detec cancer

7/31/2019 Detec Cancer

1/14

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS 1

Integrated SFCW Transceivers for UWBBreast Cancer Imaging: Architectures and

Circuit ConstraintsMatteo Bassi, Graduate Student Member, IEEE, Andrea Bevilacqua, Member, IEEE,

Andrea Gerosa, Senior Member, IEEE, and Andrea Neviani, Member, IEEE

AbstractWe present a behavioral analysis of two differenttransceiver architectures for UWB breast cancer imaging em-ploying a SFCW radar system. A mathematical model of the directconversion and super heterodyne architectures together with anumerical breast phantom are developed. FDTD simulations dataare used on the behavioral model to investigate the limits of botharchitectures from a circuit-level point of view. Insight is giveninto I/Q phase inaccuracies and their impact on the quality of

the final reconstructed images. The result is that the simplicity ofthe direct conversion makes the receiver more robust towards thecritical circuit impairments for this application, that is the randomphase mismatches between the TX and RX local oscillators.

Index TermsBreast cancer detection, CMOS, direct conver-sion, super heterodyne, UWB.

I. INTRODUCTION

BREAST cancer is by far the most incident tumor amongfemale population [1], as shown in Fig. 1. Early time pre-

vention is a key factor in delivering long term survival to breast

cancer patients; 95% cure rates are possible if it is detected inits early stages [1].X-ray mammography remains the most effective technique

to detect non palpable breast tumors. However, ionizing radia-tions together with breast compression do not lead to comfort inpatient treatment. Moreover, 1030% of tumors are missed bymammography [2]. The significant number of false negativescan be attributed to the presence of a dense glandular tissuearound the tumor, absence of microcalcifications in the earlystages and tumors located close to the chest wall or underarm.

In this context, ultrawideband (UWB) microwave radar tech-nology is an attractive alternative [3][9]. The general approachis to illuminate the breast with a UWB pulse from a number of

antenna locations. The backscattered waves are collected andpostprocessed to obtain a high resolution dielectric map of thebreast. This approach relies on the contrast between the dielec-tric properties of normal and malignant tissues at microwave fre-quencies. As shown in [10], the intrinsic contrast is estimated to

Manuscript received March 18, 2011; revised July 12, 2011 and August 30,2011; accepted September 09, 2011. This work was supported by the Universityof Padova under Grant CPDA080777. This paper was recommended by Asso-ciate Editor P.-I. Mak.

The authors are with Department of Information Engineering, Univer-sity of Padova, I-35131 Padova, Italy (e-mail: [email protected];[email protected]; [email protected]; [email protected])

Digital Object Identifier 10.1109/TCSI.2011.2173400

Fig. 1. Age-adjusted tumor incidence rates per 100000 people grouped bycancer site [1].

vary from a minimum of 2:1 to a maximum of 10:1, whereas forX-ray it is only a few percent.

Several works on microwave imaging reported over the pastyears show the feasibility of this technique [3][9], [11][13].So far, RF measurement equipment has been employed to syn-thetically generate the UWB pulse by transmitting a set of nar-rowband stepped frequency waveforms over a wide range of fre-quencies and by collecting the backscattered signals, eventuallyretrieving time-domain pulses by means of the inverse Fouriertransform [14]. Even if this approach can be considered feasible,the employed equipment is extremely expensive and bulky. For

this reason, it cannot be considered a realistic and suitable so-lution for mass screening programs. Based on these considera-tions, we investigate the design of a low cost CMOS integratedcircuit (IC) that can be connected to an array of antennas to per-form monostatic or bistatic measurements. Each IC acts as atransceiver, generating stepped frequency waveforms and col-lecting back the scattered signals from the breast. In this work,we provide insight into the choice of the possible transceiver ar-chitecture through the study of an accurate behavioral model.The performance of the system is evaluated for different valuesof the circuit level impairments. Transmission power, receiverconversion gain, , 1 dB compression point, noise figure andbaseband filter bandwidth, as well as the ADC resolution are di-

rectly set, while the amount of the I/Q phase imbalance, as well

1549-8328/$26.00 2011 IEEE
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


2/14


2 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMSI: REGULAR PAPERS

as the phase noise are modeled as stochastic processes. The im-pact of all the relevant circuit-level impairments is then care-fully assessed with respect to the signal-to-clutter ratio (SCR)of the final reconstructed breast image. Specifications for theintegrated circuit are derived based on the application require-ments.

The paper is organized as follows. In Section II, the steppedfrequency continuous wave radar approach and the investigatedtransceiver architectures are presented. Section III describesthe breast phantom numerical setup and the receiver behav-ioral model, along with the image-processing algorithm. InSection IV, the impact of the relevant circuit-level impairmentsis explained. The super heterodyne and direct conversionarchitectures are compared in Section V. Conclusions aresummarized in Section VI.

II. SYSTEM DESCRIPTION

In principle, the UWB imaging technique we consider isbased on the radar operation [15], [16]. A low power short pulse

is irradiated into the breast. The pulse bounces back in presenceof objects with different electro-magnetic (EM) properties withrespect to the surrounding breast tissue. From the waveformand the time of flight of the back scattered pulses it is possibleto derive information on the reflecting objects, such that theirdistance and size. Furthermore, by combining observationsfrom different antennas in an array, it is possible to derive a2-D or even 3-D image of the volume that has been illuminated[4], [8].

The system we investigate seeks to identify the presence andlocation of neoplastic tissues that, due to their EM properties,behave as significant microwave scatterers in the breast. Con-tinuous waves stepped in frequency rather than a single pulse

are transmitted. This corresponds to the so-called Stepped Fre-quency Continuous Wave (SFCW) radar approach, where a setof narrowband measurements is performed over a wide rangeof frequencies. High resolution is obtained by performing theinverse Fourier transform (IFFT) on the collected results, thusretrieving time-domain waveforms [17]. Such a system allowsthe choice of the desired frequency range, topological parame-ters such as the position and number of antennas, as well as ofthe transmitted power.

The frequency range from to spanned by a SFCWradar is directly related to the achievable resolution in the slant-range , i.e. in the direction of wave propagation, as

(1)

where is the wave velocity in the medium andis the synthetic bandwidth of the system. Note that since

for every frequency step narrowband sinusoidal tones are pro-cessed, the system does not need to feature a large instanta-neous bandwidth, while still achieving high resolution. While onthe one hand higher resolution is obtained by increasing , onthe other hand the attenuation displayed by the body tissues in-creases with , impairing the capability of detecting a deepertarget. Consequently, penetration depth and spatial resolutionneed to be properly balanced by the choice of the operating fre-quencies.

The step in the frequency sweep determines the max-imum unambiguous range , i.e. the maximum distance a

target can be in order to be reliably detected without spatialaliasing, as

(2)

The choice of step results in a maximum un-

ambiguous range of 0.5 m, more than enough for the intendedapplication, avoiding any issue with spatial aliasing.

Topological parameters, such as the number of antennas andthe antenna-to-antenna distance directly influence the resolutionin the cross-range, i.e. on the plane orthogonal to the directionof wave propagation. By placing equally spaced antennasnext to each other, an equivalent synthetic aperture antenna oflength is generated, yielding, in the case of UWB waveforms,a resolution in the cross-range dimension , of[18], [19]

(3)

where is the target range. The depth of a typical normal, non-lactating human breast is in the order of 4 cm [20]. This suggeststhat a mildly-compressed breast would span less than 4 cm be-tween the skin surface and the rib cage. Further, almost 50% ofall breast tumors occur in the quadrant near the armpit where thebreast is less than about 2.5 cm deep [21]. Accordingly, we havebased our system on detecting tumors to depths of about 4 cmwith a typical depth of 3 cm. The survival of patients with inva-sive carcinoma is a inverse function of tumor size, independentof the method of detection [22], [23]. Smallest tumors are in theorder of 910 mm, thus a 4 mm-diameter tumor is a significanttest case as reported in [22]. Moreover, the typical length of thesurface of a nonlacting breast is about 20 cm. This limits the

maximum length of the synthetic aperture antenna. Assumingthe wave velocity within the breast is about one-third the speedof light in vacuum, to resolve the tumor with adequate resolution(3 mm to have some margin) a bandwidth of 16 GHz is requiredas given by (1). Such a bandwidth results in a cross-range res-olution up to a range of 10 cm, see (3), whichis more than adequate for the intended application. Taking 2GHz as the minimum frequency of operation to ease some im-plementation issues (antenna operation, ac-coupling, bandpassresponse of the transceiver, etc.) the frequency range the systemhas to cover is from 2 to 18 GHz.

