1. Introduction and background

The effects of atmospheric distortions can be practically eliminated in high-resolution ground-based astronomical imaging via the use of high-brightness laser guide star (LGS) assisted adaptive optics (AO) correction loops. Such real-time systems use lasers tuned to the Sodium D_2a emission line at λ ~589.159nm, and require high laser powers. Furthermore, near-maintenance-free, ruggedized systems are preferred for LGS applications in remote observatory locations, especially when considering the reliability and availability of multiple-LGS systems, and the mounting of the lasers directly onto a gravity-vector-variant element (e.g., the telescope altitude structure, changing in elevation angle with observations) [1]. Unfortunately, no known solid-state medium has sufficient gain directly at 589nm, and existing alternative dye laser systems typically remain highly maintenance intensive and alignment sensitive. Up to 50W CW 589nm single-line radiation has until now only been produced via sum frequency generation (SFG) in a solid-state system [2]. We describe here a successful approach for generating the same 50W power levels at MHz-class 589nm linewidths, based on the frequency doubling of state-of-the-art (coherently, collinearly combined) optical fiber systems. We limit the following study to optical fiber systems.

Sum-frequency-mixed [3] or frequency-doubled infrared fiber amplifier sources are considered well suited to this unique LGS application, even if the frequency-mixed fiber-based systems currently remain somewhat power-limited. An overview of recent developments in Bismuth-doped [4], extended-wavelength Ytterbium-doped [5,6] and broadband Raman fiber systems [7,8], has been offered in [9]. In that same paper, we presented results on our frequency doubling of a single SBS-suppressed Raman fiber amplifier (RFA), achieving 14.5W CW 589nm radiation with linewidth Δν₅₈₉ < 7MHz.

We focus this report on the power scaling of the described 589nm source well beyond the results presented in [10], whilst maintaining its spectral, temporal and spatial optical qualities. Setting aside for now the challenging development of a much more heavily SBS-suppressed RFA, we present for what we believe to the first time, the cascaded coherent collinear combination of a seed-split triplet of such narrow-band high-power Raman fiber amplifiers, adding a third and higher power (40W-class) output to the already combined pair.

Fiber amplifiers (FA) and fiber lasers (FL) are indeed suitable candidates for power scaling via coherent beam combination (CBC), given their characteristic high beam qualities, potential narrow emission bandwidths and ease of polarization control. In those spectrally intense and narrow-band cases, these systems are more typically limited by nonlinear effects such as stimulated Brillouin scattering (SBS) than by material damage thresholds, although high-power fiber sources are known to exist at the kW, or multi-kW levels [11]. Good overviews of coherent addition methods are offered in [12,13] amongst others. Although not of direct interest here, we note that incoherent (wavelength, or spectral) beam combination has also been successfully completed using Yb-doped fiber systems [14]. Commonly used schemes for the CBC of FA or FL, can be either ‘aperture filling’ (providing a coherently overlapped beam in the far-field only, and typically suffering a beam quality degradation due to the necessarily less-than-unity fill factor of the array [15,16]), or via collinear interference. Given the beam quality limitations inherent to the ‘aperture filling’ schemes, we do not consider them here.

The collinear interference technique results in a spatially and coherently combined beam in both the near and far field, with combined beam quality typically similar to that of the individual inputs. Lei and Feng [17] use a Michelson-type cavity to combine a fiber laser pair (both coherently and incoherently, via tuning of one of the laser frequencies), to achieve a combined brightness ‘nearly twice’ that of the individual lasers. Cheung et al. [18], use a diffractive element to obtain better than 91% CBC efficiency using five master-oscillator power-amplifier (MOPA) configured fiber systems. The beam quality in both cases remains excellent (with M² ~1.04 in the latter case). In principle, scaling this summation technique to larger number arrays of single-mode fiber amplifiers (using bulk optics) appears possible [19].

We note that the all-in-fiber coherent beam combination of a pair of fiber lasers has also been demonstrated [20,21] and scaled to include up to 8 inputs, with high resultant output port contrast [22]. However, due to the non-polarization-maintaining (non-PM) nature of our RFA, we are limited here to the free-space collinear summation of our outputs. We are nonetheless actively pursuing an industrial development on PM RFA, and their all-in-fiber CBC and subsequent doubling to 589nm [23].

This paper is structured as follows. We present for what we believe to be the first time, a detailed analysis of the voltage signals returned to the all-in-fiber phase control elements for efficient free-space collinear combination of a pair of seed-twinned non-PM SBS-suppressed narrow-band (sub-1.5MHz) Raman amplifiers (operating here at 1178nm). We then relate the quality of this feedback to the intensity noise of the combined output, considering this an essential part of system understanding required for further power scalability. We then demonstrate, also for what we believe to be the first time, the cascaded CBC of a split-seeded triplet of such amplifiers (in a nested Mach-Zehnder-type configuration), and confirm the validity of the power scalability via frequency doubling of this cascaded-combined beam to more than 50W at 589nm.

