Scribd
Upload
7 views2 pages

c08 Appendix

The ESPRIT algorithm estimates signal parameters using rotational invariance techniques by analyzing subarrays of antenna elements. It involves computing the covariance matrix, performing eigenvalue decomposition to separate signal and noise subspaces, and determining the directions of arrival (DOAs) of signals. The algorithm is particularly effective with uniform linear arrays and relies on the shift invariance of the antenna structure.

Uploaded by

OPTICALMIMOOFDM
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views2 pages

c08 Appendix

The ESPRIT algorithm estimates signal parameters using rotational invariance techniques by analyzing subarrays of antenna elements. It involves computing the covariance matrix, performing eigenvalue decomposition to separate signal and noise subspaces, and determining the directions of arrival (DOAs) of signals. The algorithm is particularly effective with uniform linear arrays and relies on the shift invariance of the antenna structure.

Uploaded by

OPTICALMIMOOFDM
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

8.6.1 Appendix 8.

A: The ESPRIT Algorithm

An important example for high-resolution algorithms is the ESPRIT (Estimation of Signal Param-
eters by Rotational Invariance Techniques) algorithm. The basic idea is as simple as it is inge-
nious. Consider an antenna array with Nr elements. Now create a subarray with the elements
1, 2, ....Nr − 1, and a subarray with elements 2, 3, ....Nr . It is clear from Eq. (8.37) that the steering
matrices of the subarrays are linked by the following matrix equation

A2 = A1  (8.47)

where  is a diagonal matrix whose main-diagonal entries are exp(−j k0 da cos(φi )). Define now a
selection matrix Jk so that Jk A = Ak ; this allows to write Eq. (8.47) as

J1 A = J2 A (8.48)

In the next step we compute the signal space, i.e., the vector space spanned by the receive vectors
r (for a general discussion of vector spaces, see [Strang 1988]). A basis for those vector spaces
can be obtained by an eigenvalue decomposition of the covariance matrix

Rrr ≈ UU† (8.49)

where the columns of the matrix U are the basis of the signal space, and and  is a diagonal
matrix. The equation becomes exact when the noise is negligible.
If the signal is noisy, the basis for the signal subspace is defined by the columns of U that
correspond to the signal subspace Us . In practice, these correspond to the K dominant eigenvalues
of Rrr (finding the number of incident waves N, also called “source order”, is nontrivial; for a
discussion see, e.g., [Haardt and Nossek 1994]).
Assume now that N un-correlated plane waves impinge on a ULA of Nr sensor elements and
N ≤ Nr . The eigenvalue decomposition of the covariance matrix is given as

Rrr = UU†
= Us s U†s + Un n U†n
= Us s U†s + σn2 Un U†n

Therefore the basis for the signal subspace is defined by either the columns of Us or, equivalently,
only those columns of U which correspond to the N dominant eigenvalues of Rrr .
It is also obvious that the matrix A is in the signal space spanned by Us . Thus there exists a
matrix T so that A can be represented as A = Us T. Thus, Eq. (8.48) can be written as

J 1 Us  = J 2 Us (8.50)

where

 = TT−1 (8.51)

The matrix  is created from the matrix  by a transformation that preserves the eigenvalues. In
other words, if we know the eigenvalues of , we also know the eigenvalues of . The eigenvalues
of  give the angles of incidence via the relationship exp(−j k0 da cos(φi )). This completes the
search for the DOAs.
Wireless Communications, Second Edition Andreas F. Molisch
© 2011 John Wiley & Sons, Ltd
Thus, the ESPRIT algorithm uses the following steps:

• Determine the covariance matrix of the received signal, Rrr .


• Perform an eigenvalue decomposition, and separate signal subspace from noise subspace, in order
to determine the eigenbase Us .
• Determine the matrix  by solving the (usually overdetermined) system of equations (8.50)
• Find an eigendecomposition of the matrix 
• Determine the directions φi = arccos[arg(eigval())/(−k0 da )].

It has to be stressed that ESPRIT relies on the shift invariance of the antenna structure, and thus
is mainly suitable for measurements with uniform linear arrays where all elements have the same
patterns.
This description covers only the basic principles. Details of the implementation, as well as
generalizations to the multi-dimensional case, estimation of the number of incident signals, etc.,
are advanced research topics.

You might also like