Gebert TAES2009 PDF

I.
INTRODUCTION
Remote sensing of the Earth’s surface demands
Digital Beamforming on sensors that are capable of continuous global coverage
and, in addition, provide detailed imagery. Important
Receive: Techniques and applications are e.g., disaster management, land and
sea traffic observation, wide area surveillance, and
Optimization Strategies for environmental monitoring. Conventional synthetic
aperture radar (SAR) systems cannot meet these rising
High-Resolution Wide-Swath demands as the unambiguous swath width and the
achievable azimuth resolution pose contradicting
SAR Imaging requirements on system design [1]. A good azimuth
resolution ±az requires a high Doppler bandwidth BD
that results from a long synthetic aperture which is
illuminated by a short antenna of length La (cf. (1)).
NICOLAS GEBERT
Hence, a high pulse repetition frequency (PRF) is
GERHARD KRIEGER, Member, IEEE needed to sample the Doppler spectrum according to
ALBERTO MOREIRA, Fellow, IEEE the Nyquist criterion. In contrast, to unambiguously
German Aerospace Center (DLR) image a wide swath of width Wg on ground, a large
interval between subsequent pulses is favorable what
Synthetic Aperture Radar (SAR) is a well-proven imaging corresponds to a low PRF (cf. (2))
technique for remote sensing of the Earth. However, conventional
La v v
SAR systems are not capable of fulfilling the increasing demands ±az ¼ ¼ s ¸ s (1)
for improved spatial resolution and wider swath coverage. To 2 BD PRF
overcome these inherent limitations, several innovative techniques c
have been suggested which employ multiple receive-apertures Wg < (2)
to gather additional information along the synthetic aperture. 2 ¢ PRF ¢ sin(£i )
These digital beamforming (DBF) on receive techniques are where £i represents the incident angle, vs the sensor
reviewed with particular emphasis on the multi-aperture signal
processing in azimuth and a multi-aperture reconstruction
velocity, and c the speed of light. To obtain a figure
algorithm is presented that allows for the unambiguous recovery of merit that combines swath width and resolution, the
of the Doppler spectrum. The impact of Doppler aliasing is reciprocal of (1) is multiplied with (2). This yields
investigated and an analytic expression for the residual azimuth expression (3) that does not depend any more on
ambiguities is derived. Further, the influence of the processing on adjustable system parameters, which means that the
the signal-to-noise ratio (SNR) is analyzed, resulting in a pulse
SAR parameters underlie a trade-off, as the resolution
repetition frequency (PRF) dependent factor describing the SNR
scaling of the multi-aperture beamforming network. The focus is
can only be enhanced at the cost of a decreased swath
then turned to a complete high-resolution wide-swath SAR system width and vice versa
design example which demonstrates the intricate connection Wg c
between multi-aperture azimuth processing and the system < : (3)
architecture. In this regard, alternative processing approaches
±az 2 ¢ vs ¢ sin(£i )
are compared with the multi-aperture reconstruction algorithm. Alternative SAR imaging modes push this trade-off
In a next step, optimization strategies are discussed as pattern
tapering, prebeamshaping-on-receive, and modified processing
only further into one direction or another without
algorithms. In this context, the analytic expressions for both the resolving the underlying system-inherent constraint:
residual ambiguities and the SNR scaling factor are generalized the spotlight mode yields a high resolution, but no
to cascaded beamforming networks. The suggested techniques sufficient coverage [2] while burst modes as ScanSAR
can moreover be extended in many ways. Examples discussed [3, 4] and TOPS-SAR [5] map a wide swath but
are a combination with ScanSAR burst mode operation and the provide only a coarse resolution.
transfer to multistatic sparse array configurations.
Consequently, new system concepts are needed
Manuscript received January 29, 2007; revised May 16, 2007; to fulfill the increasing demands of future SAR
released for publication March 25, 2008. missions. The most promising concepts employ
IEEE Log No. T-AES/45/2/933006.
multi-channel SAR systems where the receiving
antenna is either split into multiple subapertures
Refereeing of this contribution was handled by V. Chen. with independent receiver channels or the receiver
This work was supported by EADS Defence Electronics, Ulm, apertures are distributed on multiple platforms
Germany. leading to a multistatic SAR. References [6]—[37]
Authors’ address: German Aerospace Center (DLR), Microwaves list the various approaches in chronological order.
and Radar Institute (HR), Muenchener Strasse 20, Wessling, 82230, All methods are based on the simultaneous reception
Germany, E-mail: ([email protected]). of the backscattered signal with mutually displaced
receiving apertures. The additional information
0018-9251/09/$25.00 °
c 2009 IEEE permits to overcome the aforementioned inherent
564 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 45, NO. 2 APRIL 2009
Authorized licensed use limited to: Deutsches Zentrum fuer Luft- und Raumfahrt. Downloaded on June 22, 2009 at 12:15 from IEEE Xplore. Restrictions apply.
SAR processing. The approach presented in
[24] and further elaborated in [30] introduces a
phase correction that is applied to the raw data to
resample the signal in azimuth, while the technique
proposed in [20] introduces an algorithm based on a
generalization of the sampling theorem that allows for
the unambiguous recovery of the azimuth spectrum
from multiple aliased subaperture signals. The method
is elaborated in several follow-on papers [27, 28,
31, 32, 36]. Finally, [33] and [34] bring up another
space-time approach for application in small satellite
constellations forming a sparse array. The basic idea
is to minimize the overall noise power by a trade-off
between a spatial filtering of the azimuth signal to
suppress Doppler ambiguities, which corresponds
to the method presented in [20], and a matched
filter.
Besides, alternative concepts not directly building
on conventional SAR are brought up, as a squinted
Fig. 1. Multi-aperture system with formation of different beams SAR configuration in combination with beamforming
(top) and corresponding block diagram with DBF on receive [8]. Further, a very general approach based on
principle (bottom). Each aperture is interpreted as individual a multi-satellite constellation forming a sparsely
channel whose signals are digitized and stored before combined
distributed radar sensor is developed and an optimum
coherently a posteriori.
way of processing in the space-time domain is derived
in [12] and [19]. Furthermore, [26] presents the “SAR
limitations of conventional SAR systems. The basic train” that consists of a multi-satellite constellation
concept of a multi-aperture SAR system is illustrated which is distributed in along-track direction and uses
in Fig. 1. Each of the N receiver channel’s signals spread spectrum waveforms for transmission.
is mixed, digitized, and stored. Then, a posteriori, The paper is organized as follows. Basic properties
digital beamforming (DBF) on receive is carried out of the spatial sampling of multi-aperture SAR systems
by a joint spatiotemporal processing of the recorded are summarized in Section II. The DBF algorithm
subaperture signals. In azimuth dimension this enables in azimuth introduced in [20] is briefly recalled in
a coherent combination of the N subsampled and Sections IIIA and IIIB. The new contributions of
hence aliased signals to a single output signal that is the paper start in Section IIIC with an illustrative
sampled with N ¢ PRF and free of aliasing. Compared interpretation of the multi-aperture processing,
with a mono-aperture system operated with PRF, the followed by a detailed analysis on how signal,
additional samples increase the effective sampling ambiguities, and noise are affected by the DBF
by a factor of N and allow either for an improved network in Sections IIID, IIIE, and IIIF, respectively.
resolution or a reduction of the PRF without an Then, in Section IV, a system design example is
increase of azimuth ambiguities, thereby enabling the presented that allows for verifying the theory derived
mapping of a wide swath. in Section III and demonstrates the potential of
In the following, a brief overview over different the multi-aperture reconstruction algorithm with
multi-aperture systems and processing strategies is respect to different performance parameters and in
given. More details can be found in the corresponding comparison with alternative techniques of processing
references. Concerning classical side-looking the azimuth signal. In a next step, error sources are
multi-aperture SAR systems, the development started identified and innovative strategies are derived in
with proposing an array antenna in elevation to Section VA. Optimization concepts as pattern tapering
suppress range ambiguous returns [9] and the idea on transmit (Section VB), prebeamshaping on receive
of splitting the antenna in azimuth direction to (Section VC), and adapted beamforming networks
reduce azimuth ambiguities [10] followed by the (Section VD) are presented and their performance
combination of both [11]. Then, the complexity of is shown. In the case of prebeamshaping on receive
the systems increased and interest turned to more and adapted beamforming networks, the theoretical
sophisticated processing strategies enabling so-called examination of residual errors, ambiguities, and the
“software-defined radars,” where the multi-aperture scaling of the signal-to-noise ratio (SNR) is extended
SAR signal is processed in azimuth, elevation, or both to the class of cascaded beamforming systems. The
dimensions [13—17, 20—23]. paper closes with a discussion containing an outlook
Regarding the azimuth dimension, [20], [24], and on further issues like ScanSAR in multi-aperture
[33] specify dedicated algorithms for multi-aperture systems and sparse array systems.
GEBERT ET AL.: DIGITAL BEAMFORMING ON RECEIVE: TECHNIQUES AND OPTIMIZATION STRATEGIES 565
II. MULTI-APERTURE SAR SIGNAL IN AZIMUTH
In the following, this paper focuses on the

characteristics and processing of the azimuth signal
of a single platform SAR system. In order to provide
a better understanding, this section recalls basic
relationships of multi-aperture systems regarding
spatial sampling in along-track dimension.
A. Effective Phase Center, Virtual Sample Position, and

Spatial Sampling in Along-Track
The phase center of an antenna represents the

“effective” position of the transmitted or received
signal and it is situated in the center of the respective
antenna or aperture. As in multi-aperture systems a
single transmit antenna is combined with a number
of receiving channels, the positions of transmitter
(Tx) and receiver (Rx) do in general not coincide.
Such systems of spatially separated transmit and
receive apertures can be approximated by a virtual
system where both coincide midway between the
Fig. 2. Aperture position and corresponding effective phase
transmitter and respective receiver and a constant center giving spatial position of sample in azimuth direction x
phase term is added. This position is called “effective (top) and resulting spatial sampling (bottom) for three subsequent
phase center” and represents the spatial position of pulses transmitted at t¡1 (dotted), t0 (solid), and t1 (dashed).
the received signal. The positions of the effective Sample spacing for single transmit pulse is determined by aperture
phase centers in azimuth of a single platform system distance ¢x while offset between samples of subsequent pulses
depends on sensor velocity vs and PRF.
are shown in Fig. 2 on the top. In this case the
spacing of the samples is equal to half of the receive
aperture spacing ¢x. If we consider now the spatial subapertures is constant and denoted by ¢x and all kj
sample positions for a train of pulses transmitted with of (4) become zero, resulting in expression (5) giving
PRF, we obtain the sampling scenario as shown in the optimum PRF
Fig. 2 on the bottom. For every transmitted pulse,
N signals are received whose positions in azimuth 2 ¢ vs
PRF = : (5)
are determined by the positions of the receivers N ¢ ¢x
with respect to the transmitting aperture. Hence, the The optimum value is referred to as “uniform”
gaps between samples received for a single pulse PRF as it fulfills the timing requirement for uniformly
are defined by the distance between the receiving distributed samples and yields a data array equivalent
channels, while the distance between two subsequent to that of a single-aperture (“monostatic”) system
pulses is determined by vs =PRF. Below, the relation operated with N ¢ PRF. Any deviation from the
between spatial sampling and the system parameters requirements imposed by (4) and (5) will result
as aperture position, sensor velocity vs , and PRF is in nonequally spaced samples along the synthetic
derived and the most interesting cases are presented. aperture (“nonuniform sampling”) and the gathered
data array does no longer correspond to a monostatic
B. Uniform Sampling signal and cannot be processed by conventional
monostatic algorithms without performance
In the optimum case all gathered samples are degradation. Hence, classic multi-aperture systems
distributed uniformly along the synthetic aperture. with split antenna in azimuth dimension underlie
To obtain these equally spaced samples in a system stringent operational constraints or require appropriate
of N sensors moving with vs , the following relation preprocessing to correct for perturbations caused by
between PRF and the along-track displacements xj nonuniformly spaced samples (cf. Section III).
of the receive apertures j = f2, : : : , Ng relative to the
position x1 of receiver 1 have to be fulfilled: C. Special Sampling Scenarios
μ ¶
2 ¢ vs j ¡ 1
xj ¡ x1 = + kj , kj 2 Z: (4) To complete this section, two special sampling
PRF N
scenarios are presented. Firstly, certain PRFs result
In a single platform system with an array antenna, in an interleave of samples originating from different
the along-track displacement between adjacent transmit pulses in a way such that again all samples
sampled uniformly which leads to a degraded
system performance as discussed later. In this case,
a further processing of the signal is required before
conventional monostatic algorithms can be applied
to focus the signal. In this section, a technique
called Multi-Aperture Reconstruction Algorithm
that is suited to process multi-aperture signals is
presented, investigated, and compared with alternative
approaches. This algorithm was first proposed in [20],
and it is based on solving a system of linear equations
Fig. 3. Spatial sampling in azimuth direction x originating from to unambiguously recover the formerly aliased
three subsequent transmit pulses separated by ¢t = PRF¡1 and azimuth spectrum even in case of a nonoptimum PRF,
emitted at t¡1 (dotted), t0 (solid), and t1 (dashed) yielding i.e., nonuniformly spaced data samples. Although
interleaved uniform sampling (top) and spatially coinciding in the following focus is turned to single platform
samples of channels 1 and 3 (bottom).
systems, the algorithm has great potential for any
multi-aperture system, be it a distributed SAR with
are spaced equally, yielding a uniform sampling of multiple satellites or single antenna systems like the
higher order (cf. Fig. 3, top). This means that only high-resolution wide-swath (HRWS) SAR [14] or the
after k pulses (5) is fulfilled and no samples coincide dual receive antenna mode of TerraSAR-X [25].
spatially, yielding further possible PRFs of uniform
sampling, PRFuni,k , given by (6). The inequality for k A. Multi-Aperture Impulse Response and Quadratic
excludes those values, where the timing requirement is Approximation
fulfilled, but spatial samples coincide
In a multi-aperture configuration consisting of
2 ¢ vs a receiver j separated by ¢xj from the transmitter,
PRFuni,k = k ¢ = k ¢ PRFuni
N ¢ ¢x the azimuth response hs,j (t) for a point-like target
(6)
at azimuth time t = 0, slant range R0 and for a
p¢N
k 6= , m 2 f1, : : : , N ¡ 1g, p, k 2 N: carrier wavelength ¸ is described by (9), for the time
m neglecting the azimuth envelope of the signal. The
A second interesting case arises when samples of phase is determined by the lengths of transmit and
different transmit pulses received by apertures j and i respective receive path which are approximated by the
coincide spatially as sketched in Fig. 3, bottom. This square roots assuming a straight sensor trajectory
condition is defined generally in (7) and simplified for
a single platform system in (8), yielding the respective hs,j (t, ¢xj )
PRFc specified by m and n. Note that the sampling · μq q ¶¸
2¼
becomes uniform if one of the respective coinciding = exp ¡j R02 + (vs ¢ t)2 + R02 + (vs ¢ t ¡ ¢xj )2 :
¸
channels is switched off
¢xi v ¢xj v (9)
+k¢ s = +l¢ s , k, l 2 N (7)
2 PRFc 2 PRFc Quadratic approximation of (9) yields (10) that
allows for relating the multi-aperture point target
2 ¢ vs n PRFuni n response to the monostatic response hs (t) by a
PRFc,n,m = ¢ = ¢ ,
¢x m N m (8) constant phase shift ¢'j = ¡¼ ¢ ¢xj2 =(2¸ ¢ R0 ) and
a time delay ¢tj = ¢xj =(2vs ), what corresponds to
m 2 f1, : : : , N ¡ 1g, n 2 N ^ n < m ¢ N: the effective phase center position midway between
transmitter and receiver as introduced in Section IIA.
III. DIGITAL BEAMFORMING ON RECEIVE: In Doppler frequency (‘f’) domain, this relation
MULTI-APERTURE RECONSTRUCTION between a monostatic impulse response Hs (f) and
ALGORITHM the multi-aperture response of channel j, Hs,j (f), is
described by the filter Hj (f) (cf. (11) and (12))
Most of the suggested systems propose to simply
operate a multi-aperture antenna in azimuth by · ¸ " #
» 4¼ ¼ ¢ ¢xj2
interleaving the samples of the different receiving hs,j (t) = exp ¡j R0 ¢ exp ¡j
channels without further specific processing steps ¸ 2¸ ¢ R0
[10, 11, 14, 18]. Consequently, the timing requirement " #
of (5) has to be fulfilled to obtain a signal that 2¼ ¢ vs2 (t ¡ (¢xj =2vs ))2
¢ exp ¡j
is equivalent to a monostatic signal, while in any ¸ R0
other case the sample positions deviate from the
ideal positions, but are treated as if the signal was = hs (t ¡ ¢tj ) ¢ ej¢'j (10)
Fig. 4. Left: Block diagram of multi-channel SAR system with “reconstruction” of N subsampled channels by filters Pj (f).
Right (zoom): Each Pj (f) consists of N bandpass filters Pjm (f) valid on a sub-band of width PRF.
Hs,j (f) = Hj (f) ¢ Hs (f) (11) of the respective channels can be used for a complete
" # reconstruction of the scene reflectivity, but for reasons
· ¸
¼ ¢ ¢xj2 ¼ ¢ ¢xj of simplicity only the quadratically approximated
Hj (f) = exp ¡j ¢ exp ¡j ¢f : system model of A is considered in the following.
2¸ ¢ R0 vs
The received signals are sampled in azimuth by
(12) PRF and hence the maximum signal bandwidth is
N ¢ PRF. A compact characterization of the whole
The multi-aperture response can hence be
system is then given by the matrix H(f), that contains
separated into the influence of the multi-aperture
all channel representations Hj (f) shifted by integer
system described by Hj (f) and the conventional SAR
multiples of the PRF according to (14)
impulse response given by Hs (f). Further, for single
2 ¢¢¢
3
platform systems the same azimuth antenna pattern H (f) 1 H (f) N
a(t) and A(f), respectively, can be assumed for all 6 H1 (f + PRF) ¢¢¢ HN (f + PRF) 7
6 7
channels. This yields the block diagram of Fig. 4, H(f) =6 7:
6 .
.. .. .
.. 7
where us (t) and Us (f), respectively, describe the scene 4 . 5
reflectivity. Filtering of the scene with A(f) ¢ Hs (f)
H1 (f + (N ¡ 1) ¢ PRF) ¢ ¢ ¢ HN (f + (N ¡ 1) ¢ PRF)
yields the monostatic SAR signal in Doppler domain,
U(f), with its time domain representation u(t). The (14)
multi-aperture SAR signal of channel j is then given Then, as shown in [39], the inversion of H(f)
by Uj (f) and uj (t), that are related to U(f) and u(t) yields a matrix P(f) that contains in its rows N
by Hj (f) according to (13) and its time domain functions Pj (f) that are decomposed by the columns in
representation hj (t), respectively N functions Pjm (f) defined on sub-bands m of width
PRF that make up the Doppler spectrum (cf. Fig. 4
Uj (f) = Hj (f) ¢ U(f): (13) and (15)). As we will see later, the scaling by N is
reasonable from an energy point of view. Note that
B. Theoretical Background in the considered case H(f) is invertible as long as
samples of different receive apertures do not coincide
The multi-aperture reconstruction algorithm is spatially
founded on a generalization of the sampling theorem
P(f) = N ¢ H¡1 (f)
according to which N independent representations of 2 P (f) 3
a signal, each subsampled at 1=Nth of the signal’s 11 P12 (f + PRF) ¢¢¢ P1N (f + (N ¡ 1)PRF)
Nyquist frequency, allow for the unambiguous 6 P (f) P22 (f + PRF) ¢¢¢ P2N (f + (N ¡ 1)PRF) 7
6 21 7
“reconstruction” of the original signal from the aliased =6
6 .. .. ..
7:
7
..
Doppler spectra of the N representations. This means 4 . . . . 5
that a band-limited signal U(f) is uniquely determined PN1 (f) PN2 (f + PRF) ¢ ¢ ¢ PNN (f + (N ¡ 1)PRF)
in terms of the responses Hj (f) of N linear systems
with input U(f), sampled at 1=Nth of the Nyquist (15)
frequency. The functions Hj (f) may be chosen in The aliased frequencies in the Doppler spectra
a quite general way, but not arbitrarily [38, 39]. of the individual channels are then suppressed and
Transferred to a multi-aperture SAR system, U(f) consequently the original signal U(f) is recovered
gives the monostatic SAR signal while the functions by filtering each of the multi-aperture channels j
Hj (f) represent the ‘channel’ between the transmitter with its appropriate “reconstruction” filter Pj (f) and
and each receiver j with respect to the monostatic subsequent coherent combination of all weighted
impulse response (cf. Fig. 4 and Section A). Note receiver channels (cf. Fig. 4). To complete Fig. 4, the
that in principle the complete multi-channel SAR conventional monostatic SAR focusing filter Pmf (f) is
signal model of (9) including the two-way patterns included.
C. Illustration of Principle
In this section, an illustrative approach to explain