2. Experimental set-up and procedure

For what we believe to be the first time, we report on the cascaded (collinearly interfered) CBC of a triplet of split-seeded non-PM SBS-suppressed high-power narrow-band Raman fibre amplifiers (RFA), and the frequency doubling (second harmonic generation, SHG) of their cascaded-combined output (whilst maintaining high spatial, spectral and temporal optical output qualities). A schematic of the experimental layout is offered in Fig. 1 .

 

Fig. 1 Schematic of the overall experimental configuration.

Download Full Size | PDF

We seed all three RFA using a single external cavity diode laser (DL-Pro system from Toptica Photonics AG [24]). This seed provides a fiber-coupled P₁₁₇₈ > 35mW, within a specified linewidth ∆ν < 100kHz, and is protected against eventual return power from the RFAs, using a 1210nm-centered free-space optical isolator from Linos [25] (in this case, isolation-optimized for use at 1178nm, providing ~0.6dB insertion loss and ~32dB isolation). A first (2x2) 50:50 in-fiber splitter provides the pre-amplifier to the 40W-class power amplifier with approximately half of the available seed power, and we use another similar splitter to seed the remaining RFA pair (providing nominally ¼ of the available seed power to each). We make use of the fourth (seed-side) port of each 50:50 splitter to monitor the summed fractional parts of light returning from the respective RFAs. All selected fibers and fiber components are strictly single-mode at their operating wavelengths.

The 20W-class RFA pair is pumped as before [10], but we now recycle the unconverted pump from one of them (RFA₂₀) to pump a 1W-class preamplifier (comfortably achieving a non-optimized ~20dB signal gain in that sub-system). The output from this preamplifier is optically isolated (~65dB inter-stage isolation achieved using both in-fiber and free-space isolators), and further amplified in a 40W-class SBS-suppressed power amplifier, RFA₄₀ [26]. We use a separate in-house-built 150W-class 1120nm Raman fiber oscillator (RFO) [27] for pumping the power amplifier section of this third RFA. In a system better optimized for wall-plug efficiency, one might make use of the remaining 1120nm power output from the RFA₄₀, to replace one of the 75W RFO.

We collimate the amplified 1178nm outputs (1/e² radius, ω₀ ~0.62mm), and recover the polarization to linear, using a LabView-controlled pair of quarter- and half-wave retardance plates (resultant stable polarization extinction ratio PER > 25dB, despite the use of non-PM amplifiers). All RFA are then (separately) optically isolated using transmission-optimized free-space isolators, similar to that used at the seed. Using these three free-space beams, we are able to construct a pair of (effectively nested) Mach-Zehnder interferometers (c.f. Figure 2 .).

 

Fig. 2 Schematic of the free-space cascaded CBC layout and subsequent SHG. The complete set-up is built on a 50x65cm breadboard (the SHG cavity covers approximately 15x20cm).

Download Full Size | PDF

We first align each beam to the SHG cavity (used here as a spatial filter, whilst scanning it in a Fabry-Pérot (FP) type mode and with the two other beams blocked). In practice, we use the mirror pair M₁-M₂ to define a common axis centred through the mode-matching lens and couple this single beam to the doubling cavity, using M₃-M₄. We then use M₆-M₇ to cavity-couple the output of the RFA₂₂, after using M₅-M₆ to define a common axis for the primary CBC. We finally use only the mirror pair M₈-M₉, for overlapping of the RFA₂₀ output.

We aim to actively minimize the power in one of the two arms of each the interferometers via control of the relative phase between the respective RFA outputs. We first constructively interfere the 20W-class RFA pair to obtain a single combined beam, which we then collinearly sum with the output from the 40W power-amplifier. Fine-tuning of the optical alignment is achieved by dark-port-power minimization of the two (nested) interferometers.

The phase control scheme is based on feedback to a woofer-tweeter configured all-in-fiber piezo-stretcher pair and is able to provide a continuous and mechanical-resonance-free response over the frequency range [DC-57kHz] (the soft cut-off frequency between the fast- and slow-response stretchers is empirically set in the range [50-100Hz]). The fiber stretchers (schematically represented by PZ₁ and PZ₂ in Fig. 1.) were purchased from Optiphase Inc [28]. For reasons of polarization stability, we are forced to use an entirely PM front-end, including all fiber components up until the input to the RFAs (where we are able to adjust or optimize the input polarization to these non-PM systems).

