0% found this document useful (0 votes)
72 views3 pages

Fourier Series: Yuriy Zakharov

1) The document discusses representing a signal x(t) as a linear combination of other signals φn(t) through Fourier series analysis. 2) An optimal set of expansion coefficients cn is derived that minimizes the error between the original signal x(t) and its approximation x^(t). 3) As an example, the expansion coefficients are calculated to represent the function e^t as a linear combination of three orthogonal polynomial basis functions on the interval -1 ≤ t ≤ 1.
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 views3 pages

Fourier Series: Yuriy Zakharov

1) The document discusses representing a signal x(t) as a linear combination of other signals φn(t) through Fourier series analysis. 2) An optimal set of expansion coefficients cn is derived that minimizes the error between the original signal x(t) and its approximation x^(t). 3) As an example, the expansion coefficients are calculated to represent the function e^t as a linear combination of three orthogonal polynomial basis functions on the interval -1 ≤ t ≤ 1.
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/ 3

1

Fourier series
Yuriy Zakharov

We are going to represent one signal, x(t), by a set of other To minimize Ee , a necessary condition is
signals, {ϕn (t)}N −1
n=0 , where N can be any number, in partic-  2
ular, N = ∞. Specifically, we will approximate a signal of  T /2 N
 −1
∂Ee ∂
interest x(t) as a sum of ‘simple’ signal components ϕ n (t): = x(t) − cn ϕn (t) dt = 0. (7)
∂ci ∂ci −T /2 n=0
N
 −1
This condition should be fulfilled for all i = 0, ..., N − 1. We
x̂(t) = cn ϕn (t) (1)
n=0
now expand the integrand and take the derivatives with respect
to ci . As a result we obtain:
where cn are expansion coefficients.  T /2
The material related to this lecture can be found, for example, ∂Ee ∂
= x2 (t)dt (8)
in [1]. ∂ci ∂ci −T /2
N −1  T /2
∂ 
A. Orthogonal signal set: real signals − 2 cn x(t)ϕn (t)dt (9)
∂ci n=0 −T /2
Suppose that there is a set of real functions {ϕ n (t)}∞
n=0 on N −1 N −1  T /2
an interval [−T /2, +T /2]. We call such a set orthogonal if the ∂  
+ cn cm ϕn (t)ϕm (t)dt (10)
functions possess the orthogonality property: ∂ci n=0 m=0 −T /2

 T /2  = 0.
En if m = n;
ϕn (t)ϕm (t)dt = En δnm = (2)
−T /2 0 otherwise The integral in (8) does not depend on c i ; therefore its derivative
with respect to ci is equal to zero. In (9), only one term in the
where δmn is the Kroneker delta: sum depends on c i and this dependence is linear; therefore (9)
 takes the form
1 if m = n;  T /2
δmn = (3) −2 x(t)ϕi (t)dt.
0 otherwise. (11)
−T /2

The value En is the energy of the function ϕ n (t), The function set {ϕ n (t)} is orthogonal, i.e., the properties (2)
 and (4) are hold. Then we can re-write (10) as
T /2
En = ϕ2n (t)dt. (4) N −1
−T /2 ∂  ∂  
En c2n = Ei c2i = 2Ei ci . (12)
∂ci n=0 ∂ci
If N → ∞, we expect that x̂(t) = x(t). If this happens for
any continuous function x(t), we say that the set of functions Now we can represent (7) as
{ϕn (t)} is a complete set.
 T /2
The approximation error is ∂Ee
= −2 x(t)ϕi (t)dt + 2Ei ci = 0. (13)
∂ci −T /2
N
 −1
e(t) = x(t) − x̂(t) = x(t) − cn ϕn (t). (5) This equation is easy to solve with respect to c i :
n=0
 T /2
1
Then the error energy is ci = x(t)ϕi (t)dt. (14)
Ei −T /2
 T /2
Ee = e2 (t)dt This equation gives optimal expansion coefficients that provide
−T /2 a minimum error energy.
  N −1
2
T /2 
= x(t) − cn ϕn (t) dt. (6) C. Example-1
−T /2 n=0
Find the expansion coefficients of the function x(t) = e t for
its representation by the three algebraic polynomials orthogonal
B. Optimal expansion coefficients
on the interval −1 ≤ t ≤ 1:
Thus, the error energy E e = Ee (c0 , c1 , . . . , cN ) depends on  
the expansion coefficients. We want to choose such expansion 1 3 45 2
ϕ0 (t) = √ ; ϕ1 (t) = t; ϕ2 (t) = (t − 1/3).
coefficients cn that the error energy is minimized: E e → min. 2 2 8
2

