See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/286421148

RFI mitigation techniques in radio astronomy

Conference Paper · July 2014

DOI: 10.1109/IGARSS.2014.6946399

CITATIONS READS
13 262

2 authors:

John Ford Kaushal Buch

The University of Arizona National Centre For Radio Astrophysics - Tata Institute Of Fundamental Research
43 PUBLICATIONS   309 CITATIONS    36 PUBLICATIONS   101 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

FPGA-based broad band beamformer for FPA View project

Digital noise source development View project

All content following this page was uploaded by Kaushal Buch on 19 May 2017.

The user has requested enhancement of the downloaded file.

 RFI MITIGATION TECHNIQUES IN RADIO ASTRONOMY

John M. Ford Kaushal D. Buch

National Radio Astronomy Observatory Giant Metrewave Radio Telescope

Green Bank, WV 24944, USA NCRA-TIFR, Khodad, Pune, 410504, India

ABSTRACT spurious RFI from several manmade sources contaminates the

Radio Frequency Interference (RFI) is a growing concern spectrum allocated for radio astronomy observations.
for the highly sensitive radio receivers of ground based radio The ITU-R RA.2126 defines the level of RFI as detrimen-
telescopes. RFI reduces the sensitivity of a radio telescope tal when its presence introduces more than a 10% RMS er-
and produces artifacts in the observed data. Over the years, ror in the flux measurement. RFI is classified into two types
several techniques have been proposed and implemented for based on the effects that it produces in the spectrum of the
mitigating RFI, either by controlling and eliminating sources received signal. Broadband RFI is a noise-like signal that
of RFI around the radio telescope or its removal at differ- causes an increase in the power level of the entire spectrum,
ent locations in the receiver chain. This paper examines and is usually generated by temporally impulsive events such
regulatory and management efforts as well as technological as sparking and corona discharges in high power transmis-
approaches employed in mitigating radio frequency interfer- sion lines, automobile ignitions and switching of inductive
ence in radio astronomy observations. Also, the techniques loads. They are almost always unintentional signals. Nar-
currently under development and those proposed for the up- rowband RFI is the occurrence of an unwanted discrete fre-
coming wide bandwidth and highly sensitive radio telescopes quency or a band of frequencies affecting only a narrow part
are described. of the receiver’s bandpass. Narrowband RFI is often produced
by communication broadcast transmitters for TV, commercial
Index Terms— Radio Frequency Interference, Radio As- and short distance radio services, aircraft and satellites.
tronomy, Radio Telescope As a measure towards minimizing the loss of astronomical
data to RFI, several mitigation techniques have been proposed
1. INTRODUCTION over the years. These range from establishing radio quiet
zones around the radio telescope to using advancements in
Radio telescopes detect faint signals from celestial sources technology to develop techniques that can excise RFI early in
emitting electromagnetic radiation having wavelengths rang- the receiver chain. This paper provides a summary of RFI mit-
ing from several meters to sub-millimeters. The strength igation techniques with an emphasis on the wide-bandwidth
of the signal received by the telescope receiver is very systems.
weak with typical values ranging from −150dBW m−2 to
−220dBW m−2 . For instance, the strongest radio source at
300 MHz has a power spectral density of about 105 Jansky 2. RADIO TELESCOPE RECEIVER AND THE
(1 Jansky = 10−26 W m−2 Hz −1 ). This requires the radio SIGNAL MODEL
telescope receiver system to be more sensitive than other
communication receivers by a factor of 50 to 60 dB. The in- The radio telescope receiver is a radiometer measuring the
cident radio waves from the faint and distant celestial sources received power. The signal received by the antenna(s) is am-
are detected using a single large aperture parabolic reflector plified using a low-noise amplifier, often cryogenically cooled
or a synthesized aperture with high directive gain. to reduce thermal noise. A super-heterodyne radio telescope
In order to achieve the desired signal-to-noise ratio, ra- receiver uses frequency heterodyning for converting the ra-
dio telescopes employ large parabolic reflectors, increasing dio frequency (RF) signals to an intermediate frequency (IF)
the collecting area and hence the received power, leading to for further amplification and processing. The IF signals are
better G/Tsys . Since the astronomical signal has properties further down-converted to base-band for digitization and are
similar to that of Gaussian noise, the receiver sensitivity in- processed in the digital domain thereafter. Some contem-
creases with the receiver bandwidth. The need for increased porary receivers carry out direct RF digitization and down-
bandwidth to achieve desired sensitivity necessitates the radio conversion. Following the base-band conversion and digitiza-
telescope receiver to observe in the bands outside the ones al- tion, a single dish telescope measures the total power in the
located for radio astronomy observations. Apart from this, the observing band whereas array telescopes measure correlated
power between pairs of antennas (baselines), in a manner sim- RFI mitigation techniques is shown in Fig. 1. Details of these
ilar to a correlation radiometer. are provided in the subsequent subsections.
In the time-domain, a signal with RFI as observed by a
radio telescope can be given as [1];

x(t) = xsrc (t) + xsys (t) + xRF I (t) (1)

