Telescope Array Radar (TARA) Observatory for Ultra-High Energy Cosmic Rays

R. Abbasia, M. Abou Bakr Othmana, C. Allenb, L. Beardc, J. Belza, D. Bessonb,g, M. Byrnea,B. Farhang-Boroujenya, A. Gardnera, W.H. Gillmand, W. Hanlona, J. Hansonb, C. Jayanthmurthya, S. Kunwarb,S.L. Larsone, I. Myersa,∗, S. Prohyrab, K. Ratzlaffb, P. Sokolskya, H. Takaif, G.B. Thomsona, D. Von Maluskia

aUniversity of Utah, Salt Lake City, UT 84112 U.S.A.bUniversity of Kansas, Lawrence, KS 66045 U.S.A.

cPurdue University, West Lafayette, IN 47907 U.S.A.dGillman & Associates, Salt Lake City, UT 84106 U.S.A.

eUtah State University, Logan, Utah 84322 U.S.A.fBrookhaven National Laboratory, Upton, NY 11973 U.S.A.

gMoscow Engineering and Physics Institute, 31 Kashirskaya Shosse, Moscow 115409 Russia

Abstract

Construction was completed during summer 2013 on the Telescope Array RAdar (TARA) bi-static radar observatoryfor Ultra-High Energy Cosmic Rays (UHECR). TARA is co-located with the Telescope Array, the largest “conven-tional” cosmic ray detector in the Northern Hemisphere, in radio-quiet Western Utah. TARA employs an 8 MWEffective Radiated Power (ERP) VHF transmitter and smart receiver system based on a 250 MS/s data acquisitionsystem in an effort to detect the scatter of sounding radiation by UHECR-induced atmospheric ionization. TARAseeks to demonstrate bi-static radar as a useful new remote sensing technique for UHECRs, extending their detectionaperture far beyond what is accessible by conventional means. In this report, we describe the design and performanceof the TARA transmitter and receiver systems.

Keywords: cosmic ray, FPGA, radar, digital signal processing, chirp

1. Introduction

Modern observatories of Ultra-High Energy CosmicRays (UHECR) with energies above 1018 eV (1 EeV)exploit the properties of cosmic-ray induced ExtensiveAir Showers [1] in order to determine information aboutthe primary cosmic ray. These EAS are studied primar-ily in two ways: high energy particle detectors deployedon the ground to study the footprint of air showers [2, 3],and atmospheric fluorescence telescopes to study thefull longitudinal development of the air shower [4, 5].

With ground arrays, the air shower particles are ob-served directly. The land required to instrument groundarrays is large, cf. Telescope Array’s 700 km2 surfacedetector covers roughly the same land area as New YorkCity. The costs of the equipment required to instrumentsuch a large area are substantial, and the available landcan only be found in fairly remote areas.

∗Corresponding Author. Tel.: +01 801 5879986. Addr.: 115 S1400 E #201 JFB

Email address: isaac@cosmic.utah.edu (I. Myers)

A partial solution to the difficulties and expense in-volved in ground arrays is found in the fluorescencetechnique. Here, the atmosphere itself is part of the de-tection system, and air shower properties may be deter-mined at distances as remote as 40 km. Unfortunatelyfluorescence observatories are typically limited to a tenpercent duty cycle by the sun, moon and weather.

The possibility of radar observation of cosmic raysdates to the 1940’s, when Blackett and Lovell [6] pro-posed cosmic rays as an explanation of anomalies ob-served in atmospheric radar data. At that time, a radarfacility was built at Jodrell Bank to detect cosmic rays,but no results were ever reported. Recent experimen-tal efforts utilizing atmospheric radar systems were con-ducted at Jicamarca [7] and at the MU-Radar [8]. Bothobserved a few signals of short duration indicating a rel-ativistic target. However in neither case were the mea-surements made synchronously with a conventional cos-mic ray detector.

A new approach, first attempted by the MARI-ACHI [9, 10] project, is to utilize bi-static or two-stationradar in conjunction with a conventional set of cosmic

Preprint submitted to Nucl. Instr. Meth. Phys. Res. November 13, 2019

arX

iv:1

405.

0057

v1 [

astr

o-ph

.IM

] 3

0 A

pr 2

014

Figure 1: Map of TARA Observatory sites (transmitter and receiver)along with the Telescope Array (TA) detector facilities. The transmit-ter broadcasts as station WF2XZZ near Hinckley, Utah, towards a re-ceiver site located at the TA Long Ridge Fluorescence Detector (FD).The sounding radiation illuminates the air over the central portion ofthe TA Surface Detector array.

ray detectors. Air shower particles move very close tothe speed of light, so the Doppler shift is large. The bi-static configuration minimizes the large Doppler shift infrequency expected of the reflected signal (see [11, 12],and Section 2 below.) Also, depending on the size of theradar cross section relative to the square of the sound-ing (interrogating wave) wavelength, scattering in theforward direction might be enhanced relative to backscatter [13], thus providing an advantage in detectingthe faintest echoes in comparison to mono-static radar(ranging radar). Co-location with a conventional detec-tor allows for definitive coincidence studies to be per-formed. If coincidences are detected, the conventionaldetector’s information on the shower geometry will al-low direct comparison of echo signals with the predic-tions of air shower radio frequency (RF) scattering mod-els.

The Telescope Array Radar (TARA) project is thethe next logical step in the development of the bi-staticradar technique. Whereas MARIACHI made parasiticuse of commercial television carriers as a source ofsounding radiation (now impossible due to the transitionto digital broadcasts), TARA employs a single transmit-ter in a vacant VHF band which is under the experimen-talists’ control. The TARA receiver consists of broad-band log-periodic antennas, which are read out using a250 MS/s digitizer. TARA is co-located with the Tele-scope Array, a state-of-the-art “conventional” cosmicray detector, which happens to be located in a low-noise

environment. The layout of the TA and TARA detectionfacilities are shown in Figure 1.

This work begins with a brief description of the na-ture of air shower echoes expected for the TARA config-uration. Next, we describe the transmitter and receiversystem in some detail, including tests of system per-formance. Finally we describe upgrades to the systemwhich are currently in progress.

2. Extensive Air Showers, Radar Echoes

Cosmic rays with energies per nucleon in excess of≈ 1014 eV [14, 15] create cascades of particles withelectromagnetic and hadronic components in the atmo-sphere, known as extensive air showers (EAS). Conven-tional cosmic ray experiments detect events through co-incident shower front particles in an array of surface de-tectors or through fluorescence photons in the sky thatradiate from the shower core [2, 16, 17]. Another tech-nique takes advantage of two naturally-emitted radiosignals: the Askaryan effect [18] and geo-synchrotronradiation from interactions with the Earth’s magneticfield [19, 20, 21, 22]. This section discusses the detec-tion of EAS events through radar echo detection.

s]µTime [0 5 10 15 20 25 30

Fre

quency [M

Hz]

40

60

80

100

120

140

PS

D [dB

m/H

z]

­230

­220

­210

­200

­190

­180

­170

­160

­150

Figure 2: Spectrogram of a chirp signal produced by the radar echosimulation for an EAS located midway between the transmitter and re-ceiver with a zenith angle of 30 out of the transmitter-receiver plane.A weighted fit to the power of this signal gives a -2.3 MHz/µs chirpslope. Color scale is Power Spectral Density (PSD) given as dBm/Hz.

As the shower core ionizes the atmosphere, liber-ated charges form a plasma with plasma frequencyνp = (2π)−1

√nee2/meε0, where ne is the electron num-

ber density, e is the charge of the electron, and me isthe electron mass. A shower is denoted under-dense orover-dense (See Figure 3 in [23]) relative to the sound-ing frequency ν depending on whether ne corresponds

2

to νp > ν or νp < ν. The radar cross-section of theunderdense region is expected to be greatly attenuateddue to collisional damping [24, 25, 26]. Therefore, weexpect the dominant contribution to EAS radar cross-section σEAS to be the over-dense case, which is mod-eled as a thin-wire conductor [27]. Figure 2 displays a“typical” EAS echo, where standard shower models ofparticle production and energy transport have been as-sumed [28].

The mechanism of radar echo detection of EAS dif-fers from other radio applications. The short-livedplasma created by the shower is likely to have a radarcross-section comparable to the square of the soundingwavelength [29]. Thus, the target is small, and movingnear the speed of light. However, letting RT and RR rep-resent the transmitter/shower and receiver/shower dis-tances, respectively, the bi-static geometry guaranteesthat the echo has a non-relativistic phase shift becausethe total path length L = RR + RT evolves slowly withtime. The time-dependence of the path length causesthe phase of the echo to evolve, while the transmis-sion maintains a constant frequency. The result is anecho that has a time-dependent frequency – a chirp sig-nal [11] (Figure 2).

Chirp signals are ubiquitous in nature, although CRradar echos have very unique signatures. A simula-tion [29] has been designed that requires as inputs theCR energy, geometry and transmitter and receiver de-tails, and then evolves an EAS according to standardparticle production and energy transport models [28]while tracking the phase and amplitude of the radarecho. Shower parameters are functions of the primaryparticle energy [30]. The simulation indicates (see, forexample, a “typical” TARA geometry wave form in Fig-ure 2) that CR radar echoes are short in duration (com-parable to the shower life-time, ≈ 10 µs), have chirprates of a few times 1 MHz/µs and span a bandwidth onorder of the sounding frequency (see Figures 3 and 4).