the principle and limitations of the multi-aperture
reconstruction algorithm is given. Consider a system
according to Fig. 4 with exemplary N = 2 apertures
and U(f) describing a monostatic SAR signal
while Uj (f) = Hj (f) ¢ U(f) represents the signal at
receiver j. At first, assume U(f) spectrally limited
to [¡PRF, PRF]. Then, Fig. 5 on the left shows the
spectrum of the band-limited signal “seen” at the
receiver j before sampling (top) and its periodic
continuation after sampling with PRF (middle)
yielding Ũj (f), given in (16). For convenience we
introduce the notation Ujk (f) = Uj (f + k ¢ PRF) where
the index k indicates a shift by k ¢ PRF in Doppler
domain
X1 X1
Ũj (f) = Uj (f + k ¢ PRF) = Ujk (f)
k=¡1 k=¡1
1
X Fig. 5. Signal spectrum at receiver j before (top) and after
= U(f + k ¢ PRF) ¢ Hj (f + k ¢ PRF): (middle) sampling with PRF and after reconstruction (bottom)
taking into account all receivers. Left shows signal band-limited to
k=¡1
[¡PRF, PRF] that is correctly recovered while the bandwidth on
(16) the right exceeds [¡PRF, PRF] entailing erroneous contributions ek .
As can be observed, for any frequency, the
subsampled and aliased signal consists of not more
than two (in general N) shifted and superimposed system of (19) that corresponds to (15)
spectra, as not more than one spectrum (in general
P12 (f + PRF) ¢ H1 (f + PRF) ¢ U(f + PRF)
N ¡ 1 spectra) of the periodic continuation overlaps
with the original spectrum. Hence, the spectra Ũ1 (f) + P22 (f + PRF) ¢ H2 (f + PRF) ¢ U(f + PRF)
and Ũ2 (f) can be weighted and combined in such !
a way, that the component of the original spectrum = N ¢ U(f + PRF)
is recovered (cf. (17), top), while the back-folded
component is cancelled (cf. (17), bottom). With Pjm (f) P12 (f + PRF) ¢ H1 (f) ¢ U(f) + P22 (f + PRF) ¢ H2 (f) ¢ U(f)
denoting the reconstruction filter for receiver j on the !
Doppler frequency interval m (cf. Fig. 4), this requires =0 (18)
the following equations to hold in the interval I1 (cf.
H(f) ¢ P(f) = N ¢ 1 , P(f) = N ¢ H¡1 (f): (19)
Fig. 5, middle, left):
P11 (f) ¢ H1 (f) ¢ U(f) + P21 (f) ¢ H2 (f) ¢ U(f) In a next step, we analyze a scenario where
the bandwidth of the signal exceeds N ¢ PRF. For
! N = 2, an example for the spectrum of such a
= N ¢ U(f)
signal is given in Fig. 5 on the right, before (top)
P11 (f) ¢ H1 (f + PRF) ¢ U(f + PRF) + P21 (f) (17) and after (middle) sampling. In contrast to the
band-limited case, the sampled signal Ũj (f) consists
¢ H2 (f + PRF) ¢ U(f + PRF) of up to three contributions, as the spectra of the
!
periodic continuation may overlap. For the general
= 0: case, this means that more than N spectra may
coincide at a certain Doppler frequency. From the
For a uniform sampling, no processing needs to mathematical point of view this results in a linear
be applied, i.e., the samples of all signals have to system of equations that is underdetermined and
be simply interleaved and hence it seems reasonable consequently the original spectrum can in general
to set the arbitrary normalization factor equal to N, not be reconstructed exactly. As shown in [28], a
as this ensures a magnitude equal to 1 of the filters complete suppression of the contributions from the
Pjm in this special case. Setting up the corresponding shifted spectra is not achieved, because the filters
equations on the interval I2 and shifting them to I1 Pj (f) are determined as if the signal bandwidth was
yields (18) which allows for setting up the linear limited to §N ¢ PRF=2. Hence only the ambiguous
energy within the band [¡N ¢ PRF=2, N ¢ PRF=2] of energy per sample for a higher PRF is compensated
the original signal is cancelled by the reconstruction. by the fact that more samples are gathered within the
All energy outside this band is not well suppressed integration time and then combined during azimuth
and finally gives rise to ambiguous contributions ek compression. We obtain (21), giving ps,BD that
in the reconstructed signal (cf. Fig. 5 bottom right) describes the signal power after focusing with BD ,
that is specified in the following section. How the where the mean value is calculated on the original
ambiguous energy can be reduced, e.g. by selecting interval Is and a “white” scene is assumed which
appropriate sub-bands or weighting of the azimuth means that jU(f)j can be described by the signal
spectrum during the reconstruction is discussed in envelope A(f)
Section V. ps,BD = N 2 ¢ E[jU(f) ¢ rect(f=BD )j2 ]
D. Signal Power = N 2 ¢ E[(A(f) ¢ rect(f=BD ))2 ]: (21)
In the following, the signal power is defined E. Residual Reconstruction Error, Error Power, and
as the mean energy of the unambiguous signal, Azimuth Ambiguities
which is limited by the system bandwidth in azimuth
Is = N ¢ PRF and eventually further confined by Section C illustrated that energy outside the
the processed Doppler bandwidth BD · Is . To band [¡N ¢ PRF=2, N ¢ PRF=2] of the original signal
determine the signal power, at first the output signal is not cancelled by the algorithm and disturbs the
power of the reconstruction network is determined. unambiguous reconstruction of the multi-aperture
Remember that the filters Pj (f) are chosen such that SAR signal. With knowledge of the overall
the “original” signal received by a single aperture configuration and the antenna patterns, this section
U(f) is reconstructed. Recall furthermore that the derives the spectral “error” that remains from these
spectral weighting of U(f) depends only on its aliased parts after the system of reconstruction filters.
envelope defined by the joint pattern A(f) of the The further development considers the sampled
transmitter and a single receiver element. This signal at receiver j in (16) and focuses on the kth
means that we basically obtain a signal that is not continuation of the original spectrum Uj (f + k ¢ PRF) =
dependent on the reconstruction filter network up to Ujk (f). It is now of interest, how the ambiguous
a constant scaling factor of N as given in (17). This parts of U(f), i.e., the contributions outside the
factor accounts for a digital summation yielding an original frequency band [¡N ¢ PRF=2, N ¢ PRF=2],
amplitude gain increased by a factor of N with respect are folded back by the sampling. As the further
to a single channel and consequently the power is processing of Ujk (f) by the filters Pjm (f) is defined
amplified by N 2 compared with the input signal power on sub-bands m of width PRF, the original band is
ps,el . Then, the signal power ps is determined by the split into respective intervals Im = [(¡N=2 + m ¡ 1) ¢
mean squared value of N ¢ U(f) limited to the system PRF, (¡N=2 + m) ¢ PRF], where m = 1, : : : , N. The
bandwidth in azimuth IS = [¡N ¢ PRF=2, N ¢ PRF=2] ambiguous contributions of Ujk (f) are located at
(cf. (20)). The limitation to IS is expressed by the frequencies that deviate more than §N ¢ PRF=2
rectangular window function rect(f=Is ) and the from the center frequency ¡k ¢ PRF, i.e., jf + k ¢ PRFj
calculation of the mean value within this interval is > N=2 ¢ PRF. Further, only contributions within
indicated by the operator E[:]1 [¡N ¢ PRF=2, N ¢ PRF=2] are of relevance, yielding
the following expressions for the bands where the
ps = N 2 ¢ E[jU(f) ¢ rect(f=IS )j2 ] = N 2 ¢ ps,el : (20)
ambiguous parts are located:
If the signal is focused with a defined bandwidth N N
BD only this part of IS is used for the compression, PRF ¡ k ¢ PRF · f · PRF, k>0 (22)
2 2
what can be understood as an additional lowpass filter
of width BD that is applied to the signal. This filtering N N
¡ PRF · f · ¡ PRF ¡ k ¢ PRF,
is equivalent to a fixed integration time resulting in an 2 2
azimuth spectrum of width BD . For the case where k < 0: (23)
the system is operated with a constant duty-cycle,
the average transmit power per time is constant. Consequently, depending on k, certain
This means that the resulting energy after focusing sub-bands of the spectrum are not cancelled
is independent from the PRF, as e.g. the decreasing by the reconstruction algorithm and have to be
considered when determining the error. Assuming
1 Note
a symmetrical pattern, i.e., A(f) = A(¡f), the
that for deterministic signals as U(f) the operator E[:] is
identical to an integration over frequency normalized by the interval
problem is symmetrical and it is sufficient to
width, while for stochastic processes E[:] describes the expectation concentrate on k > 0 and (22). In this case,
value. As both give a measure of the power, in the following the sub-bands Im up from index m0 = maxfN ¡ k + 1, 1g
same operator symbol is used to provide a consistent notation. contribute to the error. Note that choosing m0 = 1
does not lead to wrong results as the respective the respective processed bandwidth and the associated
summands are zero due to the inverse filtering by lowpass filtering are included, yielding the AASRN,BD
Pjm (f). In a next step the processing of the signal is for the image. Therefore e§ (f) is band-limited to
taken into account by considering the weighting of the BD by a rectangular window rect(f=BD ) before the
signal Ujk (f) with the filters Pjm (f) over all sub-bands ambiguous power is calculated
Im where the signal is not properly reconstructed.
Then we sum over contributions from all receiver E[je§ (f) ¢ rect(f=BD )j2 ]
AASRN,BD = : (27)
channels j. This yields ek (f) that describes the ps,BD
spectrum of the remaining output error due to aliasing
of the input signal caused by the kth backfolded In order to obtain the response of the residual
spectrum (cf. (24) and Fig. 5 bottom right). Uj (f) ambiguity of order k in the image, the respective
ambiguous contribution ek defined by (24) has to be
is expressed by the monostatic SAR signal U(f)
multiplied by the respective Hj (f). Again, index k focused by the used SAR processor.
indicates a shift from f to f + k ¢ PRF due to the
sampling and the signal is confined to the sub-band F. SNR Scaling in Digital Beamforming Networks–©bf
Im by a rectangular window of width PRF around
a center frequency f0,m = (¡N=2 + m ¡ 1=2) ¢ PRF In this section, the influence of the DBF network
represented by rect((f ¡ f0,m )=PRF) on signal and noise power is investigated and a
parameter that characterizes the impact of a network
N X
X N μ ¶
f ¡ f0,m of digital filters on the SNR is derived. As under
ek (f) = Uk (f) ¢ Hjk (f) ¢ Pj (f) ¢ rect certain conditions the SNR is improved by the
m=m0 j=1 | {z } PRF
U (f)
| {z } processing, we abandon the term “noise figure” but
jk
Pjm (f)
talk about “SNR scaling” in beamforming networks
N X
X N and denote it by ©bf to avoid confusion with existing
= Uk (f) ¢ Hjk (f) ¢ Pjm (f): (24) terminology.
m=m0 j=1 Consider the multi-aperture system as a linear
The complete ambiguous spectrum due to aliasing system of N channels j with an input signal uj (t) and
is obtained by summation of all individual ambiguous additive white Gaussian receiver noise components
contributions after reconstruction, ek , yielding e§ (f) in nj,B (t) that are mutually uncorrelated between the
(25), where the factor 2 accounts for both signs of k channels. The corresponding spectra are given by
Uj (f) and nj,B (f) and the signal-to-noise ratio is
1
X denominated by SNRel,j . The system model of a single
e§ (f) = 2 ¢ ek (f) channel j is shown in Fig. 6. After reception, the
k=1
0 1 signal is amplified in the low-noise amplifier (LNA)
1
X N X
X N that is characterized by its power gain Gj , amplitude
=2¢ @Uk (f) ¢ Hjk (f) ¢ Pjm (f)A : gain gj , and its noise figure with respect to power (Fj )
k=1 m=m0 j=1 and amplitude (fj ). Under the realistic assumption of
a sufficiently high Gj the nj,B (f) are the dominating
(25)
sources of thermal noise and hence all other thermal
Expression (25) can be divided into the noise sources afterwards are neglected. After
characteristics of the azimuth signal and the overall amplification the signal is sampled in azimuth with a
configuration. The signal spectrum Uk (f) is defined ratio of PRF yielding a subsampled and consequently
by size and tapering of the transmit antenna and the aliased time-discrete signal. The quantization error
single receive aperture, while the second part of (25) induced by the sampling is modelled as an additive
is defined by the products Hjk (f) ¢ Pjm (f) that depend noise source nqj and assumed to be uniformly
on the relative positions of all apertures. Relating distributed and spectrally white as proposed in [40].
the ambiguous power given by the mean squared Further, we assume the granular error dominant and
amplitude of e§ (f) with the signal power of (20) consequently no SNR loss caused by clipping is
results in an equivalent azimuth ambiguity-to-signal encountered. The signal of each channel j is then
ratio for multi-aperture SAR systems (AASRN ) digitally filtered by a Doppler frequency dependent
that takes into account the weighting and coherent filter Pj (f) that is defined on a bandwidth of N ¢ PRF.
combination of the individual channels by the Finally, the outputs uout,j [k] of all N branches are
beamforming network combined coherently yielding the output signal uout [k]
E[je§ (f)j2 ] (cf. Fig. 6). As the system is linear and only additive
AASRN = : (26) noise sources are considered, signal and noise power
ps
can be analyzed separately, when investigating the
So far the AASRN was calculated for the complete influence of the reconstruction network on the SNR.
Doppler bandwidth given by N ¢ PRF. In a next step, Note that in the following, “power” is used to describe
Fig. 6. System model for single channel of multi-aperture system. After reception of Uj (f), receiver noise is added. Then RF signal is
amplified and digitized before filtering and coherent combination of all channels.
the average energy per time or frequency, respectively. The power of nj (f) can be expressed by the power
The normalisation to the respective interval width at the point of reception, pn,el,j , i.e., the power of
is included in the mathematical operator E[:] that nj,B (f), as the sampling changes only the power
represents the mean value for deterministic signals spectral density of the noise without affecting its
and the expectation value for stochastic processes. For power or spectral appearance
reasons of simplicity, considerations are carried out
in frequency domain and after sampling in Doppler pn,el,j = E[jnj,B (f)j2 ] = E[jnj (f)j2 ]: (31)
frequency domain, both designated by f. After
sampling, only Doppler frequency is considered, as Finally, we assume all subaperture elements j
reconstruction by the filters Pj (f) is carried out in to be identical, which means that we expect the
Doppler domain. same characteristics for Gj , Fj , gj , fj and thermal
First, consider how the beamforming network noise of same power pn,el,j for all elements. Further,
affects the noise power. In channel j, the thermal the assumption of spectrally white receiver noise
noise contribution after sampling is gj ¢ fj ¢ nj (f) allows for separating the noise power and its spectral
while nqj (f) accounts for the quantization noise. weighting given by the Pj (f), which yields the
After digitization, filtering with Pj (f), and coherent following expression for the system’s output noise
summation of all channels j, (28) describes the overall power pn that consists of pn,rx and pn,q
noise n(f) in the reconstructed data. The output noise 2¯ ¯2 3
power pn is then defined by the mean square value of X
N ¯X N ¯
¯ ¯
n(f) given by expression (29) pn = pn,el ¢ G ¢ F ¢ E[jPj (f)j2 ] + E 4¯ Pj (f) ¢ nqj (f)¯ 5
j=1
¯ j=1
¯
N
X | {z } | {z }
n(f) = Pj (f) ¢ (nj (f) ¢ gj ¢ fj + nqj (f)) (28) Thermal noise pn,rx Quantization noise pn,q
j=1
= pn,rx + pn,q : (32)
2
pn = E[jn(f)j ]
2¯ ¯2 3 Due to the mutual correlation of the quantization
¯X N ¯ errors in the channels j, the scaling of the error by the
6¯¯ ¯ 7
= E 4¯ Pj (f) ¢ (nj (f) ¢ gj ¢ fj + nqj (f))¯¯ 5 : network can be stronger than the amplification of the
¯ j=1 ¯ thermal noise. Generally, we expect that the number
of bits will be chosen such that the quantization noise
(29) in the output signal will be negligible compared with
the thermal noise, which means that the overall noise
As nqj and nj are uncorrelated [40], the overall
can be approximated by pn,rx defined by the input
noise power pn simplifies to the sum of the noise
noise power and an amplification factor determined
powers induced by thermal receiver noise components
by the reconstruction filter network (cf. (32)).
pn,rx,j and the quantization (pn,q ). Assuming mutually
In a next step, the influence of the reconstruction
uncorrelated nj , the squared sum representing the
network on the signal power is investigated. Taking
receiver noise power in (29) simplifies to the sum of
into account the power gain G of the LNA and
squared values (cf. (30))
2¯ ¯2 3 considering the scaling of the signal power by N 2 by
XN ¯X N ¯ the beamforming network (cf. Section D), the signal
¯ ¯
pn = E[jnj (f) ¢ Pj (f)j2 ] ¢ Gj ¢ Fj + E 4¯ Pj (f) ¢ nqj (f)¯ 5 power after reconstruction ps is given by scaled input
| {z } ¯ ¯
j=1 j=1 noise power ps,el according to (33)
Thermal noise pn,rx,j | {z }
Quantization noise pn,q
ps = ps,el ¢ G ¢ N 2 : (33)
X
N
= pn,rx,j + pn,q : (30) Combining (32) and (33), the following expression
j=1 for the scaling of the SNR by the network is obtained
[31] TABLE I
PN 2 Mission Performance Requirements
ps,el =pn,el SNRel F¢ j=1 E[jPj (f)j ]
= = : (34)
ps =pn SNRout N2 Carrier wavelength ¸ 0.031 m
Orbit height hs 500 km—700 km
To concentrate on the effect of the reconstruction Coverage (incident angle) μi 20± —55±
network, we define the SNR scaling factor ©bf of Swath width Ls 100 km
the DBF network by (34) normalized to the optimum Geometric resolution in azimuth & ±az , ±rg ·1 m
value of F=N that is achieved for uniform sampling range
PN 2
Range ambiguity suppression RASR · ¡21 dB
SNRel =SNRout j=1 E[jPj (f)j ] (distributed targets)
©bf := = : Azimuth ambiguity suppression AASRN · ¡21 dB
(SNRel =SNRout )jPRFuni N
(distributed targets)
(35) Noise equivalent sigma zero NESZ · 19 dB
Note that ©bf considers only the relation between
SNRel and SNRout at the respective PRF. This
an expression for the noise power that remains in the
includes the gain by oversampling but does not
image, the noise power spectral density (defined by
account for a possible variation of the SNRel with the
noise power and PRF) in combination with a lowpass
PRF as it is e.g. encountered for a constant duty cycle.
filter of bandwidth BD has to be considered after the
From a matrix theoretical point of view, the
signals are filtered by the Pj (f). This yields (38) for
scaling of the receiver noise power in the data as the noise power after focusing, which is derived from
quantified by (35), can be expressed by the sum of (32) extended by the rectangular window rect(f=BD )
the eigenvalues ¸j (f) of the matrix P(f) ¢ PH (f), limiting the Doppler spectrum
where PH represents the conjugate transpose of P.
N
X
For a given PRF the sum over all eigenvalues is
pn,BD = pn,el ¢ G ¢ F ¢ E[jPj (f)j2 ¢ rect(f=BD )]:
constant and consequently the SNR scaling factor
j=1
©bf is not frequency dependent. The equality of the
(38)
representations of the SNR scaling factor in (35)
and (36) is derived using the relation between the The dependency on the PRF cannot be seen
trace of a matrix and the sum over its eigenvalues in explicitly, but it follows implicitly from the restriction
combination with the Hilbert-Schmidt norm of the filters Pj (f) to BD and the definition of the
PN Pj (f) on a bandwidth of N ¢ PRF over which the
j=1 ¸j (f) mean value is calculated. Hence an increasing PRF in
©bf = : (36)
N combination with a constant BD means a decreasing
In 2006 Wang and Bao [41] have shown that spectral part–with respect to N ¢ PRF–that
contributes to the noise power. For BD = N ¢ PRF,
for the considered system and the approximations
(38) is equivalent to (34). The SNR scaling factor with
proposed in [20] the scaling of the noise by the
respect to the image ©bf,BD cannot be given explicitly,
reconstruction network as given in (34) can be
as the relation between input and output signal power
expressed explicitly by harmonic functions as follows
depends on the shape of U(f) in combination with
in (37), where the sample positions of receivers p and
BD as defined in (21), but of course ©bf,BD can be
r are described by their respective sample times tp
calculated using (21) and (38).
and tr normalised by 1=PRF. Basically the numerator Note that receiver noise and ambiguous power
gives the deviation of the samples from the optimum (cf. Section E) are mutually independent, and thus
regular spaced positions given by 2¼ ¢ n=N, while the the resulting interfering power in the data and the
denominator takes the mutual distances between the image can be simply determined by addition of the
receivers into account respective expressions.
XN
E[jPj (f)j2 ] IV. SYSTEM DESIGN EXAMPLE

