A survey of spatially and temporally resolved radio frequency interference in the FM band at the Murchison Radio-astronomy Observatory
PPublications of the Astronomical Society of Australia (PASA)doi: 10.1017/pas.2020.xxx.
A survey of spatially and temporally resolved radiofrequency interference in the FM band at theMurchison Radio-astronomy Observatory
Tingay, S.J., Sokolowski, M., Wayth, R., & Ung, D.
International Centre for Radio Astronomy Research, Curtin University, Bentley, WA 6102, Australia
Abstract
We present the first survey of radio frequency interference (RFI) at the future site of the low frequencySquare Kilometre Array (SKA), the Murchison Radio-astronomy Observatory (MRO), that bothtemporally and spatially resolves the RFI. The survey is conducted in a 1 MHz frequency range withinthe FM band, designed to encompass the closest and strongest FM transmitters to the MRO (locatedin Geraldton, approximately 300 km distant). Conducted over approximately three days using thesecond iteration of the Engineering Development Array in an all-sky imaging mode, we find a range ofRFI signals. We are able to categorise the signals into: those received directly from the transmitters,from their horizon locations; reflections from aircraft (occupying approximately 13% of the observationduration); reflections from objects in Earth orbit; and reflections from meteor ionisation trails. In totalwe analyse 33,994 images at 7.92 s time resolution in both polarisations with angular resolution ofapproximately 3.5 ◦ , detecting approximately forty thousand RFI events. This detailed breakdown ofRFI in the MRO environment will enable future detailed analyses of the likely impacts of RFI on keyscience at low radio frequencies with the SKA. Keywords: astronomical instrumentation: radio telescopes – astronomical techniques: time domain astron-omy – radio frequency interference
The next generation of radio telescopes, built to explorethe evolution of the Universe to very high redshift, duringthe so-called Epoch of Reionisation (EoR; Furlanettoet al. (2006)), will be located at sites selected to have verylow levels of human-made Radio Frequency Interference(RFI).For example, the low frequency component of theSquare Kilometre Array (SKA ) will be built at theMurchison Radio-astronomy Observatory (MRO), lo-cated in the Mid West of Western Australia. The MROis also home to two SKA Precursor telescopes, CSIRO’sASKAP (Johnston et al., 2008) and the Murchison Wide-field Array (MWA) (Wayth et al., 2018; Tingay et al.,2013a). The MWA remains the only fully operationalPrecursor for the low frequency SKA and has been anoperational facility since 2013.With a frequency range of approximately 70 - 300MHz, the MWA spans a number of Earth and space-based broadcast bands, including the ubiquitous FM band (approximately 88 - 108 MHz in Australia), consti-tuting a primary source of RFI at the MRO. Likewise,the frequency range for SKA_low is 50 - 350 MHz, alsoencompassing the FM band.The effect of RFI on key science programs at low fre-quencies, including EoR experiments, is very significant.Aside from locating telescopes such as the MWA at thebest sites, such as the MRO, substantial general efforthas been put into the identification and removal of RFIfrom radio telescope data (RFI mitigation). An excellentassessment of the effects of RFI on EoR measurements,plus a review of the general considerations for RFI mit-igation, can be found in Wilensky et al. (2020); theyplace limits on the tolerable RFI budget before EoRmeasurements are adversely affected. However, the prac-tical realistion of an allowable RFI budget is complex.Wilensky et al. (2020) find that a detailed knowledgeof the spatial and temporal distribution of the RFI, theduration of observations, the characteristics of the tele-scope in question, and the detailed scheduling of thetelescope (where it is pointing and when) are required.Measurements that attempt to characterise the RFI en-1 a r X i v : . [ a s t r o - ph . I M ] A ug Tingay et al. vironment in which a telescope operates are thereforevery important in order to understand the overall perfor-mance of key science observations, such as for the EoRexperiment.Previously, a study of the RFI environment of theMRO, utilising the MWA over 10 nights of observationsin the frequency range 72 - 231 MHz, was conductedby Offringa et al. (2015). Using their methods of RFIdetection and excision, they found that 1.1% of datarequired excision, but that all frequency ranges remainedhighly usable, including in the FM and digital TV bands.More recently, Sokolowski et al. (2016) and Sokolowskiet al. (2017) examined RFI statistics for the MRO in thefrequency range 70 - 300 MHz using the BIGHORNSglobal EoR signal experiment (Sokolowski et al., 2015).They find examples of ducting events caused by atmo-spheric conditions that propagate signals from distanttransmitters to the MRO. A similar, highly compre-hensive survey of the LOFAR RFI environment wasconducted by Offringa et al. (2013).These previous studies of the RFI environment of theMRO at low frequencies have collected information as afunction of frequency and of time, but not as a functionof location on the celestial sphere. The previous studieshave obtained information that has been integrated spa-tially prior to analysis. In this paper, we examine theRFI characteristics of the MRO for one frequency withinthe FM band, resolved both temporally and spatially. Arange of RFI signals are expected at low radio frequen-cies, with the FM band one of the most prominent inthe SKA_low frequency range. At these frequencies, aswell as ducted propagation from distant transmitters,multi-path reception of RFI is expected due to reflec-tions off objects within the environment. For example,reflections off aircraft (Wilensky et al., 2020), ionisedmeteor trails (Zhang et al., 2018), and satellites (Prabuet al., 2020b; Zhang et al., 2018; Tingay et al., 2013b)are expected. Such signals are ubiquitous and have beenstudied in detail for other low frequency radio telescopessuch as the Owens Valley Long Wavelength Array (Mon-roe, 2018). These reflected signals are expected to belocalised in time and direction and contribute at differ-ent levels to the total RFI budget. We aim to assessand characterise these different contributions here, pre-senting the information required for a range of futurequantitative studies of the impact of RFI on key lowfrequency science for the SKA.We obtain temporally and spatially resolved RFI mea-surements by using the second realisation of the so-calledEngineering Development Array (EDA2: Wayth et al.2020, in preparation), an SKA_low station configurationcomposed of 256 individual MWA antennas on a 35 mstation footprint.In §2.1, we describe the EDA2 instrument, its charac-teristics, and the observations undertaken for this work.In §2.2, we describe the imaging and calibration of the EDA2 data undertaken to provide the raw images usedfor further analysis. In §2.3 we describe the data pro-cessing undertaken to separate the different categories ofRFI signals, and in §3 we describe the results, assessingthe components of the RFI budget in this frequencyband. Finally, in §4 we discuss the likely impacts ofthese components of the RFI budget for some areas ofkey low frequency science.