The system block diagram of the investigated UWB steppedfrequency transceiver architectures is shown in Fig. 2 [24]. The

direct conversion is the more compact and straightforward so-lution. A PLL generates the local oscillator (LO) frequencies inthe desired range and feeds them into the mixer and the poweramplifier (PA). The LO signal is irradiated through the antenna,which is supposed to be available to work in both transmit/re-ceive modes connected to a TX/RX switch, circulator, or direc-tional coupler. An approach with different TX and RX antennasis also possible, both in monostatic and multistatic form. Thedetails about the physical implementation of the antenna are be-yond the scope of this work; they are discussed in the literature,e.g. in [7], [25], [26]. For simplicity, isotropic antennas were as-sumed. However, the analysis still holds if the antenna has con-stant gain over a suitable beamwidth to cover the breast surface.

In case different TX/RX antennas are employed, sufficient iso-lation between the two is required, as discussed in Section III-B.
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


3/14


BASSI et al.: INTEGRATED SFCW TRANSCEIVERS FOR UWB BREAST CANCER IMAGING 3

Fig. 2. Proposed UWB transceiver architectures. 2(a) direct conversion. 2(b)super heterodyne.

The backscattered signal from the breast is collected by the an-tenna, amplified by the LNA and downconverted by the quadra-ture mixer. The baseband signal is low-pass filtered and digi-tized before signal processing (IFFT, imaging) is performed.

As previously discussed, the detection of the cancerous tis-sues is based on the difference in their EM properties comparedto the healthy breast tissue. However, the breast skin also showsdifferent EM parameters, behaving as an undesired scatterer thatproduces a large unwanted signal. Signal-wise, this correspondsto a large interferer that cannot be easily separated from the de-sired signal neither in the frequency, nor in the time domain [4],[5]. Together with the request of high resolution in the ADCs,

the presence of the large in-band interferer results in a pecu-liar consequence for the system. Along with the typical circuitimpairments of the direct conversion receivers (flicker noise,second-order nonlinearities, dc offsets) there is an image issuerelated to I/Q mismatches: the image is indeed the signal re-ceived in the desired band, but the latter is dominated by thespurious skin backscatter. As such I/Q imbalances are critical.

The issues with I/Q mismatches may be alleviated by a superheterodyne scheme. In this case, the image signals are out-of-band. Given the intended application and the widely availablespectral monitoring techniques for the medical screening envi-ronment, it is reasonable to assume that the breast imaging isperformed in a controlled environment such that out-of-band in-

terferers are nearly absent, thus relaxing the requirements on theI/Q mismatches. On the other hand, the super heterodyne archi-tecture adds complexity to the receiver scheme (see Fig. 2(b)).A two-step quadrature downconversion is performed to a lowintermediate frequency (IF). To enable possible hardware reusebetween the transmitter and the first receiver LO, the first IF isassumed to be around 100 MHz, while the second IF is assumedto be in the 100 kHz range. Note that the high resolution require-ment for the ADCs forces to employ a low second IF. The mainissue with the frequency generation in the super heterodyne ar-chitecture is that all the employed LOs have to be coherent, thatis they have to display a fixed and well known phase relationshipwhen the baseband signal is sampled. Such a condition is, on the

other hand, guaranteed in the direct conversion transceiver, asthe same LO is shared between the transmitter and the receiver.

III. SIMULATION DECK AND SYSTEM MODELING

The aim of the work is to explore possible transceiver archi-tectures for breast cancer imaging and evaluate them in termsof circuit-level impairments impact on the imaging capabilitiesof the system. An accurate model of the system is thus nec-essary. The modeling effort we carried out is based on three

parts. First, the propagation of the waves in the breast tissueshas to be addressed. This has been done by setting up a numer-ical model of the various breast tissues (skin, normal tissue, ma-lignant tissue) and by performing finite-difference time-domainEM simulations on a numerical breast phantom. The result is aset of transfer functions that relate the transmitted tones and thereceived backscattered signals for different scenarios (with orwithout the presence of a tumor, different tumor positions, dif-ferent antenna configurations). Second, a baseband equivalentmodel of the receiver including all the circuit-level impairmentshas been developed, such that the signals at the receiver antenna,obtained by means of the transfer functions describing the prop-agation through the breast, can be related to the signals outputby the receiver. Third, a set of simulated receiver output sig-nals, assumed to come from different antennas in an array, areprocessed to obtain an image of the breast, such that the impactof the circuit nonidealities on the quality of the image can bereadily appreciated.

A. Numerical Breast Model

In a breast cancer imaging system, the patient orientationleads to two different system configurations. In the planar con-figuration, the patient lies in a supine position, and a planar arrayof antennas scans the naturally flattened breast [25]. Inthe cylin-

drical configuration, the patient is oriented in a prone positionwith the breast naturally extended in an examination hole. In thelatter, a cylindrical array is used to scan the breast [27]. As re-ported in [4], the two configurations show equivalent properties.Since the focus of this work is on the relativeperformance of dif-ferent integrated transceiver architectures, we consider a simplescenario based on a two-dimensional (2-D) configuration, wherea linear antenna array is supposed to be placed at the surface ofthe naturally flattened breast of a patient. The obtained imagewill of course show a cross-section of the breast. Despite itssimplicity, this configuration is realistic since the propagationof the waves in the breast, as well as the scattering due to theskin and the tumor, are well captured. Moreover, it can be re-

lated to more complex scenarios, assuming the antenna array isconformally placed along or around the breast. Fig. 3 shows the2-D breast model together with a possible configuration of theantenna array and tumor position. The breast is modeled as ahalf-space of breast tissue bounded by a thin layer of skin. Alinear array of antennas is located at a distance from thesurface of the breast, immersed in air. The antenna-to-antennadistance is .

The antenna array may be a physical one, or a synthetic one,meaning that it can be obtained by mechanical repositioningof one antenna. The array of transceivers can make use of asingle antenna (or even an antenna pair for TX and RX) thatis moved step-by-step in the correct positions by a mechanical

system while being connected to the right transceiver by a ma-trix of switches. Alternatively, a single transceiver and antenna
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


4/14



Fig. 3. 2-D model of the numerical breast phantom together with the antennaarray.