j=1
PN¡1 QN A. Requirements and Timing
1X
N
n=0 r=1, r6=p
(1 ¡ cos(2¼ ¢ n=N ¡ 2¼ ¢ tr ¢ PRF))
= QN :
N
r=1, r6=p
(1 ¡ cos(2¼ ¢ PRF ¢ (tp ¡ tr ))) In the following, an X-band high-resolution
p=1
wide-swath SAR system is designed with the basic
(37) mission performance requirements of Table I and for
In a further step, we take the focusing of the two different swath definitions. First, a conventional
data into account to describe the noise in the image, approach is chosen that is suitable to cover the
pn,BD . Based on the assumption of spectrally white required incident angle range from 20± to 55± with
receiver noise, the power spectral density of the swaths of 100 km and a variable overlap of 10—20%.
noise decreases with increasing PRF while the This case takes into account the interference of
noise power remains constant [42]. Hence, to derive the transmit event and the nadir echo. The detailed
TABLE II
System Parameters
Orbit height hs 580 km § 10 km

Slant Range R0 604 km—1112 km
Sensor velocity vs 7560 m/s
PRF range PRF 1240—1470 Hz
Receive aperture size in azimuth daz,rx 1.6 m
Number of Rx channels in azimuth N 7
Overall antenna length in azimuth La 11.2 m
Transmit antenna size in azimuth daz,tx 3.0 m
Reconstruction Network SNR ©bf · 0:7 dB
Scaling Factor
Receive aperture size in elevation del,rx 0.08 m
Number of Rx apertures in elevation Nel 25
Overall antenna height in elevation Ha 2 m
Transmit antenna size in elevation del,tx 0.19 m—0.41 m
Maximum transmit antenna gain Gtx 38.7 dB—42.1 dB
Maximum gain of single receiver Grx,j 46.2 dB
channel
Transmit peak power Ptx 5 kW
Duty cycle dc 15%
Boltzmann constant k 1:38 ¢ 10¡23 J/K
System temperature T 300 K
Losses (atmospheric, system, L¢F 5.7 dB
Fig. 7. Top: Timing analysis for orbit height of 580 km suitable receiver noise, 2-way)
to cover incident angle range from 20—55± with 6 distinct swaths Azimuth loss Laz 2.7 dB
of width 100 km taking nadir echo (horizontally streaked) and
interference of transmit events (vertically streaked) into account.
Bottom: Timing analysis with nadir echo assumed to be
sufficiently suppressed by DBF in elevation. Dashed line indicates
line representing the swath’s center indicates (cf.
center of swath.
Fig. 7, bottom). When reaching the minimum PRF
of 1250 Hz, a “jump” to a PRF of 1470 Hz becomes
relation between swaths and incident angles is given necessary to guarantee continuous coverage (cf.
in Table III. Second, a system is regarded, where Fig. 7, bottom, dotted line).
the beamforming capability in elevation is sufficient To achieve the required resolution in azimuth,
for suppressing the nadir echo and consequently the simulations have shown that a processed bandwidth
timing depends only on the transmit events. In this BD = 7:6 kHz is needed. In combination with
case the coverage requirement is modified to enable the minimum PRF this determines a minimum
imaging of a 100 km swath centered on an arbitrary number of receive apertures to ensure the necessary
angle of incidence. A detailed timing analysis with effective sampling ratio on receive. In the present
a duty cycle of 15% yields a suitable orbit height case we obtain N = 7. Further, the overall antenna
of 580 km §10 km, leading to a sensor velocity of length is to be chosen to set the optimum PRF
7560 m/s. Fig. 7 on the bottom shows the result for within the required PRF range. It turns out that an
the suppressed nadir echo, while Fig. 7 on the top overall length La of 11:2 m = 7 ¢ 1:6 m fulfils the
accounts for the conventional scenario. In this case requirements concerning the SNR scaling factor of the
the small mutual shift of swaths of the same reference reconstruction network (©bf ) caused by the recovery
number is due to the variable overlap of adjacent of the nonuniform sampling. The respective value of
swaths. ©bf resulting from the PRF for each swath is listed in
Table III. An analysis of the ambiguous energy shows,
B. System Parameters that on transmit an aperture antenna of length daz,tx =
3 m and without tapering results in a sufficiently high
The timing shows that for the nominal orbit suppression (cf. Fig. 8, top). For a detailed analysis of
height a PRF range from 1250 Hz up to 1350 Hz the impact of different antenna lengths and tapering,
is sufficient to ensure the coverage of all six swaths refer to Section V. Table II summarizes the system
(cf. Fig. 7, top). If the height deviates by 10 km, parameters and dimensions.
the minimum PRF decreases to » 1240 Hz. In To ensure a slant range resolution of 1 m, a chirp
the case where the nadir is suppressed, the same bandwidth of 150 MHz is necessary and hence the
minimum PRF is chosen. To allow for the imaging appropriate bandwidth BRg for a resolution of 1 m
of an arbitrarily centered swath, the suitable PRF on ground is 150 MHz= sin(μi ), where μi denotes
decreases with increasing incident angle as the dashed the incident angle. Consequently, the required
bandwidth in range depends on the minimum μi of
the respective swath. Further, to illuminate the whole
swath for all incident angles, the effective height
del,tx of the transmit antenna is varied to guarantee
a coverage of 100 km on ground within the 3 dB
beamwidth angle μ3 dB . The value of μ3 dB in degrees
is approximated by 51 ¢ ¸=del,tx according to [43].
The selection of the respective swath can be either
achieved by an electrical steering of the transmit
beam or a mechanical steering of the antenna by a
roll-maneuver of the satellite. In the following, we
assume a mechanical steering that ensures an optimum
pointing of the antenna to the region to be imaged.
Table III lists the respective values for the necessary
chirp bandwidth, effective transmit antenna height,
and corresponding maximum antenna gain Gtx,max .
On receive, the antenna in elevation consists of a
large number of independent elements that allow for
beamforming in elevation which ensures a high gain
and a sufficient suppression of range ambiguities. For
this, one may apply the scan-on-receive (SCORE)
technique suggested in [14] and [15] that uses a
real-time beamforming to scan the reflected pulse as it
travels over the ground. Problems may occur with this
technique by topography and it might be necessary Fig. 8. Top: Simulated (solid) azimuth ambiguous energy
to apply a more sophisticated approach to cope with suppression (AASRN ) versus PRF and its prediction (diamonds)
for BD = 7:6 kHz. Bottom: AASRN achieved by reconstruction
the problems arising from height variations within algorithm (solid) compared with phase correction (dashed) [24],
the scene [44]. In order to guarantee the steering simple interleaving of samples (dotted) [10], and pattern
capability of the elevation beam to cover a swath null-steering (diamonds) [33]. PRF region of interest limited by
of 100 km, a scan angle up to §4:5± with respect to shaded areas.
antenna boresight is necessary. By scaling the solution
for a swath width of 80 km in [35], the necessary (cf. Fig. 8, top). A closer analysis of (27) reveals
element spacing in elevation to avoid grating lobes that the amplification of the ambiguous energy
is chosen to 8 cm. To obtain a sufficiently high gain outside the frequency interval [¡N=2 ¢ PRF, N=2 ¢
on receive and achieve the required low NESZ, a PRF] rises with increasing deviation from the
receiving antenna of height of 2 m is used, which optimum PRF, which is caused by the increasing
results in a number of 25 subantenna elements.
mean squared value of the filter functions Pj (f).
Table II and Table III summarize the parameters in
Next, the multi-aperture reconstruction algorithm is
elevation dimension.
compared with alternative algorithms for processing
the multi-aperture azimuth signal. As a first reference
C. System Performance: Azimuth Ambiguity
case, the samples of the multiple azimuth channels
Suppression, Resolution, NESZ
are simply interleaved yielding the output signal
In a first step, the suppression of azimuth without further processing as proposed in [10]. In a
ambiguities is simulated within the relevant PRF more complex approach, the method of [24] takes
range and compared with its analytic prediction the properties of the SAR signal into account as
in (27) derived in Section IIIE. The results show it compares the multi-aperture signal’s phase with
good agreement with a deviation of less than 0.1 dB the phase of a monostatic and uniformly sampled
TABLE III
Swaths and Respective Parameters
Swath μi BRg (μi,min ) del,tx Gtx,max PRF ©bf
1 20.0—29:0± 439 MHz 0.19 m 38.7 dB 1340 Hz ¡0.92 dB

2 27.5—35:6± 325 MHz 0.22 m 39.4 dB 1250 Hz 0.06 dB
3 34.9—41:9± 262 MHz 0.26 m 40.1 dB 1350 Hz ¡0.96 dB
4 40.9—47:0± 229 MHz 0.30 m 40.7 dB 1260 Hz ¡0.12 dB
5 46.4—51:7± 207 MHz 0.36 m 41.5 dB 1330 Hz ¡0.86 dB
6 50.3—55:0± 195 MHz 0.41 m 42.1 dB 1260 Hz ¡0.12 dB
In addition, one has to account for the azimuth
loss factor Laz that considers the decay of the joint
Tx/Rx azimuth pattern which attenuates the recorded
signal while the added noise power remains spectrally
white [45]. Hence, the SNR becomes dependent
on the Doppler frequency. Assuming a normalised
rectangular filter of bandwidth BD for focusing
in azimuth, Laz is expressed by (39), where A0 (f)
represents the normalised weighting of the Doppler
spectrum by the joint Tx/Rx azimuth antenna pattern.
A transmit antenna of 3 m and a receiving antenna of
1.6 m yield a loss of Laz = 2:7 dB
Fig. 9. Characteristic of SNR scaling factor ©bf versus PRF due
to signal reconstruction with consideration of processed Doppler BD
Laz = R B =2 : (39)
bandwidth of 7.6 kHz (solid) and for whole bandwidth (dashed). D 2
¡BD =2 jA0 (f)j df
signal. This yields a Doppler frequency-dependent Finally, the well-known expression for the NESZ
phase difference between the desired and the [1] is extended by the SNR scaling factor of the
actual signal that is corrected by the processing. reconstruction filter network ©bf,BD (cf. Section IIIF)
Finally, a space-time approach is evaluated that is which comprises a possible gain by oversampling,
based on a frequency-dependent adjustment of the but does not account for the changed input SNRel
weighting coefficients of the azimuth channels to with the PRF. As ©bf,BD is normalized to PRFuni , an
steer nulls in the joint antenna pattern to the angles additional compensation factor ©bf,NESZ = PRF=PRFuni
corresponding to the ambiguous Doppler frequencies. is required. Effectively, this considers the noise
This null-steering corresponds to a spatial filtering power at PRF with respect to PRFuni according to
of the data to suppress ambiguous frequencies in the changed number of samples. Note that all other
the azimuth signal [27, 33]. Fig. 8 on the bottom losses (system, atmospheric, etc.) and the receiver
shows that the best suppression is achieved by noise figure are summarized in the loss factor L ¢ F.
the reconstruction algorithm and the null-steering The description and values of all parameters can be
approach. As shown in Appendix C, under certain found in Table II
approximations this approach is already included in 256 ¢ ¼3 ¢ R03 (£i ) ¢ vs ¢ sin(£i )
the reconstruction algorithm. For a more detailed ¢k ¢ T ¢ Brg (£i ) ¢ ©bf,BD ©bf,NESZ ¢ L ¢ F ¢ Laz
NESZ = :
comparison of different processing methods, refer to Ptx ¢ Gtx (£i ) ¢ N ¢ Grx,j ¢ ¸3 ¢ c0 ¢ dc
[36].
Regarding the azimuth resolution, all algorithms (40)
remain below 1 m for the optimum PRF of 1350 Hz, Under consideration of the above parameters
but only the multi-channel processing approaches and the respective transmit pattern we obtain the
(phase correction, null-steering, reconstruction following characteristic of the NESZ versus ground
algorithm) provide a constant resolution over the range (cf. Fig. 10, top). In the case where an arbitrary
whole PRF range, while the resolution degrades swath of 100 km can be imaged, the variation of the
for increasing offset from the optimum PRF if no NESZ for a swath centred at a certain range is given
dedicated processing is applied and the samples are by the best and worst value within the swath (cf.
just interleaved. Fig. 7, bottom). Note that this takes into account the
Next, the noise equivalent sigma nought (NESZ) is adaptation of the chirp bandwidth and the effective
determined. In a preliminary step, the SNR scaling transmit antenna height with varying ground range.
factor ©bf , which describes the variation of the The steps in the curve result from the steps in the
SNR caused by the DBF network in dependency PRF and the corresponding SNR scaling factor of the
of the PRF, is evaluated (cf. Fig. 9). ©bf worsens reconstruction network. In both cases the NESZ is
with increasing deviation from the optimum PRF below the required ¡19 dB.
due to the rising mean square value of the Pj (f)
that tends to infinity when the PRF entails spatially V. OPTIMIZATION POTENTIALS
coinciding samples and the whole system bandwidth
in azimuth is considered. This effect is mitigated by In a further step, strategies for optimizing the
the increasing oversampling caused by rising PRF SAR system performance are developed. The first
values in combination with a constant BD entailing section identifies the main error sources, analyzes
for the image a reduced part of the input noise power their impact on the respective performance parameters
(cf. Fig. 9, solid line). The improvement for uniform and derives general strategies to minimize the errors.
sampling is directly given by 10 ¢ log[BD =(N ¢ PRF)] = Then, in the following sections, specific techniques
¡0:95 dB. according to those strategies are developed and the
nonuniform sampling of the signal due to the inverse
character of the filter system. In this context, an
investigation of the spectral properties of ©bf shows
that the degradation of the SNR for strong nonuniform
sampling with PRF values above the optimum PRF
is dominated by the spectral sub-bands m of order 1
and N, i.e., the lowest and highest sub-band within the
system bandwidth Is . Consequently, the singularity
of the noise and ambiguity scaling for spatially
coinciding samples (cf. Section IVC) is caused by
these sub-bands, while the contribution of the other
sub-bands is not critical. This seems logical from an
information theoretical point of view as the remaining
samples when skipping the coinciding channels
still yield an effective sampling ratio that fulfils the
Nyquist criterion with respect to the bandwidth of the
inner sub-bands.
2) Optimization Potential: The above analysis
of error sources yields two main areas of possible
optimization. First, the minimization of ambiguous
energy in the received signal allows for reducing the
ambiguities in the image. This means that antenna
patterns on transmit and receive are adapted to
optimally confine the desired Doppler band, which
Fig. 10. Top: NESZ versus ground range for coverage by 6 is the subject of a detailed analysis following in
distinct swaths. Bottom: Best (solid) and worst (dotted) NESZ for
Section B that introduces the idea of tapering on
arbitrarily located swaths. Ground range gives pointing location of
transmit antenna beam. transmit, while Section C includes the adaptation
of the receive characteristic. Besides, the azimuth
results of the optimized system are presented. As processing shows potential for improving the
main performance parameters to be optimized, we performance by minimizing the error amplification
focus on the signal-to-noise ratio, suppression of caused by the reconstruction filters. In a first simple
ambiguous energy, and geometric resolution. The approach, the ratio between PRF and BD can be set in
optimization aims at improving the SNR and the a way to benefit from a large oversampling and allow
ambiguity suppression while keeping the resolution for eliminating frequency bands that primarily cause
at a constant level. the degradation of the performance. Basically this
corresponds to an appropriate lowpass filtering of the
A. Error Sources and Optimization Approaches signal to achieve acceptably moderate levels of SNR
degradation even near to singular PRFs. According
1) Error Sources: In Section IIIE it was presented to the first paragraph it is necessary to filter the n
that the ambiguous energy in the signal after outermost bands at the lower and upper border of the
reconstruction is determined by two aspects. First (as Doppler spectrum to suppress completely the strong
in any conventional SAR system) the joint antenna rise in noise scaling at the nth singular PRF. Hence,
characteristic of transmit and single receive aperture the sampling ratio N ¢ PRF has to exceed the processed
contains information from azimuth angles that bandwidth BD by a factor of N=(N ¡ 2 ¢ n) to keep the
correspond to Doppler frequencies outside the band SNR degradation low:
Is = [¡PRF ¢ N=2, PRF ¢ N=2]. These contributions
BD · (N ¡ 2 ¢ n) ¢ PRF: (41)
are subsampled and give rise to azimuth ambiguities
in the image. In addition, the ambiguous energy is Although a very effective way of reducing the
weighted and possibly amplified by the power spectral SNR degradation, this requires a relatively high PRF
densities of the reconstruction filter functions Pjm (f). to guarantee the necessary oversampling with respect
In general, this amplification is more severe, the to the processed bandwidth BD . This might cause
stronger the nonuniform sampling of the signal is. problems with the timing. Simply decreasing BD
Concerning the noise power, Section IIIF showed that for a given PRF represents the easier way to ensure
the processing by the reconstruction filter functions (41), but is at the cost of a deteriorated geometric
may cause a degradation of the SNR, as the filters resolution. Nevertheless, a strongly improved
Pjm (f) possibly amplify the noise power while SNR might be obtained by a moderately degraded
preserving the signal power. Analogically to the resolution. So, finally this results in a trade-off
ambiguities, this amplification rises with increasing between azimuth resolution and SNR.
Fig. 11. Basic principle of pattern tapering on transmit.
Suppression of aliased energy in combination with optimization of
signal energy.
Furthermore one can aim at improving the Fig. 12. AASRN versus PRF for BD = 7:6 kHz and different
conditions for the processing by matching the transmit antenna sizes and patterns. sin(f)=f characteristic for an
phase centres and the PRF to obtain a data array aperture of 3 m (solid), 4 m antenna with cos(x) excitation
that is sampled as uniform as possible. This can be (dashed), 4.6 m antenna with (sin(f)=f)2 characteristic (dotted)
and approximately rectangular pattern (dotted dashed) for 11.2 m
either achieved by an adaptive management of the
antenna.
PRF to obtain an improved spatial sampling, but
this approach shows only great potential in sparse
To demonstrate the potential of pattern tapering
array systems and is hence only treated shortly in
on transmit, we consider the system of Section IV
Appendix D. In single platform systems, it is more
and investigate different combinations of transmit
promising to adapt the phase centers to the PRF by
antenna dimensions and excitations that can be either
adding a reconfigurable preprocessing network to
realized by a separate transmit antenna or by using
the conventional beamforming obtaining a cascaded
an active array that offers the flexibility to use parts
network as introduced and elaborated in Section C.
of the receiving antenna for transmit. Fig. 12 shows
In a next stage, the processing itself might be
the results for the azimuth ambiguity suppression
modified, mitigating the negative effects of the
AASRN in comparison to the 3 m transmit antenna
inverse filter by introducing a trade-off between with a uniform taper and hence a sin(f)=f pattern
ambiguity suppression and SNR optimization in (solid) as used in Section IV. The suppression is
the processing strategy. This comprises for example already clearly improved for higher PRF values by
an adapted beamforming approach that merges the applying a 4 m antenna with a cos(x) excitation
classical ambiguity suppression strategy with a beam (dashed) and becomes even better with a 4.6 m
steering approach that optimizes the signal power (cf. antenna with a simple triangular tapering which
Section D). As one will see, the transition from the entails a (sin(f)=f)2 characteristic (dotted). However,
techniques in Section C to Section D is smooth, as to provide an improvement in suppression also
both approaches are based on cascaded processing for lower PRF values, a larger transmit antenna is
networks. necessary. An unconventional realization is given by
the quasi-optimum–as it approximates a rectangular
B. Pattern Tapering on Transmit pattern ¡ sin(x)=x excitation in combination with a
transmit antenna of 11.2 m (dotted dashed) what
As derived in Section IIID, all spectral energy corresponds to the antenna length on receive. It
outside the band [¡N ¢ PRF=2, N ¢ PRF=2] causes is realized by 35 elements of 0.32 m length each.
aliasing in the reconstructed signal and finally gives Especially the sin(x)=x-excitation demonstrates
rise to ambiguities. This can be avoided by confining the potential of tapering to efficiently cancel the
the Doppler bandwidth of the signal to N ¢ PRF by spurious spectral components while preserving the
an appropriate antenna pattern. In a very simple resolution. Note that all scenarios show a resolution
approach, one could just enlarge the dimension of the below 1 m (BD = 7:6 kHz). A full exploitation of
transmit antenna resulting in a narrower pattern. But the benefits of tapering requires a fine adjustment of
as this is achieved only at the expense of resolution, antenna dimensions, PRF and BD , taking into account
a bigger antenna in combination with an adapted the trade-off between resolution and ambiguity
tapering is to be used to provide improved ambiguity suppression.
suppression without degrading the resolution. In the frame of advanced concepts of transmit
Furthermore, the better the pattern is limited to the antenna architectures in azimuth, a next step
relevant Doppler frequencies, the better the emitted comprises an antenna that allows for adaptively
power is used as less power is lost by illuminating varying the transmit phase center from pulse to pulse
unwanted areas and consequently the NESZ of the by using only certain parts of the antenna. This would
system is improved. The basic idea is visualized in enable a sliding phase center on transmit from pulse
Fig. 11. to pulse that has the potential to compensate for
Fig. 13. “Prebeamshaping on receive” network as analogue
representation of preprocessing network. Amplified RF signals of
various apertures are adaptively weighted and combined before
A/D conversion. This yields N optimized output channels that
enter the following DBF network.
a nonoptimum PRF. By choosing the direction of

