Phase unwrapping

Rüdiger Gens
 Outline
y Basics
y Factors influencing the phase
y Terminology
Phase unwrapping

y Phase filtering
y Phase unwrapping algorithms
x Path-following methods
x Minimum-norm methods
y Weighting factors
y Trends and challenges

GEOS 639 – InSAR and its applications (Fall 2006) 2

 Importance of phase

y transmission or reception of coherent signals

y coherent processing
Phase unwrapping

x synthetic aperture radar (SAR)

x synthetic aperture sonar
x seismic processing
x adaptive optics
x magnetic resonance imaging (MRI)
x aperture synthesis radio astronomy
x optical and microwave interferometry

GEOS 639 – InSAR and its applications (Fall 2006) 3

 Relation to physical quantity

y in many applications the phase relates to a

physical quantity
x adaptive optics → wavefront distortion
Phase unwrapping

x MRI → degree of magnetic field inhomogeneity in

the water/fat separation problem
x astronomical imaging → relationship between the
object phase and its
bispectrum phase
x interferometry → surface topography

GEOS 639 – InSAR and its applications (Fall 2006) 4

 Optical and SAR interferometry

y Optical interferometry y SAR interferometry

x coherent signal source: x coherent signal source:
laser synthetic aperture radar
Phase unwrapping

x application: x primary application:

holography digital elevation models

Focus on phase unwrapping in

SAR interferometry

GEOS 639 – InSAR and its applications (Fall 2006) 5

 Why phase unwrapping?

y continuous phase information is sampled in a

discrete wrapped phase
y looking for the correct integer number of phase
Phase unwrapping

cycles that needs to be added to each phase

measurement to obtain the correct slant range
distance
y absolute phase is wrapped into the interval
(-π,+π] → ambiguity problem
y solving ambiguity referred to as phase
unwrapping
GEOS 639 – InSAR and its applications (Fall 2006) 6
 Factors influencing the phase

y phase aliasing → insufficient sampling rate

y phase noise
Phase unwrapping

y thermal noise → sensor electronics

y temporal change → different backscatter
y baseline geometry → fringe density

GEOS 639 – InSAR and its applications (Fall 2006) 7

 Influence of phase aliasing
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 8

 Influence of phase aliasing
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 9

 Influence of phase noise
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 10

 Influence of phase noise
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 11

 Influence of phase noise
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 12

 Influence of phase noise
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 13

 Phase unwrapping terminology

y phase gradient
y phase discontinuity
Phase unwrapping

y residue
y polarity
y charge
y branch cut

GEOS 639 – InSAR and its applications (Fall 2006) 14

 Phase gradients
Δ4 = -0.1 y small portion of
0.1 0.2 0.3 wrapped phase
image
Δ1 = -0.2 Δ3 = 0.4
y values divided
Phase unwrapping

-0.1 -0.2 -0.4 by 2π

Δ2 = -0.1 y phase gradients
defined as phase
difference of
-0.2 -0.2 -0.3 adjacent pixels

4
q = Σ Δi = 0
i =1
GEOS 639 – InSAR and its applications (Fall 2006) 15
 Inconsistencies and residues
y integrating
Δ4 = -0.1 wrapped phase
0.1 0.2 0.3 gradients around
every 2x2 sample
Δ1 = -0.4 Δ3 = -0.3
path in the entire
Phase unwrapping

image
-0.1 -0.2 -0.4
Δ2 = -0.2 y residue
(discontinuity)
if sum of phase
-0.2 -0.2 -0.3 gradients
not zero
4
q = Σ Δ i = -1
i =1
GEOS 639 – InSAR and its applications (Fall 2006) 16
 Polarities and charges

0.1 0.2 0.3

y non-zero
0 - integrals define
Phase unwrapping

residues
-0.1 -0.2 -0.4
y sign of the
residues define
0 0 polarity or
-0.2 -0.2 -0.3 charge of a
residue

GEOS 639 – InSAR and its applications (Fall 2006) 17

 Branch cuts

y connection of residues with opposite polarity

are referred to as branch cuts
y prevent any integration path from crossing the
Phase unwrapping

branch cuts
y residues and branch cuts are essential part of
path-following phase unwrapping methods