Observations were conducted using the Engineering De-velopment Array, version 2 (EDA2: Figure 1), an arrayof 256 MWA antennas arranged in an SKA_low stationconfiguration (station diameter of 35 m) and locatedat the MRO (Wayth et al. in preparation). Analoguesignal chains from the individual antennas (both polari-sations; X and Y) are digitised and coarse-channalisedin the firmware (Comoretto et al., 2017) implemented inTile Processing Modules (TPM; Naldi et al., 2017). Thecoarse-channelised voltage streams are received on thedata acquisition computer and signals from individualantennas are correlated using the x G P U (Clark et al.,2011) software correlator. Thus, the EDA2 forms a 256element, dual-polarisation interferometer that can beused to form all-sky images using standard interferomet-ric calibration and imaging techniques.
Figure 1.
An aerial view of the EDA2 instrument used for thiswork.
EDA2 data were collected between 2020-01-31 06:30and 2020-02-03 07:20 UTC at 1.98 s temporal resolution,as part of commissioning activities for the array. Duringthese observations, 54 antennas were non-functional forcommissioning, leaving a usable array of 202 antennas.Data were collected over a 0.926 MHz bandwidth (singlecoarse channel), at a central frequency of 98.4375 MHz.This particular coarse channel covered two FM stations
FI in the FM band at the MRO
H D F 5 format ,which is envisaged as the data format for the SKAtelescope. The data were converted from native
H D F 5 formatinto UV FITS files (Greisen, 2019), averaging four con-secutive raw integrations into 7.92 s files, using the zenithas the phase centre for each file. This generated 33,994individual files to be imaged. Further processing was per-formed with the
M i r i a d data processing suite (Saultet al., 1995).Calibration (phase and amplitude) was performed us-ing the m f c a l task on a subset of data from the solartransit and a flux density model of the quiet sun (Benz,2009). Baselines shorter than 5 λ were excluded to min-imise the contribution from Galactic extended emission.The calibration was transferred to all datasets, theneach file was imaged with i n v e rt into 128 ×
128 pixelimages using robust=-0.5 weights, and deconvolved with c l e a n using 10000 iterations, speed=-1, and phat=0.1.Each image covers the entire hemisphere visible for thatfile with 1.25 degree pixel scale at the zenith in sine pro-jection, with an angular resolution of ∼ . ◦ at zenith.The two independent polarisations (corresponding to the‘XX’ or east-west oriented dipoles and the ‘YY’ or north-south oriented dipoles) were imaged independently. Thisprocess generated two zenith-centred images for eachvisibility file, which were exported as FITS images. The FITS images at the 33,994 individual time steps(at each XX and YY polarisation, for a total of 67,988images) were processed using a Python script.Each image was loaded from disk and a differenceimage was formed via subtraction of the previous imagein time. Difference images remove the component of thesky brightness distribution that remains constant fromtime step to time step, including astronomical sourcessuch as the very bright Galactic Plane. Difference imagesreveal signals that change with time between images andare sensitive to time-varying sources of RFI. Since thereare no sources of FM RFI in the immediate vicinityof the MRO, all sources of RFI in this band are dueto complex propagation or scattering effects, and aretherefore strongly time variable.The difference images were searched for point sources,using the DAOStarFinder task within the photu-tils Python module (Bradley et al., 2019), based onDAOPHOT (Stetson, 1987). At the angular resolution of EDA2, any time-variable sources of RFI will be point-like in nature, even scattering objects in motion (air-craft and satellites, for example). In the search for pointsources, a threshold of ten times the Root Mean Square(RMS) in the difference image was adopted and a FullWidth at Half Maximum (FWHM) for the source of fivepixels was adopted.The RMS values of the difference images vary. For im-ages with strong sidelobes due to sources of interferenceand/or the presence of strong astronomical sources, theRMS values can be higher than at other times. Thus, thedetection threshold of ten times the RMS varies acrossthe set of observations. Figure 2 shows the distributionsof mean and RMS values, for all difference images in theXX polarisation. The YY polarisation distributions arevery similar and are not shown. N u m b e r N u m b e r Figure 2.
Distributions of the mean of the difference images (top)and the RMS (bottom) for the XX polarisation data.