Fig. 14. Conventional multi-aperture system with subaperture size
the sliding phase center along or against the flight
d0 equal to the phase center distance (top) and system with
direction, it is possible to adjust the spatial sampling prebeamshaping on receive network (bottom). The multiple use of
resulting from too high or too low PRF values, certain elements yields mutually overlapping subapertures of
respectively. The PRF range that can be compensated increased length d10 , but decreased phase center spacing d1 .
is determined by the maximum displacement of the
phase center, while the fine-tuning ability of the
position, i.e., the “step size” of the phase center from resulting in N “virtual” output channels j. The
pulse to pulse, depends on the distance between single weights may vary from subaperture to subaperture and
transmit elements. Note that a shift of the phase center from channel to channel. This allows for “using” the
is necessary every time it has reached its outermost signal of a certain subaperture in more than one of the
position. In the case of a too high PRF this requires virtual channels resulting in a spatial overlap of these
combining or even skipping some samples that might channels. The contribution of the signal of element i
overlap while in the case of a too low PRF this results to the virtual channel j is described by the complex
in “missing” samples within the synthetic aperture, as coefficient wij (cf. Fig. 13). In the case of analogue
a gap occurs when the phase center is switched. preprocessing, each signal is amplified separately by
an LNA before weighting, combining, and digitizing
C. Cascaded Beamforming Networks I–Analogue the RF signals.
Prebeamshaping on Receive The basic idea behind the network is to modify
the received signal in way to match it to the
In the context of system optimization more reconstruction network, primarily by flexibly setting
sophisticated beamforming on receive approaches the number and phase center position of the virtual
are regarded as powerful tools to adaptively improve channels that are formed. This allows for adjusting the
the system performance. In a cascaded beamforming spatial sample positions to the actual system PRF to
network the existing system is extended by a minimize the nonuniform sampling. In addition, such
second network that is used for analogue or digital networks enable to implement pattern tapering on
preprocessing of the multi-aperture SAR signals receive thus offering a powerful tool to e.g. suppress
before the reconstruction filter network introduced ambiguities. The reconfigurable structure allows for
in Section III is passed. In this section, we stick flexibly allocating the network resources to emphasize
to an analogue preprocessing technique of the a specific system parameter that finally results in
multi-aperture RF signals, a so-called prebeamshaping an adjustable trade-off between SNR, AASRN , and
on receive network, but the results are also valid for resolution.
a digital preprocessing step introduced in Section D. In the following the basic relations regarding
The system is based on an antenna consisting of a the formation of virtual channels are presented
number of K independent elements. Then a network to demonstrate the idea and potential of the
follows, allowing for an individual and reconfigurable prebeamshaping concept. Consider a multi-aperture
weighting and combination of the elements’ signals antenna of overall length La = N0 ¢ d0 , where d0
represents the length of the aperture that forms monostatic signal U(f) filtered with an appropriate
the single channel. The spacing between the phase function Mi (f) that takes into account the offset
centers is equal to the length of the subapertures and between transmitter and receiving element. Further,
hence given by d0 , which determines the optimum in an analogue implementation the need for power
PRF (cf. Fig. 14, top). In Fig. 14, on the bottom, a dividers will result in amplitude imbalances in the
simple realization of a prebeamshaping network is signals that have to be compensated for by adapting
shown, where some of the receiving elements are the weighting functions wij according to (52) (cf.
part of multiple virtual channels yielding a number Appendix A). This allows for setting up the following
of N mutually overlapping subapertures of length relation between input signal power ps,el and output
d10 > d0 . Due to the overlap of adjacent apertures, signal power before and after focusing, given by ps,c
the resulting phase center spacing d1 decreases with in (45) and ps,c,BD in (46), respectively
respect to the reference case shown on the top, where 2¯ ¯2 3
no prebeamshaping network is applied, i.e., d1 < d0 . ¯X K ¯
¯ ¯
ps,c = ps,el ¢ N 2 ¢ G ¢ E 4¯ (Mi (f) ¢ wij (f))¯ 5 (45)
Consequently while keeping the overall antenna ¯ i=1 ¯
length constant, the resulting spatial sampling of the
system is modified as the corresponding optimum 2 ¯ K ¯2 3
¯X ¯
PRF is now defined by N and d1 . This means that ¯ ¯
ps,c,BD = N 2 ¢ G ¢ E 4A(f)2 ¢ ¯ Mi (f) ¢ wij (f) ¢ rect(f=BD )¯ 5 :
one obtains a system that allows for flexibly setting ¯ i=1 ¯
the phase centers appropriate to the operating PRF
and improving the SNR. In this context, the modified (46)
SNR scaling factor of a cascaded beamforming In a similar way, the power of the ambiguities is
networks has to be taken into account as we see in to be adapted to the prebeamshaping scenario. This
the following. In addition, a larger subaperture length means that all expressions derived in Section IIIE can
is available enabling pattern tapering of the receiving be adapted to the extended processing network by
apertures by adjusting the network’s weighting adding the power gain G of the LNA and replacing
coefficients. The enlarged subaperture length d10 U(f) with the resulting equivalent monostatic output
and the corresponding phase center distance d1 are signal of the prebeamshaping network Uc (f), which
connected by (42) and define the overlap dov of is derived in Appendix A and represents the spectral
adjacent apertures as given in (43), resulting in a new appearance of the signals before the reconstruction
optimum PRFuni,c given by (44) network, but takes into account the prebeamshaping
on receive.
La ¡ d10 N ¢ d ¡ d10
d1 = = 0 0 (42) Finally, let us consider the influence of the
N ¡1 N ¡1 prebeamshaping network on the output noise power
dov = d10 ¡ d1 (43) pn,c under the assumption of a lossless network of
weighting elements wij (f). As mentioned before,
2 ¢ vs N0 ¢ (N ¡ 1) the receiver noise is assumed to be mutually
PRFuni,c = = ¡ ¢ ¢ PRFuni :
N ¢ d1 N ¢ N0 ¡ d10 =d0 uncorrelated additive white Gaussian noise. This
(44) means that the input noise power pn,el is amplified
by the prebeamshaping network and the following
Note that the preprocessing network has an impact reconstruction, yielding the output noise power in
on signal, ambiguities and noise power and thus the the data ps,c and in the image ps,c,BD , according to
expressions describing these parameters that were (47) and (48), respectively. Note that in contrast to
derived in Sections IIID—IIIF have to be adapted to the mere DBF network discussed before, the mutual
the extended beamforming scenario. The following correlations introduced by the prebeamshaping do not
paragraph briefly summarizes the differences to the allow any more for simplifying the squared sum to a
conventional beamforming network and the resulting sum of squares
relations that are derived in detail in Appendix A. 2¯ ¯2 3
XK ¯X N ¯
The relevant signal for the output signal power is ¯ ¯
pn,c = pn,el ¢ G ¢ F ¢ E 4¯ Pj (f) ¢ wij (f)¯ 5 (47)
determined by the input signals of the reconstruction i=1
¯ j=1
¯
network (cf. Section IIID). These signals are given
2¯ ¯2 3
by the N output signals of the prebeamshaping X
K ¯X N ¯
¯ ¯
network, which are defined by the weighted sum of pn,c,BD = pn,el ¢ G ¢ F ¢ E 4¯ Pj (f) ¢ wij (f) ¢ rect(f=BD )¯ 5 :
signals received by the single elements. The respective i=1
¯ j=1 ¯
weighting is given by the coefficients wij (f), that
can be Doppler frequency dependent in the case of (48)
a digital preprocessing network (cf. Section VD). Combining the expressions for signal power (cf.
Further, similar to Section IIIA, the received signal (45)) and noise power (cf. (47)) and normalizing by
at a single element can be modeled as an equivalent the optimum value F=K yields the SNR scaling factor
Fig. 15. Predicted SNR scaling factor ©bf of the the Fig. 16. AASRN versus PRF with prebeamshaping network
conventional DBF network (solid) compared with the case where (dashed) compared with conventional DBF (solid) (BD = 7:6 kHz).
prebeamshaping network is added (dashed). Simulated values Peak in conventional approach occurs at “singular”
overlaid in diamonds (BD = 7:6 kHz). PRF = 1440 Hz (cf. Section IIC).
©bf,c of the cascaded network with respect to the data the reconstruction network shows sufficiently low
(cf. (49)). As before, the SNR scaling factor with values from 1150 Hz up to approximately 1350 Hz,
respect to the image cannot be given explicitly, but but yields unacceptably high values for the PRF
can be determined by calculating the signal power range above (cf. Fig. 15, solid line). To provide
ps,c,BD and the noise power pn,c,BD according to (46) a low SNR scaling over the complete PRF range,
and (48), respectively a reconfiguration of the prebeamshaping network
becomes necessary for higher PRF values. In the
pn,c =pn,el K
©bf,c = ¢ present case, mutually overlapping subapertures of
ps,c =ps,el F 2.625 m length are formed, resulting in a decreased
·¯ ¯2 ¸ phase center spacing that corresponds to a new
P ¯PN ¯
K¢ K i=1 E ¯ P
j=1 j (f) ¢ wij (f) ¢ rect(f=IS ¯
) optimum PRF of 1346 Hz. In order to keep the
= ·¯ ¯2 ¸ : beamwidth of the receiving pattern constant and to
2 ¯PK ¯
N ¢ E ¯ i=1 Mi (f) ¢ wij (f) ¢ rect(f=IS )¯ suppress its sidelobes, a cosine taper is applied to each
of the subapertures. The respective ©bf,c is given in
(49) Fig. 15 (dashed line). It is normalised to the optimum
As explained in detail in Appendix A, the ©bf of the conventional network and takes into
quantization noise power in an analogue realization account the impact of the subaperture dimension and
pn,q,a is described by (50), while (51) in Section D taper on noise power, signal peak power, and azimuth
will give the noise resulting from a digitization before loss Laz that has increased to 3.05 dB due to the
the preprocessing network. The comparison of (50) preprocessing network compared with Laz = 2:9 dB in
with (51) implies that digitizing as late as possible the conventional case. Fig. 15 shows that an improved
reduces the scaling of the quantization error SNR scaling factor is obtained for PRF values above
2¯ 1240 Hz.
¯2 3
¯X N ¯ In terms of the NESZ, we focus on the differences
6¯ ¯ 7
pn,q,a = E 4¯¯ Pj (f) ¢ nqj (f)¯¯ 5 : (50) of the modified system to the system of Section IV.
¯ j=1 ¯ First, one can state a degradation of 0.2 dB due to the
increased Laz . It results from an increased transmit
To demonstrate the potential of the antenna and an increased receive aperture giving rise
prebeamshaping on receive technique we assume a to an overall Laz of 2.9 dB compared with 2.7 dB
modified system where each subaperture consists of of the original system. In contrast, the new antenna
multiple independent elements. With respect to the dimensions result in an increased gain of 0.2 dB on
system presented in Section IV, the new system is transmit and 0.4 dB on receive and the SNR scaling
designed to be applicable in a larger PRF range from factor of the reconstruction network (cf. Fig. 15) tends
1150 to 1550 Hz. Hence, the subaperture length is to be better than in the original system (cf. Fig. 9),
increased to 1.75 m resulting in an overall length of which might be compensated by the decreasing SNRel
12.25 m and a decreased optimum PRF of 1234 Hz. with increasing PRF according to (40). Altogether, the
Further, a slightly increased aperture antenna on NESZ of the original system (cf. Fig. 10) gives a good
transmit of 3.15 m is required to ensure a sufficient estimation for the new NESZ.
ambiguity suppression as we see later. All other Concerning the resolution, the increased
system parameters remain unaltered. subaperture length of 1.75 m with respect to the
An analysis of the SNR scaling factor ©bf of length of 1.6 m in the original system entails a
minor degradation to 1.03 m, both for conventional to improve the SNR. This means that a number of
and prebeamshaping system operation. Finally, the degrees of freedom is used for placing nulls in the
ambiguous energy suppression of the modified group pattern while other weights are chosen to
system is investigated. As mentioned above, the optimize the signal power. Basically this approach
transmit antenna was slightly increased to 3.15 m results in the cascaded structure of two subsequent
to guarantee a minimum suppression of ¡21 dB (cf. networks where one effectuates the steering of
Fig. 16). When the preprocessing network is applied, the mainlobe while the other cancels the azimuth
the suppression becomes better for PRF values ambiguities. The order of the networks and the
above » 1200 Hz and is clearly improved for higher combinations of the subapertures are in principle
PRF values due to reduced sidelobes of the receive arbitrary as long as the structure allows for the
pattern and minimized amplification of the azimuth multiplication of the respective patterns of the two
ambiguous energy caused by the adapted phase stages. This means that the weighting functions are the
centers (cf. Fig. 16, dashed line). In combination same for each of the formed channels. Such networks
with the results for the SNR scaling factor ©bf , show flexibility with respect to the receiving pattern
it is hence favourable to use the conventional and the effective sampling ratio given by the number
configuration below a PRF of 1200 Hz and apply the of different reconstructed channels. Further, the spatial
prebeamshaping up from 1240 Hz. In between the sample positions can be adapted, but only with certain
focus can be either turned to the NESZ by choosing flexibility as this is directly linked to the number of
the conventional approach or the preprocessing is reconstructed channels. Regarding SNR, AASRN ,
applied to concentrate on the optimization of the and resolution this results in a trade-off between the
AASRN . effective sampling ratio, the receiving pattern, and an
eventual amplification by the reconstruction algorithm
depending on the spatial sampling.
D. Cascaded Beamforming Networks II–Adapted Concerning the scaling of the quantization noise,
Digital Beamforming the position of the A/D converter with respect to
the weighting networks has to be considered, as
Similar to the prebeamshaping network presented only processing steps after the digitization affect
before, this approach is based on subsequent the quantization error. This means that in contrast
processing networks, but in contrast to the analogue to (50) the preprocessing is taken into account for
prebeamshaping, all signals are sampled, digitized, determining the power of the quantization noise pn,q,d
and stored before processing. Hence, the amount of 2¯
data to be handled limits in this case the maximum ¯2 3
XK ¯X N ¯
number of individual elements but a complete 6¯¯ ¯ 7
pn,q,d = E 4¯ Pj (f) ¢ wij (f) ¢ nqi (f)¯¯ 5 : (51)
a posteriori processing becomes possible enabling i=1 ¯ j=1 ¯
wide flexibility and reconfigurability in terms of
order and structure of the networks. As the basic One possible realization of a cascaded network
principle is again a cascaded structure of networks, structure is given by Fig. 17, where the multi-aperture
