Building the Evryscope: Hardware Design and Performance

With dozens of lenses and CCD arrays available from a multitude of manufacturers, we performed a comprehensive trade study of the possible lens/CCD combinations. The pixel scale was set by the anti-crowding science requirement to be smaller than 20'', and we set the FoV to 8000 sq. deg. With those parameters fixed, we evaluated each lens/CCD combination based on the signal-to-noise ratio (S/N) that could be achieved all-sky on a m_g' = 16 source. The S/N calculations included the likely PSFs and vignetting generated by each lens/CCD combination, the expected sky background and source photon noise contributions, the detector characteristics, and many other factors, and most lens/CCD combinations were not able to achieve the required S/N because of one of those factors.

We elected to limit our CCD selections to interline-transfer chips which have electronic shutters. Our prototype systems (Law et al. 2013) both suffered mechanical shutter failures during their arctic deployments, with the achieved number of error-free exposures being just over one-tenth of the specification. Although the failures were correctable by individually adjusting the tension of internal springs every few months, this is untenable in a robotic system with dozens of cameras. The use of electronic shutters effectively eliminates this failure mode.

The trade study resulted in a single workable choice for lens/CCD combination: a Rokinon 85mm F/1.4 lens combined with a KAI29050 CCD array. All other combinations resulted in unacceptably low S/N or budgets factors-of-several times larger than our target amount. The KAI29050 array had a particular advantage in its rectangular format: most photographic lenses have rapid fall-offs in PSF quality towards the edges of the frame, and square arrays can therefore have poor image quality in the corners (Law et al. 2013). Compared with a square format, a rectangular array trades off highly off-axis image area at the corners for less off-axis area at the left and right edges of the array, and thus has more uniform PSFs across the image than a square CCD with equivalent area. On the basis of our positive experience with previous similar cameras, we elected to use thermoelectrically cooled Finger Lakes Instrumentation ML29050 units.

Next, we built a metric to optimize the camera positions in the array. Each camera produces a rectangular field on the sky, with a large enough FoV that spherical geometry must be taken into account for even simple sky-area calculations. We designed the camera array positions to (1) optimally tile over the above-airmass-two FoV and (2) avoid large areas of overlap between cameras; and (3) retain a few-degree overlap between each camera to constrain systematics.

We designed a code to project the FoV of each camera onto the sky, taking spherical geometry into account. The code then divides the sky into patches approximately 0.3° across, counts the number of cameras pointed at each patch, and measures the total sky area and overlap areas covered between different combinations of cameras. Starting with a simple arrangement of cameras divided into rows of declination, we then varied the position of each camera in the array using an annealed downhill-simplex algorithm, optimizing for overlap and covered sky area (Law et al. 2016). The optimization converged on an arrangement very similar to the input declination-separated grid of cameras; other camera arrangements we explored did not produce significantly better performance metrics. For ease of fabrication we used the simple declination-separated grid to place the cameras, with spacing parameters inherited from the fully optimized solution (Figure 3).

Zoom In Zoom Out Reset image size

Each camera assembly rotates in a circular arc around the pole-facing camera as the mushroom tracks the sky. Over the course of a typical night the system covers ≈18,000 sq. deg. (Figure 4), with each part of the sky being observed at 2-minute cadence for 4–10 hours per night.

Zoom In Zoom Out Reset image size

Each CCD is orientated so that its long axis (designated as the x-axis) is tangential to this arc; this ensures the objects in each image remain in a constant orientation throughout the night. There are seven rows, with the cameras in each row sharing the same pointing declination, equidistant from the pole camera. The camera mounting flanges (and therefore the CCDs) are normal to the surface of the mushroom dome, which ensures that the cameras are pointed in the proper direction without manual alignment being necessary. We designed the mushroom to be capable of supporting 27 telescopes; at CTIO 24 are Southern hemisphere facing and three cover positive declinations. The number of operational cameras has varied slightly during the course of the project: 22 or 23 cameras have been operational in 2015–2017, with another camera reserved for testing. We plan to fill in all available slots in the near future.

The camera hardware units fix the cameras to the mushroom, provide mechanical support of the components, and a mount for a protective window. The camera mounts have three primary constraints on their design: flexure limits, size, and weight. Although atmospheric refraction precludes keeping each star on the same pixel while tracking (Law et al. 2015), we designed the camera mounts to not contribute any extra drift throughout the Evryscope's range of motion, requiring the relative camera mount flexure to be less than 13''. The size of the mushroom was set to a 6-ft diameter by our target dome, and this set the packing requirements for the cameras. As there are two dozen camera mounts with relatively heavy CCD units, they and the systems they contain are the primary drivers of the weight of the system. A trade study of available mounts suggested that significant cost savings were possible if the total mushroom weight could be kept below 400 lbs.

