Progress In Electromagnetics Research B, Vol.

76, 87–96, 2017

Advanced Multi-Pass InSAR Imaging for Surface

Deformation Studies

Sui Ping Lee* , Yee Kit Chan, and Tien Sze Lim

Abstract—This paper describes a multi-pass InSAR imaging approach for surface deformation studies.
Such a technique extends concept of SAR tomography (TomoSAR) based on multi-pass InSAR data, in
order to produce deformation map in elevation domain and velocity domain, respectively. Compared to
conventional InSAR method, multi-pass InSAR imaging technique acquires multi-baseline information
and allows reconstruction of multiple scattering sources in Tomo-Doppler plane (Elevation-Velocity
plane). This technique offers a solution to layover issue over conventional InSAR method, but it
suffers from double-scattering problem. This paper simulates a phenomenon where double-scattering
impairs the imaging process and an improved solution method to separate single and double scatters
from inferring pixels. In real circumstance, there are still other interferometric issues such as phase
ambiguities of noise and phase discontinuity. Thus, a phase-unwrapping method associated with an
improved ordered-statistical filter is included for interferometry processing. An experiment based on
real SAR data is set up to demonstrate this technique.

1. INTRODUCTION

Interferometry synthetic aperture radar (InSAR) is a renowned topography measurement technique

which employs pairs of SAR images and extracts spatial height information over large scale of illuminated
area. Today, almost all portions of earth surface have been modelled into the digital elevation map
(DEM) based on InSAR technique via shuttle radar topography mission (SRTM) [1, 2]. Measurement
of surface deformation can be done by repeating multiple passes of InSAR imaging process on the
same target area. Thus, surface deformation can be detected by comparing these InSAR images with
time separation that enables extraction of surface deformation over time, such an approach also known
as differential-interferometry synthetic aperture radar (D-InSAR). D-InSAR approach has been widely
implemented for deformation mapping. Its applications include and not limited to land use assessment,
landslide zone mapping, volcano area monitoring, and ice glacier research [3, 4]. Aforementioned InSAR
imaging approaches employ the dominant scattering mechanism by assuming the presence of single
scatter on each sample point. However, this assumption could not be hold on the layover area with high
density of scatterer such as sharp or vertical terrain slopes and complex building structures, which leads
to estimation error. According to the literature [5, 6], such InSAR errors (or referred as InSAR noise)
could be significantly improved, if the InSAR height estimation is performed by using multiple baseline
information generated from multiple passes of SAR observations. Subsequently, SAR tomography
(TomoSAR) [7] emerged as an evolvement technique of InSAR on layover issues migration over recent
decade. TomoSAR is able to reconstruct backscattering distribution along elevation profile (slant height)
based on multiple passes of InSAR images. Meanwhile, tomographic imaging approach suffers from the
issue of multiple scatterers, which might requires research effort for its solutions [8]. In recent years,
Received 20 February 2017, Accepted 5 June 2017, Scheduled 20 June 2017
* Corresponding author: Sui Ping Lee ([email protected]).
The authors are with the Faculty of Engineering and Technology, Multimedia University, Jalan Ayer Keroh Lama, Melaka 75450,
Malaysia.
88 Lee, Chan, and Lim

these processing approaches have evolved to more advanced InSAR techniques [9–13]. In this paper,
an advanced multiple-passes InSAR imaging approach is proposed for further extended TOMOSAR
concept in time domain and not only generates elevation outcome but also extracts velocity information
based on multiple passes of InSAR measurement (with multiple spatial and temporal baseline).
It is noted that double-bound scattering problem is unsolvable in TomoSAR [14, 15]. Also, phase
ambiguity problem has long been considered as one of the most critical issues in InSAR field [2]. In
order to extract more accurate deformation information, advanced multi-pass InSAR imaging techniques
attempt to tackle both of aforementioned issues. This paper simulates double-scattering condition and
demonstrates possibility to solve double-scattering problem by a simulation model. For real practices,
Delaunay Triangular method is applied to data pairing of more than two acquisitions. After pairing,
we suggest to include the InSAR model associated with improved order-statistical filter to overcome
the phase ambiguity problem due to interferometry processing [14]. After interferometry processing,
stack of unwrapped phase images would be rewriten into vector matrix form as collection of received
signal. In order to detect multiple scatterers, beam forming estimation is applied by using an inversion
matrix. A calibration method based on threshold is employed to adjust deformation images. The final
outcome of relevant approach would be an elevation mapping of deformation and a velocity mapping of
deformation.
This paper demonstrates the conceptual models and application of an advanced multi-pass InSAR
imaging for surface deformation studies. Section 2 describes the imaging process and relevant imaging
problem. Section 3 suggests a scheme for actual implementation on real data. Its implementation steps
are described. Simulation outcomes of double-scattering and its solution are presented in Section 4.
The experimental results based on real data are also discussed in this section. Section 5 concludes the
paper.