The electronics (Digilock110) and associated software used for the relative phase detection and feedback control were purchased from Toptica Photonics AG; these implement the well-established Pound-Drever-Hall (PDH) phase stabilization technique [29]. In order to efficiently drive the fiber stretchers, the woofer-tweeter frequency-split outputs from the Digilock110 (typically in the range ± 5V) were offset and amplified to within 0-150V, and the signal to the slow PZ₁ was additionally filtered using tuned 1/RC circuitry, so as to reduce the effects of added or amplified signal noise on the quality of the feedback signal. As reported in [10], we make use of a software feature programmed to ‘snapback’ the long-range stretchers to their center positions, as they drift to a defined value close to either of their extrema. In this manner, we are able to continuously track (warm-up transient) relative phase drifts over a range significantly broader than the otherwise high-voltage-amplifier limited ± 500waves. The post-snapback relocking of the CBC to a minimum at D_CBC occurs after typically ~10ms.

The sideband modulation for the PDH lock-in signals used by the Digilocks are generated directly on the fast (PZ₁) in-fibre-stretchers of each phase control loop, and are set at 60.1kHz and 65.1kHz for the first (low-power) and second (cascaded, effectively 3-beam) CBC loops, respectively. We note that these selected frequencies are just higher than the first (narrow) mechanical resonance of the PZ₁ devices, and for this same reason, we are not able to lock the CBC with modulation frequency in the range [55-58kHz]. We optimise all tuning parameters for each loop for best dark-port minimisation (and thus, efficiency) and overall system stability (iterating in practice, electronics tuning and optical alignment, for best results).

Finally we frequency-double the cascaded collinearly combined 1178nm beam to generate 589nm using a PDH-locked, bow-tie configured SHG cavity, also purchased from Toptica Photonics AG. The only modifications made to the system were for appropriate evacuation and dumping of the non-locked 1178nm power (as reflected off the first surface of the input coupler), and further attenuation of the optical signals to the photodetectors controlling the cavity locking and monitoring (photodetectors D₂ and D₁ respectively, as shown in Fig. 2.). The generated 589nm beam was then folded onto a 200W-capacity water-cooled thermal sensor (LM200BHTD from Coherent GmbH) using a high reflector at 589nm (also, having high transmission at 1178nm), further enhancing the already extremely high 589nm-purity of the beam. No visible collimation optics were used. The combined 1178nm power was measured between mirrors M₃-M₄, using a similar 200W-capacity thermal sensor.

We present and discuss our results on the primary (seed-twinned) and cascaded (RFA-triplet) CBC, and their subsequent SHG, in the next section.

3. Results and discussion

3.1. Free-space collinear CBC

In this section, we present a detailed investigation into the quality of the primary CBC, including the PZ phase control element return voltage signal and output 1178nm intensity noise power spectra. We then present results on the cascaded CBC of a seed-split RFA triplet.

We begin by noting that due to the additional power division required for seeding the preamplifier sub-unit (c.f. Fig. 1.), we seed the 20W-class RFA at approximately half their design power, and at half the power as compared to the results presented in [10]. However, we observe no significant difference in overall system behaviour under these conditions.

We present the low-frequency power spectrum of the intensity noise of the amplified output of RFA₂₀, in Fig. 3 . (Here, as in all following figures, the power spectra are sampled at a 5Hz update rate and averaged 200 times, and are recorded using a feature of the Digilock110 control software). Only insignificant differences in behavior are observed between each of the 20W-class RFA and consequently, we consider the presented data and subsequent analysis representative of both. We address the long-term stability and spectral properties of all RFA outputs in a later section.

 

Fig. 3 Power spectrum of the output intensity noise of RFA₂₀ at P₁₁₇₈ = 17W over the range [DC; 2.5kHz]. The spectral noise remains white and featureless towards higher frequencies, measured up to 2.25MHz (limited by the detector response time).

Download Full Size | PDF

The intensity noise power spectra reveal no additional significant spectral features over the range [DC; 2.25MHz] (upper frequency limit dictated by the detector response time, and with approximately 40mW at the detector). The white noise level is approximately 30dB suppressed relative to the (resolution limited) DC component, and the remaining existing features are typically only 1.5-2dB above the noise floor. As a fraction of the mean averaged power and at P₁₁₇₈ ~17W, we measure the root-mean-square (rms) fluctuations in the temporal trace to be in the range 5.9-7.3 10⁻³ (over suitably sampled time-periods, in the range [20μs-20s]). We consider these values consistent with the (essentially featureless) white noise floor level, as described.

We allow the output power from each of the RFA to stabilize (requiring a few minutes only) then lock and optimize the primary CBC phase control loops. Near-unity coherent combination efficiency is achieved at all power levels, and we attribute this to both the high stability of our narrow-band 1178nm RFA, and the quality and tuning of the CBC control loop electronics. We note that the most efficient CBC requires the ratio of the incident powers precisely match the actual split ratio of the nominally 50:50 beam combiner (ideally nulling the power at D_CBC). To this end, we are able to fine-tune the 1120nm pump power to each RFA independently, and make only minor adjustments to the spatial overlap, even at the highest powers. Results on the efficiency of this primary CBC are discussed in a later section.