We used 3D modeling to test several hardware unit designs, with the goal to minimize weight, flexure, and complexity. The final version (Figure 5) features interlocking sections for added rigidity, weighs less than 4 lbs (supporting imaging hardware that weighs 8.0 lbs), and provides a maximum differential flexure of less than 10 arcsec. The maximum flexure in the vertical orientation is ≈.02 mm and over the course of a telescope ratchet the differential movement due to the changing camera orientation is well within our 1 pixel goal. The camera mounts are interchangeable, have locator pins to easily place the cameras into the proper orientation in the mushroom, and perform equally well in flexure for all cameras regardless of the declination row (which have considerably different gravitational vectors).

Zoom In Zoom Out Reset image size

Each mount has an outer window to protect the lenses and electronics from dust, water, and other possible contaminants, enabling easy cleaning as well a providing a backup to the observatory dome. The high transmission (over 96% in the visible range) optical window is mounted on a soft o-ring with a stainless steel retaining ring, and allows for easy cleaning of dust during maintenance.

Interline-transfer CCDs cannot take darks without extra mechanical shutters, so we elected to use a filter wheel with a blocked position to allow calibrations to be taken. The Finger Lakes Instrumentation CFW-5-1 filter wheels also provide a sunshield (Section 2.7.4) and science filter changing capability.

Camera lenses are used on SuperWASP (Pollacco et al. 2006), HAT (Bakos et al. 2004), KELT (Pepper et al. 2007), XO (McCullough et al. 2005), and other transiting exoplanet surveys to reach as much as 1000 sq. deg. fields of view. Other surveys types such as the ASAS-SN (supernova) (Shappee et al. 2014), Pi of the Ski (gamma-ray bursts) (Piotrowski et al. 2013), Fly's Eye (asteroid detection) (Csépány et al. 2013), and HATPI³ also use camera lenses to reach wide sky coverage. These types of wide-field surveys and many others including the Evryscope are susceptible to image quality tilt and focus challenges. Even a slight misalignment between the optics and the CCD causes a tilt that results in an unacceptable increase in size of the PSF FWHM toward the edges and corners of the image. For the Evryscope, the very wide FoV (380 sq. deg.), fast F# of each lens and the small 5.5 μm pixels exaggerate this effect. While the machining tolerances (+/−.005 inch in most cases) and the assembly tolerances of the mass-produced lenses, adapters, filter wheels, and CCD assemblies is reasonable for their standard usages, it is not precise enough to achieve the absence of tilt required for the needed Evryscope image quality.

We designed a robotic tilt adjustment mechanism (Figure 6) to address those challenges, with the ability to remotely and precisely realign the camera assemblies. The Robotilter (Figure 6) uses three precision servos controllable to within 4 deg. steps coupled to an 80 thread per inch adjuster to move the lens position relative to the CCD. This allows adjustment of the tilt as well as the lens/CCD separation in increments as fine as .003 inch. The design uses specialized flexible shaft couplings to prevent binding and tension springs to hold the lens accurately in place. The assembly mounts to the top plate of the filter wheel to avoid costly re-configuring of the existing filter wheel, CCD, or camera mount. A separate servo independently adjusts the lens focus position to compensate for tolerance differences due to temperature changes throughout the year. The Robotilters were installed in 2015 November and the cameras were aligned remotely in early 2016; the installation of the Robotilters was the final step in commissioning the system. The Robotilters and resulting image improvements will be described in detail in an upcoming technical paper (J. K. Ratzloff et al. 2019, in preparation).

Zoom In Zoom Out Reset image size

The camera support structure (the mushroom, Figure 1) needs to provide the same limited flexure as the camera mounts, while also bearing the 400 lbs load of up to 27 camera assemblies and related components. We chose a molded fiberglass hemisphere with support ribs along the bottom and back for extra strength and rigidity, and a sturdy mounting point. The material is hand-laid cloth weave fiberglass, providing light weight and minimal flexure with excellent durability. The mushroom also features reinforced and precision-located inner and outer camera mount flanges to provide accurate and secure mounting points. The camera flanges are normal to the surface, and the holes are CNC cut into the mushroom to ensure the precise location necessary to achieve the desired field coverage without holes or excessive overlap. The manufacturing tolerances are .020 inch on the hole locations, and based upon this the camera alignment is fixed normal to the mushroom surface and the long side CCD is perpendicular to the rotation axis. Our 3D model simulation predicts that despite the close packing of the cameras and considerable weight, the stress is mostly compression and results in absolute movements on the scale of .02 mm. Differential camera movements over the tracking cycle are on the order of microns ensuring accurate camera pointing. On-sky pointing accuracy is well within the simulated performance.

