Iridium Satellite Signals: A Case Study in Interference Characterization and Mitigation for Radio Astronomy Observations
aa r X i v : . [ a s t r o - ph . I M ] M a r April 2, 2019 1:19 desh˙lewis˙Iridium˙ms
IRIDIUM SATELLITE SIGNALS: A CASE STUDY IN INTERFERENCECHARACTERIZATION AND MITIGATION FOR RADIO ASTRONOMYOBSERVATIONS
Avinash A. Deshpande , and B. M. Lewis Raman Research Institute, Sadashivanagar P.O., Bangalore 560 080 INDIA NAIC/Arecibo Observatory, HC3 Box 53995, Arecibo, Puerto Rico 00612
Several post-detection approaches to the mitigation of radio-frequency interference (RFI) are compared by ap-plying them to the strong RFI from the Iridium satellites. These provide estimates for the desired signal in thepresence of RFI, by exploiting distinguishing characteristics of the RFI, such as its polarization, statistics, andperiodicity. Our data are dynamic spectra with full Stokes parameters and 1 ms time resolution. Moreover, sincemost man-made RFI is strongly polarized, we use the data to compare its unpolarized component with its StokesI. This approach on its own reduces the RFI intensity by many tens of dBs. A comprehensive approach that alsorecognizes non-Gaussian statistics, and the time and frequency structure inherent in the RFI permits exceedinglyeffective post-detection excision provided full Stokes intensity data are available.
Keywords : methods: data analysis, radio lines: general, radio-frequency interference, techniques: polarimetric
1. INTRODUCTION
Some sources of radio frequency interference (RFI) are inescapable. While radio astronomers can minimizethe effects of many terrestrial sources by placing their telescopes at remote sites, none can escape from RFIgenerated by satellite transmitters, such as those of the Iridium System. So astronomers are now studyinga variety of diverse approaches to mitigating the effects of RFI on their observations (Ellingson 2005, andreferences therein). The pre-detection approaches to RFI excision relate to those applied to time sequenceof signal voltage ; that is before the stage of square-law detection or translation to intensity, such that signalphase information is still available. On the other hand, the post-detection RFI mitigation methods arerelevant to a more commonly available form of data sets of intensity , estimated in general as a function oftime and frequency, that is dynamic spectra with desired temporal and spectral resolutions and spans. Morebroadly, the post-detection data refer to any quantity proportional to average intensity, such as estimates ofcross-correlation of signal voltage sequences from two elements of an interferometer or from two orthogonalpolarization feed antennas, which preserves only the relative phase between the correlated signals.Some of these exploit particular characteristics of specific sources of RFI, such as its location, if froma geostationary satellite, or its polarization. Indeed, since man-made signals are highly polarized, whereasthe inherent nature of most astronomical signals is unpolarized, Deshpande (2005) proposed the use of amitigation technique based on the unpolarized, relatively RFI-free signal. In this paper, we compare andcontrast this approach with several other post-detection approaches to the identification and mitigation ofRFI originating from the Iridium System.
