SKA site spectrum monitoring, measurement program and data processing

SKA site spectrum monitoring, measurement program and data processing Draft Report Verified: Name Signature Date Rev.nr.... 0.3 R.P. Millenaar R.T. Schilizzi Accepted:......... date: date: date: c ASTRON 2011 1

Distribution list: Group: Expert Panel on RFI and EMI For Information: R.T. Schilizzi Site Representatives Document revision: Revision Date Section Page(s) Modification 0.3 July 25, 2011 all all creation.......... 2

Acknowledgements Thanks to the Site teams for contributions to this report, specifically concerning information on the SKA 2010/2011 spectrum monitoring campaign systems and for their contributions and feedback concerning the signal processing. 3

Contents 1 Introduction 6 1.1 Measurement types........................................ 6 1.2 Deviations from the RFI 2008-2011 agreement......................... 7 2 Measurement set-up 8 2.1 Bandwidth and integration time of the modes......................... 8 3 Measurement programme 10 3.1 Fast Scanning mode (FS).................................... 10 3.2 High Sensitivity mode (HS)................................... 10 3.3 Transient mode (TR)....................................... 11 3.4 Rural Mode (RM)........................................ 11 3.5 Max hold Mode (MH)...................................... 12 4 Calibration 13 4.1 Gain calibration......................................... 13 4.2 Gain smoothing.......................................... 14 4.3 Noise temperature calibration.................................. 15 4.4 Power sanity check........................................ 16 4.5 Antenna gain........................................... 18 4.6 Flux calibration.......................................... 20 4.7 Transmitter detection and baseline correction......................... 22 4

4.8 Occupancy statistics....................................... 23 4.9 Time statistics and averaged power statistics......................... 24 4.10 Figure compression........................................ 25 5 Output description and measurement result summary 27 5.1 FS mode results.......................................... 27 5.1.1 Spectrum measurement spectra............................. 27 5.1.2 Calibration spectra.................................... 27 5.2 HS mode results......................................... 28 5.2.1 Spectrum measurement spectra............................. 28 5.2.2 Calibration spectra.................................... 29 5.3 RM mode results......................................... 29 5.4 MH mode results......................................... 30 5.5 TM mode results......................................... 30 References 31 Appendix A. Measurement report list 32 Appendix B. Allocations and services 33 Appendix C. Component calibration data 34 Appendix C1. Antenna, LNA, and noise diode........................... 34 Appendix C2. LNA, and noise diode................................. 35 Appendix A2. Cables and switches.................................. 36 5

1 Introduction As part of the SKA site selection process the Radio Frequency Interference environment has been measured at two candidate sites. This was done under the Agreement on radio frequency interference monitoring 2008, cf. [1]. This Agreement entails the development of instrumentation, control and data processing software, infrastructure and carrying out of measurements at the two candidate core sites, plus at a selection of remote sites in the hosts proposed array configuration. This spectrum monitoring report is one of a series. 1.1 Measurement types The spectrum measurements were taken at the core sites and at the remote sites. Five types of measurements have been carried out: Fast Scanning Mode (FS) High Sensitivity Mode (HS) Transient Mode (TM) Rural mode (RM) Max Hold mode (MH) The results of each mode and of each separate run are presented in separate reports. Fast Scan measurements have been taken at several times during the month-long campaign at the core sites. Each of these FS measurements are reported in individually generated documents. The measurements are interspersed with High Sensitivity Mode measurements and serve to provide an overview of the RFI environment for all pointing directions and both polarizations, while the HS measurements concern only one pointing direction and polarization per measurement set. As such the FS measurements provide an overview of the spectrum in all directions and polarizations during the course of the campaign, albeit at somewhat lower sensitivity because of the reduced integration time. The spectrum measurement reports are coded by three identifiers. The first identifier is the measurement number preceded with the letter M. The second identifier is the country at which the observations were done: C1 and C2 for the two sites. The identifier C3 is used for reports which present results from both sites. The third identifier denotes the measurement type: FS, HS, TR, RM or MH. The measurement report code is a combination of the three identifiers separated by a dot. 6

1.2 Deviations from the RFI 2008-2011 agreement Measurements have been conducted according to the RFI monitoring agreemement [1]. There are however a few deviations from the specied measurment protocol: 1. Channel bandwidth. Due to hardware limitations, complexity issues, and planning issues, it turned out to be very dicult to meet the 10 khz channel bandwidth requirement. There was consensus in the team that a moderate increase of the channel bandwidth would not hamper the monitoring campaign, nor the data quality, nor compromise the monitoring campaign goals. The channel bandwidths used are listed in table 1. 2. HS Mode integration time. There were two reasons to deviate from the suggested integration time of 24 hours net for the HS measurements: Hardware development time was much longer than anticipated, and in combination with restrictions for doing the measurements at the core during a period of absence of self generated RFI due to construction activities left only one month available. Artefacts generated by the FFT spectrometer turned out to be such that they can only be calibrated out by equal length measurement phases for noise source on, off and antenna. Therefore the measurement effciency was reduced from 90 to 33%. Both arguments required the integration time to be reduced from 24 to 2 hours, and this constitutes a reduction in sensitivity of 12 or 5.4 db. Required sensitivties in db(w m 2 Hz 1 ) can be found in the agreement [8]. For the cores sites the aim was to be close to the RA769 sensitivities [4], ranging from about -258 to -240 db(w m 2 Hz 1 ). Due to these technical and observation time limitations the achieved sensitivities are about a factor 10 higher than the (interpolated) RA769 values. The sensitivies for the other modes are lower than mentioned above as the integration times for those modes are lower. 7

