A Single Hemispiral Antenna for GNSS Interference Mitigation andDirection Estimation

Cara Yang Kataria, Jennifer Truman Bernhard, and Grace Xingxin GaoUniversity of Illinois at Urbana-Champaign

BIOGRAPHIES

Cara Yang Kataria is a PhD student at the University of Illinois at Urbana-Champaign, where her research includeswork on small reconfigurable antennas and techniques for increased channel security and capacity. Prior to that, she workedas an electrical engineer on mobile, console, and desktop platforms at Apple and Microsoft, specializing in audio systemdesign. She received her BA and B.Engs degrees from Dartmouth College and its Thayer School of Engineering, and herMS degree from the University of Illinois at Urbana-Champaign. She is a member of the IEEE.

Jennifer Truman Bernhard received the BSEE degree from Cornell University in 1988. She received MS and PhDdegrees in Electrical Engineering from Duke University in 1990 and 1994. She is the Donald Biggar Willett Professor in theDepartment of Electrical and Computer Engineering at the University of Illinois at Urbana-Champaign, where she is alsothe Associate Dean for Research and helps run the Electromagnetics Laboratory. She is a Fellow of the IEEE and servedas an elected member of the IEEE Antennas and Propagation Society’s Administrative Committee from 2004-2006.

Grace Xingxin Gao received her BS degree in mechanical engineering and her MS degree in electrical engineering fromTsinghua University, Beijing, China in 2001 and 2003. She received her PhD degree in electrical engineering from StanfordUniversity in 2008. From 2008 to 2012, she was a research associate at Stanford University. Since 2012, she has been anassistant professor in the Aerospace Engineering Department at University of Illinois at Urbana-Champaign. Her researchinterests are systems, signals, control, and robotics. She is a senior member of IEEE and a member of ION.

ABSTRACT

This work presents a single-feed hemispiral antenna that leverages both hardware and software approaches to mitigateinterference from jammers, spoofers, and unintended RF interference. The operation frequency range encompasses mostknown GNSS bands, but characterization specifically targets the L1, L2, and L5 bands for GPS. The diversity in thehemispiral’s radiation pattern enables the receiver to perform DoA checks on incoming narrowband signals, distinguishbetween data from true satellites versus spoofers, and identify potential multipath signals. This added functionality can beimplemented in parallel to a receiver’s regular processing for satellite acquisition and tracking.

I. INTRODUCTION

While most wireless communications channels are more susceptible to interference than their wired counterparts, globalnavigation satellite system (GNSS) signals are particularly weak due to their long transmission paths from orbit, yieldingpower levels way below the noise floor by the time they reach receivers on Earth. Much of the ingenuity in the design ofGNSS receivers lies in their ability to boost these signals back up with processing gains. However, strong radio frequencyinterference (RFI) from jammers can easily overcome a receiver’s ability to acquire satellite signals. Additionally, wideavailability of spread spectrum codes, particularly those for GPS C/A, means that smart spoofers are straightforward toimplement. Such spoofers can introduce errors in location or time and, even worse, lead vehicles to undesired locations [1].

To combat this, many authors propose various flavors of adaptive algorithms to run on phased arrays; a good overviewis given in [2]. Aside from calibrating out antenna “biases” [3], relatively little attention is given to the actual design ofthe antenna element aside from the age-old demands of small and cheap. Ironically, these algorithms need multiple antennainputs for good performance, driving up both the array size and the number of expensive electronic components at the feeds.Another key problem is the assumption of uncorrelated interfering sources, which renders these useless against spoofers orin multipath environments. If the design of antenna hardware is included in the approach, then the added degrees of freedomcan yield a much more efficient solution. Specifically, reconfigurable antennas have been highlighted for their use in directionof arrival (DoA) estimation [4],[5], and [6]. DoA estimation can identify spoofed signals, which typically emanate fromdirections other than their advertised locations in the navigation message. However, these solutions are also implemented onarrays or reconfigurable antennas that rely on array principles (even if many elements are not driven), so the cost, complexity,and overall system size are still major issues. Moreover, the radiation produced is almost always linearly polarized with arelatively narrow beamwidth and not the broad right-hand circularly polarized (RHCP) pattern ideal for GPS receivers.

