Möglichkeiten und Grenzen maritimer [627474]

Möglichkeiten und Grenzen maritimer
Anwendungen von Bodenwellen-
Überhorizontradaren:
Eine Systemstudie zu verteiltem
selbstkonfigurierenden HF-Überhorizontradar auf
Schiffen

Beauftragung durch WTD 71

E/E71S/9T307/9F131

Abschlussbericht 29.07.2013

Lehrstuhl für Hochfrequenztechnik
Adj. Ass. Prof. (Griffith Univ.) Thomas Fickenscher
B.Tech. Anshu Gupta

Lehrstuhl für Allgemeine Nachrichtentechnik
Univ.-Prof. Dr.-Ing. habil. Udo Zölzer
Dr.-Ing. Martin Holters
M.Sc. MBA Jan Oliver Hinz

I

Universität der Bundeswehr Hamburg Page II
Bitte Ausfüllanweisung auf der
Rückseite beachten Berichts-/Erschließungsblatt

Dieses Berichts-/Erschließungsblatt wird unabhängig vom Geheimhaltungsgrad des Dokuments immer als "of fen" behandelt, deshalb
sind keine geheimschutzwürdigen Angaben zu übernehm en.
1 Nur vom DOKFIZB w auszuf üllen Lfd Nr
2 BerichtsNr des Herausgebers/Auftragnehmers (AN) ( vollständige Buchstaben-/Ziffernfolge
3 BerichtsNr des Auftraggebers (AG)
Titel/Untertitel (VS-eingestuften Titel fingieren " ….."; bei
mehrbändigen Dokumenten BdNr und zutreffenden
Einzeltitel angeben) Systemstudie zu verteiltem selbstkonfigurierenden H F-Überhorizontradar auf
Schiffen 4
Kurztitel (max. 30 Stellen) Systemstudie HF-Überhor izontradar
4a Ins Englische übersetzter Titel/Untertitel
System study of distributed self configuring HF O ver-the-Horizon-Radar
5 Autor(en) (Name, Vorname(n) oder Institution als körperschaftlicher Urheber)

Fickenscher Thomas, Gupta Anshu, Hinz Jan O liver, Holters Martin, Zölzer Udo
6 Auftragnehmer (AN) (Institution(en), Abteilung, O rt/Sitz der beteiligten AN, SubAN, MitAN)
Helmut-Schmidt-Universität / Universität de r Bundeswehr Hamburg, Lehrstuhl für Hochfrequenztec hnik und Lehrstuhl für
Allgemeine Nachrichtentechnik
7 Auftraggeber (AG) / Aufgabesteller (ASt) / Fachli ch zuständige Stelle
AG: WTD 71
ASt: WTD 71 FWG 710
8 Kurzreferat (Inhaltsbeschreibung des Dokuments)
Möglichkeiten und Grenzen maritimer Anwendungen vo n Bodenwellen-Überhorizontradaren.
9 Schlagwörter (Schwerpunktartige Inhaltskennzeichn ung mittels Fachbegriffen)
HF-Radar, Surface Wave Radar, Over-the-Horizon-Rad ar, MIMO-Radar, Radar Signal Processing, Ship Detec tion, CFAR Detector,
Clutter Reduction, Clutter Canceller Beam Former, A rray Signal Processing, Electrical Small Antenna
10 DSt und StO (Dokument langfristig verfügbar, Auslei he)

11 Geheimhaltungsgrad Offen VS-NfD VS-Vertr. GEHEIM
12 Gesamtseiten-/blattzahl 70 13 Quellen 68
14 Tabellen 1 15 Statistiken 0
16 Techn. Zeichnungen 0 17 Abbildungen 42
18 Berichtsdatum 29.07.2013
19 Berichtsart (z.B.Zwischen-/Abschlußbericht, vgl . Feld 27) Abschlußbericht
20 Auftrags-/VertragsNr des AG (vollständige Buchst aben-/Ziffernfolge) E/E71S/9T307/9F131
21 Auftragserteilung/Vertragsabschluss 30.04.2009 2 2 Abschlussdatum/Vertragsende 31.03.2014
23 Projekt-/Programm-/Konzeptbezeichnung (z.B. ZTL 1979, FAG 1, MBB 1-85-1)
Systemstudie zu verteiltem selbstkonfigurierenden H F-Überhorizontradar auf Schiffen
24 Studien-/Aufgabenkennziffer, DateiblattNr (SKZ:, AKZ:, DateiblattNr)
9F131
25 Aktenzeichen des AG/Herausgebers oder der fachli ch zuständigen Stelle 900-310/E/E71S/9T307/9F131
26 Sperrvermerke
27 Zusätzliche Angaben/Hinweise

Universität der Bundeswehr Hamburg Page III

Universität der Bundeswehr Hamburg Page IV Table of Contents

1A Zusammenfassung … …………………… ….… ..………………………………1
1B Executive Summary … ..……………………….… …… ……………… ……….5

2 Review of Mobile and Compact HF Surface Wave Rada r …….…….…… …….9
2.1 Literature Overview…….…….…………………………………………. ….9
2.2 Key Challenges in Design of Distributed ship borne HF SWR………… …11

3 Modelling of Quasi-Monostatic FMCW HF Surface Wa ve Radar………………15
3.1 Propagation…….…….……………………………………………… …….15
3.2 Sea Clutter…………………………………………………………… ..….16
3.3 Signal Processing……………… ……………………………………… ….19

4 Compact Coastal Antenna Array Using MIMO Concept… ……… …….… …..21
4.1 Summary…………………………………………………………………..21
4.2 Motivation………………………………………………………… ..…….21
4.3 Concept of MIMO Beamforming…………… .……………………… ..….22

5 Large Sparse Sea Floating Antenna Arrays Using MI MO Concept… …………..25
5.1 Summary…………………………………………………………………..25
5.2 Sparse Array Distributed Among Naval Formati on………………… ..…..26
5.3 Implementation of MIMO Concept in FMCW Radar ………… ..………… 28
5.4 Compensation of Pattern Distortion caused by Di splacement due to ………….
Wind and Waves… …… ……………………… ……………………….….30

6 Clutter Canceller Beamformer…… …….…………………… ………….……..33
6.1 Summary…………………………………………………………………..33
6.2 Principle of CCBF…………………………… …………………………….33
6.3 Measurement Results……………………………………………… .…….35
6.4 Impact of Antenna Pattern Distortion…………………… ………… .…… 36

7 Electrically Small HF Antenna………………………………….…… ..………41
7.1 Summary…………………………………………………………………..41
7.2 Antenna Structure……………………………………………………… …41
7.3 Simulation and Measurement Results………………………… …..………43

8 Novel Sub-Clutter Detector….…… .…….……………………….……………47
8.1 Summary…………………………………………………………… .…… 47
8.2 Detector Architecture……………………………………………… .…….47
8.3 Measurement Results……………………………………………… .…….49

9 Improved CFAR Based Target Detection………………………… ..……..…..53

Universität der Bundeswehr Hamburg Page V 9.1 Resolution Capabilities……………… ……………………………… .…..53
9.2 Short Introduction to CFAR in general………………………… …….…..54
9.3 Presegmentation………………………… ……………………… ..……… 54
9.4 Scan-by-Scan Averaging………………………………………… ..………56
9.5 Adjacent Detection Merging and Target Parame ter Estimation…..……… 57

10 Future Prospects……………………………………….…………… ..………61

List of Abbreviations……… ……………………………………………… …… 63

Universität der Bundeswehr Hamburg Page 1 1A Zusammenfassung
Über gut leitendem Meerwasser folgt eine vertikal p olarisierte elektromagnetische Welle im
HF-Band (3 – 30 MHz) der Krümmung der Erde und ermö glicht damit einem HF
Oberflächenwellen Radar (HF SWR) über den Horizont hinaus bis über 100 Kilometer zu
blicken. HF SWR finden Anwendungen in verschiedenen Bereichen wie Ozeanographie,
maritimer Überwachung oder zur Erkennung niedrig fl iegender Ziele. Gegenüber flugzeug-
oder satellitengestützter Techniken bieten HF SWR d en Vorteil wesentlich niedrigere Kosten
pro km² Überwachungsgebiet. Sie können zur Aufkläru ng militärischer Bedrohungen und der
ganzen Palette von Bedrohungen durch internationale n Terrorismus und Piraterie sowie
illegaler oder unerwünschter Migration, illegaler F ischerei, Quarantäneverletzung und
Drogenschmuggel eingesetzt werden. Die Alternative einer luftgestützten Überwachung bietet
trotz höherer Kosten kein Echtzeit-Lagebild der ges amten Küste, da ein Flugzeug in der Regel
nur im Abstand von einigen Tagen ein Seegebiet über fliegt. Zudem wird die Luftaufklärung
durch schlechtes Wetter und die satellitengestützte optische Aufklärung (gilt nicht für Radar
mit synthetischer Apertur SAR) durch eingeschränkte Sicht behindert. Im vorliegenden
Bericht werden die Möglichkeiten der Detektion von Schiffen durch kleine küstenbasierte HF
SWR sowie durch mobile auf Schiffen oder schwimmend en Plattformen installierbarer HF
SWR untersucht.

Anders als bei Mikrowellenradaren wird im HF Band u nter der V oraussetzung ausreichender
Sendeleistung die Detektion von Schiffen durch den See-Clutter (Radarrückstreuung durch die
Meeresoberfläche) und weniger durch internes Empfän gerrauschen begrenzt. Somit ist die
Reduzierung des See-Clutter eine der wichtigsten He rausforderungen bei der Detektion
langsam bewegter Ziele wie Schiffe. Das See-Clutter -Signal verringert sich mit zunehmender
räumlicher Auflösung (Azimut und Entfernung) sowie mit zunehmender Dopplerauflösung
(Geschwindigkeitsauflösung) des Radars. Aufgrund de r hohen Wellenlänge im HF-Band von
10 – 100 m ist die Azimutauflösung des Radars durch die mögliche Ausdehnung der
Antennenapertur begrenzt (z. B. notwendige Länge ei nes 16-Element-Arrays in der Mitte des
HF-Bands ca. 200 m). Die Entfernungsauflösung wird durch die Bandbreite des Radars
begrenzt. Es ist sehr schwierig, einen entsprechend en freien Kanal innerhalb des HF-Bands zu
finden. Die Dopplerauflösung wird durch die kohären te Integrationszeit begrenzt. Letztere ist
wiederum durch die maximale Geschwindigkeit des Zie ls limitiert. Daher liegt der
Schwerpunkt dieser Studie in der Untersuchung neuer Ansätze zur Steigerung des Signal-zu-
Clutter-Verhältnisses (SCR) für realistische Szenar ien, zur Verbesserung der Detektion durch
Clutter maskierter Ziele (sub-clutter visibility) u nd zur Reduktion der Größe des
Antennenarrays für kompakte küstenbasierte oder gar schiffsgestützte
Oberflächenwellenradare.

Der Einsatz des Multiple-Input Multiple-Output (MIM O)-Konzeptes ermöglicht virtuelle
Antennenposition zu generieren und damit die Synthe se eines größeren Antennenarrays. So
kann nach dem MIMO-Prinzip durch Hinzufügen eines z weiten Senders die effektive Länge
eines herkömmlichen Empfängerarrays verdoppelt werd en. Dieses Konzept wurde in ein
kommerzielles küstenbasiertes HF SWR System (Remote Ocean Sensing System WERA,
Helzel Messtechnik GmbH, Kaltenkirchen, Deutschland ) implementiert und getestet. 6
Elemente eines herkömmlichen 12-Element Empfangsarr ays wurden für ein MIMO-Radar mit
zwei Sendern genutzt. Die Messergebnisse wurden mit denen verglichen, die mit dem
gesamten physikalischen 12-Element Array in Verbind ung mit einem einzelnen Sender
gewonnenen wurden. Die Präsentation dieser Ergebnis se auf einem Kongress von

Universität der Bundeswehr Hamburg Page 2 Ozeanographen hat zu einer unmittelbaren Nachfrage für eine entsprechende
Produktentwicklung geführt.
Das MIMO-Prinzip kann auch eingesetzt werden, um ei n großes virtuelles Array aus auf
mehreren Schiffen oder schwimmenden Plattformen ver teilten Antennen zu synthetisieren. Ein
Array, das auf mehrere Schiffe eines Verbandes vert eilt ist, ist dünn besetzt. D. h. der Abstand
zwischen benachbarten Antennenelementen, welche sic h auf verschiedenen Schiffen befinden,
ist größer als eine halbe Wellenlänge ( λ/2). Dies führt zu sog. Gitterkeulen. Damit nimmt d ie
Empfangsantenne nicht nur aus der Hauptstrahlrichtu ng Signale auf, sondern mit der gleichen
Empfindlichkeit auch aus weiteren unerwünschten Ric htungen. Unter Berücksichtigung des
Konzeptes der effektiven Apertur bei der Festlegung der relativen Antennenabstände können
diese Gitterkeulen vermieden werden. Das Konzept de r effektiven Apertur ist besonders
effizient, wenn es mit MIMO-Strahlformung kombinier t wird. Hierbei können die Amplituden
und Phasen für die einzelnen Signalwege zwischen de n Sende- und Empfangs-
Antennenelementen gesteuert werden, so dass letztli ch der Strahl des synthetisierten virtuellen
Arrays wie der einer realen phasengesteuerten Grupp enantenne geformt werden kann. Da im
HF-Band die Detektion von Schiffen durch den See-Cl utter und nicht durch internes
Empfängerrauschen begrenzt ist, ist die mit dem MIM O-Prinzip verknüpfte Verringerung der
Empfindlichkeit weniger bedeutsam.
Die Antennendiagramme auf schwimmenden Plattformen installierter Antennenarrays werden
durch relative gegenseitige horizontale Verschiebun g sowie durch Neigung der
Antennenelemente infolge des Einflusses von Wind un d Wellen verzerrt. Zur Korrektur der
Verzerrungen des Horizontaldiagramms des virtuellen Arrays wurde ein echtzeitfähiges
Verfahren entwickelt, welches aus der aktuellen Pos ition und Neigung der physikalischen
Antennenelemente die notwendige Amplituden- und Pha senbelegung der virtuellen
Antennenelemente ermittelt.

Die spektralen Eigenschaften des See-Clutter im HF- Band unterscheiden sich signifikant von
denen im Frequenzbereich der Mikrowellen. Wellen mi t einer Wellenlänge gleich der halben
Radarwellenlänge führen zu einer starken Bragg-Stre uung. Mit Kenntnis der Betriebsfrequenz
und Peilung eines HF SWR hat ein Schiff die Möglich keit, sich in den charakteristischen
Spektrallinien der Bragg-Resonanz erster Ordnung de s See-Clutter (Bragg-Linien) zu
verstecken. Das derzeit am meisten verbreitete Verf ahren zur Unterdrückung von See-Clutter
bei Überhorizontradaren nutzt Space Time Adaptive P rocessing (STAP) um die resonante
Rückstreuung des Meeres in Raum, Zeit und Azimut-Do mäne zu filtern. Unsere
Betrachtungen zeigen jedoch, dass aufgrund der spek tralen und statistischen Natur des See-
Clutter-Signals STAP im Zusammenhang mit HF SWR nur bedingt erfolgreich ist. Dies gilt
insbesondere für den Einsatz auf mobilen schwimmend en Plattformen. Darüber hinaus ist
dieses Verfahren nicht hilfreich, Ziele zu detektie ren, die sich in den Bragg-Linien verstecken.
STAP filtert hier Nutz- und Störsignal gleichermaße n.

Es wurde ein neuartiger Clutter-Canceller-Strahlfor mer (CCBF) untersucht, der das See-
Clutter-Signal in der Azimut-Domäne filtert und dab ei das Signal-zu-Clutter-Verhältnis SCR
verbessert. Nebenbei hat dieser CCBF gegenüber eine m herkömmlichen Chebyshev-
Strahlformer den V orteil, robuster gegenüber Verzer rungen des Antennendiagramms zu sein,
welche infolge Dislokation einzelner Antennenelemen te (z.B. verteiltes schwimmendes Array)
oder aber infolge von Störungen im Nahfeld der Ante nne (z.B. Antennenintegration auf einem
Schiff) auftreten. Unter der Annahme eines homogene n Hintergrundrauschens ergibt sich für
ein Punktziel eine Steigerung des SCR von 2,48 dB i m Vergleich zu einem herkömmlichen 16-
Element Chebyshev-Strahlformer. Unter Verwendung vo n Messdaten einer WERA Station mit
einem 16-Element Empfängerarray wurde der CCBF erfo lgreich getestet. Erstmalig in der
offenen zugänglichen Literatur konnte die Detektion eines Schiffes innerhalb einer Bragg-Line

Universität der Bundeswehr Hamburg Page 3 nachgewiesen werden. Der vorgeschlagene Strahlforme r ist effektiv in der Unterdrückung von
See-Clutter und weniger rechenintensiv als STAP.

Neben der Größe des Empfangsarrays ist die Größe de r einzelnen Antennenelemente selbst
eine weitere Einschränkung von heutigen HF SWR Syst emen. Typischerweise werden in
küstenbasierten Arrays λ/4 Monopole verwenden, deren Länge von 2,5 – 25 m v ariiert.
Kleinere HF-Antennen mit großer Betriebsbandbreite und hohem Wirkungsgrad sind höchst
wünschenswert. Sie können getarnt am Strand oder au f kleinen schwimmenden Plattformen
eingesetzt werden. So wurde eine kapazitiv beladene , induktiv gekoppelte selbstresonante
elektrisch kleine Antenne von 1,45 m Höhe und 1,6 m Breite entwickelt und getestet.
Simulations- und Messergebnisse belegen eine große Bandbreite, einen hohen Wirkungsgrad
und gute Abstimmbarkeit im mittleren HF Band (10,7 – 23 MHz). Die Güte liegt sehr nahe an
der durch das Chu-McLean-Limit gegeben physikalisch en Grenze. Das bedeutet, dass eine
weitere Verringerung der Antennenabmessungen zwangs läufig zu einer niedrigern Bandbreite
und/oder niedrigerem Wirkungsgrad führen würde. Das Bandbreiten-Wirkungsgrad-Produkt
der vorgeschlagenen Antenne ist fast viermal höher als bislang in zugänglichen Publikationen
veröffentlicht. Die spiralförmige Ausführung der Da chkapazität der Antenne trägt der in der
Praxis auftretenden starken Windlast Rechnung. Alle genannten Eigenschaften lassen diese
Antenne ideal für den Einsatz in kompakten küsten- oder schiffsbasierten HF SWR
Anwendungen erscheinen.