In equation 1, xsrc (t) is contribution due to the astronomical

source (desired signal), xsys (t) is the system noise (an unde-
sired signal) and xRF I (t) is the RFI signal (an undesired sig-
nal). System noise is a combination of sky background noise
and receiver noise. Both astronomical signal and system noise
are zero mean random signals possessing a Gaussian distribu-
tion. RFI usually has a non-Gaussian distribution [1].
Terrestrial RFI usually enters the highly directive anten-
nas of the radio telescope through the sidelobes. However, it Fig. 1. Basic Classification of RFI Mitigation Techniques
may also enter through the main lobe during observations at
lower elevation angles or through multi-path propagation or
by reflections from the antenna structure. Additional power
3.1. Regulatory methods
introduced by RFI can be seen as increase in noise tempera-
ture of the receiver. Also, due to its non-random nature, the To prevent pollution of the radio spectrum, a radio quiet zone
fluctuations in the received power do not reduce at the theo- spanning several tens of kilometers in diameter may be es-
retical rate given by the radiometer equation as (B ∗ T )−0.5 tablished around the radio telescope in coordination with the
where B is the receiver bandwidth and T is the integration domestic spectrum allocation authorities. For instance, in
time. Strong RFI drives the receiver into its non-linear region the US, the National Radio Quiet Zone sets aside an area of
and results in contamination of the resultant power spectrum, 13,000 square miles (33,670 square kilometers) where instal-
which results in undesirable effects in the radio images. lation of fixed transmitters is evaluated for impact on the ra-
For an antenna array, where signals are received simul- dio astronomy facility at Green Bank. The spectrum in and
taneously from different antenna elements and processed around the observatory is regularly monitored by the obser-
as baselines, the component of the astronomical source vatory authorities to check the power levels at the frequency
signal received by each of the antennas is mutually corre- bands of interest.
lated. The RFI, being from a different line of sight, is not Construction of future radio telescopes is planned in re-
as well correlated as the astronomical signal. This reduces mote geographical regions either in the deserts or on the
the effects of RFI in the astronomical data. As the baseline mountains [5]. This provides an environment free from ter-
length increases, the correlation of terrestrial RFI source is restrial RFI, enabling the radio telescopes to achieve their
reduced and in the case of Very Long Baseline Interferom- expected sensitivity.
etry (VLBI), the observations are mostly affected by extra- To prevent observatory-generated interference, the build-
terrestrial (space-based) sources of RFI (satellites, airplanes ing needs to have appropriate shielding to prevent the RFI
etc.) rather than terrestrial sources [2]. ITU specifies that at from various RF components, high power electrical equip-
metre wavelengths, the threshold level of interference should ment and digital devices. Devices in regular use such as com-
be lower than should be less than −190dBW m−2 in radio puters and networking equipment possess broadband as well
astronomy bands [3]. as distinct spectral line emissions. All these systems and lab-
oratories must be shielded to attenuate the emissions. In ad-
3. RFI MITIGATION TECHNIQUES dition to shielding, it is a usual practice to use instruments
which have minimal levels of leakage at the observatory site.
The spectrum observed by a radio telescope may be either The forgoing are the initial steps towards RFI mitigation in a
collocated or shared with various other communication ser- radio telescope. Advanced signal processing techniques have
vices, which can severely affect the observations. To mitigate been proposed and used for locating sources of RFI around
this source of RFI, the radio telescope site is usually built in the observatory.
an area which has sparse population density and is devoid of Over time, due to urbanization and industrialization of the
industrial development. The majority of RFI that is found in once remote observatory site, the radio observatory starts fac-
such remote sites is due to extra-terrestrial sources or radio ing increasing levels of RFI, which is the situation with most
propagation through the ionosphere and troposcatter from re- existing radio telescopes. Where regulatory efforts fail to pro-
motely located transmitters [4]. The basic classification of vide adequate protection against RFI, it becomes necessary to
implement techniques for RFI mitigation at various stages in corrupted by the RFI. It can be implemented in either the tem-
the receiver chain and in the post-processing of data. poral or spectral domains. RFI excision in the initial stages of
digital signal processing at the higher time resolution avail-