To address these issues, this work presents a single miniature hemispiral antenna to mitigate interference from bothjammers and spoofers. The operation frequency range will encompass most known GNSS bands, but characterization willspecifically target the L1, L2, and L5 bands for GPS. The whole system only requires a single RF feed and is comprised ofa hemispherical spiral element (roughly λ/3 diameter, λ/7 height at L5), a phase shifter, and a coupler. The antenna system

(a) (b)

Fig. 1: Geometry of the hemispiral with feed simulated in HFSS R©

enables the receiver to perform DoA checks on the incoming wavefronts, distinguish true satellite signals from spoofers,and identify potential multipath signals, all while performing the tasks of acquiring and tracking satellites for navigationdata. A modified multiple signal classification (MUSIC) algorithm that can handle multiple sources is implemented to testthe system in an anechoic chamber. The details of the antenna design and simulation results are presented in Section II. Atechnical overview of the algorithm for DoA and its integration into a GPS receiver are given in Section III. Measurementresults are presented in Section IV, along with analysis and discussion.

II. ANTENNA AND SYSTEM DESIGN

For a single antenna, multiple excitation modes and thus multiple radiation patterns can act as a substitute for the phasediversity provided by multi-element arrays. The well-known spiral antenna produces a broad CP pattern when its arms aredriven 180◦ out of phase due to radiation from the currents that flow in the same direction between neighboring sectionswhen the spiral radius nears λ/(2π). This would be the preferred pattern, to be used during normal operation before anyinterference is detected. Driving the arms in-phase, however, produces opposing currents in the active region, yielding a nullalong the axis of the spiral. Increasing the phase lead or lag between the arms steers this null until the traditional pattern isseen again, generating the desired pattern diversity from a single antenna.

Added functionality in the form of interference mitigation should not come at the cost of degraded performance whenthere is no interference. In addition, increases in size and cost should be kept at a minimum to be viable for applicationsthat would not consider traditional phased arrays for those reasons. As a benchmark, the target size and performance forthis design is compared against a miniature fixed pattern COTS GPS antenna with a 90 mm diameter [7]. The occupied areais in line with the smallest GPS arrays found in the literature today [8], and the final hemispiral design has a considerablysmaller diameter (though slightly more height) than the FRPA-3, a standard miniature fixed pattern L1/L2 military GPSantenna [9].

The spiral baseline is important for producing different patterns electronically, as noted. However, these have knowndrawbacks when compared to a helix antenna, another popular candidate when CP is required. Helices, of course, occupy amuch greater volume due to their height, but can offer higher (unidirectional) gain and better off-boresight axial ratio (AR)than the planar spiral.

It therefore follows that some form of hybrid design may be able to capture the best parts of both geometries. Fromsimulations of general geometries using HFSS R©, it is clear that while a flat spiral can achieve the COTS antenna’s advertisedgain with good margin at all bands, its AR is quite poor at frequencies near L2 and L5. Helix shapes produced fairly stableAR across the frequencies of interest. Several additional windings around the outer diameter were added to the spiral,with the idea of reinforcing the active region (the outer portions at lower bands) with additional turns and reduce reflectedpower from the ends that otherwise interfere. The junction between the windings and the spiral was made gradual throughparametrics and eventually formed the hemispiral’s semi-rounded profile. Figure 1 shows the geometry of the hemispiralalong with its feed, a two-wire quarter-wave transformer from a ring hybrid coupler. The diameter of the antenna is 90 mm,and the height from the ground plane to the top of the dome (at the feed) is about 38 mm.

III. OVERVIEW OF THE ALGORITHM

There is a fair number of known DoA estimation algorithms in the literature, all of which need inputs from an array ofsome kind. The technique chosen in this study is based on MUSIC, which falls into the category of subspace approaches

that generally yield better resolution of sources with less computation burden when compared to more direct methods [10].MUSIC also places no restrictions on the physical geometry of the array as long as element locations are known. Thereare other options with corresponding benefits and drawbacks, like Min-Norm, and a study of performance between differentalgorithms would be interesting as further analysis.

A. MUSIC modified with pattern vectorsFirst introduced over three decades ago, MUSIC decomposes the received signal covariance matrix into a signal and a

noise subspace, which are disjoint [11]. A brief review of the traditional operation is presented here to establish the technicalbackground for the current approach, which is detailed at the end of this section.

