Digital Phased Arrays Challanges

INVITED
PAPER
Digital Phased Arrays:

Challenges and Opportunities
This paper overviews the many interrelated aspects of the digital beamforming
trade space, covering data distribution within the array, data processing,
calibration, and fixed versus adaptive beamforming.
By Caleb Fulton, Member IEEE , Mark Yeary, Fellow IEEE ,
Daniel Thompson, Student Member IEEE , John Lake, Student Member IEEE , and
Adam Mitchell, Student Member IEEE
ABSTRACT | Digital beamforming (DBF) has long been her-

alded as the next frontier in phased array technology, and
not without reason. The digitization of transmit and receive
signals at the element level opens the door to new processing
and beamforming schemes and promises to deliver maximum
flexibility and unprecedented dynamic range in large systems.
However, it is not without inherent technological risks and
practical challenges associated with the amount of data to
process and the use of less sophisticated transceivers. This
paper provides broad overviews of several interrelated as-
pects of the resulting DBF trade spaces for these systems. In
particular, emerging concepts are highlighted for the roles
and interconnection of distributed beamforming/processing Fig. 1. Three generic beamforming architectures with digitization.
for fixed and adaptive beamforming. These are then related
to digital array calibration mechanisms that can potentially
reduce the need for data- and processing-intensive beam-
forming algorithms. [1]–[3]. In phased array systems, each antenna element
is driven by circuitry that approximates relative time de-
KEYWORDS | Calibration; digital systems; phased array radar; lays between antennas selected to steer toward a partic-
phased arrays ular direction relative to the array face [4]. A beam is
produced upon signal summation using array-level elec-
tronics on receive (Rx) or through radiation on transmit
I. INTRODUCTION TO DIGITAL (Tx). The traditional phased array hardware implemen-
PHASED ARRAYS tation is shown in Fig. 1(a). This architecture may have
Phased array technology has evolved since the early active, solid-state front-end electronics (as depicted) or
1950s to become a cornerstone of modern remote sens- passive splitting/combining and element-level electronics;
ing, communication, and electronic warfare equipment in either case, the electronics are primarily analog.
Modern “digitized subarray” phased arrays with mul-
Manuscript received July 1, 2015; revised September 28, 2015; accepted
tiple transceivers [Fig. 1(b)] are providing increasing
November 11, 2015. Date of publication February 3, 2016; date of current version functionality and performance at reduced cost, size, and
February 17, 2016.
C. Fulton, M. Yeary, D. Thompson, and J. Lake are with the Advanced Radar
weight compared to their predecessors [5]. This has been
Research Center, University of Oklahoma, Norman, OK 73019 USA (e-mail: accompanied by advances in efficient front-end circuitry,
[email protected]).
A. Mitchell is with the Department of Electrical and Computer Engineering,
improvements in packaging and integration, and an in-
Ohio State University, Columbus, OH 43210 USA. creased role of digital electronics closer to the aperture
Digital Object Identifier: 10.1109/JPROC.2015.2501804 itself [6]–[8]. In this modern architecture, Tx/Rx
0018-9219 Ó 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 487
Fulton et al.: Digital Phased Arrays: Challenges and Opportunities
beamforming is accomplished with analog combiner net- Section I-C that summarizes more recent trends. The re-
works (usually passive) and phase shifters within each mainder of the paper then provides a broad overview of
subarray to approximate true time delays. Digital trans- ongoing challenges as well as emerging research opportu-
ceivers drive the combined input/output (I/O) signals of nities to improve the underlying design trade spaces as-
each subarray, in turn connecting to digital beamformers sociated with the first three bulleted items above.
that operate at the subarray level. Digitization of multi- Section II provides a basic DBF framework to facilitate a
ple subarray channels enables adaptive beamforming [9], discussion on these topics then highlights architectures
[10], space-time adaptive processing (STAP) [11], and for basic DBF implementation. Section III discusses
multiple concurrent functions, an important requirement ADBF and means by which its processing requirements
for future systems like the multifunction phased array may be reduced. Section IV then presents calibration-
radar (MPAR) [12]. To form P simultaneous beams in specific challenges as well as techniques that leverage
arbitrary directions, the phase shifters and front-end an- element-level control and digitization to help address
alog beamformers must be instantiated P times.1 Arrays them. A wideband null-steering example is presented in
with analog beamforming are inherently constrained to Section V that illustrates how the underlying I/O, pro-
the beamforming scheme imposed by the exact configu- cessing, and calibration trade spaces are all intercon-
ration of front-end beamforming electronics. nected in the design of these modern systems. Finally,
The most generically flexible phased array architec- Section VI provides concluding remarks.
ture would feature digitization and control of both the
transmitted and received signals at each antenna ele-
ment, along with the ability to cover wide bandwidths. B. Past Examples of Digital Arrays
Such a system is depicted in Fig. 1(c). Because the Several element-level digital arrays have been devel-
element-level processing and subsequent beamforming is oped in recent decades. While details on large-scale, op-
digital, it can be reconfigured and optimized for different erational digital arrays have not been published in the
applications. Ideally, this can be done without altering open literature, the CEAFAR in Australia and the
the front-end analog electronics and packaging. A fully EL/M-2248 MF-STAR in Israel have both been described
digital array may have a digital-to-analog converter in press releases as being “fully digital” [20], [21]. The
(DAC) and a waveform generator on each element FlexDAR testbed, currently under development in the
[e.g., direct digital synthesizer (DDS), as depicted in the United States by the Office of Naval Research (ONR), is
figure]. This provides inherent flexibility for new modes also a fully digital array [22]. There are several smaller
of operation beyond that of traditional beamforming, element-level digital testbeds that have been reported in
where the same “desired” signal is transmitted and/or recent years. Many of these systems were developed to
received on each element. In summary and as further explore or illustrate the potential benefits listed above.
motivation for this technology, the primary benefits of In the 1980–1990s, a few narrowband (few mega-
element-level digitization include: hertz at most) fully digital testbeds appeared for Rx [16],
• simultaneous Rx (and even Tx) beams and dy- [23], [24] and Tx [25] demonstrations, ADBF [26], and
namic digital subarray allocation (Section II); element-level calibration [27]. These early developments
• adaptive digital beamforming (ADBF) with de- are well summarized in [15]. The team in [28] later dem-
grees of freedom (DOFs) extending to the ele- onstrated the digitization of hundreds of channels for
ment level [14] (Section III); precision beamforming on a conformal array. At the
• enhanced calibration capabilities, with the po- time, card- or rack-based commercial off-the-shelf
tential for ultralow sidelobes and wideband (COTS) components were becoming available for digiti-
equalization [15], [16] (Section IV); zation and processing, leading to a program at ONR with
• multiple-input–multiple-output (MIMO) radar Lincoln Laboratory that developed a 96-element array
and communication as well as reconfigurable Tx with offline element-level equalization [29] and subse-
subapertures with independent waveforms [17]; quent DBF. In 2003, Air Force Research Laboratory
• dynamic range (DR) improvements on the order (AFRL), Lincoln Laboratory, and Lockheed Martin devel-
of 10 log10 ðNÞ dB with N transceivers, with some oped a 16-element S-band demonstrator [30] with two
metrics limited by correlation of spurious prod- intermediate frequency (IF) down-conversion stages sup-
ucts [6], [18], [19]. porting narrowband (10 MHz) operation or stretch pro-
cessing over 500 MHz, again with offline DBF. In 2003,
the German Aerospace Center developed a 5 5
A. Paper Overview Rx-only array for satellite communications featuring in-
With these benefits in mind, Section I-B highlights tegration of the radio-frequency (RF) front-end and first
several past examples of these systems, followed by IF stage on the patch antenna panel itself; small digital
1
Overlapped subarray beamforming (analog or digital) can form
modules behind this panel performed digitization and
clusters of beams more efficiently [13]. processing [31].
488 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

Numerous other highly digital demonstrators have ap-

peared in support of research into advanced radar process-
ing functions, such as the element-level MIMO testbed in
[46]. The authors of [6] in this issue are developing digital
testbeds supporting their research. As early as 1996 [47],
several testbeds with element-level digitization have been
developed to explore the practical aspects of DBF for com-
mercial wireless, more recently including MIMO capabili-
ties [48]. There are also several digitized subarray systems
with a large number of channels that are helping inform
the design choices for future fully digital arrays with simi-
lar packaging and processing challenges. The synthetic ap-
erture radar (SAR) arrays in [49] and [50] have a large
Fig. 2. Army digital array radar (DAR) project summary, showing number of digitized subarrays. The work in [51]–[53]
the original DAR demonstrator [32] (middle), its newer were precursors to the square kilometer array (radio as-
integrated transceiver panel (left), and dual polarization (8 1) tronomy), which will utilize faster and more efficient ap-
testbed [33] (right). The block diagram shows the interconnection
plication-specific integrated circuits (ASICs) for front-end
of highly integrated transceivers, front-end electronics, and a
basic DBF in individual panels.
processing and aggregation of roughly 2.5 Pb/s of data
[54]. Overall, these demonstrators and systems have all
navigated and informed the underlying trade spaces
highlighted in the rest of this paper.
The type of panel-level integration exhibited in these
demonstrators was becoming common at the time for C. Recent Trends
front-end array electronics [34], [35], motivating the addi- Unlike the Army DAR and similar systems above,
tional integration of COTS transceivers and digitizers. The many of these DBF prototypes used larger rack- and
Army Digital Array Radar (DAR) project, funded by the card-based solutions for higher DR transceivers and digi-
U.S. Army (CERDEC) as a collaboration between industry tizers with higher sample rates [55]. For fully digital ar-
and academia from ca. 2007–2011, resulted in the system rays, this approach scales poorly to larger and higher
testbed shown in Fig. 2 and detailed in [32]. This frequency arrays because of size constraints. Develop-
16-element S-band system featured panel-level integration ments like that in [56] at the AFRL with integrated SiGe
of direct-conversion transceivers, COTS DACs and analog- electronics more closely resemble the integration needed
to-digital converters (ADCs), and field-programmable gate for future systems, but they have not yet been openly
arrays (FPGAs) for real-time beamforming of a single demonstrated in large arrays. The current DARPA Arrays
20-MHz beam on both Tx and Rx. The X-band array de- at Commercial Timescales (ACT) program is seeking to
scribed in [36] for UAV sense-and-avoid applications fea- close this gap for a wide variety of applications.
tured a similar combination of highly integrated RF As digital beamforming (DBF) technologies continue
front-ends with colocated digitizing electronics. There are to evolve [57] there has also been significant interest in
recent reports of similar efforts in China [37], [38] and lowering the cost of radar and phased array components
Norway [39]. The University of Oklahoma is developing in order to open up new application spaces (e.g., weather
an 8 8 element line-replaceable unit (LRU) for an all- radar). Because system-level DR scales with the number
digital polarimetric phased array for MPAR-like applica- of elements, digital arrays may be able to use lower cost/
tions that has, per element, two 10-W T/R modules driven complexity transceivers than their traditional array coun-
by a two-channel direct-conversion transceiver with inte- terparts. Thus, as evidenced by the examples below, the
grated ADCs, DACs, and digital filtering on a single chip. It path to digitization at every element in large arrays in-
is, in a sense, a technological successor to the Army DAR. volves the use of increasingly integrated digital trans-
During the late 2000s, the ONR was developing a ceivers. Higher levels of RF integration are enabled by
larger digital array radar (DAR) with an associated lowering the number of frequency conversion stages,
Affordable Common Radar Architecture (ACRA) effort eliminating the need for off-chip components and filters.
[40], focusing in part on maturing open architecture con- Despite the general trend of decreasing relative costs of
cepts [8], [41] for handling data and control while form- digital hardware [58], there remain potential barriers to
ing multiple simultaneous beams in real time using the widespread adoption of fully digital array technologies.
FPGAs [42]. AFRL’s wideband transformational element- For example, while traditional arrays typically have on the
level array (TELA) testbed utilizes digitization down to order of 32 b of control per element, the highly integrated
the element level and/or true time delay MMICs for transceivers for potential use in digital arrays have hun-
wideband operation [43]. Other wideband benchtop test- dreds or even thousands. This leads to complex and highly
beds have been reported in [44] and [45]. application- and hardware-dependent control schemes.