Mikrowellen-Radare weisen einen definierten interne n Rauschpegel auf. Mittels eines
Constant False Alarm Rate (CFAR)-Detektors kann dan n für eine vom Benutzer vorgegebene
konstante Fehlalarm-Wahrscheinlichkeit die Detektio ns-Wahrscheinlichkeit maximiert
werden. Im HF-Band ist das Empfängerrauschen hingeg en wesentlich geringer, so dass das
See-Clutter-Signal – dessen statistischen Eigenscha ften nicht genau bekannt sind – als
dominierende Störung in den V ordergrund tritt. Aktu elle Detektoren verwenden einen
dreidimensionalen CFAR-Algorithmus um die Detektion sschwelle adaptiv entsprechend dem
stark inhomogenen Amplitudenverlauf des Clutter-Sig nals entlang Azimut, Entfernung und
Doppler nachzuführen. Eine konstante Fehlalarm-Wahr scheinlichkeit kann aber unter diesen
Bedingungen nicht erzielt werden.
Wir schlagen ein neues zweistufiges Verfahren vor. In einem ersten Schritt wird ein Ziel
mittels eines Korrelationsverfahrens in der Entfern ungs-Doppler-Ebene detektiert. In einem
zweiten Schritt wird mittels der Methode des minima len mittleren quadratischen Fehlers der
Azimut des Ziels geschätzt. Unter Verwendung von Me ssdaten einer WERA Station wird
belegt, dass dieser Detektor Ziele in See-Clutter d ominierten Szenarien detektieren kann.
Diese Ergebnisse werden durch die Auswertung der Ec htzeitpositionen der Schiffe aus
Datensätzen des Automatic Identification System (AI S) bestätigt .

Des Weiteren wird alternativ ein modifizierter CFAR Detektor untersucht. Herkömmliche
CFAR-Detektoren bestimmen die Detektionsschwelle fü r jede zu testende Entfernungs-
Doppler-Azimut-Zelle unter der Annahme eines nahezu homogenen Hintergrundes (See-
Clutter und Rauschen). Somit kann die Detektionssch welle unter Verwendung der
Amplitudenwerte benachbarter Zellen geschätzt werd en. In der Nähe der charakteristischen
Spektrallinien der Bragg-Resonanz erster Ordnung de s See-Clutter kann hingegen nicht von
einem homogenen Hintergrund ausgegangen werden. Dah er wurde der CFAR Detektor
dahingehend modifiziert, dass zunächst die Entfernu ngs-Doppler-Azimut-Zellen geeignet in
clutter- bzw. rauschdominierte Bereiche unterteilt werden, mit jeweils individuell angepassten
Algorithmen zur Bestimmung der Detektionsschwelle.
Eine weitere Verbesserung wird durch die Berücksich tigung mehrerer aufeinanderfolgender
Scans durch Scan-by-Scan-Mittelung erreicht. Schlie ßlich wird, um Mehrfach-Detektionen

Universität der Bundeswehr Hamburg Page 4 eines Einzelziels zu vermeiden, ein Verschmelzungsa lgorithmus für benachbarte Detektionen
entwickelt. Diese Erweiterungen des herkömmlichen C FAR Verfahrens stellen eine deutliche
Verbesserung für die betrachtete Anwendung dar.

Universität der Bundeswehr Hamburg Page 5 1B Executive Summary

Over highly conductive sea water a vertically polar ized electromagnetic wave at HF band (3-
30 MHz) follows the curvature of the earth enabling HF Surface Wave Radars (HF SWR) to
look over the horizon up to more than 100 kilometer . HF SWR find applications in various
fields such as oceanography, maritime surveillance, or detection of low flying targets.
Compared to airborne or space-based techniques HF S WR offers much lower cost per square
km of surveillance area and can be used to cope wit h naval insurgency and the whole panoply
of threats posed by international terrorism and pir acy as well as illegal or undesirable
migration, poaching of fish resources, quarantine v iolation and drugs smuggling. Besides the
aspect of costs airborne surveillance does not off er real time information of the whole littoral
since a plane will fly around a patch only once in a few days. Secondly bad weather can
severely compromise with airborne surveillance as d oes limited visibility in case of optical
space-based techniques (does not apply to Synthetic Aperture Radar SAR). Therefore, the
scope of this report is to investigate possibilitie s for ship detection by small coastal HF SWR
as well as by mobile shipborne or floating HF SWR p latforms.

Provided sufficient transmit power is applied in HF band detection of ships is limited by sea
clutter rather than by internal receiver noise. Thu s, reduction of sea clutter is one of the main
challenges in detection of slow moving hard targets like ships. Sea clutter signal reduces with
increasing spatial resolution (azimuth and range) a s well as with increasing Doppler resolution
of the radar. Wavelength at HF band is 10 – 100 m and as space for antenna arrays is limited
so is the azimuth resolution of the radar (e.g. the length of a typical 16-element receive array
operating in mid HF band is in the order of 200 m). Range resolution is limited by radar
bandwidth and it is very difficult to find continuo us clear channels within the HF band. Finally,
Doppler resolution is limited by the coherent integ ration time (CIT). The maximum speed of
the target imposes an upper limit on it. Therefore, the focus of this study is to investigate new
approaches to increase the signal to clutter ratio (SCR) for realistic scenarios, to improve sub-
clutter visibility and to reduce the size of the an tenna array for more compact coastal or even
shipboard applications.

Multiple-Input Multiple-Output (MIMO) concept enabl es to synthesize virtual antenna
positions which result in a larger number of effect ive array elements. By adding a second
transmitter MIMO concept can be used to double the effective length of a conventional
receiver array. This approach has been implemented and tested on a commercial coastal HF
SWR system (Remote Ocean Sensing System WERA, Helze l Messtechnik GmbH,
Kaltenkirchen, Germany). 6 elements of a commercial 12 element receive array have been
used to demonstrate the MIMO concept enabling a dir ect comparison with the results obtained
with the physical 12 element array. Results present ed among oceanographic community have
driven immediate demand for product development.
For shipborne applications MIMO concept can also be used to overcome the challenges of
array size by making use of several ships or mobile platforms sailing within a naval formation.
An array distributed among a naval formation will b e sparse in nature, i.e. the inter-element
spacing between antennas located on different ships will be more than one-half of a
wavelength ( λ/2). This will result in gratinglobes, sidelobes of the antenna array which have
the same magnitude response as the mainlobe. This i s undesirable for a radar system. Appling
the concept of effective aperture gratinglobes can be resolved. The concept of effective
aperture is most efficient when combined with MIMO beamforming approach where complex
weights can be applied to the individual signal pat hs between each transmit (Tx) and receive
(Rx) antenna array element. As target detection of ships at HF band is mostly clutter limited
but not limited by internal receiver noise the sens itivity penalty of MIMO radar is of no

Universität der Bundeswehr Hamburg Page 6 consequence in this application. On floating platfo rms antenna array patterns will be distorted
due to displacement as well as tilt of antenna elem ents caused by wind and waves. A scheme
for array factor pattern correction of virtual arra y using fast real time phase and amplitude
compensation for the longitudinal and transversal d isplacement as well as tilt (pitch and roll)
of physical antenna elements of sparse array has be en developed.

At HF band spectral behaviour of sea clutter differ s significantly from that faced by
microwave radar. Waves having a wavelength equal t o half the radar wavelength give rise to a
strong Bragg backscatter. By knowing the operating frequency and bearing of a HF SWR a
ship has the opportunity to hide in the characteris tic spectral peaks of the resonant first order
Bragg backscatter signal of the sea. State of art c lutter reduction algorithm in HF SWR uses
Space Time Adaptive Processing (STAP) to filter out the resonant backscatter of the sea in the
space, time and azimuth domain. Our investigations indicate that due to the spectral and
statistical nature of sea clutter STAP has only lim ited success in HF SWR application
especially on floating platforms. Moreover in its a ttempt to suppress the first order Bragg
backscatter this algorithm also ends up attenuating the target signal within the Bragg lines. A
novel Clutter Canceller BeamFormer (CCBF) was inven ted which filters the sea clutter in the
azimuth domain resulting in a higher SCR. This CCBF has the advantage to be even more
robust to array distortion (e.g. displacement in a floating array) or impact of re-radiators which
is a severe problem in the context of shipborne int egration. Assuming a homogeneous
background and a point target an enhancement in SCR of 2.48 dB as compared to
conventional 16-element Chebyshev beamformer has be en achieved. The proposed CCBF was
successfully tested on measurement data obtained wi th a WERA system with 16-element
receiver array. For the first time in open literatu re ship detection within the first order sea
clutter is reported. The proposed scheme is more ef fective and less computational intensive
than STAP.

Besides the extend of the geometrical dimension of the receiver array a further limitation of
current HF SWR systems is the size of the individua l antenna elements. Typically coastal
systems use λ/4 monopoles whose length vary from 2.5 – 25 m. It is highly desirable to design
HF antennas which have a low profile and can thus b e disguised from the public view on a
beach or deployed on floating platforms. We develop ed and tested a capacitive top loaded and
inductively coupled self resonant HF Electrically S mall Antenna (ESA) of 1.45m height and
1.6m width which has good performance in terms of b andwidth, tunability and radiation
efficiency. The antenna can be tuned from 10.7 – 23 MHz with a quality factor very close the
fundamental Chu-McLean Limit. This proves that furt her reduction of antenna size will cause
lower bandwidth and/or radiation efficiency by phys ical limits. The bandwidth-efficiency
product of the proposed antenna is nearly 4 times b etter than those of any other HF ESA
published in open literature. The spiral shape of t he capacitive top loading of the antenna is
well suited to cope with strong winds. All the abov e properties make this antenna ideal for
both coastal and naval HF SWR applications.

Microwave radars have a well defined internal noise level and employ Constant False Alarm
Rate (CFAR) detector to maximize the probability of detection ( PD) while keeping the
probability of false alarm ( PFA ) at a user defined constant level. At HF band rece iver noise is
much lower and, thus, detection is limited by sea c lutter whose statistical characteristics are
not precisely known. State of art detectors use thr ee dimensional CFAR to adaptively vary the
threshold by power regression along azimuth, range and Doppler. In HF SWR this detector
does not guarantee a constant PFA and has limited adaptability especially in the Dop pler
regions dominated by sea clutter. We propose a two step detection scheme which first uses
correlation to detect targets in the range and Dopp ler domain. In a second step it uses

Universität der Bundeswehr Hamburg Page 7 Minimum Mean Square Error (MMSE) estimation scheme to localize the target azimuth. By
means of measurement data obtained by a WERA system we prove that this detector can
detect targets in sea clutter dominated scenarios. An encouraging agreement is also observed
when comparing the detections with Automatic Identi fication System (AIS) records.

In addition an alternative modified CFAR detector h as been investigated. State of art CFAR
detector derives a detection threshold for each ran ge Doppler azimuth cell under test (CUT)
assuming a nearly homogeneous level of clutter and noise power in the background. Thus
detection threshold is calculated from adjacent cel ls. Obviously, in the vicinity of the spectral
peaks of the resonant first order Bragg backscatter of the sea this background is far from
homogenous. Therefore, a pre-segmentation step has been developed to distinguish between
clutter and noise dominated cells. This information is then used to adapt the estimation process
of the local detection background, enabling better estimates, and hence, better detection
performance. A further improvement is achieved by c onsidering multiple successive scans by
applying scan-by-scan averaging. Finally, to mitiga te multi-detection of single targets, an
adjacent target merging algorithm is developed. The se extensions to the basic CFAR method
greatly improve the performance for the considered application scenario.

Universität der Bundeswehr Hamburg Page 8

Universität der Bundeswehr Hamburg Page 9 2. Review of Mobile and Compact HF Surface Wave Rad ar

2.1 Literature Overview
High Frequency (HF) propagation over the horizon us ing skywave mode was discovered in
1920s by Marconi's demonstration of transatlantic c ommunication. During World War II in
1938 the first radar called 'Chain Home' using the surface wave mode propagation was
developed in UK. Operating mostly from 20-30 MHz th e radar had limited success since it
was overwhelmed by a strong interference which was thought to be a German jammer and was
later discovered to be the intense Bragg backscatte r from the ocean waves. In 1960 USA’s
Defence Research Project Agency (DARPA) launched a multi million dollar program code
named 'MayBell'. Sander Associates (sold to Lockhee d Martin in 1986 and later to BAE
Systems in 2005) was the major player in developmen t of Over The Horizon (OTH) HF
Surface Wave Radar (SWR) and developed TOP SEA rada r for U.S. Air Force's Rome
Laboratory and US Naval Research Laboratory (NRL) [ 1]. This radar demonstrated the ability
to detect low flying aircrafts at a range of greate r than 185 km. Set up along the coast of San
Clemente's Island TOP SEA radar receiver array had 25 elements and was 500 m long.

In early 1990, Sanders Associates diversified into developing a compact shipborne HF SWR
named Advanced Technology Demonstrator (ATD) to enh ance ship's self defense [2]. ATD is
by far the most advanced shipborne HF SWR study rep orted in open literature. In 1995 Electro
Magnetic Compatibility (EMC) studies were done on f rigate ships to initiate the design and
placement of the antennas in order to minimize the Electro Magnetic Interference (EMI) due
to onboard ship communication devices. The problems of integration of the antenna array with
the superstructure of the ship and other re-radiato rs were also studied. The study concluded
with optimum locations of a 24 element receiver arr ay on broadside of a ship with
beamforming weights to compensate for non linear el ement radiation pattern. The system
never made beyond this first phase, the exact reaso ns for the project to fail are not published in
open literature and are at best speculative.

In Europe, UK was the first country to start with t he development of HF SWR. In 1980's with
a generous grant from Ministry of Defence, Rutherfo rd Appleton Laboratory (RAL) in
Oxfordshire developed a phased array HF SWR to dete ct and track ships. The receiver
structure developed at the University of Birmingham and at Admiralty Research
Establishment was groundbreaking and laid the found ation of modern day HF SWR array
processing. Marc Lesturgie at ONERA, France has dev eloped excellent space time adaptive
algorithms to suppress the overwhelming sea clutter and is a leading authority on Multiple
Input Multiple Output (MIMO) radars. In 2003, he de veloped Space Time Adaptive
Processing (STAP) algorithms for HF SWR on a floati ng platform [9]. However STAP only
had a limited success in reducing the clutter level since it requires detailed information of
clutter statistics which has not yet been well stud ied. Moreover in its attempt to suppress the
first order Bragg clutter this suppression algorith m also ends up attenuating a target signal
within the Bragg lines. In future – if the statisti cal properties of sea clutter are better
understood – STAP can prove to be a more effective tool in clutter suppression. In 2005,
Guinvarc´h at ONERA presented initial research in f orming large antenna arrays by using a
chain of buoys [10]. He has studied the modulation of the received HF signal caused by tilt
and roll of the buoys.

In addition to military establishments HF SWR was a lso used and developed by
oceanographic community for remote ocean sensing. T he jamming signal in 'Chain Home' was
analyzed by Crombie in 1955, in a ground breaking p aper he explained the basic physics

Universität der Bundeswehr Hamburg Page 10 behind the backscattering of EM waves from a rough moving surface. In 1977 with funding
from NOAA and along with handful of engineers, Barr ick developed the first commercial and
compact HF remote sensing system for measuring the sea surface current using Direction
Finding (DF) which was named Coastal Ocean Dynamics Applications Radar (CODAR). This
marked the technological shift in bearing estimatio n in HF SWR from BeamForming (BF) to
DF reducing the size of antenna system required for HF SWR significantly. In 1986 a phased
array shipborne HF SWR was developed by Teague in S tanford University, USA. The project
was called the JASIN Experiment and was a low cost system for sea current measurement
employing digital BF. Remarkably this was the first time digital BF was used for remote HF
oceanographic sensing. The trials during the JASIN experiment delivered the first measured
data sets which demonstrated the phase modulation c aused in sea clutter due to platform
motion. By this time US navy was looking for an alt ernative to the expensive HF SWR facility
and funded few experiments to compare the performan ce of DF employed by CODAR with
BF by Multifrequency Coastal Radar (MCR) for OTH sh ip detection. These experiments
always concluded with the result that DF yielded en couraging results for ship detection and
also had more widespread applicability in most of t he scenarios. A major advantage of BF
which was overlooked was its ability to filter the sea signal in the spatial domain hence
increasing the effective target signal to sea clutt er signal ratio.

HF SWR for remote oceanographic sensing also starte d in Germany when University of
Hamburg participated in Marine Remote Sensing Exper iment (MARSEN) conducted in the
North Sea in 1979 [3]. In early 1990's University o f Hamburg bought 4 CODAR for
measuring the sea currents onboard a ship using the DF systems. The experiments were
designed and conducted by Gurgel as part of his doc toral thesis [11]. The results from the
experiments were below satisfactory because for a D F system sources of errors are higher on
board a ship. The additional modulation caused by p itch, roll and yaw of a ship confuses the
inverting algorithm in the DF system. Moreover the antenna pattern distortion caused by
reradiators on board a ship adds another significan t sources of error. Following these
experiments Gurgel set about developing a simple sh ore based phased array HF SWR named
WEllen Radar (WERA). The first experiments with WER A were conducted on the Dutch
coast in 1996 and later in the fall of 1996 in Nort h of Germany at the mouth of Rhine river. In
the fall experiment the measurement results from WE RA were compared to those obtained
with the CODAR placed at the same location. Since t hen to this date the oceanographic
community is divided between DF (backed by CODAR) a nd BF (backed by WERA) to
estimate the bearing of backscatter. DF needs a rel atively simple hardware; just one
transmitter antenna and very few receiver antennas but this comes at the price of inferior
information. Only sea current values can be determi ned by inverse algorithm of DF. BF needs
an antenna array on the beach which for a state of the art coastal system (16 element linear
array) is about 120 m long. The advantage of the az imuth filter of BF is the ability to analyze
the azimuth dependency of sea backscatter which yie lds further important information beside
the surface currents like wave height and wind spee d alongside with a much better opportunity
for clutter suppression for ship detection applicat ions. In 2000 WERA was bought by a
company called Helzel Messtechnik GmbH and the deba te between CODAR and WERA took
a commercial turn.