When the phase control loops are not locked, power oscillations are naturally observed in both ports of the effective Mach-Zehnder interferometer. These follow the relative phase difference between the two RFA, and we consider the ratio of the minimum to maximum powers at the dark port (D_CBC) indicative of the achievable combination efficiency. A residual (non-zero) power remains at D_CBC, partly due to a slight mismatch in incident powers from each RFA, but also likely, due to non-cancelling phase- and frequency-dependent intensity noise between the amplifier outputs (a study of which, would constitute interesting further work).

Power spectra of the 1178nm intensity noise are recorded under both free-running and locked (D_CBC minimized) conditions, at the bright port of the first quasi-Mach-Zehnder interferometer and after attenuating the peak (or locked, combined) power at the detector to P₁₁₇₈ ~43mW. Results are presented in Fig. 4 .

 

Fig. 4 (a-d). Intensity noise power spectra of the unlocked (red), and coherently locked (blue), combined ‘bright’ 1178nm signal indicating low- and high-frequency noise properties.

Download Full Size | PDF

The intensity noise power spectrum of the free-running (unlocked) collinear output is considered indicative of the relative phase noise between the two amplified outputs. This noise spectrum is somewhat pink in the low-frequency range (typically [DC; 1kHz] as shown in Fig. 4a.), where it also carries certain distinct features; thereafter, it also becomes typically white and mostly featureless (as in Fig. 4b., but also over the measured range [1; 500kHz], where it is ~25dB suppressed relative to its DC component). This observation is in good agreement with the phase noise power spectral measurements presented in [30] and [31], despite those results quantifying phase noise in 1W-class commercial Yb-doped FAs.

We see that essentially all features in the intensity noise power spectra of the non-locked combined beam are eliminated under locking, as the residual noise spectrum becomes white over the entire measured frequency range [DC; 500kHz] (at typically −30dB with respect to the DC component, Fig. 4c.). An unexplained feature at 50kHz however appears in the locked signal spectrum (Fig. 4d.), but we determine that it carries no significant power and hereafter neglect it. We consider the intensity noise power spectrum of the combined output essentially indistinguishable from that of either single amplifier (again, c.f. Figure 3.), and conclude that our phase-difference correction scheme is adequate both in bandwidth and response.

This result is in excellent agreement with that of Musha et al. [32], who demonstrate that under their experimental conditions, the intensity noise power spectra of a single injection-locked and collinearly combined pair of similarly twin-seeded (but single-line, stabilized, bow-tie cavity, injection-locked) Nd:YAG lasers are also essentially indistinguishable. They further investigate the residual phase and frequency noise of the combined system, and achieve an excellent 0.94 combination efficiency at a 4W-level. This result is even more impressive in that they use a single piezo-driven element for the phase control, whereas we use a pair of such devices, separately compensating for the high- and low-frequency relative phase differences. However, they appear to have no means of confirming or quantifying spatial overlap in either the near- or far-field, and note that the quality of the spatial overlap of their two beams was (to some extent) coupled with the correction of the relative phase noise between arms, thus becoming a (rather minor) source of intensity noise in that combined beam. In contrast, we perform our phase correction entirely in-fiber, and so do not observe such an effect.

We investigate the time-traces of the corresponding voltage signal using an Agilent oscilloscope (sampling at 200Ms.s⁻¹), and determine that the ratio of signal rms to its time-averaged value remains more or less constant over the temporal range [20μs-20s], at around 8.6 - 8.8 10⁻³. We consider this value consistent with the white noise floor level of the power spectra of the combined beam. Assuming these intensity fluctuations are consistent with both values derived from Figs. 3. and 4c., and with the achieved combination efficiencies, we estimate that approximately half the non-combined power is lost due to intensity noise and/or spectral phase fluctuations, and hypothesize that the other half is lost due to spatial overlap imperfections (axial, tilt and curvature mismatches). As a corollary, we expect no significant intensity noise to be sourced from the use of the stretchers themselves (likely, given they carry little to no power at frequencies close to their mechanical resonances).

We conclude the study on the short-timescale behavior of this primary (2-RFA) CBC with a comparison between the above, and the woofer-tweeter frequency-split response provided to each fiber stretcher from the Digilock110. Power spectra of the corresponding signals are presented in Fig. 5 .

 

Fig. 5 (a-d). Power spectra of the voltage signals to the slow (woofer, PZ₁) and fast (tweeter, PZ₂) all-in-fiber piezo-stretcher phase control units (under CBC₁, of a pair of RFA).