There are more general practical challenges surrounding where Hm;p ð!Þ is the frequency-domain equivalent of the
the performance, packaging, cost, and synchronization/ DBF processing on the mth element for the pth beam. This
calibration of these potentially error-prone transceivers DBF process is typically implemented in a manner that is
that must be addressed. A more obvious challenge is the equivalent to a finite impulse response (FIR) form4 as
sheer amount of data that must be ingested and processed
by an array with ADCs at each of potentially thousands of
elements. These challenges lead to the trade spaces ex- M X
X K1
plored in the remainder of this paper. yp ½n ¼ am;p ½kxm ½n k (4)

m¼1 k¼0
II. DBF TRADE SPACES AND

ARCHITECTURES where xm ½n is the complex baseband equivalent sampled
The generic DBF process can be described as a frequency- data on the mth element and am;p ½k are the equivalent
dependent linear combination of the element signals, digital weights applied for beam p along time ðkÞ and
ultimately reducing a large amount of data to a smaller space ðmÞ.
number of data streams. The complex baseband equivalent There are three operations involved in guiding practi-
representation of the Rx digital signal from the mth of M cal DBF implementation details:
elements in an array can be described as • the actual DBF processing that results in the
Hm;p Xm products and subsequent aggregation
through (4) (Section II-B);
Xm ð!Þ ¼ Fm ð!; ; ÞGm ð!Þ (1) • explicit estimation of Fm and Gm errors (devia-
tions from ideal values) through calibration/
for an incident impulse from an angle pair ð; Þ; Tx op- alignment processes, and the use of this in-
eration is similar.2 The active electronics gain Gm ð!Þ cas- formation to derive the fixed portions of am;p ½k
cades with the embedded element pattern Fm ð!; ; Þ of (Section IV);
the mth antenna. This pattern includes all mutual cou- • implicit removal of subsequent residual errors
pling, finite array, and platform-related effects [4]. An im- through ADBF, the adaptive determination of N
portant example is that of a large planar array, where effective DOF weights within the am;p ½k to ulti-
most elements are several rows from the edge of the array mately form the realized Hm;p (Section III).
and patterns can be approximated as These three operations are critically interrelated and are
very system and application dependent. What follows is a
general description of the resulting trade space that sets
Fm ð!; ; Þ F0 ð!; ; Þejk0 u^ð;Þ½rm r0 (2) the stage for more specific discussions in the rest of the
paper.
Channels Gm are, in general, functions of tempera-
where F0 is the pattern of the element at location r0 ; rm is ture, frequency, and transceiver configuration (e.g., tun-
mth element location, u ^ð; Þ is the unit vector pointing able filter settings), and errors therein can be significant
toward the angle ð; Þ, and k0 ¼ ð!c þ !Þ=c is the usual in highly integrated, low-cost transceivers. Digital equali-
propagation constant near a center frequency of !c . On zation, the correction of Gm with Hm;p , is very important
the other hand, the active gain Gm ð!Þ captures all linear when wideband interference suppression is to be per-
transceiver effects [60].3 The ultimate goal of the DBF formed with a limited number of DOF with ADBF; this
process is to digitally remove nonideal aspects of Gm and is exemplified by the example in Section V. Overall
Fm in order to form P beams upon summation of the waveform quality must also be considered for many
resulting element-level signals, each with prescribed applications [62].
pattern characteristics (pointing angle, sidelobe level, null Errors in Fm can be less straightforward. For the
angle, etc.). Mathematically, this can be expressed for the large-array approximation in (2), these consist of phase
pth beam as deviations due to element location errors or slight varia-
tions in F0 due to manufacturing issues. For smaller ar-
X
M rays, variations in mutual coupling environments among
Yp ð!Þ ¼ Hm;p ð!ÞXm ð!Þ (3) the elements combine with other platform-related ef-
m¼1 fects to complicate the Fm patterns. For single-mode
(minimum scattering [64], [65]) antennas, the Fm can
2
Polarization is not included here, but can be modeled using a 2 2
matrix form for patterns [59]. Fig. 3 shows an example of subsequent be described fully with simple matrix-based equations
corrections. that capture mutual coupling effects through linear
3
The Fm Gm factorization is not always so straightforward (e.g.,
4
when there is signal leakage between transceivers), but serves here to Note that 2-D IIR examples exist [61], as well as nonlinear pro-
facilitate a generic DBF discussion. cessing [18].

Fig. 3. Digital matrix-based mutual coupling compensation illustration using the techniques in [63] on the eight-element testbed in
[33], showing improved sidelobe behavior as well as dual-polarization pattern correction in (b) relative to the pattern in (a) from
simple DBF derived from the assumptions in (2).
combinations of patterns like (2) for each digital ele- address for practical implementation; this is the topic of
ment. These can in turn be corrected with matrix-based the following sections.
processing to derive am;p [63]. More sophisticated
models are required for complex elements and for con-
A. Front-End Processing Considerations
formal arrays [66]–[68]. The important underlying fact
There are many options for practical implementation
is that these small-array effects generally require that
of (4), and processing architecture choices are highly ap-
element-level patterns be measured/known in order to
plication dependent. Martinez et al. [70] give an excep-
inform accurate pattern synthesis as part of the calibra-
tional overview of both the back-end DBF processing and
tion process [69], especially for wide bandwidths. An
data transfer or aggregation trade spaces, with examples
example of the need for—and efficacy of—these small-
mapped to smaller but generic and advanced multichan-
array techniques is shown in Fig. 3, which demonstrates
nel arrays. These topics are briefly surveyed in this
that a basic calibration that assumes the accuracy of (2)
section.
is insufficient to correct for mutual coupling effects in
Element-level DBF processing typically involves con-
this particular array.
version of ADC samples to complex baseband (I/Q) sam-
Fundamentally, errors in the correction for these
ples for subsequent DBF stages. These signal samples
channel and pattern effects will map to limits in beam-
may be generated in subbanded (frequency-domain)
forming performance, and thus there is inherently a rela-
form for applications where the signal of interest can be
tionship between beamforming, adaptation, and
logically channelized. Radio astronomy is one such appli-
calibration in pursuit of the overall DBF goal of produc-
cation, where polyphase filter banks are often employed
ing quality beams. This is illustrated further by high-
[51], [52]. The precise relationship between am;p ½k and
level examples throughout the rest of the paper.
Hm;p ð!Þ is established by such architectural choices. For
As a basic example, consider a planar array where (2)
example, for a direct-conversion implementation with
holds perfectly. Basic calibration at broadside (wherever
time-domain processing like the Army DAR [32], the I/Q
^ð; Þ ½rm r0 ¼ 0 for all elements) measures the rel-
u
samples are already complex baseband; thus, one simply
ative Gm through frequency-domain analysis of the xm ½n
has DFTfam;p ½kg ¼ Hm;p ð!Þ within the Nyquist band-
signals from (4) upon far-field illumination by a probing
width after appropriate normalization. On the other
signal. For a given beam p, am;p ½k are set to effectively
hand, an IF-sampled system with frequency-domain pro-
provide Hm;p ð!Þ ¼ Bm;p ð!Þ=Gm ð!Þ, where Bm;p ð!Þ is the
cessing may involve FFT-based implementation of the
desired overall channel response. With array tapering
convolution process in (4) in addition to prior digital
weights of Wm , one then simply has
down-conversion and decimation. In this case, the trans-
form from the am;p ½k to Hm;p ð!Þ involves frequency shift-
ing and scaling as well.
Bm;p ð!Þ ¼ Wm ejk0 u^ðp ;p Þ½rm r0 (5) Regardless of the architecture, the overall function is
to impose equivalent complex baseband channels, such
as (5), on each element’s signal, paying careful attention
to scan to ðp ; p Þ. Even in this simple and nonadaptive to the overall frequency-dependent phase in wideband
case, there are many architectural considerations to systems. This includes channel-level equalization [71],

Hierarchical beamformers [Fig. 4(d)] reduce the

number of data streams at each level of the hierarchy,
performing partial weighting and aggregation along the
way. Systolic beamformers [Fig. 4(b)] are similar, but in-
stead of aggregating data in parallel at a given “stage,”
data are sent serially down a link of nodes (or even ele-
ments), with partial beam data being aggregated along
the way to produce outputs for subsequent processing
stages. This requires that buffers of sufficient memory
depth be placed along each chain. It also increases over-
all latency, but this can be quite manageable if the links
are short. An example of such processing at the outputs
of digitized subarray was reported in [75]. Virtually every
digital array of moderate to large size known to the au-
thors uses some form of hierarchical/systolic processing
to form DFEs, including most of the larger examples
from the Introduction.
Fig. 4. Architectural building blocks for interconnection of digital Mathematically, the implementation of (4) would
front-ends (DFEs), in (a), to implement overall DBF processing.
resemble
I/Q correction [72], [73], and removal of any static local Q X