As can be seen in these figures, the RMS valuesare strongly dominated by values of less than ∼ Tingay et al. visualised as density maps in pixel coordinates, celestialcoordinates, and Galactic coordinates, in order to makean initial examination of the variety of signals presentin the data.Broadly speaking, the detected signals can be wellcategorised as the following:1) sources confirmed to a single time step occurringat random locations in the sky, with larger numbersoccurring near the horizon ( < ◦ elevation) than nearthe zenith (mostly likely meteor reflections);2) reflections from aircraft, identified as having veryhigh peak intensities and being trackable from timestep to time step, along known flight paths at angularspeeds corresponding to standard aircraft air speeds andaltitudes;3) candidate reflections from satellites, identified asbeing similar to planes but with far lower peak fluxdensities, detectable preferably near zenith, not havingthe trajectories of known flight paths, and with theexpected angular speeds of objects in Low Earth Orbit;4) persistent sources of time-variable RFI on the hori-zon and at azimuths corresponding to population centresat large distances, beyond the MWA horizon; and5) detections that followed the locations of the bright-est celestial radio sources, largely due to imperfect dif-ference imaging in the presence of the extremely brightradio sources, likely due to combinations of sidelobesaround these sources, the effect of the ionosphere, andsmall errors in imaging and calibration.Across all categories, a total of 40,133 detections weremade in the XX polarisation and a total of 39,954 inthe YY polarisation.This situation of a time variable RMS and a fixedsignal-to-noise detection threshold somewhat problem-atic in terms of a robust statistical analysis of the full67,988 image set and the different signal populations.We could adopt a running median of the image RMS,to mitigate against the variations in RMS. However, thiswould then effectively lower the signal-to-noise ratio ofthe detection threshold in a time variable manner. Forexample, for an image with a high noise (due to side-lobes caused by the presence of a very strong transientsignal) relative to the running median, and using a fixeddetection threshold, the result would be a large numberof false detections above the threshold for that image.In this case, we would be corrupting the events we claimto be real.The approach we have taken ensures that the eventswe claim to be real are real, at the price of occasionallysacrificing real events that would sit above the detec-tion threshold for other images with lower noise levels.Given the large volume of observations, and thereforetotal number of detections, we have clearly been able todistinguish between the different categories of signals,even though we have sacrificed some real signals.We cannot claim to be able to do statistically robust population analyses for these signal types, in part due tothe issue of the variable RMS and the loss of some realdetections. However, we maintain that even without thisissue, such a detailed analysis would be difficult, due to:the calibration of the signal strengths because of primarybeam effects; the intrinsic nature of the signals, whichis highly dependent on the placement and illuminationpattern of the RFI source (FM radio stations hundredsof km away); propagation effects; and the temporal andfrequency resolution of our data.Figure 3 shows a sequence of five images (left panels;intensity range from 0 Jy/beam to 5000 Jy/beam), pluscorresponding difference images (right panels; intensityrange from -600 Jy/beam to 600 Jy/beam), with exam-ples of the different categories of signal present in theEDA2 data, for the XX polarisation. In all images, theGalactic Centre is near transit at the zenith. The Sunis present and labeled and the location of Geraldton onthe horizon is labeled. From top to bottom (time order):the top two images show an aircraft moving from southto north, including high amplitude residuals due to side-lobes; the third, fourth, and fifth images show a satellitemoving toward the north-east, between the GalacticCentre and the Sun. For the aircraft and the satellite,the difference images show the distinctive streaks withpositive/negative structures (Prabu et al., 2020a). Thefifth image also shows a single timestep signal likelyfrom a meteor, near the southern horizon. Throughoutthe series of images, residuals near the Galactic Centreare apparent, also at the Sun (likely due to the strongvariability of even the quiet Sun). Variability of astro-nomical sources due to ionospheric scintillation will alsomake some contribution to difference image artifacts.The variability of the signals received from over thehorizon toward Geraldton is also apparent, in this casedue to tropospheric scattering/ducting.Movies showing the full sequences of the original im-ages, for both XX and YY polarisations, are available aselectronic supporting material in the online publication.Further processing of the data was undertaken toisolate and examine the different categories of signal inmore detail.For the category of signals associated with high am-plitude residuals from the differencing process and asso-ciated with the brightest celestial radio sources in thesky, the simplest method to isolate them is to defineexclusion regions around their locations. Thus, for eachdifference image, a set of celestial radio source coordi-nates were transferred into image pixel coordinates usingthe WCS information attached to the images (utilisingfunctions within the astropy Python module). Exclusionregions with a radius of five pixels were defined aroundthe sources and signals detected in these regions wereexcluded from further analysis. The bright radio sourcesin question are: the Sun; Cygnus A; Centaurus A; 3C273;the Crab nebula; Hydra A; Fornax A; Virgo A; Pictor A; FI in the FM band at the MRO Sun Geraldton2020-02-03 01:25:34.9 Sun GeraldtonSun Geraldton2020-02-03 01:26:45.9 Sun GeraldtonSun Geraldton2020-02-03 01:32:02.9 Sun GeraldtonSun Geraldton2020-02-03 01:32:34.9 Sun GeraldtonSun Geraldton2020-02-03 01:33:06.9 Sun Geraldton
Figure 3.