(or antenna pair) can be moved by a mechanical system to coverall the synthetic aperture length. According to [28], a significanttest case is given by a tumor with a diameter of 4 mm, which hasthus been selected as a target for the imaging. The thickness ofthe skin layer is set to 2 mm. The distance vector represents thedistance from any antenna to the tumor. The setup enables a flex-ible configuration of the number of antennas, the antenna-to-an-tenna distance and the tumor location with respect to the antennaarray.The antenna-to-skin ischosen to be1 cm. Thisis a re-alistic and feasible assumption for the breast screening [4]. Oncethe length of the antenna array is chosen, different number ofantennas result in different values of . The number of an-

tennas does not have an impact on the resolution (for a fixed ).However, a larger number of antennas increases the processinggain of the array while reducing the level of the sidelobes of thebeamforming system [19]. At the same time, isdesirable [19], especially if one is constrained by the allowabledimensions of the antenna array. All this taken into account, wechose 11 antennas over a length as a typical test case.The overall accuracy of the numerical breast model is dictatedby the approach used to capture the dispersive EM properties ofthe tissues. Each material in the model was assigned the appro-priate frequency-dependent dielectric properties (electrical con-ductance and relative dielectric constant ) in the range from2 to 18 GHz as documented in [10], a large scale study on the

UWB dielectric properties of normal, benignant and malignantbreast tissues from 0.5 to 20 GHz. The dispersive properties ofthe tissues are incorporated in the model by means of the es-tablished four-term one-pole Cole-Cole parametric dispersionmodel for complex permittivity:

(4)

In the model of the normal tissue, adipose-dominated tissue isconsidered over a glandular-dominated one, as suggested by thehistograms of distributions of tissues presented in [10]. The pa-rameters for the adipose-dominated normal tissue( ,

, , , )and the malignant tumor ( , ,

Fig. 4. Comparison between the Cole-Cole curves of the normal (solid) andmalignant (dashed) tissue used in the 2-D numerical breast model.

, , ) yield the complexpermittivity shown in Fig. 4. The data in Fig. 4 suggests thatthe contrast between normal adipose-dominated tissue and ma-lignant tissue is ranging from a minimum of 6:1 to a maximumof 8:1, depending on the frequency of interest. The differenceof the dielectric constant enables for waves at microwave fre-quencies to reconstruct the tumor image with respect to the adi-pose-dominated tissue around it.

B. Electro-Magnetic Simulations

Electro-magnetic simulations are performed on the numericalbreast phantom with the finite-difference time-domain (FDTD)

method using Meep, a freely available software package [29].A monostatic approach is followed, where each antenna inthe array sequentially transmits the signal and records thebackscatter from the breast model. The EM signal used inthe Meep FDTD simulations is a tone generated by a contin-uous-wave source applied to a isotropic cylindrical antennawith infinite length in the direction perpendicular the plane ofFig. 3. The frequency is varied as

(5)

By leveraging the linearity of the medium, the setup in Fig. 3is varied as antenna in the air only and setup with and without

tumor. This allows to discriminate between forward and re-verse waves as well as reflection due to the air-skin interfaceand that due to the presence of the tumor. Simulations are per-formed for different relative positions of the tumor and antenna,i.e. different . The electric field at the input of the antennais recorded and the results of the simulations are recast in theform of transfer functions of the paths antenna-skin-antenna

and antenna-tumor-antenna . Based on theset of the performed EM simulations, any configuration of theantenna array and tumor position can be reproduced. Once thenumber of antennas, the antenna-to-antenna distance and tumorpositions are set, and are associated to eachantenna based on the required distance . As an example, the

attenuation experienced by the transmitted signal through thetwo paths is shown in Fig. 5 for a set of ranging from 4
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


5/14



Fig. 5. Antenna-skin-antenna H ( ! ; ~ r ) and antenna-tumor-antennaH ( ! ; ~ r ) path attenuation for j ~r j ranging from 4 cm to 11 cm.

cm to 11 cm. Clearly, the radiation bouncing from the skin ex-periences little attenuation and dispersion, while this is not thecase for the tumor response. A first important result is thus thatthe skin reflection dominates the tumor response. Comparing

and one can see that the receiver has tobe able to process signals with a large difference in amplitude.The dynamic range may be in excess of 100 dB, depending onthe transmitted frequency, suggesting that the ADC resolutionhas to be in the range of 15 to 18 bits. Another result is thatany TX/RX interface that would be used must feature an iso-lation exceeding the attenuation experienced by the transmittedsignal in the antenna-skin-antenna path, which is less than 20

dB. Many commercial devices achieve such a performance, e.g.[30].

C. Receiver Behavioral Model

The developed behavioral model of the receiver is more easilydiscussed with reference to the super heterodyne architecture.For each frequency step of the measurement scan and for eachantenna, the transmitted tone

(6)

is irradiated through the antenna, where is the signal am-plitude and the transmitted frequency. The signalis scattered by both the skin and the tumor back to the receiver,where the superposition of the two effects results in the receivedsignal

(7)

The two-step quadrature downconversion shown in Fig. 2(b)can be modeled by complex multiplication of the receivedsignalby the complex signals representing the two LOs:

(8)

(9)

where and (or and ) take into account the phase inac-curacies of the in-phase and quadrature components. and

Fig. 6. Phase inaccuracies of quadrature LO signals: (a) common-mode phaseerror and (b) differential-mode phase error.

(or and ) can be decomposed in terms of common-modeand differential-mode components: and

(and similarly for and ). As shownin Fig. 6(a), the common-mode phase error does not impair theaccuracy of the quadrature, but it accounts for a rotation of the

LO signals with respect to . Therefore, it represents thephase error between the transmitted signal at the antenna andthe LO at the mixer port. Such an error may arise from any dif-ference in the two signal paths, or from the lack of precise phasecoherence among the LOs, in the case of the super heterodynearchitecture (see Fig. 2). The common-mode phase error alsoincludes the effect of the phase noise, as it is assumed that thequadrature LO signals are derived from the same source by fre-quency division or by using a multiphase oscillator.

On the other hand, the differential-mode phase error quanti-fies the quadrature error between the I and Q components of theLO, as depicted in Fig. 6(b), resulting in a limited image rejec-tion in the receiver. Both and (or and ) mayvary with the frequency of operation . Therefore, they aremodeled as random variables. The common-mode phase error

is made of a static part and the phase noise, where is the

round trip time delay of the path antenna-tumor-antenna. Thestatic components of the phase errors are modeled as gaussianrandom variables with variance and (or and

), respectively. The cumulative phase noise is mod-eled as a Gaussian stochastic process with a variance that canbe written as [17]

(10)

where is the LO phase noise at the frequency offset ,and is the bandwidth of the baseband filter. The phasenoise of the second LO is neglected as it operates at much lowerfrequencies.

As shown in details in the Appendix, the signal downcon-verted to the low-IF, and low-pass filtered, is written in a base-band equivalent fashion as

(11)

where , similarly to , is the round trip time delay asso-ciated to the skin, calculated as .
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


6/14



In the case of the direct conversion, we only have one LO, setat and the signal downcoverted to baseband becomes1

(12)

The nonlinearities of the receiver are modeled with a power se-ries approach

(13)

where is the linear conversion gain of the receiver, is the

coefficient of the second-order nonlinear term and it is related to, and , the coefficient of the third-order nonlinear term,

models the gain compression [31], [32]. Furthermore, the am-plitude of is hard limited to accommodate the operationbeyond the 1 dB compression point. Note that in the super het-erodyne case .

The thermal noise of the receiver is modeled as gaussian addi-tive noise, specified by the receiver noise figure, which is addedto . The operation of the ADC is taken into account byquantizing (plus noise) before the imaging processing isperformed:

(14)

where is the quantized signal, is the noise sample, andis the quantization operation.

D. Image Reconstruction Procedure

Once the array of antennas and the tumor position are es-tablished, the signal output by the receiver connected to eachantenna, , is calculated as described in Section III-C. Con-sequently, a set of signals carrying information in the fre-quency-domain is available. By taking the inverse Fourier trans-form (IFFT) of , time-domain waveforms are generated

(15)

where . A t this p oint, t he o btained t ime-domainwaveforms include both the skin and tumor backscatters. Theearly-time content is dominated by the reflection from the skin,whereas the late-time content contains the tumor backscatterand clutter signals, as shown in Figs. 7(a) and 7(b). Since theskin response has much greater amplitude than the tumor re-sponse, as shown in Fig. 5, we need to remove the early-timecontent without corrupting the late-time content. A calibrationsignal is generated for each antenna by averaging the time re-sponse of every other antenna. The signal is then subtracted

1As detailed in the Appendix, the second term in brackets in (12) is neglectedin (11) because it can be filtered out by means of a complex filter.