We were encouraged by a paper written by Barrick wh ere he adverts the availability of a
superdirective array in his shore based Compact EEZ -Sonde SWR system [4]. In principle a
superdirective array can achieve the high azimuth r esolution of conventional BF array but with
much smaller aperture. A literature review was done to explore the usefulness of
superdirectivity in ship borne HF SWR. Superdirecti ve arrays, earlier known as ‘supergain
arrays’ are arrays generating higher directivity th an obtained with the same array length and

Universität der Bundeswehr Hamburg Page 11 elements uniformly excited (constant amplitude and linear phase). However, practical
application of superdirectivity has major problems such as low radiation resistance (hence low
radiation efficiency), sensitivity to excitation an d position tolerances, and narrow bandwidth.
Although many papers have been published on the opt imization of array excitation for
maximum directive gain, it has never been considere d for practical design. After extensive
literature survey it is discovered that most of the research in the field of superdirective arrays
has been academic in nature. The motivation being a ble to find out the maximum possible
directivity achievable by an array rather than any application oriented approach. Since for an
array maximum directivity can be achieved in endfir e direction, only endfire directivity and
gain have been measured. No experimental data on th e radiation pattern and broadside gain of
a superdirective array has been published. Moreover due to strong mutual coupling between
the superdirective antenna elements it is difficult to steer the beam. There is no published
experimental verification of beamforming in any sup erdirective receiver array. After the
literature review, we conclude that it will be diff icult to realize a highly directive receiving
antenna system consisting of a superdirective array . To the best of the author’s knowledge it
can be concluded that the system adverted in [4] is nothing more than an awareness campaign.

Typical clutter affects only a limited part of the radar data as it exhibits a very specific
velocity: Ground clutter is usually stationary whil e sea clutter has certain dominant velocities
(see section 3.3). However, in case of a moving ra dar platform, the (relative) radial velocity
depends on the look angle. Additionally, due to the finite angular resolution, clutter
components from slightly different look angles and hence with slightly different velocity will
bleed into each other, leading to clutter spread. M ethods to alleviate this problem are DPCA,
which tries to compensate the platform motion and r econstruct the radar signals a stationary
platform would have seen, and STAP, which combines beam-forming and Doppler processing
into one operation. STAP in particular has the adva ntage of being optimal in the sense of
optimizing the signal-to-noise+clutter+interference -ratio. However, it requires an accurate
statistical model of the noise, clutter and interfe rence present in the radar data. In practice, this
model is estimated from the measured data, with any estimation errors directly deteriorating
the performance of STAP. Therefore, STAP, as well a s DPCA, have mainly been used for
airborne radar, where the effect of platform motion is obviously much more severe. While
ship-based radar might in theory also benefit from STAP or DPCA, the possible gain in
practice seems to be rather small; hence they were excluded from this study.

2.2 Key Challenges in Design of distributed shipbor ne HF SWR
Characteristics of signals in a HF SWR are very dif ferent from microwave radar; hence there
are different challenges:

1. Resonant sea clutter – At HF frequencies the sea waves having wavelengt hs equal to
half the radar wavelength correspond to a strong re sonant Bragg backscatter. Since the
velocity of a gravity wave depends on its wavelengt h hence the Bragg backscattered
signal corresponds to a specific Doppler frequency which shifts on changing the radar
wavelength. This severely hinders with the ability to detect targets. So when the target
velocity is close to the velocity of the sea wave c ausing the Bragg backscatter the sea
clutter echo can be many times higher than the targ et signal. This makes the system
external noise limited. STAP can prove to be an eff ective tool to filter noise once the
statistical properties of sea clutter are better un derstood [4]. Currently, sea clutter
models at HF band are not very well developed in te rms of predicting accurately the
statistical and spectral nature of the received sea signal. There exists no comprehensive
model which can estimate the sea clutter RCS for HF SWR onboard a floating

Universität der Bundeswehr Hamburg Page 12 Power
amplifier ≈Waveform
Generator RX frontend ≈ Mixer A/D Converter
RX frontend ≈ Mixer A/D Converter
D/A
Converter Beam Forming
Range Doppler Transform
Clutter Mitigation
Detection
Tracking
Remote Central Unit Optical or µµ µµ-Wave
Interfaces
Optical or Wireless
Interfaces
Power
amplifier ≈Waveform
Generator RX frontend ≈ Mixer A/D Converter RX frontend ≈ Mixer A/D Converter
RX frontend ≈ Mixer A/D Converter
D/A
Converter Beam Forming
Range Doppler Transform
Clutter Mitigation
Detection
Tracking
Remote Central Unit Optical or µµ µµ-Wave
Interfaces
Optical or Wireless
Interfaces

Fig. 2-1 : Distributed shipborne FMCW radar with digital be amforming with optical or microwave ( µ)
interfaces for HF signals. Central unit (collocated with the transmitter) receives and processes signa ls
from remote sensors on ship or buoys.
platform or can predict the received time domain si gnal for a given sea state. The effect
of currents, wind and varying sea floor depth also result in modulation of the gravity
waves which is very difficult to model. Few models which exist have not been proved
experimentally.

2. Limited antenna aperture – At HF band wavelengths are between 10 m to 100 m .
Hence for a reasonable angular resolution the radar receiver antenna array should be
large. However, large antenna arrays are very costl y and, in particular in the case of
shipborne platforms, the array dimensions are limit ed. ONERA, France has done some
research on forming large antenna array by using a chain of buoys [5]. Another
approach can be a sparse array using Multiple Input and Multiple Output (MIMO)
technique and several ships/buoys in a naval format ion [6].

3. Inhomogeneous background – The clutter power received for a conventional HF
SWR after range and Doppler processing is very inho mogeneous with a dynamic range
of noise signal up to 80 dB. Design of a detector w hich can detect targets in such an
inhomogeneous background with equal probability of false alarm in all regions of the
received Range Doppler power map is very challengin g and state of art detectors like
CFAR have limited success.

4. Ship RCS in resonant region – Radar Cross Section (RCS) of targets like ships/ boats
lie in the resonant region at HF frequency band wit h its main contribution caused by
the mast of the vessel. This implies that the value of RCS can fluctuate a lot with a
small change in the radar frequency. Hence the rada r frequency must be tunable across
whole HF band [2].

5. Congested HF band – HF band is frequently used by amateur radio user s and others.
Moreover free bands with bandwidth in excess of 150 kHz are very rare. The system
needs to be agile in using this crowded frequency s pectrum efficiently to maximize the
performance.

Universität der Bundeswehr Hamburg Page 13 6. Pitch and Roll compensation — Transversal and lateral motion of a platform carr ying
the transmitter / receiver over sea causes frequenc y modulation of the HF signal.
French researches proposed an adaptive technique to compensate this distortion in the
signal [7]. However this is time consuming and an e ffective technique was invented by
the group and presented in the status report [8].

7. Antenna Integration with Ship’s Superstructure – A naval frigate is already
equipped with a HF communication system along with many radar and communication
systems. Integration of an HF antenna array in term s of Electro Magnetic (EM)
compliance with other RF systems on board is essent ial. Another challenge is the
placement of an antenna array on the ship in to ord er to minimize the distortion in the
radiation pattern of the antennas caused by the sup erstructure of the ship and
surrounding reradiators. After extensive discussion s with all the project stakeholders it
was decided that both these problems are not in the immediate scope of the project due
to lack of resources. If required a dedicated platf orm like a pontoon can be deployed to
simplify the EM compliance problem. However this st udy focuses on the signal
processing issues of HF SWR hence the solution to a ntenna integration was deemed
out of the scope of this study.

8. Synchronization and communication in a distributed system – HF radar is
inherently quasi-monostatic (or bistatic) since the receiver has to be isolated from the
transmitter. An offshore system with a receiver arr ay build by a formation of ships or
buoys is even more distributed and hard to synchron ize. Moreover the information
from each antenna element must be combined together to realize a digital beamforming
system. This requires data transmission from the of fshore radar array to a central
processing unit located on board a ship or at the c oast (Fig. 2-1).

9. Interface realization in a distributed system – A distributed offshore HF SWR
system with possible interfaces is shown in Fig. 2- 1. It is clear that realization of the µ-
wave and optical links is very challenging in a rem ote offshore system.

During the discussion with all the stakeholders and keeping the limited resources in
perspective, few key decisions were made. It was ag reed that investigation of skywave
systems, passive HF radars and a hybrid over the ho rizon radars (combined surface wave and
skywave) fall outside the scope of this project [2] . It was also decided that the group will
continue with Frequency Modulated Continuous Wavefo rm (FMCW) for the radar since it
keeps the signal processing and hardware simple and cost effective [2].

Literature
[1] D. Barrick, “30 years of cmtc and codar,” IEEE/OES 9th Working Conference on Current
Measurement Technology, 2008. CMTC 2008., pp. 131–1 36, 2008.
[2] Systemstudie zu verteiltem selbstkonfigurierenden H F-Überhorizontradar auf Schiffen, 1st status
report, 17.09.2009.
[3] H. Essen, E. Mittelstaedt, and F. Schirmer, “On nea r-shore surface current measurements by means
of radar,” Deutsche Hydrographische Zeitschrift 36, pp. 1–14, 1981.
[4] Systemstudie zu verteiltem selbstkonfigurierenden H F-Überhorizontradar auf Schiffen, 2nd status
report, 22.01.2010.
[5] A. Bourges, R. Guinvarch, “Perspectives of the use of high frequency radar on buoys,” Oceans
2005- Europe, pp. 1256-1259, Brest, June 200 5.

Universität der Bundeswehr Hamburg Page 14 [6] T. Fickenscher, A. Gupta, “Element-Space Spatially Waveform Diverse FMCW Radar Distribution
on Naval Platform,” IEEE German Microwave Week 2011 , pp. 1-4, 2011.
[7] A. Bourges, R. Guinvarc´h, B. Uguen, R. Gillard, ‘A simple pattern correction approach for high
frequency surface wave radar on buoys,’ IEEE Europe an Conference on Antennas and
Propagation, pp. 1-4, 2006.
[8] Systemstudie zu verteiltem selbstkonfigurierenden H F-Überhorizontradar auf Schiffen, 5th status
report, 24.11.2010.
[9] Lesturgie, M.: ‘Use of STAP techniques to enhance t he detection of slow targets in shipborne HF
SWR’. Proc. IEEE Radar Conf. 2003, Sept. 2003, pp. 504-509.
[10] A. Bourges, R. Guinvarch, “Perspectives of the use of high frequency radar on buoys,” Oceans
2005-Europe, pp. 1256-1259, Brest, June 2005.
[11] K.W. Gurgel, Flächenhafte Messung der Oberflächens trömung vom fahrenden Schiff aus – Eine
neue Anwendung des Hochfrequenz-Radarverfahrens am Beispiel der Arktisfront, Ph.D. Thesis,
University of Hamburg.

Universität der Bundeswehr Hamburg Page 15 50 100 150 200 250 300 025 50 75 100 125 150
Distance (km) Attenuation (dB)

30 MHz (Avg. Sea)
30 MHz (Baltic Sea)
15 MHz (Avg. Sea)
15 MHz (Baltic Sea)
5 MHz (Avg. Sea)
5 MHz (Baltic Sea)

Fig. 3-1: Two way surface wave propagation losses over averag e sea water (σ =4.10 S/m, εr =80)
and Baltic sea ( σ =3.73 S/m, εr =80) at various HF frequencies calculated using GR WAVE. 3 Modelling Quasi-Monostatic FMCW HF Surface Wave
Radar

The chapter explains the models and signal processi ng chain used to estimate the received
signal of a quasi-monostatic HF SWR.

3.1 Propagation

We use a standard model based on the Ground WA VE (G RWA VE) program to calculate the
propagation losses in HF SWR. GRWA VE incorporates b oth the free space propagation losses
and the additional losses due to the propagation of the surface wave over the curved sea
surface for a given conductivity and permittivity. This program uses Sommerfeld integrals to
estimate the electric field strength and is made av ailable by the International
Telecommunications Union (ITU) [1], [2]. The GRWA VE model considers a troposphere
whose refractive index varies with height. It is ba sed on three different methods to calculate
the electric field strength at distances within and over the radio horizon, respectively. Beyond
the radio horizon, the residue series is used; at d istances within the radio horizon, the model
employs the extended form of the Sommerfeld flat-Ea rth theory [1]. On the other hand,
geometric optics is used to calculate the field str engths at distances in between the residue
series and the Sommerfeld theory [3], [4]. For us, the main parameters are the frequency and
the electrical properties of the sea: the conductiv ity and the permittivity. These electrical
parameters are dependent on the salinity and surfac e temperature of the sea surface under
consideration. Fig. 3-1 shows the amplitude of the two way attenuation (obtained from the
GRWA VE model) versus range for various frequencies for average sea water and in Baltic sea.
The electrical parameters of the average sea surfac e (salinity = 35 g/kg) are fixed to typical
values of electrical conductivity σ = 4.10 S/m and relative electrical permittivity εr = 80 while
in case of Baltic sea (salinity = 30 g/kg) they are σ = 3.73 S/m and εr = 80 (15°C, atmospheric
pressure). It can be seen from Fig. 3-1 that the at tenuation of the surface wave reduces
considerably with decreasing frequency and due to l ower salinity of Baltic sea the attenuation
is higher then average sea water. At 5 MHz the atte nuation over sea is comparatively low

Universität der Bundeswehr Hamburg Page 16

Fig. 3-2: Bragg resonance when the path difference between th e echoes from two waves crests
is a multiple of half the radar wavelength ( λ/2) [5].
√2 Bragg
f / fBNormalized Power (dB) 23/4 Bragg
√2 Bragg Receding Bragg line Approaching Bragg line
√2 Bragg
f / fBNormalized Power (dB) 23/4 Bragg
√2 Bragg Receding Bragg line Approaching Bragg line

Fig. 3-3: Typical Normalized Doppler power spectrum recorded at 8 MHz [5]. however the size of a conventional monopole is 15m so the physical dimension of the antenna
poses major challenges in terms of placement and in tegration at such low frequencies.
Propagation at higher frequencies deteriorates rapi dly. A good compromise with propagation
loss and size of the antenna structure is in the mi d of HF band. This is one of the main reasons
why most of the commercial HF SWR operates around 1 2 MHz. It is worth mentioning that
surface wave attenuation function for hard targets is a biquadratic function of distance while in
the case of sea clutter it is a cubic function of d istance.

3.2 Sea Clutter

Gravity sea waves are generated on the surface of t he ocean by the wind over a long time and
fetch either locally or by distant storms. The sea clutter signal received by an HF SWR is due
to the reflection of the HF signal by these gravity waves. In an open sea all waveheights and
wavelength are present up to some maximum value def ined by the sea state. The sea waves
travelling perpendicular to the radar beam will giv e rise to backscatter, but those having a
length equal to half the radar wavelength ( λ/2) will give a much stronger echo because the
partially reflected HF signals add together constru ctively as illustrated in Fig. 3-2. This echo
signal is known as first order Bragg scattering, af ter W.L. Bragg, who first proposed this
mechanism to explain the scattering of X rays from crystals.
A typical Doppler power spectrum for HF SWR recorde d at 8 MHz is shown in Fig. 3-3. The
first order Bragg lines are the strongest signal in the spectrum. The dispersion relation for sea
waves relates the phase velocity, Vp with the sea wavelength, λs. It states that Vp is directly
proportional to acceleration due to gravity, g and λs. As a result the Bragg frequency, fB has a

Universität der Bundeswehr Hamburg Page 17

Fig. 3-5: Example of normalized RD map (difference between me asurement and model). square root dependence on the radar frequency λ. However the Doppler frequency ftar is
directly proportional to λ. The peak at positive Doppler frequency correspond s to the Bragg
scattering from approaching sea waves and the peak at the negative Bragg scattering from the
receding sea waves. The ratio of the power of the p ositive and negative first order Bragg peaks
depends on the look direction of the radar relative to the wind direction. In the example of Fig.
3-3 the wind is blowing somehow more towards the ra dar causing the approaching Bragg line
to be stronger. The rest of the Doppler spectrum is known as second order Bragg scattering
with significant peaks at √2 and 2 3/4 times the Bragg frequency. Higher harmonic sea wave s
which fulfil the Bragg backscattering condition hav e longer wavelengths like 2 λ/2, 3 λ/2… and
correspond to peaks in Fig. 3-3 at √2, √3 … times the Bragg frequency respectively. The pea k
corresponding to √2 times the Bragg frequency is visible since the ot her peaks are masked by
the second order Bragg continuum caused by sea wave s of all wavelength. The HF signal is
also scattered back to the radar from two sea waves travelling at right angles to each other.
This corner reflector corresponds to peak at 2 3/4 times the Bragg frequency.

The sea clutter Doppler spectrum can be modelled fo r different sea states and wind speeds for
an monostatic HF SWR as shown by Barrick [6][7]. Th e model assumes a fully developed
patch of sea and is presented in details in status report [8]. This is a simple and well accepted
model used extensively by oceanographers for wave h eight and wind speed prediction. It has
been tested thoroughly by comparisons with measurem ent results.
Fig. 3 -4: Example of measured RD map (left) and model fit (ri ght)

Universität der Bundeswehr Hamburg Page 18 The validity of the propagation and clutter models was verified using measurement data from
Wangerooge. According to the propagation and the cl utter model, the clutter-and-noise
background should consist of a noise component that is independent of range and Doppler and
a clutter component with quite specific behavior al ong the range and Doppler axis. In
particular, for the clutter component, the shape of the Doppler power spectrum is expected to
be independent of the range; only its overall level is assumed to decrease with range. Analysis
of the measurement data from Wangerooge shows that this model has to be extended to cope
with different sea currents and different ranges. T hese sea currents lead to a range-dependent
shift of the clutter Doppler power spectrum, while the independence of the shape still remains
valid. To verify this model, a fit of the shift due to the sea currents, the clutter Doppler power
spectrum and the background noise to measurement da ta using numerical optimization
methods was performed. The range-dependent attenuat ion was taken as predicted with the
help of GRWA VE. A representative example is shown i n Fig. 3-4, where it is apparent that the
model provides a reasonable fit to the measurement data. Normalizing the measurement data
with the estimated model (i.e. taking the differenc e on the dB-scale) allows a better judgment.
The main differences between measurement and model are

1) the existence of targets in the measurement data
2) some deviation at very close ranges were the propag ation and sea clutter models do not
fit due to the shallowness of the water and
3) artifacts of unknown origin along a line from zero Doppler at close ranges to slightly
negative Doppler further away.
Overall, the model matches the measurement data qui te well.
So far, only the (average) power of the clutter-and -noise background has been modeled. For
simulation purposes, it is desirable to model the c omplex-valued signal. The common
assumption is then that in-phase and quadrature com ponent are both independently Gaussian
distributed with the power given by the model. This has been examined by considering
histograms of the phase, which has always exhibited a very uniform distribution, and of the in-
phase and quadrature amplitude values, shown in Fig . 3-6. The analysis has been carried out
on the normalized RD map, so if the assumptions hol d, Gaussian distributions of unit power
are to be expected. Indeed, when considering the wh ole RD map, the histogram very close
matches the Gaussian. But note that this result is dominated by the noise area. If the analysis is
restricted to the area of the first order Bragg lin e, the result is rather inconclusive due to the
limited number of data points considered.
In conclusion, the proposed clutter-and-noise model shows a fair match to the measured data
and the assumption of independently Gaussian distri buted in-phase and quadrature
Fig. 3-6: Histogram of in-phase and quadrature amplitude va lues for overall normalized RD
map (left) and a small excerpt at the first order B ragg lines (right) with the Gaussian
distribution for unit power in comparison

Universität der Bundeswehr Hamburg Page 19 components has been verified for the noise-dominate d background, while the probability
distribution of the clutter component is still an o pen research question.

