Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband,...

23
Multidim Syst Sign Process DOI 10.1007/s11045-013-0250-7 Electronically scanned RF-to-bits beam aperture arrays using 2-D IIR spatially bandpass digital filters Arjuna Madanayake · Chamith Wijenayake · Rimesh Joshi · Mohammad Almalkawi · Leonid Belostotski · Len Bruton · Vijay Devabhaktuni Received: 31 July 2012 / Revised: 5 July 2013 / Accepted: 26 August 2013 © Springer Science+Business Media New York 2013 Abstract A beamforming system based on two-dimensional (2-D) spatially bandpass infi- nite impulse response (IIR) plane wave filtering is presented in a multi-dimensional signal processing perspective and the implementation details are discussed. Real-time implemen- tation of such beamforming systems requires modeling of computational electromagnet- ics for the antennas, radio frequency (RF) analog design aspects for low-noise amplifiers (LNAs), mixed-signal aspects for signal quantization and sampling and finally, digital archi- tectures for the spatially bandpass plane wave filters proposed in Joshi et al. (IEEE Trans Very Large Scale Integr Syst 20(12):2241–2254, 2012). Multi-dimensional spatio-temporal spec- tral properties of down-converted RF plane wave signals are reviewed and derivation of the spatially bandpass filter transfer function is presented. An example of a wideband antipodal Vivaldi antenna is simulated at 1GHz. Potential RF receiver chains are identified including a design of a tunable combline microstrip bandpass filter with tuning range 0.8–1.1 GHz. The 1st-order sensitivity analysis of the beam filter 2-D z-domain transfer function shows that for a 12-bits of fixed-point precision, the maximum percentage error in the 2-D magnitude A. Madanayake · C. Wijenayake (B ) · R. Joshi Department of Electrical and Computer Engineering, The University of Akron, Akron, OH, USA e-mail: [email protected] A. Madanayake e-mail: [email protected] M. Almalkawi · V. Devabhaktuni Department of Electrical and Computer Engineering, The University of Toledo, Toledo, OH, USA e-mail: [email protected] V. Devabhaktuni e-mail: [email protected] L. Belostotski · L. Bruton Department of Electrical and Computer Engineering, The University of Calgary, Calgary, AB, Canada e-mail: [email protected] L. Bruton e-mail: [email protected] 123

Transcript of Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband,...

Page 1: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign ProcessDOI 10.1007/s11045-013-0250-7

Electronically scanned RF-to-bits beam aperture arraysusing 2-D IIR spatially bandpass digital filters

Arjuna Madanayake · Chamith Wijenayake · Rimesh Joshi ·Mohammad Almalkawi · Leonid Belostotski · Len Bruton ·Vijay Devabhaktuni

Received: 31 July 2012 / Revised: 5 July 2013 / Accepted: 26 August 2013© Springer Science+Business Media New York 2013

Abstract A beamforming system based on two-dimensional (2-D) spatially bandpass infi-nite impulse response (IIR) plane wave filtering is presented in a multi-dimensional signalprocessing perspective and the implementation details are discussed. Real-time implemen-tation of such beamforming systems requires modeling of computational electromagnet-ics for the antennas, radio frequency (RF) analog design aspects for low-noise amplifiers(LNAs), mixed-signal aspects for signal quantization and sampling and finally, digital archi-tectures for the spatially bandpass plane wave filters proposed in Joshi et al. (IEEE Trans VeryLarge Scale Integr Syst 20(12):2241–2254, 2012). Multi-dimensional spatio-temporal spec-tral properties of down-converted RF plane wave signals are reviewed and derivation of thespatially bandpass filter transfer function is presented. An example of a wideband antipodalVivaldi antenna is simulated at 1 GHz. Potential RF receiver chains are identified including adesign of a tunable combline microstrip bandpass filter with tuning range 0.8–1.1 GHz. The1st-order sensitivity analysis of the beam filter 2-D z-domain transfer function shows thatfor a 12-bits of fixed-point precision, the maximum percentage error in the 2-D magnitude

A. Madanayake · C. Wijenayake (B) · R. JoshiDepartment of Electrical and Computer Engineering, The University of Akron, Akron, OH, USAe-mail: [email protected]

A. Madanayakee-mail: [email protected]

M. Almalkawi · V. DevabhaktuniDepartment of Electrical and Computer Engineering, The University of Toledo, Toledo, OH, USAe-mail: [email protected]

V. Devabhaktunie-mail: [email protected]

L. Belostotski · L. BrutonDepartment of Electrical and Computer Engineering, The University of Calgary, Calgary, AB, Canadae-mail: [email protected]

L. Brutone-mail: [email protected]

123

Page 2: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

frequency response due to quantization is as low as 0.3 %. Monte-Carlo simulations are usedto study the effect of quantization on the bit error rate (BER) performance of the beamform-ing system. 5-bit analog to digital converter (ADC) precision with 8-bit internal arithmeticprecision provides a gain of approximately 16 dB for a BER of 10−3 with respect to the nobeamforming case. ASIC Synthesis results of the beam filter in 45 nm CMOS verifies a realtime operating frequency of 429 MHz.

Keywords Beamforming · 2-D IIR · Broadband antenna · systolic arrays

1 Introduction

Digital apertures refer to beamforming systems based on antenna element arrays for thedirectional enhancement of spatio-temporal plane waves based on their directions of arrival(DOAs) (Lin et al. 1988; Bruton 2003; Wijenayake et al. 2012; Gunaratne and Bruton 2011b;Rajapaksha et al. 2012; Wen et al. 2013; Wei et al. 2013). Discrete-domain algorithms exist forachieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digitalsignal processing methods, which are realized using RF mixed-signal integrated circuits.Multidimensional spatio-temporal IIR array processing algorithms have been previouslyreported (Madanayake and Bruton 2008b; Madanayake et al. 2008; Bruton 2003; Wijenayakeet al. 2012) for such beamforming requirements in applications including radar, remotesensing, radio astronomy aperture synthesis and wireless communication systems. Thesearray processing algorithms exploit the specific spectral properties of plane wave signals in themulti-dimensional spatio-temporal frequency space and are shown to have low computationalcomplexity compared to traditional algorithms, which are based on digital phased arrays(Wijenayake et al. 2012; Joshi et al. 2012).

In this work, we explore the potential RF mixed-signal implementation details of a classof multi-dimensional spatio-temporal IIR algorithms, known as spatially bandpass 2-D IIRbeam filters. As recently proposed in Joshi et al. (2012), spatially bandpass 2-D IIR beamfilters exploit the multi-dimensional spatio-temporal frequency domain properties of planewave signals received by a uniform linear array (ULA) of antenna elements, after subsequentdown conversion and down sampling by the front-end RF signal chain.

Figure 1 shows an overview of a typical ULA beamforming system employing a 2-D IIRspatially bandpass beam filter. Extending Joshi et al. (2012), we explore implementationdetails for the 2-D IIR spatially bandpass beam filters including antennas, RF front-endsignal processing and digital arithmetic processing circuits and also study the effect of finiteprecision quantization towards the system performance. The RF front-end systems necessitatehigh flexibility for emerging reconfiguration and multi-frequency multi-waveform wirelesssystems. In modern scenarios, reconfigurable radios and radar adapt digital signal processorsas well as microwave RF front-end circuits on-the-fly under the control of cognition engines.We explore potential design of the RF front-ends containing a frequency-agile bandpassfilters and fixed-point digital integrated circuit design of the arithmetic circuits so that weobtain an acceptable complexity to noise-performance compromise.

The rest of this paper unfolds as follows. In Sect. 2 we review the multi-dimensional spatio-temporal frequency domain properties of down converted RF plane wave signals and thederivation of spatially bandpass 2-D IIR beam filter transfer function. In Sect. 3, we describethe antennas and mixed signal RF front-end sub-systems including an example design ofa tunable combline bandpass filter. In Sect. 4, we perform a quantitative study on the finiteprecision quantization effects of filter coefficients using multi parameter transfer function

123

Page 3: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fig. 1 System overview of 2-D IIR spatially bandpass beam filter based digital beamformer

sensitivity. We then present the design of a fixed-point arithmetic hardware architecture forthe spatially bandpass beam filter, using bit-error rate (BER) as a metric for determining theprecision requirements at each quantizer. The paper is concluded in Sect. 5.