Fig. 7. The result of a successful removal of the early-time skin content for thecentral antenna of the array with a tumor 3 cm in depth. (a) the signal beforecalibration. (b) the signal after calibration.

from the corresponding antenna as in [33]. Since the skin re-sponse embedded in the early-time content is almost the samefor each antenna, the calibration removes the skin reflection andenhances the tumor response, as shown in Fig. 7(b). However,it is clear that this calibration procedure is effective as long as

each antenna has the same distance from the skin. If this is notthe case, more sophisticated algorithms can be employed to esti-mate the correct distance and efficiently remove the skin content[5].

At this point, time-domain waveforms containing only thetumor (plus clutter) response are available. The image creation isbased on a simple delay-multiply-and-sum algorithm [8]. Itcon-sists on calculating, for each pixel, its intensity. First, the roundtrip time from each antenna to the considered pixel is calculated.In this calculation, the wave velocities of the different coveredmediums (air, skin, healthy breast tissue) are taken into account.Then, the time-domain signals are time-shifted by an amountcorresponding to the calculated round-trip times. In this way,

the information on the considered pixel embedded in the varioustime-domain signals is aligned in time to the point .
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


7/14



Next, all the possible cross products between the time-shiftedwaveforms are evaluated, and summed together to obtain thesignal . Finally, the intensity of the pixel of coordi-nates is calculated as

(16)

where is approximately equal to the duration of the ex-pected tumor response.

IV. IMPACT OF THE IMPAIRMENTS

The developed behavioral model allows to evaluate theperformance of the system for different values of the circuitlevel impairments. Transmission power, receiver conversiongain, , 1 dB compression point, noise figure, basebandfilter bandwidth, ADC resolution, variance of the phase errors,as well as the variance of the cumulative phase noise are variedand their impact on the time domain signals processed by thereceivers is carefully investigated. We consider the scenariodepicted in Fig. 3, where the tumor is located 3 cm in depth,since it is the most important test case reported in the medicalliterature [20], [21]. Insights on the most critical circuit levelimpairments are given and specifications for the IC design arederived based on the application requirements.

A. Gain

Gain flatness specifies how much the receiver conversiongain varies over the specified frequency range. Variations in thegain can cause distortion on the waveforms processed by the

system. Despite the high resolution needed by this application,the system shows a good robustness toward ripples in the gain.As we will show in Section V, a peak-to-peak ripple level of 3dB does not impair the ability of the system to properly detectand enhance the tumor response while introducing a negligiblelevel of degradation on the image quality.

Process spread during the realization of the antennas and theintegrated circuits generate gain mismatches among differenttransceivers that can impair the ability of the system to correctlyenhance the tumor response. In fact, if the skin content is not cor-rectly estimated by the calibration algorithm, the residual spu-rious signal can severely deteriorate the tumor response. In thisscenario, complex algorithms can be employed to efficiently es-timate and remove the skin content [5]. Alternatively, a singletransceiver can be mechanically moved over all the syntheticaperture length of the antenna connected to a single antenna orto an antenna pair for TX and RX. Moreover, calibration canbe performed to characterize the gain of the single transceivers.The latter can limit the negative effects of ripples and gain mis-matches on the enhancing of the tumor response.

B. Noise

The impact of noise on the image quality is assessed by eval-uating the amplitude of the time domain signals pro-cessed by the receivers. Fig. 8 shows the IFFT amplitude con-

sidering 15 dBm transmitted power, 40 dB conversion gain,10 dB receiver noise figure, 100 kHz baseband filter bandwidth,

Fig. 8. IFFT amplitude of skin-and-target, target-only received signals withrespect to the noise floor at the output of the receiver. The plot refers to thecentral antenna of the setup in Fig. 3 with the tumor 3 cm in depth.

, 30 dBm 1 dB compression point, 18-bitADCs. Flicker noise, phase noise and phase inaccuracies havebeen neglected so far, hence direct conversion and super het-erodyne yield the same result. The results reported in Fig. 8show the IFFT amplitude of the received backscatterers fromthe skin and tumor together. The amplitude of the target-onlyprocessed signal lies 60 dB above the noise floor. This givesimportant insights into the circuit design of the integrated trans-ceiver. Despite the heavy attenuation experimented by the signalin the antenna-tumor-antenna path at higher frequencies (seeFig. 5), the receiver is robust towards thermal noise, even if the

noise figure is not very low, and even if the transmitted poweris moderate. This is due to the SFCW approach, that allows fornarrow baseband bandwidths, while preserving synthetic UWBperformance at radio frequencies, i.e. high resolution. More-over, the inherent processing gain of the inverse Fourier trans-form process allows for enhancing the target response with re-spect to the receiver noise.

The possibility of relaxing the noise figure specification isfavorable especially to the direct conversion approach. In thiscase, the downconverted spectrum extends to DC and the flickernoise gives a non negligible contribution to the overall systemnoise. Due to the SFCW approach, a sample of the flicker noiseis taken at each frequency step of the measurement, as ex-

plained in Section III-B. The folding process inherent in thesampling operation flattens the flicker PSD, resulting in a broad-band white noise contribution [34]. As such, the immediate ef-fect is to increase the white noise floor depicted in Fig. 8. Au-tozeroing techniques and chopper stabilization can be exploitedto mitigate the flicker noise effects. If autozeroing is employed,the low-frequency 1/f noise can be high-pass filtered and thusstrongly reduced at the cost of an increased noise floor due toaliasing of the white noise. Chopper stabilization upconverts thefrequency range of the input signal to the chopping frequency,where 1/f noise is lower than thermal noise, and then demod-ulates it back to baseband. Both these techniques allow to de-crease the flicker noise PSD to a nearly constant value that tends

to the input white noise, reducing its additional power contribu-tion to a negligible level [34].
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


8/14



Fig. 9. Normalized RMS error of the time domain signal p ( k T ; ~ r ) before andafter calibration as a function of the I I P relative to the same signal in absence

of second order distortion. The RMS error is normalized to half-LSB of a 18-bitADC with 2 V input range. The plot refers to the central antenna of the setup inFig. 3 with the tumor 3 cm in depth.

C. Nonlinearities

The nonlinear behavior of the receiver is modeled as de-scribed in (13) to take into account for the high dynamicrange required by the system. As previously explained inSection III-C, breast cancer imaging is realistically assumed tobe performed in a screened environment, where the interfererscontribution is negligible. Since the spurious reflection dueto the skin is much larger than the desired signal, it excitesthe non linearities the most. For this reason, in both super

heterodyne and direct conversion, the 1 dB compression pointis kept well above the maximum signal level due to the skinreflection. With a transmitted power of 15 dBm, the highestsignal level processed by the system is around 33 dBm andthe 1 dB compression point specification is set to 30 dBm. Ascounterintuitive as it may sound, the direct conversion receiverappears to be quite immune to second-order nonlinearities. Tothis regard, the RMS error between the time domain signal

processed by the receiver before and after calibrationand the same signals in absence of second order distortion (i.e.

) is simulated as a function of . The RMS error isnormalized to half-LSB of a 18-bit ADC with 2 V input range.

As shown in Fig. 9, the error after calibration is at least oneorder of magnitude less than the error before calibration and itscontribution is less than half-LSB for values of larger than10 dBm. That is, since most of the second-order distortion isdue to the skin reflection, it results common to all the receivedsignals at the antennas of the array and it is effectively removedby the calibration algorithm.