3.3 Signal Processing

The signal processing in a Frequency Modulated Cont inuous Wave (FMCW) radar can be
subdivided into an analog and a digital part. The c hoice of a FMCW radar can be justified by
the following four advantages: 1.) low average powe r, 2.) flexible adjustment of range and
range resolution, 3.) simplified receiver processin g (also known as stretch processing) and 4.)
the simple creation of signals. The analog part of the receiver performs a demodulation of the
received FMCW signal with the transmitted (referenc e) FMCW signal, commonly denoted as
stretch processing [9]. This leads to a reduced sam pling requirements as compared to Nyquist
sampling and for a static target is illustrated in Fig. 3-7.

After the Analog to Digital converters (ADC) each t arget can be represented as a sampled
sinusoidal signal with frequency f b, which is composed of a range frequency f r and a Doppler
frequency f d [10]. Certain restrictions regarding the speed of the respective targets and the
maximum range do apply.

Fig. 3-7 : Analog FMCW Front End – Stretch Processing

In the digital domain Azimuth Range Doppler (ADR) p rocessing is applied: This can be seen
as a three dimensional FFT processing of Beamformin g, Range Transform and Doppler
Transform, for which the Range-Doppler transform is illustrated in Fig. 3-8.

Fig. 3-8 : Range-Doppler transform

The input of Fig. 3-8 is formed by the samples from the ADC of each chirp/sweep written
row-wise into the input structure. The range transf orm is applied row wise for each sweep (fast
time), while the Doppler transform is applied colum n wise (slow time) to each of the formed
range bins. After that each target is situated at s ome range bin with frequency f R , some
Doppler bin with frequency f D and in some beam n B. The Beamforming operating is applied
across the receiver antenna array, typically formin g beams overlapping at their Half Power
Beam Width (HPBW) points. In addition, windowing (a lso known as amplitude tapering) to
adjust the resulting main to sidelobe ratio might b e applied independently to each of the three

Universität der Bundeswehr Hamburg Page 20 dimensions. Based on this 3D data cube the target d etection, as described in Section nine, is
carried out.
Literature
[1] A. N. Sommerfeld, “Propagation of waves in wireless telegraphy,” Ann. Phys. , vol. 81, pp. 1135–
1153, 1926.
[2] Ground Wave Propagation (GRWAVE) , Int. Telecommun. Union, Geneva, Switzerland. [Onl ine].
Available: http://www.itu.int/en/pages/default.aspx
[3] S. Rotheram, “Ground-wave propagation. Part 1: Theo ry for short distances,” Proc. Inst. Elect.
Eng.—Commun., Radar Signal Process. , vol. 128, no. 5, pp. 275–284, Oct. 1981.
[4] S. Rotheram, “Ground-wave propagation. Part 2: Theo ry for short distances,” Proc. Inst. Elect.
Eng.—Commun., Radar Signal Process. , vol. 128, no. 5, pp. 285–295, Oct. 1981.
[5] S. Kinsley, S. Quegan, Understanding Radar Systems, McGraw-Hill Book Company, London,
1992.
[6] Barrick, D.E. ‘Remote sensing of sea state by radar .’ Remote sensing of troposphere, Chapter 12,
Derr,V.E. : Washington D.C.: US Government Printing Office, 1972.
[7] Barrick, D.E. ‘Remote sensing of the sea state by r adar,’ in Proc. OCEAN , 1972, vol. 4, pp. 186-
192 .
[8] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 4th status
report, 26.07.2010.
[9] A. E. Carr, L. G. Cuthbert and A. D. Olver: Digital signal processing for target detection FMCW
radar, IEE Proceedings Communications, Radar and Si gnal Processing, Vol. 128, Issue 5, pages
331-336, October 1981.
[10] D. E. Barrick FM/CW Radar Signals and Digital Proce ssing: Technical Report ERL 283-WPL
26, National Oceanic and Atmospheric Administration (NOAA), July 1973

Universität der Bundeswehr Hamburg Page 21 4. Compact Coastal Antenna Array Using MIMO Concept

4.1 Summary
Multiple-Input Multiple-Output (MIMO) beamformers e nable to synthesize virtual antenna
positions which result in a larger number of effect ive virtual array elements. By adding a
second transmitter MIMO concept can be used to doub le the effective length of a conventional
receiver array. This approach has been implemented and tested on a commercial HF SWR
system (Remote Ocean Sensing System WERA, Helzel Me sstechnik GmbH, Kaltenkirchen,
Germany) [2]. 6 elements of a commercial 12 element receive array have been used to
demonstrate the MIMO concept enabeling a direct com parison with the results obtained with
the physical 12 element array. Results presented am ong oceanographic community have
driven immediate demand for product development [3] and [4].

4.2 Motivation
The investment for a coastal HF SWR is driven by th e beach property. HF SWR systems
based on direction finding instead on beamforming a re highly compact and are suitable for
oceanographic measurements such as ocean current mo nitoring. For this particular application
where the strong clutter is the desired signal it i s not necessary to gain azimuth information by
beamforming and to process the Doppler spectrum sep arately for each range azimuth cell.
Instead the frequency components with the strongest signal in each range azimuth cell are
assumed to represent the reflection of the Bragg re sonant sea waves (first order Bragg
backscatter) often referred as Bragg lines. Due to the current in the sea an azimuth dependent
Doppler shift is observed for these Bragg lines. Pr oper signal processing can estimate the
direction of arrival of the signal of the Doppler s hifted Bragg lines by exploiting the whole
Doppler spectrum of the respective range cell (cont aining the entire azimuth cells) and, thus,
generate the missing azimuth information.

However, in the case long range maritime surveillan ce the strong clutter return is no longer the
signal of interest but generates a severe clutter b ackground that can mask the weak signals of
slow moving targets like ships. Long range maritime surveillance cannot be performed
successfully without some kind of beamforming which involves an antenna array (at least a
combination of both beamforming and direction findi ng) due to the strong sea clutter
environment. The American company CODAR is publishi ng different statements but it is
known for overpromising e.g. ref. to Superdirective Arrays, chapter 2.
Fig. 4.1 : Top view of uniform linear
10-element effective aperture
(EA 1…EA 10 ) generated by 5-
element physical Rx array
(Rx 1…Rx 2 spaced λ/2) and 2 Tx
arrays (Tx 1 and Tx 2).

Universität der Bundeswehr Hamburg Page 22 Shortening the array length is of higher importance than reducing hardware complexity. By
adding a second transmitter Multiple-Input Multiple -Output (MIMO) concept can be used to
double the effective length of the receiver array. MIMO concept requires orthogonal
waveforms, i.e. the signals of both transmitters mu st be separable at the receiver [1]. This
approach has been implemented and tested on a comme rcial HF SWR system (Remote Ocean
Sensing System WERA, Helzel Messtechnik GmbH, Kalte nkirchen, Germany) [2]. Results
presented among oceanographic community have driven immediate demand for product
development [3] and [4].

4.3 Concept of MIMO Beamforming
MIMO beamformers enable to synthesize virtual anten na positions which result in a larger
number of effective virtual array elements (effecti ve aperture). The effective aperture of the
combination of a transmit and a receive array is th e receive aperture that would produce the
same two-way radiation pattern if the transmit ante nna would be a point source. Fig. 4-1
displays the example of two transmit arrays (Tx 1 and Tx 2) and one physical 5-element receive
array (Rx 1…Rx 2 blue). Both elements of each Tx array are separate d by λ/2 and are driven
180° out of phase resulting in a null in their resp ective radiation patterns along the receive
array (quenching direct path!). Spacing between Tx phase centers and respective neighboring
Rx element is half (i.e. λ/4) the inter-element spacing of Rx aperture. In co njunction with
transmit array Tx 2 the elements Rx 1…Rx 5 provide the signals of elements EA 1 … EA 5 of
virtual effective aperture. In conjunction with tra nsmit array Tx 1 the same elements provide
the signals of elements EA 6 … EA 10 . An equivalent conventional antenna array would re quire
10 Rx and 1 Tx channels. The MIMO beamformer can be considered as a conventional
beamformer in frequency domain which assigns comple x weights to the channels of the virtual
aperture.

4.4 Implementation in WERA System
Orthogonal waveforms for the individual transmitter s of FMCW radar can be realized for
instance by Time Division Multiplex (TDMA), Frequen cy Division Multiplex (FDMA), Beat
Frequency Division or Chirp Rate Division. In the c ase of Beat Frequency Division time or A
DA
D
vTx 1 (t)vb1(t)Rx 1
MIMO
Beam-
forming A
DA
D
vTx 1 (t)vb2(t)Rx 2
A
DA
D
vTx 1 (t)vb6(t)Rx 6Range-
Doppler
Transf. Separation in
range Doppler
domain
Tx 1Tx 2Beat freq.
offset
vTx (t)vb6(t)Rx 6Range
Doppler Tx 2Tx 2 Tx 1
Tx 1Tx 1 Tx 2 Tx 2Processed as originated from

Fig. 4-2: Modified WERA receiver (left) with 2 Tx and 6 physi cal Rx channels resulting in 12
element effective aperture. Transmitters operating with frequency offset equal to half a range cell.
Sketch of range Doppler map (right) with both Bragg lines (of different strength); orthogonality is
lost in Doppler domain.

Universität der Bundeswehr Hamburg Page 23 frequency staggered chirps can be used [5] and [6]. From practical implementation point of
view modification of a WERA using frequency stagger ed chirps was the most simple approach
(Fig. 4-2). All Rx channels are downmixed with the signal of Tx 1 resulting in a beat frequency
offset for the signals originating from Tx 2. Both transmitters were operated with an offset
frequency equal to half a range cell (1.92 Hz). Thu s, orthogonality of the received signals is
lost in the Doppler domain. Signals in the RD map w ith Doppler shift higher than half a range
cell +/- 0.96 Hz were processed as originating sole ly from transmitter 2 and vice versa.

Measurements were carried out at WERA station Bunke r Hill (Sylt, operator Helmholtz-
Zentrum Gesthacht). The modified radar was operated at a frequency of f = 10.85 MHz,
bandwidth B = 100 kHz and transmit powers of less then 100 mW. Due to the loss of
orthogonality some second order clutter and noise i s spread in the data of the respective part
of the RD map allocated for the other transmitter. Apart from that, the experiment has shown
successful operation of the MIMO approach (Fig. 4-3 ).
The above investigations have lead to product devel opment at Helzel Messtechnik GmbH
where further improvements are currently under inve stigation: In order to reduce the impact
of pattern distortion a 3 λ/2 instead the λ/2 spacing between Tx and Rx arrays or positioning
the Tx arrays further aside the Rx array is conside red. To avoid loss of orthogonality in the
Doppler domain a TDMA approach is currently under d evelopment.

Literature
[1] G. Frazer, Y . Ambramovich, and B. Johnson, “Spa tially Waveform Diverse Radar: Perspectives for
High Frequency OTHR ,” IEEE Radar Conference 2007, Boston, USA, pp. 385-3 90, June 2007
[2] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen , 8th status
report, 29.05.2012
[3] Fickenscher, Th. ; Gupta, A.: Novel Concept for Compact ULA Using MIMO Beamforming,
PIERS 2012, Progress in Electromagnetic Research Sy mposium, Kuala
[4] Fickenscher, Th.: Advances and Prospects in HF Surface Wave Radar Research at Helmut Schmidt
University, 3rd Remote Ocean Sensing (ROS 2011), Na to Underwater Research Centre (NURC),
La Spezia, Italy, 11. – 13. Oktober 2011
[5] J. O. Hinz, T. Fickenscher, A. Gupta, M. Holter s, U. Zölzer: Evaluation of Time-Staggered MIMO
FMCW in HF SWR, International Radar Symposium (IRS) 2011, Leipzig, September 2011 2×6 element effective aperture 12 element physical apert ure
MIMO Non-MIMO

Fig. 4-3: Measured RD map for MIMO array (left) with 2 Tx and 6 physical Rx elements
(transmitters operating with frequency offset equal to half a range cell) in comparison with results
for conventional array with 1 Tx and 12 Rx elements .

Universität der Bundeswehr Hamburg Page 24 [6] T. Fickenscher, A. Gupta, J. O. Hinz, M. Holter s, U. Zölzer: MIMO Surface Wave Radar Using
Time Staggered FMCW Chirp Signals, IEEE European Ra dar Conference (EuRAD) 2011,
Manchester, Oktober 2011

Universität der Bundeswehr Hamburg Page 25 5. Large Sparse Sea Floating Antenna Array Using MI MO
Concept

5.1 Summary
Reduction of sea clutter is one of the main challen ges in detection of slow moving hard targets
like ships. Sea clutter signal of first and second order reduces with increasing spatial
resolution of the radar. However, large antenna arr ays are very costly and, in particular in the
case of shipborne platforms, the array dimensions a re limited [1]. In order to resolve this
restriction, some research has been published on fo rming large antenna arrays by using a chain
of buoys [2]. We propose another approach to overco me the limitations of the array
dimensions for naval applications by making use of several ships or mobile platforms sailing
within a naval formation [3]. In this case the arra y dimensions are virtually unlimited.
However, the array size (element number) remains li mited and there are constraints regarding
the spatial arrangement of the antenna elements. An array distributed among a naval formation
will be a sparse array in nature, i.e. the inter-el ement spacing between antennas located on
different ships will be more than one-half of a wav elength ( λ/2). Beamforming in sparse
arrays typically results in gratinglobes because of spatial undersampling. These gratinglobes
aliased in the angle space have the same magnitude response as the mainlobe and are
indesirable for a HF SWR system. Appling the concep t of effective aperture [4] gratinglobes
can be resolved.

We consider the naval formation as a linear sparse array with uniform sparse distance D > λ/2
among N mobile receiver platforms each equipped with a sin gle antenna element. By judicious
choice of D according to transmitter array inter-element spaci ng d and transmit element
number M, gratinglobes can be avoided in the two-way antenn a pattern. The concept of
effective aperture is most efficient when combined with MIMO beamforming approach where
complex weights can be applied to the individual si gnal paths between each transmit (Tx) and
receive (Rx) antenna array element. As target detec tion of ships is mostly clutter limited but
not limited by internal receiver noise the sensitiv ity penalty of MIMO radar is of no
consequence in this application. MIMO concept requi res orthogonal waveforms, i.e. the
signals of the transmitters must be separable at th e receiver. Orthogonality can be achieved
Effective aperture dD = M d
Deff = ( M N-1) d(M-1) d
darbitrary quasi
monostatic Physical array
Tx array Rx array

Fig. 5-1: Sparse array distributed among Naval Formation of f our ships for the example of M=8
Tx and N=3 Rx antenna elements : Geometry of physical (dense) Tx and sparse Rx arr ay (top)
and equivalent effective aperture with dense unifor m linear array (bottom).

Universität der Bundeswehr Hamburg Page 26 quite easily when using FMCW waveform. An example f or necessary hardware and software
implementation is provided [5] and [6].

On floating platforms array patterns will be distor ted due to displacement as well as tilt of
antenna elements caused by wind and waves. A simple scheme for array factor pattern
correction of virtual array using easy to implement fast real time phase and amplitude
compensation for the longitudinal and transversal d isplacement as well as tilt (pitch and roll)
of physical antennas elements of sparse array has b een analysed [7]. It’s the simplicity of the
computational effort of this algorithm that makes i t superior to more sophisticated approaches
of pattern synthesis of sparse antenna arrays (for e. g. see. [8]-[10]).

5.2 Sparse Array Distributed Among Naval Formation
Fig. 5-1 displays an example of the array design fo r a HF SWR system distributed among a
naval formation of 4 ships. The transmit array with M=8 elements and inter-element spacing
of d=λ/2 is located on a single ship. N=3 ships sailing in a line – each equipped with a s ingle
antenna – compose the receive array. The signal pro cessing unit can be collocated on either of
these platforms or – if desired – located at a remo te site. It is not necessary that the ship
hosting the transmit array is in line or close by t he ships with the receive array. A quasi-
monostatic or even a bistatic architecture can be c onsidered. Communication between the
individual platforms of this distributed radar syst em (collocated MIMO radar) can be provided
via satellite communication or via a line-of-sight communication between all platforms which
can be established via microwave or optical links.

We consider the two-way antenna pattern, that is th e product of the transmit pattern and the
receive pattern. Grating lobes of this sparse array are resolved using the effective aperture
concept. The effective aperture of the combination of a transmit and a receive array is the
receive aperture that would produce the same two-wa y radiation pattern if the transmit
aperture would be an isotropic radiator. In order t o avoid grating lobes d = λ/2 element
spacing is preferred for the effective aperture, an d, hence, for the transmit array. For the
example of M=8 transmit antenna elements, N=3 receive antenna elements at a frequency of
10 MHz (wavelength λ=30 m) we obtain d = 15 m, a transmit array length of ( M-1) d = 105 m,
a sparse inter-element spacing of the receive array of D=120 m and an effective aperture of
MxN = 24 elements with length Deff = ( M N -1) d = 345 m.