2. ADVANCED MULTI-PASS INSAR IMAGING AND ITS PROBLEM

STATEMENTS

This section describes extension concept of tomography imaging into aforementioned advanced multipass
InSAR imaging based on InSAR data analysis. Consider that there is N acquisition of SAR image
(n = N − 1). Thus, tn indicates time span between two acquisitions (also known as temporal baseline),
and bn refers to the distance of spatial baseline between two acquisitions. Equation (1) shows the signal
model of TOMOSAR modelled as superposition of multiple backscatters’ sources.
 lk−1
Sn = γ(l) exp (−j2πξn l) dl (1)
li
Sn refers to received signal, and γ indicates backscattering coefficient. The slant height or elevation is
indicated by l. r denotes target distance from reference track, and λ is wavelength. Spatial frequency
can be computed as ξn = 2b n
rλ . Eq. (1) is the Fourier transform of γ(l) at ξn frequency. It is noted that
irregular sample is the main problem of elevation imaging. Equation (2) describes a linear problem after
discretization [17].
S = Aγ̄ (2)
By rewriting the relevant variables and parameters after discretization in matrix form: collection of
received signal is the vector represented by S = [S0 , . . . , SN −1 ]T , and measured complex data are now
indicated by γ̄ = [γ(l0 ), . . . , γ(lN −1 )]T . Elevation is in elevation bin form as li where i = 0, . . . , k − 1. A
is an N × k steering vector, A = [a(l0 ), . . . , a(lk−1 )] where a(lp ) = exp(j2π ε̄lp ) and ε̄ = [ξ0 , . . . , ξN −1 ]T .
In order to extract the deformation map, the advanced multipass InSAR technique extends
Equation (1) to Equation (3) where deformation term of d(l, tn) is extracted.
 lk−1  
4π
Sn = γ(l) exp (−j2πξn l) exp −j d(l, tn ) dl (3)
li λ
By further extending exp(−j 4π λ d(l, tn )) in Fourier harmonics as Equation (4) with velocity frequency of
ηn = 2tλn ,
   lk−1  vk−1
4π
exp −j d(l, tn ) = X(l, v) exp(−j2πηn v)dl dv (4)
λ li vi
Progress In Electromagnetics Research B, Vol. 76, 2017 89

Consider that γ(l)X(l, v) is substituted into γ(l). Thus, received signal of the advanced multipass
InSAR imaging approach is modelled as Equation (5)
 lk−1  vk−1
Sn = γ(l, v) exp (−j2π (ξn l + ηn v)) dldv (5)
li vi
Similar to tomography model, scatterer detection becomes a problem where irregular samples infer on
sample pixels. It is necessary to perform two-dimensional Fourier-Inversion to reconstruct backscattering
distribution in elevation-velocity plane. Discretization leads to the same linear problem as Equation (2),
but its steering model [17] is described as Equation (6) where η̄ = [η0 , . . . , ηN −1 ]T .
  
a(lp , vq ) = exp j2π ξ̄lp + η̄vq (6)
In real condition, received signal is always inferred by more scatterers that target detection could be
done by analyses on the peaks of backscattering distribution in E-V plane (l, v). Deformation signal is
constituted by linear deformation and nonlinear deformation dUL [17].
d(l, tn ) = vtn + dUL (l, tn ) (7)
The estimation error of relevant deformation is considered as ed . In this scenario, the total estimation
error, etotal , covers the phase estimation error during InSAR processing, eInSAR . This can be expressed
as Equation (8).
etotal = ed + eInSAR (8)

3. EXPERIMENTAL SET UP AND ITS IMPLEMENTATION STEPS

Figure 1 shows the proposed implementation scheme for real data experiment for the multipass
InSAR imaging approach with N acquisitions of SAR images. The following paragraph presents the
implementation steps of relevant approach. Each acquisition is referred to a single SAR image acquired
on certain times pan or path track.