The hemispherical shape of the mushroom, along with the placement of the instrument so that the dome leafs in the open position are slightly higher than the mushroom base, help make the Evryscope resilient to wind shake. The system is able to operate in 30 mph winds without a measurable change in image quality.

The base structure (Figure 7) attaches the mushroom to the Mathis 750 mount, via a mount plate attached to the tracking mount and a structure which transfers the mechanical load from the mushroom fiberglass. We tested several design ideas via finite element analysis and found a reinforced round tubing design to be most effective. Using aluminum tubing, we reduced the weight in half from a similar design made of steel and kept the total flexure within requirements. The differential camera displacement of the mounting base throughout the telescope tracking is on the order of microns, and combined with the mushroom and camera mount flexure is simulated to be within our total goal of 1 pixel, with comparable performance measured on-sky.

Zoom In Zoom Out Reset image size

The proper location of the center of mass is critical to reliable telescope mount operation. We inset the mount plate significantly into the mushroom so that the effective lever arm of the Evryscope cameras is minimized (Figure 7). The center of mass is only 10 inches from the mount plate, which greatly reduces the load on the telescope mount compared to simpler designs. The base structure positions the Evryscope so that the center of mass in the mounted position is directly over the telescope mount axis center, further reducing stress on the telescope mount and easing the balancing of the instrument.

The polar alignment of the mount is critical to the tracking performance of the system. Because the system's FoV is such a large fraction of the sky, conventional pointing models cannot be used, because they optimize the performance on one part of the sky by reducing performance on other parts of the sky. For this reason, we developed a precision polar alignment procedure specifically for Evryscope-like instruments (Appendix).

On-sky performance confirms the predictions of the flexure and center of mass simulations. The camera pointing is accurate within a tenth of a degree, providing the proper FoV overlaps without gaps (except for one initial, now corrected, misalignment caused by a contaminated bolt thread). The camera orientations remain constant throughout sky tracking. The telescope mount tracks the sky consistently without stalling or shifting, and we conclude that the total flexure is very close to the 1 pixel goal.

The Evryscope is located in an AstroHaven clamshell dome originally built for the PROMPT network of telescopes (Reichart et al. 2005). The dome had already been used for routine long-term operation, and no mechanical changes beyond a custom pier structure were necessary for the Evryscope deployment. Careful electrical design was necessary, however; the large dome opening/closing motors can induce strong transients onto power and potentially signal lines from the dome. To avoid possible interference or even damage, we separated the dome electrical systems on a separate UPS system. A Raspberry-Pi single-board computer runs the dome-control daemon and communicates with the rest of the system via an electrically isolated Ethernet connection; there are no other direct electrical links between the Evryscope and the dome.

The Evryscope is deployed at CTIO in Chile in PROMPT (Reichart et al. 2005) dome 4 (Figure 8). The site was chosen for the large number of usable nights (>320 per year), dark-sky conditions (m_v = 21.8 moonless night background average), and Southern sky visibility. UNC affiliated hardware and support synergies, especially the PROMPT Program, were also advantageous.

Zoom In Zoom Out Reset image size

The dome and observatory site introduced several design constraints: (1) a maximum power consumption of 15A/120V; (2) operation with a relatively small Internet bandwidth that precludes the real-time, off-site transport of data; (3) the potential for lightning strikes and earthquakes (Section 2.7); (4) potential external temperature ranges of −15°C to +25°C; and (5) extremely dry conditions.

The low end of the temperature range is outside that which most off-the-shelf electronics are rated for. Wherever possible, we purchased industrial components rated for low-temperature operation (typically −20°C). In some cases, we tested and used off-the-shelf consumer electronics (for example, Raspberry-Pi single-board computers); testing was performed in fridge-freezer units under a range of relative humidity (see Law et al. 2013; Law et al. 2016 for testing details).

The potential for extremely dry weather spells required careful electronic and mechanical design. For example, Nylon becomes brittle under extremely dry conditions (Pai et al. 1989); this can cause failures in cable insulation and zip-tie-type harnesses in a matter of months, leading to possible short circuits or mechanical interference between cables and moving parts. The static electricity discharges prevalent in dry conditions can cause electronic failures, especially while personnel are maintaining the system. Many power supplies and similar units are rated only to 20% relative humidity, while the CTIO site can regularly reach low-single-digit humidity. We mitigated these concerns by using only plastics, connectors, and electronics rated for long-term extremely dry conditions. All metal components are grounded, with isolators used to avoid ground loop conditions, and we take operational steps to ground personnel before working on the system.

