1

High Precision Interferometric Radar for SheetThickness Monitoring

Sebastian Mann, Christoph Will, Torsten Reissland, Fabian Lurz, Stefan Lindner, Sarah Linz, Robert Weigel,and Alexander Koelpin

Abstract—Contact-less sensing plays an important role intoday’s highly automated industrial environment, as modernsensors allow for an increased product quality in high-speedmanufacturing lines. Hereby, a measurement system for sheetthickness monitoring in aluminum and steel rolling mills ispresented. The presented concept is based on a differential mea-surement principle using two cooperating six-port radar systems.The radio frequency front-ends are designed in a high densitysubstrate integrated waveguide technology at a frequency of61 GHz. Furthermore, a detailed analysis of non-ideal behavior ofmono-static six-port radar front-ends is presented. Moreover, themeasurement principle is evaluated in a simplified measurementsetup and the results are compared to previously presentedtheory. Finally, an overview of modern sensing technology forsteel rolling mills is presented and compared to the proposedmeasurement system considering the important aspects, i.e.update rate, precision, and thickness limit.

Index Terms—radar systems, interferometry, six-port circuits,microwave circuits, substrate integrated waveguide (SIW), planarwaveguides

I. INTRODUCTION

METAL PROCESSING, e.g., aluminum, iron and steelindustry, is a market characterized by intense competi-

tion. Therefore, product quality as well as cost reduction havebecome increasingly important, especially for many Europeanand US American companies. In order to improve yield andquality of produced metal sheets, surface quality as well astheir final material thickness, which is also referred as stripexit thickness, are important characteristics to be monitoredin rolling mills. Material sheets are typically rolled to theirfinal thickness in several steps. For each operation, the in-tended sheet thickness is defined by the rolls’ gaps. However,roll eccentricity caused by non-uniform thermal expansionor inexact roll grinding can introduce a thickness error of40 µm and even more in cold and hot rolling mills [1], [2].Thus, roll eccentricity defines the lower limit of the achievablethickness tolerances. For a further reduction of the tolerances,an active compensation has to be applied to the role mill [3].In order to realize such compensation systems, inline sheet

The research project Kurzwegradar (engl. short-range radar) was supportedby the Bavarian Ministry of Economic Affairs and Media, Energy andTechnology, Munich, Germany, project grant No. MST-1208-0005// BAY175/001.

Sebastian Mann (sebastian.mann@fau.de), Christoph Will, Thorsten Reiss-land, Fabian Lurz, Stefan Lindner, Sarah Linz and Robert Weigel are withthe Institute for Electronics Engineering, Friedrich-Alexander UniversityErlangen-Nuremberg (FAU), 91058 Erlangen, Germany.

Alexander Koelpin is with the Chair for Electronics and Sensor Systems,Brandenburg University of Technology, 03046 Cottbus, Germany.

thickness sensors are inevitable as roll eccentricity changeswith temperature and operating speed [1].

Two different inline sheet thickness measurement systemsare well established since more than 50 years: mechanical rollforce measurement [4] and X-ray based thickness gauges [3],[5]. Thickness measurement using the roll force method isan indirect measurement method making use of a fix relationbetween the plastic deformation constant of the processedmaterial, the output sheet thickness deviation and the measuredroll force deviation. However, quality related changes of thematerial parameters influence the metered value [2]. X-raybased gauges are based on measuring the signal power ofa nucleonic or X-ray beam transmitted through an absorbingmaterial [5]. Photo detectors are used to obtain the transmittedradiation intensity in form of a voltage value, which is directlyrelated to the material’s absorption value as well as the sheetthickness. With today’s maximum thickness tolerances beingclearly defined with specification values below 0.8 % of theexit thickness, X-ray detectors reach their limit, as they sufferby their nature from a significant random noise componentin the detector output signal. Thus, current control technology(roll thickness measurement and X-ray gauges) can not isolateperiodic components caused by present roll eccentricity of lowamplitude [3].

The described major drawbacks of both systems highlightthe need for alternatives. However, the challenge is not onlyto monitor little changes in thickness and quality of thematerial, but doing this under difficult conditions, whichare the hazardous environment, the high operating speed oftypically 20 m/s or higher, and the big varieties of surfacedefects, e.g. cracks, blowholes, and scratches [6]. Establishedcontact-free sensing technologies like ultrasonic or opticaltechnology can hardly be used in the harsh environment,especially if high update rates are mandatory [7]. Radarsystems are perfectly qualified for harsh industrial environmentand show competitive performance at reasonable costs [7],[8]. Although classical radar architectures, i.e. pulse radarand frequency modulated continuous wave (FMCW), mayfeature the recommended accuracy, they are limited in theirupdate rate by nature [9], [10], [11]. In contrast to that,interferometric continuous wave (CW) radar systems (withoutany modulation) feature high precision displacement detectionand extraordinary high update rates at the same time [12], [13].Therefore, CW radar is a good candidate for monitoring sheetthickness variation in the described application. A special typeof CW radar is the six-port reflectometer, using a passive six-port junction with power detectors instead of a conventional

2

Fig. 1. Sketch of the system design.

mixer [14].Hereby, a metal sheet thickness monitoring concept based

on six-port interferometry is presented and validated in ex-periment. The publication is organized as follows: SectionII will introduce the system concept, followed by a detaileddescription of the radio frequency frontend design in SectionIII. Furthermore, this section will give an overview of ex-pectable non-ideal effects in hardware design. The influenceof the non-ideal behavior of the six-port radar system on thebaseband signals will be analyzed in Section IV. The basebanddesign and implementation of the sensor for the simplifiedtest setup will be shown in Section V, while Section VI dealswith the measurement setup, calibration, and dedicated signalprocessing. Additionally, the functionality of the sensor will bevalidated by acquired measurement data. Section VII gives acomprehensive comparison of the presented sensor with state-of-the-art technology. Section VIII finally concludes this paper.

II. SYSTEM DESIGN

A sketch of the system design is displayed in Fig. 1.Intending to replace conventional X-ray gauges in existing aswell as new rolling mills, the system design basically relies onthat of a conventional gauge. Therefore, one or several radarsystems pair are mounted opposite on the top and bottom sideof the transportation path, where the metal sheet is guided.As a relative measurement of the sheet thickness change issufficient in the desired application, the equally constructedradar systems are realized as six-port interferometers. Sincethe frequency f is directly related to the wavelength λ byλ = c0/f , with c0 being the speed of light in the surroundingmedia, a higher system frequency is equivalent to an increasedsensitivity regarding small displacements [13]. For millimeterwave frequencies the International Telecommunication Union(ITU) has qualified three license free frequency bands (2.4-2.5 GHz, 24-24.25 GHz, and 61-61.5 GHz) for Industrial, Sci-entific and Medical (ISM) applications [15]. Furthermore,there are several open frequency bands beyond 100 GHz.In terms of evaluation of the system concept 61 GHz has

been chosen as design frequency, as commercial-of-the-shelfcomponents are hardly available at higher frequencies. Thus,in combination with an analog-to-digital converter (ADC),providing an effective number of bits (ENOB) of 18 bits, aresolution in the two digit nanometer range can theoreticallybe achieved, which is sufficient. The dual antenna concept isnecessary by means of the thickness deviation measurementand, furthermore, helps to compensate for vibrations of themoving metal sheet. However, to allow high speed operation,exact positioning of the antenna with an opposite pointingvector is mandatory. Finally, the sampling rate of the 8-channelparallel ADC and the data transmission limits the maximumspeed of the sheet movement. For a sheet moving with 20 m/s(787.4 inch/s), assuming a sampling rate of 10 kSa/s, a samplevalue is obtained every 2 mm (78.74 mil). Although antennaswith limited directivity are used, this is much smaller than theilluminated spot size. For this reason, keeping in mind that anintegral value is obtained, it is assured that occurring thicknesschanges will be monitored.

III. RF FRONTEND DESIGN

Both radio frequency (RF) front-ends that are employed inthe measurement system are equally constructed as describedin this section. Due to the superior performance of substrateintegrated technology compared to other planar technologyat frequencies higher than 30 GHz [16], substrate integratedwaveguide (SIW) has been chosen as main transmission linetype. However, interconnections to other transmission linestyles are inevitable, as especially active components, e.g.,amplifiers and diodes, need to be integrated to the RF structure.The front-end is illustrated as block diagram in Fig. 3 and canbe divided in several components, which are the signal source,the radar coupler, the antenna, the variable attenuator, thelow noise amplifier (LNA) and the six-port junction includingthe power detectors. The generated signal SLO is equallysplit up by the radar coupler and half of the power (STx)is radiated by the antenna. The other part (S′2) is coupled tothe reference path where its power level is adjusted using avariable attenuator. The transmitted wave is reflected at thetarget’s surface and partly received by the antenna structure.An LNA is utilized in order to increase the receive signal

BalunVCO

LNA

Antenna

Attenuator

Six-port receiver

-3dB-Coupler

PLL

Fig. 2. Foto of the manufactured RF front-end.

