuning rate of the filter, but also by the time constant for building up laser operation. This manuscript will present an analysis of different operating regimes for high speed frequency tuning and suggest design approaches for optimizing performance.

In this paper, we demonstrate a high power, high speed, frequency swept laser light source at 1300 nm. The laser uses a robust and compact fiber ring design with an additional booster amplifier. The laser achieves high speed frequency sweeping up to 30 kHz repetition rates. At an operation point of 20 kHz sweep rate a peak output power of ~45 mW is achieved. We demonstrate the application of this laser for OCT imaging. We describe the laser design and the scaling of power and coherence length with frequency sweep speed. We also investigate the effect of amplification on the output spectrum, power performance and noise characteristics. We present design criteria which relate the maximum frequency sweep speeds to the cavity length and properties of the gain medium. Finally, we also present a fast recalibration method for real time resampling interferometric signals to even spacing in frequency in order to provide real time OCT image preview.

2 Experimental setup

Figure 1 (top) shows a schematic of the frequency swept laser. The laser uses a ring cavity geometry and consists of a fiber coupled semiconductor amplifier (SOA from InPhenix, Inc.), two isolators with an insertion loss of <0.5 dB each, a piezoelectric actuated fiber Fabry-Perot tunable filter (FFP-TF from Micron Optics, Inc.) and a fiber output coupler. The waveform driver consists of a digital function generator and an amplifier for driving the low impedance capacitive load of the PZT of ~2.2 µF. The physical length of the cavity is 2.4 m, resulting in an optical path length of 3.5 m. The FFP-TF acts as a narrow-band transmission filter for active wavelength selection. The isolators eliminate extraneous intra-cavity reflections and ensure unidirectional lasing of the ring cavity. The FFP-TF has a free spectral range (FSR) of 270 nm (48 THz), a bandwidth of ~0.135 nm (~24 GHz), and an insertion loss of ~2 dB. The high finesse of the FFP-TF provides narrow band spectral filtering suitable for achieving a sufficient dynamic coherence length of the source. A fiber splitter acts as 40 % output coupler. After isolation, the output is amplified with a second fiber coupled semiconductor amplifier (SOA from InPhenix, Inc.) which functions as a booster amplifier. At a sweep rate of 20 kHz the peak power from the laser is amplified from 2.3 mW to about 46 mW, providing a theoretical increase in sensitivity of ~12 dB. A final isolator avoids optical feedback from the OCT system or other apparatus.

The OCT system shown in Fig. 1 (bottom) uses a dual balanced detection scheme for recording the fringe signal of the Michelson interferometer on one channel of a two channel, high speed analog to digital converter (ADC) operating up to 100 Msamples/s with 14 bit resolution (National Instruments, model NI5122). The photodiodes (New Focus, model 1817) have a bandwidth of 80 MHz, a transimpedance gain of 50,000 V/A and an estimated responsivity of 0.8 A/W at 1300 nm. For real time recalibration of the instantaneous wavelength to frequency, a fraction of the laser output power is split off into a second, fixed fiber Fabry-Perot filter (FFPI) with a FSR of 50 GHz (0.28 nm). A 125 MHz bandwidth photodetector (New Focus, model 1811) records the comb like signal as the laser source sweeps in wavelength. Each peak in the recalibration signal corresponds to a frequency shift of 50 GHz with respect to the previous peak correlating to the transmission maxima of the FFPI. After electronic filtering, the recalibration signal is recorded on the second channel of the ADC synchronously with the sample signal from the OCT interferometer, allowing real time recalibration of the interference signal for equal spacing in frequency prior to the FFT algorithm. OCT imaging is performed by using a beam scanning system probe consisting of a 20 mm focal length fiber collimating lens and a 40 mm focal length objective lens using a 6 mm aperture XY galvanometer scanner which scans the beam post objective.

 

Fig. 1. Top: Schematic of the amplified frequency swept laser source. The laser uses a fiber coupled SOA and a FFP-TF. A SOA boosts the laser output. Bottom: Schematic of the OCT system using Fourier domain/swept source detection.

Download Full Size | PDF

 

Fig. 2. Left: Signal generation for recalibration. Right: Principle of the “nearest neighbor check” algorithm. The schematic shows the two synchronously acquired signal traces, the calibration trace (right top) and the sample trace (actual OCT signal) from the dual balanced detector (right bottom).