the expressions for the SNR scaling factor ©bf in reconstruction is followed by a beam-steering with
(45)—(49) are valid. weights wi (f). Consider an antenna of N0 receiving
Recall the “classical” reconstruction algorithm elements, where N channels are processed by the
presented in Section III that aims at recovering the multi-aperture reconstruction algorithm, respectively,
unambiguous Doppler spectrum by suppressing the to form N2 output channels that may mutually overlap
ambiguous frequency components. In terms of the or not. The resulting N2 channels are then weighted
group antenna characteristic that results from the and combined where the weighting functions wi (f)
weighting coefficients of the beamforming algorithm, are chosen to steer the maximum of the joint antenna
this corresponds to placing nulls at angles where characteristic for every f 2 [¡BD =2, BD =2] in the
ambiguous Doppler frequencies are situated. Further, direction corresponding to that frequency. As the
one observes that only for uniform sampling, the multi-aperture reconstruction processing does not
maximum in the group characteristic is steered at affect the signal envelope, the beamwidth of the
the Doppler frequency that is to be recovered. For output signal is defined by the single element length,
increasing nonuniformity, the maximum drifts away while the gain is determined by the number of
from that angle resulting in a decreased gain. Hence, channels that are combined in the second stage. This
the degradation of the SNR is caused by a loss of means that the signal amplitude gain rises by a factor
the signal energy that is compensated for by a later of N2 at the cost of an effective sampling ratio that
scaling of the signal that lifts up the noise floor. A is decreased by a factor of N0 =N and an increased
basic idea of the “adapted beamforming” by cascaded noise floor. The latter is caused by mutual correlations
networks is now to introduce a trade-off between between the noise components introduced by partly
ambiguity suppression and optimized signal energy overlapping output channels. Equations (45)—(49)
Fig. 19. Prediction (dashed) and simulation (diamond symbols)
Fig. 17. Cascaded networks for ambiguity suppression followed of SNR scaling factors ©bf for adapted beamforming compared
by beam-steering with mutual overlap of adjacent filter groups of with prediction of conventional DBF (solid).
first stage.
regarding the PRF. Note that all processing is applied
a posteriori and consequently the system setting can
be reconfigured arbitrarily to focus on the respective
performance parameter of interest, be it ambiguity
suppression, resolution, or NESZ.
In the following, the potential of the adapted
beamforming approach is demonstrated with a system
example. Consider again the system of Section C
with 7 subapertures of length 1.75 m, each, and a
PRF range of operation from 1150 Hz to 1550 Hz.
As derived in Section C, the SNR scaling factor of
the reconstruction network shows sufficiently low
values up from 1150 Hz, but yields unacceptably high
Fig. 18. Cascaded networks: Beam-steering and pattern tapering values for PRF values above 1400 Hz (cf. Fig. 19,
followed by reconstruction of N channels with mutual overlap of solid line). To operate the system in these PRF ranges,
adjacent beam-steering groups. This representation is similar to the gathered data are processed in a different way:
prebeamshaping network but with less elements and A/D
The signals of six adjacent subapertures, respectively,
conversion carried out before processing.
are processed by the multi-aperture reconstruction
algorithm, yielding two output channels that are
describe the resulting SNR scaling factor of the
weighted and combined to maximize the signal power.
network. Changing the order of the processing networks, this
The commutativity of the system allows for is equivalent to combining two adjacent receiving
changing the different stages as long as the resulting elements by the beam-steering, respectively, yielding
processing of each element’s branch remains six channels that are reconstructed unambiguously in
unaltered. Consequently, the system of Fig. 17 the second stage. A further interpretation arises, if the
can be equivalently represented by inversely beam-steering is combined with the multi-aperture
ordered networks. Fig. 18 shows an example of reconstruction to a single filter function for each
such a structure where the steering of the beam in channel. In this case two systems of six elements
combination with a possible tapering is carried out each are obtained, where the respective processing
before multi-aperture reconstruction. is equivalent to the multi-aperture reconstruction
One recognizes that the way of combining the two if the systems are considered separately. The
stages offers a wide range of flexibility, for example reconstructed signals from these systems are simply
a closer spacing of the resulting phase centers can be added afterwards.
achieved by subapertures that are no longer formed Due to the adapted processing the number of
by adjacent but distributed elements. The minimum phase centers is reduced and consequently the
possible phase center spacing is given by the distance optimum PRF is shifted to 1440 Hz. Further,
between adjacent elements. It can be achieved if mutual correlations cause the noise to rise. This
subapertures made up of N2 elements overlap by is compensated by the increased signal power due
N2 ¡ 1 elements with their neighbored subapertures to the doubled area of each subaperture, finally
or if distributed elements form the subapertures. The leading to a SNR scaling factor ©bf that is improved
maximum spacing is obtained in the conventional by 0.35 dB when comparing the values at the
case where adjacent subapertures do not overlap. respective optimum PRFs (cf. Fig. 19, dashed
In combination with the number of channels that line). Due to the reduced effective sampling, this
are formed, this offers a wide range of adaptation curve starts only at PRF = BD =6. As the azimuth
amplification of residual azimuth ambiguities. The
theory was verified in simulations when evaluating
an example system that was designed to image a
swath of 100 km with a geometric resolution of 1 m.
Concerning the performance parameters, simulations
were carried out including a comparison of the
reconstruction algorithm to alternative methods of
processing the azimuth signal. The results showed
that only the reconstruction algorithm and the very
similar null-steering technique provide high resolution
in combination with efficient ambiguity suppression.
Fig. 20. AASRN for adapted beamforming (dashed) compared
In a next step, the main error sources that may disturb
with conventional DBF (solid) (BD = 7:6 kHz). Peak in the output signal were identified to be aliased parts
conventional approach occurs at “singular” PRF at 1440 Hz in the received signal and the amplification of errors
(cf. Section IIC). and noise in the input signal due to the processing in
the case of nonuniformly sampled data. Based on this
loss remains the same due to the beam-steering, analysis, optimization strategies as pattern tapering on
the NESZ of this system can be deduced from the transmit (Section VB), prebeamshaping on receive
NESZ derived in Section C in combination with (Section VC) and adapted beamforming networks
the modified ©bf of Fig. 19. Additionally, a “loss” (Section VD) were introduced and the resulting
factor of 6=7 has to be considered, as the number of performance improvement demonstrated. Pattern
effective channels decreased from N = 7 to N = 6 (cf.
tapering on transmit enables an efficient suppression
(40)).
of ambiguous energy by adapting the pattern to
Next, the AASRN of the system is investigated
the processed Doppler bandwidth and thus limiting
and presented in Fig. 20. The solid line shows the
the aliased part in the signal. In an extension of an
suppression of the original system that is improved
advanced transmit architecture, an adaptive adjustment
by the adapted beamforming above 1340 Hz
of the transmit phase center to the PRF is possible by
(dashed line). In combination with Fig. 19, one can
letting the phase center “slide” over the whole antenna
conclude that the conventional approach is suitable
from pulse to pulse. Prebeamshaping on receive
below 1340 Hz, while the adapted beamforming is
networks can enforce the limitation of the Doppler
favourable for PRF values above.
spectrum by tapering the receive pattern. Furthermore,
Regarding the resolution, the adapted beamforming
such networks allow for adaptively adjusting the
yields a value of 1.04 m similar to the conventional
phase center positions to the PRF thus optimizing the
case.
spatial sampling. As a consequence, the system can be
operated in a wider PRF range. Finally, the adapted
VI. DISCUSSION
beamforming method established the idea of pushing
The paper summarized the state-of-the-art for SAR the processing from a pure suppression of ambiguous
systems and principles that enable HRWS imaging energy to a trade-off between SNR and ambiguity
and recalled the basic properties of multi-aperture suppression by combining a beam-steering approach
SAR signals regarding the spatial sampling. Then, with the reconstruction algorithm. In the case of
an algorithm for DBF on receive in multi-aperture prebeamshaping on receive networks and the adapted
SAR systems was presented that allows for the beamforming approach, the theoretical examination of
unambiguous recovery of the azimuth signal even in residual errors, ambiguities and the SNR scaling factor
case of a nonuniformly sampled synthetic aperture, was extended to the class of cascaded beamforming
thereby avoiding any stringent PRF restriction. This networks. The extensions consider mutual correlations
multi-aperture reconstruction algorithm combines between different channels in contrast to conventional
the individual receiver signals in a linear space-time beamforming systems.
processing and can be interpreted as solving a linear The focus of future work is turned to distributed
system of equations in a way that ambiguous parts SAR systems with multiple receiver satellites
in the azimuth signal are canceled. In a next step, the forming a sparse array SAR system. In principle
algorithm was investigated in detail with respect to the reconstruction algorithm is also applicable to
the influence of the processing in the DBF network those systems, but new aspects of importance arise
on signal power, ambiguities, and noise, yielding that have to be considered. Firstly, the increased
analytical expressions for the estimation of residual along-track separations of the receiving apertures
ambiguities and the determination of the SNR scaling have to be taken into account as they entail a high
factor ©bf of a beamforming network. This analysis sensitivity of the sampling against PRF variations
showed that especially strong nonuniform sampling due to modulo influence on the effective sample
of the azimuth signal may cause a degraded SNR due position. On the one hand, this might cause problems,
to an increased network SNR scaling factor and an as small PRF changes have a huge impact on the
sampling scenario, but on the other hand this offers a
big opportunity for optimizing the spatial sampling by
adapting the PRF as was mentioned in Section VA and
Appendix D. Further, large along-track displacements
might require a steering of the antenna footprints
to illuminate a joint area on ground resulting in a
different squint of the satellites and thus the reception Fig. 21. Multi-aperture system model taking into account
network to perform prebeamshaping on receive before signals are
of mutually shifted Doppler spectra in the receivers. sampled and pass through reconstruction filter network.
This might lead to a subsampling and require new
processing approaches, e.g. the combination of
multi-aperture reconstruction and superresolution
techniques [37]. In this context, the investigation of
the impact of different antenna gains versus Doppler
frequency on the processing will also be of interest.
Secondly, any cross-track separation of the receivers
will introduce an additional phase in the received
signals, which has to be compensated, e.g., via the Fig. 22. Multi-aperture system model including cascaded
simultaneous acquisition of a digital elevation model networks to perform preprocessing after the signals are sampled
in case of multiple satellites. Further, a sparse array and before they pass through reconstruction filter network.
configuration has to be carefully investigated with
respect to a two-dimensional processing approach as extended by a block representing the additional
different sensors might receive the signal of the same processing steps, either a prebeamshaping (PBS)
target with different ranges and range cell migrations network situated before the analog-to-digital converter
as introduced in [28]. (ADC) (cf. Fig. 21) or a network processing
Further potential fields of application for DBF the already digitized data. In the case where the
are e.g. the application of the ScanSAR or the preprocessing network is located behind the ADC,
TOPS-SAR principle in multi-aperture systems, the order of the applied filters can of course be
enabling ultrawide swaths. In this context the impact changed as long as the overall transfer function
of the multi-aperture reconstruction algorithm when of each branch remains the same. The weight
operated in combination with burst modes is to be functions are denominated by wij that (although
investigated with respect to the particular properties of not necessary) may be frequency dependent in
such modes. As the scaling of the ambiguous energy the case of a filtering after the ADC (cf. Fig. 22).
is strongly dependent on the Doppler sub-band, the We start with a linear system of K channels i with
use of different subspectra for different targets in a an input signal ui (t), a receiver noise component
ScanSAR or TOPS-SAR mode will result in a large ni,B (f), and the corresponding signal-to-noise ratio
variation of the residual azimuth ambiguities of these SNRel,i , respectively. As before, receiver noise is
targets. So, in ScanSAR operation and even in the considered to be the dominating thermal noise source
TOPS-SAR mode, this can result in a scalloping-like and is assumed to be additive, white, and mutually
effect for the azimuth ambiguities, which is currently uncorrelated between the channels. After reception,
investigated in detail. the signal is amplified before the K input signals
are weighted and combined to form N intermediate
APPENDIX A. CASCADED BEAMFORMING output signals, each representing a virtual channel
NETWORKS for the reconstruction filter network. These channels
are then (either before or after the preprocessing)
In the context of system optimization more subsampled in azimuth with a ratio of PRF inducing
sophisticated beamforming approaches are regarded a quantization error that is modelled as additive
as powerful tools to adaptively improve the system noise sources nqj (f) and nqi (f), respectively. Then
performance. Sections VC and VD established filtering by the reconstruction filters Pj (f) takes place
the idea of cascaded beamforming networks and followed by coherent combination of all N output
investigated the new concept with respect to the branches yielding the signal uout [k] (cf. Fig. 21,
network capability to adaptively position the systems’ Fig. 22).
phase centers. In the following, the impact on signal,
ambiguities and noise power that were only mentioned
briefly in Section VC are derived in more detail. B. Compensation for Power Loss
A. Extended System Model In the analogue representation of a preprocessing