D. Phase Noise and I/Q Phase Mismatches

A coherent system, in which the phase of the transmittedsignal relative to the receiver LOs is well known, is necessary torecover the difference in phase of the received backscatter rel-ative to the transmitted signal and compute the range-delay of

the target. For this reason, the I/Q phase mismatches are criticalin receivers designed for this application. It is useful to get more

insight on their impact on the received time-domain waveforms.

The received signal consists on the sum of two terms:the first is related to the tumor response while the second one tothe skin spurious backscatterer. In fact, if no phase inaccuraciesare taken into account, (11) and (12) become

(17)

showing that the tumor response is dictated by the termwhile the skin by , that is at least 20 dB greater in magni-tude, as suggested by Fig. 5. The distortion introduced on the re-ceived signal by phase inaccuracies is weighted byand as well. The net result is that, from the receivedsignal point of view, the distortion introduced by the skinterm is much larger compared to the target one. Hence, throughthe action of the spurious skin backscatterer, even a small dis-tortion generated by I/Q phase inaccuracies and/or phase noisecouples into the signal processed by the receiver. Since the setof phase inaccuracies is different for each receiver connectedto the antennas, the skin content response is different as well.The calibration signal generated by averaging the contributionof each antenna results noisy and uncorrelated to the actualskin content, adversely affecting the ability of the system toenhance the tumor response during the calibration procedure.Lets now investigate the impact of common-mode, , anddifferential-mode, , mismatches as static errors (i.e. theirvalues are set and do not depend on frequency) on the receivedtime-domain waveforms after calibration.Tothis end,we consider a one-step downconversion and very high level ofphase inaccuracies, to better appreciate the results from a quali-tative point of view. The situation is depicted in Fig. 10. For the

differential-mode mismatches, the relative power of the secondterm in (12) increases with . While the true tumor re-sponse decreases, a ghost image of the tumor is folded atfrom the maximum unambiguous range of the radar, as shownin Fig. 10(a). However, the presence of a ghost tumor imageis not a problem since the information carried in the (very)late-time response is not useful for the localization of tumors innormal-dimension breasts. Additionally, the tumor signal phaseis hard-shifted by a term proportional to , yielding a con-stant error in the evaluation of the tumor position. This secondeffect is experienced in presence of common-mode mismatches,too. In this case, the phase is shifted by a term proportionalto that leads to a larger fixed error in the evaluation of

the tumor position, as shown in Fig. 10(b). It is clear that mis-matches due to a static errors are not critical for the tumor detec-tion, although leading to errors in the evaluation of the correcttumor position.

In general, due to the frequency dependent nature of devicemismatches and the way I/Q components are generated by in-tegrated frequency synthesizers, common-mode and differen-tial-mode phase mismatches are expected to be frequency de-pendent. To properly capture this effect, and aremodeled as gaussian variables with zero mean and variance

and , respectively. The tumor response undergoesa decrease in amplitude while the clutter level increases else-where, as illustrated in Fig. 11, where the value of the variance

of the common-mode phase mismatch is set large to underlinethis effect. If differential-mode instead of common-mode phase
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


9/14



Fig. 10. Amplitude of p ( k T ; ~ r ) after calibration in presence of phasemismatches modeled as static errors: (a) differential-mode mismatches, (b)common-mode mismatches. Plots refer to the central antenna of the setup inFig. 3 with the tumor 3 cm in depth.

mismatches were considered, waveforms similar to the one inFig. 11 would be obtained from the simulation. To better under-stand the strong impact of random errors even for small vari-ance values, the RMS error of the time domain signalafter calibration relative to the same signal in absence of any

mismatch was calculated as a function of the standard devia-tion and . Fig. 12 shows the RMS error normalizedto half-LSB of a 18-bit ADC with 2 V input range. Even forsmall values of (or ), the RMS error quickly reachesvalues comparable to half-LSB. Moreover, the error generatedby common-mode mismatches grows twice as fast as the onegiven by the differential-mode mismatches. Since the clutter isclose to the tumor response and comparable in magnitude, a se-rious reduction in the ability of the system to detect and locatethe tumor is experienced.

In addition to the random I/Q phase mismatches, the phasenoise has to be considered. As shown by (11) and (12), the ef-fect of the phase noise is the same of the common-mode phase

mismatches and the variance of the cumulative phase noise canbe calculated using (10). As such, the considerations made for

Fig. 11. Amplitude of p ( k T ; ~ r ) after calibration in presence of common-mode phase mismatch modeled as a gaussian variables with variance .The plot refers to the central antenna of the setup in Fig. 3 with the tumor 3 cm

in depth.

Fig. 12. Normalized RMS error of the time domain signal p ( k T ; ~ r ) after cal-ibration relative to the same signal in absence of any mismatch as a function ofthe standard deviation and : (a) differential-mode phase mismatch,(b) common-mode phase mismatch. The RMS error is normalized to half-LSBof a 18-bit ADC with 2 V input range. The plot refers to the central antenna ofthe setup in Fig. 3 with the tumor 3 cm in depth.

the common-mode phase mismatches directly apply to the phasenoise case as well.

To summarize, frequency-independent phase errors are notcritical for the tumor detection. They lead to errors in the evalu-ation of the tumor position but not in its size and detection. Theghost tumor image is not a problem since the information car-ried in the late-time response is not useful for the image recon-struction. On the other hand, frequency dependent phase mis-matches and phase noise are critical even for small variancevalues. They enhance the clutter level with respect to the tumorresponse, leading to a possible failure in the detection of thetumor. Therefore, particular attention has to be taken in the ICdesign, where the standard deviation of the common-mode and

differential-mode mismatches has to be kept well below 1.5 , assuggested by Fig. 12.
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


10/14



TABLE IRECEIVER SPECIFICATIONS

Fig. 13. Reconstructed tumor image with an ideal receiver. In this image, 11antenna spacing 2 cm from each other irradiate a tumor 3 cm in depth from theskin.

E. Receiver Specifications

Based on the impact of the most important circuit-level im-pairments, we delineate the specifications for the design of theintegrated transceiver, summarized in Table I. The receiver op-eration was seen to be robust towards the thermal noise, thus al-lowing to relax the noise figure requirement to 10 dB. The 1 dBcompression point has to be kept at a level of at least 30 dBmto accommodate for the large in-band interferer backscatteredfrom the skin. The is set to 20 dBm, to limit the contribu-tion of its RMS error to less than half-LSB, as shown in Fig. 9.Frequency dependent phase mismatches where shown to be crit-ical. As such, the standard deviation of the common-mode anddifferential-mode phase errors is limited to

, as suggested by Fig. 12. A value of

at a frequency offsettogether with a baseband filter bandwidth of 100 kHz yields astandard deviation of the cumulative phase noise offor the case of a tumor 4 cm in depth and the last antenna ofthe array, limiting the contribution of the phase noise to thecommon-mode phase mismatches.

The specifications summarized in Table I do not change if thetumor is located closer to the skin or deeper in the breast. Thenoise floor was shown to be well below the desired signal level,while the linearity requirements are set by the large skin reflec-tion and not the tumor location itself. The tight specificationsregarding the I/Q phase mismatches were derived based on theRMS error of the waves at the output of the receiver in presence

and absence of phase mismatches, that does not depend on thetumor location as well.

If the distance between the antenna array and the skin ischanged, the antenna-skin-antenna and antenna-tumor-antennatransfer functions in Fig. 5 will both experience an equal rigidtranslation. This can be compensated by adjusting the trans-mitted power when designing the overall system.

