The technique of Fourier domain mode locking (FDML) is one way to overcome the physical limitation of cavity buildup dynamics. The idea behind it is to synchronize the optical roundtrip time of light circulating in the laser cavity with the filter sweep operation period of the optical bandpass [14, 27, 28]. This means one tuning period of the laser takes exactly as long as the light needs to propagate through the laser cavity. In other words, light with a certain wavelength is transmitted through the optical bandpass filter and then propagates through a very long laser cavity, realized by an optical delay line. Meanwhile the optical bandpass filter is swept over one entire cycle, covering all desired wavelengths, and just when the light arrives back at the filter, the filter transmits this wavelength again, but now already for the next wavelength sweep. This means light does not have to build up from ASE background; it is still there from the last wavelength sweep. In other words, the entire wavelength sweep is optically stored inside the laser cavity. The setup of an FDML laser is identical to a regular swept laser, besides an additional very long optical delay line. Since sweep repetition rates of f_sweep = ∼100 kHz require optical path length of ΔL = c/f_sweep = c/100,000 Hz = 3 km, usually a spool of optical single mode fiber is used. To make all wavelength components circulate at the same frequency, the chromatic dispersion of the fiber delay lines has to be compensated. The better the compensation, the higher the number of effective roundtrips of the photons inside the laser and the better the performance of the swept source [29, 30] (Fig. 24.3).

The synchronization condition of FDML operation leads to the effect that the light of a certain wavelength sees the spectral optical bandpass filter always in the same position (see Fig. 24.4, right), whereas in a standard tunable laser, light always is transmitted through a slightly changing filter (see Fig. 24.4, left). The fact the periodic operation makes the filter look “non-tuned” for the light results in a stationary operation of the FDML laser (see Fig. 24.4, right) [36].

The system design and the fact that FDML is a stationary laser operating mode yield many advantages which dramatically improve the performance of the swept laser and, consequently, the OCT system specifications:

Figure 24.6 shows the first implementation of an FDML laser for OCT in 2005 [59]. The laser is implemented in a ring geometry using a fiber-based resonator. The laser gain medium is a fiber-pigtailed packaged semiconductor optical amplifier (SOA) from Inphenix, Inc. The tunable optical bandpass filter is a Fiber Fabry-Perot Tunable filter (FFP-TF) (Micron Optics, Inc.). A spool with 7 km standard single mode telecom fiber (Corning SMF 28e) provided the optical delay of 34 μs. Figure 24.7 shows the output power of the laser for different values of tuning frequency of the optical bandpass filter. It can be seen that only for a very narrow range of filter drive frequencies f_filterdrive, the FDML condition f_filterdrive = c/l is fulfilled. The resonance has a width of about 7 Hz at a center of 29,015 Hz corresponding to a system finesse of 1/4,000.

As seen in Fig. 24.7, the laser resonator has an optical roundtrip frequency of 29 kHz. Since the Fabry-Perot Tunable filter is preferentially driven sinusoidally at such high speeds, two wavelength sweeps are generated for each drive cycle of the filter. So driving the filter with 29 kHz generates 29,000 forward (shorter to longer wavelength) sweeps and 29,000 backward sweeps per second. For OCT application, 58 kHz line rate is possible.

Obviously this laser can also be driven with higher harmonics, i.e., two or more sweeps are simultaneously active in the FDML laser. The laser has resonances at 2 × 29 = 58 kHz, at 3 × 29 kHz = 87 kHz, and so on. In the first demonstration in 2005, up to 290 kHz sweep rate had been demonstrated.

Conceptionally, FDML lasers have no speed limit as long as the intracavity tunable bandpass filter supports the high sweep speeds. It actually turns out that FDML lasers are more stable at higher speed, because the required delay fiber length is shorter for higher frequencies. This reduces thermal drift effects and problems with unstable polarization [29, 60].

However, the availability of fast tunable optical bandpass filters with the required specifications is limited. The filters should have <0.15 nm passband, >120 nm free spectral range, and >40 dB out of band rejection. Currently, at sweep speeds of more than 100 kHz, only Fabry-Perot filters achieve this type of performance. Custom-built research devices achieve up to 2 × 419 kHz [61], and commercially available devices achieve 2 × 170 kHz (Lambdaquest, Inc.). This is typically the highest mechanical resonance frequency with sufficient response. However, the achievable tuning amplitude in these resonances is typically much larger than the lasing range, even larger than one free spectral range. So the total tuning speed measured in nm/μs can be increased by increasing the amplitude.