2 Plane wave properties and spatially bandpass beam filter transfer function

2.1 Frequency domain properties of down-converted RF plane wave signals

Let wpw (x, y, z, ct) denote a propagating RF plane wave signal emitted by a distant trans-mitter, where (x, y, z) ∈ R

3 is the 3-D Cartesian spatial domain and ct ∈ R is time t nor-malized by the wave propagation speed c. A spatially sampled version of wpw (x, y, z, ct)is obtained by a ULA of antennas (along the x-axis) as shown in Fig. 1, and is given bywpw (n1Δx, 0, 0, ct), where Δx is the inter-antenna spacing and n1 = 0, 1, . . . , N − 1 isthe antenna index. For spatial Nyquist sampling, Δx = λmin/2, where λmin is the shortestwavelength of the RF signal of interest. Let the spatial DOA of the plane wave of interestwith respect to the ULA be ψ (Zhang et al. 2012; Liu et al. 2013; Zhu and Chen 2013),the carrier frequency be Fc and the temporal bandwidth be B. Time-synchronous samplingof wpw (n1Δx, 0, 0, ct) at the sampling frequency Fs = 2 (Fc + B) by N analog to digi-tal converts (ADCs) lead to the 2-D discrete domain spatio-temporal sequence w (n1, n2),where n2 is the temporal sampling index. The frequency domain representation ofw (n1, n2)

is obtained by following the 2-D discrete Fourier transform (DFT) (Dudgeon and Mersereau1984) and is denoted by W (ω1, ω2), where ωk is the normalized frequency variable in

123

Page 4: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

(a) (b)

Fig. 2 Spectral ROS of the RF plane wave signal received by a ULA a before and b after down-conversionand down-sampling to baseband (Gunaratne and Bruton 2011a; Joshi et al. 2012)

dimension k, where k = 1, 2. It is recalled that, the region of support (ROS) of the 2-Dspectrum W (ω1, ω2) is a line which passes through the origin in 2-D spatio-temporal fre-quency space ω ≡ (ω1, ω2) ∈ R

2 (Bruton and Bartley 1985). The angular orientation of thisline-shaped ROS is called the spatio-temporal DOA of the plane wave and is related to thespatial DOA by θ = tan−1 (sinψ), where the 2-D Nyquist sampling condition Δx = c/Fs

is assumed (Bruton and Bartley 1985). Due to angular uncertainty of the DOA ψ and thetemporally partially broadband nature of the carrier modulated RF signal, the ROS of thespectrum occupies a portion of a fan-shaped region as shown in Fig. 2a.

In typical wireless systems, RF signals received by each antenna in the ULA are downconverted to baseband or intermediate frequency. In the case of down conversion to baseband,with a temporal down-sampling factor equal to Ns = (1 + Fc/B), the spectral propertiesof W (ω1, ω2) are changed as shown in Fig. 2b (Gunaratne and Bruton 2011a; Madanayake2008; Joshi et al. 2012). The ROS of W (ω1, ω2) is shifted away from the origin along thespatial frequency axis and the angular orientation is also changed to α. The spatial frequencyshift of the ROS is given by ωx = ±π ((Fc/ (Fc + B)) sin θ) and α = tan−1 ((1/Ns) sinψ)(Madanayake 2008; Joshi et al. 2012). Directional selective enhancement of down-convertedRF plane wave signals, which are often encountered in wireless communication systemstherefore requires a 2-D spatio-temporal filter having a passband shape that can effectivelyencompass the spectral ROS shown in Fig. 2b. The spatially bandpass 2-D IIR beam filterproposed in Joshi et al. (2012) is suitable for this purpose and we briefly review the derivationof the filter transfer function next.

2.2 Spatially bandpass beam filter 2-D transfer function

As shown in Fig. 1, the spatially bandpass 2-D IIR filter has the z-transform transfer functionHB (z1, z2), where zk ∈ C is the z-transform variable along the dimension k = 1, 2. Deriva-tion of HB (z1, z2) is based on the practical-BIBO stable (Agathoklis and Bruton 1983) firstorder 2-D IIR frequency planar transfer function given by,

H (z1, z2) =(

1 + z−11

) (1 + z−1

2

)

1 + b′10z−1

1 + b′01z−1

2 + b′11z−1

1 z−12

, (1)

where the filter coefficients are recalled as (Madanayake and Bruton 2008a)

b′pq = R + (−1)p cosα + (−1)q sin α

R + cosα + sin α. (2)

123

Page 5: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

(a) (b)

Fig. 3 The 2-D magnitude frequency response of a∣∣∣H

(e jω1 , e jω2

)∣∣∣ and b∣∣∣HB

(e jω1 , e jω2

)∣∣∣ obtained by

setting zk = e jωk where k = 1, 2 in (1) and (3), respectively

It is also recalled that, the 2-D frequency response of H (z1, z2), which is obtained by eval-uating (1) on the unit bi-circle given by zk = e jωk , k = 1, 2, has a beam-shaped passbandpassing through the origin in ω (see Fig. 3a). Here, parameters α and R > 0 set the angu-lar orientation and sharpness of the beam-shaped passband, respectively (Madanayake andBruton 2008a).

However, we require the spatially bandpass beam filter transfer function HB (z1, z2) tohave a beam-shaped passband having a spatial frequency shift ωx , such that it can effec-tively encompass the 2-D spectrum of the down converted RF plane waves shown in Fig. 2b.Synthesis of this passband shape is achieved by modulating the 2-D impulse responseh (n1, n2) corresponding to H (z1, z2) in (1) along the spatial dimension (Madanayake 2008).Practical-BIBO stability of H (z1, z2) guarantees that h (n1, n2) is absolutely summablefor 0 ≤ n1 ≤ N and n2 ≥ 0. Therefore, αn1 h (n1, n2) also leads to an absolutely sum-mable sequence, hence a practical-BIBO stable 2-D filter, provided that |α| ≤ 1. Followingthis argument, by spatially modulating h (n1, n2) we obtain the required 2-D unit impulseresponse as h B (n1, n2) = cos (ωx n1) h (n1, n2) = ∑2

k=1 βkαn1k h (n1, n2), where βk = 1/2

and αk = e(−1)k jωx n1 for k = 1, 2. Following the property∑2

k=1 βkαn1k h (n1, n2)

2−D⇔∑2k=1 βk H (z1/αk, z2), we obtain the z-transform transfer function of the spatially bandpass

2-D IIR beam filter as

HB (z1, z2) =(

1 + z−12

)

2

2∑k=1

[1 + e(−1)k jωx z−1

1

1 + b′10e(−1)k jωx z−1

1 + b′01z−1

2 + b′11e(−1)k jωx z−1

1 z−12

],

which can be simplified to yield

HB(z1, z2) =∑2

i=0∑2

j=0 ai j z−i1 z− j

2

1 + ∑2i=0

∑2j=0 bi j z

−i1 z− j

2

(1 + z−12 ). (3)

The filter coefficients ai j and bi j in (3) are expressed in terms of the filter coefficients in (1)(i.e. before the spatial modulation is applied) and are given by (Joshi et al. 2012)

123

Page 6: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

ω2

ω1

(π , π)( π , π)

(π , π)( π ,π)

θω2 = ωt

( ωx,0)2-D Nyquist

squarex

y(b)(a)

Fig. 4 a Relationship between the frequency variables along a line passing through (−ωx , 0). b An examplearray factor |G B (ψ, ωt )| computed at ωt = π/4

a00 = 1 b00 = 0a01 = b′

01 b01 = 2b′01

a02 = 0 b02 = b′201

a10 = cos (ωx )(1 + b′

10

)b10 = 2b′

10 cos (ωx )

a11 = cos (ωx )(b′

01 + b′11

)b11 = 2 cos (ωx )

(b′

10b′01 + b′

11

)a12 = 0 b12 = 2b′

01b′11 cos (ωx )

a20 = b′10 b20 = b′2

10a21 = b′

11 b21 = 2b′10b′

11a22 = 0 b22 = b′2

11.

(4)

Magnitude frequency response∣∣HB

(e jω1 , e jω2

)∣∣, obtained by evaluating (3) on the unit bi-circle zk = e jωk , k = 1, 2 is shown in Fig. 3b, where the spatial frequency shift of the beamshaped passband can be observed.