Download Full Size | PDF

As expected from the difference between the free-running and locked power spectra over the range [DC; 1kHz] and as shown in Figs. 4a, c., we observe (as shown in Fig. 5a.) a pink-noise compensation returned to the stretcher PZ₁ (‘woofer’) for frequencies below the selected cut-off frequency (100Hz in this case). However, we see here that the main correction contributions from this stretcher lie in the near-DC range, given the particularly weak transferred spectral powers at all other frequencies. Interestingly, we note that the cut-off frequency between the woofer-tweeter pair is soft, and that the signal to the tweeter unit (PZ₂) actually carries significant power over this slow frequency range (also a near-pink, but heavily featured response, as shown in Fig. 5c.). Nonetheless, and given that the ‘woofer’ response is approximately 42-45 times stronger than that of the tweeter over the range [DC-10kHz] (12-16μm.V⁻¹ vs. 0.28-0.35μm.V⁻¹ respectively, with both responses slowly and monotonically increasing with increasing frequency), it becomes clear that this slower device is likely responsible for the elimination of the majority of the features in the free-running power spectra (Figs. 4a, b.). Indeed, we note the similarities between the spectral features over the range [DC; 800Hz] in Fig. 4a., compared to those in Fig. 5b.

We clearly observe the 60kHz modulation frequency applied to the tweeter unit for error signal generation, for the CBC locking (Fig. 5d.). This signal appears strongly overlaid with harmonic or near-harmonic features, and we speculate that these are sourced by some interaction between the applied compensation and modulation signals (the modulation-free compensation signal to the slow piezo-stretcher does not display such features). As previously stated, we observe no significant intensity fluctuations in the coherently combined signal (or features in its power spectrum), even at the first and strongest mechanical resonance of the device (specified close to 57kHz). We so conclude, that we do not excite this resonance. The spectral intensity of the voltage signal to the tweeter is approximately −40dB relative to the DC component at this frequency, and so consequently, we consider all other applied signals (with smaller integrated spectral power) as without significant effect on the tweeter response. A comparison with the specified device response [28] leads to the notion that in fact, the most important of the remaining harmonic features might be the one close to 20kHz (with peak spectral intensity approximately −33dB relative to the DC component, c.f. Fig. 5d.). We consider this to be responsible for the elimination of the (unlocked) features at the same frequency (as seen in Fig. 4b.), finally resulting in the near-ideal white noise residual in the coherently combined power spectra. For all practical purposes and if so required, we might limit the correction bandwidth to a few kHz without significantly compromising the quality of the collinear CBC.

A pickoff from the feedback signal to the slow stretcher, PZ₁, reveals temporal information on the slow-timescale relative phase stability between the two RFA. We present results elsewhere [10] but recall here that only a few snapbacks occur during a 90minute warm-up transient (equivalent to ~1400waves of relative phase drift, excluding the switch-on time required for validation of the proper spatial overlap between the beams, a couple of minutes only). After this transient period, we are able to operate snapback-free for over 40hours (limited only in that particular case, by a failure of one of the polarization control rotation stages; these would be completely eliminated in the case of all-in-fiber CBC of PM RFA). We calculate the average relative phase drift to be ~7.5 waves/hour over that period, and expect to be able to reduce this value further, by appropriate system engineering.

We now consider the cascaded CBC of the RFA triplet, via the addition of a third (higher power) RFA to the already combined RFA pair (as shown in Fig. 1.). As mentioned in section 2 and as reported elsewhere [26], we develop a pre-amplifier/power-amplifier system capable of 40W CW, without measurable line broadening (the measured linewidth remains resolution-limited at ∆ν₁₁₇₈ < 1.5MHz). Even if the RFA₄₀ remains SBS-limited at its highest output power, we nonetheless consider that these systems represent the state-of-the-art in both narrow-band SBS-suppressed RFA and Raman fibre laser technology, at any wavelength.

Using duplicate and appropriately tuned hardware and control loops, we first lock the primary (2-RFA) CBC to an optimized dark-port minimum at a given power level, then repeat the process optimally combining the resultant beam with the output from the RFA₄₀, over a range of power levels. The cascaded (3-RFA) CBC system is then tuned for overall stability and efficiency. Defining efficiency as the ratio of combined power to the sum of combined and dark port powers, we present results in Fig. 6 .

 

Fig. 6 Efficiencies of the collinear coherent primary- and cascaded-combinations (red circles and blue triangles respectively), including the overall combination efficiency (pink squares). All values are plotted against total cascaded-combined (useful, polarised, isolated) 1178nm power.

Download Full Size | PDF

We achieve typically 95% collinear combination efficiency at, importantly, both the primary and cascaded CBCs over the full range of SBS-free powers (η_CBC ≥ 95% for P₁₁₇₈ ≤ 50W, and η_CBC ≥ 94% for P₁₁₇₈ ≤ 58W). We consider this a meaningful demonstration of power scalability using this method and we expect further system scalability using the same technique, whilst maintaining similarly high combination efficiencies and low resultant intensity noise. Up to P₁₁₇₈ ~60W CW is obtained here, limited only by SBS onset in one or more of the RFA. To the best of our knowledge, this represents the highest achieved 1178nm power from any system and at this linewidth, and we are continuing work on the SBS suppression.