2. OBSERVATIONS
The International Telegraphic Union (ITU) granted the Radio Astronomy Service (RAS) primary statusin the 1610.6-1613.8 MHz band in 1992 to observe the 1612.235 MHz spectral line emission from the Avinash Deshpande [email protected] 1 pril 2, 2019 1:19 desh˙lewis˙Iridium˙ms Deshpande and Lewis hydroxyl molecule (OH). This is typically emitted by OH/IR stars a (see Lewis, Eder & Terzian 1985) asa pair of narrow features, with the allocated band sized to allow for Doppler shifts of the emission, as wellas guard-band separation from ITU Services using adjacent spectrum. One such is the Iridium L-Bandsystem, which presently uses a 1618.85-1626.5 MHz allocation. But this system also produces a comb ofRFI, with a characteristic 333 kHz spacing (often with 8 times finer sub-spacing), extending well beyondits licensed band. The ∼ ∼ a These are Asymptotic Giant Branch (AGB) stars that show strong OH maser emission and are unusually bright in near-infrared (IR). pril 2, 2019 1:19 desh˙lewis˙Iridium˙ms
Mitigation of RFI from IRIDIUM Satellite Signals channels from every possible pairing of spectra: they are produced as cross-correlations between spectra inthe native linears, in I, in the unpolarized flux, I u ( = I − p Q + U + V ), as well as between the twoobserved bandwidths, and as autocorrelations of spectra. After folding 2 minutes of data at the period ofthe Iridium clock cycle, the only Iridium artifacts in our RAS band data are momentary gain compressionepisodes (see Deshpande & Lewis 2005). Hence, it is not surprising that in this band, the statistical measures(such as the average spectrum and the rms values of the noise) during the cycle phases corresponding tothe Iridium RFI are found to be very similar to those for the rest of the cycle (without the RFI).The intensity of the Iridium clock synchronization signal (at 1626 MHz) generates some ringing inpril 2, 2019 1:19 desh˙lewis˙Iridium˙ms Deshpande and Lewis
Fig. 2. Similar to Fig. 1, but for a bandwidth of 3.125 MHz in the RAS band (centered at 1612.5 MHz), which is narrow andso has a correspondingly better spectral resolution, providing a more resolved view of the line features from the star. the adjacent spectrum, despite our nine-level sampling, so our data is always Hanning smoothed, as, forinstance, in Figures 1,2,3. Detailed channel by channel examination of the data, shows that there is stilla noticeable under-correction of data in frequency bins immediately before and after the synchronizationimpulse, and we find that this exaggerates the residuals in comparisons with I u : data in Figures 5 & 6 aretherefore Hanning smoothed twice. The bandpass gain calibration is applied to the linear polarization databefore computing Q & I u from the Stokes I, Q, U, & V of each channel of each spectrum as I u = I − I p (where the polarized component is estimated as I p = p Q + U + V ), which reduces the residuals from I u .
3. MITIGATING
All of the Figures, except Fig.2, use 25 MHz bandwidth data obtained towards the OH/IR star IRAS17256+0504 on 9th September 2004 (azimuth and zenith angle of ∼ ◦ and ∼ ◦ , respectively). Fig. 3shows a typical, 180 ms (two-cycle) sequence of Stokes I spectra with the OH/IR star at ∼ Mitigation of RFI from IRIDIUM Satellite Signals ∼ feature here, which is only seen for ∼
5% of the time, has ∼ b , whichonly discriminates against obviously non-Gaussian components in a series. It is computed from (and maymodify) the temporal sequence of each frequency channel. While the robust mean excises most of thestrong RFI in Fig. 3, though leaving a clear residue in the average spectrum circa 1621 & 1625.5 MHzand a broad, slight one circa 1624.5 MHz, the vertical panel shows that it otherwise tracks I. The residuesfrom the robust mean in the average spectrum in effect arise from integrating up features below the robust b Robust mean and robust rms refer to respectively the arithmetic mean < x > and the standard deviation σ associated witha distribution of an ensemble of values after excluding identifiable outliers from the ensemble. Outliers are the samples withvalues outside the expected range < x > ± γσ , where γ defines the threshold in units of σ on either side of the mean. For anensemble of size N , the γ can be specified such that the probability of value being outside the above range is less than 1/N,for a Gaussian distribution. An iterative procedure refining the ensemble is expected to converge in a few iterations. pril 2, 2019 1:19 desh˙lewis˙Iridium˙ms Deshpande and Lewis