The Evryscope cameras together require a maximum of ≈170A of 12V power; the ancillary systems with the mushroom (Robotilters, filter wheels, fans, USB hubs, etc.) together require a further ≈20A of 12V power. The AWG-1 (quarter-inch diameter) cables required to safely carry the required 200A into the mushroom would be bulky and inflexible, and risky if frayed or overheated. Powering each camera from its own 12V supply would lead to a very bulky and heavy power distribution system, beyond the load capacity of the mushroom mount. For those reasons we elected to send 120V AC power into the mushroom over a single flexible small-diameter cable, and use two 120A-capable 12V power supplies to power the main camera systems. We deliberately overspecified the power supplies to reduce the need for active cooling and the associated vibrations. Ancillary systems are powered from their own smaller 12V power supplies, with Digital Loggers Network Power Switches allowing computer-controlled switching of each component. Although it has proven reliable, this setup resulted in over 600 ft of cabling inside the main mushroom, because each camera has six separate cables going into it (3 power, 3 data). These cables are heavy and impede airflow; the Northern Evryscope, currently under commissioning, has relay and control systems built into each camera to reduce the number of required cables to two per camera.

The two 120V input/12V 80A output power supplies are mounted on panels attached to the wings of the base inside the mushroom (Figure 10) . Fused distribution blocks with custom cabling connects the power to the cameras. The filter wheels use a similar, but smaller, 120V input/12V 8A output power and distribution located on the same panels. An additional 120V input/12V 8A output power supply is also available on each panel to supply the focus servos, cooling fans, and other accessories. A panel attached to the center of the base over the mount (Figure 10) holds a Network Power Supply (NPS) and a power supply for the USB hubs used to control the cameras assemblies. The selection and placement of the power systems allows for proper balancing of the mushroom assembly, cooling of the electronics, access to all of the components, and provides a safe supply of power to many different systems confined in a small area.

Zoom In Zoom Out Reset image size

The Evryscope uses up to 1.2kW when all cameras are cooling at maximum power, producing a significant amount of heat within a 6-ft semi-enclosed space. In-lab tests showed that parasitic heating between cameras could lead to a thermal runaway under some environmental conditions: cameras pulling in warm air exhausted by the thermoelectric coolers of neighboring cameras must work harder to cool their sensors, increasing the amount of waste heat exhausted, and causing other cameras to further increase their cooling power. This process headed for runaway when the air temperature inside the mushroom exceeded ≈32°C. Although several layers of protection prevent hardware damage from overheating (Section 2.6) this could have impacted system uptime during summers.

We implemented three systems to eliminate the parasitic heating. First, we built aluminum deflectors to move the camera exhaust air toward the center of the mushroom. Second, we added a bank of 8 120 mm low-vibration 12V fans to direct cool air to the top of the mushroom. Third, we added external Vornado high-volume industrial fans to direct large amounts of external cool air to the mushroom (when rarely necessary). Together, these systems produce a coherent flow of cool air from the front-bottom of the mushroom to the top of the dome and down again out of the back of the systems. Testing showed no measurable effect on image quality when all systems are activated. The thermal protection systems have not triggered a shutdown since this system was commissioned.

We monitor the hardware status with sensors distributed around the mushroom and dome, all linked to the main control system via Ethernet or USB connections. The main control computer runs automated analysis and control scripts, and alters the state of fans as necessary to maintain stable temperatures around the cameras. Logs of all sensor values are recorded each minute.

Inside the mushroom, each camera has an external temperature sensor, measuring the air temperatures at 22 points around the dome. An environment-monitoring Raspberry-Pi is located at the center of the mushroom. Its custom-built sensor board monitors the overall mushroom temperature with a wide-angle infrared thermometer, the center-mushroom temperature with a built-in sensor, and the tilt of the mushroom using a precision three-axis accelerometer. A timing GPS system is also connected at that location. A summary of all sensors is shown in Table 3.

Outside the mushroom, two webcams continuously monitor the system from the north and the south. The northern webcam is a pan/tilt unit; the southern webcam is a Raspberry-Pi camera which, in addition to providing a view of the mushroom internals, automatically monitors the light level in the dome. If the light level is consistent with the dome being unexpectedly open in daytime, a loud alarm bell is sounded and the Evryscope team is alerted via email.

We use the PROMPT weather monitoring system (Reichart et al. 2005) for dome open/close decisions; this system has been in reliable operation for almost a decade. The PROMPT weather station monitors cloud levels, wind, and dewpoint. We use the RASICAM (Lewis & Howard Rogers 2010) system to log cloud measurements for data-quality testing.

The main control computer, watchdog, and environment-monitoring computers and data storage and analysis servers are located within the telescope dome, with optical fiber connections to a backup storage site in an adjacent PROMPT dome. The Evryscope data and control bus is a gigabit Ethernet system operating as a separate subnet behind a router connected to the main CTIO network.

A single sealed and fanless Logic Supply ML600G-30 rugged computer runs the robotic control software (Section 2.6) and the USB-controlled devices, including the cameras, filter wheels, Robotilters, and the mount.