3

BGT60

LP VCO VGA

PLL

TCXO

LNA

Six-port receiver

B3 B4 B5 B6

∆d

d

TargetTx/Rx antenna

S′1 S′

2

S1 S2

SLO STx

SRx

Fig. 3. Block diagram of the RF front-end.

power (S′1) and to achieve a low total noise figure [7].Afterward the amplified receive signal S1 and the referencesignal S2 are superimposed within the passive six-port junctionand down-converted to baseband by four diode detectors [7],[14].

A. Signal source

1) Signal generation: The radio frequency signal sourceis mainly based on a commercial V-band Infineon BGT60transceiver millimeter wave integrated circuit (MMIC) provid-ing two balanced signal output pairs for both, the millimeterwave signal (57 GHz to 64 GHz) as well as the referencedivider signal. Furthermore, the MMIC features a variablegain amplifier (VGA) enabling an adjustment of the transmitsignal power using a serial programming interface (SPI). Forsettling the frequency of the mm-wave signal, the referencedivider of the voltage controlled oscillator (VCO) is connectedto a phase-locked-loop (PLL) integrated circuit (IC) of typeHittite HMC703 with an active loop-filter using a differentialMSL. The reference clock of 50 MHz for the PLL IC’s phasediscriminator is provided by a common temperature controlledQuartz oscillator (TCXO) distributed on the system’s back-end.

2) MMIC-SIW-transition: The connection of the signalsource MMIC delivered in an embedded wafer level chip scalepackage (eWLB) to the passive SIW circuit is accomplishedin two steps. First, a two-section half-wave balun in groundedcoplanar waveguide (gCPW) technology is used to transformthe balanced MMIC output to an unbalanced signal [17],second, a tapered line transition between gCPW and SIW isemployed making use of their similar modal fields [18].

B. Radar coupler

Reflecting the work of Henry J. Riblet of 1952 [19], theradar coupler is realized as single layer short slot directionalcoupler, consisting of two adjacent waveguides with an open-ing in the common wall [18]. The length of the central aperturewhich defines the coupling ratio of the directional coupler [20],is chosen for a coupling ratio of -3 dB. In the area of the

z

y

x

ϑ

ϕ

Fig. 4. Computer aided design (CAD) draft of the antenna PCB withintegration in the metal case. The orientation of the antenna coordinate systemcorresponds to the presented simulation data.

aperture, the outer waveguide walls are slightly shifted to themiddle in order to compensate dispersion effects and enlargingits bandwidth. The exact dimensions of the geometry havebeen optimized using CST Microwave Studio software.

C. Antenna

As the antenna is forming the interface between radarsystem and measuring path, an appropriate antenna designsuitable for the desired application is essential. By meansof the described near-field application, the antenna needs tofulfill several criteria: small aperture size, medium gain withit’s maximum in boresight direction, and easy integrationinto the chosen highly integrated System-on-Substrate conceptwith SIW line type on a 127 µm Rogers RO3003 material.In literature, two main SIW antenna types can be identified,which are slotted SIW antennas (e.g., [21], [22]) and leaky-wave SIW antennas (e.g., [23], [24]). However, both types arespace-consuming, narrow-band, and their antenna efficiencydecreases with reducing substrate height. Furthermore, thoseantenna types do not meet the requirements of small aperturesize. Therefore, other concepts like tapered slot antennas aremore suitable in the purposed application [25], [26]. Hence, aSIW surface antenna has been designed based on an end-fireVivaldi concept which has been first-time presented by the au-thors in [27],[28] In order to achieve better system integrabilityand more symmetric elevation behavior, the antenna concept

4

0 30 60 90 120 150 180-180

-120

-60

0

60

120

180

ϑ in DEG

ϕin

DE

G

−30

−20

−10

0

10

-3 dB

-10 dB

Gain in dBi

Fig. 5. Surf plot of the antenna gain over azimuth (ϕ) and elevation (ϑ). The-10 dB as well as the -3 dB beamwidths are marked additionally.

has been extended by two aluminum shielding caps mountedon both sides of the PCB as shown in Fig. 4. This figure alsoprovides the alignment of the reference coordinate system,which was defined for the calculation of the antennas’ farfielddata. Fig. 5 depicts a two-dimensional surf plot of the realizedantenna gain, providing the gain in dBi as color information. InFig. 6, the antenna gain is additionally plotted in the xz-plane(delineated in Fig. 4), as this pattern is mainly relevant for thepreviously described application. For moving targets parallelto the yz-plane (refer to Fig. 4), the antenna pattern introducesa low-pass behavior to the measured distance information.As to be obtained by Fig. 6, the introduced shielding of theantenna structure results in both an increased gain for boresight(ϑ = 90 degrees, ϕ = 0 degrees) and less steep filter slopes ofthe low-pass behavior.

0 30 60 90 120 150 180

−10

0

10

ϑ(ϕ =0 DEG) (DEG)

Dire

ctiv

ity(d

Bi)

with shielding without shielding

Fig. 6. Antenna gain in elevation (xz-plane).

D. Low noise amplifier

The amplification of the received signal is performed bya commercial LNA MMIC of type UMS CHA2159-099Fsupplied by a dedicated driver network. Since the device isdelivered as bare die it needs to be bonded to the substratematerial using a cavity in order to keep the wires as shortas possible. Fig. 7 gives the layout of the LNA assembly andthe designed matching networks for the input and output signalpaths referred as S′1 and S1, respectively. An equivalent circuitof the input and output matching and the transition to SIWis displayed in Fig. 8. First, the transformation of the TE10

mode of the SIW to qTEM mode of the conductor backedcoplanar waveguide is conducted by a linear tapered microstripsection. The top layer ground conductor is tapered in order toreduce discontinuity effects, additionally. The length of thetapered section is approximately up to quarter wavelength inSIW, the dimensions have been optimized using a full-waveEM simulation tool. The matching of the chip’s input andoutput impedance Zin/Zout with the serial ribbon bond wireof inductance Lb is done by a high impedance transmissionline and a parallel capacitance, which is realized as two parallellow impedance open stubs. Therefore, the length of the bondwire is crucial to the input and output matching of the LNA.In order to achieve a good matching and low loss the length ofthe bond wires has been optimized to be as short as possible(330 µm) by mechanical testing. Furthermore, the length andthe shape were considered within the EM simulation of theLNA’s matching network.

UMS

In

S′1 S1

1 mmVg1 Vg23 Vg4

Vd1Vd23Vd4

Top Layer Cavity 130 µm Vias

Fig. 7. Layout, chip assembly and ribbon bonds of the LNA circuit on a127 µm Rogers RO3003 laminate with cavity.

TE10 qTEM

SIW gCPW Lb MMICSIW /gCPW

S′1 /

S1

Zin /Zout< λ/4

Fig. 8. Equivalent circuit of the LNA’s input/output matching network.

E. Variable attenuator

In many millimeter wave systems variable attenuators arebased on MMIC technology [29], [30], [31]. In general, bothtransmission type and reflection type attenuators are alsorealizable using discrete components [32], however, reflectiontype approaches seem to be more feasible at V-band as therequirement specification concerning the parasitics of the ac-tive element is lower [33]. Reconfigurable SIW structures havebeen shown in several publications many of them dealing withtunable filters [34] or phase shifters [35]. In [36] a reflectivevaractor tuned SIW phase shifter based on a quadrature hybridwith reflective loads is presented, making use of transformingthe TE10 mode of SIW to a qTEM mode of a parallel plate

5

line. Such design is also feasible for an attenuator structurereplacing the varactors by PIN diodes. However, a transitionof SIW mode to coplanar waveguide or microstrip line is morestraightforward, less space consuming and the design of thebiasing network is easier to accomplish.

The proposed reflective attenuator is based on the pre-sented radar coupler structure and two identical reflectiveloads connected to the coupler’s output ports. The operationprinciple of suchlike attenuator is extensively described in[33]. The reflective load is displayed in Fig. 9 and is mainlybased on a short SIW-to-MSL transformation network and anAluminum-Gallium-Arsenide (AlGaAs) flip-chip PIN diode oftype Macom MA4AGFCP910, which is qualified for operationup to more than 70 GHz. In order to reduce the transmissionline’s influence, the diode is directly soldered on the taperedsection of the transformation network and connected in shuntconfiguration. The RF ground connection and the bias networkare realized using two quarter-wave radial stubs and a highimpedance transmission line section of a quarter wavelength.The attenuation value is set by Vtune, provided by a digital-to-analog converter (DAC), connected to the diode using aserial resistor limiting the bias current. The attenuation of the

TE10

SIW λ/4 RVtune

SIW /

MSL

λ/4 λ/4

Fig. 9. Equivalent circuit of the attenuator’s reflective element.

described attenuator is adjustable using a 10-bit DAC in arange from 10 dB to 35 dB. While the minimum attenuationvalue is limited by the series impedance of the PIN diodes,the maximum attenuation value depends on the isolation valueof the utilized short-slot coupler.