Figure 1. Implementation scheme of advanced multi-pass InSAR imaging.

90 Lee, Chan, and Lim

Step 1: SAR images pairing by Delaunay Triangular method [18]. During interferometry processing,
formation of each interferogram requires a pair of SAR images. It is noted that the possible matching
combinations are more than N − 1 though only N − 1 matched pairs are required in the next step. Thus,
pairing process is optimized by referring to a triangular network structure that computes temporal
baseline and spatial baseline. The computation outcome distributed in temporal-spatial diagram is
shown as Figure 2 where the pairings are selected by referring to the triangular vector among red nodes
(red node indicates each acquisition).

Figure 2. Network structure plotted by Delaunay Triangular.

Step 2: Interferometry processing by improved-order statistical filter. During each pass of operation,
decorrelation phenomena due to nonlinear deformation and atmosphere noise create phase ambiguity
issue for interferometry processing. However, such a multipass InSAR imaging approach could only
be possible with the condition that data stacks are well-compensated. Thus, this step incorporates a
phase-unwrapping method associated with an improved order-statistical filter (which is newly developed
by [16]) to overcome aforementioned issue. The relevant working mechanisms are as follow. After the
pairing of SAR images from Step 1, each interferogram, Φ, is formed by extraction of phase value
difference between relevant matched SAR images xi and xj .
Φ = arg(xi xj ∗ ) (9)
()∗ indicates conjugate value, and arg() refers to arc tangent angle value. Such an interferogram is then
filtered by minimum order statistical filter which varies its filtering rate based on the coherency weight,
wc . The coherency weights are determined by the correlation level between xi and xj .
 
| xi xj ∗ |
wc = ∈ [0, 1] (10)
 xi  xj 
The larger local window size (3×3) is applied to low coherency area while smaller local window size (1×1)
is applied to high coherency area. The pixel value with correlation level less than 0.3 is considered total
corrupted since it could not provide useful information [19]. Equation (11) shows an improved-order
statistical filter that preserves useful information and restores corrupted pixel (with correlation level
lower than 0.3). The minimum order statistical filter [21] sorts pixel values (phase difference) within
Progress In Electromagnetics Research B, Vol. 76, 2017 91

its local window and apply restoration based on the minimum value. The local window is denoted by
windowΦ .
⎧
⎪
⎪ {Φ(x, y)} for wc  0.3
⎪
⎪
⎪
⎪ ⎧ ⎫
⎪
⎪ ⎪ Φ(x, y), ⎪
⎪
⎪ ⎪ ⎪
⎪
⎪ ⎪
⎪ Φ(x − 1, y − 1), ⎪⎪
⎪
⎪ ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪ − ⎪
⎪
⎨ ⎪ ⎪ Φ(x 1, y),
⎪ Φ(x − 1, y + 1), ⎪
⎪
⎪
f (x, y) = min {Φ(x, y)} , given that, windowΦ = ⎪
⎨ ⎪
⎬
(x,y)∈windowΦ ⎪
⎪ −
⎪
⎪ ⎪
Φ(x, y 1),
⎪
for wc < 0.3
⎪
⎪ ⎪
⎪ Φ(x, y + 1), ⎪⎪
⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪ + 1, y − 1), ⎪
⎪
⎪ ⎪
⎪ Φ(x ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ Φ(x + 1, y), ⎪
⎪
⎩ ⎩ ⎭
Φ(x + 1, y + 1)
(11)
Then, 2D Itoh-based phase unwrapping [20] is implemented into filtered outcome of the minimum
ordered-statistical filter for absolute phase estimation. Φ̂ refers to estimated phase value. kcolumn and
krow refer to the jump cycle counts adjusted due to adjacent pixel difference in horizontal term and
vertical term [19].  