X L1 XX
R1
oscillator (LO) phase offsets [18]. This is in addition to yp ½n ¼ cq;p ½l dm;p ½rxm ½n r l (6)
q¼1 l¼1 m2Sq r¼0
the amplitude tapering and phase shifting associated
with (5) as well as any pattern corrections. It is more
computationally efficient to perform several corrections
for one stage of hierarchical or systolic processing. Here,
at once, e.g., combining true time delays (TTDs) with
the Sq are the elements in the qth digital subarray, which
equalization [74]. Again, implementation can be in the
is often simply a physical DFE but need not be. The in-
time or frequency domain. Additionally, all of these
dex r and the corresponding dm;p ½r represent element-
error-correction processes require practical calibration
level filtering (amplitude and phase manipulation) for
solutions as part of the design trade space. This design
the mth element in Sq , which may vary with beam num-
space also naturally includes physical/architectural
ber p. Similarly, the index l and the corresponding cq;p ½l
considerations.
represent filtering of the qth digital subarray output for
beam p. The processing above has a frequency domain
equivalent in the form of
B. Distributed Processing and Aggregation
Considerations
A unique aspect of fully digital arrays relative to
X
Q X
other high performance computing applications is that Yp ð!Þ ¼ Cq;p ð!Þ Dm;p ð!ÞXm ð!Þ (7)
the data sources are often tied to fixed positions on a q¼1 m2Sq
physical array lattice. Mechanical and serviceability con-
siderations lead to the physical segregation of the array
into a number of digital front-end (DFE) modules, very which maps back to the effective Hm;p ð!Þ in (3). This
much like analog subarrays. Practical DFE connectedness particular “one-stage” architecture is popular because it
strategies map to physical implementations that come in essentially replicates the analog beamforming performed
a variety of forms, as summarized in Fig. 4. While some by a digitized subarray (Fig. 1), but in an inherently flex-
of the past examples in the Introduction show that it is ible way [32], [76]. Many systems use the Dm;p for fixed
practical to build smaller arrays using a centralized DBF channel equalization and subarray steering, followed by
topology [Fig. 4(c)], most larger arrays require that large ADBF with the Cq;p acting as the DOF; this is discussed
amounts of data be aggregated in some way before being more in the next section.
sent to or between a number of subsequent processing A common nonadaptive use of this architecture is
nodes. As an example, even a 64 64 element array TTD beamforming over wide bandwidths, where the fre-
with only 100 MHz of sampling bandwidth would have quency-dependent phase of k0 in (5) is corrected. This is
on the order of 10 Tb/s of total ADC data to process performed at the element level in smaller or extremely
into what might be only a few beams of 20 Mb/s wave- wideband digital [77] or analog [78], [79] arrays to pre-
form data. vent beam smearing. In larger or moderate-bandwidth

arrays, TTD can be implemented on the digital subarray

(DFE) outputs, just like on equivalent digitized subarray
(analog) architectures. In (6) and (7), R ¼ 1 and dm;p ¼
Dm;p ð0Þ ¼ Bm;p ð0Þ (a phase shift at an RF of !c ), and
Cq;p ð!Þ or cq;p ½l is set to provide a scan angle-dependent
delay to the phase center of the qth subarray for the pth
beam. When TTD is combined with the formation of
multiple beams, one must carefully trade time versus fre-
quency domain implementation costs [53]. Either way,
Fig. 5. Distributed processing through digital connections
this provides a significant reduction in the overall pro-
between DFEs or processing nodes from Fig. 4, shown as large
cessing complexity over element-level TTD (on the order dots: (a) a folded torus interconnection example, and (b) example
of Q=M) that is traded against frequency-dependent grat- of a 3 3 spanning tree to distribute processing for nine beams
ing lobes [4]. with only six links between neighboring nodes.
Importantly, and unlike analog arrays, with
hierarchical/systolic beamforming the number of beams
can be traded against the signal bandwidth in the digital functionally equivalent folded torus. A simple illustrative
domain, with a fixed overall “beam-bandwidth” product 4 4 example is shown on the left of Fig. 5 [76]. This
remaining roughly constant at every point in the front- interconnection strategy distributes the roughly uniform
end processing chain. For multitier hierarchies, the inter- traffic needed for both fixed and adaptive beamforming
connection costs scale with the logarithm of the number [80], provides means of redundancy for bypassing any
of elements M, while data and front-end processing scales nonfunctioning DFEs, and maintains a mean path length pffiffiffiffi
roughly linearly with M. Both scale with the overall sys- between nodes that scales better than linearly, with Q
tem bandwidth. These types of considerations guide the for Q nodes.
design of any front-end DBF architecture within the A second future interconnection concept is that of
overall trade space of calibration, beamforming, and spanning trees, exemplified in Fig. 5 for a simple 3 3
adaptation. mesh configuration. The assumption is that back-end
fixed or adaptive signal processing is to be equally dis-
C. Novel Concepts for Future DBF Back-End tributed among nodes in order to implement and/or ex-
Interconnections pand upon the second summation stage of (7). This
While hierarchical and systolic topologies primarily configuration allows for the maximum number of simul-
complement front-end processing needs, back-end signal taneous beams to be processed for a given link band-
processing (including ADBF processing) is commonly width between nodes or, equivalently, to evenly
performed on a centralized architecture [Fig. 4(c)]. This distribute data from all DOF sources to a number of
produces “bottlenecks” in that all data support required fixed processing nodes [76]. In the figure, nine such
for such processes must ultimately arrive at a single loca- “trees” are mapped with only six interconnections be-
tion. For ADBF, this fundamentally requires independent tween each node. For a given maximum link bandwidth
samples from all DOF sources prior to processing. between physical nodes, this configuration provides a
Because parallel processing is a fundamental mecha- maximally flexible solution. Even with optimal front-end
nism by which modern high-performance computing sys- processing and data distribution architectures, the advent
tems continue to improve, new concepts are needed for of fully digital arrays provides a number of challenges
digital-array-specific applications to enable parallelization and opportunities for adaptive processing of these data.
of both fixed and adaptive processing while maximally uti- This is explored at a high level in Section III.
lizing the data links for any required “all-to-all” communi-
cation between distributed processing nodes. Given this,
mesh-based topologies [Fig. 4(e)] are future alternatives to III . ADAPTI VE BE AMFORMI NG
centralized approaches to the final stages of beamforming. CONCE PT S
What is needed are architectures that distribute paralleliz- Adaptive beamforming represents a mature set of array
able back-end processing tasks between processing nodes signal processing techniques that is primarily used to re-
or DFEs in Fig. 4. Two such future concepts for mesh- duce the impact of interference or clutter on the system
based beamforming are summarized in Fig. 5. in the presence of uncorrected Gm and/or Fm errors from
A regular 2-D mesh network like the one shown in Section II. It is also generally required when interference
Fig. 4 will have edge DFEs or processing nodes with directions are unknown. Adaptive techniques are gener-
fewer connections than those in the middle. A more op- ally more computationally expensive than traditional,
timal configuration would interconnect the top to bottom fixed-weight beamformers and therefore operate by adap-
and left to right, forming a torus. One way to implement tively combining a limited number of overall array
such connections without long physical links is to use a DOFs. The manner in which these combining weights

are determined and the order in which the processing Table 1 Summary of ADBF Weight Calculation Complexities
and aggregation take place are all part of a broad trade
space that considers the application’s requirements as
well as practical technology limitations. After a general
overview of the standard algorithms and processes, this
section presents the rough complexities of a few repre-
sentative ADBF schemes and describes a new technique
based on the matrix inversion lemma (MIL) that can
help reduce the underlying complexity in some cases.
where xi is the ith Ns 1 column vector of training or

A. ADBF Overview and Complexity support data; 2) matrix inversion of this covariance
As indicated in Section II, ADBF can be generalized matrix; 3) calculation of adaptive weights w ¼ S1 x w0
as processing Nd DOF within the am;p ½k to achieve the based on a nominal steering vector w0 ; and 4) applica-
overall DBF goals [9], [10]. In fully digital arrays, this tion of the weights to each of the incoming data samples
usually takes a form similar to (6) and (7), with cq;p ½l x as y ¼ wH x. There may also be signal conditioning
generically representing the adaptive weights for the pth (e.g., FFTs) required before or after this processing if
beam on the qth digital subarray output sum after fixed time domain processing is not used. Step 1) by itself
processing with the equivalent dm;p ½r within each digital takes N1 ðNc ; Ns Þ 8Nc2 Ns þ 2Nc2 FLOPs. Step 2) takes
subarray. The DOFs may also operate on frequency- N2 ðNc Þ Nc3 FLOPs to get to row-reduced form,
domain data, different pulses (e.g., STAP [81]), or followed by N3 ðNc Þ 3Nc ðNc 1Þ FLOPs for each beam
Doppler bins as well. in 3) and finally 4) N4 ðNc Þ 8Nc2 FLOPs for each beam-
While it is common to derive DOFs from the outputs data point product at the output. Typically, however, a
of fixed, digitized subarrays, a unique aspect of digital ar- QR decomposition and back-substitution is used to exe-
rays is that element-level adaptation is possible, with cute steps 1)–3) for efficiency and numerical stability
Q ¼ M. Unlike analog arrays, digital arrays can in princi- reasons [70], requiring NQR ðNc ; Ns ; PÞ 8Nc2 ðNs
ple also allow for remapping of digital subarrays to best Nc =3Þ þ 4PNc2 FLOPs in total for P output beams. These
suit a given application. Ultimately, it is the nature of ex- are summarized in Table 1 for reference; more details
pected interference that drives DOF choices from a sub- are available in [70].
set of practical subarray configurations. In general, the Traditionally, a narrowband ADBF uses Nch digital
more DOF one uses, the more data support is needed for channels and potentially Np pulses from each channel to
ADBF. However, sidelobe performance and signal-to- form the Nc signals to be adaptively combined. For
noise ratio losses improve. The work in [14] provides an example, STAP uses multiple pulses from multiple chan-
excellent overview of this trade space for digital arrays, nels to adaptively form output beams in P different direc-
concluding that element-level ADBF may not be neces- tions. For wideband applications, the effective
sary or even practical, but that performance does im- Gm ð!ÞFm ð!; ; Þ products from (1) may vary signifi-
prove with a large number (hundreds) of DOF in large cantly over the signal bandwidth. Without precise cali-
arrays. bration and front-end processing to address such
If implementing ADBF on a distributed processing variations, ADBF would require more DOF to produce
architecture like those suggested in Fig. 5, one must wideband nulls at the interference locations. These extra
consider that it requires “all-to-all” communication be- DOFs could come from Nt taps in a filter [e.g., the K
tween nodes; this may lead to “bottlenecks.” One must, taps in (4)], weighting different time delays of each
of course, also consider the overall computational com- channel such that Nc ¼ Nt Nch . ADBF would then pro-
plexity. In this section, this is presented at a high level duce output time samples for P beams. Section V de-
in terms of floating point operations (FLOPs), and com- scribes an example of this. For many such applications, it
plex signals are assumed throughout. is common to instead first perform an FFT on channel
Most adaptive beamforming algorithms can be gener- data to form Nf frequency bins, separating the wideband
alized to the following steps to process Nc signals (for ADBF process into Nf narrowband ones that are executed
Nd ¼ Nc 1 DOF) using Ns independent data support in parallel. The FFTs require N5 ðNch ; Nf Þ ¼
samples from these signals: 1) estimation of the signal 5Nch Nf log2 Nf FLOPs, which scales well with Nf . This
covariance matrix may form P separate beams as well, with Nf being
roughly equivalent to Nt above. In all cases, typically
Ns ¼ 2Nd to 5Nd , with additional samples (diagonal load-
ing) used to ensure stability. This amount of statistically
1 X Ns
Sx ¼ xi xHi (8) independent data sample support can become impractical
Ns i¼1 for a large number of DOFs and/or frequency bins.