The technique of sweep buffering [62] can convert excess sweep amplitude to sweep frequency. It is used to convert lower filter sweep frequencies with a high amplitude to high sweep frequencies with lower amplitude. Figure 24.8 shows the setup of an 8× buffered FDML, and Fig. 24.9 shows the concept of sweep buffering in the case of 4× buffering. In the setup in Fig. 24.8, a 325 kHz FFP-TF is used. The filter is driven with such a high amplitude that the duty cycle of one forward sweep is only 12.5 %, and the rest of the time over one entire sweep cycle the SOA is switched off. The sweep output is now coupled into a cascade of splitters and recombiners, each with an additional length of fiber in one arm. The length of this arms is one half, one fourth, and one eight of the cavity length. This leads to multiple copies of the sweep, here 8. After the last coupler, there are two outputs, each with 8 × 325 kHz = 2,600 kHz sweep rate. Figure 24.9 shows the individual steps for a 4× buffered FDML with the steps of overdriving the FFP-TF, generating a short duty cycle, ON-OFF modulation of the SOA, copying and delaying the individual sweeps. It should be noted that sweep buffering is not limited to FDML lasers; it can be applied to all lasers that have excess tuning rate and short duty cycle [63].

The main reason for sweep buffering is to increase the repetition rate of FDML or standard swept lasers [30, 34, 62–71]. However, there are numerous additional advantages of buffering. The most important one is that there is an almost perfect optical phase relation between the different copies of the sweep, making them the ideal source for phase-sensitive and Doppler OCT imaging [68, 69].

Another advantage is that wavelength over time tuning characteristics and also the optical frequency over time tuning characteristics of the light sources’ output is usually much more linear than in standard FDML lasers. The reason is that the FFP-TF filter in FDML lasers is usually driven in a mechanical resonance sinusoidally which leads to a very slow tuning operation near the turning points. On one hand, such a huge variation in filter sweep speed is problematic from an engineering viewpoint for an OCT system, because the resulting fringe frequencies span a very wide range of RF frequencies from the photo-receiver. This requires systems with a very flat electronic phase response. On the other hand, the very slow tuning speed at the turning points at a constant exposure level adds up to the total amount of optical power on the sample, but only very few data points are collected right at the turning points, considering the equidistant optical frequency grid before FFT in SS-OCT/ODFI. This means a lot of energy is put on the sample, but very little information is acquired. So sweep buffering improves linearity which can simplify OCT system design and improve image quality.

A further advantage of sweep buffering is that the polarization state of the different sweeps can be passively controlled. For many PS-OCT applications, a sequence of wavelength sweeps with alternating polarization state is required [72, 73]. This is usually realized by active optical elements with the problem of synchronization, additional differential polarization-dependent group delay, and dispersion. With the technique of sweep buffering, different polarization states can be generated passively [60].

For these reasons, buffering stages are an integral element in many of the most advanced MHz FDML lasers.

The idea of FDML is to synchronize the optical roundtrip time of light in the laser cavity with the filter tuning period. But this would require that all the different wavelength components of the FDML laser propagate with the same speed. Due to chromatic dispersion in the delay fiber, this is not the case which leads to a more or less pronounced miss synchronization for some wavelengths [29]. The first FDML lasers have been implemented at a center wavelength of ∼1,310 nm. This is the zero dispersion point of standard optical fiber, meaning in a narrow range around 1,310 nm, all wavelengths propagate at the same speed. However, zero dispersion point means only that there is no linear group velocity dispersion (GVD) (i.e., second-order dispersion), so no term measured in ps/(nm^∗km). However there still is remaining GVD slope measured in ps/(nm^{2 ∗}km). And because typical OCT lasers sweep over a very wide spectral range of 100 nm or more, this has a significant effect on the roundtrip time. Figure 24.10 (left) shows the difference in roundtrip time for the various spectral components in a 100 kHz 1,300 nm FDML laser (2 km fiber).

It can be seen that light with 1,240 nm has a 550 ps longer propagation time through the cavity than light with 1,320 nm. At a roundtrip time of 10 μs, this is a synchronization error of 50 ppm (5e-5) which is small enough that it doesn’t affect the output power of the FDML laser. However, it also means that light will not circulate in the FDML cavity forever; after a certain number of roundtrips, the timing offset will be large enough that the light is blocked by the filter. In the shown case, this reduces the number of roundtrips of the intracavity photons in the FDML laser to approximately 12 at the very edges of the spectrum [30]. This negatively affects the instantaneous linewidth of the FDML laser reducing the instantaneous coherence length. For this laser the practical single-sided OCT ranging depth is reduced to about 5 mm, i.e., the OCT depth range over which almost no signal decay can be observed. The effect of chromatic dispersion is even more severe in FDML lasers at 1,050 nm as used for retinal imaging. There the chromatic dispersion of optical fiber has a substantial linear contribution of ∼40 ps/(nm^∗km).