2 Measurement set-up A description and technical details of the RFI campaign instrumentation can be found in [8]. This section only briefly outlines the system and some of the used parameter settings. Figure 1 shows a simplified block diagram of the system. A log periodic antenna (HL033) is connected to a low noise amplifier (LNA) via a coaxial switching box. The signals are digitized and converted to spectra by using a digital spectrometer backend. In the Fast Scanning (FS), High Sensitivity (HS), Rural (RM), and Max Hold (MH) modes, the spectra are time-averaged power spectra (resp. 1 10, 120 60, 10 60, and 90 0.1 s). In the Transient mode (TR) the power spectra are averaged over much smaller time intervals (500.000 1µs) as this mode is intended to obtain high time resolution statistics. A noise diode can be connected to the LNA in order to calibrate the system. The noise figure of the diode, the gain of the antenna and LNA, and the switch losses are given in the appendix. antenna HL023 cable 4 1 cable 2 cable 3 LNA back-end 2 switch cable 1 noise diode Figure 1: Simplified block diagram of the SKA spectrum monitoring system. Note that the calibration methods involves an RF switch and cable 1, and that the antenna and cable 4 are not included the signal path while calibrating. This will give rise to a spectral ripple in the gain, receiver temperature, and calibrated spectra. The ripple does not affect the detection of narrow-band signals, but may impede detection of weak broadband signals. 2.1 Bandwidth and integration time of the modes The used spectral ranges for the modes are shown in table 1. The table also shows the ADC sampling frequency F s, and the channel bandwidth f. 8

F s (MHz) f (Hz) 70-200 975 29.8 200-430 975 29.8 430-530 780 23.8 530-850 960 29.3 850-1050 780 23.8 1050-1330 975 29.8 1330-1550 830 25.3 1550-1800 960 29.3 1800-2000 850 25.9 Table 1: Spectral ranges, sampling frequencies, and channel bandwidths. 9

3 Measurement programme The measurements that were carried out can be classified into a number of categories, that aim to capture a particular aspect of the RFI environment. These are summarised in the following paragraphs. 3.1 Fast Scanning mode (FS) This mode has been carried out at the core sites and was intended to provide relatively rapid overviews of the entire spectrum (all 9 bands) in all directions (4 pointings, 2 polarisations). The measurements were repeated several times, at least once between instances of the long duration High Sensitivity mode (HS) measurements. The FS measurements have been carried out with the following instrumentation settings: integration time: 10 sec (per spectrum) number of repetitions: 1 net on antenna integration time per pointing/polarisation/band: 10 s calibration: equal time on noise source on/off/antenna (10s/10s/10s), each repetition One complete FS measurement (4 pointings, 2 polarisations, 9 bands), including calibration and overhead, takes 100 minutes. The Fast Scanning measurements at the core sites have been carried out 10 times at one of the sites and 9 times at the other site. The differences are caused by unforeseen circumstances dealing with issues in the field. 3.2 High Sensitivity mode (HS) The High Sensitivity mode is intended to maximise the sensitivity of the measurements by increasing the integration time to limits set by the total available time at the core site. These measurements were only carried out at the core sites. Available time at the remote sites precluded inclusion of this mode there. One HS measurement uses the following settings: integration time: 60 s (per spectrum) number of repetitions: 120 net integration time per pointing/polarisation/band: 7200 seconds (2 hours) calibration: equal time on noise source on/off/antenna (60s/60s/60s), each repetition A complete scan (1 pointing, 1 polarisation, 9 bands), including calibration and overhead, takes a total of about 2.9 days. For a complete survey, all pointings and polarisations, a total of about 23 days was required. 10

3.3 Transient mode (TR) Transient mode measurements have shown to be very time consuming because of the very large amount of time required to read out the data from the hardware. In addition there have been hardware issues that compromised the dynamic range. For this reason the transient mode was used only at the core site. The following parameters apply: integration time: 1 µs (per spectrum) number of repetitions: 5 10 5 (elapsed time per scan 500 ms; captured in 16 blocks of 64 ms continuous capture) net integration time per pointing/polarisation/band: 1 µs channel bandwidth: 1 MHz calibration: fractional calibration time (on noise source) 0.008 (noise diode on/off: 2ms/2ms; antenna meas. 0.5s) A complete TR survey thus requires 8 repetitions for all four azimuthal directions and both polarisations. Including calibration and the overhead for writing of data from hardware to system disk one complete measurement takes about 9 hours. This does not include the time required to read the data from the system disk to off-line media. 3.4 Rural Mode (RM) An intermediate integration time measurement mode is used at rural (remote) sites to obtain data at a sensitivity high enough to provide useful rural (remote) site information, while limiting the required time at the site. Therefore the sensitivity is less than the HS mode, but higher than with the FS mode. One complete RM measurement uses the following settings: integration time per cycle: 60 s (per spectrum) number of repetitions: 10 net integration time: 600 s calibration: equal time on noise source on/off/antenna (60s/60s/60s), each repetition A complete scan (1 pointing, 1 polarisation, 9 bands), including calibration and overhead, takes a total of about 5.8 hours. One complete survey, all pointings and polarisations, then takes almost 2 days. 11