We note, however, a roll-off in CBC efficiency at the high-power end of the respective curves, and attribute this to phase-irrecoverable intensity noise likely generated by a low-level onset of SBS (seen in Fig. 6. and under these experimental conditions, at 27W for CBC₁ and 59-60W for CBC₂). We further note that the high-efficiency CBC₁ power limit was a little low for these given boundary conditions (relative to previous work), and that we expect to be able to generate an additional 1-2W combined power by fine-tuning the polarization input to the RFA₂₀ (effectively, increasing its SBS threshold). In any case, we do not intend to operate in this regime. We nonetheless consider that the characterization of SBS-induced intensity noise (and its consequences on CBC quality) would constitute interesting further work. The earlier analysis on the respective intensity noise power spectra remains representative and valid, however, given they were taken well below the respective SBS onset levels. Finally, we note that the behavior of the overall CBC efficiency essentially follows a contributed-power-weighted product of the CBC₁ and CBC₂ efficiencies, and so resembles that of the CBC₂.

Given the essentially indistinguishable combination efficiency between the primary and cascaded CBC loops, and the similarities between the intensity noise power spectra of both the single RFA₂₀ and combined ‘bright port’ of the CBC₁, we consider that any eventual intensity noise contributions from the primary into the cascaded CBC do not significantly degrade the quality of this dependent CBC loop. Given the very nature of coherent (phase) combination, the same assumption applies to any eventual phase noise generated under CBC₁. Consequently, we do not consider it necessary to complete a similarly detailed investigation into the fast-timescale characteristics of the cascaded CBC properties. Furthermore, we have already reported that the CBC₁ is highly temporally stable (as measured over periods of tens of hours), and so only plan to indirectly infer the stability of this cascaded CBC through a measurement of that at 589nm.

Finally, we note that the linewidth measurements of all three RFA outputs remain resolution limited at Δν₁₁₇₈ < 1.5MHz (even at their highest output powers), using an FP optical spectrum analyzer (OSA) having free spectral range ~1GHz, and finesse ~650 (the specified seed linewidth was given as Δν_seed ≤ 100kHz). We recall that the linewidth of the first combined RFA pair also remained measurement resolution limited. Consequently, we expect to obtain the same unresolved Δν₁₁₇₈ < 1.5MHz from the cascaded-combined configuration. Similarly and given both, the fundamental physical properties of spatial cleaning under cavity-enhanced SHG, and previously reported results [9,10], we also expect a near-diffraction-limited 589nm beam at the output of the cavity under SHG of the combined RFA triplet, and do not measure this here.

We investigate the frequency doubling of the combined and cascaded-combined beams in the next section.

3.2. SHG to 589nm

We frequency double the coherently combined 1178nm beam using the same SHG cavity as for previous work [9,10,26]. Specifically, we use a 30mm long, non-critically phase matched LiB₃O₅ crystal, operating at temperature T_OP ~40°C. Further cavity-specific details can be found in [9] and are not repeated here.

We first confirm the proper alignment of the CBC beam to the frequency-doubling cavity, using it in a Fabry-Pérot-type scanning mode (a ramp voltage is applied to the piezo-element driving the flat full-mirror, c.f. Figure 2.). We see that when using the cascaded-combined beam at 50W, we obtain a contrast of the peak spectral intensity of the fundamental mode to that of the strongest higher order mode (HOM) of 157:1, with those to the next HOM being 489:1, 520:1 and 581:1, respectively. Consequently, we are confident of coupling essentially all power into the fundamental mode. Similarly and indirectly, we consider this a good confirmation that all RFA outputs (and the cascaded-combined beam) remain essentially diffraction limited over the full range of powers.

We first compare the cascaded-combined frequency-doubled powers and conversion efficiencies with those of previously published 2-RFA CBC data [10], as we continue to use an input-coupler reflectivity of R_IC = 90%. Results are presented in Fig. 7 .

 

Fig. 7 SHG to 589nm of a 2-RFA CBC pair, and of a cascaded-combined 3-RFA triplet (indicating both indistinguishable powers and efficiencies, respectively).

Download Full Size | PDF

We see that both sets of results are indistinguishable in both generated visible power and conversion efficiency, over the full range of IR powers and over the thermal measurement timescales considered here. Care was taken to suitably balance contributing IR powers under the CBC, and in each case, both the electronics tuning parameters and optical alignment were carefully optimised. Peak efficiencies η₅₈₉ > 86% were achieved in both cases over the IR power range [20-25W], gently rolling off in efficiency at the higher powers due to effective optical mode impedance mismatching at the cavity input [9]. Greater than P₅₈₉ > 30W (with corresponding efficiency η₅₈₉ > 82%) was thus achieved. The SHG to 589nm was not investigated with this setup (R_IC = 90%) for P₁₁₇₈ > 37W.