V. ARCHITECTURE COMPARISON

The direct conversion and super heterodyne receivers arecompared by generating the tumor images in the two caseswith the same set of circuit impairments. The quality of thegenerated image is quantified by means of the signal-to-clutterratio (SCR), defined as the average intensity of the image inthe area where the target is located, divided by the averageintensity of the image elsewhere

(18)

where is the area containing the target, and is the com-plementary area with respect to the region under consideration.The image of a 4 mm-diameter malignant tumor at 3 cm indepth from the skin obtained with an ideal receiver (no cir-cuit impairments) is shown in Fig. 13; the corresponding SCRis 21.5 dB. It shows the quality of the employed imaging al-gorithm, and it represents the benchmark to assess the impactof all the circuit impairments. In this case, the direct conver-sion and super heterodyne architecture yield the same result.The two architectures are then compared by generating the im-ages with the same set of impairments, based on the specifi-cations derived in Section IV-E and summarized in Table I.

Fig. 14(a) shows the result in the direct conversion case con-sidering 15 dBm transmitted power, 40 dB conversion gain, 3dB gain ripple, 10 dB receiver noise figure, 100 kHz basebandfilter bandwidth, , 30 dBm 1 dB compres-sion point, , 18-bitADCs, and a phase noise level of at

MHz, while Fig. 14(b) shows the super heterodynecase. The SCR reduces to 17.8 dB and 14.5 dB in the directconversion and super heterodyne cases, respectively. Note thatthe gain ripple decreases the SCR by only 0.4 dB. As shown inFig. 15, the system reliably detects the tumor even when it islocated deeper in the breast or closer to the skin. The SCR de-creases as the tumor is located farther from the antennas while

the difference between the SCR experienced by the direct con-version and super heterodyne architectures remains within 2 to3 dB. In Fig. 16, the tumor images were generated for differentvalues of the antenna-to-antenna distance over the samesynthetic aperture length . As the total number of antennas de-creases to 8 and 5, the SCR decreases as well, as expected fromthe discussion in Section III-A. The difference in performancebetween the two architectures does not change with the numberof antennas.

As discussed in Section IV-D, the I/Q mismatches are crit-ical to the operation of the receiver. In Fig. 17, the different im-pact of the common-mode and differential-mode impairmentson both the architectures is shown. The common-mode mis-

matches, i.e. the random phase offsets between the TX and RXLOs, are clearly the most critical source of degradation of the
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


11/14



Fig. 14. Reconstructed tumor image with the two architectures and the sameset of circuit impairments, summarized in Table I. (a) Direct conversion; (b)super heterodyne.

Fig. 15. Reconstructed tumor image with the direct conversion (top) and superheterodyne (bottom) architectures for the same set of impairments summarizedin Table I and different tumor positions. The SCR of the direct conversion is18.1 dB and 8.9 dB for the tumor placed closer to the skin and deeper in thebreast, respectively, while the SCR of the super heterodyne is 15.5 dB and 7.1dB for the tumor placed closer to the skin and deeper in the breast, respectively.

image quality. In the super heterodyne architecture, the signalis degraded during both the two downconversion steps while inthe direct conversion it undergoes a single step downconversion.To this regard, direct conversion inherently behaves better. On

the other hand, super heterodyne shows a good insensitivity toincrements of differential-mode mismatches. Since the devel-

Fig. 16. Reconstructed tumor image with the direct conversion (top) and superheterodyne (bottom) architectures for the same set of impairments summarizedin Table I with 8 (left column) and 5 (right column) antennas covering the samesynthetic aperture length l = 2 0 c m . The SCR of the direct conversion is 12.8dB and 9.2 dB for 8 and 5 antennas, respectively, while the SCR of the superheterodyne is 10.3 dB and 7.4 dB for 8 and 5 antennas, respectively.

oped super heterodyne model assumes interferers screening ispossible in the application environment, no spectral content isdownconverted from the image frequencies at and

to baseband, as explained in the Appendix. Con-sequently, (11), unlike (12), does not contain any additionalimage term. However, if spectral leakage was experimented (e.g.from the skin reflection), terms at , and

would grow and super heterodyne would ex-periment a SCR reduction similar to the direct conversion. InFig. 18, the SCR is reported as a function offor and for both

the receiver architectures. As previously explained, the two SCRcurves for andoverlap almost entirely in the super heterodyne case, showinga good insensitivity to increments of the differential-mode mis-matches. However, when both the actions of differential-modeand common-mode mismatches are considered, the direct con-version architecture seems to be preferable, thanks to its singlestep downconversion. On the other hand, if bigger devices areemployed in the second downconversion step, a lower level ofimpairments can be considered for the second LO of the superheterodyne architecture. In the case where half the impairmentsare supposed for the second LO (i.e. and

), the super heterodyne and the direct conver-

sion architectures show a comparable performance, as shown inFig. 18. By further decreasing and , the super het-erodyne architecture becomes progressively comparable to thedirect conversion, as predicted by (11) and (12). Nevertheless,the requirement of phase coherence among the LOs in the superheterodyne architecture implies a more complex implementa-tion making the direct conversion preferable. Due to the highresolution required by the application (i.e. large bandwidth), asmall random phase mismatch dramatically enhances the clutterlevel and decreases the tumor response amplitude. That is, in ahigh resolution phase-detection system, random phase offsetsbetween TX and RX LOs are extremely critical and clearly donot enable a correct recovery of the phase difference between ir-

radiated and backscattered signal, leading to errors in the eval-uation of the tumor location and detection. From this point of
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


12/14



view, direct conversion seems preferable and enjoys a muchsimpler architecture. In the super heterodyne architecture, I/Qmismatches requirements can be relaxed in the second down-conversion step by using bigger devices. However, both the LOshave to be coherent (i.e. their phase relative to the TX signalhas to be known), which does not lend itself to easy practicalsolutions. On the other hand, the same LO can be used to irra-diate the breast and downconvert the received backscatterer inthe direct conversion architecture. The reduction of the dynamicrange due to LO leakage, DC offset voltage and low frequency

noise may be critical in the direct conversion operation. Theamplitude of the induced errors has to be kept below half-LSBof the baseband ADC. Adding auto-calibration requires addi-tional hardware and software implementation, complicating theIC design. Auto-zero amplifiers achieve typical input offset of1 V with a temperature-related drift of 20 , satisfyingthe requirements of this application to deal with small signals.On the other hand, chopper stabilization technique upconverts

noise to the chopping frequency, resulting in a nearly con-

stant baseband noise [34]. Both the techniques can be appliedtogether to satisfy the application requirements. To address thewide tuning range specifications, a PLL can be used to generatethe frequencies from 9 to 18 GHz while synthesizing the twolower octaves by means frequency dividers [35]. A possible ef-ficient implementation of the dividers is as injection locked fre-quency dividers (ILFD) [36], [37]. The low phase noise level of

110 dBc/Hz at can be achieved by employinga dual-band VCO working in the band from 9 to 18 GHz [38],[39], or covering the band with multiple VCOs. The design ofa wideband LNA in the range from 2 to 18 GHz is challenging.A multi section reactive network can be designed to resonatethe input impedance over a wide bandwidth to achieve input

matching and good noise performances [40]. Alternatively, acommon-gate stage together with a noise-canceling architecturecan be used to decouple the noise figure from the matching re-quirement without the need of a global negative feedback[41].On the other hand, shunt-shunt resistive feedback can enablethe design of very compact broadband inductorless LNAs [42].Finally, careful design techniques have to be used to minimizethe coupling of the transmitted signal at the antenna to the LNAinput port while preserving the I/Q phase mismatches requiredlevel.

VI. CONCLUSIONS

Super heterodyne and direct conversion architectures for

a SFCW breast cancer detection radar system are compared.Both the architectures exhibit a comparable performance, asshown in Fig. 18. However, a marginally higher robustnessto phase inaccuracies makes the direct conversion performslightly better in terms of signal-to-clutter ratio, as reported inFig. 14(a) and 14(b). issues are shown to be solvable bysubtracting the calibration signal from each antenna and widelyknown solutions allow to properly deal with noise and DCoffset issues inherent the direct conversion architecture. Mostimportantly, super heterodyne complexity does not allow aneasy solution to the TX/RX coherence, critical requirement ina high-resolution phase-detection system. This complexity isalso likely to result in higher power consumption to achieve the

same set of specifications, further emphasizing a preference forthe direct conversion solution.