Download Full Size | PDF

3 Fast recalibration for real time display

When operating at fast frequency sweep speeds, the laser frequency varies sinusoidally rather than linearly with time. Therefore, an accurate and reliable recalibration of the interference output to equidistant spacing in frequency is necessary. AD converter cards which operate at lower frequencies often have a clock input that can be used to trigger AD conversion using a recalibration signal. However, most high speed AD cards do not have this feature or timing jitter of the generated clocking signal can be a problem. The AD converter card in this system was operated in a mode with parallel and synchronous acquisition of two channels, one channel recording the interference signal from the dual balanced detector, the second channel recording the electronically bandpass filtered comb-like recalibration signal as shown in Fig. 2 (left). An accurate recalibration can be performed in post processing mode, to recalibrate the OCT sample signal to an even spacing in frequency. However, real time preview for aiming and adjusting requires a faster recalibration algorithm. The concept of this algorithm is depicted in Fig. 2. First, the recalibration signal, the output signal of the photodiode (FFP calibration trace, Fig. 2 (top, left) is bandpass filtered with a passband matching the expected frequency range of the fringe signal in order to remove electronic noise and provide a clean sinusoidal signal (bandpass filtered signal, Fig. 2 (middle, left)). The sampling rate of the AD converter is set slightly above two times the maximum fringe frequency (Nyquist sampling, Fig. 2 (bottom, left)). A high speed, “nearest neighbor check algorithm” is applied and performs a stream processing of the recalibration signal such that at every point, the value of the sampling point before and after, are checked. If both values are higher or lower, the corresponding data point from the sample trace is added to the recalibrated signal array (Fig. 2 (right)). Because this algorithm contains only one loop with two comparisons, the speed is adequate for real time streaming for online preview. The recalibration processing time is negligible compared to the time required for Fourier transformation. In spite of the slight inaccuracy in the determination of the actual peaks this algorithm proves to provide a robust and fast method for real time recalibration in OCT imaging.

4 Limitations to frequency tuning speed

In the following discussion, we will use the term “repetition rate” or “sweep rate” to characterize the periodicity of the waveform which drives the frequency filter. In the case of bidirectional waveforms, such as sinusoidally driven filters, we will apply the term “effective sweep rate” for the number of frequency sweeps per second, accounting for both forward and backward frequency scan. Depending upon the operating conditions of the laser source, both forward (scanning from shorter to longer wavelengths) and backward scan (scanning from longer to shorter wavelength) can be used for OCT imaging, so the effective sweep rate can be two times the repetition rate or sweep rate. The term “line rate” or “axial scan rate” is used for the number of axial scans per second that can be actually acquired by using the bidirectional frequency sweep.

 

Fig. 3. Concept of cavity tuning: build up of laser activity from ASE background.

Download Full Size | PDF

In cavity-tuning the maximum tuning speed is usually limited by the time-constant of the laser to build up lasing from the amplified stimulated emission (ASE) background. This is dependent on the filter function, the ASE intensity, the saturation power, the laser gain and the cavity roundtrip time. In the following analysis we consider two distinct limits which characterize the frequency tuning behavior of the frequency swept laser source: (i) the saturation limit and (ii) the single roundtrip limit.

4.1 Saturation limit

The first limit or characteristic frequency at which a change in the laser dynamics can be expected is the “saturation limit”. This limit represents the maximum frequency tuning speed which still allows for full build-up of lasing activity from the ASE background. Figure 3 shows the concept for estimating this characteristic frequency. We assume a background intensity of ASE from the laser medium. The broad green curve shows the ASE spectrum of the laser gain medium with a characteristic mode structure inside a cavity (narrow green curves). When the filter window is tuned in wavelength, the ASE background is amplified up to the saturation power limit of the gain medium, provided that the time the filter transmits the given wavelength is enough for this build up. The maximum frequency tuning speed which allows the build up of saturated lasing from ASE can be estimated. Assuming a filter width of Δλ, the number n of roundtrips required to reach saturation from ASE is

where P_ASE is the power of the ASE within the spectral window of the filter, β is the roundtrip net gain for small signals and P_sat is the saturation power. The net gain per roundtrip β can be written as

where G is the small signal gain of the laser medium and ρ is the fraction of energy fed back after one roundtrip. The value ρ is determined by the round-trip losses and the output coupling. The power P_ASE can be estimated

where the total tuning bandwidth is Δλ_tuningrange, which is approximately the total bandwidth of the ASE. P_ASEtotal is the total spectrally integrated ASE power. The prerequisite for saturated operation is that the filter should not be tuned further than Δλ within a time window τ_buildup of n cavity roundtrips. The cavity roundtrip time is

where L is the physical cavity length of the ring cavity (2L for a linear cavity), n_ref is the refractive index of the cavity and c is the speed of light. The tuning speed in nanometers per second is determined by