We hereby confirm previous expectations that we generate no intensity noise in the IR beam under cascaded CBC that translates as a degradation of SHG conversion efficiency (the relative root-mean-squared intensity noise values in the 589nm beam generated from twinned- and triplet-CBC will be compared in a later section). Importantly, we consider that this result confirms the usefulness of cascaded CBC in a practical situation, and reasonably suggests scalability to a greater number of IR units, whilst maintaining overall system viability (even at such high power and efficiency levels, in power-optimised systems).

Using a fast Si photodiode placed in a suitably attenuated part of the generated 589nm beam, we mirror the previous CBC₁ analysis and investigate the fast temporal instabilities of this 2-RFA-combined beam under SHG (P₅₈₉ ~36mW at the detector). The corresponding intensity noise power spectra are shown in Fig. 8 .

 

Fig. 8 (a-b). Power spectra of the 589nm intensity noise (slow and fast timescales).

Download Full Size | PDF

We note, consistently with the trends observed in the coherently combined intensity noise power spectra (Figs. 4c., d.), an essentially white-noise behavior of the visible signal, always better than 20dB suppressed relative to the DC component. Neglecting the high-frequency features, the main spectral details occur in the range [1; 5kHz], yet are narrow and carry only little integrated power. Consequently, given the typical requirements for LGS laser systems (with ~1kHz update rates, or slower) we consider these insignificant within the scope of this work. Consistently with the white-noise behavior and baseline levels of both the CBC and visible spectra, the integrated visible power instabilities are around 1.5% rms at the 20W-level, as measured over these sample periods and rates (suitably sampled over a temporal range [20μs; 20s]).

Given the high quality of the visible signal and corresponding spectra, we do not investigate the control signal voltage spectral power distribution or eventual drifts in the signal to the piezo-actuated mirror of the SHG cavity. We hereby confirm the earlier conclusion, that no significant intensity noise is introduced under SHG.

We now optimise the cavity for operation at higher power (for P₁₁₇₈ ~60W), changing the IC mirror for one with lower reflectivity (R_IC = 86%). After optimisation of the cavity electronics and locking parameters, and after iterating this tuning against slight optical alignment optimisations, we obtain the power and efficiency curve pair shown in Fig. 9 .

 

Fig. 9 589nm powers and efficiencies under SHG of a cascaded-CBC RFA triplet, generating P₅₈₉ > 50W CW and demonstrating peak conversion efficiencies η₅₈₉ ~85%.

Download Full Size | PDF

We achieve P₅₈₉ = 50.9W CW from a combined P₁₁₇₈ ~60.6W (η₅₈₉ ~84%), as well as demonstrating essentially the same peak conversion efficiencies as in the 25W-optimised case (η_peak ~85% and ~86%, from 56W and 22W IR, respectively; c.f. Fig. 7). We suspect that a slight focal mismatch to the cavity fundamental mode might be responsible for the somewhat reduced peak conversion efficiency; this might be sourced in the free-space optical isolator at the RFA₄₀ output, at high powers. Although these results essentially match those of Denman et al. [2] in power, we believe the efficiency of the nonlinear frequency conversion step (SHG or SFG, accordingly) is somewhat higher, in this frequency-doubled case. With that exception and insofar as we are aware, there is not a more powerful 589nm source based on any other technology. We also suspect that this is the highest CW visible power achieved from direct SHG of an IR fibre laser system.

In comparison with previous results [9,10] and with Fig. 7. above, we note an apparent fast high-end roll-off in the conversion efficiency to 589nm, for P₁₁₇₈ > 55W. We correlate this observation with that of slightly lower cascaded CBC efficiency over the same range of powers (c.f. Fig. 6.), and hypothesise, as before, that this might be due to a slight degradation in intensity noise properties coupled with a non-critical SBS onset. A study of such phenomena would constitute interesting further work.

Given the previously stated assumption that the linewidth measurement of the cascaded-combined CBC beam remains device resolution limited (and by extension of the resolution-limited visible spectral measurements reported in [10]), we expect the linewidth of the generated visible beam to remain within the resolution limit of our 589nm-centered OSA (~2.3MHz), and do not measure this here. The spatial quality of the visible beam generated via SHG of a 2-RFA fundamental was confirmed to remain essentially diffraction limited (after 11 optical surfaces for attenuation and collimation, and using a Phasics SID4 shearing interferometer [33], we measured less than 34nm rms wavefront error within the 1/e² intensity area). We do not measure the beam quality for the visible beam generated from the cascaded-combined fundamental.

As completed in [10], we measure the slow-timescale temporal behaviour of the 589nm beam at approximately 80% of its peak power value. We use a fast Si photodiode (P₅₈₉ ~57mW at detector) together with an Agilent DSO6104A oscilloscope to record the time-trace and show results in Fig. 10 .

 

Fig. 10 Time-trace of the visible power generated under SHG of a CBC RFA triplet, demonstrating 41.45W mean power over 500s, with 0.66W rms intensity noise.