A sequence of five images (left panels), plus correspond-ing difference images (right panels), for the XX polarisation, asdescribed in the text (times are UTC). and the Vela nebula. Additionally, the Galactic Centre isthe brightest radio region in the sky and this region wasalso excluded, but in this case it was more convenient to define an exclusion region based on Galactic coordinates,such that − ◦ < b < ◦ and − ◦ < l < ◦ definedthe region.The simplest and most effective method to isolatesignals from aircraft was to manually identify the timeranges during which they were visible to EDA2 andconstruct a data set containing aircraft and a data setwith aircraft absent. Given the strong scattered signalsfrom aircraft often produced strong sidelobes, we didnot attempt to recover other categories of sources duringperiods when aircraft were present. In total, 9.82 hoursof data included strong signals from aircraft, from atotal of 75.35 hours of observations, corresponding to atemporal percentage of 13%. Monroe (2018) accuratelydescribes aircraft as “an irritating source of RFI”. Detailsof the characteristics of the aircraft signals are given in§3.1.A similar approach was taken for the isolation of satel-lite signals. In total, seven candidate satellite signalswere detected, occupying approximately 16 minutes ofthe total observation time, approximately 0.3%. Thesecandidates were confirmed via examination of the origi-nal images and were verified by comparing the observedtrajectories to trajectories predicted from orbital pa-rameters corresponding to the epoch of observation. Inall seven cases, a positive identification with an objectin orbit could be made. Detailed information on thesesatellite detections is given in §3.2. Although the frac-tion of the observations containing satellites is small, weexclude these times from the analysis of other categoriesof source in the same way that we exclude aircraft.The isolation of signals from persistent transmittersbeyond the MWA horizon is generally relatively straight-forward, since the signals occupy a constant position inthe images. However, because the difference images willonly characterise the RFI variability between time stepsof a persistent transmitter, and because we are inter-ested in the apparent intensity of the RFI, we extractsignals from the original images and describe the signalsin detail in §3.3. In terms of excluding these signals fromthe difference image analysis, we place exclusion regionsaround the transmitter locations most affected. Thestrongest RFI originates from Geraldton, south-west ofthe MRO. For this transmitter origin, a circular regionof exclusion in the difference images corresponding to aradius of ten pixels was used. Signals from this regionwere not considered further in the difference image anal-yses. Likewise, exclusion regions with a radius of tenpixels were located at the horizon positions of Karathaand Port Hedland, both due north of the MRO.While transmitters on the horizon were generallystraightforward to deal with, we did observe some ex-treme variability on the horizon (almost the entire west-ern horizon) in the last ∼ Tingay et al. in the YY polarisation data (more sensitive in the east-west direction than the XX polarisation). This activityis due to lightning and we describe it in detail in §3.3.Thus, the last 1.8 hours of data were also excluded fromfurther analysis of the difference images.After the isolation of residual errors near bright radiosources, aircraft, satellites, persistent signals on the hori-zon from transmitters, and the extreme activity notedin the final 1.8 hours of the observation period, the re-maining signals in the difference images represent signalsconfined to single time steps and occurring randomlyacross the sky. Overwhelmingly, these signals are likelyto be due to backscatter from meteor trails in the upperatmosphere. Detailed information for these signals ispresented in §3.4. 15,530 of these signals were detected,in both XX and YY polarisations, representing approxi-mately 37% of the total number of detected signals.
For each category of signal we have described and iso-lated in the previous section, we provide detailed in-formation below. In general, we examine the spatial,temporal, and intensity distributions of these signalsand show examples of individual signals. Throughoutthis section, many of the figures depicting histogramsof peak intensity for the various classes of event ap-pear to have powerlaw-like distributions. In all cases,we attempted to parameterise the distributions withpowerlaws, but in no case did a single powerlaw providea good description of any observed distribution.
Figure 4 shows the durations of periods in which aircraftare visible in the EDA2 data, as a function of time ofday. Each point in Figure 4 denotes a discrete periodduring which one or multiple aircraft are present.Figure 5 (top panel) shows example aircraft trajecto-ries, as detected in the XX polarisation via the differenceimaging analysis, over an approximate 6.5 minute pe-riod. As noted in §2.3, the integrated observation timeaffected by RFI reflected from aircraft represented inFigure 4 is 9.82 hours, or 13% of the total observationtime over three days. Over a multi-day period, this willbe representative of the general situation for the MRO.Overwhelmingly, the flight paths over the MRO arenorth-south routes, to/from Perth, which is due southof the MRO, to/from a number of mining centres in thenorth of Western Australia, and to/from a range of in-ternational locations largely north of Australia. A smallnumber of east-west routes to the north of the MRO arealso detected, and also other seldom-used flight paths.A two-dimensional histogram of the spatial occuranceof signals detected during the 9.82 hour period is alsoillustrated in Figure 5 (bottom panel), reflecting the D u r a t i o n ( h r ) Figure 4.
Durations of periods in which aircraft backscattersignals are present in the EDA2 images, as a function of timerelative to the starting time of observations (and folded on a24 hour period, given that flight schedules from day to day aresimilar). preponderance of north-south flight paths over the MRO.Finally, Figure 6 shows the distribution of peak inten-sity of signals in the XX polarisation (the distribution forthe YY polarisation is effectively identical) for periodsduring which aircraft are visible in the EDA2 data. Thesignals shown in Figures 5 and 6 account for detectionsof aircraft, but also include all other detections of othercategories of signals present during these periods. Thedetections are dominated by aircraft detections, how-ever, characterised by their very high peak intensities ascompared to other signal categories, as shown in latersections.