A factor η=1/π accounts for the higher frequency sweep speed of a sinusoidal sweep compared to a linear, unidirectional sweep assuming the filter sweep occurs over the same spectral range. The maximum sweep frequency f_sweep for lasing build up to saturation can be estimated as

This “saturation build up limit” represents the maximum sweep frequency for which the filter is tuned over its full width half maximum (FWHM) in a time that is sufficient to build up saturated lasing from ASE. Although this is only a rough estimate, it can give design rules for high speed frequency swept laser sources. It is expected that this value gives the order of magnitude of the frequency sweep above which a decrease in output power will occur.

This estimation does not account for the suppression of the ASE by the lasing activity and the fact that the effective gain of the SOA is smaller than the small signal gain. Both effects will lead to a drop in power at lower sweep speeds than the estimated saturation limit. However, both effects will become smaller for higher sweep speeds, as a drop in output power will increase the ASE level as well as increase the effective gain, thereby improving the output power. These effects cause the power to begin to decrease at lower frequencies than predicted by the saturation limit, with the power decrease occurring over a broad frequency range.

It should be noted that the discussion presented here describes a quasi-stationary model for the lasing in a saturated regime, neglecting hysteresis effects caused by the carrier dynamics in the gain medium. This is valid for the case of a semiconductor gain medium since typical carrier relaxation times are well below a nanosecond, much faster than the tuning process and the build up time for the lasing. However for gain media with slow dynamics, such as solid state media including Er-doped glass, Nd-doped glass, or Ti:sapphire with excited state lifetimes on the order of microseconds to milliseconds, rapid tuning can cause the laser to generate pulses in a Q-switched operation mode. Q-switching does not occur in the studies presented here.

Using the experimental parameters of Δλ=0.135 nm, Δλ_tuningrange=120 nm, P_ASEtotal=1 mW, P_sat=10 mW, η=1/π (accounting for the higher sweep speed in a nonlinear sinusoidal sweep), G=158 (22 dB), ρ=0.2, L=2.4 m, and n_ref=1.46, the estimated sweep speed is ~11,600 Hz. This value represents a rough estimate of how fast the laser can be tuned while preserving the maximum power output. Shorter cavities, long duty cycles and linear scans will enable higher speeds. Experimentally it was observed that tuning from shorter to longer wavelengths (i.e. to lower energy or frequency) is favored compared to tuning in the opposite direction. This effect can be related to four-wave mixing effects in the semiconductor gain medium which tend to produce frequency or energy downshifting. This assists the tuning process because the lasing does not have to build up solely from ASE. This nonlinear process is not accounted for in the analysis presented here.

Another important parameter in addition to the output power is the instantaneous linewidth, which is inversely proportional the instantaneous coherence length. Operating the laser in the saturation regime will lead to an instantaneous linewidth which is significantly narrower than the filter bandwidth. The multiple roundtrips and multiple passes through the filter will lead to mode competition and narrow the emission linewidth.

4.2 Single roundtrip – post filtering limit

For frequency sweep speeds higher than the “saturation limit” the output power will decrease until it reaches the “single roundtrip limit”, the frequency tuning speed above which the light on the average makes only one pass from the gain medium to the filter and is coupled out. The filter has tuned so rapidly that during the next roundtrip it already blocks the wavelength. The output light shows the characteristics of ASE, which is spectrally filtered and amplified once (on the average). An expression for the sweep frequency for the single roundtrip limit is given by

Using the parameters of the experimental setup, the sweep frequency for the single round trip limit is f_single is ~30,700 Hz. This limit represents an estimate of the maximum frequency for which laser feedback can occur. When the laser is frequency swept above this frequency, it operates like a post-filtered broadband light source rather than a tuned cavity. In this post filtering regime the instantaneous linewidth and instantaneous coherence length will be determined by the bandwidth of the optical filter in the cavity. In the following section the performance of the laser in this operating regime will be discussed.