GEOS 639 – InSAR and its applications (Fall 2006) 18

 Ideal phase unwrapping case

y no residues (discontinuities) in the images

y integration of the phase gradients over the

Phase unwrapping

whole data set

y integration independent from integration path

GEOS 639 – InSAR and its applications (Fall 2006) 19

 Phase unwrapping reality

y phase noise
y phase discontinuities resulting in residues
Phase unwrapping

y high fringe rates in foreshortening and layover

regions → fringes cannot be separated
y shadow regions
→ no phase unwrapping possible at all
y integration not independent from its integration
path

GEOS 639 – InSAR and its applications (Fall 2006) 20

 Phase unwrapping reality

y white:non-integrated
y black: grounding
y
Phase unwrapping

purple: branch cut

y red: neg. residue
y blue: pos. residue
y yellow: integrated

GEOS 639 – InSAR and its applications (Fall 2006) 21

 Phase filtering

y interferogram power spectra

x “white” component generated by thermal noise and
loss of coherence
Phase unwrapping

x narrow band component related to fringes

y fringe rate determined by

x look angle
x along-track changes in the baseline
x any motion of the scene along the line of sight

GEOS 639 – InSAR and its applications (Fall 2006) 22

 Phase filtering

y approach developed by Goldstein and Werner

y adaptive filtering sensitive to

Phase unwrapping

x local phase noise

x fringe rate

y segmentation of interferogram into overlapping

rectangular patches

GEOS 639 – InSAR and its applications (Fall 2006) 23

 Phase filtering

y estimation of the power spectrum

x computing by smoothing the intensity of the
two-dimensional FFT
Phase unwrapping

y spatial resolution of the filter adapts to the local

phase variation
x regions of smooth phase are strongly filtered
x regions with high phase variance are weakly filtered

GEOS 639 – InSAR and its applications (Fall 2006) 24

 Phase filtering
α
H ( u, v ) = Z ( u, v )
⎧⎪ u2
− σ uσ v + σ 2 ⎫⎪
2 uv v2
Phase unwrapping

σ u2
Z ( u, v ) = exp⎨− v
⎬
⎪⎩ 2(1 − ρ ) ⎪
2
⎭
• H(u,v): adaptive filter
• Z(u,v): power spectrum

GEOS 639 – InSAR and its applications (Fall 2006) 25

 Unfiltered phase
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 26

 Filtered phase
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 27

 Phase unwrapping algorithms

y no standard procedure to solve the phase

unwrapping problem
y large variety of algorithms developed
Phase unwrapping

y generally trade off between accuracy of

solution and computational requirements
y two types of strategy to solve the phase
unwrapping problem
x path-following methods
x minimum-norm methods

GEOS 639 – InSAR and its applications (Fall 2006) 28

 Path-following methods

y local approach

y Goldstein`s branch cut algorithm

Phase unwrapping

y Flynn`s minimum discontinuity algorithm

y minimum cost flow (MCF) networks
y minimum spanning tree algorithm

GEOS 639 – InSAR and its applications (Fall 2006) 29

 Goldstein`s branch cut algorithm

y classical path-following method

y defines branch cuts between all detected
residues
Phase unwrapping

y algorithm prevents any integration path from

crossing these cuts
y residues need to be balanced
x connection with a residue of opposite polarity
x connection with the image border

GEOS 639 – InSAR and its applications (Fall 2006) 30

 Goldstein`s branch cut algorithm

y approach minimizes the sum of the branch cut

length
y algorithm
Phase unwrapping

x is computationally very fast

x requires little memory

y lack of weighting factors that could be used for

guiding the placement of branch cuts
→ poor performance in areas of low
coherence

GEOS 639 – InSAR and its applications (Fall 2006) 31

 Goldstein`s branch cut algorithm

y algorithm tends to create isolated regions by

closed branch cuts
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 32

 Goldstein`s branch cut algorithm

y several
enhancements
suggested
Phase unwrapping

x removal of so-called
dipoles
x phase filtering
reduces number of
residues
→ higher fringe
visibility
→ reduced phase
noise