Artificial satellites in Earth orbit also produce backscat-tered signals from FM transmitters, in the same waythat aircraft do. Even though satellite passes are vastlymore frequent above the MRO horizon (constant, in fact)than aircraft, the objects are much more distant ( ∼ ∼
10 km) and generally smaller ( ∼ ∼ th as sensitive), a handful of satellitedetections have been made.Table 1 provides the list of satellites identified fromthe difference imaging process, based on being able totrack a trajectory across the sky over multiple timesteps.Figure 7 shows an example of a detected satellitetrajectory, for the BGUSAT satellite, compared to the FI in the FM band at the MRO Table 1
Identified backscatter from objects in orbit and their properties
Satellite a NORAD Start End Mean intensity RCS b name ID m )BGUSAT 41999 2020-01-31 14:40:09.9 2020-01-31 14:43:11.9 1060 < > − > < > < a TLE information for predicted trajectories from space-track.org for the epoch of observation: 2020-02-01 b Radar Cross Section (RCS) is categorised by space-track.org as: small ( < < RCS < > .MAX VALIER SAT is a small Italian/German amateursatellite, launched in June 2017; it is equipped with anX-ray telescope . FLOCK 3P 71 is a 3U cubesat that ispart of a large fleet of satellites equipped with camerasfor Earth observation; it was launched in February 2017 .Given the low Radar Cross Sections (RCSs) for thesethree objects, we would not typically expect to detectthem in reflected FM transmissions. Some small satel-lites carry long antenna systems for communicationsat frequencies near 150 MHz, which may increase theRCS at FM frequencies. However, even in these caseswe would expect detection to be marginal. In Table 1,BGUSAT is the brightest object detected, brighter thanthe ISS, which would not be expected (note howeverthat these measured intensities are not corrected for theantenna beam pattern in Table 1.It is possible that we are not seeing these small satel-lites in reflected FM, but are seeing direct transmissionsfrom the satallites in the EDA2’s frequency range. Usingthe MWA for non-coherent passive radar observations,Prabu et al. (2020a) found two cubesats to be trans- https://in.bgu.ac.il/en/epif/Pages/BGU-SAT.aspx https://space.skyrocket.de/doc_sdat/flock-1.htm mitting broadband signals. One was the Israeli studentsatellite Duchifat-1 and the other was the UK satelliteUKube-1. Indeed, there is evidence that BGUSAT isin the same situation; EDA2 data similar to those pre-sented here, but at a frequency of 70 MHz (well belowthe FM band), shows a detected signal from BGUSAT(X. Zhang, Private Communication). An investigation ofMAX VELIER SAT and FLOCK 3P 71 should be con-ducted at other frequencies, to determine if they are alsogenerating broadband transmissions. These satellites arenot licensed for broadband transmissions, which raises apotentially concerning situation for ground-based radioastronomy.Finally, Figure 8 shows the distribution of peak inten-sity of signals in the XX polarisation (the distributionfor the YY polarisation is virtually identical) for peri-ods during which satellites are detected in the EDA2data (note that values much higher than the mean peakintensity listed in Table 1 are seen in Figure 8). As opposed to reflected signals from aircraft and satel-lites, the (more-or-less) direct reception of signals fromFM transmitters is also strongly present in our data.Although these transmitters do not have a line-of-sightto the EDA2, the signals can be propagated long dis-tances via tropospheric scattering, and occasionally verystrongly via tropospheric ducting events (Sokolowskiet al., 2017, 2016). FM transmitters are also present inproximity to virtually all towns and settlements acrossWestern Australia, so the reception at EDA2 is not un-expected. In particular, we persistently detect highlyvariable signals from the directions of Geraldton (south-west of the MRO), Karatha (north of the MRO), andPort Headland (north of the MRO). Evidence for similarsignals toward other directions, at much lower levelswere noted in the data but are not expanded upon here.We present some examples of the types of behaviours
Tingay et al.
Figure 5.
Example of two aircraft trajectories across the skyon north-south flight paths (top panel) and the two dimensionalhistogram of detected signals during the periods when aircraft arevisible from the MRO (bottom panel). The red markers in the toppanel denote the positions of strong astrophysical radio sourcesabove the horizon at this time. In both panels, the yellow markerdenotes the location of Geraldton on the horizon. seen toward Geraldton, as the closest location with thestrongest transmitters at this particular FM frequency.In order to illustrate these signals, we take two ap-proaches. First, because these signals are persistent ata fixed location in our images, a difference image ap-proach will only detect the variation of the signal withtime. More useful is to understand the fixed and variablecomponents of the flux density, extracting these mea-surements from the original images rather than from thedifference images. However, we need to accept in thiscase that these measurements will also pick up any as-tronomical contribution at this image location as strongastronomical sources set in the direction of Geraldton. N u m b e r Figure 6.
Peak intensity of signals detected during periods inwhich aircraft are above the horizon at the MRO, for the XXpolarisation.
Figure 7.
Example satellite trajectory detected as described inthe text, for the BGUSAT satellite as listed in Table 1, comparedto the trajectory predicted from orbital elements at the epoch ofobservation. Only the middle portion of the predicted trajectory isshown as dark blue markers, to allow comparison to the observedtrajectory (light blue markers).
In Figure 9, we show the intensity at the horizon inthe direction of Geraldton, as a function of time, forboth XX and YY polarisation data. As can be seen,the signal is persistent but highly variable. The baselineintensity also clearly shows the diurnal signal due toastronomical sources, repeating three times over thethree days. Particularly prominent is the setting of theGalaxy in the direction of Geraldton.Second, we examine the difference image data, butacross the full 360 ◦ horizon, in order to highlight someunusual behaviour noted earlier in the paper duringthe last ∼ FI in the FM band at the MRO N u m b e r Figure 8.
Peak intensity of signals detected during periods inwhich satellites are detected in the EDA2 data (reflections and/ordirect transmissions, as discussed in the text). images for an approximate 1.8 hour period during thesecond of the three days of observations, for both XXand YY polarisations. The intensities are plotted asa function of azimuth angle (east of north) and time.Clearly seen in both polarisations is the persistent andvariable transmitter in Geraldton, with little activity atother azimuth angles. This is typical of the behaviourthroughout the observations.However, Figure 11 shows the same measurements asin Figure 10 but 24 hours later, in the final 1.8 hours ofthe observations, showing strikingly different behaviour.In these data, strong signals are seen apparently propa-gating around almost the entire western horizon, overperiods of 100s to 1000s of seconds. These signals areattributed to lightning to the west of the MRO and areverified with reference to contemporaneous data fromthe BIGHORNS antenna, which show broadband signalstypical of lightning at the same time (Sokokowski, priv.comm.). As broadband signals, lightning will affect amuch wider bandwidth than just the FM band.