3.5 Max hold Mode (MH) The Transient mode (TR) measurements were not carried out at the remote sites. To capture RFI of short duration at maximum levels not time averaged, integrated levels another measurement mode was used at these remote sites: the pseudo Max Hold (MH) mode. In essence in this mode the hardware operates in the same manner as in the FS and HS modes, but with very short internal spectrometer integration time (100 msec), for a large number of repetitions. In software, the statistics including the maximum and various percentile levels can be derived. One complete MH measurement uses the following settings: integration time per cycle: 0.1 s (per spectrum) number of repetitions: 90 net integration time: 0.1 s (for statistics), 9 s when integrated calibration: one calibration cycle per 90 repeats A complete scan (4 pointings, 2 polarisations, 9 bands), including calibration and overhead, takes a total of almost 1.3 days (the large number of files generated causes substantial overhead). 12

4 Calibration This section describes the calibration and main signal processing algorithms used in the spectrum monitoring. Only the formulas used are described, no detailed background information is given as this can be found in literature, and for example in [3, 5, 6, 7]. In this section (and throughout the report) all quantities are expressed in SI units. 4.1 Gain calibration The gain of the monitoring system as depicted in figure 1 can be estimated using a noise diode. In this calibration process three measurements are required: (1) measurements with a the noise diode switched on, (2) a measurement with the diode switched off (being at ambient temperature and therefore presenting as a matched load of ambient temperature) and (3) a measurement in which the antenna is connected to the RF/IF chain. For the noise source on/off measurements, the diode output power needs to be known. This power is related to the excess noise ratio ENR calibration data of the diode, which is available. The noise temperature T hot of the noise diode when switched on is given by T hot = (ENR + 1)T amb (1) where T amb is the ambient temperature in K. Given an ambient temperature T amb and known values for ENR, P hot is computed with the formula above. In the following a description is given of how the gain is computed from the three observations mentioned above. Let the receiver temperature be denoted by T rec, the noise source on temperature by T hot, the noise source off temperature by T cold, the antenna temperature by T ant, and the system temperature by T sys. Denote the received power when the antenna is connected by P ant, the received power when the noise source in off state is connected by P cold, and the received power when the noise source in on state is connected by P on. Denote the electronic gain by G e, then the following relations hold. P sys = G e kb(t rec + T ant ) (2) P hot = G e kb(t rec + T hot ) (3) P cold = G e kb(t rec + T cold ) (4) P sys, P hot, and P cold values are obtained from measurements, figure 2 shows examples. Here, the Boltzmann constant is denoted by k, and the channel bandwidth is denoted by B (in the actual computations, the factor kb is incorporated in the gain factor G e ). Initially, G e, T rec and T ant are unknown and must be calibrated. The T cold temperature of the noise source is equal to its ambient temperature, assumed to be 300 K. Equations (3) and (4) imply that the gain can be deduced from hot-cold measurements yielding values for P hot, P cold, and P sys : G e = P hot P cold T hot T cold (5) Please note that the gain value computed here is the electronic gain, the antenna gain is not included. Figure 3 shows an example of the estimated gain for the fast scanning mode. 13

200 180 160 raw data, measurement raw data, "cold" raw data, "hot" SKA monitoring 2010/2011, power, Virtual Site power (db) 140 120 100 80 60 40 20 0 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 2: Spectrum analyzer observed power: noise source on, noise source off, and measurements using the HL033 antenna. Clearly visible are the transitions between the nine processing bands.the bands have an artifical offset of multiples of 10 db for ease of inspection. Both the pass-band and blocking part of each of the nine bands are shown. 4.2 Gain smoothing For several reasons the integration times for the noise source on, noise source off and for the spectrum measurements were equal for the FS, HS, and RM modes. As the electronic gain is used for calibrating the power flux spectral densities, noise fluctuations of the gain estimates will increase the rms noise of the calibrated spectra. In order to reduce this effect, a gain smoothing algorithm is available. Applying this algorithm would lead to a small systematic error due to the spectral ripple. Although the gain smoothing function is available, it is not used in the final data processing. The reason is that it turned out that there are narrowband digital artefacts produced by the the digital system hardware and firmware. These artefacts will of course also be present in the gain curve. Smoothing the gain curve (and increasing the spectral resolution again by interpolation) will spread these artefacts to a larger fraction of the band. For this reason it was chosen not to use the gain smoothing function. The gain smoothing algorithm averages the estimated gains over frequency. The spectrum is divided into blocks of N gainsm spectral pixels and the median value is computed both of the gain and of the frequencies of the gain data. The blocks are fixed, the processing is not done on running intervals. If the number of frequency bins is not a multiple of N gainsm then the last data point is copied so as to fill 14

65 SKA monitoring 2010/2011, gain, virtual site 60 55 G (db) 50 45 40 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 3: Estimated gain for four pointing s and two polarizations, clearly visible are the transitions between the nine processing bands. a full N gainsm interval. The spectral resolution of the smoothed data is restored to its original size by interpolating the median values. Edge effects are avoided by including the first and last spectral points before interpolation. Figure 4 shows an example of the effect of gain smoothing on the estimated gains. 4.3 Noise temperature calibration As a sanity check for the spectrum monitoring measurements the system temperature T sys, receiver temperature T rec, and the antenna temperature T ant are estimated and plotted. The system temperature is defined by T sys = T rec + T ant (6) Given an estimated gain G e (eq.5), T sys can be deduced using equation (2) and a measured value for 15

SKA monitoring 2010/2011, gain, virtual site 49.15 49.1 49.05 G (db) 49 48.95 48.9 48.85 compr 140 141 142 143 144 145 146 147 Figure 4: Estimated gain for four pointings and two polarization, without and without 20-point median filtering (gain smoothing). P sys ; the receiver temperature can be estimated by using equation (4), assuming T cold is known: T sys = P sys kbg e (7) T rec = P cold kbg e T cold (8) The antenna temperature can be estimated by using equations (2) and (8): T ant = P sys P cold kbg e + T cold (9) Examples of estimated T sys and T rec curves are shown in figure 5. The system temperature estimates are, obviously, affected by RFI. T sys estimates can be improved by applying filtering techniques in the spectral and/or temporal domains. 4.4 Power sanity check As the monitoring antenna is a directive antenna pointed at the horizon, one would expect that for high frequencies the observed power in a noise source off measurement would be a few db higher than when the directive antenna is connected. The reason is that the noise source off temperature is T amb whereas the directive antenna looks partly at the cold sky and partly at the warm ground (at T amb ). At 16