5 Power performance with and without booster amplifier

Figure 4 (top, left) shows the maximum output power versus the sweep rate for the laser without the booster amplifier for the forward scan (black) as well as for the backward scan (red) normalized to the peak power at slow sweep rates. It can be seen that up to sweep frequencies of about 1,000 Hz, there is no significant decrease in output power for the forward scan and only a small decrease for the backward scan. This asymmetry is due to nonlinearities in the gain medium which tend to produce a downshift in energy. Therefore, forward sweeps (shorter to longer wavelength) have higher energy than backward sweeps (longer to shorter wavelength). At a sweep frequency of 10 kHz (the estimated saturation limit) the output power decreases to about 50 % and 5 % of its low frequency value for the forward and backward sweep, respectively. At higher sweep frequencies than the saturation limit the power starts to decrease rapidly because there is insufficient time for build up to saturation. At the single roundtrip limit of 30 kHz sweep frequency, the power in both the forward scan and the backward scan has decreased to 0.4 % of its low frequency value.

Because of this significant decrease in output power at high sweep frequencies, the operating point of the laser should be chosen at a frequency somewhere between the saturation limit and the single roundtrip limit. The dramatic decrease in power above the single roundtrip limit suggests that an operation far above will not be feasible without post amplification. Figure 4 (top, right) shows the influence of the external booster amplifier stage on the power output versus sweep frequency. The normalized power is shown for forward scan direction for the laser without (black) and with (blue) the booster amplifier. The booster stage is driven in saturation. At a sweep rate of 20 kHz, the peak power directly from the laser was 2.3 mW and the maximum peak power after the booster stage was 46 mW. The effective gain is 12 dB and this value is ~10 dB smaller than the small signal gain of the amplifier of 22 dB, indicating that the SOA is well saturated. This saturation leads to a much flatter curve for the drop in output power versus sweep frequency. As a result, the booster stage helps to maintain a high output power even for frequencies approaching the single roundtrip limit. The saturation effect of the booster amplifier also helps to provide a much higher power in the backward frequency scan, shown in Fig. 4 (bottom, left). This means both frequency scanning directions can be used up to much higher frequency sweep rates than for the unamplified laser. Figure 4 (bottom, right) shows the comparison between the power of the forward and backward scans with the booster amplifier. It can be seen that the difference in power between the forward and backward scans is small even up to sweep frequencies of 6–8 kHz. These sweeps frequencies would result in “effective line rates” of 12 to 16 kHz for frequency domain reflectometry or OCT imaging. This demonstrates that the booster amplifier stage not only improves the output power, but also enables operation at higher frequencies in the bidirectional scan mode, thereby achieving higher effective line rates. The change in the temporal intensity profiles for different sweep speeds is depicted in Fig. 5. Figure 5 shows the temporal intensity profiles of the laser for the three characteristic operation regimes, as discussed above.

 

Fig. 4. Maximum output power of the forward- (from shorter to longer wavelengths) and backward- (from longer to shorter wavelengths) frequency scan versus the sweep frequency normalized to the power at low frequencies. The boxes mark the regions between the “saturation limit” and the “single roundtrip limit.” Top left: Energy of forward (black) and backward (red) scan without an external booster. Top right: Comparison of the relative drop in the maximum power output of the laser without (black) and with (blue) booster for the forward scan. Bottom left: Comparison of the relative drop in the maximum power output of the laser without (red) and with (green) booster for the backward scan. Bottom right: Comparison of the relative drop in the maximum power output of the laser with booster for the forward (blue) and the backward scan (scan).

Download Full Size | PDF

 

Fig. 5. Temporal intensity profile for different frequency sweep rates. Top left: Saturation regime: for sweep speeds significantly slower than the saturation limit (2 kHz drive waveform). Top right: Multiple-roundtrip regime with frequency sweep speed between the saturation limit and single roundtrip limit (20 kHz drive waveform). Bottom left: Frequency sweep speed at single roundtrip limit with (left) and without (right) booster (30 kHz drive waveform).

Download Full Size | PDF

5.1 Saturation regime

Figure 5 (top, left) shows the temporal intensity profile of the laser with booster for a sinusuodal drive waveform of 2 kHz. As discussed above, this corresponds to operating regime significantly below the saturation limit. It can be seen that both forward (slightly higher peaks) and backward scan (slightly lower peaks) have almost the same power. For frequency domain reflectometry or OCT imaging this would result in an “effective sweep rate” of 4 kHz. The maximum instantaneous power of ~50 mW is determined by the saturation power of the booster stage.