F. Six-port receiver

1) Passive six-port junction: Analogous to the six-portdesigns in microstrip technology, six-port junctions are mostlybased on combinations of power dividers, couplers and phaseshifters [37]. For the design of single layer six-port junctionsin SIW technology two types of power dividers are feasi-ble. Those are the T-junction and the Y-junction [38], [39].Furthermore, two coupler designs are interesting for a singlelayer design: The well-known short-slot coupler as well as acruciform shaped directional coupler [40]. Although the useof cruciform shaped couplers allows for highly space efficientsix-port designs [41], its bandwidth as well as its requirementfor low manufacturing tolerances plead for utilization ofparallel line short-slot couplers. Fig. 10 shows the designedsix-port wave correlator structure using three short-slot cou-plers and one T-junction power divider. In order to achievehigh relative bandwidth and stability against manufacturingtolerances, a symmetrical design without the use of lineardelay elements has been realized [42], [37]. The input portshave the designators P1 and P2, while P3-P6 are the output

ports of the structure. Furthermore, port P7 is terminated bya match, realized using a SIW-MSL transition and a 50 Ωflip-chip high frequency thin film resistor of type VishaySfernice CH0201650RGFT in shunt configuration. By meansof the symmetric structure, the amplitude and phase imbalanceof the six-port junction is directly related to the imbalanceof the designed short-slot couplers. Therefore, the amplitudeimbalance ∆ac of the designed passive six-port junction is lessor equal to 0.6 dB, while the phase imbalance ∆ϕc is smallerthan 3 degrees over the interesting frequency range.

2 mm

RogersRO3003127 µm

P1

P2 P7

P3

P4

P6

P5

Fig. 10. Six-port wave correlator structure.

2) Power detector: The conversion of the four six-portcorrelator output signals S3-S6 is achieved by four matchedzero bias Schottky diode detectors [14]. The detector structureshown in Fig. 11 is mainly based on a prior design utilizinga low-cost Skyworks SMS7630-061 silicon zero bias diode,accurately described in [42]. As the influence of parasiticelements gets crucial with increasing frequency, a Gallium-Arsenide (GaAs) zero bias Schottky diode of type KeysightHSCH9161 in discrete beam lead housing has been chosenfor a redesign featuring higher voltage sensitivity and lowernoise floor due to its alternative semiconductor technologywith lower parasitics. The reactive matching network as wellas the output filter have been adapted as shown in Fig. 11.

Si

Bi

1 mm

Fig. 11. Layout of the detector circuit (R=51 kΩ, C=12 pF).

It is common knowledge and specially shown for six-portreceivers in [43], that the ambient temperature has a stronginfluence on a Schottky diode detector’s characteristic. Thus,temperature effects can’t be neglected in six-port radar sys-tems. Related to the temperature distribution over the completepassive circuit board, a temperature compensation needs to beapplied to every single digitized baseband signal within thesignal processing routine.

6

IV. NON-IDEAL BEHAVIOR OF SIX-PORT RADAR SYSTEMS

Non-ideal effects degrade the Six-Port radar system’s per-formance [7]. In general, there are three non-ideal effects to beidentified, which are noise, limited isolation and the receivernon-linearity [44], which is mainly based on the non-idealdetector behavior. The entire described non-ideal behavior canbe categorized in systematic and random effects. As randomnoise effects introduced by the signal source and the receiverstructure have little impact on short-range systems and areeasy to overcome, the primary concern in the presented CW(Six-Port) radar system is the minimization of the systematicmeasurement error [45], which results from limited isolation(crosstalk) between the transmit (Tx) and the receive (Rx) pathand detector non-linearity.

A. Influence of impairments

Fig. 3 showed the complete frontend of the radar systemwith the passive six-port junction. In this section non-idealitiesof the proposed RF frontend will be identified and theirinfluence on the baseband signals will be evaluated by ananalytic derivation of the six-port equations with a crosstalksignal. The reasons for limited isolation between the transmitand the receive paths are manifold. Parasitic coupling of theoscillator signal to the receiver structure occurs due to limitedshielding, limited isolation of the radar coupler (Sleakage), an-tenna mismatch (Santenna), and reflections of static targets inthe antenna’s field-of-view (

∑SRx,par) [46], [47]. Therefore,

the sum signal SRx,∑ at the receiver’s input port is composedas follows:

SRx,∑ = SRx + Sleakage + Santenna +∑

SRx,par . (1)

Thus, the crosstalk signal SX can be expressed by the sum ofall parasitic signal parts:

SX = Sleakage + Santenna +∑

SRx,par . (2)

As the parasitic signals directly interferes with the receivesignal, the LNA and the reconfigurable attenuator must adjustthe power level of the reference signal S2 = A2ejω2t+φ2 andthe receive signal S1 = A1ejω1t+φ1 to optimize the basebandvoltages so that the dynamic range of the baseband circuitrycan be efficiently used.

To analyze the influence of parasitic signals in the re-ceive path and to find the optimal power level settings, thesix-port equations [48] with an additional parasitic signalSX = AXejωXt+φX have to be calculated for the homodynearchitecture (ω1 = ω2 = ωX ):

B3/k3 = A21 +A2

2 + 2A1A2 sin(φ1 − φ2) (3)

+A2X + 2A2AX sin(φX − φ2) + 2A1AX cos(φ1 − φX),

B4/k4 = A21 +A2

2 − 2A1A2 sin(φ1 − φ2) (4)

+A2X − 2A2AX sin(φX − φ2) + 2A1AX cos(φ1 − φX),

B5/k5 = A21 +A2

2 + 2A1A2 cos(φ1 − φ2) (5)

+A2X + 2A2AX cos(φX − φ2) + 2A1AX cos(φ1 − φX),

B6/k6 = A21 +A2

2 − 2A1A2 cos(φ1 − φ2) (6)

+A2X − 2A2AX cos(φX − φ2) + 2A1AX cos(φ1 − φX),

where ki are conversion factors of the four power detectors.When analyzing the obtained baseband signals, each signal

again can be split into two parts: the wanted signal that isindependent from the crosstalk signal, and the unwanted signalthat depends on the crosstalk signal’s amplitude and phase.The unwanted signal itself consists of three components:

1) an offset equal to the square of the crosstalk signal’samplitude

2) an offset proportional to the amplitudes of the referenceand the crosstalk signal and dependent on the phasedifference between these two signals (note, that φ2 andφX are constant)

3) a sinusoidal signal that varies with the rotating phaseφ1 (note, that the phase of the receive signal is varyingwith the target’s movement)

For a moving target, the wanted signal and the third componentof the unwanted signal vary with the same rotating phaseφ1(t). Thus, the sum of both signals also varies with therotating phase φ1(t), but has a different phase offset andamplitude. For instance, the sum of the varying part ofbaseband voltage B3 can be calculated to be:

2A1A2 sin(φ1−φ2)+2A1AX cos(φ1−φX) = As sin(φ2+φs)(7)

with

As = 2A1

√A2

2 +A2X + 2A2AX sin(φX − φ2), (8)

φs = arccos

(A2 cos(φ2) +AX sin(φX)√

A22 +A2

X + 2A2AX sin(φX − φ2)

). (9)

This demonstrates, that a crosstalk signal results in threedifferent types of errors:

1) Offset: even if ideal detectors are supposed with k =k3 = k4 = k5 = k6, the offsets in (3) to (6) do notfully cancel out by calculating the IQ signals, but anoffset 4kA2AX sin(φX − φ2) and 4kA2AX cos(φX −φ2), respectively, is remaining.

2) Amplitude imbalance: the amplitudes of the basebandsignals do not only depend on the amplitude of the ref-erence and the received signal, but also on the crosstalksignal’s amplitude and the phase shift between referenceand crosstalk signal.

3) Phase imbalance: as to be obtained by (7) and (9) thereis also a phase imbalance that has to be considered. Thismeans that the four baseband signals are not separatedby 90 anymore.

These errors usually can be eliminated with an IQ imbalancecompensation [49]. However, due to the characteristics of theantenna, the amplitude A1 of the received signal is exponen-tially decaying with increasing distance to the target. Thus, thebaseband signals show, additionally to the above mentionederrors, a spiral characteristic with a decaying diameter overthe distance in the near field region [50].