Solution: The expansion coefficients c n , n = 0, 1, 2, are Then, the expansion coefficient c 2 is


found as  1 
 1 45 2
1 1 c2 = et (t − 1/3)dt (33)
cn = x(t)ϕn (t)dt (15) E2 −1 8
En   1   1
−1 5 5
2 t
= 3 t e dt − et dt (34)
where 8 −1 8 −1
 
1 5 2 t 1
En = ϕ2n (t)dt. (16) = 3(t e − 2tet + 2et ) − et −1 (35)
8
−1 

5 1 2 2 1
The energy E 0 of the basis function ϕ 0 (t) is = 3(e − 2e + 2e) − e − 3 + + +
8 e e e e
 1
1 ≈ 0.79057(5.43656 − 5.15131) (36)
E0 = dt = 1. (17)
−1 2 ≈ 0.2263. (37)
2 t 2 t t t
Then, the expansion coefficient c 0 is Here we use the formula: t e dt = t e − 2te + 2e .
 1
1 1 D. Minimum error energy
c0 = et √ dt (18)
E0 −1 2 To find the minimum error energy, we need to substitute the
 1
1 optimal coefficients in (6). Then we obtain
= √ et dt (19)  T /2
2 −1
1 Ee = x2 (t)dt
1 t −T /2
= √ e (20)
2 −1 N
 −1  T /2

− 2 cn x(t)ϕn (t)dt
1 1
= √ e− (21) n=0 −T /2
2 e
N −1 
 T /2
≈ 1.6620. (22) + c2n ϕ2n (t)dt
n=0 −T /2
Here we use the formula: et dt = et .
N
 −1
The energy E 1 of the basis function ϕ 1 (t) is
= Ex − c2n En (38)
 1
3 1 2 t3 n=0
E1 = t dt = = 1. (23)
2 −1 2 −1 where Ex is the energy of signal x(t).

Then, the expansion coefficient c 1 is E. Example-2


 1  We now continue with the previous example and calculate
1 3
c1 = et tdt (24) the error energy according to (38):
E1 −1 2
  1 2
 2

3 Ee = Ex − c2n En = Ex − c2n . (39)
= tet dt (25)
2 −1 n=0 n=0

3 t 1 The last relationship is due to the fact that E n = 1 for n =
= te − et −1 (26) 0, 1, 2. The signal energy E x is
2
  +1  +1 +1
3 1
= [(e − e) − (−1/e − 1/e)] (27) Ex = x2 (t)dt = e2t dt = e2t
2 −1 −1 2 −1
 √
3 2 6 1 2 −2

= · = ≈ 0.9011. (28) = e −e = 0.5(7.3891 − 0.1353)
2 e e 2
= 3.6269. (40)
Here we use the formula: tet dt = tet − et .
The energy E 2 of the basis function ϕ 2 (t) is Then we obtain:
 c20 = 1.66202 = 2.7622,
45 1 2
E2 = (t − 1/3)2 dt (29) c21 = 0.90112 = 0.8120,
8 −1

2 c22 = 0.22632 = 0.0512. (41)
45 1 4 2 2 1
= t − t + dt (30)
8 −1 3 9 Finally, we have
1
45 t5 2 t3 1 Ee = 3.6269 − 2.7622 − 0.8120 − 0.0512 = 0.0015. (42)
= − + t (31) 
8 5 3 3 9 −1 Thus, the relative error is ε = Ee /Ex ≈ 0.02, which is a
= 1. (32) quite small value.
3

F. Generalisation to complex signals


The above results can be generalised to complex signals. A
set of functions ϕ0 (t), ϕ1 (t), . . . , ϕN −1 (t) is mutually orthog-
onal over the interval [−T /2, T /2] if
 T /2 
En if m = n;
ϕn (t)ϕ∗m (t)dt = En δnm = (43)
−T /2 0 otherwise.

In this equation, one of the functions, ϕ m (t), is taken in a


complex-conjugate form, ϕ ∗m (t). The value En is the energy
of the function ϕ n (t), which is now defined as
 T /2  T /2
En = |ϕn (t)|2 dt = ϕn (t)ϕ∗n (t)dt. (44)
−T /2 −T /2

The optimal expansion coefficients are found as


 T /2
1
cn = x(t)ϕ∗n (t)dt. (45)
En −T /2

The error energy can be calculated as


N
 −1
Ee = Ex − |cn |2 En . (46)
n=0

R EFERENCES
[1] B. P. Lathi, Signal processing and linear systems, Berkeley-Cambridge
Press, 1998.

You might also like