As implemented in [11], an array of M elements can resolve M − 1 sources. In matrix form, the response to the inputvector ~X , which represents incident wavefronts from N narrowband sources, is a vector ~Y of length M. This is representedby Equation 1.

~Y = A~X + ~W , (1)

where

A =[~a(θ1) . . . ~a(θN)

]=

a1(θ1) a1(θ2) . . . a1(θN)a2(θ1) a2(θ2) . . . a2(θN)

......

. . ....

aM(θ1) aM(θ2) . . . aM(θN)

The columns of A (“mode” or “steering” vectors) represent the array response to a single source xm in ~X . The term ~Wrepresents measurement or receiver noise and is assumed to be uncorrelated with ~X . The covariance matrix S of the incomingdata is constructed in Equation 2. Eigendecomposition reveals the eigenvectors spanning the signal and noise subspaces,whose columns form the matrices Em and Ew, respectively. Λmin is the diagonal matrix of the (repeated) minimum eigenvalue,which would be 0 in the absence of noise. M minus the number of times λmin appears indicates how many sources exist inthe signal.

S =1M~Y~Y H = A~X~XHAH +EwΛminEw

H

=[Em Ew

]λm . . . 0...

. . ....

0 . . . λmin

[EmH

EwH

](2)

The inverse squared Euclidean distance between the signal subspace and the mode vectors, Equation 3, is the MUSICspectrum as a function of θ (in 2D space), and the N largest peaks represent the estimated sources. This is shown here for ageneral array with polarization diversity, where two orthogonally polarized mode vectors are required for each θ to resolvean incoming wavefront of any polarization [11].

PMU (θ) =1[

~av(θ)H

~ah(θ)H

]EwEw

H [~av(θ) ~ah(θ)] (3)

Given that the algorithm works under no assumptions about the array geometry, the mode vectors may be replaced bylinearly independent functions of θ other than the array response, so long as the incoming signals are actually receivedwith those complex coefficients across θ . It naturally follows that certain radiation patterns from the hemispiral may replacethese mode vectors. Each pattern is used over a sampling period of T to receive the signal, and after M patterns are cycledthrough, a new ~Y must be reconstructed for each previous timestep. The extra step of processing to find the covariancematrix is shown in Equation 4 [12].

S =1T

T

∑t=1

~Y (t)~Y (t)H (4)

The approach taken in this paper is very similar to [12], with a few changes in implementation to allow for multiple sourceswith shifted carrier frequencies, as is always the case for moving transmitters like satellites.

B. Software integrationThe receiver integration details will specifically target GPS, but adapting the approach for other GNSS systems would

be straightforward. During acquisition, a receiver finds the coarse estimate of each available satellite’s Doppler frequencyand code delay, usually via correlation with replicas on parallel channels. It then enters tracking operation, where finerestimates are found via one or more tracking loops, depending on the implementation. Based on satellite availability, the

DoA steps occur before or after acquisition. Figure 2 shows a flowchart for the method, in the context of a general receiverimplementation. M total patterns are used and K satellite PRN codes are found on the incoming signal. Acquisition is donein the first (preferred) pattern state, and the code-wiped carrier signals are saved in M received signal vectors length T , whereT is the number of samples in a period. If there are enough satellites, i.e. K > 4, tracking can immediately commence usingthe data from the preferred pattern (180◦ phase on the feed). As a parallel path, DoA steps corresponding to Equations 4, 2,and 3 are done on each of the K channels, and any channels with DoA error above a certain threshold are deemed spoofedor multipath signals. These are removed from the tracked group and the total remaining number of useful channels afterDoA processing determines whether source separation mode is entered. Once in separation mode, tracking is interrupted andacquisition is redone with increasing numbers of patterns until all sources are resolved or the resolution limit is reached.Note that only the channels with spoofed signals need be processed during this mode. If a single strong interferer is found,a pattern with a null in its direction can be chosen, and acquisition is attempted again. If successful, the remaining stepsare the same as in a regular GPS receiver.

