INTRODUCTION

Adaptive optics (AO) systems cancel wavefront distortions in real-time by driving wavefront correctors such as deformable mirrors (DMs) with wavefront sensors (WFSs), typically in closed-loop control. Such systems have found use in a wide range of imaging and telecommunications applications, includ-ing ground-to-space optical communications [1], deep tissue 10 microscopy [2], and remote sensing [3]. In astronomy, AO sys-11 tems correct for wavefront distortions which may originate from 12 the overhead turbulent atmosphere (in the case of ground-based 13 observation) or time-varying imperfections in the scientific in-14 strument. AO has enabled highly-resolved imaging of the galac-15 tic center [4], exoplanets [5], and more; [6] provides a review. 16 One of the primary goals for astronomy for the next decades 17 is the direct imaging of an Earth-like exoplanet and the iden-18 tification of potential biosignatures [7]. This achievement will 19 require the separation of exoplanetary light from that of a host 20 star which outshines its companion by 10 orders of magnitude 21 in visible wavelengths [8], for instance by using coronagraphy 22 [9]. However, coronagraphic contrast is highly sensitive to wave- the other [10]. Other sources of wavefront error include the 32 low-wind effect [11] and petaling [12], both of which arise from posed (e.g. DRWHO [13]), as well as dedicated sensor designs 40 like the self-coherent camera [14]. Recently, focal-plane sens- HARDWARE (SCExAO) SOFTWARE (CACAO) Detector image (cropped) Control matrix Leaky integrator Response matrix S V D Pixel # Mode # Mode # Pixel # Measure response to DM modes Image calibration 90:10 BS Artifical source fiber Dispersing optics (R = 300-700) ... Starlight, post primary AO (Subaru AO3K) b c DM correction photonic lantern cross-sectional views jacket cladding core to SMF outputs a As shown in Figure 1a, the PL is a slowly varying waveguide similar to a tapered multicore fiber which can efficiently couple 84 we measure the slope of the PL's response to the first 100 non-85 piston Zernike modes. This gives the response matrix, which 86 approximates the relation between the pupil phase and WFS 87 output; a singular value decomposition (SVD) of this matrix 88 determines our control modes. To compute the phase correction, 89 the phase retrieved by the PL WFS is fed into a leaky integra-90 tor. We then closed the AO loop (i.e. applied the correction 91 to the DM in real-time) at a framerate of 1.5 kHz. After run-92 ning a "self-test" of the loop, where a static amount of each 93 control mode is injected into the system one-by-one on a sepa-94 rate channel of the DM, we found that the first 52 control modes 95 -each a linear combination of Zernike modes -were clearly 96 corrected, with correction of the remainder too slow to be use-97 ful. Note that an un-dispersed 19-port PL senses strictly less 98 than 19 aberration modes [21], and that previous experiments 99 from [23] demonstrated control over only 15 modes. We provide 100 an explanation for this increase in §3. Next, to test the loop, 101 we generated phase screens with an RMS amplitude of 150 nm 102 and windspeed of 10 m/s, and applied them to an indepen-103 dent channel of the DM. We used high-pass-filtered Kolmogorov 104 phase screens to emulate first-stage correction of atmospheric 105 turbulence by Subaru's facility AO. Finally, we recorded the 106 amplitudes for each control mode, as measured through the 107 PL WFS, in both open-and closed-loop operation. The ratio of 108 the closed-loop and open-loop power spectral densities (PSDs) 109 estimates the squared modulus of the rejection transfer function. 110 We find that the correction loop is stable up to a limit of 150 -111 200 nm RMS of wavefront error, dependent on alignment and 112 of the PL, dispersing some at low spectral resolution for fast