3.2. Technical methods able there reduces the overall loss of data [6].
Time domain excision using robust estimates of vari-
The central approach in technical methods is to minimize the ance [7] is used for mitigating impulsive broadband RFI and
amount of desired signal that is lost while mitigating the in- has been implemented using the Median Absolute Devia-
terfering signal. Different types of observations are sensitive tion (MAD) estimator on an FPGA for real-time removal of
to different interference types, and require different mitiga- power-line RFI in the Giant Metrewave Radio Telescope’s
tion strategies. For instance, pulsar search is normally carried (GMRT) wideband backend [8]. Excision of time-domain
out over a wide bandwidth at low frequencies where the worst RFI resulting from high voltage power-line transmission lines
RFI is present. Pulsar searches are sensitive to periodic broad- has also been effectively demonstrated using the MAD esti-
band interference sources. Similarly, sensitive continuum and mator [9] in the offline mode of the GMRT software backend
spectral line observations are disturbed by narrow band RFI (GSB) implemented using a CPU cluster. This robust esti-
that is present in the band of interest, regardless of the tem- mation technique is effective when the RFI is much stronger
poral qualities of the interference. RFI mitigation is carried than the astronomical signal. Upon computation of the ex-
out at different stages in the receiver chain (i.e. RF, IF and cision threshold through a robust estimator, the RFI outside
digital). this specified threshold is selectively removed (or replaced
with a particular value or noise). Figure 2 shows the input
3.2.1. Mitigation in the RF & Analog domains time series with impulsive RFI (in blue) and the filtered time
series (in red) resulting from real-time MAD-based filtering
Antenna patterns of radio telescopes are usually designed to
and replacement with Gaussian noise at 3σ threshold on an
be highly directive with very low level of sidelobes which
FPGA operating at 200 MHz and 8-bit data precision. The
provides first level of protection from RFI. The front-end RF
design [8] is undergoing detailed tests along with the wide-
electronics of the receiver is designed using high dynamic
band backend chain at GMRT. A current area of research
range devices to prevent strong RFI from driving the receiver
into the non-linear operating region. In most radio telescope
receivers, the 1 dB compression and 3rd order intercept points
throughout the system are designed to resist the effects of
gain compression given the RFI environment. For rejection of
known strong narrowband RFI signals like television broad-
cast transmitters and GSM transmitters, notch filters early in
the signal chain provide a typical stop-band attenuation of
60dB. Another option for keeping the analog components in
the linear region is the use of automatic gain control, which
is a closed loop control of gain proportional to the strength of
the signal.
Fig. 2. Results of MAD-based RFI Excision Implemented on
3.2.2. Mitigation in the digital domain FPGA
The algorithms in the digital domain are either implemented
in real-time or as a part of the offline processing chain. With and development in the area of real-time RFI excision is the
the availability of high speed signal processing using the Field study and characterization of the effects of RFI filtering on
Programmable Gate Array (FPGA) and Graphical Process- the end result and the propagation of the flagging/blanking
ing Unit (GPU) platforms, real-time implementations of tem- information from the real-time excision system to the final
poral, spectral and spatial RFI filtering is being carried out. output for appropriate weighting by the post-processing soft-
ITU-R RA.2126 classifies the real-time RFI mitigation tech- ware. Robust estimation techniques can also be applied to the
niques into three categories, namely, Excision, Cancellation real-time data in the spectral domain for excising narrowband
and Anti-coincidence. In general, the number of computa- RFI.
tions required for these algorithms operating on data from Another technique for excision is the detection of abnor-
each individual antenna scales as N (N is the number of an- mal signals in the data set. RFI contaminates the normally
tennas) whereas those operating on baselines scale as (N ∗ Gaussian distribution of the astronomical signal. A Kurtosis
(N − 1))/2. estimator can be used on time or frequency domain data for
Excision is the technique of detecting RFI from astronom- detecting presence of RFI in the signal. Spectral kurtosis has
ical data and mitigating it by blanking or clipping the data been shown to be effective in detecting and mitigating nar-
 rowband RFI and has been implemented in real-time on an 5. REFERENCES
FPGA [10].
Spatial excision involves forming a null in the beam pat- [1] PA Fridman and WA Baan, “RFI mitigation
