[go: up one dir, main page]

0% found this document useful (0 votes)
72 views36 pages

Lecture 2: Background: - Linear Algebra

This document provides an overview of key concepts in linear algebra including: - Vectors such as column vectors, row vectors, and Hermitian transposes. - Norms of vectors like the L1, L2, and L∞ norms. - Inner products of vectors and their properties including the Cauchy-Schwarz inequality. - Concepts related to matrices including symmetric, Hermitian, and convolution matrices. Properties of the matrix rank, inverse, determinant, and trace are also discussed. - Solutions to systems of linear equations in matrix form including underdetermined, overdetermined, and minimum norm solutions.

Uploaded by

Sana Saad
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
0% found this document useful (0 votes)
72 views36 pages

Lecture 2: Background: - Linear Algebra

This document provides an overview of key concepts in linear algebra including: - Vectors such as column vectors, row vectors, and Hermitian transposes. - Norms of vectors like the L1, L2, and L∞ norms. - Inner products of vectors and their properties including the Cauchy-Schwarz inequality. - Concepts related to matrices including symmetric, Hermitian, and convolution matrices. Properties of the matrix rank, inverse, determinant, and trace are also discussed. - Solutions to systems of linear equations in matrix form including underdetermined, overdetermined, and minimum norm solutions.

Uploaded by

Sana Saad
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/ 36

Lecture 2: Background

• Linear Algebra

1
Vectors
• Column vector of dimension (length) N
 x1 
 
 x 2 
x=
 
 
x N 
• Row vector

x = x 1, x 2, , x N 
T

2
Vectors
• Hermitian transpose
H
x = x ( )
T ∗
= x 1∗, x 2∗, , x N∗ 
• Other notations
 x (0)   x (n ) 
   
x (1) x (n − 1)
x=  x(n ) =  
     
   
x (N − 1) x (n − N + 1) 3
Vector Norms
• Provide a metric for vector length. Can be used to
measure the distance between two vectors.
• L1 norm: N
x 1 =  xi
i =1
• L2 norm (Euclidean norm):
1/2
 N
2
x 2 = x =  x i 
 i =1 
• L∞ norm:
x ∞ = max x i
i

• Unit norm vector: x


vx =
x 4
Inner Product
• Complex vectors N
a, b = a H b =  ai∗bi
i =1
• Real vectors N
a, b = a b =  aibi
T

i =1
• Geometrical relationship between two vectors
a, b = a b cos q
• Orthogonal vectors
a, b = 0
• Orthonormal vectors: orthogonal vectors having
unit norm. See Example 2.3.1 5
Inner Product
• Cauchy-Schwarz inequality:
a, b ≤ a b
Equality holds iff a and b colinear i.e., a = α b
for some constant α
• Another useful inequality
2 2
2 a, b ≤ a + b
Equality hold if and only if ||a|| = ||b||

6
Inner Product
• FIR filter Convolution Sum:
N −1
y(n ) =  h(k )x (n − k )
k =0

• impulse response in vector form:


 h(0) 
 
 h(1) 
h=
  
 
h(N − 1)
• Output expressed in terms of inner product:
y(n ) = hT x(n ) 7
Linear Independence
• A set of n vectors, v1, v2,..., vn is linearly
independent if
a1v1 + a2 v2 +  + an vn = 0
implies that αi = 0 ∀ i.
• Linearly dependent vectors: At least one of
the vectors may be expressed as a linear
combination of the others
v1 = b2 v2 + b3 v 3 +  + bn vn
for some set of scalars, βi .
See Example 2.3.2 8
Vector Spaces
• Given a set of N vectors,
V = {v1, v2, , vN }
• Vector space  is the set of all vectors that
may be formed from a linear combination of
vectors vi , N
v =  ai vi
i =1

• Vectors vi are said to span the space 

9
Basis Vectors
• If the vectors vi are linearly independent,
then they are said to form a basis for the
space 
• The number of vectors N in the basis is called
the dimension of the space.
• Reading: Section 2.3.2

10
Matrices
• Notation: n × m matrix A with elements aij
• Convolution matrix: Consider the output of
the FIR LSI filter,

y(n ) = hT x(n ) = xT (n )h

• If x(n) = 0 for n < 0 then y(n) for n ≥ 0 is given


by
X0h = y

11
Matrices
where the convolution matrix

 x (0) 0 0  0 
 
 x (1) x (0) 0  0 
 x (2) x (1) x (0)  0 
X0 =  
    
x (N − 1) x (N − 2) x (N − 3)  x (0)
 
      

12
Matrices
• Symmetric matrix: A square matrix is
symmetric if
A = AT
• Hermitian matrix: A n × m complex matrix is
Hermitian if
H ∗ T T ∗
A = A = (A ) = (A )
• See Section 2.3.3 for some properties of
Hermitian matrices
13
Matrices

14
Matrix Rank
• Rank of matrix A, denoted by ρ (A) is defined as the
number of linearly independent columns in A
• ρ (A) = ρ (AH) => if A is partitioned in terms of n
row vectors,
 r1H 
 H
 r2 
A=
 
 H
 rn 
then the rank of A is equivalently equal to the
number of linealy independent row vectors.
15
Matrix Rank
• ρ (A) = ρ (AAH) = ρ (AHA)
• for an n × m matrix A
r(A) ≤ min(m, n )
• Full rank matrix:
r(A) = min(m, n )