network, amplitude imbalances might occur as
Based on the system model presented in not all elements’ signals are to be split equally.
Section IIIF and on the block diagram given in Hence, depending on the necessary branching of the
Fig. 13, we obtain new system models that are respective signal, the gain loss is different and has to
be compensated to ensure uniform amplitude of all interpretation allows for expressing the signals by a
signals when entering the first processing network. equivalent monostatic signal Uc (f) and functions Hj (f)
This can be achieved by including the compensation characterizing the virtual channels j as done in (12).
in the weighting functions of this first stage by We can simplify (55)—(56), where Uc (f) corresponds
replacing the amplitude gain jwij j by the new factor to the signal that arises by the weighted sum of the
jwij0 j according to (52), where Nch,i represents the single elements, if the resulting phase center was
number of signals in which the original signal of situated at the same position as the transmitter. The
element i is split. By this, the amplitude of the signal LNA’s power gain G is mentioned separately as it will
after being split is recovered as given in (53), where cancel later
Ui,split (f) gives the signal after splitting while Ui (f)
denotes the original signal of channel i. With these Uc,j (f) = G ¢ Uc (f) ¢ Hj (f): (56)
adapted weighting functions all following equations
Only the prebeamshaping changes the
remain valid
q characteristic of Uc (f) while the DBF network that
jwij0 j = jwij j ¢ Nch,i (52) reconstructs the signal only introduces a scaling
factor of N without modifying the pattern of Uc (f).
U (f) Hence, the signal power is determined by the spectral
Ui,split (f) ¢ wij0 (f) = pi ¢ w0 (f) = Ui (f) ¢ wij (f):
Nch,i ij appearance of Uc (f) given by the resulting pattern
(53) characteristic of the combined subapertures and the
number of virtual channels N. Using (57), we obtain
C. Signal Power the following output signal after reconstruction in the
data (cf. (58)) and in the image after focusing (cf.
The desired signal that is relevant for the output
(59)) that correspond to (45) and (46), respectively
signal power is determined by the input signal of the ¯ ¯
reconstruction network (cf. Section IIID) and does ¯ XK ¯
¯ ¯
not depend on the location of the ADC. In contrast jUc (f)j = ¯U(f) ¢ Mi (f) ¢ wij (f)¯
¯ ¯
to the case where no prebeamshaping is applied, this i=1
signal is given by Uc,j (f) that forms the input of the ¯ K ¯