5.2 Multiple-roundtrip regime

Figure 5 (top, right) shows the temporal intensity profile of the laser with the booster amplifier for a sinusoidal drive waveform of 20 kHz. This corresponds to operation between the saturation limit and the single roundtrip limit. The ASE makes multiple roundtrips and is amplified and filtered several times. However, there is insufficient time to build up to full saturation. Nonlinear processes in the laser gain medium leading to a downshift in energy promote frequency scanning in forward direction (higher peaks) and hinder scanning in backward direction (smaller peaks). For this reason, a significant asymmetry in the two frequency scan directions can be observed. For frequency domain reflectometry or OCT imaging the “effective sweep rate” would be 20 kHz, the same as the drive frequency. The maximum peak power of ~45 mW for the forward scan direction is close to the saturation limit of the booster amplifier. At this sweep frequency, there are only ~2 roundtrips and this suggests that lasing does not start from only the ASE level, but that both ASE and nonlinear frequency shifting contribute. The peak power is less than 5 mW in the backward scan direction and this emphasizes that, for this case, higher numbers of roundtrips would be required to build up lasing.

5.3 Single roundtrip – post filtering regime

Figure 5 (bottom, left) shows the temporal intensity profiles for the laser with the booster amplifier at a sinusoidal drive frequency of 30 kHz. This corresponds to operation in the single roundtrip regime. The ASE on the average makes approximately one roundtrip and is filtered once, but there is no subsequent feedback and amplification within the cavity. However, the filtered ASE is amplified by the booster amplifier. The laser does not reach saturation and the instantaneous power is ~1.5 mW. Although this power might be sufficient for frequency domain reflectometry or OCT imaging, there is a significant amount of power in the ASE background from the amplifier. This background, which is subtracted in Fig. 5 (bottom, left), has an average power of about 2 mW, while the average power of frequency swept light is less than 0.5 mW. The high ASE occurs because there is insufficient power from the laser to saturate the amplifier stage. The output of the laser without booster is shown in Fig. 5 (bottom, right) and is only several tens of microwatts. This power level is consistent with what is expected from post filtering the ASE of the gain medium in the laser cavity. At these high sweep frequencies, the system does not operate as a laser with repetitive feedback, but instead operates as a post filtered ASE source followed by amplification in the booster stage. It can be seen that the intensity of forward and backward scan are identical. Since there is no feedback, the nonlinear frequency shifting effects in the laser gain medium no longer influence the laser operation.

In spite of the disadvantages of increased ASE background, an amplified post filtered light source such as this would have the advantage that both scan directions can be used. Using a 30 kHz drive frequency, this would yield an effective sweep rate of 60 kHz. Furthermore, there would be no drop in output power for higher frequency sweep speeds and the maximum sweep rate would be limited only by the maximum filter sweep speed. It should be noted that the instantaneous spectral intensity of the ASE is much lower than the output of the amplified, post filtered configuration even though the average power is higher. Assuming a filter bandwidth Δλ=0.135 nm and an ASE bandwidth of Δλ_tuningrange=120 nm, the spectral intensity of ASE within the filter bandwidth is estimated to be ~2 µW, about 23 dB lower than the intensity of the amplified filtered ASE. This performance could be optimized using specially designed booster stages, cascaded filters or special ASE sources.

6 Spectral shaping by the booster amplifier

In addition to increasing the output power, the booster amplifier stage can also improve the spectral bandwidth and shape of the output spectrum. The spectral bandwidth and shape is important because it determines the axial resolution and pointspread function in frequency domain reflectometry and OCT imaging applications. Figure 6 (top) shows the output spectra of the laser operating at 15 kHz frequency sweep (red), the laser with an external booster amplifier identical to the laser gain device (green), and the laser with an external booster amplifier with a gain spectrum which is blue shifted compared to the laser gain device (black). The spectra were taken by calibrating the time-to-wavelength measurement of the laser output power measured by a photodiode. The asymmetric shape of the spectrum directly from the laser can be attributed to the nonlinear frequency downshifting mechanism in the laser gain medium, which tends to shift the spectrum to longer wavelengths during the forward frequency sweep (from shorter to longer wavelengths) as well as to the higher gain of the SOA at longer wavelengths due to the characteristic joint density of states and population distribution. The application of the booster amplifier partly compensates the asymmetry caused by nonlinear frequency downshifting and also increases the FWHM bandwidth from 59 nm to 78 nm because the spectrum tends to follow the gain spectrum of the booster. Finally, using a booster amplifier with a blue-shifted gain spectrum, results in an even more symmetrically shaped spectrum. The laser SOA and the booster SOA used in these studies have ASE maxima at 1300 nm and 1274 nm, respectively. An optimized choice of the booster gain spectrum should result in an even broader bandwidth and more symmetric spectrum.