2.3 Array factor (2-D polar response) of HB (z1, z2)

We compute the array factor (i.e. the 2-D polar response) (Kraus and Marhefka 2003) pro-duced by HB (z1, z2) as follows. Let G B (ψ, jωt ) denote the array factor, which is com-puted at the given temporal frequency ω2 = ωt as a function of spatial angle ψ . Con-sider a line passing through (−ωx , 0) and having the spatio-temporal inclination θ with theω2 axis (see Fig. 4a). At ω2 = ωt , along this line, the frequency variables are related byω1 = −ωt tan θ − ωx . By following the relationship tan θ = sinψ , where ψ is in the spa-tial domain (i.e. (x, y) ∈ R

2), we have ω1 = −ωt sinψ − ωx . Using this relationship, weevaluate the 2-D frequency response HB

(e jω1 , e jω2

)along the spatio-temporal line shown

in Fig. 4a for 0 ≤ θ ≤ π/4 (equivalently 0 ≤ ψ ≤ π/2) to obtain the array factor given byG B (ψ, jωt ) = HB

(e− j(ωt sinψ+ωx ), e jωt

). Figure 4b shows the magnitude of an example

array factor computed at ωt = π/4, where |GB (ψ, jωt )| = 1 at the design DOA.

2.4 Differential and integral form design equations

The transfer function in (3) leads to direct-form digital hardware realizations, which is dis-cussed in detail in Sect. 4. However, differential-form and integral-form realizations are alsopossible (Chen et al. 2011; Khoo et al. 2003; Reddy et al. 1997). It has been shown thatdifferential-form realizations result in low computational complexity whereas integral formexhibits low sensitivity to perturbations in filter coefficients (Bruton and Strecker 1983).In this paper, we only focus on the digital hardware realization of the direct-form version.

123

Page 7: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

However, we present the design equations (i.e. filter coefficients) for differential- and integral-form realizations, that can be used by digital designers.

In general, both differential- and integral-form realizations have the transfer function ofthe form

HB (z1, z2) =∑2

i=0∑2

j=0 ai j

(z−1

1

1+(−1)k z−11

)−i (z−1

2

1+(−1)k z−12

)− j

1 + ∑2i=0

∑2j=0 bi j

(z−1

1

1+(−1)k z−11

)−i (z−1

2

1+(−1)k z−12

)− j(1 + z−1

2 ), (5)

where k = 1 corresponds to integral-form and k = 2 corresponds to differential-form. The

new filter coefficients γi j ∈{

ai j , bi j

}in (5) are found in terms of the direct-form coefficients

γi j ∈ {ai j , bi j

}in (3) by equating the coefficients of z−i

1 z j2, where i, j = 0, 1, 2 and are

given by⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

γ00

γ10

γ01

γ11

γ20

γ02

γ12

γ21

γ22

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

=

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

1 0 0 0 0 0 0 0 0−(−1)k 2 1 0 0 0 0 0 0 0−(−1)k 2 0 1 0 0 0 0 0 0

4 −(−1)k 2 −(−1)k 2 1 0 0 0 0 01 −(−1)k 0 0 1 0 0 0 01 0 −(−1)k 0 0 1 0 0 0

−(−1)k 2 1 2 −(−1)k 0 −(−1)k 2 1 0 0−(−1)k 2 2 1 −(−1)k −(−1)k 2 0 0 1 0

1 −(−1)k −(−1)k 1 1 1 −(−1)k −(−1)k 1

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣

1γ10

γ01

γ11

γ20

γ02

γ12

γ21

γ22

⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦

(6)

where k = 1 is for integral-form and k = 2 is for differential-form. Note that, both in (5) and(6) b00 = 0.

3 Mixed-signal RF front-end

3.1 Antennas

The antennas are located in a ULA configuration (see Fig. 1), with inter-antenna spacingΔx .The combined transfer-function of the digital beamformer and antenna array, evaluated at thetemporal frequency ωt , as a function of the spatial angle ψ is obtained using the principle ofpattern multiplication (Kraus and Marhefka 2003) as

HULA (ψ, jωt ) = AE (ψ, jωt ) · GB (ψ, jωt ) , (7)

where, AE (ψ, jωt ) is the broadband polar response of one element in the antenna array inisolation and G B (ψ, jωt ) is the array factor produced by the spatially bandpass beam filter.For simplicity, we assume omnidirectional broadband elements such that |AE (ψ, jωt )| ≈ 1.Effects due to mutual coupling between ULA elements is not considered at this stage (Warnicket al. 2009) and is kept for further investigations.

In Fig. 5a, we provide an example design of a broadband antipodal Vivaldi antenna element,which can potentially be used with the spatially bandpass beam filter (Yngvesson et al. 1989;Jasik 1993). The simulated scattering parameter S11 from 0.5–1.5 GHz and the far-fieldradiation pattern at 1 GHz are shown in Fig. 5b, c, respectively. Such Vivaldi elements offernearly omni directional far-field radiation patterns through a wide band of frequencies. The

123

Page 8: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

l1

l3

LA = 190 mm

material: RO4003, εr = 3.55Metal: Copper

l1 = 158.55 mml2 = 142.7 mml3 = 32.77 mmlt = 5 mmwt = 3.3 mm

WA = 120 mm

(a)

(c) (d)

broadbandmatchingnetwork

Antenna

X

lt

wt

LA

WA

(futurework)

Substrate:thickness: 1.527mm

(b)

l2

Fig. 5 Broadband antipodal Vivaldi antenna element: a example design, b simulated S11 parameter and csimulated far-filed radiation pattern. d The use of a broadband matching network

pattern shown in Fig. 5c falls on the approximate center frequency of the band of interest 800–1,100 MHz. It can be observed from the S11 curve in Fig. 5b that, the antenna achieves betterthan 10 dB return loss up to 1.5 GHz for a nearly octave bandwidth. As a future extension forthe current design, we propose to use a broadband matching network (Sussman-Fort 1994;Sussman-Fort and Rudish 2009) in between the antenna and the subsequent RF receiver chain(see Fig. 5d).

3.2 RF receiver chain

A number of potential RF receiver chains are available to convert an incident plane wavesignal received by the antenna to a voltage at the input to an analog to digital converter(ADC). The main performance metrics to be considered for a receiver are its sensitivity andits response to interference. The sensitivity of a receiver defines its ability to detect weaksignals while still achieving acceptable signal to noise ratios (SNR). The potential receptionof interference signals, known as blockers, by the antenna makes receiver design challengingand necessitates the need for proper filtering (Andrews and Molnar 2010c).

The sensitivity of a receiver is strongly dependent on its noise figure (Behzad 2008; Leeet al. 1999). A typical communication receiver would have noise figure specifications in the5–6 dB range. For example, a noise figure of a WLAN receiver is usually specified at 6 dB andthat of a GSM receiver is 5 dB. Out of 5–6 dB of noise figure, 2–3 dB is usually assigned to thelosses in front-end filter and circuitry in front of the LNA. In the case of the filter discussedin Sect. 3.4, its noise figure is relatively small as can be seen from its in-band insertion loss(<2 dB). This relaxes the requirements for the LNA. To deal with the interferers, an RFcommunication system usually has two types of filtering: the front-end filtering to define theinput bandwidth and filter out out-of-band blockers and the channel select filleting to selectthe desirable signal and remove in-band blockers.

123

Page 9: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Most planar filters, including the front-end filter considered in this work, produce lowrejection at odd harmonics of the filter center frequency and appear as having re-entrancebands at those frequencies. Out-of-band blockers at the frequency of the re-entrance bandsmay not be rejected sufficiently. Therefore, in addition to the standard architecture whichemploys only one front-end bandpass filter, a low pass filter is also required to remove theseblockers. In general, blockers can be much stronger than the desired signal. For examplein case of a GSM system, in-band blockers 3 MHz away from the signal channel can beas large as −23 dBm and out-of-band blockers 80 MHz away from the channel can be ashigh as 0 dBm, while the desired signal can be as low as −100 dBm. To deal with the highpower in-band blockers, the receiver linearity must be high enough1 to avoid receiver satu-ration and non-linear behavior. To reject the out-of-band blockers, a surface acoustic wave(SAW) filter with very steep transition from its passband to stopband is often required. Inthis work, in place of a fixed passband SAW filter, a tunable filter is proposed. In its cur-rent implementation, the proposed filter may not have as much rejection as a SAW filter butthis can be improved by the rejection in the receiver itself as will be discussed later in thissection.