Over 50 individual USB devices are connected to the control computer, which produces challenges to reliable system operations (Ethernet control was not available for our chosen cameras at the time of system design). We initially connected groups of 4–8 USB devices together using powered USB hubs. However, lab testing showed occasional USB-bus-voltage brownouts, where the 5V power supply in a typical computer could be pulled out of voltage specification just by connecting dozens of USB devices, even when the devices were powered off and connected via powered hubs. This could prevent the control computer starting up or cause unreliable operation, and occurred for all tested brands of USB hubs. We eliminated this problem by finding and removing an undocumented jumper inside Starlink ST7200USBM rugged USB hubs which completely disconnects the upstream USB power rails from the downstream devices; this produces reliable operation with at least 60 USB devices connected.

The Evryscope uses up to 1.2kW of power when all systems are simultaneously operating, within a fairly small enclosure. Two 120A/12V power supplies supply power to the camera systems, and a short-circuit on a 120A-capable line could easily produce enough heat to ignite surrounding material. We mitigated these concerns by (a) breaking apart the high-current lines very close to the power supplies for individual camera power; (b) individually fusing each power supply line; (c) powering the system via GFCI breakers to produce a rapid shutdown in the event of a ground fault; (d) specifying all plastics to be flame retardant; (e) wrapping all exposed cables in flame-retardant material; (f) placing an omni-directional infrared temperature sensor in the dome which shuts the power down on detection of an overheat condition; and (g) placing a Raspberry-Pi connected smoke detector in the dome to rapidly shut off power and sound an alarm if smoke is detected.

Electrical storms are rare at CTIO, but the Evryscope has so far experienced one extremely nearby lightning strike that damaged equipment in nearby domes. To mitigate the possible lightning impact, we applied surge protectors to every power line and isolators to every USB and Ethernet cable longer than three feet; this also mitigates the effects of possible ground loops. No lightning damage has been experienced by the system.

Chile regularly experiences large earthquakes, and telescope systems must be designed to survive large ground accelerations. As with the other main instrument components (Section 2.4), we evaluated the Evryscope pier mount design using 3D modeling finite element analysis. We simulated the telescope weight on the pier design over several angles to mimic positions during the ratchet cycle. The final pier design is 1/2'' wall structural grade steel box tubing, with a strength failure several orders of magnitude above any level the Evryscope is likely to see. An accelerometer inside the mushroom measures the tilt of the mushroom and any other accelerations, and places the system in safe mode if limits are exceeded. On 2015 September 16, CTIO was hit by a 8.3 magnitude earthquake at a distance of 115 miles. The Evryscope automatically went into safe mode; no structural damage occurred and after quick manual checks, the system was able to restart with no maintenance required.

With a telescope pointing at almost the entire sky, if the dome is opened during the day at least one camera would be pointing directly at the Sun. The resulting heat buildup in the Sun-pointing region of the CCD chip would be likely sufficient to cause significant CCD damage. If the dome was left open for an entire day, during maintenance or as a result of equipment failure, it is possible that an entire row of cameras could be damaged or destroyed. We addressed this with 1) a daylight alarm which sounds a loud bell and contacts the Evryscope team; 2) sunshields built into each camera.

The sunshields are contained within the cameras' filter wheels and consist of a 3mm-thick steel washer backed by a mirror; sunlight entering the lens will be very out of focus at the filter position, preventing the formation of hotspots. Experiment at Chapel Hill showed no dangerous heating of the lens over hours of Sun exposure. The sunshields are a primary safety system and as such are engaged immediately upon error conditions; each morning the system engages the sunshields as part of the shutdown procedure (apart from fans, the sunshields are the only moving parts inside the mushroom that are used nightly).

The Evryscope generates approximately 6500 55 MB science images each night. This data volume precludes transmitting the data for off-site processing with the current CTIO Internet link. All data is therefore stored and processed on-site. Images are stored in an FPACK-compressed format across multiple Synology DS-2415+ network storage appliances, each of which is equipped with twelve 8 or 12 TB drives. In addition to image storage, we have provisioned a separate data store exclusively for our photometry database, consisting of 12 helium-filled 8 TB drives directly attached via a SAS backplane to our database server.

Data processing is split between two servers, both housed in the PROMPT domes at CTIO. The original server, a 12-core Intel Xeon-based machine, was installed with the system. Post deployment, the mainboard of this server suffered some mechanical damage, limiting its RAM capacity to 112 GiB. In January of 2016, a second server was installed and the original was reprovisioned to support a calibrations and image indexing database, while all other analysis tasks were migrated. The second server is also based on the Intel Xeon platform, with 36 physical cores and 256 GiB of RAM.