In the following, we will prove that though the ori ginal receive array is sparse, but convolving
it with a filled array with appropriate element num ber and spacing an aperture without grating
lobes can be obtained [11]. The steering vector for the transmit array (time dependence of
transmitted signal e jωt) with the angle θ from boresight is

Td M d] e… e 1 [)() sin( ) 1 (2j ) sin( 2j θλπθλπ
θ−= P

where [.] T denotes transpose. Similarily, the steering vector of sparse receive array can be
written as

TD N D] e… e 1 [)() sin( ) 1 (2j ) sin( 2j θλπθλπ
θ−= S .

Thus, the total steering vector of the two way ante nna pattern is

Universität der Bundeswehr Hamburg Page 27 T TD NTDT)] ( e… )( e )([ )( )( )() sin( ) 1 (2j ) sin( 2jθ θ θ θ θ θθ
λπθ
λπ
P P P S P G−= ⊗ =

where ⊗ denotes Kronecker product. With D=Md the phase of the total steering vector will
become contiguous and the virtual effective apertur e will be equivalent to a uniform linear
dense array with inter-element spacing d=λ/2. Grating lobes of the distributed sparse array a re
resolved and – making use of the concept of MIMO be amsteering – individual weights in
amplitude and phase can be applied to each of the v irtual M x N elements of the effective
aperture.

As an illustrative example simulation results are g iven for a look angle of 0° from boresight.
For the sake of simplicity, uniform horizontal ante nna patterns are assumed for the individual
antenna elements. Fig. 5-2(a) depicts the azimuth c ut of the antenna pattern of the dense
transmit array with uniform amplitude weights. Due to the limited array size that can be hosted
on a ship a comparatively wide beamwidth of approxi mately 13° is achieved. In the same
graph the azimuth cut of the antenna pattern of the large but sparse receive array with uniform
amplitude weights is given. The beamwidth (6°) is s maller than that for the transmit array but
– apart from sidelobes which also occur in the case of the transmit array – grating lobes are
obvious for this sparse array. However, these grati ng lobes are rigorously located in the nulls
of the transmit beam pattern. Thus grating lobes ar e effectively canceled in the two-way
antenna pattern shown in Fig. 6-2(b) where – for co mparison – uniform amplitude weights as
well as -40 dB Chebyshev weights have been applied. In both cases no grating lobes appear
and – as expected – the sidelobe level is reduced d own to -40 dB in the later case where a
beamwidth of 6° is achieved. This value is the same as for the original sparse receive array
with uniform weights. This is possible as the lengt h of the effective aperture is longer by ( M –
1) d. -90 -60 -30 0 30 60 90 -80 -60 -40 -20 0
θ (° ) Magnitude response (dB) Dense transmit
array Sparse receive
array

(a)

-90 -60 -30 0 30 60 90 -80 -60 -40 -20 0
θ (° ) Magnitude response (dB) -40 dB Chebyshev weights
uniform weights

(b)

Fig. 5-2 : Resolving Grating lobes of effective aperture us ing nulls of transmit array pattern: (a)
azimuth cuts of transmit and receive array pattern with uniform amplitude weights and (b) azimuth cut
of two-way antenna pattern with uniform and with -4 0dB Chebyshev weights; M=8, N=3.

Universität der Bundeswehr Hamburg Page 28
5.3 Implementation of MIMO Concept in FMCW Radar
MIMO concept requires orthogonal waveforms, i.e. th e signals of the transmitters must be
separable at the receiver in order to individually weight the amplitude and phase of the MxN
signals of the combination of each transmit antenna array element i and each receive antenna
array element k. Orthogonal waveforms for the individual transmitt ers of FMCW radar can be
realized for instance by Time Division Multiplex (T DMA), Frequency Division Multiplex
(FDMA), Beat Frequency Division or Chirp Rate Divis ion. In the case of Beat Frequency
Division time or frequency staggered chirps can be used. Fig 5-3(a) displays frequency
staggered FMCW chirp signals with chirp rate α and respective instantaneous transmitter
frequency fti as individual waveform for each Tx element i ( i=1,…, M)

t ifftfst i α+− += ) 1( )(0 t

where f0 +fst (i-1) denotes the start frequency of the individual c hirps. In order to distinguish in
the beat frequency domain between the signals origi nating from different transmit antenna
array elements of a quasi-monostatic radar the freq uency increment fst has to be chosen
according to fst > ατd, where τd is the maximum round-trip time of the signals give n by i
tf0
Tc 05
4
3
2
i=1 fi
f0+B
f0+f st
≈≈ ≈≈

(a)

A
DA
D
vTx1 (t)vb11 (t)
vTx ivb1i (t)
vTx Mvb1M (t)Splitter Rx 1
Rx k
Rx NRx 1Rx 1
Rx kRx k
Rx NRx NMIMO
beam-
forming A
DA
D
A
DA
DRange-
Doppler
Transf.

(b)

Fig. 5-3 : Implementation of MIMO Concept in FMCW radar by beat frequency division: (a)
frequency staggered chirps of individual transmitte rs i ( i=1,2,.., M) and (b) block diagram of receiver
hardware and software functions to separate all M Tx channels for all N Rx channels .

Universität der Bundeswehr Hamburg Page 29 Dxi
Dyi
θyaw xyd(M-1) dd(M-1) d
dD = M d
θθθ

(a)
-90 -60 -30 0 30 60 90 -60 -50 -40 -30 -20 -10 0
angle (° ) antenna pattern (dB)

(b)
-90 -60 -30 0 30 60 90 -60 -50 -40 -30 -20 -10 0
angle (° ) antenna pattern (dB)

(c)
Dx= [0 1 5 0 ] m , D y= [ 0 0 0 ]m
Dx= [0 1 5 0 ] m , D y= [ 0 1 5 0 ]m 3 0 Dx= [ 0 0 0 ]m , D y= [0 1 5 0 ]m 10 10
10 10 Dx=[0 10 0] m ,Dy=[0 0 0] m , θyaw =0 Dx=[0 0 0] m ,Dy=[0 10 0] m , θyaw =0
Dx=[0 10 0] m ,Dy=[0 10 0] m , θyaw =5°
Fig. 5-4: Mobile surface wave MIMO radar: (a) example of dis tortion of effective aperture due to θyaw as
well as surge Dxi and sway Dyi , of individual Rx platform i = 1…3, resulting antenna pattern of effective
aperture for various patterns of misalignment (b) w ithout and (c) with phase correction.

Universität der Bundeswehr Hamburg Page 30
cR
dmax 2=τ (2)
with c denoting the speed of light in vacuum and Rmax the maximum usable range. For a HF
SWR system Rmax is limited due to the attenuation of the surface w ave. Assuming Rmax = 200
km results in τd ≤ 1.33 ms. Thus, constraints due to maximum unambigu ous range are not a
problem and ( M-1) fst << B can be fulfilled using repetition frequencies Tc-1 which allow for
unambiguous Doppler covering the speed of ships. Th is implies that the implementation of the
MIMO concept does not lead to significant increase in bandwidth requirement. Note that for
graphical reasons Fig. 5-3(a) is not drawn to scale and the constraint ( M-1) fst << B is not
fulfilled.
The signal of each receive array element (Fig. 5-3( b)) is splitted after amplification and down
mixed with the individual transmit chirp signals wi th instantaneous frequency fti. With
adequate bandpass filters the beat signals vbi,k(t) originating from each combination of transmit
element i and receive element k can be obtained separately. Filtering could be omi tted if the
individual transmit elements would radiate sequenti ally. Effectively this would be the TDM
concept. However, it reduces the rate between succe ssive range transforms by a factor of M-1
and, thus, the maximum resolvable Doppler frequency .

5.4 Compensation of Pattern Distortion caused by Di splacement due to Wind and
Waves
Swell and wind can cause some misalignment of the g eometrical antenna arrangement. This
will cause a distortion of the virtual effective ap erture. We consider a MIMO radar with M=8
transmit elements and N=3 receive elements synthesizing a virtual linear a rray with an effective
aperture of MxN = 24 elements with Chebyshev weighting (side lobe level SSL =40 dB) and
spacing d=λ/2. The element pattern are assumed to be omnidirec tional. The starting frequency
and the bandwidth of the chirp signals are f0=3 MHz ( λ=100m) and B=αTc=100 kHz,
respectively. Assuming a maximum range of 200 km w e obtain a maximum round-trip time of
τd = 1.33 ms and, thus, chose fst > ατd = 2.66 kHz.
Fig. 5-4(a) displays a naval formation hosting all M=8 Tx antenna elements on a single ship
and N=3 Rx antenna elements each on a single ship. Fig. 6-4(b) shows the resulting distorted
antenna patterns for various values of the paramete r of yaw angle θlyaw (Tx ship) and the
individual surge (vector Dx) as well as sway (vector Dy) of the Rx platforms i=1..3. Surge and
sway of the Tx platform has no impact on the virtua l effective aperture – as long as it can be
considered static within the coherent integration t ime. Similarily there is no impact due to a
possible yaw of the Rx platforms. From Fig. 5-4(b) it can be concluded that a misaligned inter-
ship formation pattern as well as non-synchronous r oll of Rx platforms has severe impact on
antenna pattern of effective aperture which results in mainlobe broadening and in increased
sidelobe level.

Fig. 5-5: Compensation of pattern distortion due to misplace d antenna phase centres: Compensate for
look angle θ dependant path difference L.

Universität der Bundeswehr Hamburg Page 31 Taking the feeding point of the undisplaced virtual antenna as a reference, the total phase
change observed from a point in the far field due t o horizontal displacement and tilt is
calculated as the sum of the phase change due to an tenna tilt and the phase variation due to
longitudinal and transversal shift Dxi and Dyi of the feeding point of physical Rx element i and
yaw angle θyaw of Tx platform (Fig. 5-5) [12]. As the actual posi tions of elementary phase
centers of floating sparse array are known a dynami c adaptive algorithm for a look angle
dependant compensation of phase and amplitude error s of each virtual antenna element can be
used for array pattern correction. Mainlobe broaden ing is resolved and sidelobe level is
decreased significantly (Fig. 5-4(c)). The algorith m is simple and real time applicable. Pattern
cannot be perfectly reconstructed as actual angle o f arrival θa of individual received signal is
not known but scan angle of the array. Therefore co rrection works well in the vincinity of the
main lobe but gets worse for off axis beams (i.e. s idelobes).

Literature
[1] R. Dinger, “Development of a shipboard high-frequen cy surface wave radar for anti-ship missile
detection,” 3rd NATO/IRIS Joint Symposium , Quebec, Canada, 19.-23. Oct. 1998.
[2] A. Bourges, R. Guinvarch, “Perspectives of the use of high frequency radar on buoys,” Oceans
2005-Europe , pp. 1256-1259, Brest, June 2005.
[3] Fickenscher, Th.; Gupta, A.; Hinz, J. O.; Holter s, M.; Zölzer, U.: MIMO Surface Wave Radar
Using Time Staggered FMCW Chirp Signals, European R adar Conference EuRAD 2011,
Manchester, Oktober 2011
[4] Z. Long, and X. Liu, “Gratinglobes Resolving in Spa rse Array Beamforming,” Int. Conf. on Radar
2006, CIE’2006, pp. 1-4, Shanghai, Oct. 2006.
[5] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 4 th status
report, 26.07.2010
[6] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 6th status
report, 18.05.2011
[7] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 5th status
report, 24.11.2010
[8] S. Berger, “Nonuniform Sampling Reconstruction Appl ied to Sparse Array Beamforming”, 2002
IEEE Radar Conference, Long Beach, CA., pp. 98-103.
[9] G. Lockwood, and S. Foster, “Optimizing the Radioti on Pattern of Sparse Periodic Two-
Dimensional Arrays”, IEEE Trans. Ultrosonics, Ferroelectrics, And Freque ncy Control , Vol. 43,
pp. 15-19, 1996.
[10] R. Haupt, “Thinned Arrays Using Genetic Algorithms” , IEEE Trans. Antennas & Propagation ,
Vol. 42, pp. 993-999, 1994.
[11] G. Frazer, Y. Ambramovich, and B. Johnson, “Spatial ly Waveform Diverse Radar: Perspectives
for High Frequency OTHR ,” IEEE Radar Conference 2007 , Boston, USA, pp. 385-390, June
2007.
[12] Fickenscher, Th.; Gupta, A.; Ziehm, S. ; Karstä dt, P.: Horizontal Displacement and Tilt Angle
Compensation for Large Sea Floating Sparse Linear A rray for Surface Wave Radar, International
Radar Symposium IRS 2011, Leipzig, September 2011.

Universität der Bundeswehr Hamburg Page 32

Universität der Bundeswehr Hamburg Page 33 6. Clutter Canceller Beamformer for Clutter Suppres sion

6.1 Summary
At HF band the sea clutter power dominates the inte rnal receiver noise and clutter has a
characteristic behavior. In a fully developed sea t he waves having wavelength equal to half the
radar wavelength corresponds to a strong Bragg back scatter. This severely hinders the ability
to detect targets. When the target velocity is near to the velocity of the sea wave causing the
Bragg backscatter of the radar signal the sea clutt er echo can be many times higher than that of
a big sailing ship. The state of art clutter reduct ion algorithm uses Space Time Adaptive Filter
(STAP) to filter out the Bragg lines in the space, time and azimuth domain. Our research
indicates that due to the spectral and statistical nature of sea clutter STAP has limited
application in HF SWR. Moreover in its attempt to s uppress the first order Bragg clutter this
clutter suppression algorithm also ends up attenuat ing the target signal within the Bragg lines.
Hence with the state of art STAP algorithm detectio n of targets within the sea clutter is not
possible.

A novel clutter suppression scheme is invented whic h filters the sea clutter in the azimuth
domain. An estimate of the sea clutter signal is ma de with the help of an auxiliary azimuth
window called as Estimator Beam (EB). The signal fr om EB is coherently subtracted from the
received signal which results in higher Signal to C lutter Ratio (SCR). Since the dominant
noise signals are external in nature a further incr ease in SNR is achieved by a non-linear
amplifier. This beamformer is termed as Clutter Can celler BeamFormer (CCBF).

The proposed CCBF has the advantage to be even more robust to possible array distortion
(e.g. position displacement in a floating array). I t is shown that the proposed CCBF shows
lesser degradation as compared to conventional Cheb yshev BeamFormer (CBF) in low SNR
scenarios. Moreover the smallest target which can b e spatially resolved by CCBF has a RCS
of -10 dBsm which in case of CBF is -8 dBsm. The de gradation in the beamwidth when the
beamformer is steered to large angles (| ψL|> 30) is also less in case of CCBF than convention al
CBF. Assuming ideal antenna patterns for all the be ams, a homogeneous background and a
point target an enhancement in SCR of 2.48 dB as co mpared to conventional 16 element
Chebyshev BF (SLL=-40 dB is reported over a wide sc an angle. When introducing significant
pattern distortion by misplacing antenna elements o r adding a reradiator in the vicinity of the
array only very little impact on clutter cancellati on is observed. For the first time in open
literature sea clutter suppression within the first order sea clutter is reported. The proposed
scheme is more effective and less computational int ensive than STAP.

6.2 Principle of CCBF
The clutter signal in each range Doppler cell is pr oportional to the resolution of the radar in
azimuth and range. The beamwidth of the two way rad iation pattern is approximately given by
the ratio of radar wavelength λ to length of the antenna aperture (effective apert ure in case of
the MIMO approach presented in chapters 4 and 5). F or microwave radars monopulse clutter
cancellation techniques has been investigated that slightly improve SCR for a given antenna
aperture. When applying monopulse cancellation tech nique with sum and difference beam to
HF radar scenario, results are shown to be less sa tisfactory. However, using a CCBF utilizing
three beams (Fig. 6.1), a Flat Beam (FB), a Differe nce Beam (DB) and an Estimator Beam
(EB) appreciable increase in SCR is demonstrated in practical application of coastal HF SWR
[4] and [5]. All beams have identically located nul ls except for the DB which has an additional
null in the centre of the main lobe. The EB has a 3 dB dip at the centre of its main lobe

Universität der Bundeswehr Hamburg Page 34 -80 -60 -40 -20 0 20 40 60 80 -60 -50 -40 -30 -20 -10 0
ψ (° ) Normalized Power (dB)

Difference Beam
Estimator Beam
Flat Top Beam

Fig. 6.1 : Horizontal antenna pattern of Flat Top Beam (FB) , Difference Beam (DB) and
Estimator Beam (EB) for 16 element monopole array ( d=λ/2)
U1R
D
UFB
(ψL)R
DUEB
(ψL)R
DUDB
(ψL)R
D∑ ∑ ∑…
wF1 (ψL)wE1 (ψL)wD1 (ψL)1rnr
),,(ˆFB L j iDRy ψUnR
D
wFn(ψL)wEn(ψL)wDn(ψL)
1
FB )] ( [max −
Lψ U
∑ ∑1
FB )] ( [max −
Lψ U1
FB )] ( [max −
Lψ U
),,(CCB L j iDR y ψ),,(ˆDB L j iDRy ψ+-+- +-Range Doppler
transformed complex
digital antenna signals …
U1R
D
UFB
(ψL)R
DUEB
(ψL)R
DUDB
(ψL)R
D∑ ∑ ∑…
wF1 (ψL)wE1 (ψL)wD1 (ψL)1rnr
),,(ˆFB L j iDRy ψUnR
D
wFn(ψL)wEn(ψL)wDn(ψL)
1
FB )] ( [max −
Lψ U
∑ ∑1
FB )] ( [max −
Lψ U1
FB )] ( [max −
Lψ U
),,(CCB L j iDR y ψ),,(ˆDB L j iDRy ψ+-+- +-Range Doppler
transformed complex
digital antenna signals …

Fig. 6.2 : Block diagram of the CCBF with range Doppler transf ormed complex signals from
individual antenna elements, beamforming, normaliza tion, coherent clutter subtraction and non linear
amplification. compared to the maximum of the FB. The EB is used t o dynamically estimate the clutter
signal which is coherently subtracted form the sign als of the FB and the DB, respectively. As
this technique represents a filtering in the spatia l domain, radar platform motion – linear
motion as well as pitch and roll of a floating plat form – does not affect the principle of
operation.