Φ̂(x, y) = Φ(x, y) + 2π kcolumn (x, y) + krow (x, y) (12)
A first-order neighbour averaging filter (with 3 × 3 window) is employed for noise suppression purpose.
This step repeats N − 1 times for the interferogram stack and generates N − 1 outcomes of unwrapped
phase image (stack of output images, S). This step has generated stack of enhanced deformation
information directly related to spatial difference and mean velocity estimation.

Step 3: Deformation mapping by multi-pass imaging approach. Section 2 describes the imaging process
of discretization as Equation (2). To overcome the linear problem after discretization, beam forming
can be done by back-projection in Equation (2). The estimation of scattering count γ̂ is modelled as
Equation (13) where ()H refers to Hermitian.
γ̂ = AH S (13)
Since the estimation model of backscatters is considered as a function responding to elevation and
velocity [21], maximum value of estimated scatterer becomes the main reference to generate the elevation
map, l, and velocity map, v. Such a process can be modelled into a normalized correlation term as
Equation (14).  
| aH S |
C = max ∈ [0, 1] (14)
l,v  aH  S 

Step 4: Calibration of deformation map. By referring to the correlation term of Equation (14), two
deformation images which cover an elevation map and a velocity map can be extracted as follows.
  
ˆ | aH S |
(l, v̂) = arg max (15)
l,v  aH  S 
These images may still contain noise on the low coherency area, i.e., the place with high density scatterers
such as vegetation and road with moving vehicles. In order to establish ready-to-use outcomes for surface
deformation studies, the noisy and less accurate part should be removed. Such calibration can be easily
done by setting up a threshold filter to filter out unwanted information based on its coherency level.
C ≷ Threshold ∈ [0, 1] (16)
The relevant threshold is set by referring to desired value of false alarm probability, PFA under following
hypotheses test.
H0 : S = w
(17)
H1 : S = γ1 (l)a(l1 , v1 ) + γ2 (l)a(l2 , v2 ) + w
92 Lee, Chan, and Lim

The initial hypothesis H0 refers to the circumstance of only background noise (without presence of
any scatterer) where w indicates N × k disturbance samples modelled as zero mean complex circular
Gaussian as stated in [22]. The alternative hypothesis test H1 shows the presence of double scatterers
with noise. In order to detect the targets without background noise, H0 is tested against the alternative,
H̄0 . PF A is defined as below.
 
PF A = P H̄0 | H0 = P (C  Threshold | H0 ) (18)
It is noted that the false alarm probabilities can be very different and varying on different data sets.
Thus, the threshold can also be empirically set for many cases of real data experiment.
The working principal of double scatterers detection approach is described as above where an
inversion matrix (Equation (13)) is applied to reconstruct target scatterers. If above working steps
are implemented without the choice of inversion matrix (by skipping Equation (13)), the correlation
term will be compared to a threshold that identifies single scatterer. Such procedures form a technique
similar to standard permanent scatterer identification (PSI), but it is framed in advanced multi-pass
imaging approach [23].

4. RESULTS AND ANALYSES

4.1. Simulation Results

This section presents the simulation result in noise free condition. Figure 3 shows a simulation model
of beam forming process where back scatterer is reconstructed to a function of elevation and velocity.
Figure 4 shows simulation results of double-scattering mechanism of two different height targets. Target
scatters could be well-estimated and separated from double-scatter mechanism in the first simulation
scene of Figure 4. The second simulation scene of Figure 4 shows an estimation error case of double-
scattering. This is because the target heights between two these scatters is too close to be over the limit
of baseline resolution. Thus, this is the constraint of the current solution model.

Figure 3. Simulation of beam forming.

4.2. Experimental Results Based on Real Data

The experimental results based on real SAR data are shown in this section. The input data are
constituted of 29 acquisitions of SAR observations on the same target area. The master image of
the SAR acquisition is shown in Figure 5, which is the street view from satellite radar on Rome, Italy.
Both Figure 6 and Figure 9 show coherency mapping of input data stacks on step 3 (based on single-
scatterer and double-scatterer detections, respectively). Figure 7 and Figure 8 display deformation
map (in term of elevation and velocity) via single-scatterer detection. The advanced multi-pass InSAR
Progress In Electromagnetics Research B, Vol. 76, 2017 93

Figure 4. Solution model for double-scattering and its constraint.