7

B. Influence of the Six-Port receiver’s non-linearity

The basic concept of Six-Port technology in measurementapplications is mainly based on two requirements, whichare defined relative phase shifts between the detector inputsignals and a known transfer function of the used powerdetectors, providing square-law behavior. Deviations will af-fect the obtained baseband signals and, therefore, the valueof the complex signal. Consequently, receiver non-linearitycan be directly assigned to the non-linearity of the detectorcharacteristics. However, the amplitude and phase imbalances∆αc and ∆φc of the passive Six-Port junction, which canbe totally compensated for an ideal detector characteristic, in-crease the non-ideal behavior of the receiver. While amplitudeand phase imbalances ∆αc and ∆φc can be addressed by aproper design of the used couplers of the Six-Port junction,the influence of the four detector transfer functions is morecomplex by means of different reasons: Each detector diodeis subject of individual tolerances in placement as well as inthe semiconductor manufacturing process. Therefore, each ofthe four detector devices has an individual transfer function.Due to the necessary high dynamic range as well as offsetrelated power imbalance deviation from square law occurs byreasons of the diodes’ non-linearity [51]. The effect of non-linear detector characteristics on the IQ data was analyzed forconstant power ratios in [44] and a strong deformation of theellipse could be determined at high input powers. However,in the present application each detector runs at different inputpower levels.

−0.2 −0.1 0 0.1 0.2−0.2

−0.1

0

0.1

0.2

I

Q

(a) IQ (square-law detectors)

−0.2 0 0.2

−0.2

0

0.2

deformation

I

(b) IQ (non-linear detectors)

Fig. 12. Calculated baseband signals in the IQ representation using (a) square-law detector characteristic and (b) non-linear characteristic.

The impact of the described hardware deviations on theoffset compensated simulated IQ-data of the application isdisplayed in Fig. 12. The presented data are based on a detailedsystem simulation, considering the described effects of impair-ments, power imbalance, amplitude and phase imbalances, anddetector non-linearity. The simulation data are generated for anincreasing distance sweep of 5 mm range, starting at 22.5 mm.The antenna is simulated with an ideal pattern providing8 dBi. Multiple reflections between antenna and target arenot considered in the system simulation. Fig. 12(a) shows theobtained offset-compensated spiral for square law detectors,while Fig. 12(b) is based on a non-linear detector behavior. Itcan be obtained, that the second spiral shows deformations,which can be related to the changing input power level.

V. NF BACKEND

The backend PCB contains all components for voltage andcontrol signal generation, the analog baseband processing,as well as signal processing and communication. The boardconsists of six layers and its U-shape is optimized to align theantennas of the RF front-end modules directly opposite in adistance of 5 cm (1.97 inch) including a cut-out for the metalsheet, as depicted in Fig. 13b.

A. Analog signal conditioning and digitization

The analog baseband consists of an analog signal condi-tioning stage with digital adjustable gain and an 8-channelsimultaneous sampling ADC. At first a low-pass filteringand amplification of the susceptible detector output signals isperformed by THS4524 operational amplifiers from Texas In-struments (TI). Furthermore, a digitally adjustable potentiome-ter MCP41-series from Microchip is used to realize variablegain amplifiers with a run-time adjustable gain up to 255.Based on the design considerations from [52] an 8-channelsimultaneous sampling 24-bit ADC (ADS1278 from TI) hasbeen selected. Due to the Delta-Sigma (∆Σ) architecture thefilter requirements of the previous anti-aliasing stage can berelaxed while still maintaining a fast sampling rate of up to105 kSa/s that is more than sufficient even for fast movingtargets.

B. Microcontroller and communication

For controlling and monitoring of the system an InfineonXMC4500 microcontroller (MCU) is used, which can beaccessed by an 100-Mbps-ethernet-link to change the systemparameters, i.e., the VCO output frequencies, the output powerof the RF modules, and the baseband amplification, usingseveral SPI connections. Furthermore, the MCU reads the datafrom the ADC’s output registers and sends the raw data as wellas some monitoring signals, e.g., the temperatures of the RFMMICs, to a personal computer, which performs the furthersignal processing.

VI. CALIBRATION AND MEASUREMENT

Testing the sensor in a real roll milling application is bothcomplicated and not mandatory for initial testing and review ofthe proposed sensor. Therefore, the evaluation of the sensor’smeasurement principle is based on a simplified measurementsetup, which also can be undertaken under laboratory condi-tions.

A. Measurement Setup

The measurement setup mainly consists of the previouslydescribed hardware and an aluminum plate which is mountedon two orthogonal linear stages. A photograph of the completesetup is depicted in Fig. 13a. The schematic top view of thissetup is shown in Fig. 13b, illustrating the range directions (X-Range and Y-Range) and the distances between antennas andtarget (dA and dB). Here, the two front-ends are named FE-A and FE-B, respectively. The orthogonal positioned linearstages can be moved with an accuracy of about 500 nm

8

Ethernet NF Backend

FE-B

FE-A

Target

VoltageSupply

Range X

Range Y

(a) Photograph of the measurement system.

FE-A FE-B

dA dB

NF Backend

Target Range Y

Range X(b) Schematic top view of the measurement system.

Fig. 13. (a) Photograph and (b) schematic top view of the measurement setupwhich is used for calibration and test measurements. Two orthogonal mountedlinear stages with a stepped plate are utilized to simulate thickness changes.The two radar systems are placed on the back side of the U-shaped back-endPCB.

(≈ 0.2 mil). Both are controlled by a computer which alsocommunicates with the microcontroller. The stage moving inX-Range is named Stage-X in the following, while the otherone is termed as Stage-Y.

The aluminum plate is shaped with five steps, each havinga height of approximately 100 µm (≈ 3.937 mil) and a widthof about 5 cm (≈ 1.96 inch). The structured side of the plateis facing FE-A while the opposite surface is completely flat.The manufacturing quality of the plate corresponds to the pre-cision of the employed manufacturing computer numericallycontrolled milling process, which is 10 µm (≈ 0.4 mil) in alldirections.

B. Calibration

Moving the plate in X-Range ideally results in a linear phaseshift. Due to the non-idealities in the RF front-ends presentedin Sec.IV, the measured quantity values are impaired by aconstant offset and phase imbalances. The resulting complexsignal curve is an elliptic appearance of the before mentioned

spiral [50]. As a result of the high reflectivity of the target andthe short distance between the foil and the radio frequencyfront-end multiple reflections will occur. This effect resultsin a distance dependent offset vector which is superimposedto the receiving signal, therefore, resulting in an additionaldeformation of the IQ-data. To detect and correct these non-idealities several calibration methods have been published.Next to hardware based approaches with known calibrationstandards [53] also software based solutions are possible [54].While in [54] the individual six-port baseband signals getcalibrated, the calibration procedure for this measurementsetup is based on a phase error correction.

Before starting the calibration, the variable attenuators inthe reference path need to be initially settled to ensuremaximum usage of the ADC’s input range. As described insection IV, the optimum attenuation value depends not onlyon the distance between antenna and target, but also on thesignal strength of the receiving signal, hence, on the target’sreflectivity. Therefore, using the variable attenuator, the radarsystem can be adapted to the measurement scenario by eitheran evaluation of the receive power or direct evaluation of thebaseband signals during a distance sweep.

In the first step of the calibration routine Stage-X is movedwithin a range of ±2 mm (±78.7 e-3 inch) with a stepsize of20 µm (0.787 mil). The processing starts with acquiring thebaseband signals of FE-A and FE-B, which are depicted inFig.14(a) and (b). The four baseband signals of each front-end are referred to as Bη,i[k,n], with η ∈ A,B, i ∈ 3..6,k ∈ 1..K and n ∈ 1..N. N denotes the number ofcalibration points along the X-Range and K denotes thenumber of acquired samples per calibration point which areaveraged to improve the signal-to-noise ratio. The resultingbaseband signals can be calculated from Bη,i[k,n] and aredenoted by Bη,i[n]. Since the characteristics of the front-endsare slightly temperature dependent, which is mainly caused bythe detectors, a temperature compensation has to be applied[43]. The compensated baseband voltages are referred to byBtemp,η,i[n]. These voltages can be used to calculate thecomplex data IQη for each front-end η data with the followingformula:

IQη = (Bη,5 −Bη,6) + j(Bη,3 −Bη,4) . (10)

Since these values are impaired by offsets and phase imbal-ances, the center of the measured spiral is estimated by usingthe technique introduced in [50]. The offset-compensatedellipse, denoted as IQcal,η,i[n], is shown in Fig. 14(c) and (d).The remaining phase non-linearity is illustrated in Fig. 14 (e)and (f). To compensate these phase imbalances, the measuredphase values are mapped on the ideal phase values, which arederived from the positions of Stage-X by means of

φideal,η = dideal,η4π

λ, (11)

wherein dideal,η denotes the ideal distances between front-endantennas and the target and λ denotes the wavelength of theemitted electromagnetic wave. The ideal phase combined withthe measured phase can subsequently be used to correct thephase during actual measurements which are described in thefollowing subsection.

