- •Contents
- •Preface
- •1.1 Smart Antenna Architecture
- •1.2 Overview of This Book
- •1.3 Notations
- •2.1 Single Transmit Antenna
- •2.1.1 Directivity and Gain
- •2.1.2 Radiation Pattern
- •2.1.3 Equivalent Resonant Circuits and Bandwidth
- •2.2 Single Receive Antenna
- •2.3 Antenna Array
- •2.4 Conclusion
- •Reference
- •3.1 Introduction
- •3.2 Data Model
- •3.2.1 Uniform Linear Array (ULA)
- •3.3 Centro-Symmetric Sensor Arrays
- •3.3.1 Uniform Linear Array
- •3.3.2 Uniform Rectangular Array (URA)
- •3.3.3 Covariance Matrices
- •3.4 Beamforming Techniques
- •3.4.1 Conventional Beamformer
- •3.4.2 Capon’s Beamformer
- •3.4.3 Linear Prediction
- •3.5 Maximum Likelihood Techniques
- •3.6 Subspace-Based Techniques
- •3.6.1 Concept of Subspaces
- •3.6.2 MUSIC
- •3.6.3 Minimum Norm
- •3.6.4 ESPRIT
- •3.7 Conclusion
- •References
- •4.1 Introduction
- •4.2 Preprocessing Schemes
- •4.2.2 Spatial Smoothing
- •4.3 Model Order Estimators
- •4.3.1 Classical Technique
- •4.3.2 Minimum Descriptive Length Criterion
- •4.3.3 Akaike Information Theoretic Criterion
- •4.4 Conclusion
- •References
- •5.1 Introduction
- •5.2 Basic Principle
- •5.2.1 Signal and Data Model
- •5.2.2 Signal Subspace Estimation
- •5.2.3 Estimation of the Subspace Rotating Operator
- •5.3 Standard ESPRIT
- •5.3.1 Signal Subspace Estimation
- •5.3.2 Solution of Invariance Equation
- •5.3.3 Spatial Frequency and DOA Estimation
- •5.4 Real-Valued Transformation
- •5.5 Unitary ESPRIT in Element Space
- •5.6 Beamspace Transformation
- •5.6.1 DFT Beamspace Invariance Structure
- •5.6.2 DFT Beamspace in a Reduced Dimension
- •5.7 Unitary ESPRIT in DFT Beamspace
- •5.8 Conclusion
- •References
- •6.1 Introduction
- •6.2 Performance Analysis
- •6.2.1 Standard ESPRIT
- •6.3 Comparative Analysis
- •6.4 Discussions
- •6.5 Conclusion
- •References
- •7.1 Summary
- •7.2 Advanced Topics on DOA Estimations
- •References
- •Appendix
- •A.1 Kronecker Product
- •A.2 Special Vectors and Matrix Notations
- •A.3 FLOPS
- •List of Abbreviations
- •About the Authors
- •Index
|
|
DOA Estimations with ESPRIT Algorithms |
113 |
||||||||
|
|
|
|
|
|
|
|
|
|
||
( |
|
|
μ |
i |
|
( |
|
|
|
||
Γ |
B )b |
( μ ) tan |
|
|
= Γ |
B )b |
|
( μ ), 1≤ i ≤ d |
(5.82) |
||
|
|
|
|||||||||
1 |
|
L i |
2 |
2 |
|
L |
i |
|
|||
|
|
|
|
|
|
|
|||||
Beam patterns of a ULA with M = 8 sensors was computed. A DFT |
|||||||||||
beamspace transformation W 8h |
was employed to form L = 3 beams. |
When rows 4, 5, and 6 were employed to form the beams, the corresponding beam patterns b1(μi), b4(μi), and b5(μi) are obtained as shown in Figure 5.6(a). Another computation was made to verify that first and last components are physically adjacent to each other. Rows 1, 2, and 8 of W 8h are employed to form L = 3 beams. The corresponding beam pat-
terns are adjacent as shown in Figure 5.6(b). Employing these insights, the following sections explain the working of unitary ESPRIT in DFT beamspace.
5.7 Unitary ESPRIT in DFT Beamspace
In this section, the unitary ESPRIT algorithm is applied to uniform linear arrays in DFT beamspace. A real-valued beamspace array steering matrix is obtained by choosing the reference point at the center of the array and by applying beamspace transformation. This real-valued transformation results in a considerable reduction in computational complexity and is readily extended to uniform rectangular array [9].