The energy and geometry of a distribution of 10000cosmic rays detected at the TA surface detector arrayhave been simulated. Figures 3 and 4 show distribu-tions of the chirp slope and duration for these events.Data obtained from the simulation have been used toguide the design of the DAQ, transmitter system, andthe receiver antennas. A 54.1 MHz radar soundingfrequency implies the need to resolve a bandwidth ofroughly 100 MHz and therefore implement a DAQ witha 200 MHz ADC. An FPGA based design is necessaryto implement real-time digital filters that will trigger theDAQ on signals that resemble chirp radar echos.

Air showers are uniquely defined by their radar echosignatures with the exception of a lateral symmetry with

Figure 3: Simulated chirp slope distribution from a set of 10000 TAcosmic ray events. The slope is calculated from a weighted fit (bypower) to the spectrogram of the simulated signal.

Figure 4: Chirp duration distribution from 10000 simulated radarechoes from TA cosmic ray events. Duration is defined as d = t1 − t0,where t0 is the time where the maximum power is received and t1is the later time when the received power drops by 20 dB below themaximum, which approximates the end of the shower.

respect to a plane connecting the transmitter and re-ceiver and also a rotational symmetry about a line con-necting the transmitter and receiver. Stereo detection isnecessary (at minimum) to break this symmetry. Sec-tion 8 discusses the remote station prototype that willsupplement our primary receiver for stereo detection.

The actual radar cross section σEAS is currently un-known. The bi-static radar equation gives the receivedpower PR as a function of transmitter power PT . Giventhe transmitter wavelength λ and receiver and transmit-ter antenna gains GR and GT , the bi-static radar equationis written as

3

PR

PT=

GT

4πR2T

σEAS

GR

4πR2R

λ2

4π. (1)

Detection possibility depends on the signal-to-noiseratio (SNR). The TARA DAQ can trigger on realisticchirp signals as low as 7 dB below the noise (see Sec-tion 7. A simple calculation will show that, if the thinwire approximation σtw is assumed to correctly modelthe actual radar cross section (RCS) σEAS , TARA ex-pects radar echoes with positive SNR (in dB).

TARA detector parameters are given in Table 1. Con-sider a 60 MHz Doppler shifted tone, scattered froman EAS located midway between the transmitter andreceiver, which have a 39.5 km separation distance.Received power can be calculated from Equation 1 ifσEAS ' σtw is known. Some basic assumptions al-low a quick calculation of σtw: Shower Xmax occursroughtly 2 km from the ground; the antennas’ polar-ization vector and shower axis are in the same plane;the length L of the scattering region of the shower isthe speed of light multiplied by the electron attach-ment/recombination lifetime τ = 10 ns [24], which im-plies L = 3 m; the over-dense region radius near Xmaxis the thin wire radius [23] a = 0.01 m. With these as-sumptions the thin-wire cross section [27] is σtw ∼ 1 m2

and the received power is -79 dBm.

UHECR energy 1019 eVPT 40 kWGT 22.6 dBi (Section 4.3)GR 12.6 dBi (Section 5)

RT = RR 19.75 km

Table 1: The list of the parameters assumed for calculating receivedpower from a 60 MHz Doppler shifted radar echo scattered from anEAS.

Section 5 shows a plot of galactic noise superimposedwith receiver system background noise. Receiver sam-ple rate and Fourier transform window size used in thecalculation were 250 MS/s and 32768, respectively. Thepower spectral density (PSD) of a -79 dBm tone de-tected by this system is −79 + 10 log10 (32768/250 ·106) = −117 dBm/Hz. The reader should note that an-tenna patterns are both assumed to be at their maximum,which rarely occurs in practice. Further, polarizationangle differences (φ) between the shower axis and an-tennas can yield another reduction in power ∝ cos4 (φ).

The receiver background noise plot demonstratesthat, in the TARA frequency band of interest, back-grounds are dominated by galactic noise. At 60 MHz,the background noise PSD is -160 dBm/Hz, much lower

than that of a narrow-band Doppler shifted radar echoat -117 dBm/Hz scattered from an ideal thin wire. Un-der reasonable assumptions for signal parameters, com-bined with our measured irreducible backgrounds andsystem response, the thin wire approximation for theradar cross-section σEAS implies high values of signal-to-noise.

3. Transmitter

3.1. Hardware

TARA operates a high power, Continuous Wave(CW), low frequency radar transmitter built from repur-posed analog TV transmitter equipment with FCC callsign WF2XZZ, an experimental license. The transmit-ter site (39 20′ 19.82400′′ N, 112 42′ 3.24000′′ W) isjust outside Hinckley, UT city limits where human ex-posure to RF fields is of little concern. A high gain Yagiarray (Section 4) focuses the radar wave towards the re-ceiver station (Section 5) located 40 km away. Figure 1shows the transmitter location near Hinckley and rela-tive to the TA SD array [2]. The geometry was chosento maximize the possibility of coincident SD and radarecho events.

Figure 5: Schematic of the transmitter hardware configuration. Acomputer connected to RF sensor equipment, an arbitrary functiongenerator and transmitter control electronics orchestrates the two dis-tinct transmitters and provides remote control and logging. RF powerfrom each transmitter’s two amplifier cabinets is combined with outof phase power rejected into a 50 Ω load. A hybrid combiner sumsthe combined output of each transmitter and sends that power to theantenna. Power reflected back into the hybrid combiner is directed toa third RF load.

Figure 5 shows a schematic of the transmitter hard-ware configuration. A Tektronix arbitrary function gen-erator (AFG 3101; Tektronix, Inc.) provides the pri-mary sine wave, which is amplified by over nine ordersof magnitude before reaching the antenna. 54.1 MHzwas chosen as the sounding frequency because of thelack of interference in the vacated analog channel two

4

TV band and the 100 kHz buffer between it and the am-ateur radio band which ends at 54.0 MHz.

Two 20 kW analog channel 2 TV transmitters havea combined 40 kW power output. The primary signalfrom the function generator is split to feed both trans-mitters (Harris Platinum HT20LS, p/n 994-9236-001;Harris Broadcast) with the same level of gain. Eachtransmitter includes a control cabinet and two cabinetsof power amplifier modules. RF power from each cabi-net is combined in a passive RF combiner (620-2620-002; Myat, Inc.) that routes any out-of-phase signalto a 50 Ω load. The combined output of each trans-mitter is sent to a 90 hybrid combiner (RCHC-332-6LVF; Jampro, Inc.) that sums the total output of eachtransmitter. Between the final combined input and eachtransmitters’ combined output there is an inline analogchannel 2 low pass filter (visual low-pass filter, 3 1/8”;Myat, Inc.) to minimize harmonics. RF power leavesthe building through 53 m of semi-flexible 3 1/8” circu-lar air-dielectric wave guide (HJ8-50B; Andrew, Inc.).

Modifications were made to the transmitters to by-pass interlocks that detect the presence of aural and vi-sual inputs and video sync pulses necessary for stan-dard TV transmission. Control cabinet electronics werecalibrated to measure the correct forward and reflectedpower of the 54.1 MHz tone instead of the RF envelopeduring the sync pulse. Currently, total power output islimited to 25 kW because of limitations that arise fromamplifying a single tone versus the full 6 MHz TV band.

Air conditioning and ventilation are critical to highpower transmitter performance. Currently, transmitterefficiency is slightly better than 30%, which implies thatnearly 75 kW of heat must be removed from the build-ing. The environment at the site is very dry and dusty,so all of the air brought into the building is filtered andpositive gauge pressure is maintained. A single 25 tonAC unit filters and pumps cool air into the building. Aneconomizer will shut down the compressor if the outsideair temperature drops below 15.6 C (60 F). However,if the room is not cooling quickly with low outside am-bient temperature, the compressor will be turned backon. Hot air near the ceiling is vented as necessary tomaintain a slight positive pressure.

Future improvements to the transmitter will includebiasing the power amplifiers for class B operation, inwhich amplification is applied to only half the 54.1 MHzcycle. Resonance in the transmitter and antenna allowthe second half of the wave to complete the cycle. Ef-ficiency will nearly double compared with the currentdesign.

3.2. Remote Monitoring and ControlRemote monitoring and control of the transmitter is

important for two reasons. First, Federal Communica-tions Commission (FCC) regulations require that non-staffed transmitter facilities be remotely controlled andseveral key parameters monitored. Second, forwardpower and other parameters must be logged for receiverdata analysis.

A computer interfaces with digital I/O and analoginput devices that, in turn, are connected to the trans-mitters’ built in digital I/O and analog output interface.RF power sensors (PWR-4GHS; Mini-Circuits) mea-sure the final forward and reflected power via stronglyattenuating sample ports on the wave guide near thebuilding exit port. The sum of the two control cabinets’forward and reflected power measurements can be com-pared with the separate RF final forward and reflectedpower measurements.

The host computer monitors transmitter digital status,analog outputs and RF power sensors and controls thefunction generator. Logs are updated every five min-utes with forward and reflected power for each trans-mitter, final (re: antenna) forward and reflected power,room temperature and various transmitter status and er-ror states. Warning and error thresholds can triggeremails to the operators and initiate automatic shutdown.The program also provides a simple interface that allowsthe operator to remotely turn the transmitter on and off,increase or decrease forward power, and add a text logentry.