After taking account of false detections associated withbright astronomical radio sources, signals due to directreception of transmissions from distant population cen-tres, and reflections and/or transmissions from aircraftand satellites (at least those obviously identifiable), weare left with signals that are confined to single timesteps and are distributed across the sky.Overwhelmingly, these signals are likely to be due toFM reflections off meteor trails. We cannot rule out thatsome of these signals are due to short duration ’glints’off satellites in Earth orbit. However, given the resultsdescribed in §3.2, this is likely to form a very minorcontribution.Figure 12 shows the two-dimensional histogram ofthese events over the sky, for the XX polarisation, with P e a k i n t e n s i t y ( J y / b e a m ) P e a k i n t e n s i t y ( J y / b e a m ) Figure 9.
The intensity as a function of time at the horizon inthe direction of Geraldton, obtained from the original images forXX polarision (top) and YY polarisation (bottom) data.
Figure 13 showing the distribution of events as a func-tion of elevation (between azimuths of 50 ◦ and 200 ◦ , inorder to avoid the significant transmitters on the horizonshown in Figures 10 and 11). These distributions aretypical of expectations for all-sky meteor radars, as sim-ulated in Holdsworth et al. (2004). Within this azimuthrange, we detect 3,352 meteors in the XX polaristion and3,252 in the YY polarisation. Adjusting these numbersfor the full sky, we estimate approximately 8,000 meteorsover the three days, or approximately 2,700 meteors perday.The detailed count rates of meteor reflections as func-tions of time and position on the sky (not during meteorshowers) are functions of the distribution of the six spo-radic meteor radiants, the illumination pattern of theradar transmitter(s), and the beam pattern of the receiv-ing antenna system (Cervera, 2004). In the case here,where the illumination is provided by an ensemble ofcommercial FM transmitters, the illumination patternis not well known. However, the estimated daily ratesfound here are comparable to those found by Holdsworthet al. (2004).Finally, Figure 14 shows the distribution of peak in-tensity for the meteor reflections for the XX polarisation(the YY polarisation results are similar and are notshown).0 Tingay et al. T i m e f r o m - - : : . U T ( h r ) T i m e f r o m - - : : . U T ( h r ) Figure 10.
The horizon intensity (colourbar units of Jy/beam)extracted from the difference images for a 1.8 hour period onthe second day of observations, in XX (top) and YY( bottom)polarisations, as a function of azimuth angle (horizontal axis) andtime (vertical axis).
The data presented above describe the signals enter-ing the EDA2 system via the MWA antennas and thusrepresent the power injected into the system after mod-ification due to the beam pattern due to the antenna(variable sensitivity across the sky). In order to considerthe signals we have studied here in general terms, forexample to consider the effect of these signals on a dif-ferent antenna type with a different beam pattern, we T i m e f r o m - - : : . U T ( h r ) T i m e f r o m - - : : . U T ( h r ) Figure 11.
The horizon intensity (colourbar units of Jy/beam)extracted from the difference images for a 1.8 hour period on thethird day of observations (24 hours after the data shown in Figure10), in XX (top) and YY (bottom) polarisations, as a function ofazimuth angle (horizontal axis) and time (vertical axis). need to correct the observed apparent intensities with amodel for the MWA antenna beam.We have used a simulation of the beam pattern for asingle MWA dual-polarisation dipole antenna over aninfinite ground plane, as used previously to compare thesensitivity expectations to measurements for the firstgeneration EDA instrument by Wayth et al. (2017), tocorrect the observed apparent intensities into calibratedintensities. The beam patterns for the X and Y polari-sations, at the frequency of observation (98 MHz), areshown in Figure 15.
FI in the FM band at the MRO Figure 12.
The two dimension histogram of the 3,352 detectionslikely to be reflections from meteor trails, for the XX polarisation. N u m b e r , < A z < Figure 13.
The elevation angle dependence of meteor reflections,in the azimuth range 50 ◦ < Az < ◦ , for the XX polarisation. Near the horizon, the beam correction is largest andgenerally most uncertain in simulated beam patterns.Additionally, mutual coupling effects between antennasin the array will generally produce deviations from thebeam expected from a single isolated antenna, resultingin the so-called embedded element pattern. For the MWAbow-tie antennas used in the EDA2 configuration, thesemutual coupling effects are extremely minimal, eventoward the horizon and do not represent a significantcomplication to the calibration. A comprehensive set ofdetailed embedded element electromagnetic simulationshave been performed to evaluate the EDA2, as part ofSKA design activities, to be published by Davidson etal. (2020, in preparation).For the aircraft reflections, satellite reflections, andmeteor reflections described above, we have produced alist of all detections for both polarisations. These lists N u m b e r , < A z < Figure 14.
Peak intensity of signals due to reflections from meteortrails in the XX polarisation. The distribution is truncated at10,000 Jy/beam, due to the very long tail of the distribution thatextends to ∼ Figure 15.
X (left) and Y (right) polarisation simulated beampower patterns for an MWA dipole antenna over an infinite groundplane, used to correct observed apparent intensities. The peakvalues are normalised to unity at the zenith. contain: date; time (UTC); azimuth; elevation; right as-cension; declination; observed peak intensity; and beam-corrected peak intensity. For horizon transmitters, weconstruct a list of detections for the Geraldton transmit-ters only, with the same fields, excluding azimuth andelevation and only for apparent intensities (not beamcorrected intensities). The (azimuth, elevation) is fixedat ( − N ∝ F − α ) with α = 1 . ± .