The Evryscope currently runs a forced-aperture-photometry pipeline. The pipeline takes incoming images, calibrates them with darks and flats, generates a precision astrometric solution from the bright stars; estimates local background light and noise across each image; and measures aperture photometry for all sources from a reference catalog. The Evryscope pipeline consists of ∼50,000 lines of custom Python and C++ code, with custom code performing flat-fielding; astrometric distortion correction; local background and noise estimation; precision aperture photometry; transient detection; and large-volume data storage. We expect to upgrade the pipeline to full image subtraction in the future.

We extensively tested standard data analysis software with Evryscope images (for example, the SExtractor (Bertin & Arnouts 1996) and astrometry.net (Lang et al. 2010) software suite used in PTF (Law et al. 2009) and the AWCams (Law et al. 2013)). However, we found that the standard software struggles with our crowded images with large lens distortions: astrometry.net had a >20% probability of failing to find a good astrometric solution at the edges of the frames, often producing distortion solutions several pixels off. SExtractor often could not attain a good background noise estimate for our crowded images, and therefore set the source-detection requirements extremely high; often several-degree-wide regions of the Evryscope images did not show any detections despite tens of thousands of stars being clearly visible by eye. A few percent of the Evryscope images also showed SExtractor photometry very divergent from adjacent images, with stars' brightness measurements changing by tens of percent with no discernible by-eye difference in the input images; these problems persisted regardless of the input settings. For these reasons we developed a completely custom pipeline, although we do use astrometry.net for initial rough astrometric solution and SExtractor for quick source-detection for camera focusing; both codes work very well for those applications.

Each processed night consists of ≈360GB of raw imaging data, resulting in several hundred new data points for each of ≈10M stars. On our current computing hardware, the pipeline is capable of processing ∼7 nights (2.5 TB of imaging data) every 24 hours. This speed is necessary to allow us to re-reduce our current three-year data set in a reasonable time.

Each Evryscope science image is subjected to an initial quality control script which evaluates the image quality based on the presence of stars in the image, PSF shape (avoiding rare tracking errors), and background levels. Images that pass (>90%) are masked for known bad pixels and columns.

Darks are taken daily with the filter wheel in the closed position, and monthly midnight darks are taken for comparison to check for light leaks. Masterdarks are generated by combining and median averaging several hundred darks. Our CCD characteristics are sufficiently stable to use the masterdarks for a season.

Twilight and sunrise flats are taken daily and evaluated with a quality control check for stars and clouds. Residual point sources are removed. Lens vignetting and small-scale interpixel variation in CCD sensitivity are removable to the 1% level with standard flattening procedures, however the large-scale sky gradient due to the extremely wide FoV necessitates a more complex procedure. We constrain the large-scale variations on using on-sky photometric measurements of starfields, and measure the small-scale variations from the high-frequency structure in twilight flats.

Our current data set includes 9.3M stars with an average of 32,600 photometric measurement points. The photometric points are stored in a flat-file-based custom backend storage system written in Python. The system is partitioned by sky position using HEALPix pixels (Gorski et al. 2005). HEALPix pixels divide the sky into equal area regions; we selected a 3.5 sq. deg. HEALPix pixel size for convenience to limit the number of stars in a particular region. This aids in processing of the light curves (done per HEALPix pixel) and allows for multi-threading and tiling the database writing steps. We evaluated database management systems (DBMS), but found that for our extremely consistent-format numerical data our custom system could reduce storage requirements by a factor of five compared to PostgreSQL while increasing access speed by a factor of 10. We also evaluated similar commercial and open-source flat-file numerical data storage systems and found that the performance was generally comparable to our flat-file-based system, but with significantly higher implementation complexity and programming overhead. The flat-field storage system stores approximately 15 TB/yr of light-curve data.

Each star's photometry is measured in five different photometric apertures, allowing an optimization of the S/N for each star (for example, selecting larger apertures for brighter stars; this technique is used by several surveys, e.g., Pollacco et al. 2006). Each measured data point also includes the star's measured R.A. and decl., CCD position, estimated S/N, limiting magnitude at that point, background light level, peak flux level, and a GPS-based precision timing signal with tested 1 s accuracy (Corcoran et al. 2018).