9

-2 -1 0 1 20

0.1

0.2

0.3

X-Range (mm)

AD

Cou

tput

volta

ge(V

)

(a) BB signals (FE-A)

B3 B4 B5 B6

-2 -1 0 1 20

0.5

1.0

1.5

2.0

X-Range (mm)

(b) BB signals (FE-B)

B3 B4 B5 B6

-0.02 -0.01 0 0.01 0.02-0.02

-0.01

0

0.01

0.02

I

Q

(c) IQ (FE-A)

-0.04-0.02 0 0.02 0.04

-0.04

-0.02

0

0.02

0.04

I

(d) IQ (FE-B)

-2 -1 0 1 2-6

-4

-2

0

2

4

6

X-Range (mm)

Pha

se(R

AD

)

(e) Unwrapped phase (FE-A)

idealmeasured

-2 -1 0 1 2-6

-4

-2

0

2

4

6

X-Range (mm)

(f) Unwrapped phase (FE-B)

idealmeasured

Fig. 14. Baseband signals (a) and (b), offset-compensated signal in IQ-plane(c) and (d) and unwrapped phase together with ideal phase (e) and (f).

C. Measurements

The complete signal processing flow of the actual measure-ments is shown in Fig. 15, in which the first four steps areequal to the calibration steps. The offset compensation of themeasurement signal is accomplished by subtracting the offsetvalue determined by the prior calibration routine. The resultingIQ signals are denoted as IQcomp,η[n]. The resulting phasenon-linearity due to the elliptic distortion is compensated bya phase error correction. In general, there are several ways torealize such a phase error correction. One possibility is a biglook-up-table with a lot of calibration points which implies ahuge calibration effort and requires a lot of memory space.Another approach is to significantly reduce the number ofcalibration points and use a segmental polynomial approxi-mation for interpolation [55]. In this measurement scenario acombination of a look-up-table and a spline-interpolation isutilized.

Fig. 14 (e) and (f) show that the measured phase valuesφmeas,η can be mapped on the ideal phase values φideal,ηalong the X-Range by a bijective function Φ. This funtion canbe otained at least point-wise from the calibration data while

Rawbaseband signal

Phase andamplitude

corr. IQ data

FilteredIQ data

Averagedbaseband signal

Offset comp.IQ data

Phasedata

Temperaturecompensated

baseband signalIQ data Relative

distance

Fig. 15. The signal processing flow from the raw ADC data to the relativedistance values.

-6 -4 -2 0 2 4 60.02

0.03

0.04

0.05

φcal (RAD)

|IQ

cal|

(a)

-4 -3 -2 -1 00.02

0.03

0.04

0.05

φcomp (RAD)

|IQ

com

p|

(b)

Fig. 16. (a) Magnitude of the calibration data and (b) magnitude of theoffset-compensated measurement data of FE-B.

continuity is realized by spline interpolation. The inversion ofthe function, denoted as Φ−1, can be employed to obtain thecorrected phase values φcorr,n.

φcal,η = Φ(φideal,η) (12)

φcorr,η = Φ−1(arg IQcomp,η) (13)

Comparing the baseband signals of both frontends, whichare depicted in Fig. 14(a),(b), the detectors are used in diverseoperating points which leads to differently distorted IQ signals.Since the magnitude distortion can be expressed as a functionof the phase it is possible to correct the magnitude errors with:

|IQcorr,η| =|IQcomp,η|

|IQcal,η (arg IQcomp,η)|(14)

IQcorr,η = |IQcorr,η| · ejφcorr,η . (15)

Fig. 16 illustrates the magnitudes of the calibration data aswell as the offset-compensated output signal both of FE-B.The amplitude distortions of the measurement data in sub-plot (b) can be reduced by incorporationg the knowledgefrom the calibration measurements depicted in sub-plot (a)using (14). Only phase information is needed to calculate therelative distances. The magnitude is linearized to get the validcomplex signal IQcorr,η . A low-pass filter is applied to reducethe influence of noise. To get a reasonable cut-off frequencyfor this filter, the spectrum of IQcorr,η has to be considered,which is determined by two factors. The first factor is the

10

surface of the plate, which is unknown a-priori, while thesecond factor is the antenna pattern. A wider antenna patternapplies a stronger low-pass effect on the measured data. Thisinfluence can be described quantitatively by convolving thecomplex surface reflection function with a convolution kernel,that incorporates the antenna pattern as well as reflectionsin the near-field. The convolution kernel is determined bysimulation and is denoted as θ. The low-pass effect can now bedescribed quantitatively by considering the Fourier transformof the convolution kernel. A Bessel filter is applied on thecorrected IQ-data due to its constant group delay. The cutofffrequency ωcutoff is chosen as

40 dB = |Fθ(0)|dB − |Fθ(ωcutoff )|dB . (16)

A Bessel filter with the cutoff frequency ωcutoff and a filterorder of 15 delivers an appropriate performance in terms ofnoise reduction combined with insignificant distortions of thephase data.

The filtered IQ data signal is used to calculate the phase databy means of φ = 6 IQ. Regarding (11) the relative distanceswhich are illustrated in Fig. 17 for the target moving 15 cm(5.9 inch) along the Y-Range can finally be computed.

The steps on the side of the aluminum plate facing FE-A can be recognized. Since the plate is not aligned perfectlyorthogonal to the antenna beams, linear distance changes aredetected along the Y-Range by both frontends. The relativethickness of the plate is calculated utilizing the calculatedrelative distances of both front-ends:

D = − (dA + dB) . (17)

The relative thickness of the aluminum plate along a Y-Rangeof 15 cm is depicted in Fig. 18. As a result of the differentialmeasurement principle, the linear distance changes are can-celed out, here. Furhermore, the figure depicts the expectedtarget surface with steps of a height of 110 µm and a size of5 cm on one side of the aluminum plate. The decreased steep-ness of the steps is caused by the low-pass effect mentionedabove, while the ripple on the obtained thickness curve can bedirectly related to the convolution of the antenna diagram andthe steps of the surface. In a sheet thickness application suchripple will not occur, as thickness changes are expected tobe smooth and frequency dependent reflection effects with theexcept of multiple reflections between metal sheet and antenna

0 0.05 0.10 0.150

0.4

0.8

1.2

Y-Range (m)

Dis

tanc

e(m

m)

(a) Stepped surface facing FE-A

0 0.05 0.10 0.15-1.5

-1.0

-0.5

0

Y-Range (m)

(b) Flat surface facing FE-B

Fig. 17. Measured distances of both frontends.

0 0.05 0.10 0.150

0.1

0.2

0.3

0.4

Y-Range (m)

Thic

knes

s(m

m)

Approximate true relative thicknessMeasured relative thickness

Fig. 18. Calculated relative thickness of the target-under-test plotted togetherwith the approximate true relative thickness.

are not present. Slow thickness drifts, however, will always beobtained by the measurement principle. For distance variationslimited to a quarter wavelength between two measurementvalues, the implemented algorithm is able to follow, even ifboth values are in different unambiguous ranges. However,such immediate changes will only occur during serious errorevents.

VII. COMPARISON WITH STATE-OF-THE-ART

There is a wide range of different measurement principlesbeing used in industrial applications ranging from ultrasonic toradiography. However, only a few of them meet the demandingrequirements of accurately measuring sheet thickness in thedescribed environment. Table I gives an overview of contact-less state-of-the-art metrology systems, which are in principlecapable of the sheet thickness measurement application, orsystems which are specially designed with focus on this in-tended purpose. Therefore, the schedule is divided in state-of-the-art commercial products for sheet thickness measurementand radar systems which have not been considered for beingused in this application, yet.

The existing difficulties of the application have been exten-sively described in Section I and II. Considering those, thesystems need to be evaluated by their measurement range,their precision, their update rate, and the spot diameter of thearea under observation. As being related to the wavelength,optical systems show best behavior in terms of precision,update rate as well as the size of the monitored spot. Opticalsystems also allow for an extraordinary accurate scanningof the surface profiles of fast moving sheets. However, theirusability is limited due to well known facts. First of all, theysuffer from environmental conditions, such as dust and fog aswell as changing illumination. Furthermore, they hardly canmonitor glowing surfaces. For those reasons, optical systemsare mainly applied in cold mills, where mechanical gaugesare also a good alternative. X-ray based systems can moreor less be used independent of the surrounding environment.However, for good results the strip’s material compound aswell as its temperature need to be known precisely, as theobtained thickness value is directly related to the materialabsorption coefficient. By nature, noise is also an importantissue for X-ray gauges [3] as it is influenced by severalparameters which are the air gap between transmitter anddetector, the source power and plate current of the tube, the

11

TABLE ITYPICAL VALUES OF STATE OF THE ART SENSOR TECHNOLOGY.

Principle Range Precision Update Rate Spot Diameter Thickness LimitOptical Interferometry Status Pro µLine F1 (1) 0-30000 mm not stated acc.: ±0.4 µm/m 100 MSa/s <1 mm none, differential meas.

Laser line triangulation MTS 8201 LLT (2) 40-100 mm ±1 µm not stated <1 mm none, differential meas.X-ray gauge VIS 2000 (3) 150-1000 mm not stated 1 kSa/s 3-20 mm 20-150 mm, material dep.

FMCW radar 24-26 GHz (4) 0.2-80 m ±1 mm 10 Sa/s * 8.4 DEG beam width ** none, differential meas.six-port 24 GHz (5) 0.01-2 m ±40 µm 20 Sa/s * 40 DEG beam width none, differential meas.