02, with analmost identical result (within the uncertainties) for theYY polarisation.The signals recorded by the EDA2 are for an ap-proximate 1 MHz of bandwidth, containing two FMtransmitter frequencies, at 98.1 and 98.9 MHz. It is im-portant to point out that the FM transmitters each have2
Tingay et al. an approximate transmission bandwidth of 180 kHz andthus do not fill the 1 MHz recorded bandwidth. Thepeak intensity measurements are therefore diluted bythe 1 MHz band. We cannot accurately disentangle therelative contributions of the individual FM transmis-sions to the 1 MHz band, but can make an approximateaverage correction.If we assume that both transmitters make an equalcontribution to the recorded signals, half the receivedpower can be attributed to each transmitter and thesum of their transmitter bandwidths makes up 40% ofthe recorded bandwidth, thereby causing a factor of 2.5bandwidth dilution. Thus, an approximation of the peakintensity for each FM transmitter is 2.5/2=1.25 higherthan the measured (and beam pattern corrected) peakintensities recorded in the online data accompanyingthis paper. Additionally, the amplitude calibration ofthe array utilised the Sun, which was at 9 ◦ from thezenith, leading to a small underestimate of the calibratedintensity, approximately 5%. Finally, the beam correc-tions are more uncertain at the horizion and very lowelevations than at significant elevations ( (cid:38) ◦ ). Eachof these uncertainties are small in the context of theintensity distributions shown earlier and the propaga-tion effects due to the atmosphere, in the case of lowelevation signals.A temporal dilution of the signals is also likely, butcannot be estimated. For the first time, this study has spatially and temporallyresolved sources of RFI at the MRO, at one frequencywithin the FM band. This has allowed a decompositionof RFI sources into well-isolated categories that gener-ally represent the main forms of RFI experienced bywell designed radio telescopes (i.e. excluding sources ofself-generated RFI by electronic equipment and poorshielding).At low radio frequencies, even at a remote and radioquiet location, such as the MRO, complex propagationeffects allow the reception of ducted and scattered RFIorigining at large distances. In the data presented here,we see all of these effects, making the data a valuableprobe of the RFI environment at the MRO in the FMband, as well as the general characteristics of theseeffects.The primary sources of scattered RFI we observe aredue to reflections off aircraft and meteors, originatingfrom transmitters at large distances. In both cases, thereceived power at the telescope reaches into the tensof thousands of Jy for a very significant proportion ofthe events recorded. We found that reflections fromaircraft are common at the MRO, due to the routineflight paths that cross the MRO on north-south tracks,occupying 13% of the observing time over a three day period. These results are for a limited frequency range,approximately 1 MHz centred at 98.4375 MHz, designedto contain signals from two Geraldton-based FM radiostations. Therefore, our data represent a worst case insome form, as we have chosen the nearest and strongesttransmitters. However, we would expect qualitativelysimilar results at other frequencies in the FM band,from more distant transmitters. Moreover, we wouldexpect similar results at any other frequency in theSKA band where transmitters exist at similar distancesand transmission powers. Thus, the results here are areasonably general representation of the effects that willbe seen at other applicable frequencies.
Measurements of this type are important in order tounderstand the detailed environment in which an inter-ferometer operates, to obtain an overall understandingof the performance of a system for key science goals.In particular, a significant part of the science missionfor the MWA and the low frequency component of theSKA, is the search for signals from the so-called Epochof Reionisation (EoR), requiring very high sensitivity,high stability of the system in frequency and time, andan exquisite understanding the “foreground” signals (allpower that enters the signal path that has been producedin the last 13 billion years!)The most recent estimates of the cosmic reionisationhistory, based on Planck cosmic microwave backgroundanisotropy data, indicate that later onset and shorter du-ration reionisation scenarios are favoured (Paoletti et al.,2020). Figure 16 shows the currently favoured scenario(2018 SROLL2) for ionisation fraction as a function ofredshift, relative to the redshift range corresponding tothe FM band relevant to the RFI analysis presented here( z = ∼
12 – ∼ FI in the FM band at the MRO Figure 16.
Cosmic reionisation history scenarios (ionisation frac-tion as function of redshift) adapted from Paoletti et al. (2020).The green shaded region represents the most recent expectationfrom Planck data. The vertical blue shaded block represents theFM band in redshift space.
Thus, assessing if a particular observational scenario,such as for the EoR experiment, satisfies the limitsderived by Wilensky et al. (2020) is a complex taskthat requires a detailed model for the response of thetelescope, a model that effectively describes where thetelescope is pointing and when, and a detailed temporaland spatial model for the RFI environment. This analysisis beyond the scope of this paper, but our work doesprovide one of these crucial ingredients, that has beenmissing until now, an empirical measurement of thetemporally and spatially resolved RFI environment.As a result of our analysis, we have generated separatelists for meteor reflections, satellite reflections, aircraftreflections, and horizon transmitters in both polarisa-tions, and recording time of event, azimuth and elevation(and corresponding Right Ascension and Declination),and peak intensity, over an approximate 72 hour period(both observed and corrected for antenna beam patterneffects). We will utilise this information, using the meth-ods in Wilensky et al. (2020) to evaluate the detailedimpact of RFI on the EoR experiment for instrumentsbased at the MRO, and will report the results in a futurepublication.These lists are available as online data accompanyingthis paper.