B. MIL and Applications

The computationally expensive OðN 3 Þ QR decomposi-
tion step must be performed after all the training data
are received. If the number of DOFs increases, it be-
comes beneficial to mitigate the computational cost of
the inversion. One potential technique for this complex-
ity reduction is the application of the MIL to update pre-
vious inverse covariance estimates as data arrives [82].
If Sx ðiÞ is defined to be the estimate of the covariance
matrix using the first i samples, then the covariance esti-
mate (8) can be calculated recursively with an update
constant ks as Sx ðiÞ ¼ Sx ði 1Þþð1=ks Þxi xHi . As mentioned,
1
the MIL ðA þ UCVÞ1 ¼ A1 A1 UðC þ VA1 UÞ VA1
can ease the computational burden of covariance matrix Fig. 6. Calibration concepts: 1) external calibration equipment;
2) dedicated calibration networks; 3) probe-based calibration;
inverse calculation. An inverse of the estimate of the co- and (4) mutual-coupling-based calibration.
variance matrix can be updated by settingpffiffiffi the
ffi variables
from the MIL to
pffiffiffiffi A ¼ S x ði 1Þ, U ¼ ð1= ks Þxi , C ¼ I,
and V ¼ ð1= ks ÞxHi and inserting into the equation as
shown based estimation of Fm and Gm in (1) to derive optimal
values of the fixed (nonadaptive) DBF weights making
up the effective am;p ½k in (4). What follows is a summary
S1 H 1
x ði 1Þxi xi Sx ði 1Þ
S1 1
x ðiÞ ¼ Sx ði 1Þ 1 : (9) of the survey in [84] that highlights practical issues for
ks þ xi Sx ði 1Þxi
H
digital array calibration, followed by updates that high-
light more recent results and applications derived from
the advanced mutual coupling algorithms presented
The efficacy of this approach has been confirmed using
therein.
data from the National Weather Radar Testbed (NWRT)
phased array radar, a SPY-1A repurposed for weather sur-
A. General Considerations and Nomenclature
veillance. Meteorological data were collected using the
For all types of arrays, an important general distinc-
NWRT’s main channel and sidelobe canceler channels,
tion is whether a given calibration routine is part of the
and the MIL-based algorithm was shown to produce vir-
initial characterization of a new system (i.e., initial cali-
tually identical results to traditional techniques in adap-
bration), or if it is meant to operate when the system is
tively removing clutter [83]. Its primary advantage is in
in an operational state (i.e., in situ calibration). Initial
the efficiency of (9), with a beam weight computational
calibration based on far-field measurements in an an-
complexity as shown in Table 1 for each iteration as new
echoic chamber or outdoors requires a spacing between
data come in. It is being explored for future digital array
the array and the test antenna on the order of 2D2 = to
weather radar applications because it would enable
8D2 = for an array of maximum dimension D, depending
ADBF with a large number of digital DOF, as it scales
on the techniques used [10]; this can often be prohibi-
fundamentally with Nc2 instead of Nc3 . High DOF counts
tive. Fortunately, near-field scanning has reached a point
would lead to minimization of peak sidelobe levels [14]
where it is widely considered to be a mature process
while simultaneously minimizing ground clutter beyond
[10], [85]. Near-field probe scans allow for the calcula-
nominal sidelobe levels. Its primary disadvantage is that
tion of far-field and “back-projected” near-field patterns.
the covariance estimates are slow to change to dynamic
This a powerful tool for overall array diagnostics, espe-
clutter/interference requirements, and thus it is not ap-
cially for large arrays. Once the array has been fielded,
propriate for all applications. This is yet another example
in situ calibration often becomes critical in maintaining
of the application dependence of overall DBF trade
performance over temperature and time. For example,
spaces.
the German Aerospace Center predicts that their space-
based DBF imaging systems will require almost continu-
ous amplitude and phase corrections [86].
IV. CALIBRATI ON OF DIGITAL ARRAYS Another important distinction is whether a given
Most phased array systems require a calibration process calibration technique makes use of internal or external
for optimum performance. While the details vary, the calibration equipment; see Fig. 6. External processes
general goal is to ensure that each element’s frequency- (item 1) use equipment that is not intended to be fielded
dependent amplitudes and phases are set correctly in or- along with the array, e.g., anechoic chambers. Internal
der to achieve the desired beam pattern(s). As indicated calibration (items 2–4, discussed in Section IV-C) refers
in Section II, this ultimately consists of measurement- to processes that utilize the array hardware itself or any

dedicated calibration equipment that is fielded along details. Digital arrays usually allow for direct
with the array. measurement of element-level signals on Rx, and
can implement phase toggling techniques with
B. Practical Digital Array Calibration high precision on Tx.
Considerations In the midst of these low-level issues, one must con-
There are important practical challenges that compli- sider the higher level trend of digital arrays moving from
cate these processes in digital arrays, particularly during test racks with cabled antennas to operational platforms
initial calibration. with integrated, low-cost transceiver electronics. With
• While analog arrays can often be simply mea- this trend comes the risk of increased static and time/
sured using a network analyzer (or equivalent temperature-dependent errors in the active gains Gm
“RF-to-RF” electronics), digital transceiver-based (Section II) [92]. Given the overall goal described at the
arrays require that any external equipment be top of this section, it is clear that robust calibration pro-
phase locked to provide a true “RF-to-digital” in- cesses are themselves an enabling technology for digital
terface. This typically involves careful reference arrays in some applications. There are important addi-
clock distribution or the use of dedicated auxil- tional digital array benefits that help with these. First,
iary transceiver channels from the native digital digital arrays naturally provide more processing DOF for
array. equalization/TTD than their analog counterparts, gener-
• A typical hierarchical/systolic DBF may not be alizing to (4) or (6) and (7). With enough processing, er-
able to pass element-level data to a central pro- ror correction is limited only by stability of the
cessor in real time. Thus, for calibration it is electronics and calibration accuracies. Second, there are
common to buffer and then stream these data to a number of emerging in situ calibration techniques to
a central calibration processor. This should be maintain performance over time that are well suited for
considered early in the design process. digital arrays. This is the topic of Section IV-C.
• Multiple LOs may be used to drive the digital
transceivers to provide improvements in phase C. In Situ Calibration and Mutual-Coupling-Based
noise [87] and array flexibility. These LOs may Techniques
require phase synchronization during calibration It is increasingly common to include dedicated in-
[84]. Similarly, digital sample clock synchroniza- ternal calibration hardware on fielded phased arrays to
tion must be ensured through proper design; support calibration, as depicted in Fig. 6. The discus-
small, residual offsets can be compensated with sion that follows here focuses on emerging techniques
wideband calibration using TTD [77]. that seek to make maximal use of element-level digiti-
At the same time, all calibration routines performed zation to overcome calibration challenges using these
on fully digital arrays have potential calibration advan- techniques.
tages relative to analog arrays because the waveform in- A high-quality, embedded calibration network near
terfaces are digital. the array face could source or receive nearly identical
• Digital arrays have mathematical precision lim- signal levels at each element, as depicted as item 2 in
ited by processing (e.g., 16-b integer math), not Fig. 6. Early examples include series feed microstrip
analog phase shifting and/or time delay resolution lines [27], [93]. The TerraSAR-X satellite uses embedded
(e.g., 6 b). Though such high element-level accu- coupling to each element for in situ calibration [94],
racy is not usually required for large arrays where maintaining 0.2 dB and 2 overall element-level ac-
random errors are more tolerable, this may be curacy [95]. Both of the digital arrays in [30] and [31]
necessary for smaller arrays and for general sub- used coupled injection for receive-only in situ calibration.
array-level processing. These types of networks have been recently incorporated
• Analog arrays may need to be calibrated/ into single-chip arrays as well [96], and remain common
measured over a significant subset of their phase/ in general.
amplitude states because of attenuation-dependent While these embedded networks are effective for
phase shifts, phase shift-dependent amplitude smaller arrays on a single module or circuit board, it
shifts, etc. [88]. Digital domain signal manipula- may be difficult to realize error-free RF signal distribu-
tion obviates this need on Rx and may simplify tion over a large array. An alternative mechanism makes
the treatment of Tx electronic control nonlinear- use of probe antennas in the near field of the array to
ities (such as amplifier compression). monitor changes to array behavior (item 3). Examples in-
• Analog arrays generally only allow for direct mea- clude the fixed probe in [97], or built-in near-field scan-
surement of combined signals on either Tx or Rx. ners for initial and in situ calibration [98]. The relative
To measure individual elements, one must switch merits of these techniques were detailed in [84].
elements on and off or use phase toggling tech- The ideal digital array calibration mechanism would
niques; see [89]–[91] for examples and [84] for use the inherent hardware available within the array,