250 SKA monitoring 2010/2011, receiver temperature, virtual site 200 T rec (K) 150 100 50 0 0 200 400 600 800 1000 1200 1400 1600 1800 2000 500 SKA monitoring 2010/2011, system temperature, virtual site 450 400 350 T sys (K) 300 250 200 150 100 50 0 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 5: Estimated receiver temperature and estimated system temperature frequencies below about 200 MHz, sky power dominates and the observed power for the measurements with the antenna connected will be higher than for the noise source off measurements. This reasoning, 17

obviously, is only valid for spectral bands in which no RFI is present. Figure 8 shows an example of measured spectra for the case in which an antenna is connected, compared to a connected noise-source off case. The solid curves in the upper figure are observations in which the noise source was connected; the dashed curves represent data obtained with the monitoring antenna connected. The lower curve shows the difference in power between the two: P ant - P cold 4.5 Antenna gain The isotropic effective antenna area A iso eff is given by A iso eff = λ2 4π (m2 ) (10) where λ is the observed wavelength in m. The antenna factor is defined as the ratio of the field strength (E) of the impinging electromagnetic wave to the voltage (V) on the line connecting an antenna. The linear antenna factor K a is given by 1 z 0 K a = A iso eff G (m) (11) z with z 0 = 120π the free space impedance, z the impedance of the line connecting the antenna, and D the antenna directivity. The logarithmic antenna factor AF is given by AF = 10 log 10 (K a ) (dbm 1 ) (12) For the spectrum monitoring measurements an log periodic antenna, HL033, is used. This antenna has a gain of 6 db at 80 MHz to 7 db at 2000 MHz. Intermediate values are interpolated. The effective aperture for the antenna, A eff is then computed by using A eff = λ2 4π G ant (13) Note: the antenna gain and effective area specified in this way are only correct in case the transmitter is located in the direction in which the antenna points and in case the transmit and receive antenna polarizations match. In case these do not match then the observed power flux density is an underestimate of the actual power flux density. As a directive gain antenna is used in the measurements (R&S HL033), with a gain of 6-7 db, all measurements in the monitoring survey were conducted in four directions (North, East, South, West) and two polarizations (horizontal and vertical). In this way in at least one in eight observations the power flux density is not underestimated. For high elevation observations, the observed power flux densities are always underestimated. For this reason satellite signals, not originating from the direction of the horizon, should be assessed with levels up to 6 db higher than indicated in the spectra. No correction for this effect is incorporated in the data processing. 18

85 SKA monitoring 2010/2011, matched power, virtual site 80 power (db) 75 70 65 60 0 200 400 600 800 1000 1200 1400 1600 1800 2000 6 SKA monitoring 2010/2011, matched power difference, virtual site 4 2 power (db) 0 2 4 6 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 6: Upper figure: matched power observations P cold (black) compared to an observation in which the antenna is connected P ant (dashed curves); lower figure: P cold P ant 19

4.6 Flux calibration Power flux density calibration Let k denote the Boltzmann constant (k = 1.38 10 23 JK 1 ), define the channel bandwidth under consideration by f (Hz), and let Ψ sky (W m 2 Hz 1 ) denote the apparent sky power flux density. The observed (system) power density P sys (in W Hz 1 or arbitrary units 1 can then be written as P sys = kt sys f, which can be related to the apparent sky flux density Ψ sys by P sys = Ψ sys A eff. Assuming the transmit and receive antenna polarizations match and the receive antenna points in the direction of the transmitter, the equations above lead to the following relation: Ψ sys = kt sys A eff (14) Using the definition for T sys (6) and using relation (2) give the relation between the measured power spectrum P sys and T sys : T sys = P sys (15) kbg e hence the calibrated flux density P sys is given by the equation Ψ sys = kp sys kbg e A eff (16) In this equation P sys is known from observations, G e is known from equation (5), and A eff is known from equation (13). However, it was found that the digital backend contributed additive spectral features to the data. Analysis shows that these features could be suppressed by subtracting noise diode off data from the measurements. So in the formula above, P sys (antenna measurement data) is replaced by P sys (P cold P cold,av ). The P cold data has a nonzero average value and a slope which must be removed in order not to influence the flux estimation too much. For this reason the P cold,av curve is introduced: this is a straight line with the same average value and slope as the original P cold data. A consequence of this procedure, implemented in the data processing, is that the (apparent) sensitivity decreases with a factor 2 (1.5 db). Interpretation In frequency channels where signals from transmitters are the dominating power contributions to the system, equation (15) gives the relation between the observed transmitter power and the estimated power flux density. For frequency channels in which there is no transmitter signal present, equation (15) gives an apparent sky flux related to system noise contributions. This is a baseline curve. Assuming approximate values T sys = 300 K and an antenna gain ranging from 6 db at 80 MHz to 7 db at 2 GHz, the apparent baseline flux ranges from approximately -210 db(w m 2 Hz 1 ) at 80 MHz to approximately - 180 db(w m 2 Hz 1 ) at 2 GHz. In case the observed transmitter power is of the same order of magnitude 1 in case the observed power P syshas an unknown scaling factor, this factor will be calibrated and incorporated in the electronic gain factor G e ; as mentioned before, in the data processing the factor kb is also incorporated in the gain G e. 20