 

Fig. 6. Top: Output spectra of the laser (red), the laser with an external booster amplifier identical to the laser gain (green), and the laser with an external booster amplifier with a gain spectrum, which is blue shifted compared to the laser gain (black). Bottom: Calculated pointspread functions for the different spectra on a linear scale (left) and a logarithmic scale (right).

Download Full Size | PDF

The asymmetric non-Gaussian spectrum directly from the laser results in significant sidelobes in the pointspread function (Fig. 6 (bottom, left)) which are visible even on a linear scale. On a logarithmic scale (Fig. 6 (bottom, right)) it can be seen that the sharp edges of the spectrum directly from the laser without the booster amplifier produce severe sidelobes in the pointspread function, which resembles the Fourier-transform of a rectangular spectrum. The spectral shaping provided by the booster amplifier improves the pointspread function significantly, thereby improving image quality for OCT applications. The application of a booster amplifier with a blue shifted gain spectrum can further improve the pointspread function (Fig. 6 (bottom, right)). Finally, it is important to note that these studies were performed by biasing the booster amplifier at constant current. It is also possible to dynamically modulate the booster amplifier (or the laser) as it is frequency swept to even more precisely tailor the shape of the spectrum and the pointspread function.

7 Performance of the frequency swept laser source for OCT

7.1 Sensitivity and resolution

As discussed previously, the use of a booster amplifier offers benefits in tuning speed, power and spectral shape of the laser output. In the following section we will discuss the influence of the amplification process on the instantaneous coherence length and the sensitivity, parameters which are important for frequency domain reflectometry and OCT applications.

For these measurements the laser was operated at 20 kHz frequency sweep rates. The peak output power of the unamplified and amplified laser was 2.3 mW and 46.5 mW, respectively. Significant output powers were obtained only during the forward scan (short to long wavelengths) and the duty cycle was ~40 %. The detection sensitivity was measured using a calibrated, attenuated isolated reflection from a mirror. The reference arm power was attenuated to several tens of microwatts. We measured a range of more than one order of magnitude of reference arm power over which this maximum sensitivity value could be achieved before heterodyne gain became too low or excess noise became too high. System losses of 12.8 dB arising from losses in the optics and beam scanning apparatus were subtracted from the measured sensitivity values. To measure the drop in sensitivity over depth the maximum amplitude of the interference fringe signal was measured with an oscilloscope characterizing the instantaneous linewidth and coherence length of the source. To include additional errors caused by the recalibration procedure and to determine the actual point spread function dependent on depth, the Fourier transform of the interference fringe signal after recalibration by the presented “nearest neighbour check” algorithm was analyzed.

Figure 7 (left) shows the sensitivity versus the delay/depth with and without the booster amplifier as well as the relative fringe contrast (diamonds vs. circles) relative to the measured sensitivity at a 500 µm delay. The sensitivity for the amplified system is shown by multiple pointspread functions (colored lines) where the values of the peaks represent the sensitivity at different delays. The sensitivity values where measured as 20x the logarithm of the ratio between the peak signal and the noise floor. The measurement was performed with 61 dB attenuation in the sample arm, including losses in the setup. The thick black line shows the sensitivity values for the unamplified system measured using the Fourier transform of the fringe signal and comparing the peaks to the standard deviation of the noise floor. The graphs displaying the fringe contrast (blue diamonds and red circles) show the relative height of the interference fringe signal measured with an oscilloscope before Fourier transforming. Therefore, the difference in the values of the measured height of the pointspread function and the fringe contrast are the result of phase jitter of the source and errors in the recalibration from time to frequency.