We periodically generate precision light curves for each star based on the typically tens of thousands of photometric points recorded for each star. The light-curve generation code processes each HEALPix pixel separately, performing differential photometry on the contained group of several thousand stars. Atmospheric extinction variations from clouds and airmass are corrected for using differential photometry among the thousands of stars in each HEALPix pixel. First, images pass through an image quality check which rejects images with high background, low numbers of detectable sources, or suspect PSF shapes. Next, the least-variable stars are automatically selected to form a consistent set of reference stars (this procedure is iterated with the differential photometry to find the stars most indicative of the overall photometric variations). For each single-camera image accepted by the pipeline for processing, which typically have a few 100,000 stars, each source is checked for possible blending, local background issues, non-detection and saturation. Flags are issued for suspect data points. Flux errors are estimated based on the local background noise for all epochs, for all sources. Airmass and differential chromaticity errors are removed by SysREM (Tamuz et al. 2005) in the default pipeline operation; we tested removing explicit correlations with star color and measured airmass, but did not find a significant improvement in photometric precision. These procedures work for the large majority of the data set, but a small fraction (<20%) of the epochs are subject to largely unremovable variability due to thin clouds with spatial scales smaller than a HEALPix pixel. We detect and remove these epochs by searching for periods of higher than average photometric variability among all sources in the healpix, as well as higher than average extinction. We are currently developing methods to instead flag and recover these epochs for usable data.

We have implemented several layers of systematics removal, which can be applied depending on the science goals. All light curves are automatically decorrelated by two iterations of SysREM (Tamuz et al. 2005). Further iterations of SysREM further remove systematic errors, but there is also a risk of removing astrophysical variability. If only short-term variability is to be measured, such as in a transit or eclipse search, we add decorrellations of photometric variability with CCD chip position and airmass. We found that some long-term variables such as low-amplitude long-period rotation curves correlate with those telescope variables, and so we offer users the option of using uncleaned light curves.

Processed light data is inserted to a PostgreSQL database, also partitioned into HEALPix pixels to increase performance. This database does not include much of the per-epoch metadata, and only contains results from the optimal photometric aperture. Each of the 6000 populated HEALPix pixels contains 0.2–2 GB of light-curve data, for a total light-curve database size of ∼10TB. We query the database for target groups, and download the results to Chapel Hill for astrophysical analysis.

The Robotilter camera/CCD automated alignment system is designed to remove tilt, minimize PSF distortions, optimize the focal plane, and defocus the image center. The PSF FWHM map of a well aligned, representative camera is shown Figure 12). Very little tilt across the image is evident, and PSF widening toward the corners due to lens coma, focus, and vignetting is within the expected range for our lenses. The PSFs range from 1–5 pixel FWHM across much of the image: 60% are less than 4 pixels, and 90% are less than 6 pixels. Figure 13 shows point-spread functions for the central region and edges of a representative camera.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We calculate the limiting magnitude achieved by the system in each epoch by taking the faintest stars in each healpix and fitting the S/N decrease as a function of the g-band magnitude as measured by APASS. The dark-sky limiting magnitude (Figure 14) reaches our expectation of ${m}_{g^{\prime} }\approx 16$, with crowding from the galaxy reducing the limiting magnitude by approximately a magnitude in low-galactic-latitude areas. A horizontal stripe pattern is visible in the limiting-magnitude map; this is caused by the falloff in PSF quality toward the edge of camera fields of view.

Zoom In Zoom Out Reset image size

The camera gains, data compression and calibration fidelities are selected so that coadding the data achieves greatly improved S/N, with depth increasing with approximately the square root of the number of exposures (Figure 15). In uncrowded regions of the sky during dark nights, the system typically achieves ${m}_{g^{\prime} }=17$ in 8 minutes coadding (four exposures), ${m}_{g^{\prime} }=17.5$ in 32 minutes, ${m}_{g^{\prime} }=17.8$ in 64 minutes, and ${m}_{g^{\prime} }=18.5$ in 360 minutes (the latter crowding-limited over much of the sky; see Figure 16).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Light-curve performance reaches our expected performance levels of near 1% rms on bright stars and ∼10% on dim stars, over three years of data under all moon and cloud conditions (Figure 17). With binning and/or aggressive removal of poor conditions data and systematics, the performance is improved to the 6-millimag level. These levels are greatly improved when coadding epochs for the detection of periodic objects, where we have published clear signals at the few-millimag level (Tokovinin et al. 2018).

Zoom In Zoom Out Reset image size

A test search limited to the northern region (declinations from +5 to +10), filtering the targets by magnitude (bright stars) and color (likely K-dwarfs or M-dwarfs) yielded 59 new eclipsing binaries and variables. Representative examples of an eclipsing binary and a low-amplitude variable are shown in Figure 19. The search was run by selecting all of the sources in the Evryscope database with light curves with greater than 5000 epochs, with magnitudes brighter than 14.5, and with sources that matched to PPMXL (Roeser et al. 2010) and APASS-DR9 (Henden et al. 2015) catalogs which could be classified as potential K-dwarf or M-dwarfs based on reduced proper motion (RPM) and B-V colors. After removing known variables, BLS and LS were run on the filtered list; the example eclipsing binary and low-amplitude variable BLS and LS detections are shown in Figure 20. The BLS and LS results were ordered by significance and the top 10% were inspected using the detection panel plots. Those passing the visual inspection and systematics test were sent to the next stage. Eclipsing Binaries were fit with a Gaussian to measure the eclipse depth using the detected period and phase as the prior. Variables were fit using Lomb-Scargle to determine the amplitude. Example eclipsing binary and low-amplitude variable fits are shown in Figure 21. Tables 4–6 in the Appendix contain the full discovery list; Figures 22 and 23 display the light curves.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