16
Matrix Inverse
• Exists for square matrices of full rank
• Some properties:
( )
−1
−1 −1
AB = B A
(A ) ( )
−1 H
H −1
= A
• Matrix Inversion lemma:

( A + BCD) ( )
−1 −1
−1 −1 −1 −1 −1
= A − A B C + DA B DA
• Where A is n x n, B is n x m, C is m x m, and D is m x n
with A and C nonsingular matrices.
17
Matrix Inverse
• Woodbury’s Identity:
−1 H −1
A uv A
( )
−1
A + uvH = A −1 − H −1
1+ v A u
• Is obtained from the Matrix Inversion lemma when C
= 1, B = u, and D = vH where u and v are n x 1 vectors.
• Special case, A = I:
1
( I + uv )
−1
H H
=I− uv
1 + vH u

18
Determinant and Trace
• Determinant and its properties (See Section 2.3.5)
• Condition for inversion of n x n matrix A:
det(A) ≠ 0

• Trace of n x n matrix A: sum of the terms along the


diagonal n
tr(A) =  aii
i =1

19
Linear Equations
• Set of n linear equations in m unknowns, xi = 1,2,...,m

• In matrix form:
Ax = b
where A is n x m, x is m x 1, b is n x 1
• The above equation can be viewed as an expansion
of b in terms of a linear combination of the column
vectors ai of A m
b =  x i ai (2.32)
20
i =1
Solutions to Ax = b
• A is n x n square matrix, m = n:
x = A −1b
• If A singular => no solution or many solutions
• If A singular => linearly dependent columns, then there
exists non-zero solutions to the homogeneous
equation
Az = 0
– k = n − ρ(A) linearly independent solutions to the
homogeneous equations
– if at least one vector x0 exists that solves Ax = b, then any
vector of the form
x = x 0 + a1z1 +  + ak zk
will also be a solution where zi , i = 1,...,k are linearly
independent solution to Az = 0
21
Solutions to Ax = b
• A is rectangular matrix, n < m: Fewer equations than
unknowns => many solutions exist, solution
underdetermined or incompletely specified
– A unique minimum norm solution is to find the vector
satisfying the equations that has min. norm
min x such that Ax = b
– Minimum norm solution:
x 0 = AH (AAH )−1 b
– where AAH is invertible and the matrix
A + = AH (AAH )−1
is the pseudo-inverse of matrix A
22
Solutions to Ax = b
• A is rectangular matrix, m < n: More equations than
unknowns => equations are inconsistent and solution is said
to be overdetermined
• We need to find an approximate solution
• Fig. 2.5: Case of 3 equations in 2 unknowns

23
Solutions to Ax = b
• An arbitrary vector b in this case cannot be represented as a
linear combination of the colums of A (Eq. 2.32)
• Goal: Find coefficient xi that produce best approximation to b
m
ˆ=
b x a
i =1
i i

• Least Squares solution: Find vector x that minimizes the norm


of the error 2 2
e = b − Ax
• Least square solution has the property that the error
e = b − Ax
• is orthogonal to the column vectors ai of A used to
approximate b
AH e = 0 (2.37)
24
Solutions to Ax = b
or
AH Ax = AH b
which are known as normal equations. If columns of A are
linearly independent (A full rank) then AHA is invertible and
the least squares solution is
x = (AH A)−1 AH b
or
x = A+ b
where
A + = (AH A)−1 AH
is the pseudo-inverse of A for the overdetermined problem.

25
Solutions to Ax = b
The best approximation to b is given by the projection of b
onto the subspace spanned by vectors ai
ˆ = Ax 0 = A(AH A)−1 AH b
b
or
ˆ = PA b
b
where
PA = A(AH A)−1 AH
is called the projection matrix
• Minimum least squares error: using the orthogonality
condition (Eq. 2.37)
2
min e = bH e = bH b − bH Ax 0

26
Special matrix Forms
• Diagonal matrix

27
Special matrix Forms
• Block Diagonal Matrix

28
Special matrix Forms
• Upper Triangular Matrix

29
Special matrix Forms
• Toeplitz Matrix: all of the elements along each of
the diagonals have the same value

• Special case of persymmetric matrices: symmetric


about the cross-diagonal aij = an-j+1,n-i+1
• A persymmetric matrix may not necessarily be
Toeplitz
30
Special matrix Forms
• Symmetric Toeplitz Matrix (special case of
centrosymmetric matrices => both symmetric and
persymmetric)

• Hermitian Toeplitz Matrix

31
Special matrix Forms
• Orthogonal matrix: real n x n matrix having
orthonormal columns (and rows)

32
Special matrix Forms
• Unitary matrix: complex n x n matrix having
orthogonal columns (rows)

Reading: Section 2.3.7 33


Quadratic and Hermitian Forms
• Quadratic form of a real symmetric n x n matrix A is
the scalar

• Example:

34
Quadratic and Hermitian Forms
• Hermitian form of a Hermitian n x n matrix A is the
scalar

• Positive definite matrix A, written A > 0: quadratic


form of A is positive for all nonzero vectors x

35
Quadratic and Hermitian Forms
• Positive semidefinite matrix A: quadratic form of A is
nonnegative for all nonzero vectors x

• Negative definite matrix A: for all nonzero vectors x,

• Negative semidefinite matrix A: for all nonzero


vectors x,

• Indefinite: none of the aforementioned


Reading: Section 2.3.8 36

You might also like