3.3. PerformanceTV transmitters are designed for 100% duty cycle op-

eration. Similarly, the TARA transmitter is intendedfor continuous operation to maximize the probability ofdetection of UHECRs. With fixed gain and input sig-nal, power is strongly correlated with transmitter roomambient temperature. Large temperature fluctuations inApril 2013 resulted in a ∼ 3 kW spread in output power(Figure 6).

Transmitter forward power is more stable if roomtemperature is kept lower than 300 K (80 F). Figure 7shows that forward power fluctuations in August 2013are much smaller than April. Built-in automatic gaincontrol was increased during this period as well. Theaverage power in December is higher than the averagepower in April because a slightly higher power inputsignal was used in later months. Reflected power is typ-ically ∼ 100 W, which is very low for such a high powersystem. This can be attributed to very good impedancematching with the extremely narrow-band Yagi antennaarray.

5

Date in 2013 [mm­dd]04­20 04­22 04­23 04­25 04­26 04­27 04­29 04­30

Pow

er

Outp

ut [k

W]

22

22.5

23

23.5

24

24.5

25

25.5

26

Room

Tem

p. [K

]

280

285

290

295

300

305

310

315

Figure 6: Transmitter forward power (black) and room temperature(red) during April 2013. Poor air conditioning calibration resulted indaily temperature fluctuations which caused large output power mod-ulation.

Date in 2013 [mm­dd]12­12 12­13 12­13 12­13 12­14 12­14 12­15 12­15

Pow

er

Outp

ut [k

W]

22

22.5

23

23.5

24

24.5

25

25.5

26

Room

Tem

p. [K

]

280

285

290

295

300

305

310

315

Figure 7: Transmitter forward power (black) and room temperature(red) during December 2013. A well-calibrated air conditioning sys-tem keeps room temperature stable and increased automatic gain con-trol minimizes forward power fluctuations.

Figure 8 shows the total forward and reflected powerin red and blue, respectively, referenced to the right ver-tical axis and the integrated on-time in black, referencedto the left vertical axis, since its commissioning in lateMarch, 2013. The transmitter has been turned off sev-eral times for maintenance and testing and during pe-riods when our receiver equipment was removed fromthe field for upgrades. Although forward power is notcontinuous and fluctuations were large in the past, weconsider 200 days of operation in the first year to bodewell for future data collection.

Harmonics have been measured to confirm compli-ance with FCC regulations and to avoid interfering with

Date [mm­dd]04­15 05­15 06­14 07­14 08­13 09­12 10­12 11­11 12­11

TX

On­t

ime [day]

0

20

40

60

80

100

120

140

160

180

200

Pow

er

[kW

]

0

5

10

15

20

25

Figure 8: Transmitter on-time in days (black, left vertical axis) andforward and reflected power in units of kW (red and blue, right verticalaxis) during 2013. Total duty cycle during this period is 83%.

other stations. With total forward output at 25 kW,the fundamental and several harmonic frequencies weremeasured from a low power RF sample port. The firstfive harmonics are about 60 dB below the fundamental(see Table 2). Harmonics will be further attenuated byabout 30 dB by the intrinsic bandpass of the antenna.

Frequency (MHz) Power (dBm)54.1 8.5108.2 -66.0162.3 -68.3216.4 -84.4270.5 -89*324.6 -77*378.7 -94*432.8 -87*486.9 -98*541.0 -91*

Table 2: Power of fundamental frequency and first ten harmonics forthe 54.1 MHz radar sounding wave. These measurements were takenfrom a highly attenuated final forward power RF sample port. Totaltransmitted power was approximately 25 kW. FM and TV stations arerequired by the FCC to limit the first ten harmonics to at least 60 dBbelow the approved power. Experimental station WF2XZZ is exemptfrom this requirement although it readily meets it. (*fluctuating value,±5 dB)

4. Transmitting Antenna

4.1. Physical DesignAs the bi-static radar equation (Equation 1) shows,

the received power is the product of the scattering cross

6

section, transmitted power, antenna gain, receiver gainand receiver aperture. Because the physics of the radarscattering cross section is not well understood, an an-tenna with high gain and directivity was chosen to max-imize received power.

The TARA transmitting antenna is composed of 8narrow band Yagi antennas designed and manufacturedby M2 Antenna Systems, Inc. Each Yagi is constructedof aluminum and capable of handling 10 kW of contin-uous RF power. The specifications for each Yagi are afrequency range of 53.9 - 54.3 MHz, 12 dBi free spacegain, front to back ratio of 18 dB, and beam widths of27 and 23 in the vertical and horizontal planes respec-tively. Each Yagi antenna is composed of five elements:a reflector, driven element, and three directors, and aremounted on a 21.6 ft long, 2 ′′ diameter boom. Eachantenna’s balanced t-match is fed from a 4:1 coaxialbalun which transforms the unbalanced 50 Ω input tothe balanced 200 Ω used to drive the antenna. A 50 Ω

7/8 ′′ cable connects each antenna balun to the four portpower dividers. Table 3 describes the lengths and posi-tions of the antenna elements on the boom. All elementsare constructed of aluminum tubing of 3/4 ′′ outer di-ameter. Each element, except for the driven element isconstructed of two equal sections that are joined at theboom via 7/8 ′′ outer diameter sleeve elements. Eachantenna weighs 35 lbs when completely assembled.

Element Length (in) Position (in)Reflector 107.625 -44.375Driven Element 100.500 0.000Director 1 99.500 51.125Director 2 97.250 131.625Director 3 97.000 193.625

Table 3: Length and relative boom position of antenna elements of theTARA Yagi antennas. All elements have a diameter of 0.75 ′′.

Transmitter output power is delivered to the antennaarray via approximately 100 feet of CommScope HJ8-50B 3 1/8 ′′ Heliax air dielectric coaxial wave guide.The Heliax then connects to a two port power divider lo-cated at the base of the antenna array. Each output portof the power divider feeds equal length 1 5/8 ′′ coax-ial cables, which in turn feed a four port power divider.Each four port power divider then delivers power to theindividual Yagi antennas via equal length 7/8 ′′ coaxialcables. All components in the transmission line chainare impedance matched to 50 Ω.

The antennas are mounted on four wooden telephonepoles, two stacked vertically on each pole. The bottomand top antennas on each pole are located 10 ft and 30 ft

Figure 9: Configuration of the eight Yagi antennas and mounting poleswhich comprise the TARA transmitting antenna array.

above the ground respectively. Currently, the antennasare mounted in a configuration that provides a horizon-tally polarized signal. Wooden poles were used to allowa change of polarization. The poles, separated by 20 ft,are aligned in a plane perpendicular to the line point-ing toward the receiver site located at the Long Ridgefluorescence detector 39 km to the southwest. Figure 9shows the antenna array configuration.

4.2. Theoretical Performance

The eight Yagi antennas are operated as a phased ar-ray to take advantage of pattern multiplication to im-prove gain and directivity relative to the individual an-tennas. The design philosophy of the antenna array isto deliver a large amount of power in the forward di-rection in a very narrow beam to maximize the powerdensity over the TA surface detector. High power den-sity is equivalent to a large PT GT factor in the bi-staticradar equation, which is needed to increase the chanceof detection of a cosmic ray air shower via radar echogiven the uncertainty in the radar scattering cross sec-tion σEAS . Before construction, modeling of the arraywas performed using version two of NEC [31], an an-tenna modeling and optimization software package.

Figure 10 shows the radiation pattern of the full eightYagi array when configured as shown in Figure 9. For-ward gain is 22.6 dBi, horizontal beam width is 12,vertical beam width is 10, the front-to-back (F/B) ratiois 11.8 dB and the elevation angle of the main lobe is 9.

Simulations were performed to find the best spacingbetween the mounting poles, vertical separation of an-tennas and height above ground to shape and direct the

7

0° 15°30°

45°

60°

75°

90°

105°

120°

135°

150°165°180°-165°

-150°

-135°

-120°

-105°

-90°

-75°

-60°

-45°

-30°-15°

−3

3

9

15

21

0° 15°30°

45°

60°

75°

90°

105°

120°

135°

150°165°180°-165°

-150°

-135°

-120°

-105°

-90°

-75°

-60°

-45°

-30°-15°

−3

3

9

15

21

Vertical Gain (dBi)

Figure 10: Simulated horizontal (top) and vertical (bottom) radiationpatterns of the eight Yagi TARA antenna array shown in blue. Redpoints are measured data that have been uniformly scaled to best fitthe model. Forward gain is 22.6 dBi, beam width is 12 horizontal,10 vertical, and the F/B ratio is 11.8 dB.

main lobe in a preferred direction. The antenna polespacing influences the main lobe beam width. A nar-rower beam width can be obtained at the expense oftransferring power to the side lobes which do not directRF energy over the TA surface detector. Elevation an-gle is manipulated by the antenna height above ground.Changing this parameter does little else to the mainlobe. The elevation angle and beam width were selectedto increase the probability that air shower Xmax wouldfall in the path of the main lobe where the charged par-

53.0 53.2 53.4 53.6 53.8 54.0 54.2 54.4 54.6 54.8 55.0 55.2Freq (MHz)

50

45

40

35

30

25

20

15

10

5

0

S11

(dB)

-37.25 dB @ 54.1 MHz

Figure 11: Reflection coefficient (S 11) for the eight Yagi array.

ticle density is the greatest. The 9 main lobe elevationangle is chosen such that the sounding wave illuminatesthe mean Xmax for showers of order 1019 EeV midwaybetween transmitter and receiver [32].