We consider a typical communication system in order to analyze the effects of digitalfiltering on its BER. Receiver systems such as described in Locher et al. (2008), Gianniniet al. (2009) are considered as typical and this paper will base the forthcoming discussionson a receiver in Giannini et al. (2009), which is particularly interesting because it is capableof receiving a wide frequency range of signals, meets specifications of most communicationprotocols, and uses passive mixing for additional RF filtering performed at IF (Andrews andMolnar 2010a,b,c). The extra RF filtering can remove the re-entrance bands and enhancethe RF bandwidth definition. The IF output of the receiver in Giannini et al. (2009) willbe sampled by a 200 MHz ADC. Some of ADCs that can meet these requirements includeVerma and Razavi (2009), Lin et al. (2010).

3.3 Bandpass sampling

In theory, perfect temporal sampling of bandpass signal having bandwidth B Hz centeredat frequency Fc Hz enables the direct down-conversion and digitization using ADCs thatoperate at the lowest possible clock frequency of Fclock = B Hz. This is known as bandpasssampling (BPS) and leads to the elimination of the down-conversion step using an analogmixer circuit. The application of multi rate signal sampling can lead to both in-phase (I) andquadrature (Q) components of the bandpass signal to be directly available at the output of twoADCs that operate at frequency B Hz using polyphase non-overlapping clocks φ1 and φ2.A specification of 40 dB attenuation at fc ± 100 MHz where fc is the RF center frequencyof interest and ADC sampling rate of Fclock = 200 MHz is assumed for both φ1 and φ2,leading to a bandpass sampling bandwidth of B = 200 MHz per channel, which ensuresalias components are at least 40 dB below signals of interest.

However, BPS, also known as sub-sampling, is hard to realize in practice because ofcomplexities due to noise folding (Pekau and Haslett 2006, 2005). Furthermore, bandpasssampling requires very fast track-and-hold amplifiers to be used in the sub-sampling ADC,which usually implies custom integrated circuits. Therefore, RF chain in Fig. 6 is the practicalchoice at present. The fixed low pass filtering in Fig. 6 is used to overcome the effect of higherharmonic passbands of the tunable bandpass filter.

1 The input IP3 for a GSM receiver is typically −21 dBm.

123

Page 10: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fig. 6 Two possible RF chains a single-stage I-Q mix down, and b sub-sampling direct-conversion receiver

Fig. 7 Lumped equivalent circuit of the electronically-tunable combline bandpass filter

3.4 Tunable combline bandpass filter

A tunable filter is required to create a system, which is capable of adjusting its input fre-quency range while still rejecting interference. Our example design uses electronically tun-able combline bandpass filter (Hunter and Rhodes 1982) at the output of the LNA havinglumped parameter model shown in Fig. 7. Combline filters are compact, have high qualityfactors (unloaded) and can be designed to operate over a wide frequency range. In thesefilters, capacitive end-loading of the resonators leads to smaller sizes when fabricated incomparison with filters based on a quarter-wave length resonance. Combline filters consistof an array of parallel resonators that are short circuited at one end with capacitive load atthe other. The resonators are positioned such that the short circuits are on one side of thefilter with the capacitors on the other. External coupling consist of same sided impedancetransformers (Levy et al. 2002), which appear as extra resonators at each end of the filter.Such a topology leads to the desired tuning range at constant bandwidth with no adjustmentto couplings needed.

Figure 7 shows the two port circuit model of a 2nd-order combline bandpass filter with tun-ing capability achieved by varying the capacitance (C). Inter-resonator coupling is inductive

123

Page 11: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fig. 8 Layout of the tunablecombline bandpass filter. Thetunable capacitors are shown as C Cs

C

50 Ω line 50 Ω line

w1 w2 w2 w1

ls1 s2 s1

C

Cs

Table 1 Combline designparameters

Symbol Value

C 0.5–4.5 pF

Cs 0.1μF

l 18.3 mm

wl 3.4 mm

w2 2.1 mm

S1 0.25 mm

S2 5.3 mm

(LC ) modeled by a series branch impedance. The port impedances are implemented at 50Ωcharacteristic impedance.

The combline bandpass filter structure in Fig. 8 was designed using full-wave electromag-netic (EM) simulation using Ansys HFSS. The dimensions of the microstrip lines are givenin Table 1. The design was printed on Rogers (RO4003C) substrate with a dielectric constantof 3.55 and a loss tangent of 0.0027. The height of the substrate is 0.813 mm. Both sourceand load couplings as well as the inter-resonator couplings were adjusted for best electricalperformance resulting in the simulated electrical performances for the forward scatteringparameter (S21) shown in Fig. 9. Tunable center frequency for several values of variablecapacitance is shown.

3.5 The effect of non-identical RF front-end filters

So far we assumed that the RF front-end filters at each signal path n1 = 0, 1, . . . , N − 1in Fig. 1 have identical behavior. However, in practice, these front-end filters will have non-identical frequency responses due mismatches in passive electromagnetic components and

123

Page 12: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fig. 9 Forward transfer-functionof the combline bandpass filterfor S21

process, voltage and temperature variations in CMOS integrated circuits. We now present amulti-dimensional signal processing approach to model this effect.

Let Tn1 ( jΩ2) denote the input transfer function of the nth1 RF front-end filter (see Fig. 6),

where n1 = 0, 1, . . . N − 1 and |Ω2| < ∞ is the frequency variable corresponding to thecontinuous time domain. In the ideal scenario, where all front-end filters are identical wehave T0 ( jΩ2) = T1 ( jΩ2) = · · · = TN−1 ( jΩ2) = T ( jΩ2), leading to 2-D input-outputrelationship given by

Y(

e jω1 , e jω2)

= T ( jΩ2) HB

(e jω1 , e jω2

)W

(e jω1 , e jω2

), (8)

where W(e jω1 , e jω2

)and Y

(e jω1 , e jω2

)are the 2-D input and output transforms of the

spatially bandpass beam filter, respectively. However, when the RF front-end filters are non-identical we get a 2-D windowing function of the form Tw

(e jω1 , jΩ2

) = F [Tn1 ( jΩ2)

],

where F [ ] denotes the spatial discrete Fourier transform. This leads to a modified 2-Dinput-output relationship given by

Y(

e jω1 , e jω2)

= Tw(

e jω1 , jΩ2

)∗ HB

(e jω1 , e jω2

)W

(e jω1 , e jω2

), (9)

where ∗ denotes 1-D convolution along spatial dimension. By measuring the S-parameters(i.e. S21) of each RF front-end filter using a vector network analyzer, under matched condi-tions, one can obtain Tn1 ( jΩ2) for n1 = 0, 1, . . . , N − 1 and use (9) to predict the effect ofnon-identical RF front-ends on the 2-D spatio-temporal filtering operation.

4 Fixed-point digital hardware realization of spatially bandpass 2-D beam filter

Having discussed potential receiver architecture, the beamformer realization is discussednext. Because the transfer function HB(z1, z2) in (3) is derived based on a 2-D passive pro-totype network (Joshi et al. 2012; Fettweis and Basu 2011; Bruton and Bartley 1985), itleads to practical-BIBO stable discrete domain realizations (Agathoklis and Bruton 1983).Massively parallel systolic array based real-time hardware architecture to realize HB (z1, z2)

in (3) has been proposed in Joshi et al. (2012). However, the quantization noise inherent indigital realizations and its effect on wireless system performance was not studied in Joshi etal. (2012). In this section, we present a quantitative study on the finite precision digital arith-metic quantization effects using the 1st-order multi-parameter transfer function sensitivity.Quantization is an unavoidable problem in digital arithmetic circuits based on fixed-point

123

Page 13: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

number systems. Coefficient quantization leads to deviation in the filter frequency response.Quantization in the data path causes noise injection, which in turn reduces the signal-to-noiseratio at the filter output and manifests as a degradation of the BER of the wireless system(Rizvi et al. 2008).

Following the 2-D inverse z-transform of (3) under zero initial conditions (ZICs), weobtain the 2-D difference equation

y (n1, n2) =2∑

p=0

2∑q=0

apqw(n1 − p, n2 − q)− bpq y(n1 − p, n2 − q), (10)

where the filter coefficients apq , bpq are given in (4).Digital hardware implementation of (10) is achieved using a massively parallel systolic

array architecture (Madanayake and Bruton 2008a; Joshi et al. 2012; Khoo et al. 2013), whichconsists of identical interconnected parallel processing core modules (PPCMs) as shown inFig. 1. Recursions along temporal dimension happen inside each PPCM and recursions alongthe spatial dimension manifest as parallel interconnections between the PPCMs. As shownin Fig. 1, the first two PPCMs are provided with spatial ZICs as required for the practical-BIBO stability of HB (z1, z2) (Agathoklis and Bruton 1983; Lin and Bruton 1989). The

separable FIR operation(

1 + z−12

)in (3) is performed at the output of the last PPCM.