The solution to this problem is to compensate the intracavity dispersion of the FDML. The different approaches that have successfully been demonstrated are the use of a wideband zero dispersion photonics crystal fiber [64, 75], the use of a sequence of different fibers to cancel out the dispersion contributions of the individual pieces [29, 76], and most recently the application of chirped fiber Bragg gratings (cFBG) [30, 34, 35]. A cFBG is a piece of fiber with a periodic refractive index modulation where the period changes over length. This way, light with different wavelengths is reflected in different depths yielding a wavelength-dependent propagation time. The cFBGs can now be designed in a way that they exactly compensate the chromatic dispersion of the FDML cavity. Since usually cFBGs are preferred with a monotonic group delay over wavelength, a pair of matched cFBGs is used at 1,320 nm, because the dispersion zero of standard optical fiber leads to a minimum of propagation time. Figure 24.10 (right) shows the setup of a compensated 1,300 nm FDML, and a 4-port circulator is used in the light path to generate the two reflections from the cFBGs. The pair of cFBGs was designed and manufactured by TeraXion, Inc. (Quebec City, Canada). Figure 24.11 (left) shows the group delay of the two cFBGs and the total group delay. Starting from more than 200 ps timing error for the different wavelength components of a 100 nm range, less than 4 ps remain after compensation (Fig. 24.11 right); this is a reduction of more than 50×. The relative timing error of 4 ps compared to the 10 μs tuning period corresponds to a relative error of 400 ppb (4e-7). This value theoretically enables several 1,000 roundtrips of light in the cavity [30].

Because of this higher number, light is effectively filtered more often improving the instantaneous linewidth. The sensitivity roll off is reduced; the OCT system ranging depth improved to more than 10 mm (>21 mm 6 dB coherence length). Figure 24.12 shows the improvement; the point spread functions are plotted over OCT imaging depth. For the non-compensated FDML, a 6 dB roll off over 5 mm is observed. Using dispersion compensation this value is more than doubled. The measurements in Fig. 24.12 may have been limited by the detection electronics rather than by the laser linewidth properties, so the real roll-off performance of this laser can be even much better. This increase in total OCT imaging range is most important for applications in intravascular imaging, where a >5 mm range is usually desired to better image some anatomic features like arterial branches [30].

Due to their setup with a long length of fiber, FDML lasers offer the ideal starting point for Raman amplification as laser gain process [46, 49–51]. Raman amplification in fibers is widely used, because the small mode field diameter in optical single mode fibers over a very long interaction length compensates for the very small Raman scattering cross section. The advantage of Raman gain is the potentially very flexible range of operating wavelengths and the good noise performance. It has also been used to study the photon life times and loss mechanisms in FDML lasers [51].

In the 1,550 nm wavelength range, which can also be used for OCT, especially if samples with low water concentration are imaged [15], an erbium-doped fiber amplifier (EDFA) can be used instead of an SOA as gain medium [47, 48]. EDFAs have also a better noise performance than SOAs, and they can achieve higher output powers. Disadvantages of EDFAs are the smaller amplification bandwidth and the tendency to Q-switch because of the long carrier lifetime of ∼20 ms. If EDFAs are used for fast swept lasers and the sweep rate is increased, carrier relaxation oscillations get stronger and high output intensity fluctuations are observed. At some point the laser goes into Q-switching mode emitting a sequence of pulses. Due to the high power of such pulses, they can destroy the intracavity filter.

At 1,050 nm an ytterbium-doped fiber amplifier (YDFA) can be used as amplification element [44–46]. In [45] this was used in combination with an SOA. YDFAs have the advantage of good noise performance and the capabilities for very high-power output. Their gain profile covers an extremely wide wavelength range of much more than 100 nm around 1,050 nm.

The output powers achievable with YDFAs are scalable into the Watt region. Figure 24.13 shows the case for an FDML laser used for retinal imaging. Considering the duty cycle of this laser, the peak power was larger 400 mW. Especially since in the 1,050 nm region swept laser sources often do not exhibit sufficient output power, YDFA-based FDML lasers appear very attractive.