4.3. Measured PerformanceThe ability of an antenna to project energy is best

characterized by the reflection coefficient S 11 (alsocalled return loss when expressed in dB). It is a measureof the ratio of the voltage reflected from a transmissionline relative to input. Large reflection coefficient impliessignificant energy is reflected back into the transmitterbuilding which can interfere with other electronics, el-evate ambient temperature and even damage the trans-mitter. Figure 11 shows the reflection coefficient for theYagi array. It shows a return loss of -37.25 dB at thesounding frequency, which is excellent. S 11 of -20 dBor less is considered good.

To verify that the transmitting antenna is operating asdesigned, an RF power meter or similar device can beused to measure the power as a function of position rel-ative to the antenna. This measurement is challengingbecause it must be performed in the far field of the an-tenna (typically r λ). To fully probe the radiationpattern of the TARA transmitting antenna, power mea-surements must be made high above the ground sincethe main lobe is inclined 9 relative to horizontal.

Vertical radiation pattern measurements were takenby using antenna transmitting/receiving symmetry. Atethered weather balloon was floated with a custombattery powered 54.1 MHz signal generator that fed adipole antenna. Over a range of discrete heights, re-ceived power was recorded at the output (normally theinput) of the Yagi array.

The horizontal (azimuthal) radiation pattern wasmeasured using a spectrum analyzer on the ground to

8

determine the pointing direction and shape of the mainlobe. Measurements of transmitted RF power weretaken at distances between 650 and 1000 m radiallyfrom the center of the array. Power was measured alonga road that does not run perpendicular to the pointing di-rection of the transmitter so a 1/r2 correction was made.Figure 10 shows the measured points for the horizon-tal and vertical patterns overlayed on the models. Thesemeasurements are all relative, not absolute, so a uniformscale factor was determined by minimizing χ2 betweenthe model and data. The measured pattern agrees verywell with the model in pointing direction and shape.

5. Receiver Antenna

Figure 12: Dual polarized TARA Log Periodic Dipole Antenna(LPDA).

The TARA receiver antenna site is located at theTelescope Array Long Ridge Fluorescence Detector, at(39 12′ 27.75420′′ N, 113 7′ 15.56760′′ W).

The receiver antennas are dual-polarized log periodicdipole antennas (LPDA) designed to match the expected< 100 MHz signal frequency characteristics. Due tonoise below 30 MHz and the FM band above 88 MHz,the effective band is reduced to 40 to 80 MHz. Each an-tenna channel is comprised of a series of six λ/2 dipoles.The ratio of successive dipole lengths is equal to the hor-izontal spacing between two dipoles (the defining char-acteristic of LPDA units), with the longest elements far-thest from the feed-point to mitigate large group delayacross the passband. Table 4 gives the lengths and po-sitions of the antenna elements on the boom from thefront edge to the back. All elements are constructed of

aluminum tubing of 1/4 ′′ outer diameter. Figure 12shows a schematic of the receiver LPDA.

Element Length (in) Position (in)1 21.875 3.6252 26.625 18.06253 32.5 35.6254 39.625 57.05 48.3125 83.1256 58.3125 115.0

Table 4: Length and relative boom position of antenna elements of theTARA Log Periodic Dipole Antennas. All elements have a diameterof 0.25 ′′.

30 40 50 60 70 80 90 100Frequency (MHz)

0

2

4

6

8

10

V.S.

W.R

MeasuredSimulation

Figure 13: SWR of a horizontally polarized TARA LPDA as measuredin an anechoic chamber.

The impedance of the antenna against a 50 Ω trans-mission line was measured in an anechoic chamber atthe University of Kansas. The standing wave ratio(SWR), the magnitude of the complex reflection coef-ficient (S 11), is shown as a function of frequency in Fig-ure 13. An SWR of 3.0 implies greater than 75% signalpower is transmitted from the antenna to the receiver ata given frequency.

The complex S 11 measurement also quantifies theeffective height of the LPDA. The effective height trans-lates the incident electric field strength, in (V/m), to avoltage at the antenna terminals. It is given as Einc ·

heff = |Einc||he f f | cos(θ) = V , where θ is the polariza-tion angle and the antenna is assumed to be horizon-tally polarized. The boresight effective height can beexpressed [33] as

9

h(ν) = 2 ∗

√Gc2|Zin|

4πν2Z0. (2)

In the effective height expression, G is the measuredgain of 12.6 dBi (see Figure 15), c is the speed of light,Zin is the complex antenna impedance, ν is the fre-quency, and Z0 = 120π is the impedance of free space.In terms of the measured complex reflection coefficientS 11, the impedance is given by |Zin| =

∣∣∣∣ 1+S 111−S 11

∣∣∣∣ 50 Ω. Thefrequency-dependent magnitude of the effective heightis plotted in Figure 14.

0

5

10

15

20

40 50 60 70 80 90 100

Eff

ecti

ve

Hei

gh

t (m

)

Frequency (MHz)

Black: w/ Measured SWRRed: w/ Simulated SWR (NEC)Blue: w/ Ave. Simulated SWR (NEC)

Figure 14: Effective height in meters vs. frequency in MHz of theTARA receiver LPDA. The black and red curves represent measuredand simulated data, respectively.

Receiver antenna gain is a factor in the bi-static radarequation that affects detection threshold. NumericalElectromagnetic Code was used in simulating the ra-diation pattern of the antenna to confirm directionality(see Figure 15). Simulated forward gain is 12.6 dBi andthe vertical beam width is 23 at the carrier frequency,54.1 MHz.

From the radiation pattern (transmission coefficient,S 21) measurements, the beam width was obtained.Here, we will define “beam width” as the angle inthe plane under consideration over which the radiatedpower is within three dB of the maximum. Figure 16)displays beam width over the band of interest.

In any RF receiver system, sensitivity is limited bythe combination of external noise entering through theantenna and internal noise from various sources like lownoise amplifiers and other resistive losses from filters,cables and couplers. Noise entering the antenna is gen-erated by the sky, earth and antenna resistive loss. Dif-fuse radio noise from the galactic plane is non-polarizedand is the dominant noise source in the TARA frequencyband. Anthropogenic noise sources are transient and

0°

15°

30°

45°

60°

75°90°

105°

120°

135°

150°

165°

180°

195°

210°

225°

240°

255°270°

285°

300°

315°

330°

345°

Horizontal Gain(dBi)

129

50

510

180°165°

150°

135°

120°

105°

90°

75°

60°

45°

30°

15°0°

-15°

-30°

-45°

-60°

-75°

-90°

-105°

-120°

-135°

-150°

-165°

Vertical Gain(dBi)

12

9

5

0

5

10

Figure 15: Simulated horizontal (top) and vertical (bottom) radia-tion pattern of a horizontally polarized TARA LPDA at the transmittersounding frequency of 54.1 MHz. Beam widths (−3 dB below peakgain) are shown with red lines.

stationary noise is absent within our measurement banddue to the receiver site’s remote location. In this fre-quency region, galactic noise dominates thermal noise.By accounting for amplifier and instrumental gains andlosses, the observed noise background can be comparedwith the irreducible galactic noise background. Ourmeasured system noise calibrated against the galacticnoise standard [34] is shown in Figure 17. RF back-ground at Long Ridge is observed to be close to the ex-pected galactic noise floor [35]. In order to measure the

10

50 60 70 80 90 100Frequency(MHz)

30

40

50

60

70

80

90

100

Hori

zonta

l B

eam

wid

th (

Deg)

Figure 16: Beam width of a single channel LPDA measured as mea-sured in the CReSIS anechoic chamber.

Frequency [MHz]40 45 50 55 60 65 70 75 80

Ave

rag

e P

SD

[d

Bm

/Hz]

­180

­160

­140

­120

­100

­80

­60

Figure 17: Receiver system noise floor (black) Power Spectral Density(PSD) in dBm/Hz shown with a fit to measured galactic backgroundnoise and its associated error [34] (red band). System attenuation,filters and amplifiers were accounted for.

observed curve in Figure 17 the frequency dependentpower as seen by the FlexRio DAQ (see Section 7) wasaveraged over a thousand forced triggered events. TheS 21 measurement of the filters and amplifiers and at-tenuation by the LMR-400 transmission line were alsoaccounted for in this calculation.

6. Receiver Front-end

There are three dual-polarization antennas at the re-ceiver site, two of which are currently connected to theDAQ. RF signal from the antennas pass through a bank

of filters and amplifiers. The components include anRF limiter (VLM-33-S+; Mini-Circuits), broad bandamplifier, low pass filter (NLP - 100+; Mini-Circuits),high pass filter and an FM band stop filter (NSBP-108+; Mini-Circuits). Both polarizations from one an-tenna are filtered (37 MHz cutoff frequency high passfilter, SHP-50+; Mini-Circuits) and amplified (40 dB,ZKL-1R5+; Mini-Circuits) at the antenna, where a biastee (ZFBT-4R2G+; Mini-Circuits) is used to bring DCpower from the control room. The second antenna’schannels are filtered (25 MHz high pass filter, NHP-25+; Mini-Circuits) and amplified (30 dB, ZKL-2R5+;Mini-Circuits) inside the control room. The lightningarrestor (LSS0001; Inscape Data) minimizes damageto sensitive amplifiers by electric potentials that accrueduring thunderstorms. The RF limiter prevents damageby transient high amplitude pulses (see Section 7.2).