¯X ¯
reconstruction filter network and is determined by the ¯ ¯
= A(f) ¢ ¯ Mi (f) ¢ wij (f)¯ (57)
weighted sum of signals Ui (f) received by the single ¯ i=1 ¯
elements and respective weighting coefficients
ps,c = N 2 ¢ G ¢ E[jUc (f) ¢ rect(f=IS )j2 ]
K
X 2¯ ¯2 3
Uc,j (f) = Gi ¢ Ui (f) ¢ wij (f): (54) ¯X K ¯
¯ ¯
i=1 = ps,el ¢ N 2 ¢ G ¢ E 4¯ Mi (f) ¢ wij (f)¯ 5 (58)
¯ i=1 ¯
As derived in Section IIIA, Ui (f) can be modelled
as an equivalent monostatic function U(f) filtered 2¯ ¯2 3
with an appropriate function that corresponds basically ¯ XK ¯
¯ ¯
to a time-shift and the addition of a constant phase. ps,c,BD = N 2 ¢ G ¢ E 4¯U(f) ¢ Mi (f) ¢ wij (f) ¢ rect(f=BD )¯ 5 :
¯ i=1
¯
To avoid confusion with the reference functions
Hj (f) that characterize the virtual channels, we (59)
denominate these functions Mi (f). We obtain (again
assuming identical LNAs in all channels and taking D. Residual Reconstruction Error
into account (52) if analogue preprocessing is applied)
the following, where U(f) represents the signal If a preprocessing network is used before
received by a single element at the same position as reconstruction is applied, the power of the ambiguities
the transmit aperture is to be adapted to the prebeamshaping scenario. This
means that all expressions derived in Section IIIE
K
X can be adapted to the extended processing network
Uc,j (f) = Gi ¢ Ui (f) ¢ wij (f) by adding the gain factor G and replacing U(f) by
i=1
the resulting output signal of the prebeamshaping
K
X network Uc (f) defined by (56), that represents
= G ¢ U(f) ¢ Mi (f) ¢ wij (f): (55) the spectral appearance of the signal before the
i=1 reconstruction network, but takes into account the
In a next step we assume the wij (f) being the preprocessing (either prebeamshaping on receive
same for every virtual channel j which means or adapted beamforming). Assuming that all virtual
that all resulting virtual channels j are identical in channels that enter the reconstruction filter network
terms of their envelope and only shifted relative to are identical in terms of weighting of the elements that
each other with respect to their phase centers. This form the channel, Uc (f) is determined by the weighted
sum given in (56) but does not include the gain G of approximated by pn,c,rx . Analogously to (34) that gives
the LNA. the SNR scaling factor of conventional beamforming
networks, the SNR scaling factor of the cascaded
E. SNR Scaling of Cascaded Beamforming network with respect to the data ©bf,c (cf. (63))
Networks–©bf,c is determined by expression (58) (that gives the
output signal power) and the first term of (60) that
Let us consider the influence of the approximates the noise power. As before, the SNR
prebeamshaping network on the noise power under scaling factor for the focused image cannot be given
the assumption of a lossless network of weighting explicitly, as the relation between the input and output
elements wij (f). As mentioned above, the receiver signal power depends on the shape of U(f) and other
noise is assumed to be mutually uncorrelated additive
parameters as defined in (59). Of course ©bf,c can
white Gaussian noise ni,B (f) with a power spectral
be calculated using (59) for the signal power and
density of N0,B with respect to the system bandwidth
expression (64) giving the remaining the noise power
B. Further, quantization noise is assumed to be
after focusing
uniformly distributed and spectrally white as proposed
in [40]. Again identical elements i, implying the same pn,c =pn,el K
©bf,c = ¢
characteristics for Gi , Fi, gi , fi for all i are assumed, ps,c =ps,el F
yielding the following output noise power pn,c of the ·¯
PK PN ¯2 ¸
system ¯
K ¢ i=1 E ¯ j=1 Pj (f) ¢ wij (f) ¢ rect(f=IS )¯
¯
2¯ ¯2 3
¯X ¯ ·¯
XK
6¯¯
N
¯ 7 = ¯2 ¸
pn,c = pn,el ¢ G ¢ F ¢ E 4¯ Pj (f) ¢ wij (f)¯¯ 5 ¯PK ¯
N 2 ¢ E ¯ i=1 Mi (f) ¢ wij (f) ¢ rect(f=IS )¯
i=1 ¯ j=1 ¯
| {z } (63)
Receiver noise pn,c,rx
2¯ ¯2 3
+ pn,q XK ¯X N ¯
¯
E 4¯
¯
Pj (f) ¢ wij (f) ¢ rect(f=BD )¯ 5 :
|{z} pn,c,B = pn,el ¢ G ¢ F ¢
Quantization noise
D
i=1
¯ j=1
¯
= pn,c,rx + pn,q : (60) (64)