Even though in most cases FDML lasers are built with FFP-TFs, the FDML concept is not linked to a special type of optical bandpass. Leung, Liu, and Mao et al. demonstrated an FDML laser using a grating filter based on a polygon scanner [40–42, 79]. The advantages of this device are unidirectional scanning operation and very good sweep linearity without buffering.

One problem of FDML operation, especially if an FFP-TF is used, is the control and automatic stabilization of the wavelength sweep filter. Sweep amplitude, offset, and frequency have to be controlled; thermal drift has to be compensated. Different active control loops have been demonstrated. A very interesting concept is the application of a regenerative mode locking concept [79]. The optical output of the FDML laser is used to control an electronic signal that drives the filter. This way, temperature-dependent changes in the optical roundtrip frequency are automatically compensated for, and the laser always finds the best frequency.

At the time of the first introduction in 2005, the performance of FDML lasers was quite unique, and hardly any other technology could achieve the combination of sweep speed, linewidth performance, and spectral sweep range. However, over the last few years, other swept source technologies have been developed that now approach and may even exceed FDML performance in some aspects. In the following sections the state of the art in the year 2013 will be shortly reviewed with respect to performance characteristics of these new sources compared to FDML. Currently there are many promising ideas and concepts for fast wavelength-swept light sources intended for use in OCT, but in several cases, it has not finally been proven, whether these sources can really achieve good imaging performance in biomedical applications [17, 80–82]. So here the focus lies only on three non-FDML techniques and light sources where OCT imaging has already been demonstrated at speeds in excess of 200 kHz, because the high sweep speed performance is the most outstanding characteristics of FDML laser.

The first class of high-speed swept light sources are lasers where the cavity length is reduced so far that the lasing buildup dynamics is fast enough to achieve high-speed tuning. Since this class of sources does not solve the problem of laser buildup dynamics, it can be expected that they are not arbitrarily scalable to higher speeds and probably only several 100 kHz sweep rates are possible, in case other typical OCT performance criteria are maintained. It should be noted that a reduction of the laser resonator cavity length in classic swept laser designs using a separated gain medium and an optical bandpass filter is only possible up to a certain amount. Since the cavity length defines the mode spacing of the laser resonator and this mode spacing becomes larger as the laser resonator length is decreased, the possible spectral sampling density becomes coarser. In other words, if the cavity length of a linear laser resonator is, e.g., 5 mm, the optical mode spacing of the resonator modes is Δf = c/2 l = 30 GHz. At 1,300 nm wavelength, this corresponds to 0.17 nm spectral separation which is, in OCT application, the best achievable spectral sampling density. This corresponds to an OCT imaging range of 2.5 mm. So with a standard laser, it is not possible to build an SS-OCT/OFDI system that has a single-sided ranging depth of more than the laser cavity length. In reality, the laser resonator length is typically chosen significantly longer to avoid beat noise between individual modes. Therefore highly integrated short cavity lasers have the laser end mirror outside the package in the fiber pigtail, resulting in typical cavity roundtrip lengths 30 cm, about 1 GHz optical mode spacing. According to the single roundtrip limit presented in [14], it can be expected that such lasers are limited to <300 kHz sweep rate if a 0.05 nm linewidth and 100 nm sweep range are chosen. The problem of discrete mode spacing may theoretically be overcome by advanced variable cavity designs or an active phase section in the laser resonator, as used in telecom applications [17]. However, the laser dynamics at fast sweep operation and the influence on OCT image quality have not been investigated yet. The most prominent examples of fast short cavity lasers are the ones from Oh et al. [63] as research systems and the ones from Axsun technologies [84] and Santec [85]. A more detailed discussion of this laser type can be found in ..” reference to Bart Johnson (AXSUN), Bill Ahern chapter in this book.

The second class of successful high-speed tunable lasers are mechanically tunable vertical-external-cavity surface-emitting lasers (VECSEL) (see Chap.​ 22, “VCSEL Swept Light Sources”). These devices can achieve very fast tuning operation. It is often argued that the short cavity length of these devices promotes very rapid buildup. However, the most recent results showing meter long coherence lengths, equivalent to >10 ns coherence time, are not possible at the typical sweep rate of >>1 GHz/ns tuning if lasing is repetitively built up, since this would violate the time bandwidth product. So the reason for the good coherence performance at high tuning speed is that the Doppler shift caused by the reflection of light from the moving end mirror automatically adjusts the wavelength of the light field such that it always matches the resonance condition of the tuned laser cavity. The calculation can be found in [26]. For this reason, VECSELs have also no fundamental tuning speed limit.