The measured maximum sensitivity value of 108 dB was obtained with 8.6 mW peak incident power or 1.7 mW average incident power (at a ~40 % duty cycle) on the sample. This agrees well with the calculated shot noise limit of about 110 dB for the given source parameters. Sensitivity values are adjusted by 12.8 dB to account for system losses such as mirror reflection and losses in the backcoupled light into the fiber interferometer. The 11.7 dB increase in sensitivity obtained by using the booster amplifier is in good agreement with the 13 dB increase in peak power, implying that little excess noise is added by the booster amplifier and the full benefit of the increase in peak power can be obtained. It should be pointed out that an amplification process always adds intensity and phase noise to the signal, but dual balanced detection can partially compensate this excess noise. These measurements show that the increased intensity noise does not significantly reduce sensitivity.

The fringe contrast drops by ~6 dB at a depth (interferometer displacement) of 3 mm. An instantaneous linewidth of 0.135 nm at 1300 nm center wavelength should correspond to a fringe amplitude envelope signal with a FWHM of 11 mm. This corresponds to a depth (interferometer displacement) of 5.5 mm. Accounting for the symmetry after Fourier transformation, the fringe amplitude should drop by 50 % at a depth of 2.7 mm, which corresponds to a 6 dB decrease in signal intensity and agrees well with the experiment.

 

Fig. 7. Left: (colored lines) Pointspread functions for different delays, measured with -61 dB sample arm attenuation including losses. The scale is calibrated such that the peak values of the point spread functions match the sensitivity for the different depths. Sensitivity versus delay for the amplified and unamplified laser measured at 20 kHz frequency scan rates calculated by the drop in fringe contrast and calibrated to the sensitivity at 500 µm delay (diamonds and circles, respectively). Sensitivity of unamplified laser measured using Fourier transform of fringe signal (black line). Right: Axial resolution plotted as FWHM of the amplitude signal of the Fourier transformed fringe signal in air versus delay.

Download Full Size | PDF

As discussed before when operating at high frequency sweep speeds close to, or faster than the single roundtrip limit, the instantaneous linewidth is determined by the filter bandwidth and additional linewidth narrowing is not expected. It can also be seen that the drop in the maximum fringe contrast is similar for the amplified and unamplified laser source, so the instantaneous coherence lengths are comparable and the booster amplifier does not change the instantaneous linewidth. Even when the booster amplifier is in the saturated regime, this can be expected if the gain medium exhibits homogeneous broadening over a spectral range larger than the instantaneous linewidth, as in the case of a semiconductor-based gain medium. All wavelengths within the instantaneous laser bandwidth experience almost the same gain. If saturation occurs, the gain is suppressed for all wavelengths simultaneously by the same factor. Therefore the amplification process does not change the spectral shape of the input light. However during the frequency sweep, if the carrier recovery time is significantly shorter than the sweep duration, the wavelengths with lower intensity near the edges of the sweep spectrum will experience higher amplification by the booster amplifier than in the center of the spectrum where the booster amplifier is saturated. In this case, saturation at one point in time (when the swept source is at one wavelength) does not affect the gain at another point in time (when the swept source is at another wavelength). When saturation occurs during the frequency sweep, wavelengths with lower intensity will experience higher gain than wavelengths with higher intensity. For this reason, the sweep spectrum is broadened but not the instantaneous linewidth.

Figure 7 (right) shows the axial resolution versus delay/depth for sweep frequencies of 15 and 20 kHz using the laser source with and without the booster amplifier. A small loss of resolution with increasing delay/depth can be observed which can be attributed to errors in the fast recalibration algorithm which become more pronounced at higher fringe frequencies corresponding to longer delays/depths. However, the width of the pointspread function is not increased by the amplification.

7.2 OCT imaging performance

The application of the frequency swept light source for high speed OCT imaging is demonstrated in Figure 8. The images shown are acquired with axial scan rates of 7, 20 and 27 kHz. The average power on the sample was 2 mW, 1.4 mW and 0.4 mW respectively. The sampling rates were 25 Msamples/s for the image at 7 kHz line rate and 100 Msamples/s for the images at 20 kHz and 27 kHz line rates. The laser configuration with the wavelength shifted booster amplifier was used. The axial image resolution is 12 to 14 µm in air, corresponding to 9 to 11 µm in tissue. The images consist of 3,500 axial scans (transverse pixels) with 400 points (axial pixels) per scan. The number of axial pixels is determined by the frequency spacing of the recalibration signal and the spectral bandwidth of the source. In this case, the interference signal is zero padded up to a power of 2 to increase the final pixel density to 512 points (1024 points before FFT). In the measurements shown, the DC level was suppressed sufficiently by dual balancing and no additional background subtraction was applied. The images are generated using a postprocessing algorithm which interpolates between the sampling points and finds the maximum more accurately than the real time nearest neighbor check algorithm. A slightly lower background noise was achieved using this algorithm than the real time, “nearest neighbor check”, algorithm. The transverse spot size was ~20 µm.