Signal conditioning in the amplifier/filter banks ischaracterized by the transmission coefficient (Figure 18)S 21. It is a measure of the ratio of the voltage at the endof a transmission line relative to the input. Impedancemismatch relative to a 50 Ω transmission line, insertionloss for the various devices and gain from the amplifiersare combined in S 21 data.

0 20 40 60 80 100 120−50

−40

−30

−20

−10

0

10

20

30

40

Frequency(MHz)

S21(d

B)

S21 of filterbank 3

Figure 18: S 21 (transmission coefficient) of the filter and amplifierbank connected to the triggering channel of the DAQ.

7. Receiver DAQ

7.1. DAQ Structure

The National Instruments FlexRIO system providesan integrated hardware and software solution for a cus-tom software defined radio DAQ. It is composed of three

11

RF LimiterFM Filters

AND Amplifiers

ControllerGPS

ReceiverUnit

Analog Input

AdapterFPGA

Trigger%

NI$FlexRIO$Device$NI$PXIe?8133$

Bista4c5Radar%Receiver%Sta4on%

LMR$cable$

Radar%Target%Echoes%Emulator%

Arbitrary Waveform Generator

TXDipole

Antenna

RX Multiple

Antennas

Figure 19: Elements of the radar receiver station.

basic parts: adapter module, FPGA module and hostcontroller (as shown in the lower box of Figure 19). Adescription of each of these subsystems follows.

The NI-5761 RF adapter module is a high-performance digitizer that defines the physical inputsand outputs of the DAQ system. It digitizes four analoginput channels at a rate of 250 MS/s with 14-bit reso-lution. Eight TTL I/O lines are available for additionalcontrol, some of which are used in custom DAQ trigger-ing schemes.

The NI-7965R FPGA module is based on the PXI ex-press platform which uses a Xilinx Virtex-5 FPGA with128MB on board DRAM. FPGA design provides accu-rate timing and intelligent triggering. The PXI-expressplatform has a high-speed data link to the host con-troller, which is connected to the development machine,a Windows based computer, which uses the LabVIEWenvironment to design and compile FPGA code. A hostcontroller application, also designed in LabVIEW, runson the development machine.

7.2. Design Challenges

Based on the high velocity of the radar target, echoesare excepted to be characterized by a rapid phasemodulation-induced frequency shift, covering tens ofMHz in 10 µs. As the magnitude of the Doppler blueshift decreases as the shower develops in the atmo-sphere, these signals sweep (approximately) linearlyfrom high to low frequency and are categorized aslinear-downward chirp signals. Echo parameters are de-pendent on the physical parameters of the air showers.Thus, unlike existing chirp applications, we are inter-

ested in the detection of chirp echoes of variable ampli-tude, center frequency and frequency rates within a rel-atively wide band. In addition, the detection thresholdmust be minimized in order to increase the probabilityof detecting radar echoes with SNR less than one.

Furthermore, UHECR events are rare and random intime. TA receives only several > 1019 eV events perweek, so background noise and spurious RF activitydominate.

Figure 20 shows a spectrogram of data acquired in thefield using the complete receiver and test system (Fig-ure 19), where FM radio and noise below ∼ 30 MHz isfiltered out. The time-frequency representation showsthat the background noise of our radar environment isrich with multiple undesirable components includingstationary tones outside the 40-80 MHz effective bandlocated at 28.5 MHz and, inside the band, the carrier at54.1 MHz as well as broadband transients. Sudden am-plitude modulation of stationary sources and powerful,short-duration broadband noise can cause false-alarms.A robust signal processing technique is needed to con-front these challenges [36].

7.3. DAQ Implementation

The DAQ is designed to detect chirp echoes and con-front the problem of a variable noise environment. Twoantenna channels, each with two channels for horizon-tal and vertical polarization, are acquired to assist inpost-processing and analysis. Data are collected simul-taneously from each of the four analog channels, thensampled using a 250 MS/s ADC (Texas Instruments;ADS62P49). Analog to digital conversion is followed

12

Figure 20: Spectrogram of background noise at the receiver site.Frequency and time are on the vertical and horizontal axes, respec-tively, with color representing the power in a particular frequencycomponent. The carrier signal is represented by the horizontal lineat 54.1 MHz. Broadband transients are the vertical lines and station-ary noise sources are the horizontal band near 30 MHz.

by fast digital memory storage on the FPGA chip, whichstores the incoming samples from each channel sequen-tially, in a 131 µs (32744 sample) continuous circularbuffer such that data in each buffer are continually over-written. Three distinct trigger modes are implemented:“snapshot”, “Fluorescence Detector (FD) external”, and“matched-filter bank”.

Timing Diagram

6

131 µsec Acquired Window

Te =

T1 = 95 µsec

Pre-trigger

T2 = 36 µsec

Post-trigger

Figure 21: Position of the triggering pulse within the data window thatis written to disk.

When a trigger occurs, the circular buffer informationis sent to the host controller to be permanently stored onthe computer’s disk. A 320 µs dead-time is required toaccount for FPGA-host data transfer limitations, duringwhich the DAQ cannot accept triggers. As depicted inFigure 21, pre/post trigger acquisition is set to 95 µs and36 µs, respectively, to allow for jitter in the FD triggertiming (which turns out to be very small) and sufficientpost-trigger data to see an entire echo wave form. AGPS time stamp is retrieved from a programmable hard-ware module [37] and recorded for each trigger with an

absolute error of ±20 ns.The snapshot trigger is an unbiased trigger scheme

initiated once every minute that writes out an event todisk. These events will (likely) contain backgroundnoise only. Unbiased triggers are crucial for backgroundnoise estimation and analysis.

During active FD data acquisition periods, the LongRidge FD (the location of the TARA receiver site) emitsa NIM (Nuclear Instrumentation Module) pulse for eachlow level trigger, an OR of individual mirror triggers.The FlexRIO is forced to trigger by each pulse receivedfrom the FD. Each FD run will result in hundreds ofthousands of triggers which can be narrowed to ∼ 100events that coincide with real events found in recon-structed TA data.

t0# t0+#Tc#Rela+ve#Time#(sec)###

Amp.#(V

olt)#

Freq

.#(MHz)

#

t0#

fH#

fL#

t0+#Tc#Rela+ve#Time#(sec)###

Slope#=#κ#

(a)# (b)#

Figure 22: Linear down-chirp signal. (a) Signal in time-domain. (b)Signal in time-frequency domain.

The matched filter (MF) bank is a solution for theproblem of detecting radar chirp echoes in a challengingnoise background using signal processing techniques.The signal of interest is assumed to be a down-chirp sig-nal that has duration Tc seconds with a constant ampli-tude, start (high) frequency fH, center frequency fC, end(low) frequency fL and chirp rate κ Hz/sec. An exampleof the signal of interest is shown in Figure 22. Assum-ing that it is centered around time t = 0, such a chirpsignal is written as

s(t) = rect(

tTc

)cos(2π fCt − πκt2). (3)

where rect(x) is the rectangle function and t is the timein seconds.

We limit our interest to detecting the presence of s(t)within a certain bandwidth, without prior knowledge ofthe chirp rate κ. Based on simulation of the physicaltarget, reflected echoes are expected to have a peak am-plitude within or near the range [65-60] MHz. Thus, weconsider fH to be 65 MHz and fL to be 60 MHz.

Since the chirp slope varies, we use a bank of filtersmatched to a number of quantized chirp rates, κ1, κ2,· · · , κM . A functional block diagram of the detectionprocess is illustrated in Figure 23.

13

Band%Pass((Filtering(and(

Decima3on(

MF(1)(Κ1(MHz/μsec(

MF(2)(Κ2(MHz/μsec(

MF(M%1)(ΚM%1(MHz/μsec(

MF(M)(ΚM(MHz/μsec(

Magnitude(Detec3on(

and(Threshold(Comparison(

Input( Amplitude(Limiter(

Decision(

Figure 23: Block diagram of the matched-filter-type detector.

Let ym denote the output samples of the mth matchedfilter and γm the threshold at the filter output. As de-picted in Figure 23, a trigger decision is made at the out-put of the matched-filter bank by comparing magnitudesof the elements of y1, y2, · · · , yM, each, against the cor-responding threshold levels γ1, γ2, · · · , γM , respectively.Threshold levels are defined as nγ units of the signallevel (equivalently, noise standard deviation) at the out-put of each filter, denoted by σm for the mth matchedfilter. Every time a trigger condition (the presence of achirp) is met, an event is declared. Since the backgroundnoise level varies with time, σm is measured every fiveseconds to maintain a constant data acquisition rate.