scale naturally to large arrays, and provide both in situ

and initial calibration. This suggests the use of the inher-
ent mutual coupling between antenna elements (item 4
in Fig. 6) as feedback paths to replace the aforemen-
tioned probe antennas with other array elements. Digital
arrays that have element-level transceivers naturally sup-
port this. Starting in the late 1980s, digital array testbeds
were used to explore initial calibration. Under the large-
array approximation of (2), one also has that pairs of ele-
ments sharing the same geometrical separation should
have identical mutual coupling [99], [100]. By examining
pairwise coupling comparisons across the array, trans-
ceiver error correction and even pattern prediction were
demonstrated. The same narrowband techniques were
used in [101] and [102] a decade later.
Motivated by these examples, the technique was re-
cently extended to focus on precise in situ calibration on Fig. 7. Mutual-coupling-based self-correction on the Army DAR
smaller systems where the large-array approximation is demonstrator as one array corner was heated. All data points
correspond to complex (amplitude/phase) channels measured
invalid. The resulting algorithm, summarized here from
externally for the Tx and Rx elements shown. The final plot
[72], [84], and [103], seeks to ensure that temperature- shows the digital correction resulting from an application of
and system-induced errors in these highly integrated the SVD-based algorithm.
arrays can be corrected. If a group of Tx elements in-
dexed by n transmit separately to a separate group of
Rx elements indexed by m, one can build a matrix
Kð!Þ whose entries Km;n ð!Þ contain the measured com- and receiving on nearby elements. Early implementations
plex couplings as seen through the digital transceivers. like the one in Fig. 7 were also limited by awkwardness
Subsequent identical measurements, forming the matrix in grouping the Tx and Rx elements to cover the entire
K0 ð!Þ, will reflect changes to the active transmit and array. The large-array initial calibration routine
receive channels, denoted as GTn ð!Þ and GRm ð!Þ, respec- highlighted above was recently significantly improved by
tively. The singular value decomposition (SVD) of the the work in [106]. This in turn inspired improvements to
0
matrix whose ðm; nÞth entry is Km;n ð!Þ=Km;n ð!Þ holds the in situ routine to track both high-power and low-
the errors in Gm ð!Þ and Gn ð!Þ in its left and right
R T
power sets of couplings with generalized, array-level
dominant singular vectors, respectively, to within an coupling matrices that together help overcome these
overall complex constant. These are then digitally cor- limitations. These new techniques employ the logarithm
rected. The Army DAR example shown in Fig. 7 dem- of coupling ratios, rather than the ratios themselves, and
onstrated that narrowband errors can be self-corrected are currently being applied to the conformal array in
through this basic process. Errors were kept to within [107]. Similar algorithms have emerged for performing
a few degrees and tenths of a decibel despite large calibration based on reflections from nearby targets
temperature-induced errors. [108], single-Tx versions of these algorithms for larger
Changes to antenna S-parameters and the nearby Rx arrays [109], MIMO radar returns to estimate phase
electromagnetic environment will also change the effec- errors on Tx [110], and even calibration based entirely
tive couplings K0 ð!Þ, either for mutual-coupling- or on clutter returns [111].
probe-based techniques. The work in [84] and [103] Both the initial and in situ mutual coupling tech-
showed that the sum of the nondominant singular values, niques were extended to wide bandwidths in [103].
denoted as , tended to track errors induced by such Fig. 8(b) shows an example of the initial estimation of a
changes, either random or localized (e.g., through dis- small simulated channel imperfection using this wide-
connecting elements). Fig. 8(a) shows an example of band extension of [106]. Application of the algorithm
versus the associated complex root mean square (RMS) resulted in Gm ð!Þ being estimated to within 1% (0.6 or
element errors in Gm resulting from simulated errors 0.1 dB) RMS of its nominal value for the element shown.
being added to the S-parameters of the 9 9 array from This study showed similar estimation errors on other el-
[104]. By setting a threshold on for tolerable Gm ements and for larger intrinsic channel imperfections.
errors, these techniques can provide robustness against In a real-time system, such estimators inform the effec-
temporary environmental effects, such as the flexing of a tive Hm;p ð!Þ corrections in (3). The final section exem-
radome in the wind [105]. plifies the type of processing reductions that may
These techniques are not without limitations. Re- become possible as these digital array calibration mecha-
ceiver linearity must be considered when transmitting nisms continue to mature in applicability and accuracy.

Fig. 8. Further mutual coupling calibration simulation studies on 9 9 array from [104]: (a) example of error indicator , the sum of
nondominant singular values of K0 ð!Þ, versus associated element-level complex errors after applying the in situ algorithm to a subset
of the central 5 5 group in the presence of random array S-parameter errors; (b) example of frequency response estimate error
after applying large-array initial calibration algorithm on the same 5 5 group; other element errors were of similar magnitude or
better [103].
V. WIDEBAND ADAPTIVE described in Section III-A to maintain a precise null in

BE AMFORMING AND CALIBRAT ION the presence of calibration errors in correcting for errors
TRADEOFF EXAMPLE in Gm and Fm of Section II. For the numbers here, the re-
The prospect of element-level digital equalization and sulting complexity for ADBF weight calculations would
coupling correction through advanced calibration could, be fairly high—on the order of 4 GFLOPs—while scal-
in principle, lead to new modes of operation that reduce ing roughly with ðMKÞ3 based on Table 1. A comparable
the inherent need for adaptivity. Many of the citations FFT-based frequency-domain approach would require
at the end of Section IV are attempting to do just that. 4 MFLOPs in this case, still scaling roughly with M3 K
For the practical STAP application in [112], it was shown for larger M while requiring similar data support.
that calibration leads to simplified adaptive processing A more efficient option would be to use beam-space
through a priori knowledge of the underlying signal co- processing. First, consider a TTD-processed beam steered
variance that takes errors into account. These approaches in the desired direction d using the techniques of [74]
reduce the need for many independent data samples to and [77] to derive the 32-tap FIR filter weights, taking
support large-DOF ADBF. This is important because sam- equalization and coupling compensation into account as
ple independence is often limited when discrete clutter described in Section II. Denote the response of the re-
and signal artifacts are present. In general, the extent to sulting beam when scanning to d as Fd ð!; ; d Þ, which
which calibration and initial measurements can mitigate varies with frequency ! and angle . A second beam
the pattern and transceiver effects described in Section II could, in parallel, be processed to form a beam peak in a
may be traded against the required complexity of ADBF. known desired null direction n , resulting in a response
Instead of exploring this high-level concept in a compre- of Fu ð!; ; n Þ. One could then form the following com-
hensive manner, the following demonstrates through a posite beam:
simplified but wideband example how the need for adap-
tive processing in some cases could be minimized
through precise array calibration. Ft ð!; ; d ; n Þ ¼ Fd ð!; ; d Þ Hr ð!ÞFu ð!; ; n Þ (10)
A. Overall Scenario and Adaptive Approach

Consider a situation where one wants to steer a where Hr ð!Þ ¼ Fd ð!; n ; d Þ=Fu ð!; n ; n Þ is the ratio of
M ¼ 16 element array in a desired direction d (e.g., the two aforementioned beams evaluated at the null
þ30 from broadside) while placing a null in another di- angle n . This response would be approximated through
rection n (e.g., 45 ) to minimize clutter or interfer- filtering of the Fu data stream with an Nr -tap digital
ence over a broad bandwidth. To maintain precise beam filter. This composite beam Ft achieves the simultaneous
control versus frequency, each channel has a K ¼ 32-tap beam pointing/null steering goal outlined above, and is
FIR filter for TTD and equalization. This fast space-time effectively a wideband form of a generalized sidelobe
processing could be performed with wideband ADBF as canceler [14].

This ideal processing, of course, requires that the

response Hr be precisely known, and that Nr be suffi-
ciently large. It is possible to adaptively estimate the Nr
filter taps based on data from the Fu and Fd data streams,
requiring a complexity of 40Nr3 FLOPs based on
arguments surrounding Table 1. For Nr ¼ 16, this
evaluates to a much more manageable 150 KFLOPs for
each adaptation. This type of processing could be ex-
tended to adaptively accommodate more than one dis-
crete interference source with the addition of Nn other
filters and null-steered beams, so long as the increased
number of adapted beams and resulting DOF scaling is
supported by sufficient processing resources. Impor-
tantly, it must also be supported by increasing amounts
of truly independent data samples, which are not always
available [112].
B. Calibration-Based Approach and Results

For cases where interference angles are truly dis-
crete and relatively stationary, a more ideal solution Fig. 9. Simulated pattern for a 16-element linear array in which a
beam is pointed to 30˚ with a null at —45˚ with (a) no errors and
would implement (10) using deterministic (fixed) (b) 3.3% calibrations errors; (c) 1-D cut of responses at q = —45˚
weights, eliminating ADBF altogether. This would re- with varying element-level complex errors relative to their
quire precise calibration to stabilize any changes to the nominal values.
active channels such that previous and precise knowledge
of the Gm and Fm values from Section II can be trusted
in forming and predicting the main and null-steered
beam pattern. factor. The top right figure shows the resulting overall
For reference, the calibration computational com- Ft pattern for ¼ 3.3%, and the bottom plot shows
plexity corresponding to this scenario is as little as the value of Ft at n versus ! for several values of .
75 KFLOPs using the algorithms in [103], including With the 1.0% RMS errors like those in Figs. 7 and
FFTs, matrix manipulations, and updates to channel 8, a respectable 45-dB null would be produced for
equalizations. Scaling with M would lie between M2 K this small array under realistic calibration limitations.
and M3 K, depending on how many transmit elements are At the center frequency, where the gain slope and
included for training, and does not scale with the num- time delay errors are zero, the theoretical null level
ber of nulls Nn because weights are fixed as a function of can be readily shown to be on the order of
the angles n and d . Changes to these angles only re- 10 log10 þ 10 log10 M dB [4]. For the error levels in
quire on the order of 5Nr Nn FLOPs to update the Hr the figure, this corresponds to 72, 62, 52, 42, and 32 dB,
based on stored initial manifold information from careful respectively; the results are clearly very close to this.
but one-time (initial) calibration. Such a rule of thumb is useful in mapping array-level
A concrete example of this fixed processing is shown performance to the stability of individual channels in
in Fig. 9, using the numbers above in a simulated system these arrays, in turn helping to guide the extent to which
with a 100% fractional bandwidth. The top left plot a given algorithm may rely on calibration accuracy to
shows the total response Ft from (10) as a function of an- drive down the complexity of more advanced processing
gle as well as normalized complex baseband frequency functions.
!= for the 16-element array. The beam steers towards This example has shown that sufficient calibration
d ¼ 30 at all frequencies while providing a deep can provide wideband interference suppression with a
(70 dBc) null at n ¼ 45 with Nr ¼ 16 taps imple- sufficiently stationary interference direction. Like the
menting a perfectly calibrated Hr ð!Þ. MIL algorithm in Section III-B, this has natural limits in
To explore the limits of this technique relative to dynamic and high-rank interference scenarios. Some
calibration accuracy, errors were added to the Gm ð!Þ form of direction of arrival estimation [113] would be
without correction of the Hr filter. Overall complex needed if the interference direction changes, or the
gain errors of (%) RMS at ! ¼ 0 were cascaded null(s) could be steered iteratively until interference is
with fractional time delay errors and gain slope errors minimized. Either way, the data support and processing
which, when normalized to one unit delay and 100%, requirements could be substantially reduced over that of
respectively, also had RMS values of . Overall, this traditional ADBF for certain applications. This example
provides a convenient, overall calibration error scaling simply serves to illustrate how the calibration and data