Fig. 17. Impact of the differential-mode and common-mode mismatches on theSCR for super heterodyne and direct conversion architectures.

Fig. 18. SCR for super heterodyne and direct conversion architectures for dif-ferent levels of I/Q phase imbalance.

APPENDIX

In this section, we address the problem of modeling the phasenoise and the phase inaccuracies of a 2-step quadrature down-conversion receiver, i.e. a complex Weaver receiver.

The input RF signal is (6) while the two complex LO sig-nals and are modeled as in (8) and (9), respec-tively, where, without lack of generality, and and

(or and ) represent the phase noise and/or the phase in-accuracies of the in-phase and quadrature components, respec-tively. By decomposing and (or and ) in terms ofcommon-mode and differential-mode components, as explainedin Section III-C, (8) and (9) can be recast as

(19)
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


13/14



(20)

The downconversion can be modeled as a multiplication ofby the complex local oscillator signals and

. We assume now that , and, without loss ofgenerality, that and , such that the

desired intermediate frequency is and. The IF signal can be written as

(21)

where and arethe sum and difference of the common-mode phase error while

and correspondto the frequency at which the images at frequency and

are downconverted, respectively. Further, low-passfiltering is performed on the output (second) IF signal with a

pass-band lower than the first IF, i.e. . In thiscase, the IF signal is written as

(22)

At this point it would still be possible to get rid of the secondterm in (22) (image signal) by means of a complex filter,yielding, in this case,

(23)In the direct conversion case, , , and

. Assuming low-pass filtering at base-band, the IF signal iswritten as

(24)

Since the transmitted signal is scattered by both the tumor andthe skin back to the receiver, the superposition of effects is usedto obtain (11) and (12) from (23) and (24), respectively. The

term at in (23) is dropped in (11) because of the equivalentbaseband modeling.

Lets consider now the impact of the phase noise on the re-ceived signal, assuming the receiver is used in a SFCW radarsystem. In such a system, the TX and RX carrier must be co-herent. Therefore, the error due to the phase noise is the differ-ence between the instantaneous phase noise at the receiver andthe sample transmitted seconds before

(25)

where is the round-trip delay associated to the path an-tenna-tumor-antenna and the phase noise. Since the I/Qcomponents are generated from the same oscillator, the phasenoise appears as a common-mode phase error. As such,is added t o and , a s described i n Section III-C. Further,we assume that is coherent with the TX (RF) carrier andthat the phase noise of is negligible.

REFERENCES

[1] Surveillance, epidemiology, and end results (SEER) program researchdata(19732008), National Cancer Institute, DCCPS, Surveillance Re-search Program, Cancer Statistics Branch, released April 2011 basedon the November 2010 submission [Online]. Available: http://www.seer.cancer.gov

[2] P. T. Huynh, A. M. Jarolimek, and S. Daye, The false-negative mam-mogram, Radiograph, vol. 18, no. 5, pp. 11371154, 1998.

[3] E. Fear, P. Meaney, and M. Stuchly, Microwaves for breast cancerdetection, IEEE Potentials, vol. 22, no. 1, pp. 1218, 2003.

[4] E. Fear, X. Li, S. Hagness, and M. Stuchly, Confocal microwaveimaging for breast cancer detection: Localization of tumors in threedimensions, IEEE Trans. Biomed. Eng., vol. 49, no. 8, pp. 812822,2002.

[5] E. Bond, X. Li, S. Hagness, and B. Van Veen, Microwave imaging viaspace-time beamforming for early detection of breast cancer, IEEETrans. Antennas Propag., vol. 51, no. 8, pp. 16901705, 2003.

[6] N. Irishina, M. Moscoso, and O. Dorn, Microwave imaging for earlybreast cancer detection using a shape-based strategy, IEEE Trans.

Biomed. Eng. , vol. 56, no. 4, pp. 11431153, 2009.[7] D. Winters, J. Shea, P. Kosmas, B. Van Veen, and S. Hagness,

Three-dimensional microwave breast imaging: Dispersive dielectricproperties estimation using patient-specific basis functions, IEEETrans. Med. Imag., vol. 28, no. 7, pp. 969981, 2009.

[8] H. B. Lim, N. T. T. Nhung, E.-P. Li, and N. D. Thang, Confocal mi-crowave imaging for breast cancer detection: Delay-multiply-and-sumimage reconstruction algorithm, IEEE Trans. Biomed. Eng., vol. 55,no. 6, pp. 16971704, 2008.

[9] S. Davis, B. Van Veen, S. Hagness, and F. Kelcz, Breast tumorcharacterization based on ultrawideband microwave backscatter,

IEEE Trans. Biomed. Eng., vol. 55, no. 1, pp. 237246, 2008.[10] M. Lazebnik, D. Popovic, L. McCartney, C. B. Watkins, M. J. Lind-

strom, J. Harter, S. Sewall, T. Ogilvie, A. Magliocco, T. M. Breslin, W.Temple, D. Mew, J. H. Booske, M. Okoniewski, and S. C. Hagness, Alarge-scale study of the ultrawideband microwave dielectric propertiesof normal, benign and malignant breast tissues obtained from cancer

surgeries, Phys. Med. Biol., vol. 52, no. 20, p. 6093, 2007.[11] M. Klemm, I. Craddock, J. Leendertz, A. Preece, and R. Benjamin,Experimental and clinical results of breast cancer detection usingUWB microwave radar, in Proc. IEEE Antennas Propag. Soc. Int.Symp. (AP-S 2008), pp. 14.

[12] P. Meaney, M. Fanning, T. Zhou, A. Golnabi, S. Geimer, and K.Paulsen, Clinical microwave breast imaging2D results and theevolution to 3D, in Proc. Int. Conf. Electromagn. Adv. Appl., 2009,pp. 881884.

[13] D. Kurrant, E. Fear, and D. Westwick, Tumor response estimation inradar-based microwave breast cancer detection, IEEE Trans. Biomed.

Eng., vol. 55, no. 12, pp. 28012811, 2008.[14] M. Klemm, D. Gibbins, J. Leendertz, T. Horseman, A. Preece, R. Ben-

jamin, and I. Craddock, Development and testing of a 60-elementUWB conformal array for breast cancer imaging, in Proc. 5th Eur.Conf. Antennas Propag., Apr. 2011, pp. 30773079.

[15] X. Li, S. Davis, S. Hagness, D. van der Weide, and B. Van Veen, Mi-

crowave imaging via space-time beamforming: Experimental investi-gation of tumor detection in multilayer breast phantoms, IEEE Trans.Microw. Theory Tech., vol. 52, no. 8, pp. 18561865, 2004.
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


14/14



[16] X. Li, E. Bond, B. Van Veen, and S. Hagness, An overview of ultra-wideband microwave imaging via space-time beamforming for early-stage breast-cancer detection, IEEE Antennas Propag. Mag., vol. 47,no. 1, pp. 1934, Feb. 2005.

[17] D. R. Wehner, High Resolution Radar. Boston, MA: Artech HouseRadar, 1994.

[18] M. Hussain, Ultra-wideband impulse radar-an overview of the prin-ciples, IEEE Trans. Aerosp. Electron. Syst., vol. 13, no. 9, pp. 914,Sep. 1998.

[19] M. Hussain, Principles of space-time array processing for ultra-wide-band impulse radar and radio communications, IEEE Trans.Veh. Technol., vol. 51, no. 3, pp. 393403, May 2002.

[20] C. Clemente, Grays Anatomy of the Human Body. Philadelphia, PA:Lea & Febiger, 1985.