Each PPCM has 5 inputs and a single output and is characterized by the signal flow graph(SFG) shown in Fig. 11. In Sect. 4.2, we further describe the finite precision quantizationpresent inside each PPCM which are represented as Qk [] for k = 1, 2, . . . , 14 in the SFG inFig. 11.

4.1 Transfer function sensitivity analysis

The fixed-point systolic-array processor realization of (10) requires the ideal values of coef-ficients γpq ∈ {

apq , bpq}

be approximated by finite-precision multipliers gpq ≈ γpq . Quan-tization of type rounding of γpq to Dpq bits of fractional precision leads to uniformly dis-tributed errors −2−Dpq+1 ≤ Δγpq < +2−(Dpq+1) such that γpq = gpq − Δγpq leading tothe percentage errors

(Δγpq/γpq

) × 100 %.Such quantization errors in coefficient γpq cause the magnitude frequency response∣∣HB

(e jω1 , e jω2

)∣∣ to deviate from the ideal value leading to percentage errorΔHpq given by(11)

ΔHγpq (ω1, ω2) = S∣∣HB

(e jω1 ,e jω2

)∣∣γpq

(Δγpq

γpq

)× 100 % (11)

where S∣∣HB

(e jω1 ,e jω2

)∣∣γpq is the 1st-order sensitivity function of

∣∣HB(e jω1 , e jω2

)∣∣ with respectto the coefficient γpq , given by (12).

S∣∣HB

(e jω1 ,e jω2

)∣∣γpq = Re

[∂

∂γpq

{HB

(e jω1 , e jω2

)} γpq

HB(e jω1 , e jω2

)]

(12)

Following (11) and (12) we obtain the percentage magnitude frequency response errordue to quantization of numerator coefficients apq and denominator coefficients bpq as givenby (13) and (14), respectively.

123

Page 14: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

ΔH

b 21

(ω1,ω

2)

ω1

ω2

ω1,2 ≤ 3π4

(a) ω1,2 ≤ 3π4

ω2

ω1

ω1,2 ≤ 3π4

ω2

ω1

ΔH

Max

(ω1,

ω2)

(c)

ΔH

b 10

(ω1,ω

2)(b)

Fig. 10 Percentage magnitude frequency error in due to quantization errors in a b10 and b b21. The maximumpercentage error ΔHMax (ω1, ω2) given in (15)

ΔHapq (ω1, ω2) = Re

⎡⎢⎢⎢⎣

e− j(pω1+qω2)Δapq

(1 + e− jω2

) (2∑

l=0

2∑m=0

alme− j(lω1+mω2)

)

⎤⎥⎥⎥⎦ × 100 % (13)

ΔHbpq (ω1, ω2) = Re

⎡⎢⎢⎢⎣

−e− j(pω1+qω2)Δbpq

1 +2∑

l=0

2∑m=0

blme− j(lω1+mω2)

⎤⎥⎥⎥⎦ × 100 % (14)

We then evaluateΔHmax (ω1, ω2) by considering the algebraic summation of error termsin magnitude frequency response

∣∣HB(e jω1 , e jω2

)∣∣ due to perturbations in individual filtercoefficients γpq ∈ {

apq , bpq}, where p, q ∈ {0, 1, 2}. ΔHmax (ω1, ω2) represents the total

error in the magnitude response and is given by

ΔHMax (ω1, ω2) =2∑

p=0

2∑q=0

(ΔHapq (ω1, ω2)+ΔHbpq (ω1, ω2)

)(15)

Figure 10a, b shows examples of ΔHapq (ω1, ω2) and ΔHbpq (ω1, ω2), respectively andFig. 10c shows ΔHMax (ω1, ω2) for 12 bits of fractional precision. It can be observed that,HB (z1, z2) has a lower percentage error in magnitude around 0.3 %.

123

Page 15: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

+

++

+

+

+

+

+

++

+ +

+

y( n1 ,n

2 )

Q1 [ ]

ΔT

Q1 [ ]

Q1 [ ]

Q4 [ ]

Q4 [ ]

Q2 [ ]

Q2 []

Q3 [ ]

Q2 [ ]

Q2 [ ]

Q3 [ ]

Q6 [ ] Q7 [ ]

Q7 [ ]

Q8 [ ]

Q8 [ ]

Q9 [ ]

Q6 [ ] Q7 [ ]

Q7 [ ]

Q8 [ ]

Q8 [ ]

Q9 [ ]

Q4

[]

Q5 [ ]

ΔT

ΔT

ΔT

ΔT

ΔT

ΔT

ΔT

ΔT

a01

a10

a11

a20

a21

− b02

− b01

− b22

− b21

− b20

− b11

− b10

− b12

An1

Bn1

Cn1

Dn1

En1

Fn1

PPCM at spatial location n1n1 = 0, 1, ..., N1 − 1

finite precision quantizers

Qk [ ] where k = 1, 2, ..., 14

Q10 [ ]

Q11 [ ] Q12 [ ]

Q14 [ ]

Q13 [ ]

Q13 [ ]

y (n1 − 2, n2)

y (n1 − 1, n2)

w (n1 − 2, n2)

w (n1 − 1, n2)

w (n1, n2)

Fig. 11 Signal flow graph (SFG) of PPCM at spatial location n1 including the finite precision quantizationrepresented by Qk [ ] where k = 1, 2, . . . , 14

4.2 Quantization inside the PPCMs

Figure 11 shows the internal SFG of a single PPCM, which contains 12 adders and 13multipliers. Finite precision quantization at each arithmetic operation is represented by thequantizers Qk [ ], k = 1, 2, . . . , NQuant in Fig. 11 where NQuant = 14 for this design. Quan-tization in the data path results in the injection of additive white Gaussian noise (AGWN)

of power σ 2k = Δ2

k12 at each quantizer k = 1, 2, . . . NQuant where the step size is Δk .

For finite impulse response (FIR) realizations containing only few quantizers, the effectof this additional noise due to σ 2

k , k = 1, 2, . . . , NQuant can be modeled in closed-form,leading to an estimate in BER penalty that results due to noise sources within the filtercircuit.

However, for massively-parallel systolic-array realizations for 2-D IIR (recursive) dig-ital filters, the number of quantizers can be up to several hundred, depending on transfer-function selectivity (i.e. order) and the number of antenna elements in the ULA. The highcomplexity and presence of multiple spatio-temporal feed-back paths in the 2-D SFG ofsuch non-separable IIR multidimensional filter circuits make it difficult to derive closed-form expressions for the BER penalty due to quantization effects. To address this problem,our approach is to use empirical methods together with Monte-Carlo analysis to deter-mine the penalty for BER by channel simulation in the spatio-temporal domain, whichin turn is connected with hardware design choices that set each precision level for thequantizers.

Let the word length of the ADC output present at each antenna’s down-conversion chain isdenoted by W with binary point location at D from the least significant bit (LSB). The PPCMinternal signal word length for this given ADC resolution (W, D) is governed by the twoparameters A and B, where A controls the word length growth in the signal and B controlsthe position of the binary point. Here, A and B serve as parameters that can be varied in anoptimization scheme such that acceptable performance-complexity-speed trade-offs can bearrived at in reasonable time.

Table 2 shows the internal quantizer sizes for Qk [ ] where k = 1, 2, . . . , 14 given inFig. 11.

123

Page 16: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Table 2 Internal quantizer sizesfor ADC resolution of wordlength W , binary point position Dand design parameters A and B

Quantizer Word length Binary point

Q1 [ ] W D

Q2 [ ], Q7 [ ] W + A − 5 D + B − 2

Q3 [ ], Q8 [ ] W + A − 4 D + B − 2

Q4 [ ], Q9 [ ] W + A − 3 D + B − 1

Q5 [ ], Q10 [ ], Q13 [ ] W + A − 2 D + B − 1

Q6 [ ], Q12 [ ] W + A D + B

Q11 [ ], Q14 [ ] W + A − 1 D + B

4.3 Penalty on BER performance versus fixed-point precision

The effects of quantization in systolic 2-D IIR beam filters have been analyzed in Madanayakeet al. (2012). Here, the effects of quantization on BER of the 2nd-order spatially bandpass2-D IIR beam filter for different cases of ADC resolutions and PPCM internal precisions arestudied and simulated.