This work: six-port 61 GHz 0.005-0.5 m ±10 µm up to 105 kSa/s 50 DEG beam width 40 mm ***, differential meas.(1) Status Pro Maschinenmesstechnik GmbH; (2) Micro-Epsilon Group; (3) Friedrich Vollmer Feinmessgeraetebau GmbH;(4) Krohne Messtechnik GmbH, Level Meter Optiwave 7300C; (5) InnoSent GmbH, ISYS 5001, [56];* No measurement update rate stated, update rate limited by use of the serial interface ** Value from [57] *** Limited by mechanical design ofthe prototype, no general limit by means of the measurement principle

detector area and finally the material-under-test itself [5]. Inorder to avoid high statistical noise, the parameters of X-raygauges need to be adapted for each material change whichis a major drawback to the system. Finally, the lifetime ofthe tubes is strictly limited. Therefore, X-ray detectors requireextensive maintenance.

Radar based systems do not suffer from previously de-scribed drawbacks and therefore, represent an appropriatealternative. In comparison to optical systems, the wavelengthis larger thereby, also the diameter of the monitored spot. Anapproximation of the spot diameter dia3−dB is given by:

dia3−dB = d arctan(α3−dB) , (18)

with α3−dB being the antenna’s 3-dB beamwidth and dbeing the propagation distance. However, some part of thepower is radiated and received outside the 3-dB spot and theapplicability of (18) raises with higher antenna directivity.

FMCW systems are not suitable for the application dueto the high required bandwidth, which increases the rampduration and therefore, decreases the update rate. However, ahigh update rate is mandatory for the application. Furthermore,bandwidth is strictly limited by regulatory authorities (e.g.,ITU), especially as no encapsulated operation is provided.However, single frequency CW radar meets the requirementswith an unambiguous range limited to half its wavelength andits update rate only depends on the speed of the ADC, thesignal processing, and the speed of communication interface.While the 24 GHz six-port radar system presented in [56] doesnot show an acceptable update rate as well as lower precision,the proposed 61 GHz radar system is capable of update ratesof more than 100 kSa/s, while providing 10 µm precision.

By means of the differential measurement principle, thesystem is capable of compensating misalignment, while keep-ing scalability to varying sheet thickness. The measurementspresented in Section VI show a low-pass effect resulting fromthe steps in the measured target. The low-pass and convolutioneffect will not degrade suitability to the application as on theone hand smooth changes should be monitored and on theother hand rapid steps need to be detected immediately asthey are an indicator for a massive failure of the rolling mill.

VIII. CONCLUSION

A new concept for high-speed measurement of sheet-thickness in rolling mills based on two six-port interferometric

radar systems in a differential configuration has been pre-sented. The proposed system design is based on a detailedanalysis of the special demands resulting from the application.The radio frequency frontends are realized as system on sub-strate in a mixed SIW-MSL-technology to make the systemsinsensitive to external interference. SIW technology enablesan encapsulated frontend design allowing for an appropriateantenna design as well as enabling a good thermal design.

Furthermore, a detailed analysis of occuring non-ideal ef-fects of the radio-frequency frontend was provided and adedicated calibration strategy has been presented. The valida-tion of the proposed measurement system has been performedunder laboratory conditions delivering promising results forthe desired application.

A discussion and comparison with present state-of-the-artindicates, that the millimeter-wave interferometric measure-ment principle is a promising alternative to current technology.Especially the suitability to the demanding environmentalconditions and high update rates as well as the need forcomparably low maintenance are factors that underline thesignificant difference to currently used technology.

REFERENCES

[1] K. Aistleitner, L. Mattersdorfer, W. Haas, and A. Kugi, “Neural net-work for identification of roll eccentricity in rolling mills,” Journal ofmaterials processing technology, vol. 60, no. 1-4, pp. 387–392, June1996.

[2] A. Kugi, W. Haas, K. Schlacher, K. Aistleitner, H. M. Frank, and G. W.Rigler, “Active compensation of roll eccentricity in rolling mills,” IEEETrans. Ind. Appl., vol. 36, no. 2, pp. 625–632, Mar.-Apr. 2000.

[3] B. Vuksanovic and A. Bousbaine, “Intelligent control of aluminiumrolling mills using two dimensional adaptive filters,” InternationalJournal of Modeling and Optimization, vol. 3, no. 5, p. 412, Oct. 2013.

[4] W. Hessenberg and R. Sims, “Principles of continuous gauge control insheet and strip rolling,” Proceedings of The Institution of MechanicalEngineers, vol. 166, no. 1, pp. 75–90, June 1952.

[5] C. Clapp and R. Pohl, “X-ray thickness gauge for hot strip rolling mills,”Electrical Engineering, vol. 67, no. 5, pp. 441–444, May 1948.

[6] N. Neogi, D. K. Mohanta, and P. K. Dutta, “Review of vision-basedsteel surface inspection systems,” EURASIP Journal on Image and VideoProcess., vol. 2014, no. 1, p. 50, Nov. 2014.

[7] A. Koelpin, F. Lurz, S. Linz, S. Mann, C. Will, and S. Lindner, “Six-Port based interferometry for precise radar and sensing applications,”Sensors, vol. 16, no. 10, p. 1556, Sept. 2016. [Online]. Available:http://www.mdpi.com/1424-8220/16/10/1556

[8] T. Jaeschke, C. Bredendiek, S. Kuppers, and N. Pohl, “High-precisionD-band FMCW-radar sensor based on a wideband SiGe-transceiverMMIC,” IEEE Trans. Microw. Theory Techn., vol. 62, no. 12, pp. 3582–3597, Dec. 2014.

[9] Z.-M. Tsai, P.-H. Jau, N.-C. Kuo, J.-C. Kao, K.-Y. Lin, F.-R. Chang,E.-C. Yang, and H. Wang, “A high-range-accuracy and high-sensitivityharmonic radar using pulse pseudorandom code for bee searching,” IEEETrans. Microw. Theory Techn., vol. 61, no. 1, pp. 666–675, Jan. 2013.

12

[10] N. Pohl, T. Jaeschke, and M. Vogt, “An SiGe-chip-based 80 GHzFMCW-radar system with 25 GHz bandwidth for high resolution imag-ing,” in 2013 14th International Radar Symposium, vol. 1. IEEE, 2013,pp. 239–244.

[11] N. Pohl, T. Jaeschke, and K. Aufinger, “An ultra-wideband 80 GHzFMCW radar system using a SiGe bipolar transceiver chip stabilized bya fractional-N PLL synthesizer,” IEEE Trans. Microw. Theory Techn.,vol. 60, no. 3, pp. 757–765, Mar. 2012.

[12] S. O. Tatu, A. Serban, M. Helaoui, and A. Koelpin, “Multiport tech-nology: The new rise of an old concept,” IEEE Microw. Mag., vol. 15,no. 7, pp. S34–S44, Nov.-Dec. 2014.

[13] C. Li, V. M. Lubecke, O. Boric-Lubecke, and J. Lin, “A review on recentadvances in doppler radar sensors for noncontact healthcare monitoring,”IEEE Trans. Microw. Theory Techn., vol. 61, no. 5, pp. 2046–2060, May2013.

[14] G. Vinci, S. Lindner, F. Barbon, R. Weigel, and A. Koelpin, “Promiseof a better position,” IEEE Microw. Mag., vol. 13, no. 7, pp. S41–S49,Nov.-Dec. 2012.

[15] “ITU Radio Regulations, Edition of 2012,” http://www.itu.int/pub/R-REG-RR-2012, accessed: 2017-05-12.

[16] M. Bozzi, A. Georgiadis, and K. Wu, “Review of substrate-integratedwaveguide circuits and antennas,” IET Microw. Antennas Propag., vol. 5,no. 8, pp. 909–920, June 2011.

[17] B. Mayer and R. Knoechel, “Biasable balanced mixers and frequencydoublers using a new planar balun,” in 1990. 20th European MicrowaveConference, vol. 2. IEEE, 1990, pp. 1027–1032.

[18] R. Garg, I. Bahl, and M. Bozzi, Microstrip lines and slotlines. Boston-London: Artech House Microwave Library, 2013.

[19] H. J. Riblet, “The short-slot hybrid junction,” Proceedings of the IRE,vol. 40, no. 2, pp. 180–184, Feb. 1952.

[20] Z. Hao, W. Hong, J. Chen, H. Zhou, and K. Wu, “Single-layer substrateintegrated waveguide directional couplers,” IEE Proc.-Microw. AntennasPropag., vol. 153, no. 5, pp. 426–431, Oct. 2006.

[21] Z. Li and K. Wu, “24-GHz frequency-modulation continuous-wave radarfront-end system-on-substrate,” IEEE Trans. Microw. Theory Techn.,vol. 56, no. 2, pp. 278–285, Feb. 2008.