5.7.1One-Dimensional Unitary ESPRIT in DFT Beamspace
Consider a ULA of M elements with maximum overlap. Assume the data and signal model discussed in Chapter 3 and an L-dimensional beamspace. The unitary ESPRIT in DFT beamspace can be formulated in the following steps.
5.7.1.1 Transformation to Beamspace
The element space data vector X is first transformed into the beamspace data vector Y as
Y = W b X |
(5.83) |
The transformation to DFT beamspace implemented using an FFT that exploits the
Y = W b X C M × N can be Vandermonde form of the
114 |
Introduction to Direction-of-Arrival Estimation |
Beam patterns for ULA and standard DFT beams
8 |
6 |
4 |
2 |
0 |
−2 |
|
|
|
Beampatterns |
Beam patterns for ULA and standard DFT beams
8 |
6 |
4 |
2 |
0 |
−2 |
|
|
|
Beampatterns |
8 |
1 |
2 |
|
b |
b |
b |
|
· |
- |
|
|
|
|
||
· |
|
|
|
|
- |
|
|
· |
|
|
|
· |
- |
|
|
· |
- |
|
|
|
|
|
|
−4 |
−6 |
4 |
5 |
6 |
|
b |
b |
b |
|
· |
- |
|
|
|
|
||
· |
|
|
|
|
- |
|
|
· |
|
|
|
· |
- |
|
|
· |
- |
|
|
−4 |
−6 |
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−8
−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 1
−8
(b)
(a)
(a, b) Beam pattern of an eight-element uniform linear array.
Figure 5.6
DOA Estimations with ESPRIT Algorithms |
115 |
|
|
rows of the FFT matrix, followed by an appropriate scaling of the rows of resulting matrix.
5.7.1.2 Real-Valued Signal Subspace Estimation
The unitary transformation is now applied as discussed in Section 5.4. An SVD of the real-valued matrix
[Re Y Im Y ] R M × 2N with Y = YW b X C M × N |
(5.84) |
is computed to estimate the d left singular vectors that correspond to the d largest singular values of (5.84); otherwise, an EVD of
Re{X}Re{Y}H + Im{X}Im{Y}H |
(5.85) |
can also be computed (i.e., covariance approach). This transformation, as described in Section 5.4, automatically achieves forward-backward averaging.
5.7.1.3 Real-Valued Invariance Equation
Let the d singular vectors corresponding to the d largest singular values of (5.84) be denoted by Es R M × d . Asymptotically, the real-valued matri-
ces Es and B span the same d-dimensional signal subspace, so there is a nonsingular matrix TA such that
|
|
B ≈ E s T A |
(5.86) |
||
Substituting this relation into (5.45), we have the real-valued |
|||||
invariance equation |
|
|
|
|
|
Γ E |
s |
Υ ≈ Γ |
2s |
R M × d |
|
1 |
|
|
|
||
where |
|
|
|
|
|
|
Υ = T A ΩT A−1 |
(5.87) |
which is solved by LS or TLS as explained in previous sections.
116 |
Introduction to Direction-of-Arrival Estimation |
5.7.1.3 DOA Estimation
The spatial frequencies or DOAs are then estimated by taking eigenvalues of the solution of the invariance equation given by
Γ = T A ΩT A−1
where
Ω = diag{ω1 , ω 2 , , ωi , , ω d }
contains the desired DOA information. The whole algorithm that operates in an L-dimensional DFT beamspace is summarized in Table 5.4 and its performance is analyzed in Chapter 6 [1].
In the next section, the concept of DFT beamspace is extended to two-dimensional arrays.