as the other system noise components or lower, then the noise contributions must be subtracted from the data in order to get corrected power flux density estimates. This process is described in the next paragraph. Sensitivity 180 190 200 SKA monitoring 2010/2011, sensitivity, Virtual Site sensitivity (T sys ) sensitivity (T rec ) RA769 line RA769 continuum Ψ (db(wm 2 Hz 1 )) 210 220 230 240 250 260 270 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 7: Sensitivity example, computed according to equation (17) using estimates for T sys and T rec compared with RA769 values [4]. Given calibrated auto-correlation spectra with apparent power flux densities Ψ sys, the standard deviation σ of the noise fluctuations, or sensitivity, is given by Ψ sys σ = 2 (17) f τ where f is the channel bandwidth (Hz), and t is the integration time (s). The factor 2 stems from the noise diode signal subtraction procedure mentioned in the previous section. Given estimated values for T sys, table values for A eff, and specified values for f and τ, one can compute expected values for σ. Estimated values of T sys may include observed transmitter signals, so an alternative is to use estimated values for T rec instead. A third way to estimate the sensitivity is by computing σ directly from the data using the standard deviation: σ = stdev(ψ sys ) (18) 21

this computation can be preceded by a detrending operation to remove a slope in the data. In this computation of the median one has to take care not to include spectral channels affected by transmitter signals. In the signal processing this is implemented by separating the spectrum in intervals of length N det = 100. The values in each of these intervals are sorted and the median is computed over the lowest x med % of the sorted data, with typically x med 75 = %: Ψ med = median(ψ sys ) (19) In the data processing, the power flux density and is computed using equations (17) and (18). The spectral resolution of this data is reduced by a factor N det as this is the interval length the quantities are computed from. If the number of samples is not an integer multiple of N det then the last estimate of Ψ med and σ of the data sequence is replaced by the estimate in the one but last interval. The two quantities Ψ med and σ are restored to their original spectral resolution by interpolation. Figure 7 shows an example of the sensitivity Ψ as a function of frequency computed using equation (17) using values for T sys and T rec, for an integration time of 10 s and for a bandwidth of 30 khz. The sensitivities are compared to the ITU-R-RA769 values [4] 4.7 Transmitter detection and baseline correction Transmitter detection and baseline subtraction Given frequency averaged measurement data Ψ med, calibrated original data Ψ sys, and estimates of σ, one can compute the Boolean function Ψ det which values are 0 in case no transmitter is detected and 1 in case a transmitter is detected: Ψ det = { 1 if Ψsys > (Ψ med + N σ σ) 0 if Ψ sys (Ψ med + N σ σ) (20) Here N σ is an integer value for setting the detection threshold level. For those frequency channels in which a transmitter is detected, the flux estimate Ψ c (the baseline corrected flux) is computed according to: Ψ det c = Ψ det (Ψ sys Ψ med ) (21) For frequency channels in which no interference is detected the Ψ rfi values are set to the detection level: Ψ nodet c = (1 Ψ det )N σ σ (22) The baseline corrected power flux Ψ c is then obtained by adding the detected spectra ψc det non-detected spectra Ψ nodet c : and the Ψ c = Ψ det (Ψ sys Ψ med ) + (1 Ψ det )N σ σ (23) 22

The measurement protocol [1] specifies that the sensitivity curve N σ σ should be added as separate plots. However, the Ψ c data already include threshold levels for the majority of the frequency channels (in general the occupancy factor is much smaller than one). In the Ψ c curves, the detected channels are plotted with colours different from the not-detected channels. Therefore no separate additional sensitivity plot is added for every measurement, however, a generic sensitivity curve is added based on equation (18). Ψ (db(wm 2 Hz 1 )) 140 150 160 170 180 190 200 210 SKA monitoring 2010/2011, power flux spectra, corrected, Virtual Site data, Ψ: baseline (noise) + RF baseline corrected data, Ψ c 1σ base line detection flag 220 230 240 0 200 400 600 800 1000 1200 1400 1600 1800 2000 Figure 8: Power flux density Ψ in db(w m 2 Hz 1 ), obtained for two polarizations and four pointings. For each of the nine bands eight curves are plotted. Green (upper) set of curves are calibrated observed data, the blue set of curves are the estimated flux rms levels using a N σ σ threshold level (N σ = 6). The red curve is the estimated 1σ level. The small 3 db spikes on the lower black horizontal line at -235 db(w m 2 Hz 1 ) (the detection flag) indicate detected interference. 4.8 Occupancy statistics The detection data computed from (20) averaged over N sp consecutive spectra is plotted as a function of frequency. The occupancy numbers are the fraction of time in which an interferer was detected given a 30 khz wide channel. In addition to these spectra, bar charts are also included in which the occupancy statistics in wider frequency bands are computed. For the bar charts the frequency intervals used are the same as for mode two in the 2005/2006 campaign [2], cf table 2. The 2005/2006 frequency ranges were not specified for frequencies below 150 MHz, so the interval 80-150 MHz was added. Also, the 2006/2007 interval starting at 1723 MHz was truncated at 2000 MHz. 23