[21] W. H. Parsons, Cancer of the Breast. Springfield, IL: Thomas, 1959.[22] J. Y. Wo, K. Chen, B. A. Neville, N. U. Lin, and R. S. Punglia, Effect

of very small tumor size on cancer-specific mortality in node-positivebreast cancer, J. Clin. Oncol., vol. 29, no. 19, pp. 26192627, Jul.2011.

[23] J. S. Michaelson, M. Silverstein, W. J. , G. Weber, R. Moore, E.Halpern, D. Kopans, and K. Hughes, Predicting the survival of pa-tients with breast carcinoma using tumor size, J. Cancer, Aug. 2002.

[24] B. Razavi, RF Microelectronics. Upper Saddle River, NJ: Prentice-Hall PTR, 1998.

[25] S. Hagness, A. Taflove, and J. Bridges, Three-dimensional FDTDanalysis of a pulsed microwave confocal system for breast cancer de-tection: Design of an antenna-array element, IEEE Trans. AntennasPropag., vol. 47, no. 5, pp. 783791, May 1999.

[26] G. Liu, C. Guo, P. Zheng, and A. Ren, 2.320.1 GHz band notchedUWB monopole antenna, in Proc. 9th Int. Symp. Antennas Propag.

EM Theory (ISAPE), 2010, pp. 323326.[27] E. Fear and M. Stuchly, Microwave system for breast tumor detec-

tion, IEEE Microw. Guided Wave Lett., vol. 9, no. 11, pp. 470472,Nov. 1999.

[28] S. Hagness, A. Taflove, and J. Bridges, Two-dimensional FDTD anal-ysisof a pulsed microwave confocal system for breast cancer detection:Fixed-focus and antenna-array sensors, IEEE Trans. Biomed. Eng.,vol. 45, no. 12, pp. 14701479, Dec. 1998.

[29] A. F. Oskooi, D. Roundy, M. Ibanescu, P. Bermel, J. D. Joannopoulos,and S. G. Johnson, MEEP: A flexible free-software package forelectromagnetic simulations by the FDTD method, Comput. Phys.Commun., vol. 181, pp. 687702, Jan. 2010.

[30] ET Industries, 180 degree hybrid coupler-stripline design technicaldatasheet, [Online]. Available: http://www.etiworld.com/rf.pass.comp/180.Degree.Hybrid.Stripline.html

[31] J. E. Chen, Modeling RF systems, 2005 [Online]. Available: http://www.designers-guide.org/Modeling/

[32] W. Sheng, A. Emira, and E. Sanchez-Sinencio, CMOS RF receiversystem design: A systematic approach, IEEE Trans. Circuits Syst. I,

Reg. Papers, vol. 53, no. 5, pp. 10231034, May 2006.[33] X. Li and S. Hagness, A confocal microwave imaging algorithm for

breast cancer detection, IEEE Microw. Wireless Compon. Lett., vol.11, no. 3, pp. 130132, Mar. 2001.

[34] C. Enz and G. Temes, Circuit techniques for reducing the effects ofop-amp imperfections: Autozeroing, correlated double sampling, andchopper stabilization, Proc. IEEE, vol. 84, no. 11, pp. 15841614,Nov. 1996.

[35] S. Rong and H. Luong, A 0.05-to-10 GHz 19-to-22 GHz and 38-to-44GHz SDR frequency synthesizer in 0.13 m CMOS, in Proc. IEEE

Int. Solid- State Circuits C onf. (ISSCC ), Feb. 2011, pp. 464466.[36] S. Toso, A. Bevilacqua, M. Tiebout, N. Da Dalt, A. Gerosa, and A.Neviani, A 0.06 m m 11 mw local oscillator for the GSM standardin 65 nm CMOS, IEEE J. Solid-State Circuits, vol. 45, no. 7, pp.12951304, Jul. 2010.

[37] S. Dal Toso, A. Bevilacqua, M. Tiebout, N. Da Dalt, A. Gerosa, andA. Neviani, An integrated divide-by-two direct injection-locking fre-quency divider for bands s through k , IEEE Trans. Microw. TheoryTech., vol. 58, no. 7, pp. 16861695, Jul. 2010.

[38] A. Bevilacqua, F. P. Pavan, C. Sandner, A. Gerosa, and A. Neviani,A 3.47 GHz transformer-based dual-mode wideband VCO, inProc. 32nd Eur. Solid-State Circuits Conf. (ESSCIRC) , Sep. 2006, pp.440443.

[39] J. Borremans, A. Bevilacqua, S. Bronckers, M. Dehan, M. Kuijk, P.Wambacq, and J. Craninckx, A compact wideband front-end usinga single-inductor dual-band VCO in 90 nm digital CMOS, IEEE J.Solid-State Circuits, vol. 43, no. 12, pp. 26932705, Dec. 2008.

[40] A. Bevilacqua and A. Niknejad, An ultrawideband CMOS low-noiseamplifier for 3.110.6-GHz wireless receivers, IEEE J. Solid-StateCircuits, vol. 39, no. 12, pp. 22592268, Dec. 2004.

[41] F. Bruccoleri, E. Klumperink, and B. Nauta, Wide-band CMOS low-noise amplifier exploiting thermal noise canceling,IEEE J. Solid-StateCircuits, vol. 39, no. 2, pp. 275282, Feb. 2004.

[42] A. Bevilacqua, M. Camponeschi, M. Tiebout, A. Gerosa, and A. Ne-viani, Design of broadband inductorless LNAs in ultra-scaled CMOStechnologies, in Proc. IEEE Int. Symp. Circuits Syst. (ISCAS), May2008, pp. 13001303.

Matteo Bassi (GS11) was born in Padova, Italy, in1985. He received the B.S. and M.S. degrees in elec-tronics engineering from the University of Padova,Italy, in 2007 and 2009, respectively. He is currentlyworking toward the Ph.D. degree at the University ofPadova, Italy.

In 2008 and 2009 he was an EAP student atthe University of California, San Diego. His mainresearch interests are in the field of RFIC design.

Presently he is working on the design of ultrawide-band RF transceivers for bio-medical imaging.

Andrea Bevilacqua (S02M04) received theLaurea and Ph.D. degrees in electronics engineeringfrom the University of Padova, Italy, in 2000, and2004, respectively.

Presently, he is an Assistant Professor with the De-partment of Information Engineering, University ofPadova, Italy. His current research interests includethe design of RF/microwave integrated circuits, andthe analysis of wireless communication systems.

Dr. Bevilacqua serves as a member of the TPC ofIEEE ESSCIRC, and was a member of the TPC of

IEEE ICUWB. He is an Associate Editor of IEEE T RANSACTIONS ON CIRCUITS

AND SYSTEMSPART II: EXPRESS BRIEFS.

Andrea Gerosa (SM07) received the M.S. andPh.D. degrees in electrical engineering from theUniversity of Padova, Italy, in 1995 and 1998,respectively.

He is now an Associate Professor at the Univer-sity of Padova. He has published more than 70 papersin international journals or conference proceedings.His research activities include UWB transceivers andradars.

Andrea Neviani (M05) received the Laurea degree(cum laude) in physics from the University ofModena, Italy, in 1989 and the Ph.D. degree inelectronics and telecomunication engineering fromthe University of Padova, Italy, in 1994.

Since 1998 he is an Associate Professor at theUniversity of Padova, Italy. His main interest is thedesign of RFICs for communication and radar ap-plications, and mixed-signal circuits for biomedicalapplications. In his career, he has been coauthor ofabout 100 journal articles and conference papers.

Dr. Neviani has served as an Associate Editor of the IEEE T RANSACTIONSON CIRCUITS AND SYSTEMSPART I: REGULAR PAPERS since 2010.

detec cancer

Documents