The test signal considered for the simulation was a partially-broadband cosine modulatedGaussian signal S(t) = cos(2π fC t)e−At2

having bandwidth B = 250 MHz at a carrierfrequency of fC = 1 GHz with fractional bandwidth of F BW = 2B/ fC = 50 % containingone desired signal having spatial DOA of ψ = 35◦ and four interfering cosine modulatedGaussian signals at spatial DOAs 20◦, 50◦, 65◦ and 80◦. Interference power was varied cor-responding to a range of signal to interference ratio (SIR) −20 to 5 dB. The composite signalreceived by the sensors (N = 16 sensors and 60 samples per symbol in our simulation) arebi-phase modulated by random streams of data bits, down-converted to intermediate base-band, low-pass filtered and down-sampled before sending it to the spatially bandpass 2-D IIRbeam filter circuit to obtain the desired signal at the output. The simulations assumed an SNRof 18 dB, which includes the quantization effects corresponding to an ADC of 3 bit precision.

4.4 Effect of ADC resolution

Normalized input signals were sampled by the ADC having resolution of (W,D) and passedto the filter containing N = 16 identical PPCMs (see Fig. 1) with internal precision asindicated in Table 1. To observe the effect of ADC resolution on the BER performance, aseries of Monte-Carlo simulations for the BER versus SIR curves of the filter were carried outfor ADCs having resolutions of 8 to 4 bits (W, D) ∈ {(8, 3) , (7, 6) , . . . , (4, 3)} at internalPPCM precision of (A, B) = (8, 3). The result of the simulation is plotted in Fig. 12, whichshows that the performance of the 2-D IIR spatially bandpass beam filter for various ADCresolutions is higher compared to the case of non beamforming (approximately 17 dB fora BER of 10−3 at W = 6, 7 and 8). It can also be observed that the beam filter with ADCresolution of W = 4 bits is only suitable for systems with high interferences. The performanceapproaches the ideal infinite precision performance as we choose larger values of W . In fact,the beam filter circuit with a 6 bit ADC, i.e. (W, D) = (6, 5), and PPCM internal precisionsat (A, B) = (8, 3) has almost similar BER performance as the ideal infinite precision case.

4.5 Effect of PPCM internal precision

Another set of simulations were carried out for various cases of PPCM internal precisions(A, B) ∈ {(9, 4) , (8, 3) , (7, 2) , (6, 1)} (such that A − B = 5) at a fixed ADC resolution of

123

Page 17: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fig. 12 Investigation of ADC precision on BER for A = 8 and B = 3

Fig. 13 Investigation of PPCM internal precisions on BER for ADC resolution of W = 6 and D = 5

W = 6 and W = 5 bits. The result of the plot for a 6 bit ADC resolution at varying PPCMinternal precisions is shown in Fig. 13, which shows a gain of approximately 16 dB for a BERof 10−3 at (A, B) = (7, 2) compared to the case without beamforming. With (A, B) = (7, 2)as PPCM internal precisions for ADC resolution of W = 6 bits, the performance severelydegrades at higher SIR values making it unsuitable for systems dealing with SIR> −5 dB.Figure 14 shows a similar BER versus SIR plot for a W = 5 bit ADC resolution. It canbe seen that the performance close to infinite precision case was observed for A > 8, forW = 5 bits ADCs (with a gain of approximately 16 dB for a BER of 10−3 with respectto no beamforming case), while the performance severely degrades for A ≤ 7. Then if the2nd-order spatially bandpass 2-D IIR beam filter is used with W = 5 bits ADCs, the PPCM

123

Page 18: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fig. 14 Investigation of PPCM internal precisions on BER for ADC resolution of W = 5 and D = 4

Table 3 Hardware resources andperformance parameters fromASIC synthesis using 45 nmCMOS technology

Parameter Value

Area (A) 1.02 mm2

Critical path delay (TC P D ) 2.33 ns

Frequency Fclk 429.2 MHz

Leakage power 8.06 mW

Dynamic power 1354.3 mW

Total power 1362.3 mW

ATC P D 2.38 mm2ns

AT 2C P D 5.54 mm2 (ns)2

internal processor word sizes should be at least of A = 8 bits for desirable BER versus SIRperformance.

4.6 Hardware resources and performance in 45 nm CMOS

The 2-D IIR spatially bandpass beam filter was synthesized to application specific integratedcircuit (ASIC) using 45 nm CMOS technology with the fixed point arithmetic precision(W, D, A, B) = (5, 4, 8, 4) using Cadence Encounter RTL compiler tool. Table 3 reportsthe ASIC synthesis results including area, critical path delay, operating frequency, powerconsumption and area-time complexity metrics. The ASIC synthesis assumes an operatingvoltage of 1.1 V and temperature 27 ◦C.

5 Conclusions

In this contribution, we present an RF mixed signal implementation scheme suitable forbeamforming using recently proposed 2-D IIR spatially-bandpass beam filters in Joshi et al.(2012). Such beamforming systems have potential applications in wireless communication

123

Page 19: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

systems, where down-converted RF plane waves are typically encountered. The proposedimplementation includes antenna element design, RF front-end receiver chains includinga tunable combline bandpass filter design and finite precision digital arithmetic processordesign for the real-time implementation of the spatially bandpass beam filter 2-D z-domaintransfer function HB (z1, z2). Simulations suggest that broadband Vivaldi antenna elementwith a maximum directivity around 6 dBi at 1 GHz can be potentially used in a uniformlinear array configuration to obtain the 2-D spatio-temporal input signal to the beamformer.Two potential RF receiver chains are identified for front-end signal conditioning, includinga microstrip design of a tunable combline bandpass filter. The forward scattering transferfunction of the bandpass filter is simulated for the tunable center frequency range 0.8–1.1 GHz.

Following the front-end RF signal processing, the digitized antenna signals are processedby HB (z1, z2) using a massively parallel systolic array processor having interconnectedparallel processing core modules (PPCMs). Finite precision quantization effects of the filtercoefficients are analyzed using 1st-order transfer function sensitivity theory, which shows thatfor a 12 bits of fractional precision the maximum error in 2-D magnitude frequency responseis around 0.3 %. Extensive Monte-Carlo simulations are used to investigate the effect of ADCprecision and PPCM internal precision towards the BER performance of the system. It wasobserved that, for ADC word lengths 6–8 bits with the PPCM internal precision of 8 bits, theproposed beamforming system achieves a SIR gain of around 17 dB for the given BER of10−3 compared to the case with no beamforming. Digital hardware realization of the spatiallybandpass beam filter is mapped to standard cell ASIC using 45 nm CMOS technology up tothe synthesis level with a clock speed of 429 MHz.

Acknowledgments Dr. Arjuna Madanayake gratefully acknowledges the financial support from the USNational Science Foundation (NSF) Grant #1247940 for EARS and the Office of Naval Research (ONR)Grant #N000141310079.

References

Agathoklis, P., & Bruton, L. T. (1983). Practical-BIBO stability of N-dimensional discrete systems. IEEProceedings, 130(6), 236–242.

Andrews, C., & Molnar, A. C. (2010a). Implications of passive mixer transparency for impedance matchingand noise figure in passive mixer-first receivers. IEEE Transactions on Circuits and Systems I: RegularPapers, 57(12), 3092–3103.

Andrews, C., & Molnar, A. C. (2010b). A passive-mixer-first receiver with baseband-controlled RF impedancematching,< 6dB NF, and>27 dBm wideband IIP3. In: IEEE international solid-state circuits conference,pp. 46–47.

Andrews, C., & Molnar, A. C. (2010c). A passive mixer-first receiver with digitally controlled and widelytunable RF interface. IEEE Journal of Solid-State Circuits, 45, 2696–2708.

Behzad, A. (2008). Wireless LAN radios: System definition to transistor design. Wiley-IEEE Press,ISBN:9780470209301.

Bruton, L. T. (2003). Three-dimensional cone filter banks. IEEE Transactions on Circuits and Systems I:Fundamental Theory and Applications, 50(2), 208–216.