[22] K. Samanta, D. Stephens, and I. Robertson, “60 GHz multi-chip-module receiver with substrate integrated waveguide antenna and filter,”Electronics Letters, vol. 42, no. 12, pp. 701–702, June 2006.

[23] F. Xu, K. Wu, and X. Zhang, “Periodic leaky-wave antenna for millime-ter wave applications based on substrate integrated waveguide,” IEEETrans. Antennas Propag., vol. 58, no. 2, pp. 340–347, Feb. 2010.

[24] J. Liu, D. R. Jackson, and Y. Long, “Substrate integrated waveguide(SIW) leaky-wave antenna with transverse slots,” IEEE Trans. AntennasPropag., vol. 60, no. 1, pp. 20–29, Jan. 2012.

[25] K. S. Yngvesson, T. Korzeniowski, Y.-S. Kim, E. L. Kollberg, andJ. F. Johansson, “The tapered slot antenna-a new integrated elementfor millimeter-wave applications,” IEEE Trans. Microw. Theory Techn.,vol. 37, no. 2, pp. 365–374, Feb. 1989.

[26] K. Wu, Y. J. Cheng, T. Djerafi, and W. Hong, “Substrate-integratedmillimeter-wave and terahertz antenna technology,” Proc. IEEE, vol.100, no. 7, pp. 2219–2232, July 2012.

[27] S. Mann, F. Lurz, S. Linz, S. Lindner, C. Will, S. Wibbing, R. Weigel,and A. Koelpin, “Substrate integrated waveguide fed antenna for 61 GHzultra-short-range interferometric radar systems,” in 2016 IEEE TopicalConference on Wireless Sensors and Sensor Networks, 2016, pp. 64–66.

[28] S. Mann, S. Linz, S. Erhardt, S. Lindner, F. Lurz, H. Maune, R. Weigel,and A. Koelpin, “Differential measuring dual six-port concept andantenna design for an inline foil thickness sensor,” in 2016 IEEE GeMiC- The 10th German Microwave Conference. IEEE, 2016, pp. 165–168.

[29] Y.-S. Dai, D.-G. Fang, and Y.-X. Guo, “A novel UWB (0.045–50 GHz)digital/analog compatible mmic variable attenuator with low insertionphase shift and large dynamic range,” IEEE Microw. Compon. Lett.,vol. 17, no. 1, pp. 61–63, Jan. 2007.

[30] B.-W. Min and G. M. Rebeiz, “A 10–50-GHz CMOS distributed stepattenuator with low loss and low phase imbalance,” IEEE J. Solid-StateCircuits, vol. 42, no. 11, pp. 2547–2554, Nov. 2007.

[31] H. Kondoh, “DC-50 GHz MMIC variable attenuator with a 30 dB dy-namic range,” in 1988 IEEE MTT-S International Microwave SymposiumDigest. IEEE, 1988, pp. 499–502.

[32] S. Bulja and A. Grebennikov, “Variable reflection-type attenuators basedon varactor diodes,” IEEE Trans. Microw. Theory Techn., vol. 60, no. 12,pp. 3719–3727, Dec. 2012.

[33] S. Mann, F. Lurz, S. Lindner, F. Barbon, S. Linz, R. Weigel, andA. Koelpin, “61 GHz millimeter wave voltage variable attenuator basedon flip-chip mounted pin-diodes,” in 2014 20th International Conference

on Microwaves, Radar, and Wireless Communication. IEEE, 2014, pp.1–4.

[34] X.-P. Chen and K. Wu, “Substrate integrated waveguide filters: Designtechniques and structure innovations,” IEEE Microw. Mag., vol. 15,no. 6, pp. 121–133, Sept.-Oct. 2014.

[35] H. Peng, X. Xia, and T. Yang, “Slotted substrate integrated waveguidephase shifter,” in 2016 IEEE Information Technology, Networking,Electronic and Automation Control Conference. IEEE, 2016, pp. 1036–1039.

[36] Y. Ding and K. Wu, “Varactor-tuned substrate integrated waveguidephase shifter,” in 2011 IEEE MTT-S International Microwave SymposiumDigest. IEEE, 2011, pp. 1–4.

[37] X. Xu, R. G. Bosisio, and K. Wu, “A new six-port junction based onsubstrate integrated waveguide technology,” IEEE Trans. Microw. TheoryTechn., vol. 53, no. 7, pp. 2267–2273, July 2005.

[38] S. Germain, D. Deslandes, and K. Wu, “Development of substrate inte-grated waveguide power dividers,” in 2003. IEEE Canadian Conferenceon Electrical and Computer Engineering, vol. 3. IEEE, 2003, pp.1921–1924.

[39] Z. Hao, W. Hong, H. Li, H. Zhang, and K. Wu, “Multiway broadbandsubstrate integrated waveguide (siw) power divider,” in 2005 IEEEAntennas and Propagation Society International Symposium, vol. 1.IEEE, 2005, pp. 639–642.

[40] T. Djerafi and K. Wu, “Super-compact substrate integrated waveguidecruciform directional coupler,” IEEE Microw. Compon. Lett., vol. 17,no. 11, pp. 757–759, Nov. 2007.

[41] T. Djerafi, M. Daigle, H. Boutayeb, X. Zhang, and K. Wu, “Substrateintegrated waveguide six-port broadband front-end circuit for millimeter-wave radio and radar systems,” in 2009 European Microwave Confer-ence. IEEE, 2009, pp. 77–80.

[42] S. Mann, S. Erhardt, S. Lindner, F. Lurz, S. Linz, F. Barbon, R. Weigel,and A. Koelpin, “Diode detector design for 61 GHz substrate integratedwaveguide six-port radar systems,” in 2015 IEEE Topical Conferenceon Wireless Sensors and Sensor Networks. IEEE, 2015, pp. 44–46.

[43] C. Will, S. Mann, F. Michler, T. Reissland, F. Lurz, R. Weigel,and A. Koelpin, “Error compensation of the temperature influence onradar based displacement measurements,” in 2017 IEEE Asia PacificMicrowave Conference, November 2017.

[44] S. Linz, C. Will, F. Lurz, S. Lindner, S. Mann, R. Weigel, andA. Koelpin, “Detector nonlinearity in six-port radar,” in 2017 IEEETopical Meeting on Wireless Sensors and Sensor Networks. IEEE,2017, pp. 19–22.

[45] D. L. Richardson and T. E. Szmurlo, “Poly-frequency CW doppler radarsystem with leakage cancellation and method,” U.S. Patent 5,861,837,Jan. 19 1999.

[46] B.-K. Park, O. Boric-Lubecke, and V. M. Lubecke, “Arctangent demod-ulation with dc offset compensation in quadrature doppler radar receiversystems,” IEEE Trans. Microw. Theory Techn., vol. 55, no. 5, pp. 1073–1079, May 2007.

[47] A. Boaventura, J. Santos, A. Oliveira, and N. B. Carvalho, “Perfectisolation: Dealing with self-jamming in passive RFID systems,” IEEEMicrow. Mag., vol. 17, no. 11, pp. 20–39, Nov. 2016.

[48] C. A. Hoer, Using six-port and eight-port junctions to measure activeand passive circuit parameters. National Bureau of Standards (U.S.),Sep. 1975.

[49] M. Zakrzewski, A. Singh, E. Yavari, X. Gao, O. Boric-Lubecke,J. Vanhala, and K. Palovuori, “Quadrature imbalance compensation withellipse-fitting methods for microwave radar physiological sensing,” IEEETrans. Microw. Theory Techn., vol. 62, no. 6, pp. 1400–1408, June 2014.

[50] S. Linz, F. Lurz, M. Sporer, S. Lindner, S. Mann, R. Weigel, andA. Koelpin, “Ultra-short-range, precise displacement measurement setupwith a near field slot-line antenna and a dedicated spiral calibration,” in2015 IEEE MTT-S International Microwave Symposium, May 2015, pp.1–4.

[51] C. A. Hoer, K. C. Roe, and C. M. Allred, “Measuring and minimizingdiode detector nonlinearity,” IEEE Trans. Instrum. Meas., vol. 1001,no. 4, pp. 324–329, Dec. 1976.

[52] S. Lindner, F. Barbon, S. Linz, F. Lurz, S. Mann, R. Weigel, andA. Koelpin, “ADC depending limitations for six-port based distancemeasurement systems,” in 2015 IEEE Topical Conference on WirelessSensors and Sensor Networks, Jan 2015, pp. 29–31.

[53] S. Li and R. G. Bosisio, “Calibration of multiport reflectometers bymeans of four open/short circuits,” IEEE Trans. Microw. Theory Techn.,vol. 30, no. 7, pp. 1085–1090, Jul 1982.

[54] K. Staszek, S. Linz, F. Lurz, S. Mann, R. Weigel, and A. Koelpin,“Improved calibration procedure for Six-Port based precise displacement

13

measurements,” in 2016 IEEE Topical Conference on Wireless Sensorsand Sensor Networks, Jan 2016, pp. 60–63.