After range Doppler transform and beamforming appli ed for all three beams the absolute
voltage values for the respective Range Doppler (RD ) map at look angle ψL are stored in the
matrixes UFB (ψL), UDB (ψL), and UEB (ψL), respectively (Fig. 6-2). Subsequently the indivi dual
RD maps are normalized by the maximum value of UFB (ψL). It is assumed that apart from
some target signal, the EB contains only external n oise which is in practice dominated by
clutter returns for most regions of interest in the RD map for ship detection in low and

Universität der Bundeswehr Hamburg Page 35 -90 -60 -30 0 30 60 90 -50 -40 -30 -20 -10 0
ψ (° ) Normalized Power (dB)

Clutter Canceller BF
Chebyshev BF

Fig. 6.3 : Normalized azimuth response of CCBF 20*log 10 (y CCBF /y CCBF max ) in comparison with radiation
pattern of Chebyshev BF 20*log 10 (y C/y Cmax ) with SLL = -40 dB; 16 element array scan angle ψL=0°

medium ranges. Furthermore in HF band internal nois e of the receiver can also be neglected.
The cancellation of sea clutter (or to be more prec ise external noise) is performed by
subtracting the estimated external noise signal fro m the FB and DB outputs, respectively. The
outcome of this operation in the range-Doppler-azim uth space are the elements
),,(ˆL j i FB DRy ψand ),,(ˆL j i DB DRy ψrespectively. Finally, a nonlinear amplification is applied by
element wise multiplication of ),,(ˆL j i FB DRy ψand ),,(ˆL j i DB DRy ψ resulting in the output
signals ),,(L j i CCB DR y ψof the CCBF.

From Fig. 6-3 it can be concluded that in the look direction ψL=0 the difference signal of FB
and EB as well as that of EB and DB FB yˆand DB yˆ, respectively, are 3 dB below the output
signal yC of a Chebyshev BF. Thus, considering the case of a point target there is a loss of
target signal of 6 dB in the look direction. Keepin g in mind that normalization of the curves in
Fig. 6-3 has removed the offset of 6 dB, the areas below both graphs in Fig. 6-3 are a measure
for the reduction of clutter power by the CCBF. The amount of reduction in clutter power (8.5
dB for a homogeneous background) is significantly h igher than the loss in signal power (6dB),
thus, SCR is effectively increased by 2.5 dB in the case of a point target. However, this value
is slightly dependant on the side lobe level (SLL) chosen for both beamformers.

6.3 Measurement Results
We use the measurement results obtained by a coasta l FMCW radar station (WERA) run by
the Helmhotz-Zentrum Gesthacht which is located at the North West German coast at
Wangerooge [4]. Operating frequency was set to 12.2 7 MHz and transmitter power to 4 W.
The 165 m long receiver array consists of 16 elemen ts and transmitter was a flood light 4
element square antenna array. A single chirp used f or each range transform was 0.26 s long
with a bandwidth of 100 kHz resulting in a range re solution of ∆R = 1.5 km. A set of 512
chirps was used for Doppler discrete Fourier transf orm (Doppler resolution ∆f = 0.015 Hz).
Fig. 6-4 shows the RD plot with a look angle of –3. 5° (measured clockwise from boresight)
with conventional CBF and Fig. 6-5 displays the sam e data set with CCBF. An appreciable
reduction in the general noise floor level is obser vable. The reduction in the clutter power in
the first order and second order Bragg is clearly v isible. To analyze the performance of the
clutter suppression in detail a set of 7 targets ar e selected and marked in Fig. 6-5. Targets 1, 2,

Universität der Bundeswehr Hamburg Page 36
Doppler (Hz) Range (km)

-1 -0.5 0 0.5 120
30
40
50
60
-80 -70 -60 -50 -40 -30 -20 -10 0

Fig. 6-4: RD map for look angle of -3.5° of a 16 element Che byshev beamformer (SLL = – 40
dB)
Doppler (Hz) Range (km)

-1 -0.5 0 0.5 120
30
40
50
60
-80 -70 -60 -50 -40 -30 -20 -10 0
Target 1
Target 2 Target 3
Target 4 Target 5 Target 6
Target 7

Fig. 6.5: RD map for look angle of -3.5° of a 16 element sea clutter canceller beamformer. Effective
increase in SNR is achieved by non linear amplifica tion and coherent clutter cancellation.

6 and 7 are more clearly visible within the first o rder sea clutter. This proves the effectiveness
of the proposed clutter suppression scheme within c lutter dominated regions. For the first time
improvement in SCR for targets within the first ord er Bragg line is reported in open literature.
The spreading in the power levels of target 3 is al so significantly reduced. Targets 4 and 5
which lie very close to each other are more clearly visible after CCBF processing due to the
finer azimuth filtering. A detailed evaluation of C CBF over many data sets is presented in the
status report [5].

6.4 Impact of Antenna Pattern Distortion
For practical applications like coastal or sea floa ting arrays as well as for shipboard antenna
systems time variable distortion of the antenna pat tern has to be considered. In the case of
coastal arrays these distortions are caused by chan ges of the coupling to the surface wave due
to tidal effects (intertidal mudflats), changes of the electrical conductivity distribution or
topology of the ground or degradation of the ground screen, antenna matching or feeding

Universität der Bundeswehr Hamburg Page 37

Fig. 6-6: Impact of antenna element displacement on radiation pattern of FB, EB, and DB of CCB in
comparison with Chebychev BF (SSL=-40 dB); 16 eleme nt array, scan angle ψL=0°.
-80 -60 -40 -20 0 20 40 60 80 -50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0
ψ (° ) Normalized Power (dB)

CCBF (MATLAB)
CCBF (FEKO)
CBF (MATLAB)
CBF (FEKO)

Fig. 6-7: Impact of antenna element displacement on normaliz ed azimuth response yCCB of CCB
in comparison with radiation pattern of Chebyshev B F simulated with MatLab (array factor
pattern) and FEKO; 16 element array, scan angle ψL=0°. network, respectively. Proper operation of the CCBF allows only for little relative distortion of
the antenna pattern of the FB, DB, and EB, respecti vely. Otherwise subtraction of the signals
of FB and EB as well as subtraction of the signals of EB and DB will result in significant
error. Thus we investigate the impact of antenna pa ttern distortion on the output of the CCBF
and compare the results with that for conventional Chebyshev beamformer (designed for -40
dB SSL in the undistorted case) for a 16 element mo nopole array at a frequency of f = 15
MHz. In the first step antenna element coupling is neglected. Antenna patterns are calculated
analytically from the superposition of individual e lement signals considering the individual
weights and omnidirectional radiation patterns. Pattern distortion is introduced by
misalignment of antenna elements. The displacements are calculated considering a random
function with Gaussian distribution and a standard deviation of λ/16.

Fig. 6.6 displays the distorted pattern of the arra y for the three beams of the CCBF in

Universität der Bundeswehr Hamburg Page 38 -8 0 -60 -40 -2 0 0 20 4 0 60 80 -50 -40 -30 -20 -10 0
ψ (°) Normalized Power (dB)

CCBF Chebyshev BF
-8 0 -60 -40 -2 0 0 20 4 0 60 80 -50 -40 -30 -20 -10 0
ψ (°) Normalized Power (dB)

CCBF Chebyshev BF

Fig. 6-8: Impact of reradiator on normalized azimuth response yCCBF of CCBF in comparison with
radiation pattern of Chebyshev BF simulated with FE KO; 16 element array, scan angle ψL=0°.
comparison with that for the Chebyshev BF for the s can angle ψL=0. The corresponding
normalized azimuth response y CCBF of the CCBF calculated from the array pattern of t he 3
beams using the respective array factor pattern (Ma tLab) as well as using 3D field solver
(FEKO) is shown in Fig. 6.7. The Half Power BeamWid th (HPBW) of CCBF is increased
from 4.8° to 5.5° and for Chebyshev BF from 9.0° to 9.5°. The SLL of the CCBF – with the
exception of both close by sidelobes – is still sig nificantly (more than 10 dB) lower than that
for the Chebyshev BF. Considering a point target th e increase of SCR still remains high at a
high value of 2.55 dB.

In a next step, the impact of a reradiator in the v icinity of the array on the azimuth response of
the CCBF is investigated and compared with the ante nna pattern of the corresponding
Chebyshev BF (Fig. 6.8). Therefore 3D field simulat ions have been carried out for an
equispaced linear array with a reradiator built out of a metallic cuboid of width 2 λ, depth λ and
height λ /4 located on the ground plane behind the antenna. The centre of the cuboid is offset
from the centre of the array by 2 λ along endfire direction and 1.5 λ along boresight with a
skew angle of 10° with respect to the array axis. H PBW of the CCBF is increased from 4.8° to
5.0° and for Chebyshev BF from 9.0° to 9.25°. The S LL of the CCBF – with the exception of
both close by sidelobes – is more than 15 dB lower than that for the Chebyshev BF.
Considering a point target the increase of SCR stil l remains high at a value of 2.21 dB.
Literature
[1] R. Dinger, “Development of a shipboard high-frequen cy surface wave radar for anti-ship missile
detection,” 3rd NATO/IRIS Joint Symposium , Quebec, Canada, 19.-23. Oct. 1998.
[2] A. Bourges, R. Guinvarch, “Perspectives of the use of high frequency radar on buoys,” Oceans
2005-Europe , pp. 1256-1259, Brest, June 2005.
[3] Fickenscher, Th.; Gupta, A.; Hinz, J. O.; Holter s, M.; Zölzer, U.: MIMO Surface Wave Radar
Using Time Staggered FMCW Chirp Signals, European R adar Conference EuRAD 2011,
Manchester, Oktober 2011
[4] Z. Long, and X. Liu, “Gratinglobes Resolving in Spa rse Array Beamforming,” Int. Conf. on Radar
2006, CIE’2006, pp. 1-4, Shanghai, Oct. 2006.
[5] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 4 th status
report, 26.07.2010

Universität der Bundeswehr Hamburg Page 39 [6] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 6 th status
report, 18.05.2011
[7] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 5 th status
report, 24.11.2010
[8] S. Berger, “Nonuniform Sampling Reconstruction Appl ied to Sparse Array Beamforming”, 2002
IEEE Radar Conference, Long Beach, CA., pp. 98-103.
[9] G. Lockwood, and S. Foster, “Optimizing the Radioti on Pattern of Sparse Periodic Two-
Dimensional Arrays”, IEEE Trans. Ultrosonics, Ferroelectrics, And Freque ncy Control , Vol. 43,
pp. 15-19, 1996.

Universität der Bundeswehr Hamburg Page 40

Universität der Bundeswehr Hamburg Page 41 7. Electrically Small HF Antenna

7.1 Summary
Due to long wavelengths at HF band, a major limitat ion of current HF SWR system is the
large physical dimension of the transmitter and rec eiver antenna/array. To achieve vertical
polarization many coastal systems use λ/4 monopoles whose length can vary from 2.5 m up to
25 m. It is highly desirable to design HF antennas which have a low profile and can thus be
disguised from the public view on a beach. This lim ited visual impact of the antennas can
provide more flexibility in site planning making th e system more efficient and commonplace.
Homeland security and coastguard application also d esire an antenna which is easy to deploy
on floating platforms like a buoy, barge, ship or a n offshore rig. Large monopoles are less
suitable for such remote and mobile applications.

A capacitive top loaded and inductively coupled sel f resonating HF Electrically Small Antenna
(ESA) with dimensions of 1.45 m in height and 1.6 m in width is proposed in the report which
has a good performance in terms of bandwidth, tunab ility and radiation efficiency [1]. The
proposed antenna can be tuned from 10.7 MHz up to 2 3 MHz [1], [2]. State of the art antennas
which try to maximize the length of radiator in a g iven antenna volume have strong coupling
between antenna components and suffer from low effi ciency [3], [4], [5]. As a design rule the
electrical length of the antenna is increased by ca pacitive loading which keeps the physical
length small. The real part of the impedance is mat ched by an inductive coupling. The
proposed antenna is compact and easy to deploy in r emote coastal locations and floating
platforms. Operating at middle and higher HF bands the dimensions of the antenna are small
enough so that it can be concealed from the public view.
The most important aspect of the antenna is the hig h efficiency and broadband matching with
the line impedance. The antenna is easily tunable w ith a lumped capacitor placed between the
top loading plate and radiating monopole (Fig.7-1). This capacitor can be bypassed with a
shunt switch causing the antenna to operate at its lowest frequency. The measured value of the
Quality factor (Q) of the proposed antenna is 39.1 which is very close to 25.2; the fundamental
limit calculated form Chu Mclean equation [6]. The Chu Mclean equation defines the lower
limit on the size of an antenna for a given efficie ncy and Q factor. This means that for the
given efficiency further reduction in size of anten na will lead to lower bandwidth. The product
of efficiency and bandwidth of the proposed antenna is nearly 4 times better than any other HF
ESA in the open literature. This parameter is a cri tical figure of merit for any antenna since it
relates two important and conflicting bottlenecks i n design. All the above properties make this
antenna ideal for both coastal and mobile HF SWR ap plications.

7.2 Antenna Structure
The proposed antenna is a capacitive loaded and ind uctively coupled self resonant monopole
which is designed to operate from a frequency range of 10.7 MHz up to 23 MHz. The
proposed design consists of six sections [1]. A hal f scaled model of this antenna shown in Fig.
6-1 was built and tested (frequency range 21.4 to 4 6 MHz). The dimension and description of
each section of the half scaled antenna are:

1) Capacitive Plate – It has a radius of 0.8 m and since it is not pra ctical to realize a
capacitive cap with a solid copper disk, it has bee n approximated with 4 flat spirals
resting on 8 radials. The capacitive top spiral pla te does not contribute to radiation but
ensures a nearly constant current distribution alon g the monopole.

Universität der Bundeswehr Hamburg Page 42

Fig. 7-1 : Capacitive loaded and inductively coupled ESA fo r HF Surface Wave propagation. The 1:2
scaled model has a total height of 0.743 m (0.053 λ @ 21.4 MHz) and width of 0.799 m (0.057 λ @ 21.4
MHz).
2) Tuning capacitor — Ceramic capacitors of values from 10 pF to 470 pF t o achieve
resonance from 23 MHz upto 11 MHz respectively.

3) Monopole – A monopole of length 0.743 m (0.053 lambda @ 21. 4 MHz) is the main
radiating element of the ESA since the current dens ity is highest at this part of antenna.
Moreover due to the capacitance between the cap and its image there is a constant
current distribution along the monopole. The curren t amplitude from the bottom of the
monopole to the top only varies by 5%.

4) Primary Coil – The feeding to the antenna is through an air cor e transformer which
transforms the input impedance of the monopole to a 50 ohm line. Its dimensions are
(0.1 x 0.2) m 2. Primary coil of the transformer consists of a sin gle coil and is connected
to the feeding port of the antenna on one terminal and another terminal is shorted to the
ground plane. The current density on the primary co il is low at resonance so it doesn’t
radiate, and only facilitate matching.

5) Secondary Coil – Monopole is fed via the one and a half turn seco ndary coil of the
transformer. Its dimensions are (0.1 x 0.2) m 2. One of its terminals is connected to the
main radiating structure hence has high current den sity while another terminal is
grounded.

6) Ground Plane – The ground plane used is made from (1 x 1) m 2 copper plate which
was extended by 24 ground radials of length 10 m ea ch.

7) Feed Port – At lower MHz range SMA connector is used to real ize the feeding point to
the antenna .

Universität der Bundeswehr Hamburg Page 43
7.5 10 12.5 15 17.5 20 -30 -20 -10 0
Frequency (MHz) S11 (dB)

Without Tuning Cap (Experimental)
56pF Tuning Cap (Experimental)
Without Tuning Cap (Simulated)
56pF Tuning Cap (Simulated)

Fig. 7-2 : Proposed HF ESA: Simulated and measured return l oss versus frequency (scaled to full size
antenna).
7.3 Simulation and Measurement Results
The simulation and optimization of the ESA were car ried through FEKO field simulator [7].
The tunable capacitor is placed between the monopol e and the cap as shown in Fig. 7-1. The
top cap is isolated from the monopole by using a sm all plastic plate. With the help of a two
way switch the capacitor can either be in series wi th the top cap and the monopole or can be
totally bypassed. The results for specific antenna performance parameters are presented in the
following subsections [1] [2]:

A. Return Loss
The return loss of the antenna is shown in Fig. 7-2 . When the tuning capacitor was bypassed,
the simulated resonant frequency was expected to be at 10.7 MHz but the measured resonance
of 1:2 scale model was observed at 23.8 MHz. This c orresponds to an unscaled resonance at
11.9 MHz which is seen in the plot. The shift in th e resonance frequency is due to the parasitic
capacitance of the isolator between the main radiat ing monopole and the metallic capacitive
cap. The measured -10 dB bandwidth is 360 kHz which is large enough for both radar
applications and supporting many voice communicatio n channels. The variable capacitor is
now switched in series with the monopole and the to p cap. The capacitor is set to a value of 56
pF. The measured resonance frequency shifts to 13.7 MHz with a -10 dB bandwidth of 370
kHz. It has been proven by simulation that capacito r ranging from 5 pF to 100 pF corresponds
to a frequency variation from 23 MHz to 13 MHz. Thi s validates the tunability of the antenna
across the HF band which is critical in application s of frequency hopping and avoidance of
broadcasting noise/jammers.

B. Radiation Pattern
Fig. 7-3 shows the simulated radiation pattern of t he HF ESA which is similar to that of a
quarter wave monopole. To avoid reflections from th e ionosphere one desires a deep null
directly above the antenna and maximum gain in the azimuth plane. Hence the radiation
pattern of the proposed HF ESA is satisfactory.

Universität der Bundeswehr Hamburg Page 44

(a) (b)

Fig. 7-3 : (a) Normalized radiation pattern of the simulate d HF ESA in dB at 10.7 MHz azimuth cut
(theta=90°). (b) Elevation cut (phi=0°) .
0 0.398 0.796 1.194 1.592 1.989 025 50 75 100
r (m) Quality Factor

Cap Disk
Two arm helix loaded ESA [4]
Proposed HF ESA(Simulation)
Proposed HF ESA(Experiment)
2D Meander line ESA [5]
η=100% η= 75%
η=60%
η=10%

Fig. 7-4: Plots of the fundamental Chu-Mclean limit for diffe rent efficiency (blue curves) for ESA
operational at 12 MHz.
C. Radiation Efficiency
The radiation efficiency is the most critical perfo rmance parameter for an ESA. The radiation
efficiency calculated from the field simulator for the proposed HF ESA was 88% and this was
measured experimentally by Wheeler Cap Method. The measured efficiency was 80%. The
loss in efficiency is due to the insufficient groun d plane which was created using 24 ground
radials spread uniformly around the antenna.