Fig. 4. Dynamic spectrum similar to that in Fig. 3, but now for the unpolarized component I u , i.e. after removing thepolarized contribution from the Stokes I, and thus mostly free of RFI. threshold. On the other hand a similar plot, as in Fig. 4, for the second mitigation approach using I u showsno evident sign of an Iridium signal. So I u over short, one-on-one, time comparisons appears to providebetter mitigation, and its output is improved further by applying a robust mean. Yet the efficacy of I u relieson the estimation of the polarized component I p , which is both over-estimated (due to direct contributionfrom variance of random noise in the Q , U and V on squaring) and least accurate for weak features: indeedthe strongest RFI features are the most accurately excised, as the estimation of the polarized flux is thendone with the best S/N, while that of the weakest is subject to a statistical bias, for which no allowanceis made here. And I u also leaves residuals when the apparent polarization of RFI is reduced as a result ofdepolarization due to averaging of different polarization states (such as while folding), as Fig. 5 shows at ∼ I u , to emerge frompril 2, 2019 1:19 desh˙lewis˙Iridium˙ms Mitigation of RFI from IRIDIUM Satellite Signals I average from the first 40 ms of each Iridium cycle (sea-green), afterlimiting its constituents to band means that are within 3 σ of the overall mean. The band averages for each millisecond of theIridium cycle are shown on the left, with I offset by 0.5 and the robust mean, z , displayed as ( z ( t ) − h z i ) ∗
100 + h z i (so as toamplify the deviations for ready visibility). the noise. Finally the folded spectra of Fig. 5 offer a third approach to mitigation, as portions of theIridium cycle without RFI are readily identified from its left-hand panel. We select “RFI-free” spectra byadditionally requiring that their integrated total power lies within 3 σ of the mean of the set. Figures 5 &6 include RFI-free means.Fig. 6 enables the integrated-up, coherent deviations to be exhibited. It shows that much the largestdeviation occurs in I u on the ascending edge of the bandpass at ∼ Deshpande and Lewis
Fig. 6. The two-minute folded mean intensity spectra in descending order: (1) arithmetic average Stokes I (black); (2) I u (green); (3) robust mean of I (sea-green); (4) robust mean of I u (blue); (5) “RFI-free” mean (violet); (6) I u mean from theRFI-free set (brown). The intensity scale in the top panel is in normalized units with respect to the system temperature T sys .Successive spectra are deliberately offset along vertical axis in multiples of 0.5 units in normalized intensity for providing clearview of their individual spectral variations. The second (bottom) panel shows the differences (3)-(5) in black & (4)-(6) in greenon a magnified scale, in units of expected rms (i.e. standard deviation of noise). to treatment by the robust approach on its own, as its occupancy of its band is both episodic and less than50%. In cases where strong RFI has a much larger fractional occupancy, so there is little uncontaminateddata, the robust mean on its own will fail, whereas that situation is usually greatly improved if I u isappropriate and available.Our analysis has been carried through on a data stream corrected for the polarized flux estimate on thetime-scale of the shortest integrations, which is usually the worst case. This processing could be adjusted toimprove on Hanning smoothing, and by using statistical bias corrections. It could also be better adapted tothe needs of spectral-line observers, by iterating the analysis on data processed for longer integration timesgenerated after synchronously averaging the dynamic spectra from each sub-interval within the Iridiumpril 2, 2019 1:19 desh˙lewis˙Iridium˙ms Mitigation of RFI from IRIDIUM Satellite Signals I u ) -32 -22 -14 -21 -18 -08 -20 -08 -13 I u + robust -32 -28 -38 -27 -28 -15 -33 -18 -23 cycle. On the other hand when only a band-averaged power level measurement is wanted, any source ofnarrow-band RFI can be attenuated by a further 15 dB, which is the factor granted by the square root ofthe number of spectral channels being averaged.In summary, we find that a comprehensive approach which recognizes the inherent time, frequency andpolarization structure of an RFI source allows for its exceedingly effective excision in the post-detectionstage when full Stokes data with suitable time and frequency resolutions are available. Acknowledgments
This work was supported by the Arecibo Observatory, which was then operated by Cornell University onbehalf of the National Science Foundation under a cooperative management agreement.
References
Ellingson, S. W. [2005]
Radio Science
40 (5) , RS5S01.Deshpande, A. A. [2005]
Radio Science
40 (5) , RS5S12.Deshpande, A. A. & Lewis, B. M. [2005] ∼ desh/IRIDIUM Lewis, B. M., Eder, J. & Terzian, Y. [1985],
Nature ,313