The most probable chirp-rate interval for a distribu-tion of simulated radar echoes isK = [−3,−1] MHz/µs.We choose M = 5 and the chirp rates (in MHz/µs) as

κ1 = −1.1161, κ2 = −1.3904, κ3 = −1.7321,

κ4 = −2.1577, κ5 = −2.6879 .

7.3.1. Amplitude Limiter

Radio background at the remote receiver site is clear ofstationary interference signals in the frequency band ofinterest, 30 - 88 MHz. Therefore, the broadband tran-sients mentioned in Section 7.2 are the primary sourceof false alarms. Consequently, the threshold of the Like-lihood Ratio Test (LRT) detector must be raised in orderto maintain the desired false alarm rate. A digital am-plitude limiter applied immediately before the input tothe LRT detector helps to minimize false alarms whilekeeping the detection threshold as low as possible andwithout significantly degrading detection efficiency.

The amplitude limiter clips the amplitude of the re-ceived signal to a fraction k of its RMS value before

clipping. Its mathematical expression isy = x, |x| < kσs

y = kσs, x > kσs

y = −kσs, x < −kσs.

(4)

where x is the raw input, y is the amplitude limited out-put, and σs is the RMS value of the signal before clip-ping. The result is a reduced relative power ratio ofthe spurious impulses to the non-perturbed background.Clipping also lowers the wave form RMS in proportionto the clipping level.

7.3.2. Band-Pass Filtering

We observe considerable CW noise within the 30-88 MHz band, including the carrier signal. The carrierand other persistent tones can have large amplitudes andlead to a high matched filter RMS output which can, asshown in the next section, cause non-detection of lowSNR chirp signals. Such tones, including the carrier,can be easily filtered out. Before the amplitude limiter,a narrow band-pass filter eliminates all frequencies out-side a 60-65 MHz band with -80 dB stop band attenu-ation. Data stored in the ring buffer are not filtered insuch a way.

7.4. Performance EvaluationDetection performance of the LRT detector has been

evaluated under two test signal conditions: noise onlyor signal plus noise. For each test, the boolean re-sult of the threshold comparison with the MF outputs isrecorded. The probability of signal plus noise exceedingMF thresholds is the efficiency and the average rate oferroneous detection decisions caused by filtered noise isfalse alarm rate.

The ability to detect a received chirp signal in back-ground noise depends on the ratio of the signal power tothe background noise power. Radar carrier power dom-inates the background so two quantities are used to de-scribe the background noise. First, we define the ratio ofthe test chirp signal power to the radar carrier power asthe signal-to-carrier ratio (SCR). Second, we define theratio of the test chirp signal power to the noise powerat the input to the matched-filter bank (after filtering outthe radar carrier) as the signal-to-noise ratio (SNR), viz.,

SNR =Pc

σ2ν

, (5)

where Pc is the chirp signal power andσν is the standarddeviation of the background noise after filtering out thepowerful carrier signal.

14

Consider the following observations about perfor-mance analysis. First, it is clear that system perfor-mance depends on the chosen threshold level nγ (userdefined, a multiple of σm as defined previously) foreach SNR value. False alarm rate is expected to de-crease as the threshold level increases, at the expenseof detection efficiency of low SNR chirp signals. Con-versely, detection efficiency increases as the thresholddecreases. Second, the false alarm rate is expected todecrease as the amplitude limiter level decreases be-cause high amplitude transients are effectively removed.To this date, radar echoes from CR air showers have notbeen detected, so it’s unlikely that the EAS cross sec-tion is large enough to produce such large amplitudeimpulses. Therefore such signals are dismissed a pri-ori. Our strategy is to choose the threshold and ampli-tude limiter level that gives high detection efficiency fora given SNR and low false alarm rate.

Two tests are conducted to determine the ideal am-plitude limiter level and the efficiency as a function ofMF threshold. The goal of the first test is to measure theaverage false alarm rate of the non-Gaussian noise en-vironment and evaluate the improvement that could beachieved by adding the amplitude limiter. Results areshown in Figure 24 for three different amplitude lim-iter levels, which clearly show that the limiter level hasa significant effect on the false alarm rate. Efficiencycurves for different amplitude limiter levels (describedin the next paragraphs) show that the amplitude limiterdoes not decrease detection performance of chirp sig-nals, although they are also clipped.

Consider the following interpretation of Figure 24. Inorder to achieve a 2 Hz false alarm rate, nγ has a value ofsix for k=3 and 9.5 for k=10 (black dashed line). Thus,detection thresholds can be decreased which enhancespositive detection of low SNR signals.

The second test applies a theoretical chirp signal withvarious chirp slopes and SNR values that correspond toa reasonable false alarm rate. Based on data storage andpost-processing computational requirements, we havedecided that a false alarm rate of ∼ 1 Hz is reasonable.Artificially generated chirp signals are transmitted insitu to the receiving antennas by an arbitrary waveformgenerator (AFG 3101; Tektronix, Inc.) and a dipole an-tenna. Both linear chirp signals and a simulated radarecho (see Section 2) are used in measuring detectionperformance.

7.4.1. Linear chirp signalA periodic, linear chirp with -1 MHz/µs slope is em-

bedded in a real receiver site background wave form.Figure 25 shows the time domain and spectrogram of

3 4 5 6 7 8 9 10 1110−3

10−2

10−1

100

101

102

103

Threshold (na)

FA

R (

eve

nts

/se

c)

k=10k=3k=1

Figure 24: False-Alarm Rate versus relative threshold (nγ units of thestandard deviation at each filter output) for different amplitude limiterlevels.

Figure 25: Linear, -1 MHz/µs, -10 dB SNR received chirp signalas recorded by the DAQ system. Left: time domain representation.Right: time-frequency (spectrogram) representation.

a chirp embedded with -10 dB SNR and -40 dB SCRvalue.

−30 −25 −20 −15 −10 −5 0 5 10 15 200

0.2

0.4

0.6

0.8

1

Signal−to−noise ratio (dB)

PC

D

No Amp. Limiterk=10k=3

Figure 26: Probability of detection for the matched-filter-type detectorwith γ = 6.

15

Figure 26 shows detection performance for a 2 Hzfalse alarm rate. Efficiency is shown for cases where theamplitude limiter is removed and at two different levelsthat result in the same false alarm rate, each with dif-ferent threshold levels. The minimum SNR for whichcomplete detection is achieved is 5 dB when no ampli-tude limiter is applied, 0 dB for k=10 (soft clipping),-6 dB for k=3 (hard clipping). These results imply thatby using the amplitude limiter, high detection perfor-mance can be achieved with low complexity. To maxi-mize detection ability, the amplitude limiter is currentlyset to k=3.

7.4.2. Simulated Air Shower

Figure 27: Spectrogram of simulated air shower radar echo with 5 dBSNR. The radar echo is from a simulated shower inclined 30 out ofthe T X → RX plane and located midway between the transmitter andreceiver.

In a more realistic test a simulated radar echo from a10 EeV air shower, inclined 30 out of the T X → RXplane and located midway between the transmitter andreceiver is transmitted to the receiving antennas. Fig-ure 27 shows a spectrogram of the received waveformwith 5 dB SNR and -25 dB SCR. The echo is broad-band (about 25 MHz) and short in duration (10 µs). De-tection efficiency of the emulated chirp is shown in Fig-ure 28. The minimum SNR for which complete detec-tion is achieved is -7 dB.

8. Remote Receiver Station

In addition to signal detection using matched filteringin the FlexRIO, an alternate technique is being devel-oped, based on signal detection at sites several kilome-

−30 −20 −10 0 10 200

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Signal−to−noise ratio (dB)

PC

D

Figure 28: Probability of correct detection for the matched-filter de-tector using γ = 6 for a simulated air-shower echo.

ters from the main LR site. Remote stations, by defini-tion, are generally subject to less radio interference, andadd stereoscopic measurement capabilities which the-oretically allow unique determination of CR geometryand core location. In contrast to the FlexRio system, amostly analog data acquisition system has lower powerconsumption at a cost which is also comparatively inex-pensive. Triggering for such a receiver station and somespecific details of hardware components are discussedin the next several subsections, followed by an analysisof noise sources at candidate future remote sites.

8.1. Remote Triggering

The alternative analog approach is based on an ana-log frequency mixer. The input signal is mixed witha delayed copy of itself i.e s(t) ⊗ s(t − τ). For an in-cident chirp signal, the non-linear components in themixer result in a product term that yields a monotoneat a beat frequency fbeat = βτ; dependent only on thedelay time τ and the chirp rate β. The delay is createdwith 100 ft of LMR-600 cable, which produces negligi-ble losses and removes the need for power consumingactive components. With appropriate filtering, the prob-lem of chirp detection is ultimately reduced to that ofdetecting the down-converted monotone. This is illus-trated in Figure 29. Portrayed here with an oscilloscopeis a -10 MHz/µs chirp which has been converted to a1 MHz monotone by mixing with a delayed copy of it-self.