processing flexibility afforded by modern digital arrays leverage the opportunities provided by element-level digi-
creates new trade spaces in adaptive arrays. tization to mitigate errors and performance limitations as-
sociated with the use of lower cost transceiver hardware
while providing a fundamentally scalable and flexible digi-
VI . CONCLUSION tal beamforming architecture. This broad survey of the en-
Digital phased array technology promises to deliver new gineering trade spaces surrounding these systems has
levels of performance and multifunctionality in future ra- highlighted a number of specific challenges and opportuni-
dar and communication systems. The move to arrays with ties, ultimately suggesting that advanced calibration and
digitization at the element level comes with a number of networking schemes should seek to make maximal use of
practical risks associated with signal processing, data man- the inherent flexibility in these digital architectures to en-
agement, and the complexity of increasingly integrated hance system-level performance through novel calibration
electronics. The ultimate goal for future systems will be to and digital beamforming techniques. h
REFERENCES [16] L. Pettersson, M. Danestig, and [29] B. Cantrell et al., “Development of a digital
U. Sjostrom, “An experimental S-band array radar (DAR),” in Proc. IEEE Radar
[1] E. Brookner, “Recent developments and digital beamforming antenna,” in Proc. Conf., 2001, DOI: 10.1109/
future trends in phased arrays,” in Proc. IEEE Int. Symp. Phased Array Syst. Technol., NRC.2001.922970.
IEEE Int. Symp. Phased Array Syst. Technol., Oct. 1996, pp. 93–98.
Oct. 2013, pp. 43–53. [30] D. J. Rabideau, R. J. Galejs, F. G. Willwerth,
[17] E. Brookner, “MIMO radar demystified and D. S. McQueen, “An S-band digital
[2] J. M. Loomis, “Army radar requirements and where it makes sense to use,” in Proc. array radar testbed,” in Proc. IEEE Int. Symp.
for the 21st century,” in Proc. IEEE Radar IEEE Int. Symp. Phased Array Syst. Technol., Phased Array Syst. Technol., Oct. 2003,
Conf., Apr. 2007, DOI: 10.1109/RADAR. Oct. 2013, pp. 399–407. pp. 113–118.
2007.374182.
[18] B. James and C. Fulton, “Decorrelation [31] A. Dreher, N. Niklasch, F. Klefenz, and
[3] S. van den Berg and A. Tonnaer, “Evolution and mitigation of spurious products in A. Schroth, “Antenna and receiver system
of naval AESA radars,” in Proc. 8th Eur. phased arrays with direct conversion with digital beamforming for satellite
Conf. Antennas Propag., Apr. 2014, transceivers,” in Proc. IEEE MTT-S Int. navigation and communications,” IEEE
pp. 557–559. Microw. Symp. Dig., May 2015, Trans. Microw. Theory Tech., vol. 51, no. 7,
[4] R. Mailloux, Phased Array Antenna DOI: 10.1109/MWSYM.2015.7166990. pp. 1815–1821, Jul. 2003.
Handbook, 2nd ed. Boston, MA, [19] D. Rabideau, “Hybrid mitigation of [32] C. Fulton, P. Clough, V. Pai, and
USA: Artech House, 2005. distortion in digital arrays,” in Proc. IEEE W. Chappell, “A digital array radar with
[5] J. Herd and M. D. Conway, “The evolution Int. Radar Conf., May 2005, pp. 236–241. a hierarchical system architecture,” in
to modern phased array architectures,” [20] F. Catta-Pretta, “Northrop Grumman IEEE MTT-S Int. Microw. Symp. Dig.,
Proc. IEEE, vol. 104, no. 3, Mar. 2016, and CEA demonstrate scalable CEAFAR Jun. 2009, pp. 89–92.
DOI: 10.1109/JPROC.2015.2494879. next-generation phased array sensor [33] C. Fulton and W. Chappell, “Calibration
[6] S. Talisa, K. OHaver, T. Comberiate, system,” Sep. 2011. [Online]. Available: of panelized polarimetric phased array
M. Sharp, and O. Somerlock, “Benefits of http://www.irconnect.com/noc/press/pages/ radar antennas: A case study,” in Proc.
digital phased-array radars,” Proc. IEEE, news_releases.html?d=232192 IEEE Int. Symp. Phased Array Syst. Technol.,
vol. 104, no. 3, Mar. 2016, DOI: 10.1109/ [21] ASDNews, “1st deliveries of IAI multi Oct. 2010, pp. 860–867.
JPROC.2016.2515842. function surveillance & threat alert radar,” [34] J. Herd et al., “Low cost multifunction
[7] M. Sarcione and A. Puzella, “Technology Oct. 2010. [Online]. Available: http://www. phased array radar concept,” in Proc. IEEE
trends for future low cost phased arrays,” asdnews.com/mobile/news-31439/ Int. Symp. Phased Array Syst. Technol.,
in IEEE MTT-S Int. Microw. Symp. Dig., 1st_Deliveries_of_IAI_Multi_Function_ Oct. 2010, DOI: 10.1109/
May 2010, DOI: 10.1109/ Surveillance_&_Threat_Alert_Radar.htm ARRAY.2010.5613327.
MWSYM.2010.5515004. [22] M. Peck, “Raytheon takes FlexDAR [35] A. Puzella and R. Alm, “Air-cooled, active
[8] H. van Bezouwen, H. Feldle, and contract,” May 2014. [Online]. Available: transmit/receive panel array,” in Proc. IEEE
W. Holpp, “Status and trends in http://archive.c4isrnet.com/article/ Radar Conf., May 2008, DOI: 10.1109/
AESA-based radar,” in IEEE MTT-S Int. 20140505/C4ISRNET08/305050001/ RADAR.2008.4720740.
Microw. Symp. Dig., May 2010, Raytheon-takes-FlexDAR-contract [36] S. Kemkemian, M. Nouvel-Fiani, P. Cornic,
DOI: 10.1109/MWSYM.2010.5517507. [23] D. Chang, W. Klimczak, and G. Busche, and P. Garrec, “MIMO radar for sense and
[9] W. D. Wirth, Radar Techniques Using Array “An experimental digital beamforming avoid for UAV,” in Proc. IEEE Int. Symp.
Antennas. London, U.K.: IEE, 2001. array,” in Proc. Antennas Propag. Soc. Int. Phased Array Syst. Technol., Oct. 2010,
[10] A. J. Fenn, Adaptive Antennas and Phased Symp. Dig., May 2005, DOI: 10.1109/ DOI: 10.1109/ARRAY.2010.5613309.
Arrays for Radar and Communications. APS.1988.94337. [37] L. Qian and X. Wang, “A new wideband
Boston, MA, USA: Artech House, 2007. [24] J. Rose, B. Worley, and M. Lee, “Antenna digital array radar (WB-DAR) experiment
[11] W. Melvin, “A STAP overview,” IEEE patterns for prototype two-dimensional system,” in Proc. IEEE Region 10 Symp.,
Aerosp. Electron. Syst. Mag., vol. 19, no. 1, digital beamforming array,” in Proc. Antennas Apr. 2014, pp. 440–445.
pp. 19–35, Jan. 2004. Propag. Soc. Int. Symp. Dig., Jun. 1993, [38] W. Wiesbeck and L. Sit, “Radar 2020: The
[12] J. Stailey and K. Hondl, “Multifunction DOI: 10.1109/APS.1993.385490. future of radar systems,” in Proc. Int. Radar
phased array radar for aircraft and weather [25] A. Garrod, “Digital modules for phased Conf., Oct. 2014, pp. 1–6.
surveillance,” Proc. IEEE, vol. 104, no. 3, array radar,” in Proc. IEEE Int. Symp. Phased [39] F. Gulbrandsen, A. Nysaeter, and
Mar. 2016, DOI: 10.1109/ Array Syst. Technol., Oct. 1996, pp. 81–86. Y. Paichard, “Design of a wide-angle scan,
JPROC.2016.2491179. [26] L. Simonangeli and A. Agrawal, “A C-band X-band, digital array radar antenna,” in
[13] J. Herd, S. Duffy, and H. Steyskal, digital beamforming array,” in Proc. Antennas Proc. IEEE Radar Conf., May 2014,
“Design considerations and results for an Propag. Soc. Int. Symp. Dig., Jun. 1988, pp. 1373–1377.
overlapped subarray radar antenna,” in DOI: 10.1109/APS.1988.94360. [40] physorg.com, “ONR pursuing affordable
Proc. IEEE Aerosp. Conf., Mar. 2005, [27] J. Herd, “Experimental results from a common radar for surface ships,”
pp. 1087–1092. self-calibrating digital beamforming array,” Sep. 2011. [Online]. Available: http://www.
[14] M. Zatman, “Digitization requirements for in Proc. Antennas Propag. Soc. Int. Symp. Dig., physorg.com/news/2011-09-onr-pursuing-
digital radar arrays,” in Proc. IEEE Radar Jun. 1988, DOI: 10.1109/APS.1990.115127. common-radar-surface.html
Conf., 2001, pp. 163–168. [28] M. Kanno et al., “Digital beam forming for [41] S. Lucero et al., “DoD’s perspective on radar
[15] H. Steyskal, “Digital beamforming antennas: conformal active array antenna,” in Proc. open architectures,” MITRE Tech. Papers,
An introduction,” Microw. J., pp. 107–123, IEEE Int. Symp. Phased Array Syst. Technol., 2010.
Jan. 1987. Oct. 1996, pp. 37–40.