tern towards the source of RFI. The method is suitable for an methods in radio astronomy,” ASTRONOMY AND
antenna array with steerable beams. Adaptive beamforming ASTROPHYSICS-BERLIN-, vol. 378, no. 1, pp. 327–
for steering the beam and null forming in the direction of RFI 344, 2001.
has shown to be useful particular in case of phased-array feeds [2] P Crane and L Hillenbrand, “Estimating harmful levels
on single-dish telescopes. of radio frequency interference,” in IAU Colloq. 112:
Cancellation involves subtracting RFI from the astronom- Light Pollution, Radio Interference, and Space Debris,
ical signal and is most effective when either a strong signal is 1991, vol. 17, p. 258.
available as a reference or a good model exists of the sig-
nal [11, 12]. The process of cancellation involves detection [3] ITU-RA Series, “Protection criteria used for
and estimation of RFI, synthesis of the RFI signal (without radio astronomical measurements,” 2003,
noise) and subtraction of this synthesized RFI signal from the http://www.itu.int/rec/R-REC-RA.769/en.
RFI corrupted data. The interfering signal is observed using a [4] Swarup G., “Surveys of radio frequency interference
separate directional antenna receiving the RFI signal. This re- (RFI) at the GMRT site from terrestial transmitters,”
ceived reference signal is then input to the adaptive filter for http://ncralib1.ncra.tifr.res.in:8080/jspui/bitstream/
optimizing the filter coefficients to mitigate the RFI signal. 2301/137/1/R00191.pdf.
This approach is better than excision for removal of low level
RFI. Cancellation implemented before correlation can poten- [5] Jacob WM Baars, Larry R D’Addario, and A Richard
tially recover most of the data from a contaminated sample. Thompson, “Radio astronomy in the early twenty-first
century,” Proceedings of the IEEE, vol. 97, no. 8, pp.
1377–1381, 2009.
3.2.3. Post-processing (Offline) RFI excision
[6] JR Fisher, “RFI and how to deal with it,” in Single-Dish
RFI excision in the post processing domain is carried out of- Radio Astronomy: Techniques and Applications, 2002,
fline on recorded data (the integrated spectrogram). It is the vol. 278, pp. 433–445.
first step before actual astronomical processing is done. The
data is either manually flagged through visual inspection or [7] PA Fridman, “Statistically stable estimates of vari-
is processed through an automated pipeline. Manual flagging ance in radio-astronomy observations as tools for radio-
can be carried out in standard astronomical imaging software. frequency interference mitigation,” The Astronomical
Automated RFI flagging pipelines typically use robust statis- Journal, vol. 135, no. 5, pp. 1810, 2008.
tical techniques which flag or blank the data based on user [8] K Buch, “Impulsive RFI mitigation: CASPER li-
defined parameters, and also assigns appropriate weights to brary block,” https://casper.berkeley.edu/wiki/ Impul-
the flagged data proportional to the amount of flagging in a sive RFI Excision: CASPER Library Block.
given data set. Offline RFI processing is a preferred tech-
nique, but with the increasing data volume from the higher [9] Jayanta Roy, Yashwant Gupta, Ue-Li Pen, Jeffrey B Pe-
spectral and temporal resolution available, manual interven- terson, Sanjay Kudale, and Jitendra Kodilkar, “A real-
tion is cumbersome and causes larger data loss compared to time software backend for the GMRT,” Experimental
its real-time counterpart operating on an integrated data set. Astronomy, vol. 28, no. 1, pp. 25–60, 2010.
[10] DE Gary, GM Nita, X Wang, and H Ge, “Spectral do-
4. CONCLUSION main algorithms for rfi excision in real time,” Proceed-
ings of the URSI General Assembly, 2008.
Ground based radio telescopes remain vulnerable to RFI, de-
[11] Weizhen Dong, Brian D Jeffs, and J Richard Fisher, “A
spite being located in sparsely populated areas. Technological
Kalman-tracker-based Bayesian detector for radar inter-
advancements have equipped the radio telescopes to achieve
ference in radio astronomy,” in Acoustics, Speech, and
better sensitivity by implementing techniques for mitigating
Signal Processing, 2005. Proceedings.(ICASSP’05).
RFI in the receiver chain and in real-time baseband signal
IEEE International Conference on. IEEE, 2005, vol. 4,
processing. The paper summarized different approaches to
pp. iv–657.
RFI mitigation with respect to their effectiveness under vary-
ing levels and types of RFI. A glimpse of the current state [12] Steven W Ellingson, John D Bunton, and Jon F Bell,
of research and development was described. Some of these “Removal of the GLONASS C/A signal from OH spec-
techniques would also find applications in the areas of digi- tral line observations using a parametric modeling tech-
tal communications and passive microwave radiometry where nique,” The Astrophysical Journal Supplement Series,
RFI presents similar problems. vol. 135, no. 1, pp. 87, 2001.

View publication stats