After de-chirping, the signal is passed through an en-velope detector (8471D; Agilent, Inc.). The entire timedomain signal chain is illustrated in Figure 30. In thisoscilloscope based example, a chirp with 0 dB SNR

16

0 20 40 60 80 100Frequency(MHz)

180

160

140

120

100

80

Pow

er(d

Bm)

0 20 40 60 80 100Frequency(MHz)

200

180

160

140

120

100

80

Pow

er(d

Bm)

Figure 29: Top: the power spectrum of a -10 MHz/µs chirp created by asignal generator, prior to mixing. Bottom: the power spectrum of a 1 MHzmonotone signal after signal mixing and passing through a low pass filter.The chirp is evident as the left-most peak in this distribution.

at a rate of -1 MHz/µs is first band-pass filtered (41-100 MHz) and then amplified by 20 dB. The signalis then de-chirped followed by passing through a DC-1.9 MHz low pass filter and eventually passed throughthe Agilent power detector. We define the SNR as theamplitude of the input signal to the RMS of noise in thesystem.

The expected value of chirp slopes from EAS echosis typically between -1 to -10 MHz/µs (see Section 2).Consequently, with 100 ns delay, the down-convertedsignal has a frequency between 100 kHz and 1 MHz.To trigger on such signals, the de-chirped signal is splitinto multiple copies. Each copy is then passed throughcustom band-pass filters and an envelope detector. Dif-ferent frequency bands are then compared by majoritylogic in an FPGA, requiring no more than one bandto form a trigger in order to suppress impulsive noise.Each of the frequency banded outputs corresponds toseparate range of chirp rates. The block diagram in Fig-ure 31 outlines this triggering procedure.

0 20 40 60 80 100Time(micro-second)

0.6

0.4

0.2

0.0

0.2

0.4

0.6

Ampl

itude

(V)

0 20 40 60 80 100Time(micro-second)

0.15

0.10

0.05

0.00

0.05

0.10

0.15

Ampl

itude

(V)

0 20 40 60 80 100Time(micro-second)

0.005

0.000

0.005

0.010

0.015

0.020

Ampl

itude

(V)

Figure 30: Top: A 0 dB SNR and 1 MHz/µs chirp embed-ded in noise prior to de-chirping. Second from top: Themonotone signal after input chirp is mixed with delayedcopy of itself and passed through a low-pass filter. Bot-tom: Monotone passed through the Agilent 8471D powerdetector.

8.2. Remote StationThe layout of the full remote station, including the

Chirp Acquisition Module (CAM), power systems, ac-quisition electronics, and communications blocks isshown in Figure 32.

The Chirp Acquisition Module has a modular de-sign enabling quick debugging of the constituent com-ponents. This unit is comprised of a custom triggeringboard encompassing 4 band pass filters and envelope de-tectors, and a 4 channel ADC (AD80066; Analog De-vices) with 16 bit resolution sampling at 4 MSa/s perchannel to sample the signal out of the envelope detec-tors. A high speed ADC (AD9634 evaluation board;Analog Devices) sampling at 200 MSa/s directly sam-

17

TO

FPGA

+V

-V

+V

-V

BPF 41 - 88 MHz

BPFDC - 1.9 MHz

LO

RF

IF

Delay

Antenna

200 MSa/s

ADC

ADC

ADC

ADC

ADC

4 MSa/s

4 MSa/s

4 MSa/s

4 MSa/s

100ns

BPF A

BPF B

BPF C

BPF D

Figure 31: Block diagram of the event triggering to be employed in the remote station.

ples the raw data from the Antenna and an FPGA (Spar-tan - 6 LX16; Xilinx) performs the majority compar-ison logic to trigger and capture triggered data beforetransferring to a single boad computer. The single boardcomputer is a Raspberry Pi Rev. 2, that stores the trig-gered data in an SD card along with GPS time stamps(M12M; i-Lotus).

Component Power Consumption(W)Single Board Computer 5.0Low Speed 4 Ch. ADC 0.5High Speed 1 Ch. ADC 0.4FPGA 3.0RMS Counter 2.0System Health Monitor 1.060 dB Amplifier (x2) 4.025 dB Amplifier (x2) 0.4GPS 0.2GPS and GPS Antenna 0.4Communication Antenna 3.0Total 19.9

Table 5: Estimated power budget for the remote station.

Another major component is the System Health Mon-itor [38] (SHM), which both monitors performance andcontrols, via Solid State Relays (SSRs), the system solar(two 100 Watt photovoltaic panels) and battery (sealedlead acid) power. The SHM also records the antennadata digitized by the TDA receiver on local SD flash

memory. The TDA (Transient Detector Apparatus) re-ceiver has two channels with front-end amplifiers, fol-lowed by filters and a logarithmic amplifier. Finally,the SHM and CAM are connected to a 5 GHz ethernettransceiver via a switch for remote system control andaccess to the data.

8.3. Remote Station Prototype Studies

By definition, a remote station detector must be sus-tainably powered. To understand the required powerbudget (Table 5) from the perspective of solar resourcesin Western Utah, a prototype with system requirementssimilar to those of the full scale remote detector stationwas deployed at the Telescope Array Fluorescence De-tector site at Long Ridge, Utah in the spring of 2013.

The deployed hardware was comprised of a sys-tem health monitor (SHM) to monitor performance andprovision of power, four data acquisition channels, a12 W dummy-load and ethernet communications. Thefirst prototype remote site was deployed several hun-dred meters from the LR site. The four detector chan-nels include horizontal and vertical polarizations of thestandard TARA receiver LPDA, a spiral (frequency-independent) antenna, and 50 Ω terminator for compar-ison to system noise. The detector antennas feed intofour bulkhead connectors through LMR-400 coaxial ca-ble, where the signals are amplified and fed into TDAdetectors.

The TDA detectors record a hit when the TDA volt-age rises above a tunable threshold (set to 100 mV). In-

18

Figure 32: Schematic block diagram of the remote detector electronics. Chirp acquisition module (CAM), power systems, acquisition electronics,and communications blocks are all shown in this figure.

11

13.5

0 1 2 3 4 5 6 7 8Bat

t. V

olt

. (V

)

UTC days since Aug 19 (midnight)

0 25 50 75

100

0 1 2 3 4 5 6 7 8

Tem

p.

(C)

0 25 50 75

100

0 1 2 3 4 5 6 7 8So

lar

Po

w.

(W)

0

5

10

0 1 2 3 4 5 6 7 8

Sw

itch

Sta

t.

Figure 33: Remote station solar power data for August.

dependent of the presence of “hits”, the trigger rate isreported in software over regular 10 s intervals. Thesoftware controlling the detectors and power manage-ment of the station is located in micro-controllers on theSystem Health Monitor (SHM).

The SHM also supports remote communications,however the prototype station was connected to the FDfacilities through 200 m of CAT6 ethernet cable, withpower-over-ethernet (PoE) offering ample capability toinclude system monitoring.

The SHM unit recorded a series of environmental

variables in addition to antenna TDA voltage rates. Theprimary variables are solar panel current and voltage,battery voltage, the status of the SSRs (the dummyload), temperature measurements, and support for ananemometer. The prototype remote station recorded so-lar panel power throughout the summer of 2013, quan-tifying the amount of solar energy available over time.Figures 33 displays the results. Each day, the stationconsumed approximately 13 W, or ≈ 26 AHr over thecourse of 24 hours. There are data gaps due to lapsesin connectivity to the site, however the data was storedlocally as well.

The data show a clear diurnal variation. With the100 W rated solar panel oriented towards the sun at12 p.m. on June 1st, approximately 75 W peak powerdelivery was observed. The curves have a full-widthhalf-maximum of 3.6 hr, meaning the station collected21.5 AHr per day. Thus, after 40 days, the station be-gan to switch off the dummy load at night via the SSRs,as the battery buffering was depleted. The y-axis of theupper graph is a binary number representing the switchstatus. The number 10 (1010 in binary) indicates thatboth the solar power and dummy load are connected.The number 2 (0010 in binary) indicates that the dummyload switch has been disconnected by the SHM.

19

8.4. Noise Studies

In preparation for the deployment of remote stationswith the analog trigger, the noise characteristics of anarray of remote sites were studied in March, 2013. TheGPS coordinates of the sites studied are given in Table 6.Sites were chosen to be at least 2 km from the FD site atLong Ridge, in the desert, to minimize any local noiseinterference. The second criterion for each site was thatthe TARA transmitter had to be detectable, given thatthe transmitter beam width limits the lateral range ofthe remote sites.

Site Latitude (N) Longitude (W)FD (Long Ridge, UT) 39 12’ 28” 113 17’ 16”

Front Site 39 13’ 1” 113 5’ 58”Site G 39 11’ 40” 113 6’ 22”Site H 39 11’ 15” 113 5’ 32”Site I 39 11’ 36” 113 5’ 40”Site J 39 11’ 34” 113 8’ 56”Site K 39 12’ 37” 113 9’ 37”Site U 39 13’ 19” 113 6’ 46”Site T 39 13’ 25” 113 7’ 39”

Table 6: GPS coordinates of the eight sites at which background mea-surements were taken, in the vicinity of the FD at Long Ridge. Ateach site the TARA transmitter was detected above the thermal noisefloor.

At each of the indicated sites, a standard TARA re-ceiver LPDA was connected to 70 dB of amplificationand a spectrum analyzer. If a CW noise source at a givensite exceeded 10 dBm (the maximum allowed power bythe analyzer), a 10-20 dB limiter was inserted. The largeamplification was employed to ensure that the naturalthermal noise floor was reached and visible above theintrinsic noise in the measurement system. Spectra wererecorded in the four cardinal directions at each site. Toanalyze these data in the presence of many CW noisespikes, a peak finding algorithm was used to record thepeak power and frequency of each CW source at theFD site, as long as the amplified power was larger thanan analysis threshold of -30 dBm. The power at thosesame frequencies was then measured for each directionat each remote site. The measurements are presented inFigure 34.