This means that the input noise power is
amplified by the prebeamshaping and the following APPENDIX B. MULTI-APERTURE RECONSTRUCTION
reconstruction and weighted quantization noise is IN THE FRAME OF CLASSICAL STAP
added. Further, in contrast to a mere DBF network as
Consider, according to [47], a multi-channel
discussed before, the mutual correlations introduced
beamforming system consisting of N channels. Each
by the prebeamshaping do not allow any more
channel i is described by its input signal si , noise
for simplifying the squared sum that gives the
component ni , and weighting function ai that are
thermal output noise power to a sum of squares. This
summarized by s = [s1 , : : : , sN ]T , n = [n1 , : : : , nN ]T
amplification of the receiver noise is independent
of the system, while the quantization error depends and a = [a1 , : : : , aN ]T , respectively. The output y of
on the location of the ADC. If the preprocessing is the system can then be described by (65), where the
applied before digitization, the quantization noise at operator ± denotes the scalar product
the output is given by (61) while (62) is valid if the
y = a ± s + a ± n: (65)
quantization error is introduced in the signal before
the preprocessing network is passed Further, if we assume an incident plane wave
2¯ ¯2 3 and use quasi-parallel approximation, each si can be
¯X N ¯
6 ¯ ¯ 7 related to s1 by a phase term that accounts for the
pn,q,a = E 4¯¯ Pj (f) ¢ nqj (f)¯¯ 5 (61)
¯ j=1 ¯ azimuth angle-dependent path difference between
receivers i and 1. Hence, s simplifies to a steering
2¯ ¯2 3 vector, where μ gives the azimuth angle and ¢xi
XK ¯X N ¯
6 ¯ ¯ 7 denotes the distance of receiver i to receiver 1.
pn,q,d = E 4¯¯ Pj (f) ¢ wij (f) ¢ nqi (f)¯¯ 5 : (62) Without restriction of generality we assume ¢x1 = 0
i=1 ¯ j=1 ¯
and finally obtain
Again, the number of bits is assumed to be chosen · · ¸ · ¸¸T
2¼ 2¼
such that the quantization error in the output signal s = 1, exp j sin(£)¢xi : : : , exp j sin(£)¢xN :
¸ ¸
will be negligible compared with the (thermal)
receiver noise and consequently the overall noise is (66)
If coloured noise is assumed, the optimum
beamforming weights a to minimize the noise power
are defined by
R¡1 ± s
a¤ = (67)
k
where * denotes the complex conjugate, k represents
a normalization factor and R describes the noise
covariance matrix that consists of the covariance
matrix of the spatially structured noise Rc and the
unity matrix I representing the white noise of power
¾2 :
R = E[n ± n¤ ] = Rc + ¾ 2 I: (68)
In the special case when interpreting the Fig. 23. Incident signal sin (μ), arriving under angle μ on linear
ambiguous Doppler frequencies as a directive array of N receiving apertures. After reception, each receiver’s
interference arriving from the respective angles and signal si (μ) is weighted by ai and summed up yielding output
further assuming that no white noise is present, signal s(μ).
i.e., ¾ = 0, one can show that (69) holds true, if the
dependency on the azimuth angle μ is replaced by output signal s(μ) is determined by (71), where s1 (μ)
Doppler frequency f represents the signal at receiver 1
R(f)¡1 ± s(f) Rc (f)¡1 ± s(f) N
X
a¤ (f) = = = P¤ (f): s(£) = si (£) ¢ ai
k k
i=1
(69)
N
X
The matrix P represents the reconstruction filter
= s1 (£) ¢ (vi ¢ ai ) = s1 (£) ¢ v(£) ± a: (71)
functions derived in Section IIIB and contains in its
i=1
ith line the weighting function ai . Hence P offers an
alternative way of deriving the optimum beamformer Consider now N output signals for input signals
to suppress azimuth ambiguities that are interpreted as arriving from N different angles μj , j = f1, 2, : : : , Ng.
directive intereferences in the STAP context. The respective output signals are given by the vector
s that can be expressed by the matrix-vector operation
APPENDIX C. ANALOGY OF NULL-STEERING AND of (72), where V represents the manifold vectors for
MULTI-APERTURE RECONSTRUCTION the different angles as given in (73)
s1 (£) ¢ V ¢ a = s (72)
Consider a linear array of N receiving apertures 2 3
where the distance of aperture i to aperture 0 is v1 (£1 ) ¢¢¢ vN (£1 )
denoted by ¢xi and the look direction is designated 6 7
V = [v 1 , : : : , v N ] = 4 ¢ ¢ ¢ vi (£j ) ¢¢¢ 5:
by μ (cf. Fig. 23). By definition we set ¢x1 = 0. After
receiving the incident signal sin (μ), an individual v1 (£N ) ¢¢¢ vN (£N )
complex weighting coefficient ai is applied to each (73)
of the channels’ signals si (μ). Finally these signals are The weighting coefficients which are necessary to
combined coherently, what is indicated by the “§,” steer nulls in N ¡ 1 directions μj in order to “extract”
yielding the output signal s(μ). the signal from the Nth direction are then determined
Applying the quasi-parallel approximation, the by the inversion of V. One obtains a matrix A = V¡1
well-known array manifold vector v that characterizes that contains in its kth column the respective weights
the signal at each receiver i with respect to the signal to steer the nulls in the pattern to all directions μj ,
at receiver 1 looks as follows [46]: j 6= k.
· μ ¶
2¢¼ In the next step, consider a frequency f1 and
v(£) = 1, exp ¡j ¢ ¢x2 ¢ sin £ , : : : , its N ¡ 1 ambiguous frequencies within the system
¸
μ ¶¸T azimuth bandwidth N ¢ PRF, that are separated by
2¢¼ integer multiples of PRF (cf. (74)). Using the relation
exp ¡j ¢ ¢xN ¢ sin £
¸ between angle μ and Doppler frequency f of (75), the
variable μj in (73) can be replaced by the respective
= [v1 , : : : , vi , : : : , vN ]T : (70) fj , what yields V(f) in (76). Then the matrix A(f) =
Taking into account the weighting coefficients V¡1 (f) contains in its kth column the respective
given by the vector a = [a1 , : : : , aN ]T and the weights to steer the nulls in the pattern to all
subsequent coherent summation, the resulting ambiguous frequencies fj , j 6= k, while recovering fk .
Hence, calculating and applying A(f) for all f1 within the inter-sample distance is smaller than the
the interval I1 = [¡N ¢ PRF=2, ¡(N ¡ 2) ¢ PRF=2] and inter-pulse distance. Hence in such a system, the
the respective ambiguous Doppler frequency sets in potential is limited as small changes in the PRF
(74) allows for unambiguously recovering the original do not allow for large variations in the sampling.
spectrum of bandwidth N ¢ PRF In contrast, in sparse arrays, the spatial position of
h i
N ¢ PRF (N ¡ 2) ¢ PRF
f1 + [0, PRF, : : : , (N ¡ 1) ¢ PRF] =: [f1 , : : : , fj , : : : , fN ] with f1 2 ¡ ,¡ (74)
2 2
2¢v
f(£) = ¢ sin £ (75)
¸
2 · ¸ 3
h ¼ ¢ ¢x1
i ¼ ¢ ¢xN
6 exp ¡j ¢ ¢f ¢¢¢ exp ¡j ¢ ¢ f1 7
6 v 1 v 7
6 · ¸ 7
6 7
6
V(f) = 6
¼ ¢ ¢xi 7:
¢¢¢ exp ¡j ¢ ¢ (f1 + (j ¡ 1) ¢ PRF) ¢¢¢ 7
6 v 7
6 · ¸7
6 h i ¼ ¢ ¢xN 7
4 exp ¡j ¢ ¼ ¢ ¢x1 ¢ (f + (N ¡ 1) ¢ PRF) ¢¢¢ exp ¡j ¢ ¢ (f1 + (N ¡ 1) ¢ PRF) 5
v 1 v
(76)
Let us now compare above shown matrix V(f) adjacent samples is not bound to the same transmitted
to the system matrix H(f) derived in Section III. We pulse as in single platform systems. One recognizes
observe that under the assumed approximations, V(f) that variations in the PRF result in variations of the
is nearly identical to H(f). The only difference is the position ¢xj,i amplified by a factor (k ¡ l) and hence
constant phase term relating the bi- and monostatic only small changes in the PRF might already have a
sample position, that is accounted for in H(f) while large impact on the spatial distribution of the samples.
it is neglected in V(f). Hence the multi-aperture This offers wide flexibility to adapt the sampling even
reconstruction algorithm can be regarded as an if only a small variation of the PRF is possible and
extension of the classical null-steering approach that allows thus for compensating for the high sensitivity
also accounts for large receiver spacing. In the case of the systems’ SNR and AASRN regarding PRF
of a single-platform system, the distances between the variations.
receiving apertures are small, and consequently the
REFERENCES
resulting phase differences are negligible.
[1] Curlander, J., and McDonough, R.
APPENDIX D. ADAPTIVE PRF MANAGEMENT IN Synthetic Aperture Radar–Systems and Signal Processing.
SPARSE ARRAY SYSTEMS New York: Wiley, 1991.
[2] Carrara, W., Goodman, R., and Majewski, R.
As mentioned in Section VA, the conditions Spotlight Synthetic Aperture Radar: Signal Processing
Algorithm.
for the multi-aperture processing can be improved
Boston: Artech House, 1995.
by harmonizing the phase centers with the PRF to [3] Tomiyasu, K.
obtain a sampling as uniform as possible. In sparse Conceptual performance of a satellite borne, wide swath
array systems, an adaptive management of the PRF synthetic aperture radar.
shows great potential to adjust the virtual sample IEEE Transactions on Geoscience and Remote Sensing,
GRS-19 (1981), 108—116.
positions. Recalling the results derived in Section II,
[4] Moore, R. K., Claassen, J. P., and Lin, Y. H.
we can express the distance between adjacent samples Scanning spaceborne synthetic aperture radar with
originating from transmit pulse k at channel j and integrated radiometer.
transmit pulse l at channel i by (77), where ¢xj IEEE Transactions on Aerospace and Electronic Systems,
denotes the distance of the receiver to the transmitter AES-17 (1981), 410—421.
[5] De Zan, F., and Monti Guarnieri, A. M.
¢xj ¢xi v TOPSAR: Terrain observation by progressive scans.
¢xj,i = xj,k ¡ xi,l = ¡ + (k ¡ l) ¢ s ,
2 2 PRF IEEE Transactions on Geoscience and Remote Sensing, 44
(2006), 2352—2360.
k, l 2 Z: (77) [6] Claassen, J. P., and Eckerman, J.
A system for wide swath constant incident angle
In a conventionally operated single platform
coverage.
system, adjacent samples originate from the same or In Proceedings of Synthetic Aperture Radar Technology
subsequent pulses, i.e., jk ¡ lj · 1, and consequently Conference, Las Cruces, NM, 1978.
[7] Jain, A. [21] Krieger, G., and Moreira, A.
Multibeam synthetic aperture radar for global Potentials of digital beamforming in bi- and multistatic
oceanography. SAR.
IEEE Transactions on Antennas and Propagation, AP-27 In Proceedings of IEEE International Geoscience and
(1979), 535—538. Remote Sensing Symposium (IGARSS), Toulouse, France,
[8] Jean, B. R., and Rouse, J. W. 2003, 527—529.
A multiple beam synthetic aperture radar design concept [22] Heer, C., Soualle, F., Zahn, R., and Reber, R.
for geoscience applications. Investigations on a new high resolution wide swath SAR
IEEE Transactions on Geoscience and Remote Sensing, concept.
GRS-21 (1983), 201—207. In Proceedings of IEEE International Geoscience and
[9] Griffiths, H., and Mancini, P. Remote Sensing Symposium (IGARSS), Toulouse, France,
Ambiguity suppression in SARs using adaptive array 2003, 521—523, 2003.
techniques. [23] Younis, M., Fischer, C., and Wiesbeck, W.
In Proceedings of IEEE International Geoscience and Digital beamforming in SAR systems.
Remote Sensing Symposium (IGARSS), Espoo, Finland, IEEE Transactions on Geoscience and Remote Sensing, 41,
1991, 1015—1018. 7 (July 2003), 1735—1739.
[10] Currie, A., and Brown, M. A. [24] Younis, M., Venot, Y., and Wiesbeck, W.
Wide-swath SAR. A simulator for digital beam forming SAR.
IEE Proceedings–Radar Sonar and Navigation, 139, 2 In Proceedings of International Radar Symposium (IRS),
(1992), 122—135. Dresden, Germany, 2003.
[11] Callaghan, G. D., and Longstaff, I. D. [25] Mittermayer, J., and Runge, H.
Wide swath spaceborne SAR using a quad element array. Conceptual studies for exploiting the TerraSAR-X dual
IEE Proceedings–Radar Sonar and Navigation, 146, 3 receiving antenna.
(1999), 159—165. In Proceedings of Geoscience and Remote Sensing
Symposium (IGARSS), Toulouse, France, 2003.
[12] Goodman, N., Rajakrishna, D., and Stiles, J.
Wide swath, high resolution SAR using multiple receive [26] Aguttes, J. P.
apertures. The SAR train concept: Required antenna area distributed
In Proceedings of IEEE Geoscience and Remote Sensing over N smaller satellites, increase of performance by N.
Symposium (IGARSS), vol. 3, Hamburg, Germany, 1999, In Proceedings of Geoscience and Remote Sensing
1767—1769. Symposium (IGARSS), Toulouse, France, 2003.
[27] Krieger, G., Gebert, N., and Moreira, A.
[13] Younis, M., and Wiesbeck, W.
Unambiguous SAR signal reconstruction from
SAR with digital beamforming on receive only.
nonuniform displaced phase center sampling.
In Proceedings of IEEE Geoscience and Remote Sensing
IEEE Geoscience and Remote Sensing Letters, 1 (2004),
Symposium (IGARSS), vol. 3, Hamburg, Germany, 1999,
260—264.
1773—1775.
[14] Suess, M., Grafmüller, B., and Zahn, R.
SAR signal reconstruction from non-uniform displaced
A novel high resolution, wide swath SAR system.
phase center sampling.
In Proceedings of IEEE Geoscience and Remote
In Proceedings of Geoscience and Remote Sensing
Sensing Symposium (IGARSS), Sydney, Australia, 2001,
Symposium (IGARSS), Anchorage, AK, Sept. 2004.
1013—1015.
[15] Suess, M., and Wiesbeck, W. Digital beamforming and non-uniform displaced phase
Side-looking synthetic aperture system. centre sampling in bi- and multistatic SAR.
EP 1 241 487 A 1, 2001. In Proceedings of European Conference on Synthetic
[16] Wiesbeck, W. Aperture Radar (EUSAR), Ulm, Germany, 2004.
SDRS: Software-defined radar sensors. [30] Younis, M.
In Proceedings of IEEE Geoscience and Remote Sensing Digital beam-forming for high-resolution wide swath real
Symposium (IGARSS), Sydney, Australia, 2001, and synthetic aperture radar.
3259—3261. Karlsruhe, Germany, 2004.
[17] Suess, M., Zubler, M., and Zahn, R. [31] Gebert, N., Krieger, G., and Moreira, A.
Performance investigation on the high resolution, wide SAR signal reconstruction from non-uniform displaced
swath SAR system. phase centre sampling in the presence of perturbations.
In Proceedings of European Conference on Synthetic In Proceedings of Geoscience and Remote Sensing
Aperture Radar (EUSAR), Ulm, Germany, 2002, Symposium (IGARSS), Seoul, South Korea, 2005.
187—191. [32] Gebert, N., Krieger, G., and Moreira, A.
[18] Brule, L., and Baeggli, H. High resolution wide swath SAR imaging–System
Radarsat-2 program update. performance and influence of perturbations.
In Proceedings of IEEE Geoscience and Remote Sensing In Proceedings of International Radar Symposium (IRS),
Symposium (IGARSS), Toronto, Canada, 2002. Berlin, Germany, 2005.
[19] Goodman, N., Lin, S., Rajakrishna, D., and Stiles, J. [33] Li, Z., Wang, H., Su, T., and Bao, Z.
Processing of multiple-receiver spaceborne arrays for Generation of wide-swath and high-resolution SAR
wide area SAR. images from multichannel small spaceborne SAR system.
IEEE Transactions on Geoscience and Remote Sensing, 40, IEEE Geoscience and Remote Sensing Letters, 2, 1 (Jan.
4 (Apr. 2002), 841—852. 2005), 82—86.
[20] Krieger, G., Fiedler, H., Rodriguez-Cassola, M., Hounam, [34] Li, Z., Bao, Z., Wang, H., and Liao, G.
D., and Moreira, A. Performance improvement for constellation SAR using
System concepts for bi- and multistatic SAR missions. signal processing techniques.
In Proceedings of the ASAR Workshop 2003, Saint-Hubert, IEEE Transactions on Aerospace and Electronic Systems,
Quebec, Canada, June 2003. 42, 2 (2006), 436—452.
[35] Fischer, C., Heer, C., Krieger, G., and Werninghaus, R. [42] Márquez-Martínez, J., Mittermayer, J., and
A high resolution wide swath SAR. Rodríguez-Cassolà, M.
In Proceedings of European Conference on Synthetic Radiometric resolution optimization for future SAR
Aperture Radar (EUSAR), Dresden, Germany, 2006. systems.
[36] Gebert, N., Krieger, G., and Moreira, A. In Proceedings of Geoscience and Remote Sensing
High resolution wide swath SAR imaging with digital Symposium (IGARSS), Anchorage, AK, Sept. 2004.
beamforming–Performance analysis, optimization and [43] Kraus, J.
system design. Antennas.
In Proceedings of European Conference on Synthetic New York: McGraw-Hill, 1950.
Aperture Radar (EUSAR), Dresden, Germany, 2006. [44] Krieger, G., Gebert, N., and Moreira, A.
[37] Krieger, G., and Moreira, A. Multidimensional waveform encoding: A new digital
Spaceborne bi- and multistatic SAR: Potential and beamforming technique for synthetic aperture radar
challenges. remote sensing.
IEE Proceedings–Radar, Sonar and Navigation, 153 IEEE Transactions on Geoscience and Remote Sensing, 46,
(2006), 184—198. 1 (2008), 31—46.
[38] Papoulis, A. [45] Krieger, G., et al.
Generalized sampling expansion. TanDEM-X: A satellite formation for high-resolution
IEEE Transactions on Circuits and Systems, CAS-24, 11 SAR interferometry.
(1977), 652—654. IEEE Transactions on Geoscience and Remote Sensing, 45,
[39] Brown, J. 11, 1 (2007), 3317—3341.
Multi-channel sampling of low-pass signals. [46] van Trees, H. L.
IEEE Transactions on Circuits and Systems, CAS-28, 2 Optimum Array Processing–Part IV of Detection,
(1981), 101—106. Estimation, and Modulation Theory.
[40] Oppenheim, A. New York: Wiley, 2002.
Discrete-Time Processing. [47] Guerci, J. R.
Upper Saddle River, NJ: Prentice-Hall, 1999. Space-Time Adaptive Processing.
[41] Wang, T., and Bao, Z. Boston: Artech House, 2003.
Improving the image quality of spaceborne
mulitple-aperture SAR under minimization of sidelobe
clutter and noise.
IEEE Geoscience and Remote Sensing Letters, 3, 3 (July
2006), 297—301.
Nicolas Gebert received the Dipl.-Ing (M.S.) from the Munich University of
Technology (TU München), Munich, Germany, in 2003.
He is currently as a Ph.D. student with the German Aerospace Center (DLR)
in Oberpfaffenhofen, Germany. His major research areas are digital beamforming
concepts with focus on signal processing, system design, and optimization
of multi-channel SAR systems for high performance imaging applications.
Further interests are the development of innovative and advanced concepts for
ultra-wide-swath imaging with SAR systems.
Gerhard Krieger (M’03) received the Dipl.-Ing. (M.S.) and Dr.-Ing. (Ph.D.)
degrees (with honors) in electrical and communication engineering from the
Technical University of Munich, Germany, in 1992 and 1999, respectively.
From 1992 to 1999, he was with the Ludwig-Maximilians University, Munich,
where he worked as an interdisciplinary research scientist on the modeling of
biological and technical vision systems. In 1999, he joined the Microwaves and
Radar Institute (HR) of the German Aerospace Center (DLR), Oberpfaffenhofen,
Germany, where he developed signal processing algorithms for an innovative
forward looking radar system. From 2001 to 2007 he led the New SAR
Missions Group which devised new bistatic and multistatic radar systems like
the forthcoming TanDEM-X mission as well as innovative digital beamforming
techniques for advanced multi-channel SAR. Since 2008, he has been Head of the
new Radar Concepts Department of the Microwaves and Radar Institute, DLR,
Oberpfaffenhofen, Germany. His research interests include innovative remote
sensing systems and applications based on radar interferometry, tomography, bi-
and multistatic satellite formations, digital beamforming, and the development of
advanced signal and image processing algorithms.
Alberto Moreira (M’92–SM’96–F’04) was born in Sõo José dos Campos,

Brazil, in 1962. He received the B.S.E.E. and the M.S.E.E. degrees, in 1984 and
1986, respectively, from the Aeronautical Technological Institute ITA, Brazil and
the Eng. Dr. degree (Honors) from the Technical University of Munich, Germany,
1993.
In 2003, he received a full professorship from the University of Karlsruhe,
Germany, in the field of microwave remote sensing. As its chief scientist and
engineer, he managed from 1996 to 2001 the SAR Technology Department of the
Microwaves and Radar Institute at the German Aerospace Center (DLR). Under
his leadership, the DLR airborne SAR system, E-SAR, has been upgraded to
operate in innovative imaging modes like polarimetric SAR interferometry and
SAR tomography. Since 2001, he is the director of the Microwaves and Radar
Institute at DLR. The Institute contributes to several scientific programs and
space projects for actual and future air- and space-borne SAR missions. Recently,
the mission proposal TanDEM-X lead by his Institute has been approved for
the realization phase. He is the principal investigator for this mission. His
professional interests and research areas encompass radar end-to-end system
design and analysis, innovative microwave techniques and system concepts, signal
processing, and remote sensing applications.
Dr. Moreira is serving as a member of the IEEE GRSS Administrative
Committee (1999—2001, 2004—2006), is the chair of the German Chapter
of the GRSS since 2003, and is actively serving as an associate editor for
the Geoscience and Remote Sensing Letters. Since 2003 he is also serving as
a member of the Board of Directors of the ITG (Information Technology
Society) of VDE (German Association for Electrical, Electronic and Information
Technologies). In 1995, he was the recipient of the DLR Science Award. He and
his colleagues received the GRSS Transactions Prize Paper Awards in 1997 and
2001, respectively. He is also the recipient of the IEEE Nathanson Award (1999)
and the IEEE Kiyo Tomiyasu Award (2007). He has contributed to the successful
series of the European SAR conferences (EUSAR) since 1996 as member of the
Technical Program Committee, Technical Chairman (2000), Awards Chairman
(2002—2004) and General Chairman (2006).

Gebert TAES2009 PDF

Uploaded by

Copyright:

Available Formats

Gebert TAES2009 PDF

Uploaded by

Document Information

Original Title

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

Gebert TAES2009 PDF

Uploaded by

Copyright:

Available Formats

I.

In the following, this paper focuses on the

A. Effective Phase Center, Virtual Sample Position, and

The phase center of an antenna represents the

In this section, an illustrative approach to explain

D. Signal Power = N 2 ¢ E[(A(f) ¢ rect(f=BD ))2 ]: (21)

E[jPj (f)j2 ] IV. SYSTEM DESIGN EXAMPLE

Orbit height hs 580 km § 10 km

Swath μi BRg (μi,min ) del,tx Gtx,max PRF ©bf

1 20.0—29:0± 439 MHz 0.19 m 38.7 dB 1340 Hz ¡0.92 dB

a nonoptimum PRF. By choosing the direction of

A. Extended System Model In the analogue representation of a preprocessing

signal is given by Uc,j (f) that forms the input of the ¯ K ¯

= pn,c,rx + pn,q : (60) (64)

Alberto Moreira (M’92–SM’96–F’04) was born in Sõo José dos Campos,

You might also like