[55] C. Will, S. Linz, S. Mann, F. Lurz, S. Lindner, R. Weigel, andA. Koelpin, “Segmental polynomial approximation based phase errorcorrection for precise near field displacement measurements using Six-Port microwave interferometers,” in 2017 IEEE Topical Meeting onWireless Sensors and Sensor Networks, Jan 2017.

[56] S. Linz, G. Vinci, S. Mann, S. Lindner, F. Barbon, R. Weigel, andA. Koelpin, “A compact, versatile six-port radar module for industrialand medical applications,” Journal of Electrical and Computer Engi-neering, vol. 2013, p. 2, Jan. 2013.

[57] N. Pohl and M. Gerding, “A dielectric lens-based antenna concept forhigh-precision industrial radar measurements at 24 GHz,” in 2012 42ndEuropean Microwave Conference. IEEE, 2012, pp. 731–734.

Sebastian Mann (S’12) was born in Nuremberg,Germany, in 1987. He received the Diploma de-gree in Electrical, Electronics and CommunicationEngineering from the Friedrich-Alexander Univer-sity of Erlangen-Nuremberg, Erlangen, Germany, in2012. In 2012, he joined the Institute for Electron-ics Engineering, Friedrich-Alexander University ofErlangen-Nuremberg, as a Research Assistant, con-ducting his Ph.D.. His research interests are in thearea of microwave circuits and component design.At present, he investigates on high-precision 61-GHz

radar sensors for industrial applications based on Six-Port technology. Mr.Mann was a recipient of the 1st prize of the High Sensitivity Radar StudentDesign Competition at the 2014 IEEE International Microwave Symphosium(IMS) in Tampa, FL and winner of the First and Second Prize at the IEEE2014 Radio and Wireless Week Student Paper Competition, Newport Beach,Ca, in 2014.

Christoph Will was born in Mellrichstadt, Germany,in 1988. He received the M.Sc. degree (with distinc-tion) in information and communication technologyfrom the University of Erlangen-Nuremberg, Erlan-gen, Germany, in 2014. Since 2012, he has been withthe Institute for Electronics Engineering, Universityof Erlangen-Nuremberg, Erlangen, Germany, wherehe is currently working as a Research Assistanttoward the Ph.D. degree. His research topics arerelated to innovative radar technology based on Six-Port interferometry for industrial and biomedical ap-

plications. His main topics are digital signal processing, baseband optimizationand calibration, as well as researching algorithms for vital sign monitoring.Mr. Will is a member of the IEEE Microwave Theory and TechniquesSociety (IEEE MTT-S), the IEEE Engineering in Medicine and BiologySociety (EMBS), and the German Association for Electrical, Electronic andInformation Technologies (VDE). He received the Second Prize at the IEEE2017 Radio Wireless Week Student Paper Competition, Phoenix, AZ, in 2017.

Torsten Reissland was born in Erfurt, Germanyin 1990. He received the M.Sc. degree in Elec-trical, Electronics and Communication Engineeringfrom the Friedrich-Alexander University Erlangen-Nuremberg (FAU) in 2016. In 2016 he joined theInstitute for Electronics Engineering, FAU, as a Re-search Assistant. Primarily he is working on digitalsignal processing for radar applications as well asbaseband design.

Fabian Lurz (S’13) received the B.Sc. and M.Sc.degrees in information and communication tech-nology from the Friedrich-Alexander UniversityErlangen-Nuremberg (FAU), Erlangen, Germany, in2010 and 2013, respectively. In 2013, he joinedthe Institute for Electronics Engineering, FAU, asa Research Assistant and became group leader ofthe Electronic Systems Group in 2017. His cur-rent research interests include microwave circuitsand system, especially for low-cost and low-powermetrology applications. Mr. Lurz was a twice a

recipient of the 1st prize of the High Sensitivity Radar Student DesignCompetition at the IEEE International Microwave Symposium in 2014 and2017, respectivly. Furthermore, he received the IEEE Microwave Theory andTechniques Society (MTT-S) Graduate Fellowship Award in 2016.

Stefan Lindner (S’12) received his diploma inMechatronics in 2011 at the University of Erlangen-Nuremberg, Germany. Since 2012 he is with theInstitute for Electronics Engineering, University ofErlangen-Nuremberg, Germany, conducting his PhDstudies on research topics related to innovative radartechnology based on the six-port receiver. Primarilyhe is working with system simulation, baseband de-sign as well as digital signal processing. He receivedthe 1st Prize in Student Challenge at the EuropeanMicrowave Week 2012 in Amsterdam. In 2013 he

won the SEW-Eurodrive diploma prize for his diploma thesis.

Sarah Linz (S’13) received the B.Sc. and M.Sc.degree in Electrical, Electronics and Communi-cation Engineering from the Friedrich-AlexanderUniversity of Erlangen-Nuremberg, Germany, in2010 and 2012, respectively. In 2013, she joinedthe Institute for Electronics Engineering, Friedrich-Alexander University of Erlangen-Nuremberg, as aResearch Assistant. Her research interests includeRF and millimeter-wave circuits for high-precisiondisplacement sensors and six-port technology.

14

Robert Weigel (S’88–M’89–SM’95–F’02) was bornin Ebermannstadt, Germany, in 1956. He receivedthe Dr.-Ing. and the Dr.-Ing.habil. degrees, both inelectrical engineering and computer science, fromthe Munich University of Technology in Germanywhere he respectively was a Research Engineer, aSenior Research Engineer, and a Professor for RFCircuits and Systems until 1996. During 1994 to1995 he was a Guest Professor for SAW Technologyat Vienna University of Technology in Austria. From1996 to 2002, he has been Director of the Institute

for Communications and Information Engineering at the University of Linz,Austria where, in August 1999, he co-founded the company DICE, meanwhilesplit into an Infineon Technologies (DICE) and an Intel (DMCE) companywhich are devoted to the design of RFICs for mobile radio and MMICs forvehicular radar applications. In 2000, he has been appointed a Professor forRF Engineering at the Tongji University in Shanghai, China. Since 2002 he isHead of the Institute for Electronics Engineering at the University of Erlangen-Nuremberg, Germany. There, respectively in 2009, in 2012, and in 2015 heco-founded the companies eesy-id, eesy-ic and eesy-innovation.

Dr. Weigel has been engaged in research and development of microwavetheory and techniques, electronic circuits and systems, and communication andsensing systems. In these fields, he has published more than 900 papers. Hereceived the 2002 VDE ITG-Award, the 2007 IEEE Microwave ApplicationsAward and the 2016 IEEE MTT-S Distinguished Educator Award.

Dr. Weigel is a Fellow of the IEEE, an Elected Member of the GermanNational Academy of Science and Engineering (acatech), and an ElectedMember of the Senate of the German Research Foundation (DFG). He isand has been serving on numerous advisory boards of government bodies,research institutes and companies in Europe and Asia. Furthermore, he is andhas been serving on various editorial boards such as that of the Proceedingsof the IEEE. He also has been founding editor of the Proceedings of theEuropean Microwave Association (EuMA) which in 2008 where transformedinto EuMAs International Journal of Microwave and Wireless Technologies.He has been member of numerous conference steering and technical programcommittees. He was General Chair of several conferences such as the 2013European Microwave Week in Nuremberg, Germany and Technical ProgramChair of several conferences such as the 2002 IEEE International UltrasonicsSymposium in Munich, Germany. He served in many roles for the IEEE MTTand UFFC Societies. He has been Founding Chair of the Austrian COM/MTTJoint Chapter, Region 8 MTT-S Coordinator, Distinguished Microwave Lec-turer, MTT-S AdCom Member, and the 2014 MTT-S President.

Alexander Koelpin (S’07–M’11–SM’16) receivedthe Diploma in electrical engineering, the Ph.D.degree, and the ”venia legendi” from the Friedrich-Alexander University Erlangen-Nuremberg (FAU),Erlangen, Germany, in 2005, 2010, and 2014, re-spectively. From 2005 to 2017 he has been with theInstitute for Electronics Engineering, FAU, where hewas a Team Leader from 2007 to 2010; a GroupLeader for circuits, systems, and hardware test from2010 to 2015; and a Leader of the Electronic Sys-tems Group from 2015 to 2017. In 2017 he became a

full professor and head of the Chair for Electronics and Sensor Systems at theBrandenburg University of Technology, Cottbus, Germany. He has authoredor co-authored more than 120 publications. His current research interestsinclude microwave circuits and systems, wireless communication systems,local positioning, and six-port technology. Dr. Koelpin serves as a reviewerfor several journals and conferences. He was a recipient of the IEEE MTT-SOutstanding Young Engineer Award in 2016. He is the Co-Chair of the IEEEMTT-S Technical Committee MTT-16, the Commission A: ElectromagneticMetrology of U.R.S.I., and has been the Conference Co-Chair of the IEEETopical Conference on Wireless Sensors and Sensor Networks since 2012.