[42] R. Sexton and M. Pollock, “R&D in Navy radar,” in IEEE MTT-S Int. Microw. Symp. [77] C. Cheung, R. Shah, and M. Parker,
phased array radar research and Dig., May 2010. “Time delay digital beamforming for
development,” presented at the 2nd Nat. [60] D. Kelley and W. Stutzman, “Array wideband pulsed radar implementation,” in
Symp. Multi-Function Phased Array antenna pattern modeling methods that Proc. IEEE Int. Symp. Phased Array Syst.
Technol. Innovat. Develop. [Online]. include mutual coupling effects,” IEEE Technol., Oct. 2013, pp. 448–455.
Available: http://www.ofcm.gov Trans. Antennas Propag., vol. 41, no. 12, [78] R. Rotman and M. Tur, “Wideband phased
[43] M. Longbrake et al., “Transformational pp. 1625–1632, Dec. 1993. arrays with true time delay beamformers
element level arrays (TELA) testbed,” in [61] C. Wijenayake, A. Madanayake, challenges and progress,” in Proc. 8th Eur.
Proc. Government Microcircuit Appl. Crit. L. Belostotski, Y. Xu, and L. Bruton, Conf. Antennas Propag., Apr. 2014,
Technol. Conf., Mar. 2010. “All-pass filter-based 2-d IIR pp. 743–744.
[44] H. Mir and L. Albasha, “A low-cost filter-enhanced beamformers for AESA [79] Q. Ma, D. Leenaerts, and R. Mahmoudi,
high-performance digital radar test bed,” receivers,” IEEE Trans. Circuits Syst. I, “A 10–50 GHz true-time-delay phase
IEEE Trans. Instrum. Meas., vol. 62, no. 1, Reg. Papers, vol. 61, no. 5, pp. 1331–1342, shifter with max 3.9% delay variation,” in
pp. 221–229, Jan. 2013. May 2014. Proc. IEEE Radio Freq. Integr. Circuits
[45] Y. Zhang, Q. Bao, J. Wu, and S. Li, [62] F. Uysal, M. Yeary, N. Goodman, Symp., Jun. 2014, pp. 83–86.
“Design and implementation of wideband R. Rincon, and B. Osmanoglu, “Waveform [80] M. Mirza-Aghatabar, S. Koohi, S. Hessabi,
all digital array radar test-bed,” in Proc. design for wideband beampattern and and M. Pedram, “An empirical investigation
44th Eur. Microw. Conf., Oct. 2014, beamforming,” in Proc. IEEE Radar Conf., of mesh and torus NoC topologies under
pp. 1920–1923. May 2015, pp. 1062–1066. different routing algorithms and traffic
[46] J. Kantor and S. Davis, “Airborne GMTI [63] H. Steyskal and J. Herd, “Mutual coupling models,” in Proc. 10th Euromicro Conf. Digital
using MIMO techniques,” in Proc. IEEE compensation in small array antennas,” Syst. Design Architect. Methods Tools,
Radar Conf., May 2010, pp. 1344–1349. IEEE Trans. Antennas Propag., vol. 38, Aug. 2007, pp. 19–26.
[47] I. Chiba, “Digital beamforming (DBF) no. 12, pp. 1971–1975, Dec. 1990. [81] A. Paine, K. Homer, J. Medley, and
antenna system for mobile communications,” [64] W. Kahn and H. Kurss, “Minimum-scattering P. Richardson, “Real-time STAP hardware
in Proc. IEEE Int. Symp. Phased Array Syst. antennas,” IEEE Trans. Antennas Propag., demonstrator for airborne radar
Technol., Oct. 1996, pp. 243–248. vol. AP-13, no. 5, pp. 671–675, Sep. 1965. applications,” in Proc. Radar Conf.,
May 2008, pp. 1–5.
[48] A. Pugliell et al., “Design of energy and [65] J. Rubio and J. Izquierdo, “Relation
cost efficient massive MIMO arrays,” Proc. between the array pattern approach in [82] W. W. Hager, “Updating the inverse of a
IEEE, vol. 104, no. 3, Mar. 2016. terms of coupling coefficients and matrix,” SIAM Rev., vol. 31, no. 2,
minimum scattering antennas,” IEEE pp. 221–239, Jun. 1989.
[49] J. del Castillo, S. Sanchez, R. de Porras,
A. Pedreira, and J. Larranaga, “L-band Trans. Antennas Propag., vol. 59, no. 7, [83] J. Lake, M. Yeary, and C. Curtis,
digital array radar demonstrator for next pp. 2532–2537, Jul. 2011. “Adaptive nullforming at the National
generation multichannel SAR systems,” [66] J. Rubio, J. Izquierdo, and J. Corcoles, Weather Radar Testbed,” in Proc. IEEE
IEEE J. Sel. Top. Appl. Earth Observat. “Mutual coupling compensation matrices Int. Radar Conf., Arlington, VA, USA,
Remote Sens., 2015, DOI: 10.1109/ for transmitting and receiving arrays,” IEEE May 10–15, 2015, pp. 1072–1077.
JSTARS.2015.2430931. Trans. Antennas Propag., vol. 63, no. 2, [84] C. Fulton, “Digital array radar calibration
[50] R. F. Rincon et al., “ECOSAR: P-band pp. 839–843, Feb. 2015. and performance monitoring techniques,
digital beamforming polarimetric and single [67] L. Josefsson and P. Persson, Conformal including direct conversion and dual
pass interferometric SAR,” in Proc. IEEE Array Antenna Theory and Design. polarization architectures,” Ph.D. dissertation,
Radar Conf., May 2015, pp. 0699–0703. New York, NY, USA: Wiley-IEEE Press, Purdue Univ., West Lafayette, IN, USA,
2006. 2011.
[51] R. Armstrong, J. Hickish, K. Zarb-Adami,
and M. Jones, “Polarisation performance [68] C. Fulton, “Phase mode analysis of a [85] W. Patton and L. Yorinks, “Near-field
and calibration of the digital beamforming cylindrical polarimetric phased array alignment of phased-array antennas,” IEEE
system for 2-PAD,” in Proc. IEEE Int. Symp. antenna,” in Proc. Allerton Antenna Appl. Trans. Antennas Propag., vol. 47, no. 3,
Phased Array Syst. Technol., Oct. 2010. Symp., Sep. 2014. pp. 584–591, Mar. 1999.
[52] K. Zarb-Adami, A. Faulkner, [69] J. Quijano and G. Vecchi, “Alternating [86] F. Bordoni, M. Younis, and G. Krieger,
J. G. Bij de Vaate, G. W. Kant, and adaptive projections in antenna synthesis,” “Calibration issue in SMART synthetic
P. Picard, “Beamforming techniques for IEEE Trans. Antennas Propag., vol. 58, aperture radar based on scan-on-receive,”
large-N aperture arrays,” in Proc. IEEE no. 3, pp. 727–737, Mar. 2010. Proc. Adv. RF Sensors Earth Observat., 2009.
Int. Symp. Phased Array Syst. Technol., [70] D. Martinez, R. Bond, and M. Vai, High [87] T. M. Comberiate, J. P. Vant Hof,
Oct. 2010. Performance Embedded Computing L. B. Ruppalt, K. C. Lauritzen, and
[53] V. Khelbnikov, K. Zarb-Adami, Handbook: A Systems Perspective. S. H. Talisa, “Phase noise model for an
R. Armstrong, and M. Jones, “All-digital Boca Raton, FL, USA: CRC Press, 2008. array of combined sources using direct
wideband space-frequency beamforming digital synthesis (DDS),” in Proc. 40th
[71] L. Liou et al., “Digital wideband phased
for the SKA aperture array,” in Proc. IEEE Annual Precise Time Time Interval (PTTI)
array calibration and beamforming using
Int. Symp. Phased Array Syst. Technol., Meeting, Dec. 2008, pp. 301–314.
time reversal technique,” in Proc. IEEE Int.
Oct. 2010. Symp. Phased Array Syst. Technol., [88] J. Mulcahey and M. Sarcione, “Calibration
[54] M. Schmatz et al., “Scalable, efficient Oct. 2010, pp. 261–266. and diagnostics of the THAAD solid state
ASICS for the square kilometre array: phased array in a planar nearfield facility,”
[72] C. Fulton and W. Chappell, “Calibration
From A/D conversion to central correlation,” in Proc. IEEE Int. Symp. Phased Array Syst.
techniques for digital phased arrays,” in
in Proc. IEEE Int. Conf. Acoust. Speech Signal Technol., Oct. 1996.
Proc. IEEE Int. Conf. Microw. Commun.
Process., May 2014, pp. 7505–7509. Antennas Electron. Syst., Nov. 2009. [89] E. Lier, M. Zemlyansky, D. Purdy, and
[55] W. Weedon, “Phased array digital D. Farina, “Phased array calibration and
[73] A. Hakkarainen, J. Werner, and
beamforming hardware development at characterization based on orthogonal
M. Valkama, “RF imperfections in antenna
Applied Radar,” in Proc. IEEE Int. Symp. coding: Theory and experimental
arrays: Response analysis and widely-linear
Phased Array Syst. Technol., Oct. 2010. validation,” in Proc. IEEE Int. Symp. Phased
digital beamforming,” in Proc. IEEE Radio
Array Syst. Technol., Oct. 2010.
[56] M. Longbrake et al., “Digital beamforming Wireless Symp., Jan. 2013, pp. 187–189.
using highly integrated receiver-on-chip [90] S. Wang, H. Qi, and W. Yu, “Polarimetric
[74] Y. Yao, X. Huang, G. Wu, and K. Wei,
modules,” in Proc. IEEE Int. Symp. Phased SAR internal calibration scheme based on
“Joint equalization and fractional delay
Array Syst. Technol., Oct. 2010, T/R module orthogonal phase coding,”
filter design for wideband digital
pp. 196–201. IEEE Trans. Geosci. Remote Sens., vol. 47,
beamforming,” in Proc. IEEE Radar Conf.,
no. 12, pp. 3969–3980, Dec. 2009.
[57] W. Manqing, “Digital array radar: May 2015, pp. 0823–0827.
Technology and trends,” in Proc. IEEE [91] T. Takahashi, Y. Konishi, S. Makino,
[75] C. Heer, C. Fischer, and C. Shaefer,
CIE Int. Conf. Radar, Oct. 2011. H. Ohmine, and H. Nakaguro, “Fast
“Spaceborne SAR systems and technologies,”
measurement technique for phased array
[58] M. Russell, “Future of RF technology and in IEEE MTT-S Int. Microw. Symp. Dig.,
calibration,” IEEE Trans. Antennas Propag.,
radars,” in Proc. IEEE Radar Conf., May 2010.
vol. 56, no. 7, pp. 1888–1899, Jul. 2008.
Apr. 2007. [76] M. Harger, “A digital architecture for phased
[92] T. Nuteson, J. Stocker, J. Clark, D. Haque,
[59] C. Fulton and W. Chappell, “Calibration array radar,” M.S. thesis, Purdue Univ.,
and G. Mitchell, “Performance
of a digital phased array for polarimetric West Lafayette, IN, USA, 2012.
characterization of FPGA techniques for