Download Full Size | PDF

We measure a mean P₅₈₉ ~41.45W with ~0.66W rms instabilities over the 500s of interest (limited only by data acquisition length). This represents a 1.6% relative rms noise with respect to signal average power, which is highly comparable to the corresponding 1.5% measured under SHG of the 2-RFA CBC beam. We observe no slow degradation of conversion efficiency (or drop in P₅₈₉) over this period and infer that in addition to the previously demonstrated primary CBC stability (and high stability of the 589nm cavity locking), we are able to demonstrate at least similar stability levels of the cascaded-combined CBC. These results further confirm the validity of the method for reliable power scaling.

We hereby complete the demonstration of system power scalability whilst satisfying all basic requirements for the demanding LGS-AO applications. We recall that due to its very nature, the developed RFA technology remains highly wavelength-versatile, and its emission wavelengths (both, IR and visible) are essentially only limited by seed-pump-pair availability. Given the low-intensity-noise quality of the SHG beam, we also consider this visible output suitable for higher-order harmonic generation (or frequency mixing with another source), at power levels of multiple tens of Watts. We finally suggest that the power, linewidth and beam quality of the system might allow its use in other scientific fields [34–36] such as atomic physics, spectroscopy and microscopy, and even due to its potentially rugged design, for medical or laser-display applications.

4. Conclusions and scope

We demonstrate here for what we believe to be the first time, the cascaded collinear CBC of a seed-split triplet of SBS-suppressed narrow-band CW Raman fiber amplifiers (operating here at 1178nm). We then demonstrate the frequency doubling of the resultant cascaded-combined beam to achieve more than 50W CW at 589nm, thereby confirming excellent system power scalability at both wavelengths.

In particular, we first investigate the quality of the primary (2-RFA) CBC through a power spectral analysis of the voltage signals to its woofer/tweeter configured phase control elements, and compare the quality of this feedback and correction with the resultant intensity noise in the combined beam. We thereby also determine, for all practical purposes, a useful relative phase correction bandwidth for highly efficient CBC. The result is a single, collinearly summed beam with an intensity noise power spectrum essentially indistinguishable from that of either one of its inputs, strongly suggestive of further power scalability.

We then consider overall system scalability via the cascaded combination of a split-seed RFA triplet, using an effectively nested pair of bulk-optic Mach-Zehnder-type interferometers configured with active, reference-beam-free, all-in-fiber phase control. To this end, we develop and incorporate a third, state-of-the-art, 40W-class RFA based on the same technology, and demonstrate (at essentially all power levels up to a maximum combined and optically isolated P₁₁₇₈ ~60W) that the cascaded combination efficiencies of the sub-1.5MHz linewidth RFA triplet are essentially indistinguishable from those of the seed-twinned pair (η_CBC ~0.95). To the best of our knowledge, this result represents the first cascaded collinear CBC of narrow-band RFA.

The combined 1178nm beam was then frequency doubled using an intra-cavity LiB₃O₅ crystal. No differences were seen in either the behavior of the SHG cavity or its frequency doubled output, using this 3-way combined fundamental as compared to previous results using a more basic constructively interfered RFA pair (or indeed, single fundamental). In both combined and cascaded combined cases and for a system optimized for P₅₈₉ ~25W, peak conversion efficiencies of η₅₈₉ ≥ 0.86 were comfortably achieved. We consider that this result reinforces the notion of potential further power scalability using similar cascaded-combined schemes, whilst essentially maintaining the optical properties of a single system. Furthermore, we demonstrate low intensity noise in the generated visible beam, and by extension of the above results, expect this beam to be equally well suited for higher-order harmonic generation (or sum-frequency-mixing with another beam) at power levels of multiple tens of Watts.

The SHG cavity was then re-optimised for operation at higher power, and P₅₈₉ > 50W was obtained (with corresponding peak conversion efficiency η₅₈₉ ≥ 0.85) from the cascaded-combined fundamental operating at full power. To the best of our knowledge, this represents the highest reported CW visible power achieved under direct SHG of any fiber-based system. The temporal stability of the system was measured over a 500s period and at P₅₈₉ > 40W, indicating only minimal rms noise levels and no signal degradation. Based on results measured in the case of the combined RFA pair, we expect the high-power visible output to remain quasi-diffraction-limited and with MHz-class linewidth or less. We also expect the system to become easily field-hardened to satisfy our original goal of LGS-AO applications, for which it clearly and comfortably satisfies all basic requirements.

Finally, we consider this state-of-the-art cascaded coherently collinearly combined and frequency doubled RFA triplet to be one of the first highly-wavelength-versatile, flexible and scalable systems now providing access to power-stable 50W-class CW spectroscopic-quality sources, with emission wavelengths essentially limited only by seed and pump pair availability. We expect such a technology to find applications not only in LGS-AO, but also for other laser projection and display systems, as well as in the fields of microscopy, spectroscopy, atomic cooling and trapping, coherent lidar and medical amongst others.