One major Evryscope transit survey has been completed and two are underway, with several others in the planning stages. A transit search for variable stars in the southern polar region led to 300 variable and eclipsing binary discoveries, with six of the eclipsing binaries having low-mass secondaries (Ratzloff et al. 2019). An exoplanet survey of ≈2500 southern sky white dwarf (WD) targets m_v < 15.0 is underway. A transit survey of ≈3500 hot subdwarf (HSD) targets is in progress and has already discovered several rare systems: 2 HSD/low-mass-secondary eclipsing binaries (HW Vir systems), 4 HSD reflection binaries, and 2 HSD/WD short-period binaries (J. K. Ratzloff et al. 2019, in preparation). From these surveys, there have been five planet candidate detections; subsequent follow-up showed these candidates to be grazing eclipsing binaries with almost identical stars or low-mass stellar companions. These detections demonstrate the Evryscope is capable of detecting planets orbiting post main-sequence stars as well as M and K-dwarfs with our current light curves and search algorithms. We have used the initial results of these first surveys to refine our transit searches; we briefly describe the status of the key Evryscope transit surveys below.

White dwarfs (WD). Recent discoveries of WD debris disks and disintegrating planetesimals have fueled the speculation that planets could be present in WD systems (Loeb & Maoz 2013; Veras et al. 2018). WD exoplanets would have very short (few minutes to tens of minutes) transit duration and very deep (∼100% for Earth-size planets) transit depths. WDs are extremely numerous in the sky as >90% of main-sequence stars will eventually become WDs, however the low luminosity and small size make these stars observationally challenging. We leverage the Evryscope fast cadence and all-sky coverage to search for WD planets. Our first results from ≈2500 southern sky WD targets m_v < 15.0 did not return any candidates. We have improved our systematics removal, increased our coverage to 3.5 years, and added targets down to m_v < 16.0 and will search again once the database processing is complete (J. K. Ratzloff et al. 2019, in preparation). In the event of a null detection, we can provide upper limit constraints on WD planetary populations.

Hot subdwarfs (HSD). HSD planet or low-mass-secondary transit durations are on the order of tens of minutes, and reasonably deep transit depths (∼10% for Neptune size planets). A transit survey of HSD planets and other variability from a target list (Geier et al. 2017) of ≈3500 known HSD is in progress (J. K. Ratzloff et al. 2019, in preparation). Although the survey is currently underway, several candidates, including the 8 mentioned above, have been identified and are pending further follow-up.

M- and K-dwarfs. The Evryscope is capable of detecting ∼2 Earth radii M-dwarf planets and gas giant K-dwarf planets. A transit search for variable stars in the southern polar region detected a 1.7 R_J planet candidate with a late K-dwarf primary. This system was later shown to be a grazing eclipsing binary, but demonstrated the Evryscope detection capability. An exoplanet survey of M and K-dwarf stars based on identifying candidates in our fields from spectral classification is planned for the entire sky when the HSD and WD surveys are completed.

Flares and coronal mass ejections (CMEs) are capable of severely affecting the survivability of potentially habitable worlds. A comprehensive flare survey of M-dwarf stars (including known exoplanet hosts) of the southern sky is underway (W. Howard et al. 2019, in preparation). These results, when combined with CME observations will be used to estimate the effects on long-term habitability of rocky planets orbiting M-dwarf stars.

We have developed tools for rapidly generating small cutouts from full-frame Evryscope images and performing high-precision photometry on uncataloged sources not included in our primary forced-photometry reduction, including difference image analysis for objects in crowded regions of the sky. This tool chain is designed to provide early prediscovery photometry to help constrain the evolution of novae and supernovae.

An example of Evryscope transient capability is a recent classical nova (Nova Carinae 2018) with prediscovery Evryscope coverage (Corbett et al. 2018), which is currently under analysis and a detailed light curve will be presented in an upcoming paper (H. Corbett, et al. 2019, in preparation). The Evryscope data complements the later discovery by the All Sky Automated Survey for SuperNovae (ASASSN; Stanek et al. 2018) and the serendipitous space-based photometry of the Bright Target Explorer (BRITE Kuschnig et al. 2018). High-cadence and high-coverage observations of classical novae can provide insight into the shock physics that drive light-curve evolution (Li et al. 2017). Also shown in Figure 24 is transient discovery from the variable star test search (Section 4.2.1).

Zoom In Zoom Out Reset image size