With the increasing utilisation of the space environmentfor a variety of purposes, supported by new technologiesthat allow small, inexpensive satellites and mass launch capabilities, concerns regarding the impact on ground-based astronomy are also increasing (McDowell, 2020;Gallozzi et al., 2020; Hainaut & Williams, 2020; Mallama,2020). A lot of the focus to date has been on the impactto ground-based optical telescopes, the most prominentexample being with respect to the StarLink constellationof satellites.However, all small satellites have uplink/downlinkcommunications systems that utilise radio frequencies.Popular frequencies include those within the 2 m ama-teur band, with downlink transmitters in the 144 - 148MHz range. This band is in the middle of the frequencyrange of the MWA and the low frequency SKA. There-fore the performance of these transmitters is of veryhigh interest in the radio astronomy community. Withinexpensive electronics, potentially limited testing, andthe potential for malfunction on orbit, these transmit-ters can pose a challenge for telescopes such as the SKA.A recent study shows that communications subsystemscontribute to 14%, 16%, and 29% of cubesat missionfailures within the first 0 (Dead On Arrival: DOA), 30,and 90 days on orbit (Langer & Bouwmeester, 2016).Thus, these systems are prone to malfunctions of varioustypes.Our results in §3.2 indicate that the small satellitesMAX VALIER SAT and FLOCK 3P 71 may be trans-mitting out of their designated transmission bands andthat the BGUSAT satellite is very likely to be trans-mitting out of band. These results add to previous outof band detections of DUCHIFAT-1 and UKube-1 byPrabu et al. (2020a) and Zhang et al. (2018). Thus,there is a growing list of small satellites for which thereis evidence for significant out of band transmission.It is not straightforward to determine the character-isation of out of band or spurious transmissions froma satellite, or determine conclusively if a given satel-lite is violating its communications license, according todocumentation of the International TelecommunicationsUnion (ITU) Tingay et al. table/table.pdf, both DUCHIFAT-1 and UKube-1utilise the ISIS TRXUV transceiver, operating withdownlink frequencies near 145 MHz . The MAXVALIER SAT may also utilise the same transceiver We thank the anonymous referee for constructive commentsthat improved aspects of the paper. This research has madeuse of the NASA/IPAC Extragalactic Database (NED),which is funded by the National Aeronautics and SpaceAdministration and operated by the California Institute ofTechnology. This scientific work makes use of the Murchi-son Radio-astronomy Observatory, operated by CSIRO. Weacknowledge the Wajarri Yamatji people as the traditionalowners of the Observatory site. This research made use ofPhotutils, an Astropy package for detection and photometryof astronomical sources (Bradley et al., 2019). The acquisi-tion system was designed and purchased by INAF/OxfordUniversity and the RX chain was design by INAF, as partof the SKA design and prototyping program.
REFERENCES
Benz A. O., 2009, 4.1.1.6 Radio emission of the quiet Sun:Datasheet from Landolt-Börnstein - Group VI Astron-omy and Astrophysics · Volume 4B: “Solar System”in SpringerMaterials (https://doi.org/10.1007/978-3-540-88055-4_5), doi:10.1007/978-3-540-88055-4_5, https://materials.springer.com/lb/docs/sm_lbs_978-3-540-88055-4_5
Bradley L., et al., 2019, astropy/photutils: v0.6, doi:10.5281/zenodo.2533376, https://doi.org/10.5281/zenodo.2533376 Cervera M. A., 2004, Journal of Geophysical Research,109Clark M. A., La Plante P. C., Greenhill L. J., 2011,arXiv e-prints, p. arXiv:1107.4264Comoretto G., et al., 2017, Journal of AstronomicalInstrumentation, 6, 1641015Furlanetto S. R., Oh S. P., Briggs F. H., 2006, Phys. Rep.,433, 181Gallozzi S., Scardia M., Maris M., 2020, Concerns aboutground based astronomical observations: a step tosafeguard the astronomical sky ( arXiv:2001.10952 )Greisen E. W., 2019, Aips memo series, AIPS FITS FileFormat. AIPS Memo 117Hainaut O. R., Williams A. P., 2020, Astronomy &Astrophysics, 636, A121Hennessy B., et al., 2019, in 2019 IEEERadar Conference (RadarConf). pp 1–6,doi:10.1109/RADAR.2019.8835821Hennessy B., Rutten M., Tingay S., Young R., 2020,arXiv e-prints, p. arXiv:2003.03947Holdsworth D. A., Tsutsumi M., Reid I. M., NakamuraT., Tsuda T., 2004, Radio Science, 39, n/aJohnston S., et al., 2008, Experimental Astronomy, 22,151Langer M., Bouwmeester J., 2016.Mallama A., 2020, arXiv preprint arXiv:2003.07805McDowell J. C., 2020, The Low Earth Orbit SatellitePopulation and Impacts of the SpaceX Starlink Con-stellation ( arXiv:2003.07446 )Monroe R. M., 2018, Gigahertz Bandwidth andNanosecond Timescales: New Frontiers in Ra-dio Astronomy Through Peak PerformanceSignal Processing, doi:10.7907/25DP-J474, https://resolver.caltech.edu/CaltechTHESIS:06042018-004220017
Naldi G., et al., 2017, Journal of Astronomical Instru-mentation, 6, 1641014Offringa A. R., et al., 2013, A&A, 549, A11Offringa A. R., et al., 2015, PASA, 32, e008Palmer J. E., et al., 2017, in 2017 IEEERadar Conference (RadarConf). pp 1715–1720,doi:10.1109/RADAR.2017.7944483Paoletti D., Hazra D. K., Finelli F., Smoot G. F., 2020,arXiv e-prints, p. arXiv:2005.12222Prabu S., Hancock P. J., Zhang X., Tingay S. J., 2020a,arXiv e-prints, p. arXiv:2002.01674Prabu S., Hancock P. J., Zhang X., Tingay S. J., 2020b,PASA, 37, e010Sault R. J., Teuben P. J., Wright M. C. H., 1995, inShaw R. A., Payne H. E., Hayes J. J. E., eds, Astro-nomical Society of the Pacific Conference Series Vol.77, Astronomical Data Analysis Software and Systems
FI in the FM band at the MRO arXiv:astro-ph/0612759arXiv:astro-ph/0612759