Figure 5. Real data image of SAR observation Figure 6. Correlation mapping based on single
on target area. scatterer detection.

imaging techniques refer to an image extraction approach based on double-scatterer detection, and its
outcomes of deformation map are displayed in elevation term (Figure 10) and velocity term (Figure 11)
as well.
From the amplitude image (Figure 5) and coherence image (Figure 6), we can observe that
stationary building area and concrete infrastructure show high correlation index above 0.6 while the
roads’ busy traffic shows low correlation index. In this experiment, we focus on observing the hard
construction buildings. Thus, busy walking streets paths and roads with traffic will be considered as
background noise. 0.6 is considered as the threshold value for false alarm probability. As a result, we
can notice that all of the road area with moving traffic is totally filtered away during calibration step.
94 Lee, Chan, and Lim

Figure 7. Deformation map in elevation based on Figure 8. Deformation map in velocity based on
single scatterer detection. single scatterer detection.

Figure 9. Correlation mapping based on double scatterers detection..

Figure 10. Deformation in elevation based on Figure 11. Deformation in velocity based on
double scatterers detection. double scatterers detection.

Based on the processing outcomes, correlation map of double-scatterer detection (Figure 9) shows
higher correlation level than correlation map of single-scatterer detection (Figure 6). From Figure 7 and
Figure 8, we can still observe that the noise whereby the noise level is high on low coherency area (the
Progress In Electromagnetics Research B, Vol. 76, 2017 95

darker part of coherency map of Figure 6). Compared with Figure 10 and Figure 11 (double-scatterer
detection), the quality of single-scatterer detected deformation images seems degraded. Overall, the
images generated by double-scatterer detection increase sensitivity level of similarity index and yield
outcome images with better accuracy.
In these deformation images, red colour part indicates the area with higher deformation rate while
deep blue indicates slow or stationary deformation. From the elevation map, we discover that the red
colour areas are mostly formed by the place with vegetation. Since the vegetation areas contribute
to high volume of scatters, the portion of red colour indicates inferring source (instead of fast surface
deformation). Figure 11 of velocity map presents the velocity information of surface deformation on
buildings. Apparently, such an imaging technique works better on stationary buildings and concrete
infrastructure.

5. CONCLUSION

The modelling process of advanced multi pass InSAR is described in this paper. The issues of double
scatterings are investigated, and a solution method is simulated. A scheme is proposed when the
implementation steps are described. The real data experiment based on double scatterers detection is
demonstrated for deformation mapping. Compared to single-scatterer detection, advanced multi-pass
approach can provide InSAR outcome images with better accuracy. Such techniques are also appropriate
for being applied to buildings or bridge monitoring.

ACKNOWLEDGMENT

The authors would like to thank the reviewers for valuable comments and express the gratitude to Diego
Reale and Gianfranco Fornaro for providing real satellite data training through International Summer
School on Radar/SAR 2016, organised by Fraunhofer Institute.

REFERENCES

1. Massonnnet, D. and K. L. Feigl, “Radar interferometry and its application to changes in earth
surface,” Rev. Geophsys., Vol. 36, No. 4, 441–500, Nov. 1998.
2. Roland, B., P. A. Rosen, and E. J. Fielding, “Synthetic aperture radar interferometry to measure
Earth’s surface topography and its deformation,” Annu. Rev. Earth Planet. Sci., Vol. 28, 169–209,
May 2000.
3. Rosen, P. A., S. Hensley, I. R. Joughin, S. N. Madsen, E. Rodriguez, and R. M. Goldstein,
“Synthetic aperture radar interferometry,” Proceeding of IEEE , Vol. 88, No. 3, Mar. 2000.
4. Tough, R. J. A., “Interferometric detection of sea surface features,” Signals and Radar
Establishment, memorandum 4446, UK, 1991.
5. Wu, B. I., M. C. Yeuing, Y. Hara, and J. A. Kong, “Insar heigh inversion by using 3-D phase
projection with multiple baselines,” Progress In Electromagnetics Reseach, Vol. 91, 173–193, 2009.
6. Ren, X. Z., L. H. Qiao, and Y. Qin, “A three-dimensional imaging algorithm for tomography
SAR based on improved interpolated array transform1,” Progress In Electromagnetics Research,
Vol. 120, 181–193, 2011.
7. Gini, F., F. Lombardini, and M. Montanari, “Layover solution in multibaseline SAR
interferometry,” IEEE Trans. Aerosp. Electron. Syst., Vol. 38, No. 4, 1344–1356, Oct. 2002.
8. Zhang, W., “Two-dimensional microwave tomographic algorithm for radar imaging through
multilayered media,” Progress In Electromagnetics Research, Vol. 144, 261–270, 2014.
9. Lombardini, F., “Differential tomography: A new framework for SAR interferometry,” IEEE Trans.
Geosci. Remote Sens., Vol. 43, No. 1, 37–44, Jan. 2005.
10. Reale, D., G. Fornaro, A. Pauciullo, X. Zhu, and R. Bamler, “Tomographic imaging and monitoring
of buildings with very high resolution SAR data,” IEEE Geosci. Remote Sens. Lett., Vol. 8, No. 4,
661–665, Jul. 2011.
96 Lee, Chan, and Lim