5.7.2Two-Dimensional Unitary ESPRIT in DFT Beamspace
In this section, DFT beamspace technique is applied to two-dimensional arrays, more specifically to uniform rectangular arrays. Consider a uniform rectangular array of size M = Mx × My. Assume the signal and data
Table 5.4
Summary of One-Dimensional Unitary ESPRIT in DFT Beamspace
1.Transformation to Beamspace: Y = WbX
2.Signal Subspace Estimation: Compute Es
as the d dominant left singular vectors of [Re{Y} Im{Y}] (square-root approach)
or as the d dominant eigenvectors of Re{X}Re{Y}H + Im{X}Im{Y}H (covariance approach)
3. Solution of the Invariance Equation: Solve the following equation for ϒ
γ1(B )Es ϒ ≈ Γ2(B )Es
by means of LS or TLS
4. DOA Estimation: Calculate the eigenvalues of the real-valued solution
|
|
|
|
|
|
|
γ = TΩT−1 with Ω = diag{ωi}id=1 |
Then extract the DOA angular information via |
|||||||
μ |
i |
|
( |
i ) |
≤ i ≤ d |
||
|
= 2 arctan |
ω |
, 1 |
||||
|
|
|
|
λ |
|
|
|
θ i |
= arcsin |
− |
|
|
μi |
|
|
2π |
|
||||||
|
|
|
|
|
|
|
DOA Estimations with ESPRIT Algorithms |
117 |
|
|
model discussed in Chapter 3. By following the similar line for beamspace processing of one-dimensional ULA, we can compute the two-dimen- sional DFT beamspace ESPRIT in a straightforward manner.
5.7.2.1 Transformation to Beamspace
Consider the case of reduced dimensionality directly and assume that Lx out of Mx beams in the x direction and Ly out of My beams in the y direc-
tion are formed, respectively. The total number of beams is L = Lx × Ly. |
|
The |
corresponding scaled DFT matrices W Lb x C B x × M x and |
W b |
C B y × M y are formed as discussed in Section 5.6.2. |
L y |
|
Let the noise corrupted signals or data received by the array at any given time tn be given by the matrix χ(t n ) C M x × M y . Premultiplying this
matrix by W b |
and postmultiplying by W b |
and applying the vec{•} |
|||
L x |
|
|
L y |
||
operator, we can have |
|
|
|
|
|
y (t n ) = vec{W LHx |
Χ(t n ) |
|
Lb y } C B , L = L x L y (5.88) |
||
W |
We place this column vector as a column of the matrix Y C L×N. The vec{•} operator maps a Bx × By matrix to a B × 1 vector by stacking the columns of the matrix. Again by applying the property of Kronecker products (Appendix, Section A.1) to (5.88), we get
Y = [y (t 1 ) y (t 2 ) y (t n )] |
|
|
)] |
|
||||||||
= W b |
W b |
[ |
x(t |
1 |
) |
x(t |
1 |
) |
x(t |
n |
(5.89) |
|
( B y |
B x ) |
|
|
|
|
|
|
|
||||
= (WBby |
WBbx ) X C B × N |
|
|
|
|
|
Signal subspace now can be estimated from this data.
5.7.2.2 Real-Valued Subspace Estimation
Let Es denote the real-valued signal subspace. The columns of Es RL×d contain the d left singular vectors of
[ |
Im{Y} |
] |
R L × 2N |
(5.90) |
Re{Y} |
|
118 |
Introduction to Direction-of-Arrival Estimation |
corresponding to its d largest singular values. Otherwise, using EVD, the d dominant eigenvectors of
Re{Y}Re{Y}H + Im{Y}Im{Y}H |
(5.91) |
spans the real-valued signal space.
5.7.2.3 Real-Valued Invariance Equation
The transformed real-valued array manifold matrix is now given by
B ( μ |
,v |
|
) = W b |
A( μ |
,v |
|
) |
|
b |
|
|
|
|
|||||||||
i |
i |
W |
|
|
|
|
||||||||||||||||
i |
|
|
|
L x |
|
|
i |
|
|
|
|
|
|
L y |
|
|
|
|
||||
|
|
|
|
= W b |
a |
( μ |
|
)aT |
(v |
|
) |
|
b |
(5.92) |
||||||||
|
|
|
|
i |
i |
W |
||||||||||||||||
|
|
|
|
|
L x |
|
|
|
M x |
|
|
|
M y |
|
|
|
|
L y |
|
|||
|
|
|
|
= b |
( |
μ |
i |
)bT (v |
i |
|
) R L x × L y |
|
|
|||||||||
|
|
|
|
|
L x |
|
|
L y |
|
|
|
|
|
|
|
|
|
|
|
|
||
where |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
b L x ( μi ) = W Lb x a M x |
( μi ) R L x |
|
and b L y |
(vi ) = W Lv y |
a M y |
(vi ) R L y |
are the one-dimensional DFT beamspace manifold vectors in a reduced
dimensional space. |
The one-dimensional |
array |
steering vectors |
a M x (μi ) C M x and |
a M y (ν i ) C My are the |
same as |
those defined in |
(2.24).