There are two major conclusions that may be drawnfrom Figure 34. First, at the frequencies measured at theremote sites, CW sources are almost always 20-50 dBlower in power compared to the FD site. There are rela-tively few data points in the field that display power thatis large compared to the amplified noise background(in this case, -65 dBm). Thus, we conclude that the

Figure 34: The four graphs shown correspond to an LPDA receiverheading of 80, 170, 250 and 350 during the remote noise studies. Theangular bins were chosen at the first site for convenience, and retainedfor consistency. Each dot corresponds to a CW source measured atthe FD site. The color scale is the amplified power in dBm. The dotsclustered around the remote site GPS coordinates represent the powermeasured at the same frequencies as the FD site noise sources. Eachdot corresponding to a remote site measurement is plotted at a fixedradius from the true coordinates.

remote sites typically represent excellent RF environ-ments. The second major conclusion is the dominancein the number of measured noise sources in the North-South direction relative to the East-West direction. TheNorth and South orientations of the antenna align withhuman traffic and radio communications between localcommunities and Salt Lake City, whereas the TARAtransmitter is located to the East at a heading of 80.Thus, the TARA transmitter-receiver systems is opti-mized to look only in directions with minimal noise.

9. Conclusion

The TARA project represents the most ambitious ef-fort to date to detect the radar signature of cosmic rayinduced atmospheric ionization. These signals will becharacterized by their low power, large Doppler shift(several tens of MHz), and short duration (< 15 µsec).TARA combines a high-power transmitter with a state-of-the-art high sampling rate receiver in a low-noise en-vironment in order to maximize the likelihood of cos-mic ray echo detection. Importantly, TARA is co-located with the Telescope Array astroparticle observa-tory, which will allow for definitive confirmation thatany echoes observed are the result of cosmic ray inter-actions in the atmosphere.

20

10. Acknowledgments

This work is supported by the U.S. National ScienceFoundation grants NSF/PHY-0969865 and NSF/MRI-1126353, by the Vice President for Research of theUniversity of Utah, and by the W.M. Keck Foundation.L. B. acknowledges the support of NSF/REU-1263394.We would also like to acknowledge the generous dona-tion of analog television transmitter equipment by SaltLake City KUTV Channel 2 and ABC Channel 4, andthe cooperation of the Telescope Array collaboration.

We would like to specifically thank D. Barr andG. McDonough from Telescope Array for their services.

References

[1] P. Auger, et al., Extensive Cosmic-Ray Showers, Rev. Mod.Phys. 11 (1939) 288–291.

[2] T. Abu-Zayyad, et al., The Surface Detector Array of the Tele-scope Array Experiment, Nucl. Instrum. Meth. A 689 (2012) 87– 97.

[3] I. Allekotte, et al., The Surface Detector System of the PierreAuger Observatory, Nucl. Instrum. Meth. A 586 (2008) 409–420.

[4] H. Tokuno, et al., New Air Fluorescence Detectors Employedin the Telescope Array Experiment, Nucl. Instrum. Meth. A 676(2013) 54–65.

[5] J. Abraham, et al., The Fluorescence Detector of the PierreAuger Observatory, Nucl. Instrum. Meth. A 620 (2010) 227–251.

[6] P.M.S. Blackett, et al., Radio Echoes and Cosmic Ray Showers,Proc. Roy. Soc. A 177 (1941) 183–186.

[7] D. Wahl, et al., The Search for Vertical Extended Air ShowerSignals at the Jicamarca Radio Observatory, in: Proceedings ofthe 30th ICRC, volume 5, 2008, pp. 957–960.

[8] T. Terasawa, et al., Search for Radio Echoes from EAS with theMU Radar, Shigaraki, Japan, in: Proceedings of the 31st ICRC,2009.

[9] M.F. Bugallo, et. al, MARIACHI: A Multidisciplinary Effort toBring Science and Engineering to the Classroom, in: IEEE In-ternation Conference on Acoust., Speech, Signal Process., 2008,pp. 2661–2664.

[10] M.F. Bugallo, et. al, Hands-on Engineering and Science: Dis-covering Cosmic Rays Using Radar-based Techniques and Mo-bile Technology, in: IEEE Internation Conference on Acoust.,Speech, Signal Process., 2009, pp. 2321–2324.

[11] D.G. Underwood, Large Doppler Shift in RADAR Detectionof Ultra-high Energy Eosmic Rays, in: Radar Conference,RADAR ’08. IEEE, 2008, pp. 1–5.

[12] M.I. Bakunov, et al., Relativistic Effects in Radar Detection ofIonization Fronts Produced by Ultra-high Energy Cosmic Rays,Astropart. Phys. 33 (2010) 335–340.

[13] N.J. Willis, Bistatic Radar, Scitech Publishing, 1995.[14] A.A. Watson, High Energy Cosmic Rays and the Greisen-

Zatzepin-Kuzmin Effect, Rep. Prog. Phys. 76 (2013).[15] K. Kampert, et al., Measurements of the Cosmic Ray Composi-

tion with Air Shower Experiments, Astropart. Phys. 35 (2012).[16] N.M. Budnev, et. al, The Tunka Experiment: Towards a 1-km-

square EAS Cherenkov Light Array in the Tunka Valley, Int. J.Mod. Phys. A 20 (2005).

[17] The Pierre Auger Collaboration, Correlation of the Highest-Energy Cosmic Rays with Nearby Extragalactic Objects, Sci-ence 318 (2007).

[18] G. Askaryan, Excess Negative Charge of an Electron-PhotonShower And Its Coherent Radio Emission, Sov. J. Exp. Theor.Phys. 14 (1962).

[19] M.P. van Haarlem, et al., LOFAR: The Low-Frequency Array,Astron. Astrophys. 556 (2013).

[20] F.G. Schroder, et al., Tunka-Rex: A radio antenna array forthe Tunka experiment, in: Proceedings of the AIP Conference,2013.

[21] T. Huege, Simulating Radio Emission from Air Showers withCoREAS, arxiv:1301.2132 (2013).

[22] P. Abreu, et. al, Antennas for the Detection of Radio Emis-sion Pulses from Cosmic-ray Induced Air Showers at the PierreAuger Observatory, J. Instrum. 7 (2012).

[23] P. Gorham, On the Possibility of Radar Echo Detection of Ultra-high Energy Cosmic Ray - and Neutrino - induced Extensive AirShowers, Astropart. Phys. 15 (2001).

[24] R.J. Vidmar, On the Use of Atmospheric Pressure Plasmas asElectromagnetic Reflectors and Absorbers, IEEE T. Plasma Sci.18 (2002).

[25] Y. Itakawa, Effective Collision Frequency of Electrons In At-mospheric Gases, Planet. Space Sci. 19 (1971) 993.

[26] Y. Itakawa, Effective Collision Frequency of Electrons In Gases,Phys. Fluids 16 (1973) 831.

[27] J.W. Crispin and A.L. Maffett, Radar Cross-Section Estimationof Simple Shapes, P. IEEE 53 (1965) 833–848.

[28] T.K. Gaisser and A.M. Hillas, in: Proceedings of the 15th ICRC,volume 8, 1977, p. 353.

[29] H. Takai, et al., Forward Scattering Radar for Ultra High EnergyCosmic Rays, in: Proceedings of the 32nd International CosmicRay Conference, 2011.

[30] A. Zech, A Measurement of the Ultra-High Energy Cosmic RayFlux with the HIRES FADC Detector, Ph.D. thesis, Rutgers Uni-versity, 2004.

[31] G. Burke, et al., NEC - Numerical Electromagnetics Code forAntennas and Scattering, in: Antennas and Propagation SocietyInternational Symposium, volume 17, 1979.

[32] R.U. Abbasi, et al., Indications of Proton-Dominated Cosmic-Ray Composition Above 1.6 EeV, Phys. Rev. Lett. 104 (2010).

[33] J. Kraus, R. Marhefka, Antennas For All Applications, 3 ed.,MgGraw-Hill, 2003.

[34] H.V. Cane, Spectra of the Non-thermal Radio Radiation fromthe Galactic Polar Regions, Mon. Not. R. Astron. Soc. 189(1979).

[35] G.A. Dulk, et al., Calibration of Low-frequency Radio Tele-scopes Using the Galactic Background Radiation, Astron. As-trophys. 365 (2001) 294–300.

[36] M. Othman, et al., On Radar Detection of Chirp Signalswith Nondeterministic Parameters in Challenging Noise Back-ground, in: Radar Conference (RADAR), 2013 IEEE, 2013, pp.1–6.

[37] J.D. Smith, et al., GPSY2: A Programmable Hardware Modulefor Precise Absolute Time Event Generation and Measurement,in: Proceedings of the 30th ICRC, volume 5, 2008, pp. 997–1000.

[38] D. Besson, D. Kennedy, K. Ratzlaff, R. Young, Design, mod-eling and testing of the askaryan radio array south pole au-tonomous renewable power stations, 2014. Submitted to Phys.Rev. D.

21