WGRM frequency ranges (MHz) 80-150 608-614 150-153 614-1000 153-322 1000-1370 322-329 1370-1427 329-406 1427-1606 406-410 1606-1723 410-608 1723-2000 Table 2: List of frequencies as specified in [2] for mode two. 4.9 Time statistics and averaged power statistics Time statistics As the sheer number of data makes it impractical to display time-frequency spectra over the full time and spectral resolutions the reporting includes percentile values over time. This kind of plot gives an impression of the time-behavior of transmitters. Percentile values of 50 (median), 90 and 99 are used and plotted in each figure. This kind of display is used for flux density data; in other cases (calibration measurement results) other percentile values are used, e.g. 25, 50 and 75 percentile values. An overview of occupancy statistics will be made available in the summary annex report. Power statistics Observed power flux densities Ψ sys, averaged over frequency intervals as specified in table 2 (as numbers or in the form of a particular figure of merit (FOM)) is outside the scope of the measurement campaign (cf. [1]) and are not included in the measurement reports. Interpretation of these power (FOM) data, if/when available, for example in terms of required ADC bits is complicated and should be done with great care as misinterpretation is very easy. There are many aspects which should not be forgotten when interpreting, for example: Satellite signals usually do not enter via the main lobe of the receiving antenna, therefore an ( 6dB) fudge factor is needed (no corrections were made in the signal processing for this effect) A power FOM for the core probably will be different from the FOM for remote sites, and it is not clear how these should be weighted There is day-night variation which is not sampled well temporally Power summation for a SKA receiver band will depend obviously on the selected band, and these 24

have not yet been fixed Transmitter duty cycle and the integration time of the digital receiver in most cases do not match, this will lead to underestimates if uncorrected The spectrum environment is subject to change 4.10 Figure compression The spectrum monitoring reports use eps formatted figures. This format is a vector format. Without compression the figure data size would be of the order of a few Mbyte or larger. This would lead to reports with a file size well over 100 Mbyte. For this reason an additional compression step (ranging between 10 and 100) is introduced when eps figures are produced. The compression is applied as a last step: all data processing, calculation of RFI metrics and calibration is done with full spectral resolution. Figure 9 and shows the effect of the applied filter. The filter computes for a specified interval (between 10 and 100 samples) the minimum and maximum value and plots both. As long as the pixel size of a computer screen or the resolution in a document is larger than the compression interval, the impression/presentation of the data is not affected. The fine structure related to the compression, as is visible in figure 9, is not visible when the spectrum is zoomed-out. The compression is applied to fixed length intervals, not on running intervals. 25

150 SKA monitoring 2010/2011, power flux spectra, virtual site 160 170 Ψ (dbwm 2 Hz 1 ) 180 190 200 210 220 130 131 132 133 134 135 136 137 138 139 140 150 SKA monitoring 2010/2011, power flux spectra, virtual site 160 170 Ψ (dbwm 2 Hz 1 ) 180 190 200 210 220 130 131 132 133 134 135 136 137 138 139 140 Figure 9: Calibrated flux data, compressed using 10 point compression (upper figure) and without compression (lower figure). 26

5 Output description and measurement result summary 5.1 FS mode results The FS mode graphical output spectra are given for the actual calibrated spectra and sensitivity curves, and for calibration spectra. 5.1.1 Spectrum measurement spectra 1. Calibrated and baseline-corrected power flux density. Calibrated and baseline-corrected (mean) power flux density curves Ψ c in db(w m 2 Hz 1 ) obtained for two polarizations (H,V) and four pointings (N,E,S,W): for each of the nine bands 2x4 curves are plotted. The N σ σ detection flag curve (N σ =6) is plotted in black at -235 db(w m 2 Hz 1 ), detections are indicated by 3 db vertical spikes. 2. Sensitivity. Sensitivity Ψ in db(w m 2 Hz 1 ) estimates are obtained based on calibrated T sys and on T rec curves, for two polarizations and four pointings; in total two times eight curves are plotted. The depicted estimations are 1σ values, the actual detection level used is N σ σ (N σ = 6). The RA769 values for line and continuum sensitivities are included for reference; the values are connected via dotted lines. 5.1.2 Calibration spectra 1. Raw spectra. Raw measurement data (db) are presented for noise source on, noise source off and for spectrum monitoring measurements, obtained for two polarizations and four pointings. In total eight curves are plotted for each band. For reference the entire passbands are plotted, not just the in-band part. The bands are plotted with multiple offsets of 10 db (increasing with bandnumber) to more clearly separate the bands for visual inspection. 2. Gain. Calibrated gain curves G in db, obtained for two polarizations and four pointings are presented. In total eight curves are plotted for each of the nine bands. Only the in-band sections are depicted; the jumps are band transitions. 3. Receiver temperature. Receiver temperature T rec in K, obtained for two polarizations and four pointings are presented. In total eight curves are plotted for each of the nine bands. Only the in-band sections are depicted; the jumps are band transitions. 4. System temperature. System temperature T sys in K, obtained for two polarizations and four pointings are presented. In total eight curves are plotted for each of the nine bands. Only the in-band sections are depicted. 27