GEOS 639 – InSAR and its applications (Fall 2006) 33

 Flynn`s minimum
discontinuity algorithm

y finds a solution that actually minimizes the

discontinuities
Phase unwrapping

y high memory and computational requirements

y tree-growing approach
x traces paths of discontinuity in the phase
x detects paths that form loops
x minimizes the discontinuities by adding multiple of
2π to the phase values enclosed by the loops

y works with or without weighting factors

GEOS 639 – InSAR and its applications (Fall 2006) 34
 Minimum cost flow networks
y formulates the phase unwrapping problem as
global minimization problem with integer
variables
x uses the fact that phase differences of neighboring
Phase unwrapping

pixels can be estimated with a potential error that is

an integer multiple of 2π

y optimization using MCF networks provides

position of branch cuts
y definition of costs assigned to flows within
network includes weighting factors in the
process
GEOS 639 – InSAR and its applications (Fall 2006) 35
 Minimum cost flow networks

y relatively new approach

y uses general purpose software packages
x MCF networks widely available
Phase unwrapping

x large field of research in itself

y designing MCF networks more adapted to the

specific constraints of phase unwrapping still a
major research issue

GEOS 639 – InSAR and its applications (Fall 2006) 36

 Minimum spanning tree algorithm

y adaptation of Goldstein`s algorithm

y approximates a minimum Steiner tree
Phase unwrapping

y builds a single tree containing all charges

x drawing branch cuts to next nearest charge to the
tree when charge of current tree becomes neutral

y definition of weights on phase gradients

x searching for the next charge to the tree with
Dijkstra`s shortest path algorithm

GEOS 639 – InSAR and its applications (Fall 2006) 37

 Minimum spanning tree algorithm

y cuts are associated with the phase

differences
x guarantees that the tree does not close on itself
Phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 38

 Minimum-norm methods

y global approach

y least-squares phase unwrapping

Phase unwrapping

y minimum Lp-norm phase unwrapping

GEOS 639 – InSAR and its applications (Fall 2006) 39

 Least-squares
phase unwrapping
y solution of phase unwrapping by discretized
partial differential equations (PDEs)
y least-squares favorable for solution of PDEs
Phase unwrapping

x solution leads to a linear equation

→ integrates the residues to minimize the gradient
differences

y works in weighted and unweighted form

GEOS 639 – InSAR and its applications (Fall 2006) 40

 Least-squares
phase unwrapping
y unweighted least-square problem described as
discretized Poisson equation that can be
solved by
Phase unwrapping

x Fast Fourier Transformations (FFTs)

x discrete cosine transforms (DCTs)
x unweighted multigrid method

y weighted least-squares approach requires

iterative methods
x Picard iteration method
x preconditioned conjugate gradient (PCG) method
x weighted multigrid method
GEOS 639 – InSAR and its applications (Fall 2006) 41
 Minimum Lp-norm
phase unwrapping
y generalization of weighted least-squares
approach
y requires solution of a non-linear PDE
Phase unwrapping

implemented in an iterative scheme

y double iterative structure makes algorithm
computationally very intensive
y generates data dependent weights (optional)

GEOS 639 – InSAR and its applications (Fall 2006) 42

 Weighting factors

y important feature for a large number of

algorithms for their improved performance
x also referred to as quality maps
Phase unwrapping

y define the quality of phase data on pixel level

y increasing interest with the introduction of
minimum cost flow networks
y various sources
y number of combinations countless

GEOS 639 – InSAR and its applications (Fall 2006) 43

 Sources for weighting factors

y correlation coefficient (coherence)

x enhanced and re-scaled

y pseudo-correlation
Phase unwrapping

x correlation with uniform magnitude

y phase derivative variance

x local sample variance of the partial derivatives of
the phase data

y maximum phase gradient

x magnitude of the largest phase gradient

GEOS 639 – InSAR and its applications (Fall 2006) 44

 Sources for weighting factors

y residue density
y flatness of unwrapped phase
Phase unwrapping

y smoothness of unwrapped phase

x sum of absolute values of the phase gradient

y statistically derived values

y masks used for excluding data from phase
unwrapping process

GEOS 639 – InSAR and its applications (Fall 2006) 45

 Trends and challenges

y complexity of many approaches increases the

demand in memory and computational
efficiency
Phase unwrapping

y improved hardware performance compensated

by size of data sets used
y results of shuttle radar topography mission
(SRTM) could improve phase unwrapping
results
y dealing with large volume data requires the
independence of human interaction
GEOS 639 – InSAR and its applications (Fall 2006) 46