D. Q factor
The Q factor of the proposed antenna is plotted in Fig. 6-4 alongside with the Chu-Mclean
limit for ESA at different efficiencies. The abscis sa represents the radius of the smallest sphere
required to circumscribe the antenna which is desig ned to operate at 12 MHz. It is evident that
the simulated and measured antenna is very close to the theoretical limit of Q achievable.
Since a higher value of Q will imply a lower 3-dB bandwidth, hence the propo sed HF ESA has
the good bandwidth with high efficiency.

Universität der Bundeswehr Hamburg Page 45 Fig. 7-4 is the graphical comparison of the propose d HF ESA with respect to the state of art
systems. The Efficiency Bandwidth product (EB) for different HF ESA can be calculated from
Fig. 7-4. EB is defined as the product of the effic iency with the 3-dB bandwidth of an antenna.
The proposed HF ESA has an EB of 0.045 Hz, the two arm helix loaded ESA has an EB of
0.008 Hz and the 2D meander line ESA has an EB of less than 0.001 Hz. HF Radar
application demand a high bandwidth of 100 kHz or m ore and it is clear that only the proposed
antenna can deliver such high requirement on bandwi dth while maintaining small dimensions
and high efficiency.

Literature
[1] Systemstudie zu verteiltem selbstkonfigurierenden H F-Überhorizontradar auf Schiffen, 7th
status report, 30.11.2011.
[2] A. Gupta, T. Fickenscher, P. Karstädt, D. Grant, ‘A n efficient electrically small antenna at HF
band’, IEEE APMC 2011, pp. 856 – 859.
[3] S. Best, “A Discussion on the Properties of Electri cally Small Self-Resonating Wire Antenna,”
IEEE Antenna and Propagation Mag. , vol. 46, issue 6, pp. 9-22, Dec. 2004.
[4] S. Lim, R. Rogers, H. Ling, “A Tunable Electrically Small Antenna for Ground Wave
Trasmission,” IEEE Trans. on Antenna and Propagation , vol. 54, No. 2, pp. 417-421, 2006.
[5] E. Bronner, Improvements of HF Surface Wave Radars Performance by Compact Antenna
Study and Adaptive Filtering Used to Reduce Sea Clu tter , PhD dissertation, The University of
Paris 6, 2005.
[6] S. Best, “A Discussion on the Properties of Electri cally Small Self-Resonating Wire Antenna,”
IEEE Antenna and Propagation Mag ., vol. 46, issue 6, pp. 9-22, Dec. 2004.
[7] FEKO-EM Software and Systems, www.feko.info .

Universität der Bundeswehr Hamburg Page 46

Universität der Bundeswehr Hamburg Page 47 8. Novel Sub-Clutter Detector

8.1 Summary
Typically radars have a well defined internal noise level and employ Constant False Alarm
Rate (CFAR) detector to maximize the probability of detection ( PD) while keeping the
probability of false alarm ( PFA ) at a user defined constant level. However, detect ion in HF
SWR is limited by sea clutter whose statistical cha racteristics are not well defined. Due to the
resonant nature of sea waves at HF band the power l evel of sea clutter is much higher than in
typical radar scenario, making the detection of tar gets a very challenging task. The state of art
detector uses three dimensional CFAR to adaptively vary the threshold by power regression
along azimuth, range and Doppler cells [1]. This de tector does not guarantee a constant PFA
and has limited adaptability especially in the Dopp ler regions dominated by sea clutter. A
comprehensive literature review proves that no targ et within the first order Bragg has been
detected by this detector.

We propose a two step detection scheme which first uses correlation to detect targets in the
range and Doppler domain. In the second step it use s Minimum Mean Square Error (MMSE)
estimation scheme to localize it in the azimuth dom ain. Using empirical oceanographic
observations we prove that in the presence of a tar get within a RD cell the azimuth variation of
the received power shows higher correlation across adjacent RD cells as compared to the RD
cells without target [2] [3] [4] [5]. This serves a s the basis for the proposed correlation
detector. Our research proves with the help of meas urement data that correlation detector can
detect targets in sea clutter dominated scenarios a s well as discriminate between close by
targets in the range and Doppler domain. An encoura ging agreement is also observed with the
detected targets and AIS records. In the second ste p a unique combination of beamforming
and MMSE Direction Finding (DF) algorithm is used t o resolve the bearing of the target. The
MMSE beamformer estimates the clutter signal to mak e a better estimate of the target bearing
The results prove that the MMSE DF performs better than state of art Chebyshev beamformer
both in near and far ranges [6].

8.2 Detector Architecture
For the example of 16 receiver antennas forming a l inear array the complex voltage signals are
filtered into Nψ azimuth cells, NR range cells and ND Doppler cells (as explained in Section
3.(c) of this report). This 3 dimensional matrix is known as Range Doppler (RD) Azimuth
cube. A vector containing the signals across all lo ok angles at a given range cell i and Doppler
cell j is sliced out from the RD Azimuth cube. This azimu th vector is termed as Vector Under
Test (VUT) and is shown in red in Fig. 8-1. As the first step we use correlation detector to
detect the presence of a target within the VUT. Cor relation detector uses neighbouring RD
vectors to calculate the threshold value in terms o f correlation. The architecture of the
correlation detector and threshold calculation is s hown in Fig. 8-2. The correlation detector
can only localize a target in the RD domain. For ea ch VUT the output from the correlation
detector is either a 1 or 0. If the output from the correlation detector is zero implying that a
target is not present in the VUT, then the detector proceeds to test next RD vector. If the
output of this step is 1 then in the second step of the detector the a priori knowledge of the
presence of a target helps to estimate the bearing of a target much more accurately than only
by beamforming which is the state of the art. The c lutter signal is estimated and is used for
bearing estimation making sure that the mean square estimation error of target bearing is
minimized. The output of this stage of the detector is the bearing of the target. Using the

Universität der Bundeswehr Hamburg Page 48

Fig. 8-1: Flow chart for detection of a target in a given Vec tor Under Test (VUT) = R iDj.
proposed 2 step detection the Range, Doppler and Az imuth bearing of a target can be
estimated.

The R iDj Vector Under Test (VUT), which is the azimuth vect or for range cell R i and Doppler
cell Dj contains Nψ azimuth cells with corresponding look angles ψk ( k=1… N ψ). The respective
received signal power is stored in the vector X with elements xk the received signal power for
the azimuth vector of adjacent range bin R i+1 Dj is saved in the vector Y with elements yk,
respectively (Fig. 8-2). A target in the VUT will i mpact the correlation between X and Y
which is calculated by Pearson Product Moment Corre lation Coefficient (PMCC) r. The
distribution of signal power consisting of target s ignal and sea clutter power in both the
vectors can be approximated by bivariate normal dis tribution so Fisher transform is applied on
r to stabilize its variance, providing the correlati on value zVUT for the VUT. The threshold T is
estimated adaptively by using all the range cells i n the Doppler neighborhood of VUT as
described in [2] [3] [4].
Azimuth estimation is performed by MMSE estimator i f a target is detected in the VUT by the
correlation detector. The measured signal in VUT (s tored in X) is the sum power of target
ψRiDjRi+1 Dj
ψNX1X
2X
ψNY2Y1Y
VUT
ψRiDjRi+1 Dj
ψNX1X
2X
ψNY2Y1Y
VUT ψRiDjRi+1 Dj
ψNX1X
2X
ψNY2Y1Y
ψNX1X
2X
ψNY2Y1Y
VUT

Fig. 8-2 : Block Diagram of correlation detector. The value of threshold T changes adaptively as
VUT shifts from sea clutter dominated region to oth er noise (ionospheric,
thermal )

Universität der Bundeswehr Hamburg Page 49

Fig. 8-3: Flow chart for MMSE azimuth estimation of a target in VUT = R iDj.
signal, sea clutter and other noise (ionospheric, t hermal). It is proved in [3] [4] that clutter
signal in VUT can be approximated by its Doppler ne ighbours. We make two assumptions;
firstly there are no targets in the neighbourhood o f VUT and secondly that the external noise is
constant around the neighbourhood of VUT. So the to tal noise (clutter and other noise) signal
in VUT is estimated by the power in neighbouring Do ppler cells as shown in Fig. 8-3. The
target signal, ij Tˆ in VUT (R iDj) is estimated by presuming the target azimuth ( ψα) and spatially
correlating it to the two way array radiation patte rn. The estimated target signal and noise
signal is summed up to get a final estimate of the signal in VUT, ij Xˆ. The MSE is calculated
and the presumed target azimuth ( ψα) is varied recursively until the MSE is minimized.

8.3 Measurement Results
We use the measurement results obtained by a coasta l FMCW radar station (WERA) run by
the Helmhotz-Zentrum Gesthacht which is located at the North West German coast at
Wangerooge [4]. Operating frequency was set to 12.2 7 MHz and transmitter power to 4 W.
The 165 m long receiver array consists of 16 elemen ts and transmitter was a flood light 4
element square antenna array. A single chirp used f or each range transform was 0.26 s long
with a bandwidth of 100 kHz resulting in a range re solution of ∆R = 1.5 km. A set of 512
chirps was used for Doppler discrete Fourier transf orm (Doppler resolution ∆f = 0.015 Hz).
Correlation detection algorithm is applied on the c ollected data sets. Fig. 8-4 shows the RD
plot with a look angle of –7.5° (measured clockwise from boresight) for 16 element
Chebyshev beamformer. The first order Bragg peak is visible around ± 0.7 Hz and is spread in
a region from ±0.6 Hz to ±0.8 Hz. Fig. 8-5 shows th e variation of z for the corresponding data
set obtained with first stage of the detector. On c omparison of Fig. 8-4 and Fig. 8-5 it can be
observed that sea clutter and external noise have l ower zij values as compared to the targets in
the RD map. As a result targets are clearly visible in Fig. 8-5. Another point to be noted is that
the zij values for external noise are not zero which means that the external noise is not totally
uncorrelated. For many RD cells the corresponding zij have a high value in Fig. 8-5 but a
corresponding target is not visible in the RD power map (Fig. 8-4). This is due to the fact that
these targets do not lie in the azimuth cell corres ponding to look angle of -7.5°.

Universität der Bundeswehr Hamburg Page 50
Doppler (Hz) Range (km)

-1 -0.5 0 0.5 120
25
30
35
40
45
50
55
60
-80 -70 -60 -50 -40 -30 -20 -10 0
2156 4
7 89
10
3

Fig. 8-4: Normalized RD map for look angle = -7.5° for 16 element Chebys hev beamformer. Range
and Doppler bins of Targets 1, 2, 3, 9 and 10 (arou nd first order Bragg peak) have been detected in
the first step by correlation detector.
Doppler (Hz) Range (km)

-1 -0.5 0 0.5 120
25
30
35
40
45
50
55
60
0.5 11.5 22.5 33.5 4
1
2
3456
8 79
10

Fig. 8-5: Normalized z map for look angles from -35° to +35° correspondin g to the RD localization
in the first step of detection process (correlation detector). Targets have higher z ij values than noise
and sea clutter.
For verifying the performance of the first step of the detection algorithm, 10 targets are
selected across the RD map. Targets 1, 2, 3, 9 and 10 are detected in the middle of the Bragg
lines. Targets 4, 5 and 6 lie very close to each ot her within the second order sea clutter and are
successfully resolved. Very close targets with simi lar power like Target 7 and 8 are
successfully detected in external noise dominated s cenario. Details on the performance of
detectors in various regions are reported in [4]. D uring the measurement the AIS data of ships
for 10 data sets are also recorded and are shown in Fig. 8-6 Alongside with the AIS sightings
the locations of targets detected close to the AIS data are also plotted. It must be noted that
these targets are not the only targets detected by the radar but the targets which help in
drawing a comparison of the detector performance wi th AIS recording. Ships at a far range of
120 km are also detected correctly.

Universität der Bundeswehr Hamburg Page 51

Fig. 8-6: Detected target versus 367 AIS contacts recorded fr om 100 data sets for correlation detector.
For verification of the second step of the detector , the bearing information of ships from 367
AIS sightings were compared with the one estimated from the proposed MMSE algorithms
and the results are shown in Fig. 8-7. The MMSE bea ring estimate considerably outperforms
the standard Chebyschev beamformer both at close by range and far off range. The
Chebyschev beamformer tends to miss targets when st eered to angles in excess of 25°. It can
be observed that MMSE estimator does not suffer tha t much from this drawback since there is
successful detection of targets at and beyond 30°. As the power of the target decreases with
distance the error in the estimation also increases . One observes an increased level of wrong
bearing resolution around 100 km. There is some inc onsistency in the detected position of
targets by the detector and AIS records in Fig. 8.6 and Fig. 8.7. Few ships which have an AIS
contact are not detected. These ships are also also not visible on the RD map. Many small
boats which are too small to be detected by HF rada r are common in this sea patch and these
AIS contacts are assumed to be of such small vessel s. The AIS data set was recorded every 45
sec and the RD map was updated every 33 sec, so whi le comparing the two a small amount of
disagreement is expected. Moreover AIS data was fou nd to be inconsistent on more than one
occasion. This is because many boats are not traine d to use and update the factory settings of
the AIS transponder. Our findings in the uncertaint y of AIS data were supported by our
discussions with Centre for Maritime Research and E xperimentation (CMRE), LaSpezia, Italy
who also made similar observations. Even with this drawback of AIS sightings in perspective
it can be safely concluded that the MMSE bearing es timate fairly agree with the bearing
provided by the AIS data.

Universität der Bundeswehr Hamburg Page 52 -30 -20 -10 0 10 20 30 10 20 30 40 50 60 70 80 90 100 110 120
Angle from boresight (° ) Range (km)

AIS Contacts
MMSE Estimates
Chebyshev Beamformer

Fig. 8-7: Target bearing estimated by state of the art Chebys hev beamformer and MMSE azimuth
estimator (second step in detector) versus the reco rded AIS contacts.

Literature
[1] A. Dzvonkovskaya, H. Rohling,‘Target Detectoin with Adaptive Power Regression Thresholding
for HF Radar,’ CIE Internation Conference on Radar 2006, pp.1-4, Oct. 2006.
[2] A. Gupta, T. Fickenscher, “Oceanographic Knowle dge Based Correlation Detector for HF Surface
Wave Radar,” International Conference on Radar Syst ems, Radar 2012, Glasgow, Oct. 2012.
[3] Systemstudie zu verteiltem selbstkonfigurierend en HF-Überhorizontradar auf Schiffen,
8 th status report, 29.05.2012.
[4] Systemstudie zu verteiltem selbstkonfigurierend en HF-Überhorizontradar auf Schiffen,
6th status report, 18.05.2011.
[5] T. Fickenscher, A. Gupta, P. Ludwig ‘Performan ce Evaluation of Correlation Detector for HF
Surface Wave Radar’, Asia-Pacific Microwave Confere nce, APMC 2012, Kaohsiung, Taiwan, 4-7.
Dec. 2012.
[6] A. Gupta, Th. Fickenscher, ‘Oceanographic Model based Azimuth Estimator for HF Surface wave
Radar’, EURAD 2012, pp. 154-157.

Universität der Bundeswehr Hamburg Page 53 9 Improved CFAR-Based Target Detection
The most commonly used detection methods in radar a pplications are CFAR (Constant False
Alarm Rate) algorithms. They work by estimating the local detection background from
adjacent ADR cells and deriving a detection thresho ld from it. Usually, a nearly homogeneous
level of clutter and noise power in the background is assumed. Obviously, in the vicinity of the
spectral peaks of the resonant first order Bragg ba ckscatter of the sea, this background is far
from homogenous, and hence, off-the-shelf CFAR dete ctors offer very poor performance for
HF OTHR applications. In addition to the detection approach investigated in section 8,
improvements and extensions of the CFAR scheme have been developed as a possible
alternative. The key idea is to employ a newly deve loped pre-segmentation step to distinguish
between clutter and noise dominated cells. Based on this segmentation, the estimation of the
background power is adapted to consider only suffic iently similar cells, i.e. other clutter-
dominated cells if the cell-under-test is itself cl utter-dominated and other noise-dominated
cells if it is noise-dominated. A further improveme nt is achieved by considering multiple
successive scans by applying scan-by-scan averaging . As the maximum velocity of ships is
limited, they will appear in the same ADR cell for multiple scans, so that averaging will
improve the contrast to the fluctuating background. Finally, to mitigate multi-detection of
single targets, an adjacent target merging algorith m is developed that fuses detection in
adjacent cells due to power leakage of a single tar get. These extensions to the basic CFAR
method greatly improve the performance for the cons idered application scenario

9.1 Resolution Capabilities
Table 9-1 displays the typical resolution cell size based on 3dB beamwidth (azimuth),
bandwidth (range) and CIT (Doppler). Even if the HF SWR resolution cell size in azimuth and
range is coarse, the long CIT provides a good Doppl er resolution. Thus, the probability of two
or more targets in the ADR vicinity of each other i s quite small. However, this does not apply
for Naval formations where ships are sailing with s ame speed.
This resolution can be increased reasonably in all dimensions up to one order of magnitude by
applying Super Resolution Techniques. In this case cell size is reduced and number of cells is
increased accordingly. However for this technique s everal limitations apply for ship detection
with HF SWR. Fast targets as well as manoeuvring or accelerating targets may cross cell
boundaries during CIT causing target smearing and, thus, reduced probability of detection.
Secondly separation of adjacent targets faces furth er limitation and cannot be expected to be
increased by the same amount as the resolution. As in the case without super resolution a weak
target can always be masked by a strong close by ta rget.
In chapter 8.2 super resolution in azimuth domain w ith a cell size of 1° has been used. In the
following no super resolution is applied, and, thus , in our dataset distance between adjacent
targets to be separable has to be either 1.5 km in range, 6.38° in azimuth or 0.09 m/s in radial
speed. Nevertheless it should be emphasized that a single target can be localized with up to an
order of magnitude higher precision (chapter 9.5).