5. Matched power. Matched power in db for a noise source-off measurement and for a measurement in which the antenna is connected, obtained for two polarizations and four pointings, are presented. In total eight curves are depicted for each of the nine bands. The spikes in the measurement data are (obviously) caused by external RFI. 6. Matched power difference. Matched power difference curves in db for a noise source-off measurement and a measurement in which the antenna is connected, obtained for two polarizations and four pointings, are presented. In total eight curves are depicted for each of the nine bands. The spikes in the data are (obviously) caused by external RFI. 7. Calibrated spectra and detection level. Power flux density Ψ in db(w m 2 Hz 1 ), obtained for two polarizations and four pointings are presented. For each of the nine bands eight curves are plotted. Green (upper) set of curves are calibrated observed data, the blue set of curves are the estimated flux rms levels using a N σ σ threshold level. The red curve is the estimated 1σ level. The small 3 db spikes on the lower horizontal line at -235 db(w m 2 Hz 1 ) (the detection flag) indicate detected interference. 8. Data error table. A table is presented with numbers indicating whether a particular band is affected by saturation of by a frequency offset error (1) or not (0). The header B denotes bandnumber, A denotes azimuth (N,E,S,W), and P denotes polarization (H,V). The next four columns ϵ s,m, ϵ f,m, ϵ f,h, and ϵ f,c indicate respectively: saturation flag for antenna measurements, frequency error for antenna measurements, noise diode hot and noise diode cold error flag. The last column ϵ all indicates the aggregate error flag. 5.2 HS mode results The HS mode graphical output spectra are given for the actual calibrated spectra and sensitivity curves, and for calibration spectra. 5.2.1 Spectrum measurement spectra 1. Calibrated and baseline-corrected power flux density. Baseline corrected power flux density Ψ c in db(w m 2 Hz 1 ), obtained for one polarization and one pointing using N σ = 6 are given. For each of the nine bands three percentile curves are depicted: 50 percentile in blue, 90 percentile in green, and 99 percentile in red. The N σ σ detection flag curve (N σ =6) for all data (not just the depicted percentile data) is plotted in black at -255 db(w m 2 Hz 1 ), detections are indicated by 3 db vertical spikes. 2. Sensitivity. Sensitivity estimates Ψ in db(w m 2 Hz 1 ) based on calibrated T sys and on T rec curves are given. The depicted estimations are 1σ values, the actual detection level used is N σ σ. The RA769 values for line and continuum sensitivities are included for reference. 28

3. Occupancy spectra. Occupancy spectra, obtained for one polarization and one pointing, are given. The number of one-minute integrated spectra used for the occupancy estimate is: 120 minus the number of erroneous spectra (cf. N flag parameter in the figure shown, and section A6 of the measurement report). The 1σ detection levels are displayed in section A5 of the measurement report; the actual detection level used is N σ = 6. 5.2.2 Calibration spectra 1. Raw measurement data. Raw measurement median power data (db) for noise source on, noise source off and for spectrum monitoring measurements obtained for one polarization and one pointing. The median values are computed using 120 one-minute integrated spectra. 2. Gain. Calibrated gain curves G in db, obtained for one polarization and one pointing. Only the in-band sections are depicted; the jumps are band transitions. Plotted are the 25, 50, and 75 percentile curves, respectively. 3. Receiver temperature. Receiver temperature T rec in K, obtained for one polarization and one pointing. Only the in-band sections are depicted; the jumps are band transitions, the spikes are self-genererated interference. Plotted are the 25, 50, and 75 percentile curves, respectively. 4. System temperature. System temperature T sys in K, obtained for one polarization and one pointing. Only the in-band sections are depicted. The pikes in the spectrum are caused by RFI. Plotted are the 25, 50, and 75 percentile curves, respectively. 5. Matched power difference. Matched power difference (median value) in db for a noise source-off measurement and a measurement in which the antenna is connected, obtained for one polarization and one pointing. The spikes in the data are caused by RFI. 6. Baseline and calibrated spectra. Power flux density Ψ in db(w m 2 Hz 1 ), obtained for one polarization and one pointing. The upper green curve is the median of the calibrated observed data, the blue curve is the median of the baseline corrected data using an N σ σ threshold level with N σ = 6. The red curve is the estimated 1σ level. The N σ σ detection flag curve (N σ =6) for all data (not just the depicted percentile data) is plotted in black at -255 db(w m 2 Hz 1 ), detections are indicated by 3 db vertical spikes. 7. Data error flag. Data flag for data affected either by a frequency offset error or by saturation. The title includes the number of flagged measurements for each of the nine bands. 5.3 RM mode results To be written. 29

5.4 MH mode results To be written. 5.5 TM mode results To be written. 30

References [1] B.J. Boyle, P.J. Diamond, B. Fanaroff, M.A.Garrett, and R.T. Schilizzi. Agreement on Radio Frequency Interference (RFI) Monitoring 2008-2011. Technical report, April 2008. [2] R.Ambrosini et al. RFI measurement protocol for candidate SKA sites. Technical report, SKA, S.Ellingson ed., May, 2003. [3] S.W. Elingson et al. et al. et al. et al. Assessment of Radio Frequency Interference at Candidate SKA Sites. Technical report, RFI Assessment Task Force (Limited Access), May 16, 2006. [4] ITU. ITU-R-RA769. Technical report, ITU, 2005. [5] John D. Kraus. Radio Astronomy. Cygnus-Quasar Books, Powell (Ohio), USA, second edition, 1988. [6] R.P. Millenaar. SKA Site Spectrum Monitoring System Technical Description. Technical Report RP-108, ASTRON, April 2006. [7] R.P. Millenaar. SSSM System Design Considerations. Technical Report RP-13, ASTRON, February 21 2006. [8] R.P. Millenaar, A.J. Boonstra, R. Beresford, and A. Tiplady. System description SKA RFI measurement instrumentation. Technical report, SPDO, 2011. 31