Bruton, L. T., & Bartley, N. R. (1985). Three-dimensional image processing using the concept of networkresonance. IEEE Transactions on Circuits and Systems, 32, 664–672.

Bruton, L. T., & Strecker, T. E. (1983). Two-dimensional discrete filters using spatial integrators. IEE Pro-ceedings of Electronic Circuits and Systems, 130, 271–275.

Chen, P. Y., Van, L. D., Khoo, I. H., Reddy, H. C., & Lin, C. T. (2011). Power-efficient and cost-effective 2-Dsymmetry filter architectures. IEEE Transactions on Circuits and Systems I: Regular Papers, 58, 112–125.

Dudgeon, D. E., & Mersereau, R. M. (1984). Multidimensional digital signal processing. Englewood Cliffs,NJ, USA: Prentice-Hall. 07632.

123

Page 20: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Fettweis, A., & Basu, S. (2011). Multidimensional causality and passivity of linear and nonlinear systemsarising from physics. Multidimensional Systems and Signal Processing, 22, 5–25.

Giannini, V., Nuzzo, P., Soens, C., Vengattaramane, K., Ryckaert, J., Goffioul, M., et al. (2009). A 2-mm2

0.1–5 GHz software-defined radio receiver in 45-nm digital CMOS. IEEE Journal of Solid-State Circuits,44(12), 3486–3498.

Gunaratne, T., & Bruton, L. (2011a). Adaptive complex-coefficient 2D FIR trapezoidal filters for broadbandbeamforming in cognitive radio systems. Circuits, Systems, and Signal Processing, 30, 587–608.

Gunaratne, T., & Bruton, L. (2011b). Broadband beamforming of dense aperture array (DAA) and focal planearray (FPA) signals using 3D spatio-temporal filters for applications in aperture synthesis radio astronomy.Multidimensional Systems and Signal Processing, 22, 213–236.

Hunter, I. C., & Rhodes, J. D. (1982). Electronically tunable microwave bandpass filters. IEEE Transactionson Microwave Theory and Techniques, 30(9), 1354–1360.

Jasik, H. (1993) Antenna engineering handbook. New York: McGraw-Hill, ISBN:9780070323810Joshi, R. M., Madayanake, A., Bruton, L. T., & Adikari, J. (2012). Synthesis and array processor realization of

a 2D IIR beam filter for wireless applications. IEEE Transactions on Very Large Scale Integration Systems,20(12), 2241–2254.

Khoo, I. H., Reddy, H. C., & Rajan, P. K. (2003). Delta operator based 2-D filters: symmetry, stability, anddesign. In IEEE international symposium on circuits and systems (ISCAS).

Khoo, I.-H., Reddy, H., Van, L.-D., & Lin, C.-T. (2013). General formulation of shift and delta operator based2-D VLSI filter structures without global broadcast and incorporation of the symmetry. MultidimensionalSystems and Signal Processing, 1–34. doi:10.1007/s11045-013-0232-9.

Kraus, J. D., & Marhefka, R. J. (2003). Antennas for all applications. New York: McGraw-Hill. ISBN:0072321032.

Lee, C. S., Jung, I. S., & Tchah, K. H. (1999). RF receiver sensitivity of mobile station in CDMA. In FifthAsia-Pacific conference on communications and fourth optoelectronics and communications conference,pp. 637–640.

Levy, R., Snyder, R. V., & Matthaei, G. (2002). Design of microwave filters. IEEE Transactions on MicrowaveTheory and Techniques, 50(3), 783–793.

Lin, Y. Z., Liu, C. C., Huang, G. Y., Shyu, Y. T., & Chang, S. J. (2010). A 9-bit 150-MS/s 1.53-mW subrangedSAR ADC in 90-nm CMOS. In IEEE symposium VLSI circuits, pp. 243–244.

Lin, Z., & Bruton, L. T. (1989). BIBO stability of inverse 2-D digital filters in the presence of nonessentialsingularities of the second kind. IEEE Transactions on Circuits and Systems, 36, 244–254.

Lin, Z., Bruton, L. T., & Bartley, N. R. (1988). Design of highly selective two-dimensional recursive fan filtersby relaxing symmetry constraints. Electronics Letters, 24, 1361–1362.

Liu, Z., Ruan, X., & He, J. (2013). Efficient 2-D DOA estimation for coherent sources with a sparse acousticvector-sensor array. Multidimensional Systems and Signal Processing, 24(1), 105–120.

Locher, M., Kuenen, J., Daanen, A., Visser, H., Essink, B. H., Vervoort, P. P., et al. (2008). A versatile, lowpower, high performance BiCMOS MIMO/diversity direct conversion transceiver IC for WiBro/WiMAX(802.16e). IEEE Journal of Solid-State Circuits, 43(8), 1731–1740.

Madanayake, A. (2008). Real-time FPGA architectures for frequency-planar MDSP. PhD Thesis, Universityof Calgary, Supervisor: L T Bruton.

Madanayake, A., & Bruton, L. T. (2008a). A speed-optimized systolic array processor architecture for spatio-temporal 2-D IIR broadband beam filters. IEEE Transactions on Circuits and Systems-I: Regular Papers,55(7), 1953–1966.

Madanayake, A., & Bruton, L. T. (2008b). A systolic-array architecture for first-order 3D IIR frequency-planarfilters. IEEE Transactions on Circuits and Systems-I: Regular Papers, 55(6), 1546–1559.

Madanayake, A., Hum, S. V., & Bruton, L. T. (2008). A systolic array 2D IIR broadband RF beamformer.IEEE Transactions on Circuits and Systems-II: Express Briefs, 55(12), 1244–1248.

Madanayake, A., Hum, S. V., & Bruton, L. T. (2012). Effects of quantization in systolic 2D IIR beam filterson UWB wireless communications. Circuits, Systems, and Signal Processing, 31(2), 595–610.

Pekau, H., & Haslett, J. W. (2005). A comparison of analog front end architectures for digital receivers. InProceedings of the Canadian conference on electrical and computer engineering, pp. 1073–1077.

Pekau, H., & Haslett, J. W. (2006). Cascaded noise figure calculations for radio receiver circuits with noise-aliasing properties. IEE Proceedings-Circuits, Devices and Systems, 153(6), 517–524.

Rajapaksha, N., Madanayake, A., & Bruton, L. T. (2012). 2D space-time wave-digital multi-fan filter banksfor signals consisting of multiple plane waves. Multidimensional Systems and Signal Processing, 1–23.doi:10.1007/s11045-012-0183-6.

Reddy, H. C., Khoo, I. H., Moschytz, G. S., & Stubberud, A. R. (1997). Theory and test procedure forsymmetries in the frequency response of complex two-dimensional delta operator formulated discrete-timesystems. In IEEE international symposium on circuits and systems (ISCAS).

123

Page 21: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Rizvi, U. H., Janssen, G. J. M., Weber, J. H. (2008). BER analysis of single-carrier MPAM in the presenceof ADC quantization noise. In 19th IEEE international symposium on personal, indoor and mobile radiocommunications.

Sussman-Fort, S. E. (1994). An NIC-based negative resistance circuit for microwave active filters. InternationalJournal of Microwave and Millimeter-Wave Computer-Aided Engineering, 4(2), 130–139.

Sussman-Fort, S. E., & Rudish, R. M. (2009). Non-foster impedance matching of electrically-small antennas.IEEE Transactions on Antennas and Propagation, 57, 2230–2241.

Verma, A., & Razavi, B. (2009). A 10b 500MHz 55mW CMOS ADC. InIEEE international solid-state circuitsconference, pp. 84–85.

Warnick, K. F., Woestenburg, E. E. M., Belostotski, L., & Russer, P. (2009). Minimizing the noise penalty dueto mutual coupling for a receiving array. IEEE Transactions on Antennas and Propagation, 57, 1634–1644.

Wei, Y., Zhang, X., Bai, Y., & Tang, L. (2013). A novel range alignment method for ISAR based on linear T/Rarray model. Multidimensional Systems and Signal Processing, 1–15. doi:10.1007/s11045-013-0229-4.

Wen, F., Ng, B., & Reddy, V. (2013). Extending the concept of IIR filtering to array processing using approx-imate spatial IIR structure. Multidimensional Systems and Signal Processing, 24(1), 157–179.