Table. 9-1 : Typical HF SWR resolution cell size

Equation Typical values
range resolution ∆r = c / (2·B) ∆r ≥ 1500 m with B = 100 kHz
Doppler resolution
speed resolution ∆fD = 1 / (T CIT )
∆vr = – ( ∆fD·c)/(2·f 0) ∆fD ≥ 0.0075 Hz with T CIT = 133s
∆vr ≥ 0.09 m/s with f 0 = 12.273 MHz
azimuth resolution ∆θ ≈ 102° / M ∆θ ≥ 6.38° with M=16

Universität der Bundeswehr Hamburg Page 54 9.2 Short Introduction to CFAR in General
Constant False Alarm Rate (CFAR) Target Detection ( TD) is iterative processing applied to
each ADR resolution cell to make a binary decision between "target present" or "noise/clutter
only", with the currently investigated resolution c ell denoted as Cell Under Test (CUT).
As opposed to other local peak detection approaches , CFAR TD has the important property of
a constant but adjustable (design) false alarm rate , which is a desired property in military
applications.

For each CUT the group of CFAR algorithms determine a local detection threshold S based on
reference cells in the local neighbourhood of the C UT. The optimal CFAR detector in
homogeneous noise/clutter is the Cell Averaging (CA ) CFAR detector with a large number of
reference cells, where the reference window can be arbitrarily oriented in the ADR domain.
For CA CFAR the threshold S is based on the mean va lue of the considered reference cells and
a constant scale factor T to maintain the desired f alse alarm rate.

Using a large reference window in the 1D/2D/3D ADR domain also increases the risk of a non
homogeneous background in the reference cells, char acterized by the following two cases: 1.)
multiple targets and 2.) clutter noise transitions. The second case of clutter noise transitions in
HF SWRs is likely to occur at the boundaries betwee n sea clutter and external noise and
results in a so-called clutter edge.

To cope with the two types of non homogeneous detec tion background, approaches like Cell
Averaging Smallest Of (CASO) CFAR, Cell Averaging G reatest Of (CAGO) CFAR or
Ordered Statistics (OS) CFAR have been developed. W hile CASO is aimed to solve the
multiple target situation, CAGO is aimed at mitigat ion of the increase of false alarms at clutter
edges. Depending on selected parameters, OS CFAR ca n be used to mitigate multiple targets
or clutter edges at a moderate detection loss of lo w SNR targets.

All CFAR approaches assume the noise or clutter cel l power to be distributed according to
some probability density function (pdf), usually an exponential distribution. Only if this
assumption holds the TD provides the desired false alarm rate. For that reason, the I/Q
components of external noise and first order sea cl utter cells of measured radar data have been
analyzed in [6] and show good agreement with a Gaus sian distribution, respectively an
exponential distribution in power.

According to available literature, current HF SWR o perate on the whole ADR data cube with
the same CFAR parameters (such as reference window shape, desired probability of false
alarm, type of CFAR algorithm) irrespective of the local detection background. This leads to a
suboptimal detection performance in homogeneous or non homogeneous regions or both. For
that reason we proposed an approach called adaptive CFAR detection with presegmentation
[7], which is presented in the following section.

9.3 Presegmentation

The adaptive CFAR detection with presegmentation [1 ] is a two-step process: It first performs
a global presegmentation of the ADR data into exter nal noise and sea clutter dominated cells.
Based on the results from the presegmentation, the CFAR parameters (such as reference
window shape and design probability of false alarm) are chosen adaptively [1].
In contrast to conventional CFAR approaches, this a daptation is based on global information
from the presegmentation and not derived locally fr om the predefined local CFAR reference

Universität der Bundeswehr Hamburg Page 55 window. This global a-priori information can be use d to avoid crossing of clutter edges in the
detection process. In addition, this global knowled ge makes it possible to operate two different
probabilities of false alarms: one in the noise- an d one in the sea clutter dominated areas.
The presegmentation starts by a segmentation betwee n first order sea clutter and external
noise. It uses the fact that the approximate Dopple r frequency position of the Bragg lines is
very stable and can be well approximated by
c) (πf) (g ±=fD ⋅ ⋅/ , (9.1)
in which g is used to denote the standard gravity, f denotes the radar operating frequency and c
is used to denote the speed of light. The calculate d Doppler positions are now compared to the
true sea clutter spectrum and a decision about the Doppler extend of the first order sea clutter
is made.

Fig. 9-1 : Simulated input range Doppler map and segmentatio n of first- and second-order sea clutter –
'white' indicating sea clutter dominated cells

In the following, a combined estimation of the aver age external noise power and an estimation
of the first order sea clutter extend in range are performed. Based on these results the second
order segmentation is performed. A scheme of the pr esegmentation is illustrated in Fig. 9-1,
showing the input range Doppler map, the first- and second order segmentation mask and the
combined segmentation mask. The segmentation mask c an now be used to 1.) adapt the
reference window shape as well as 2.) the probabili ty of false alarm in each of the two
segments. An example of the reference window shape adaptation is illustrated in Fig. 9-2,
showing a situation in which the CUT is situated in between two clutter edges.

Fig. 9-2 : CUT in mixed clutter/noise background with nonada pted (8×8) and adapted (15×1) reference
window shape

Thus, for the non adapted case, the reference windo w is partly dominated by external noise
and partly by sea clutter. The adapted reference wi ndow shape, based on the presegmentation
process, is adapted in such a way to avoid crossing the clutter edges. Although the adapted
window can be asymmetric (15×1) the same total numb er of reference cells before and after
the adaptation is considered. The global adaptation is clearly superior to non adapted
approaches since it operates on a homogeneous refer ence window. Regarding the actual CFAR
algorithm it can be combined with any of above pres ented, with the smallest detection loss
offered by CA CFAR.

Universität der Bundeswehr Hamburg Page 56

(a) (b) (c)
Fig. 9-3 : Comparison between (a) conventional CA CFAR detec tion, (b) CA CFAR with reference
window adaptation and two constant false alarm rate s, (c) conventional CA CFAR with generally
higher false alarm rate for 21 targets aligned in d istinct range-Doppler positions

The second part of the adaptation is the independen t selection of the false alarm rate in the sea
clutter and the external noise dominated parts. By looking at Fig. 9-1 it can be seen that the
external noise dominated cells outnumber their sea clutter dominated counterpart manifold.
This makes it possible to use a higher false alarm rate (and thus higher probability of detecting
small SCR targets) in the clutter dominated cells a nd maintain a lower false alarm rate in the
external noise dominated sections. A comparison bet ween 1.) the non-adapted case, 2.) the
reference window and false alarm rate adaptation an d 3.) the global reduction of the constant
false alarm rate is illustrated in Fig. 9-3.

9.4 Scan-by-Scan Averaging

Typically, most published signal processing approac hes for HF SWR are applied to a single
scan or coherent integration interval (CIT). Scan-b y-Scan Averaging (SBSA) is an approach to
combine the ADR data from several scans to reduce t he false alarm rate and/or improve the
detection performance. To evaluate the potential of SBSA and the following parameter
selection, the target behaviour in the context of t he coarse HF SWR resolution is investigated.
After that the impact on the number of detections i s illustrated. The resolution capabilities of a
typical HF SWR are presented in Table 9-1. The coar se HF SWR range resolution ∆r of 1500
meters and the coarse azimuth resolution ∆θ of 6.38° can be explained by the limited usable
bandwidth B and the limited space for the M element antenna array, respectively. Approaches
like MIMO Beamforming (refer to chapter 4 and 5) ca n help to improve the azimuth
resolution.

The long coherent integration time T CIT of 133 s leads to a fine Doppler resolution of 0.0 075
Hz (or 0.09 m/s in terms of speed resolution ∆vr). To reduce the Doppler resolution cell size,
the only parameter to be changed is the CIT, but ex actly this increase in CIT is not possible
due to the following reason: The detection is based on the assumption that during each CIT the
target is situated in the same ADR. This is certain ly fulfilled for far distant ship targets with
moderate speed in terms of coarse Azimuth dimension , but reaches its limitations in terms of
target manoeuvring and the fine Doppler domain as s hown in [5]. An additional increase of the
CIT would further restrict any target manoeuvres (r adial speed changes during the CIT). A
violation will lead to a spreading into adjacent Do ppler bins and the detection probability is
decreased. The analysis of the power of several tar get cells across different scans shows that
the power of each target is very stable for several scans, except for long term effects due to the
increase or decrease in range. SBSA uses Finite Imp ulse Response (FIR) and Infinite Impulse
Response (IIR) filters to perform a running power a verage for each ADR cell across different
scans. Depending on the radial target speed of a po tential target, a certain number of scans can
be used in SBSA. For the above mentioned T CIT with a 3/4 overlap and assuming a target with

Universität der Bundeswehr Hamburg Page 57 a radial speed of 5.87 m/s, a maximum of 4.67 scans can be used in the averaging process until
the target passes from one range cell to the next a nd spreading occurs. Based on an analysis of
Automated Identification System (AIS) data we have derived that 90% of all AIS contacts
have an absolute radial speed of less than 5.1 m/s and SBSA with an FIR averaging filter of
order two is save to be used. The following evaluat ion has been performed on real data and
shows the total number of detections N D across 101 scans by using an FIR filter with
b[0]=b[1]=b[2]=1/3 and a IIR filter with b[0]=2/3 a nd a[1]=1/3.

(a) (b) (c)
Fig. 9-4 : (a) Number of detections (ND) with and without SB SA, (b) CFAR detection on input, (c)
CFAR detection on FIR SBSA data

It can be seen in Fig. 9-4 (a) that the total numbe r of detections N D can be significantly
reduced by the use of SBSA. Still, it has to be kep t in mind that N D is in fact the sum of target
detections (either the target itself or the spreadi ng of targets into adjacent bins) and false
alarms. Even if AIS is not a perfect ground truth w ith the help of Fig. 9-4 (b)(c) it seems
justified to state that the reduction of SBSA in N D is mainly due to reduction in the number of
false alarms. The problem of how to treat multiple detections of the same target (due to
spreading of power into adjacent bins) is considere d in the next subsection.

9.5 Adjacent Detection Merging and Target Parameter Estimation

To solve the multiple detection problem, an adjacen t detection merging algorithm (ADMA) for
HF SWRs has been developed. In the general radar li terature, ADMA is also denoted as Plot
Extractor (PE) or Hit Processor (HP), with most lit erature only explaining the basic concept
[4] [5]. In the optimal case, all adjacent detectio ns of the same target are mapped into one hit.
Besides the correct assignment there are three more cases to consider: 1.) several targets are
merged into one hit, 2.) several adjacent noise/clu tter detections are merged and 3.) adjacent
noise/clutter detections and target detections are merged. Clearly Case 1 is of most concern, as
one or several targets are lost. Case 3 is not desi red, but can be tolerated. Case 2 even helps us
in maintaining a low probability of false alarm.

As the number of detections from the same target is a function of target power to surrounding
noise power, a fixed solution of merging a predefin ed number of cells in the ADR domain is
prohibitive. A better solution is found by using Co nnected Component Labeling (CCL) [3], an
algorithm widespread in the field of computer visio n and in the detection of connected regions
in binary images. The algorithm iteratively compare s each pixel to the local neighbourhood
(4/8 connectivity) and assigns the pixel a new labe l if it is different from the neighbourhood.
This makes CCL with 4 connectivity (2D) or 6 connec tivity (3D) a good candidate to be
applied to the 3D binary detection cube.

Universität der Bundeswehr Hamburg Page 58

Fig. 9-5 : Total Number of hits and number of detections ass ociated to each hit

It solves the multiple detection problem with no l oss of targets, as long as any two hits are not
directly adjacent in range or Doppler or Azimuth. I n addition, it is a valuable input to target
parameter estimation (TPE). Figure 9-5 shows how N D from Fig. 9-4 is reduced to half of the
number of hits N H. For the TPE a Center of Gravity (COG) algorit hm is proposed, which
uses the set of detections associated to each hit t o estimate the target position in terms of
range, Doppler and azimuth position. An accurate TP E is essential for subsequent processing,
like target tracking, to produce reliable tracks. T he COG has been selected to be a good match
to CCL, as it directly takes the associated detecti ons of the hits and performs a low-complexity
estimation based on the power distribution of the c onsidered cell and the relative distance to
each other.

But also a combination of CCL with interpolation t echniques is possible. In the end Figure 9-
6 is shows the sequential steps of CFAR, CCL and CO G.

Fig. 9-6: CFAR CCL and COG on RD map

Literature
[1] J. O. Hinz, M. Holters, U. Zölzer, A. Gupta and T. Fickenscher: Presegmentation-Based Adaptive
CFAR Detection for HF SWR, IEEE Radar Conference (R adarCon) 2012, Atlanta, May 2012.
[2] J. O. Hinz, M. Holters and U. Zölzer: Scan-by-Scan Averaging and Adjacent Detection Merging
to improve Ship Detection in HF SWR, International Conference on Signal Processing and
Communication Systems (ICSPCS) 2012, Gold Coast, Au stralia, December 2012.
[3] R. M. Haralick, L. G. Shapiro: Computer and Robot V ision, Volume I, Addison-Wesley, 1992.
[4] S. Kingsley and S. Quegan, Understanding Radar Syst ems, Scitech Publications, 1999
[5] G. V. Trunk, Range Resolution of Targets Using Auto matic Detectors, IEEE Transactions on
Aerospace and Electronic Systems, AES-14, Issue 5, Sep. 1978.

Universität der Bundeswehr Hamburg Page 59 [6] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 7th status
report, 24.11.2011.
[7] Systemstudie zu verteiltem selbstkonfigurierenden HF-Überhorizontradar auf Schiffen, 8 th status
report, 29.05.2012

Universität der Bundeswehr Hamburg Page 60

Universität der Bundeswehr Hamburg Page 61 10. Future Prospects

HF SWR is an old radar technology which nowadays is adapting fast to the new challenges of
cost efficient littoral monitoring. It has been dem onstrated that with the novel approaches
proposed in this report the performance of HF SWR w ith respect to ship detection can be
increased significantly. However, there are many do mains yet to be explored to make it a more
effective technology for ship detection from a comp act mobile platform. Future HF SWR will
need higher angular resolution, range enhancement o f more than 200 km and will be equipped
with credible sub clutter detector. These enhanceme nts will increase the capability of the radar
to detect small go fast boats used by pirates, ille gal traffickers and smugglers.

Accurate modeling of sea clutter in the time and sp ectral domain for monostatic and
multistatic HF SWR on a floating platform can enhan ce the design and detection process of
the receiver chain. There has been initial study do ne by ONERA, France, but the theoretical
model does not fully match with the few measurement results conducted on a moored platform.
Both for coastal and mobile HF SWR the statistics o f sea clutter are also not well studied and
modeled. As a result the theoretical estimation of probability of detection PD, probability of
miss PM and probability of false alarm PFA is not possible. In state of the art systems empir ical
estimation of PM and P FA is also not possible since not all the ships in a patch of sea do use
Automatic Identification System (AIS). Moreover as we observed that the AIS data collected
can be partly erroneous. The state of art detectors which claim to achieve Constant False
Alarm Rate (CFAR) can not in reality assure a fixed PFA for a HF SWR. Having a better
understanding of the statistics can help design opt imum detection and tracking algorithm. It
will then be possible to have a fair comparison of CFAR detector with other detection schemes
and determine what is the theoretical maximum PD achievable for a given sea state, radar
frequency and PFA . To achieve such an estimate of sea clutter statis tics measured data for a
long duration of measurement campaign – both for co astal and mobile HF SWR – will have to
be analyzed. Since radar returns from ships and boa ts will corrupt the measured clutter
estimate ideally all the AIS records of ships in th e measurement are required so that their
undesired echoes can be filtered out. An empirical estimate of PM and PFA can be achieved by
a measurement campaign in which a known target with varying level of maneuverability is
placed in the sea (for different sea states) and de tected using various detectors.

New antenna integration techniques are needed to ov ercome the challenge in the placement of
an antenna array onboard a ship in order to minimiz e the distortion of the radiation patterns
caused by the superstructure of the ship. A possibl e solution to this problem is optimized
beamforming which can dynamically minimize the effe ct of pattern distortion of the array.
Another innovative solution to the problem of array integration can be the use of existing HF
antennas on board a frigate/naval formation of frig ates to have a combination of beamforming
and direction finding. However, as soon as multiple ships are involved, a suitable
communication among the various platforms will be d emanded. This Another essential
challenge is the ElectromMagnetic (EM) compliance o f the HF SWR board the ship.

Universität der Bundeswehr Hamburg Page 62

Universität der Bundeswehr Hamburg Page 63 List of Abbreviations

ADC analog to digital converter
ADMA adjacent detection merging algorithm
ADR azimuth Doppler range
AIS automatic identification system
ATD advanced technology demonstrator
BF beam forming/former
CA cell averaging
CAGO cell averaging greatest of
CASO cell averaging smallest of
CBF Chebyshev Beamformer
CCBF clutter canceller beamformer
CCL connected component labeling
CFAR constant false alarm rate
CIT coherent integration time
COG center of gravity
CUT cell under test
DARPA defense research project agency
DB difference beam
DF direction finding
DPCA displaced phase center antenna
EB estimator beam
EB efficiency bandwidth product
EM electro-magnetic
EMC electro magnetic compatibility
ESA electrically small antenna
FB flat beam
FDMA frequency division multiple access
FIR finite impulse response
FMCW frequency modulated continuous wave
GRWAVE ground wave
HF high frequency
HP hit processor
HPBW half power beam width
I/Q inphase/quadrature
IIR infinite impulse response
MARSEN marine remote sensing experiment
MCR multifrequency coastal radar
MIMO multiple input multiple output
MMSE minimum mean square error
NRL naval research laboratory
OS ordered statistics
OTH over the horizon
OTHR over-the-horizon radar
PE plot extractor
PMCC product moment correlation coefficient
Q quality factor
RAL Rutherford appleton laboratory
RCS radar cross section
RD range Doppler

Universität der Bundeswehr Hamburg Page 64 RF radio frequency
Rx receive(r)
SBSA scan-by-scan averaging
SCR signal to clutter ratio
SLL side lobe level
SNR signal to noise ratio
STAP space time adaptive processing
SWR surface wave radar
TD target detection
TDMA time division multiple access
TPE targer parameter estimation
Tx transmit(ter)
VUT vector under test
WERA WEllen RAdar