The number of sources that can be resolved is theoretically M − 1, where M is the total number of distinct antennapatterns available. In practice, M is limited by the accuracy of the phase shifting mechanism in the feed. The number ofpatterns initially used at acquisition depends on the SNR level after correlation. Results from Section IV indicate that forSNR greater than 20 dB, as few as two patterns may be used for good performance, if they are the right patterns.

Fig. 2: Functional flowchart for GPS receiver integration

(a) (b)

Fig. 3: Realized geometry of the hemispiral and feed

(a) Measured vs. simulated, input phase = 180◦ (b) Measured patterns at various input phases (◦), L1

Fig. 4: Gain patterns (dB, RHCP)

IV. RESULTS AND DISCUSSION

The initial prototype of the hemispiral was made by hand and is shown in Figure 3. The supporting dome was 3D printedfrom PLA filament with a 40% honeycomb infill, which was later found experimentally to have a relative permittivity ofaround 3.8. This must be treated as a rough estimate, as the material is not perfectly uniform throughout. Simulations doinclude this structure but models it as a homogenous dielectric. The spiral arms are 18 AWG inductor wire hand-woundover the dome, and the twin-lead transformer is 12 AWG wire. The small ground plane is about 30 to 35 mm from theactive region at L1, but the effective wavelength is also shorter due to the higher permittivity in the plastic portion of thepath. A discrete coupler and an analog phase shifter are connected behind the ground plane for the measurements. Figure4a shows the RHCP gain patterns for this prototype measured in an anechoic chamber using a known standard gain horn.This broad pattern is used for acquisition/tracking in the absence of spoofers or jammers. Table I summarizes the gain atboresight and axial ratio from simulations and measurements, compared to information from the COTS antenna’s datasheet.Roughly 1-2 dB of attenuation is observed in the measurements and is attributed to the combined interconnect loss betweenthe antenna feed and the discrete coupler/phase-shifter used in the measurement setup. From Figure 4b, it is observed that

TABLE I: Performance comparison of several CP antenna elements

Band Hemispiral (simulated) Hemispiral (measured) COTS antenna (datasheet)Gain (dB, RHCP)

L5 2.07 -0.83 1.6L2 2.75 1.73 2.6L1 4.69 5.46 4.3

Axial Ratio (dB)L5 2.91 4 3L2 2.09 1.6 3L1 3.3 3.6 3

the gain is much lower with the null-steering patterns, where the arms are driven in-phase. The common mode impedance inthis case is no longer matched properly by the transformer-antenna combination, and the lower realized gain is also evidentfrom simulations. The DoA method used in this study is fairly tolerant to this, but for the purpose of being able to operatein this mode, it is desirable to improve the gain. The setup for data collection to test DoA estimation for a single source andtwo sources is shown in Figure 5. The signal generators were set at the L1 carrier of 1.57542 GHz but not synchronized,so there was about 50 MHz of frequency offset. The proof of concept is done in the anechoic chamber, with the hemispiralpointing to 0◦ and standard horns arranged at various relative angles. For the single source case, it is not important whichantenna is configured as the transmitter or receiver, but for consistency to the multiple-source case, the hemispiral is thereceiver and the horns are the transmitters. A USRP N210 with a DBSRX2 card provided the RF front-end/ADC/FPGA

Fig. 5: Measurement setup for two sources

(a) −40◦ and 25◦ (b) −29◦ and 25◦ (c) 8◦ and 17◦

Fig. 6: Modified MUSIC spectrum for various transmitter locations

used to collect the received data via GNU Radio on a Macbook Pro. There was no control link between the USRP and thephase shifter, so data was captured manually at different pattern settings.

Figure 6 shows the MUSIC spectrum plotted for sources placed at several different angles. Due to the tolerance in therotation mechanism during gain pattern measurements, the resolution of the estimates is expected to be up to 3◦ beforemeasurement/receiver noise. The average SNR of the received signals after the front-end is 48 dB, but much lower SNRenvironments must be analyzed. Figure 7a shows the variation of error in the estimated angle with increasing SNR whenusing 2, 4, 6, and 8 patterns. The discrete jumps in error are, again, due to the discrete angles of data collection in the gainpattern. As expected, using the minimum 2 patterns for a single source results in the most error in a low SNR environment.However, beyond 4 patterns, the rate of added accuracy diminishes.