 

Fig. 8. OCT images taken with swept source at different sweep rates. Top left: Nailfold imaged at 20 kHz line rate. Top right: Skin in the area of the palm imaged at 20 kHz line rate. Middle left: Human skin in the area of the finger tip imaged at 20 kHz line rate. Middle right: Human skin in the area of the finger tip imaged at 27 kHz line rate. Bottom: Hamster cheek pouch ex vivo images at 7 kHz line rate.

Download Full Size | PDF

Figure 8 (top, left) shows an OCT image of human skin in vivo in the nailfold region, imaged at an axial scan rate of 20 kHz. Figure 8 (top, right) shows an OCT image of human skin on the palm imaged at 20 kHz. Figure 8 (middle) shows images of the finger at axial scan rates of 20 kHz (left) and 27 kHz (right). These OCT images are comparable to those obtained using time domain OCT systems and features such as the stratum corneum (SC), the dermal epidermal junction (DEJ), the epidermis (ED), sweat ducts (SD) and the nail (N) can clearly be identified. The images in Fig. 8 (middle) show a decrease in imaging quality between 20 kHz and 27 kHz, because of a decrease in laser power and increased ASE (as discussed in the previous section). This high scan rate corresponds to a frequency sweep which is slightly below the single roundtrip limit. Figure 8 (bottom) shows an image of a hamster cheek pouch in vitro at an axial scan rate of 7 kHz. The image shows high contrast and high penetration down to the connective tissue (CT) layer.

8 Conclusion

This study demonstrates a new, high speed, high power, frequency swept laser for frequency domain reflectometry and swept source OCT imaging. We have developed simple models which can predict the performance of these lasers as a function of frequency sweep speed. Two limits which characterize the dynamics of the lasers are identified, corresponding to the “saturation limit” and the “single roundtrip limit”. A frequency swept laser can be operated in three characteristic regimes: (1) the “saturation regime” for frequency sweep rates below the saturation limit, (2) the “multiple-roundtrip regime” for frequency sweep rates between the saturation limit and the single roundtrip limit and the (3) “post filtering regime” for frequency sweep rates above the single roundtrip limit. Comparison between our simple model and experimental studies supports the prediction of a bi-phasic decrease in power with increasing frequency sweep speeds. The laser cavity length or round trip time is identified as an important parameter which determines the frequency sweep speed. Higher sweep speeds can be expected using shorter laser cavities and faster round trip times. It was shown that an external booster amplifier stage can not only significantly improve the output power of the laser source, but also help to increase the frequency sweep speed. Post amplification could enable operation in the “post filtering” regime where extremely high sweep speed, limited only by the maximum sweep speed of the filter can be achieved. A real time recalibration “nearest neighbor check” algorithm for real time image preview was also described.

The amplified frequency swept laser was demonstrated for OCT imaging with axial scan rates up to 27 kHz. At 20 kHz axial scan rates, an instantaneous output power of ~45 mW was achieved. The tuning range was ~120 nm full width, or ~78 nm FWHM, at ~1300 nm center wavelength. This yields a 12.7 µm axial image resolution in air or 9.5 um in tissue. The use of a booster amplifier increased the detection sensitivity by 12 dB, up to 108 dB, without adding significant excess noise. The amplification process does not decrease the dynamic coherence length and can shape the spectrum to achieve a better pointspread function. Since the laser and amplifier system does not contain any bulk optics, it is robust, stable, and maintenance free. These results demonstrate the potential of this high speed, frequency swept laser light source for high speed and high sensitivity OCT imaging at 1300 nm wavelengths. Linear scaling of the sweep speed to higher frequencies by reducing the cavity length can be expected. In the future, it should be possible to reduce the cavity length to about 50 cm using modified fiber optic components. This suggests that frequency sweep speeds of almost 100 kHz could be achieved in the future using this all-fiber design. Finally, this paper provides general design criteria for developing and optimizing high-speed, frequency swept laser sources.