Wijenayake, C., Madanayake, A., & Bruton, L. T. (2012). Broadband multiple cone-beam 3D IIR digitalfilters applied to planar dense aperture arrays. IEEE Transactions on Antennas and Propagation, 60(11),5136–5146.

Yngvesson, K., Korzeniowski, T., Kim, Y. S., Kollberg, E., & Johansson, J. (1989). The tapered slot antennaa new integrated element for millimeter-wave applications. IEEE Transactions on Microwave Theory andTechniques, 37, 365–374.

Zhang, X., Chen, C., Li, J., & Xu, D. (2012). Blind DOA and polarization estimation for polarization-sensitivearray using dimension reduction MUSIC. Multidimensional Systems and Signal Processing, 1–16. doi:10.1007/s11045-012-0186-3.

Zhu, W., & Chen, B. X. (2013). Novel methods of DOA estimation based on compressed sensing. Multidi-mensional Systems and Signal Processing, 1–11. doi:10.1007/s11045-013-0239-2.

Author Biographies

Arjuna Madanayake is an assistant professor of electrical and com-puter engineering (ECE) at the University of Akron, Ohio, USA, andan adjunct assistant professor at the department of ECE, University ofCalgary, Alberta, Canada. He completed the PhD in Electrical Engi-neering, at the University of Calgary, in 2008, and was a post doctoralassociate at the Multi-Dimensional Signal Processing group and ISIS,University of Calgary, from March 2008 to November 2009. He com-pleted his Masters degree at the University of Calgary in 2004, and aBSc degree in Electronic and Telecommunication Engineering, fromthe University of Moratuwa, Sri Lanka, in 2001, with First Class Hon-ours. Arjuna has won an NSERC Post Doctoral Fellowship Award in2009 where he was chosen as the most outstanding candidate in electri-cal engineering and computing science by the NSERC selection com-mittee. The award was gratefully declined. Arjuna Madanayake alsowon the iCORE International Doctoral Award for graduate studies in2005 and the Deans Entrance Scholarship (2004) for doctoral work. DrMadanayake’s research interests include MD circuits, systems and sig-nal processing, RF antenna and active circuits, analog CMOS circuits,digital VLSI and computer architecture.

123

Page 22: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Chamith Wijenayake is a Ph.D. candidate at the department of ECEat the University of Akron. He completed his B.Sc. degree in Electronicand Telecommunication Engineering with first class honors from theUniversity of Moratuwa, Sri Lanka in 2007. Prior to doctoral studiesChamith Wijenayake was a telecommunication engineer at HUAWEItechnologies Lanka pvt. Ltd and also an instructor at the University ofMoratuwa. Chamith has won a number of awards including the out-standing graduate student research award and the Dean’s summer fel-lowship at the university of Akron. Chamith’s research interests includemultidimensional analog/digital and mixed signal processing, smartantenna array signal processing, FPGA based system design, digitalVLSI.

Rimesh Joshi is with General Motors Company, U.S. as an Elec-trical Validation Engineer. He received the M.S. degree in ElectricalEngineering from the University of Akron, Akron, OH, USA, in 2012,and the B.E. degree in Electronics and Communication Engineeringfrom Tribhuvan University, Kathmandu, Nepal, in 2008. He was withHuawei Technologies Co., Ltd. Nepal as an Applications and Soft-ware Engineer prior to the graduate studies. Rimesh’s research inter-ests include digital array signal processing, multi-dimensional digitalfiltercircuits for wireless applications, VLSI and FPGA-based digitaldesigns.

Mohammad Almalkawi received his B.Sc. degree in Telecommuni-cations Engineering (with honors) from Yarmouk University, Irbid,Jordan, in 2007. In 2009, he received his M.S. degree in ElectricalEngineering from South Dakota School of Mines & Technology, SD,USA. In 2011, he earned his Ph.D. in Engineering from the Universityof Toledo, OH, USA. He was an Intern-RF Modeling Engineer withFreescale Semiconductor, Inc., Tempe, AZ, in the summer of 2011,working on modeling RF/microwave components for the developmentof product-level models. In January of 2012, he joined the EECSDepartment at the University of Toledo, as a Post-Doctoral ResearchAssociate and he is currently a Research Assistant Professor. His cur-rent research interests include RF/microwave circuit design and model-ing, UWB antennas, computational electromagnetics, and electromag-netic compatibility. Dr. Almalkawi currently a member of the Editor-ial Board of the International Journal of RF and Microwave Computer-Aided Engineering. He is also a member of the IEEE.

123

Page 23: Electronically scanned RF-to-bits beam aperture arrays using ......achieving ultra-wideband, bandpass, as well as narrowband apertures using real-time digital signal processing methods,

Multidim Syst Sign Process

Leonid Belostotski received B.Sc. and M.Sc. degrees in electricalengineering from the University of Alberta, Edmonton, AB, Canada,in 1997 and 2000, respectively, and the Ph.D. degree from the Univer-sity of Calgary, Calgary, AB, Canada, in 2007. A large portion of hisM.Sc. thesis program was spent with the Dominion Radio Astrophys-ical Observatory, National Research Council (NRC), Penticton, BC,Canada, where he designed and prototyped a distance measurement andphase synchronization system for the Canadian Large Adaptive Reflec-tor telescope. Following his graduation, he was with Murandi Com-munications Ltd., as an RF Engineer, during which time he designeddevices for high-volume consumer applications and low-volume high-performance devices for the James Clerk Maxwell Telescope, MaunaKea, HI. He is currently an Associate Professor with the University ofCalgary, the Director of the Micro/Nano Technologies (MiNT) Labo-ratory, University of Calgary, and the President of his own engineer-ing consulting firm. His research interests include RF and mixed-signalICs, high-sensitivity receiver systems and antenna arrays, and terahertz

systems. Dr. Belostotski was the winner of the IEEE Microwave Theory and Techniques 2008MTT-11 Con-test on “Creativity and Originality in Microwave Measurements.” He was the recipient of the Analog DevicesInc. Outstanding Student Designer Award in 2007.

Len Bruton is a Professor Emeritus at the University of Calgary,Alberta, Canada. He obtained the B.Sc. from the University of Londonin 1964, the M.Eng. from Carleton University in 1967 and the Ph.D. infrom the University of Newcastle Upon Tyne in 1970, all in ElectricalEngineering. He has worked as a scientist in the telecommunicationsindustry and is currently a consultant to multimedia, signal process-ing and multimedia companies. His research interests include real-timeanalog and digital filtering with emphasis on applications in image andvideo processing as well as the directional filtering of signals in space-time. He is the inventor of the Frequency Dependent Negative Resis-tance (FDNR) class of microelectronic analog filters and the LosslessDiscrete Integrator (LDI) class of digital filters, both of which havebeen employed on a worldwide basis in telecommunication systems.

Vijay Devabhaktuni received the B.Eng. degree in EEE and the M.Sc.degree in Physics from BITS, Pilani, in 1996, and the Ph.D. in Elec-tronics from Carleton University, Canada, in 2003. He was awardedthe Carleton University’s senate medal. During 2003–2004, he held theNSERC postdoctoral fellowship and spent the tenure researching at theUniversity of Calgary. In spring of 2005, he taught at Penn State Erie.During 2005–2008, he held the prestigious Canada Research Chairin Computer-Aided High-Frequency Modeling and Design at Concor-dia University, Montreal. Since 2008, he is an Associate Professor inthe EECS Department at the University of Toledo. In 2012, he hasbeen appointed as the Director of the College of Engineering for Inter-disciplinary Research Initiatives. Dr. Devabhaktuni’s interests includeapplied electromagnetics, biomedical applications of wireless sensornetworks, computer aided design, device modeling, image process-ing, infrastructure monitoring, neural networks, optimization methods,RF/microwave devices, and virtual reality. In these areas, he secured

external funding close to $4M (sponsoring agencies include AFOSR, CFI, ODOT, NASA, NSERC, NSF,as well as industry). He co-authored more than 150 peer-reviewed papers, and is advising 15 MS/PhD stu-dents and 2 research faculty. He has won teaching excellence awards in Canada and USA. Dr. Devabhaktuniserves as the Associate Editor of the International Journal of RF and Microwave Computer-Aided Engineer-ing as well as the IEEE Microwave Magazine. He is a registered member of the Association of ProfessionalEngineers, Geologists and Geophysicists of Alberta (APEGGA) and is a Senior Member of the IEEE.

123