This leads to a question of whether each pattern improves the estimate by the same amount. Figure 7b includes thetop performers from all dual combinations of the measured patterns, where the x-axis numbers represent the input phasecorresponding to each pattern. The y-axis reports the mean error from a sweep of incoming signals with SNR ranging from15 dB to 64 dB. It is generally the case that a high-gain pattern coupled with an asymmetrical pattern will give the bestestimates, which is intuitive as these provide the best resolution of received signals and avoid potential angular ambiguities.

V. CONCLUSION

In this paper, a hemispiral antenna has been proposed and evaluated for the purpose of increasing GNSS security. Whilethe receiver integration details and testing specifically target GPS, adapting the approach for other GNSS systems should bestraightforward. This functionality is intended to be a parallel processing step during tracking, where it can identify spoofedor multipath signals.

(a) Against SNR of incoming signals (b) Against patterns, top 10 cases shown

Fig. 7: Angle estimation errors

The performance of the initial prototype has been quantified over a range of SNR environments and pattern choices.Despite construction inaccuracy introduced by hand-winding the turns and the low pattern measurement resolution, thesystem can still produce good estimates with fairly low input SNR when three or more patterns are used.

Future work includes adjusting the hemispiral feed and/or the feed locations to improve the common-mode match andincrease the realized gain when the antenna is configured for null steering. The next prototype will be machined or printedto reduce error from inaccuracies in the realized geometry. The hemispiral will also be adapted to steer nulls in a fullhemisphere versus one plane, which was the case for this proof of concept. Additionally, a study of algorithms based onmethods other than MUSIC will be done to see if improvements can be made in terms of performance or computationsavings.

REFERENCES

[1] T. Fox-Brewster, “Watch GPS attacks that can kill DJI drones or bypass white house ban,” Aug. 2015, Forbes. [Online]. Available:https://www.forbes.com/sites/thomasbrewster/2015/08/08/qihoo-hacks-drone-gps/78b2b1e12bf5

[2] I. J. Gupta, I. Weiss, and A. Morrison, “Desired Features of Adaptive Antenna Arrays for GNSS Receivers,” in Proc. IEEE, vol. 104, no. 6, June2016, pp. 1195–1206.

[3] A. J. OBrien and I. J. Gupta, “Mitigation of adaptive antenna induced bias errors in GNSS receivers,” IEEE Trans. Aerosp. Electron. Syst., vol. 47,no. 1, pp. 524–538, Jan. 2011.

[4] K. R. Schab, E. L. Daly, and J. T. Bernhard, “Direction estimation using compressive array sensing and pattern reconfigurable antennas,” in 2013Asilomar Conf. Sig. Sys. Comput., Nov. 2013, pp. 927–930.

[5] H. Paaso, A. Mammela, D. Patron, and K. R. Dandekar, “Modified MUSIC algorithm for DOA estimation using CRLH leaky-wave antennas,” in 8thInt. Conf. Cognitive Radio Oriented Wireless Networks, July 2013, pp. 166–171.

[6] E. Kaderli, I. Bahceci, K. M. Kaplan, and B. A. Cetiner, “On the use of reconfigurable antenna arrays for DoA estimation of correlated signals,” inIEEE Radar Conf., May 2016, pp. 1–5.

[7] ANTCOM CORPORATION, “3.5in. dia. antenna for GNSS applications,” 2016.[8] J. L. Volakis, A. O’Brien, and C.-C. Chen, “Small and Adaptive Antennas and Arrays for GNSS Applications,” in Proc. of the IEEE, vol. 104, no. 6,

June 2016, pp. 1221–1232.[9] Antenna Research Associates, Inc., “FRPA-3 Description,” Dec. 2016. [Online]. Available: http://ara-inc.com/product/frpa-3/

[10] F. Li and R. J. Vacarro, “Analysis of min-norm and MUSIC with arbitrary array geometry,” IEEE Trans. Aerosp. Electron. Syst., vol. 26, no. 6, pp.976–985, Nov. 1990.

[11] R. Schmidt, “Multiple emitter location and signal parameter estimation,” IEEE Trans. Antennas Propag., vol. 34, pp. 276–280, Mar. 1986.[12] N. C. Karmakar, Ed., Handbook of Smart Antennas for RFID Systems. Hoboken, NJ: John Wiley and Sons, 2010.