calibration and beamforming in smart phased-array radar technology,” Lincoln demonstrator and its calibration,” in Proc.
antenna applications,” IEEE Trans. Lab. J., vol. 12, no. 2, 2000. IEEE Int. Conf. Microw. Commun. Antennas
Microw. Theory Tech., vol. 50, no. 12, [100] H. Shnitkin, “Rapid fast Fourier transform Electron. Syst., Oct. 2013, pp. 1–5.
pp. 3043–3051, Dec. 2002. phase alignment of an electronically [108] M. Harter, A. Ziroff, J. Hildebrandt, and
[93] K. Lee, R. Chu, and S. Liu, “A built-in scanned antenna,” in Proc. 20th Eur. T. Zwick, “Error analysis and
performance-monitoring/fault isolation and Microw. Conf., Sep. 1990. self-calibration of a digital beamforming
correction (PM/FIC) system for active [101] T. Gao, Y. Guo, J. Wang, and X. Chen, radar system,” in Proc. IEEE MTT-S Int.
phased-array antennas,” IEEE Trans. “Large active phased array antenna Conf. Microw. Intell. Mobility, Apr. 2015,
Antennas Propag., vol. 41, no. 11, calibration using MCM,” in IEEE Antennas pp. 1–4.
pp. 1530–1540, Nov. 1993. Propag. Soc. Int. Symp. Dig., 2001. [109] L. Xie et al., “Hybrid analog-digital antenna
[94] B. Brautigam, M. Schwerdt, M. Bachmann, [102] C. Shipley and D. Woods, “Mutual array with built-in image injection
and M. Stangl, “Individual T/R module coupling-based calibration of phased array calibration,” IEEE Trans. Antennas Propag.,
characterisation of the terraSAR-X active antennas,” in Proc. IEEE Int. Conf. vol. 62, no. 11, pp. 5513–5523, Nov. 2014.
phased array antenna by calibration pulse Phased Array Syst. Technol., 2000, [110] M. Cattenoz and P. Brouard, “An
sequences with orthogonal codes,” in Proc. pp. 529–532. experimental demonstration of a posteriori
IEEE Int. Conf. Geosci. Remote Sens. Symp., digital calibration of MIMO radar system,”
[103] A. Mitchell, “Coupling-based wideband
Jul. 2007. in Proc. Int. Radar Conf., Oct. 2014,
digital phased array calibration techniques,”
[95] M. Schwerdt et al., “Precise calibration M.S. thesis, Univ. Oklahoma, Norman, pp. 1–5.
techniques for complex SAR systems based OK, USA, 2014. [111] F. Robey, D. Fuhrmann, and S. Krich,
on active phased array antennas,” in Proc. “Array calibration utilizing clutter
[104] C. Fulton and W. Chappell, “A
IEEE Int. Symp. Phased Array Syst. Technol., scattering,” in Proc. IEEE 7th SP Workshop
dual-polarized patch antenna for weather
Oct. 2010. Stat. Signal Array Process., Jun. 1994,
radar applications,” in Proc. IEEE Int. Conf.
[96] O. Inac, S. Kim, D. Shin, C. Kim, and Microw. Commun. Antennas Electron. Syst., pp. 317–320.
G. Rebeiz, “Built-in self test systems for Nov. 2011, pp. 1–5. [112] J. H. Bang, W. Melvin, and A. Lanterman,
silicon-based phased arrays,” in IEEE MTT-S “Model-based clutter cancellation based on
[105] M. Lanne et al., “Wideband array antenna
Int. Microw. Symp. Dig., Jun. 2012, pp. 1–3. enhanced knowledge-aided parametric
system development,” in Proc. IEEE Int.
[97] A. Agrawal and A. Jablon, “A calibration Symp. Phased Array Syst. Technol., covariance estimation,” IEEE Trans. Aerosp.
technique for active phased array antennas,” Oct. 2010. Electron. Syst., vol. 51, no. 1, pp. 154–166,
in Proc. IEEE Int. Symp. Phased Array Syst. Jan. 2015.
[106] D. Bekers, R. van Dijk, and F. van Vliet,
Technol., Oct. 2003, pp. 223–228. [113] T. Ferreira, S. Netto, and P. Diniz, “Low
“Mutual-coupling based phased-array
[98] J. Nicolas, “In situ array antenna diagnosis calibration: A robust and versatile approach,” complexity covariance-based DOA
using microwave circular holography,” in in Proc. IEEE Int. Symp. Phased Array Syst. estimation algorithm,” in Proc. 15th Eur.
Proc. IEEE Int. Symp. Phased Array Syst. Technol., Oct. 2013, pp. 630–637. Signal Process. Conf., Sep. 2007,
Technol., Oct. 2010. pp. 100–104.
[107] C. Fulton et al., “Cylindrical polarimetric
[99] A. Fenn, D. Temme, W. Delaney, and W. phased array radar: A multi-function
Courtney, “The development of
ABOUT THE AUTHORS

Caleb Fulton (Member, IEEE) received the Mark Yeary (Fellow, IEEE) received the B.S.
B.S. and Ph.D. degrees in electrical and com- (honors), M.S., and Ph.D. degrees from the De-
puter engineering from Purdue University, West partment of Electrical Engineering, Texas A&M
Lafayette, IN, USA, in 2006 and 2011, University (TAMU), College Station, TX, USA,
respectively. in 1992, 1994, and 1999, respectively.
He is now an Assistant Professor in Electrical Since fall 2002, he has been with the School
and Computer Engineering at the Advanced of Electrical & Computer Engineering, University
Radar Research Center, University of Oklahoma, of Oklahoma (OU), Norman, OK, USA, where he
Norman, OK, USA. His work focuses on antenna was named the endowed Hudson-Torchmark
design, digital phased array calibration and com- Presidential Professor in 2011. He is also one of
pensation for transceiver errors, calibration for high-quality polarimet- the founding members of the Advanced Radar Research Center (ARRC)
ric radar measurements, integration of low-complexity transceivers at OU. His research interests are in the areas of digital signal process-
and high-power GaN devices, and advanced digital beamforming design ing (DSP) as applied to customized DSP systems and instrumentation
considerations. He is currently involved in a number of digital phased for radar systems with an emphasis on hardware prototype develop-
array research and development efforts for a variety of applications. ment and practical measurements. For instance, he was the principle
Dr. Fulton received the Purdue University Eaton Alumni Award for investigator (PI) on the multichannel receiver development for the SPY-
Design Excellence in 2009 for his work on the Army Digital Array Radar 1A antenna at the National Weather Radar Testbed (NWRT) in Norman,
(DAR) Project. He also received the Meritorious Paper Award for a OK, USA. He has served as a PI or Co-PI on grants from NASA, NSF-
summary of these efforts at the 2010 Government Microcircuit Applica- ATM, NSF-DUE, NSF-ECCS, DoD-EPSCoR, NOAA-CSTAR, NOAA-NSSL,
tions and Critical Technologies Conference. More recently, he received Raytheon, DARPA (ACT Program with Rockwell-Collins), and AFRL. He
a 2015 DARPA Young Faculty Award for his ongoing digital phased has also spent 13 summers (2002–2015) at Raytheon in or near Dallas,
array research. He is a member of the IEEE Antennas and Propagation TX, USA on a variety of radar projects. In fall 2012 and spring 2013, he
Society, the IEEE Aerospace and Electronic Systems Society, and the joined the Lincoln Laboratory, Massachusetts Institute of Technology
IEEE Microwave Theory and Techniques Society, and serves on the (MIT), Lexington, MA, USA, on sabbatical to make technical contribu-
Education Committee of the latter. tions related to one of the preliminary radar panels associated with the
national multifunction phased array radar (MPAR) effort. He is a li-
censed Professional Engineer (PE).

Daniel Thompson (Student Member, IEEE) re- Adam Mitchell (Student Member, IEEE) was
ceived the B.S. degree in computer engineering born in Bartlesville, OK, USA, in 1991. He re-
and the M.S. degree in electrical and computer ceived the B.S. and M.S. degrees in electrical en-
engineering from the University of Oklahoma gineering from the University of Oklahoma (OU),
(OU), Norman, OK, USA, in 2010 and 2011, respec- Norman, OK, USA, in 2013 and 2014, respectively.
tively, where he is currently working toward the He is currently working toward the Ph.D. degree
Ph.D. degree in electrical and computer in electrical engineering at the Ohio State Uni-
engineering. versity, Columbus, OH, USA.
He is currently a Research Assistant with the From 2012 to 2014, he was a Graduate Re-
Advanced Radar Research Center (ARRC), OU. His search Assistant for the Advanced Radar Re-
research interests include real-time, high-performance signal process- search Center (ARRC), OU, where he studied mutual-coupling-based
ing and devices, direct conversion transceivers, and phased array radar calibration methods for digital phased array radar systems. Since 2014,
calibration. he has been a Graduate Research Assistant at the ElectroScience Labo-
Mr. Thompson has received an ARRC Outstanding Student Paper ratory, Ohio State University. His current research interests are cogni-
Award in 2015. tive radar systems and cognitive radar imaging.
John Lake (Student Member, IEEE) graduated

summa cum laude from Texas A&M University
(TAMU), College Station, TX, USA, in 2012, from
a dual degree program, receiving the B.S. degree
in meteorology from the Department of Atmo-
spheric Sciences and the B.S. degree in com-
puter engineering from the Department of
Computer Science and Engineering. Currently,
he is working toward the Ph.D. degree in the
Advanced Radar Research Center (ARRC), Uni-
versity of Oklahoma (OU), Norman, OK, USA.
He has experience working as an operational meteorologist at
the National Weather Service Weather Forecast Office in Fort Worth,
TX, USA, where he served as a meteorologist intern in summer and
winter breaks from 2009 to 2011. More recently, he worked in the
Plains Elevated Convection at Night (PECAN) research project where he
helped collect radar data across the central United States. His research
interests lie in the intersection of meteorology and radar engineering,
with a special focus on spectrum management and mitigation of the ef-
fects of more stringent spectrum requirements on radar operations.
Mr. Lake is a recipient of the Centennial Fellowship at OU. As a stu-
dent at TAMU, he was a President’s Endowed Scholar from 2007 to
2011 and was on the President’s Honor Roll from 2008 to 2012.

Digital Phased Arrays Challanges

Uploaded by

Copyright:

Available Formats

Digital Phased Arrays Challanges

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Digital Phased Arrays Challanges

Uploaded by

Copyright:

Available Formats

INVITED

Digital Phased Arrays:

ABSTRACT | Digital beamforming (DBF) has long been her-

488 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

Numerous other highly digital demonstrators have ap-

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 489

plored in the remainder of this paper. yp ½n ¼ am;p ½kxm ½n  k (4)

II. DBF TRADE SPACES AND

490 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 491

Hierarchical beamformers [Fig. 4(d)] reduce the

I/Q correction [72], [73], and removal of any static local Q X

492 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

arrays, TTD can be implemented on the digital subarray

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 493

where xi is the ith Ns 1 column vector of training or

494 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

B. MIL and Applications

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 495

496 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

scale naturally to large arrays, and provide both in situ

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 497

V. WIDEBAND ADAPTIVE described in Section III-A to maintain a precise null in

A. Overall Scenario and Adaptive Approach

498 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

This ideal processing, of course, requires that the

B. Calibration-Based Approach and Results

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 499

500 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 501

ABOUT THE AUTHORS

502 Proceedings of the IEEE | Vol. 104, No. 3, March 2016

John Lake (Student Member, IEEE) graduated

Vol. 104, No. 3, March 2016 | Proceedings of the IEEE 503

You might also like

plored in the remainder of this paper. yp ½n ¼ am;p ½kxm ½n k (4)