Consider the one-dimensional DFT beamspace vectors in the x direction. As bL x (μi ) satisfies the invariance relationship in (5.82), it follows that B(μi, νi) satisfies
Γ (L x )B ( μ ,v ) tan |
|
μi |
|
= Γ (L x )B ( μ ,v |
|
) |
(5.93) |
||
|
2 |
|
|
||||||
1 |
i i |
2 |
i |
i |
|
|
|||
|
|
|
|
|
where the one-dimensional selection matrices in the x direction, Γ1( L x ) and Γ2( L x ) , were defined as before for the one-dimensional DFT beamspace. By using the property of the vec{•} operator in the Appendix, Section A.1, L = Lx × Ly dimensional DFT beamspace steering vectors b(μi, νi) = vec{B(μi, νi)}satisfy
|
DOA Estimations with ESPRIT Algorithms |
119 |
|||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
μ |
i |
|
|
|
|
|
|
|
Γμ1b( μi |
,vi |
) tan |
|
|
|
= Γμ2b( μi ,vi |
), 1≤ i ≤ d |
(5.94) |
|||||
|
|
|
|||||||||||
|
|
|
|
|
2 |
|
|
|
|
|
|
||
where the two-dimensional selection matrices in the x direction |
|
||||||||||||
Γ |
μ1 |
= I |
Γ (L x ) |
andΓ |
μ2 |
= I |
L y |
Γ (L x ) |
(5.95) |
||||
|
|
L y |
1 |
|
|
|
|
2 |
|
are of size bx × L with bx = (Lx − 1) Ly. Similarly, the one-dimensional DFT beamspace manifold vectors in the y direction, bL y (ν i ), satisfy
|
|
(L y ) |
|
(v |
|
) |
|
|
|
vi |
|
|
|
(L y ) |
|
|
|
(v |
|
) |
|
||||||||
|
Γ |
|
|
b |
|
tan |
|
|
|
= Γ |
|
|
|
b |
|
|
|
(5.96) |
|||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||||||||||
|
1 |
|
|
|
L y |
|
i |
|
|
|
|
2 |
|
|
|
2 |
|
|
|
L y |
|
i |
|
|
|
||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||
such that |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
B ( μ |
|
,v )Γ |
(L |
y ) |
|
|
|
v |
i |
|
= B ( μ ,v |
|
)Γ |
(L |
y ) |
|
|||||||||||||
|
|
tan |
|
|
|
|
|
(5.97) |
|||||||||||||||||||||
|
|
|
|
|
|
|
|||||||||||||||||||||||
|
|
|
i |
|
i |
1 |
|
|
|
|
|
2 |
|
|
|
|
|
i |
|
|
i 2 |
|
|
|
|||||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||
Again after the vec{•} operator is applied, |
|
|
|
|
|
|
|
|
|
||||||||||||||||||||
|
|
|
|
|
|
|
|
|
v |
i |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Γ |
b( μ |
|
,v |
|
) tan |
|
|
= Γ |
|
b( μ ,v ), |
|
1≤ i ≤ d |
(5.98) |
||||||||||||||||
|
|
2 |
|
|
|||||||||||||||||||||||||
v 1 |
|
|
i |
|
i |
|
|
|
|
|
|
|
v 2 |
|
|
i |
|
i |
|
|
|
|
|
|
|
|
|||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|||||||
where the two-dimensional selection matrices in the y direction |
|
||||||||||||||||||||||||||||
|
Γ |
|
|
= Γ (L x ) |
I |
L x |
andΓ |
|
= Γ |
(L y ) |
I |
L y |
|
(5.99) |
|||||||||||||||
|
v 1 |
|
|
|
1 |
|
|
|
|
|
|
|
v 2 |
|
|
2 |
|
|
|
|
|
|
|
are of size by × L with by = Lx (Ly −1).
The real-valued two-dimensional DFT beamspace steering matrix is then given as
B = (W LHy |
W LHx )A c |
(5.100) |
= [b ( μ1 ,v1 ) b ( μ 2 ,v 2 ) b ( μ d ,v d )] R L × d
Hence, B satisfies the following two invariance properties,