Filename Mode Polarization Azimuth SKAmon.M1.C*.FS.pdf FS H,V N,E,S,W SKAmon.M3.C*.FS.pdf FS H,V N,E,S,W SKAmon.M5.C*.FS.pdf FS H,V N,E,S,W SKAmon.M7.C*.FS.pdf FS H,V N,E,S,W SKAmon.M9.C*.FS.pdf FS H,V N,E,S,W SKAmon.M11.C*.FS.pdf FS H,V N,E,S,W SKAmon.M13.C*.FS.pdf FS H,V N,E,S,W SKAmon.M15.C*.FS.pdf FS H,V N,E,S,W SKAmon.M2.C*.HS.pdf HS H N SKAmon.M4.C*.HS.pdf HS H E SKAmon.M6.C*.HS.pdf HS H S SKAmon.M8.C*.HS.pdf HS H W SKAmon.M10.C*.HS.pdf HS V N SKAmon.M12.C*.HS.pdf HS V E SKAmon.M14.C*.HS.pdf HS V S SKAmon.M16.C*.HS.pdf HS V W Table 3: List of SKA monitoring measurement reports for the FS and the HS modi. Appendix A. Measurement report list Table 3 shows the filenames of the measurement reports produced. The table also lists the measurement mode (FS: Fast Scanning mode, HS: High Sensitivity mode), the polarization (H: horizontal, V: vertical), and azimuth (N: north, E: east, S: south, W: west). 32

Service / system Service / system 87.5-108 FM radio 1381-1381 GPS L3 150.05-153 RAS (1) 1330-1400 RAS (3) 137-138 METEOSAT 1400-1427 RAS (1) 174.0-230 Band III 1470-1480 AFRISTAR 245-270 FLTSATCOM 1525-1559 INMARSAT 322-328.6 RAS (1) 1570.42-1580.42 GPS L1 406.1-410.0 RAS (1) 1598.0625-1625.5 GLONASS 470.0-862 Band IV/V 1610.6-1613.8 RAS (1) 608.0-614.0 RAS (1a) 1616-1625.5 IRIDIUM 880.0-960 GSM900 1660-1670 RAS (1) 960.0-1164 DME 1710.1-1879.0 GSM1800 1226.6-1229.6 GPS L2 1718.8-1722.2 RAS Table 4: Incomplete list of services and allocations in the bands relevant to the SKA monitoring 2010/2011 campaign. This table is included for ease of reference when inspecting spectrum monitoring graphs in this document and its annexes. Note: (1) primary allocation, (1a) primary allocation in region 2, (2) secondary allocation, (3) notification of use. Boldface bands indicate some of the satellite bands present. Appendix B. Allocations and services Table 4 shows a list of selected services and bands for ease of reference when inspecting spectrum SKA monitoring figures. This list contains only a few services which may appear in monitoring spectra; this list is not intended to be complete. 33

Appendix C. Component calibration data For reference this appendix shows calibration data of components used: their gains, attenuation factors and noise figures. Appendix C1. Antenna gain This section shows a typical gain curve of the HL033 antenna in figure 10 (upper figure) as given by Rhode & Schwarz in the antenna fact sheet (http://www2.rohde-schwarz.com/). As it is expected that the frequency dependent gain variation between different antennas of the same type is of the order of 0.5 db, the gain curve used in the data processing can be a smoothed curve, as presented in the same figure, below. 7 Component calibration data, HL033, specified frequency range: 80 2000 MHz 6.8 gain (dbi) 6.6 6.4 6.2 6 10 2 10 3 Figure 10: Gain curves of the HL033 antenna, a typical curve as specified by Rhode & Scwarz (upper figure) and a smoothed curve used in the data processing (lower figure) 34

Appendix C2. LNA and noise diode properties This section shows the measured LNA gain in figure 11 and the Excess Noise Ratio (ENR) of the noise diode in figure 12 as specified in the noise diode fact sheet. The noise diode is specified in a much wider freqeuncy range than is used for the SKA monitoring. As only three data points are available in the SKA monitoring range, intermediate points are linearly interpolated. The expected deviations from the true values are of the order of 0.05 db or less. 51 Component calibration data, LNA4, specified frequency range: 70 2000 MHz gain (db) 50.5 50 49.5 49 48.5 48 47.5 47 10 2 10 3 Figure 11: Component calibration data, LNA AFS4 15.8 Component calibration data, noise diode, specified frequency range: 10 18000 MHz 15.7 ENR (db) 15.6 15.5 15.4 15.3 15.2 10 2 10 3 Figure 12: Component calibration data, noise diode ENR (db) 35

Appendix C3. Cable and switch attenuation This section shows the attenuation of the cables and of the switches in the system in figures 13, 14, 15, 16, 17, and 18. Component calibration data, cable1, specified frequency range: 70 2000 MHz 0.1 0.2 0.3 gain (db) 0.4 0.5 0.6 0.7 0.8 0.9 10 2 10 3 Figure 13: Component calibration data, cable one gain (db) 0 0.2 0.4 0.6 0.8 1 1.2 1.4 Component calibration data, cable2, specified frequency range: 70 2000 MHz 10 2 10 3 Figure 14: Component calibration data, cable two 36

0 Component calibration data, cable3, specified frequency range: 70 2000 MHz 0.5 1 gain (db) 1.5 2 2.5 3 10 2 10 3 Figure 15: Component calibration data, cable three Component calibration data, cable4, specified frequency range: 70 2000 MHz gain (db) 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 10 2 10 3 Figure 16: Component calibration data, cable four 37

0 Component calibration data, switch1, specified frequency range: 70 2000 MHz 0.05 0.1 gain (db) 0.15 0.2 0.25 0.3 0.35 10 2 10 3 Figure 17: Component calibration data, SPDT switch, setting one 0 Component calibration data, switch2, specified frequency range: 70 2000 MHz 0.05 0.1 gain (db) 0.15 0.2 0.25 0.3 0.35 10 2 10 3 Figure 18: Component calibration data, SPDT switch, setting two 38