11. Reale, D., D. O. Nitti, D. Peduto, R. Nutricato, F. Bovenga, and G. Fornaro, “Postseismic
deformation monitoring with the COSMO/SKYMED constellation,” IEEE Geosci. Remote Sens.
Lett., Vol. 8, No. 4, 696–700, Jul. 2011.
12. Fornaro, G., A. Pauciullo, and F. Serafino, “Deformation monitoring over large areas with
multipass differential SAR interferometry: A new approach based on the use of spatial differences,”
International Journal of Remote Sensing, Vol. 30, No. 6, 1455–1478, Apr. 2009.
13. Berardino, P., G. Fornaro, R. Lanari, and E. Sansosti, “A new algorithm for surface deformation
monitoring based on small baseline differential SAR interferograms,” IEEE Trans. Geosci. Remote
Sens., Vol. 40, No. 11, 2375–2383, Nov. 2002.
14. Fornaro, G., D. Reale, and F. Serafino, “Four-dimensional SAR imaging for height estimation and
monitoring of single and double scatterers,” IEEE Trans. Geosci. Remote Sens., Vol. 47, No. 1,
224–237, Jan. 2009.
15. Solimene, R., A. Brancaccio, R. Di Napoli, and R. Pierri, “3D sliced tomographic inverse scattering
experimental results,” Progress In Electromagnetics Research, Vol. 105, 1–13, 2010.
16. Lee, S. P., Y. K. Chan, T. S. Lim, and V. C. Koo, “An InSAR phase estimation method for
surface derformation studies,” International Conference on Space, Aeronautical and Navigational
Electronics 2016, IEICE Tech. Rep., Vol. 116, No. 319, 109–114, SANE2016-73, Nov. 2016.
17. Reale, D., G. Fornaro, and A. Pauciullo, “Extension of 4-D SAR imaging to the monitoring of
thermally dilating scatterers,” IEEE Trans. Geosci. Remote Sens., Vol. 51, No. 12, Dec. 2013.
18. Mora, O., J. J. Mallorqui, and A. Broquetas, “Linear and nonlinear terrain deformation maps from
a reduced set of interferometric SAR images,” IEEE Trans. Geosci. Remote Sens., Vol. 41, No. 10,
2243, 2253, 2003.
19. Wei, M. and T. Sandwell, “Decorrelation of L-band and C-band interferometry over vegetated areas
in california,” IEEE Trans. Geosci. Remote Sens., Vol. 48, No. 7, 2942–2952, 2010.
20. Ghighlia, D. C. and M. D. Pritt, Two-Dimensional Phase Unwrapping: Theory,Algorithms, and
Software, 1st Edition, Wiley-Interscience, New York, 1998.
21. Stella, A. and B. Trivedi, “Implementation of order statistic filters on digital image: A comparative
study,” International Journal of Modern Engineering Research, Vol. 2, No. 5, 3143–3145, 2012.
22. De Maio, A., G. Fornaro, and A. Pauciullo, “Detection of single scatterers in multidimensional
SAR imaging,” IEEE Trans. Geosci. Remote Sens., Vol. 47, No. 7, 2284–2297, Jul. 2009.
23. De Maio, A., G. Fornaro, A. Pauciullo, and D. Reale, “Detection of double scatterers in SAR
tomography,” IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2009 ,
Vol. 3, III-172, IEEE, Jul. 2009.