DTFT continue (c.f.

Shenoi, 2006)
We have introduced DTFT and showed some of its properties. We will investigate them in more detail by showing the associated derivations later. We have also given a motivation of DFT which is both discrete in time and frequency domains. We will also introduce DFT in more detail below.

DC response When w=0, the complex exponential ejw becomes a constant signal, and the frequency response X(ejw) is often called the DC response when w=0.
The term DC stands for direct current, which is a constant current.

We will represent the spectrum of DTFT either by H(ejwT) or more often by H(ejw) for convenience. When represented as H(ejw), it has the frequency range [,]. In this case, the frequency variable is to be understood as the normalized frequency. The range [0, ] corresponds to [0, ws/2] (where wsT=2), and the normalized frequency corresponds to the Nyquist frequency (and 2 corresponds to the sampling frequency).

DC response When w=0, the complex exponential ejw becomes a constant signal, and the frequency response X(ejw) is often called the DC response when w=0.
The term DC stands for direct current, which is a constant current.

DTFT Properties Revisited Time shifting

Frequency shifting

Time reversal

DTFT of (n)

n =

(n )e jwn = e jw0 = 1

DTFT of (n+k)+ (n-k)

According to the time-shifting property,

DTFT of (n + k ) is e jwk , DTFT of (n k ) is e jwk

Hence

DTFT of (n + k ) + (n k ) is e jwk + e jwk = 2 cos(wk )

DTFT of x(n) = 1 (for all n)

x(n) can be represented as We prove that its DTFT is

Hence

jwn k (w 2k )e dw =

= (w 2k )dw k =

= (w)dw = (w)dw + (w)dw + (w)dw

= (w)dw = 1

for all n

From another point of view According to the sampling property: the DTFT of a continuous signal xa(t) sampled with period T is obtained by a periodic duplication of the continuous Fourier transform Xa(jw) with a period 2/T = ws and scaled by T. Since the continuous F.T. of x(t)=1 (for all t) is (t), the DTFT of x(n)=1 shall be a impulse train (or impulse comb), and it turns out to be

DTFT of anu(n)
let then

(|a|<1)

This infinite sequence converges to when |a|<1.

DTFT of Unit Step Sequence

Adding these two results, we have the final result

Differentiation Property

DTFT of a rectangular pulse

and get

Convolution
Convolution of two discrete-time signals Let x[n] and h[n] be two signals, the convolution of x and h is

y[n] =

k =

x[k ]h[n k ]

can be written in short by y[n] = x[n] h[n]. Convolution of two continuous-time signals

y (t ) =

x( )h(t )d

can be written in short by y(t) = x(t) h(t)

From Kuhn 2005

Continuous convolution: optics example

From Kuhn 2005

Continuous convolution: electronics example

Properties of convolution Communitive x[n] h[n] = h[n] x[n] this means that y[n] can also be represented as

y[n] =

k =

x[n k ]h[k ]

Associative x[n] (h1[n] h2[n]) = (x[n] h1[n]) h2[n]. Linear x[n] (ah1[n] + bh2[n]) = ax[n] h1[n] + bx[n] h2[n]. Sequence shifting is equivalent to convolute with a shifted impulse

x[n-d]= x[n] [n-d]

An illustrative example x[n][n] x[n]h[n]

Convolution can be realized by

Reflecting h[k] about the origin to obtain h[-k]. Shifting the origin of the reflected sequences to k=n. Computing the weighted moving average of x[k] by using the weights given by h[n-k].

Convolution can be explained as arithmetic product.

Eg., x[n] = 0, 0, 5, 2, 3, 0, 0 h[n] = 0, 0, 1, 4, 3, 0, 0 x[n] h[n]:

0 , 0, 5 , 2, 3, 0, 0, ... *) 0, 0, 1, 4, 3, 0, 0, ... 0, 0, 5 , 2, 3, 0, 0, 0 0, 0, 0, 20, 8, 12, 0, 0 0, 0, 0, 0, 15, 6, 9, 0 0, 0, 5, 22,26,18, 9, 0

Convolution vs. Fourier Transform

Multiplication Property: For continuous F.T. and DTFT, if we perform multiplication in time domain, then it is equivalent to performing convolution in the frequency domain, and vice versa.
DTFT convolution theorem: Let x[n] X(ejw) and h[n] H(ejw). If y[n] = x[n] h[n], then Y(ejw) = X(ejw)H(ejw)

modulation/windowing theorem (or multiplication property)

Let x[n] X(ejw) and w[n] W(ejw). If y[n] = x[n]w[n], then

Ye

( )
jw

1 = 2

X (e )W (e
j

j (w )

)d

a periodic convolution

For continuous F. T.

In summary,

Sampling theorem revisited (oppenheim et al. 1999)

An ideal continuous-to-discrete-time (C/D) converter

Let s(t) be a continuous signal, which is a periodic impulse train:

s(t ) = (t nT )

We modulate s(t) with xc(t), obtaining

xs (t ) = xc (t )s(t ) = xc (t ) (t nT )

Examples of xs(t) for two sampling rates

Sampling with a periodic impulse followed by conversion to a discrete-time sequence

Through the sifting property of the impulse function, xs(t) can be expressed as

xs (t ) = xc (nT ) (t nT )

Let us now consider the continuous Fourier transform of xs(t). Since xs(t) is a product of xc(t) and s(t), its continuous Fourier transform is the convolution of Xc(j) and S(j).

Note that the continuous Fourier transform of a periodic impulse train is a periodic impulse train.

1 S ( j ) = ( k s ) T k =
where s =2/T is the sampling frequency in radians/s. Since X s ( j ) = X c ( j ) S ( j )

1 If follows that X s ( j ) = X c ( j ( k s )) T k =
Again, we see that the copies of Xc(j) are shifted by integer multiples of the sampling frequency, and then superimposed to product the periodic Fourier transform of the impulse train of samples.

Sampling of Bandpass Signals (c.f. Shenoi, 2006)

Correlation Given a pair of sequences x[n] and y[n], their cross correlation sequence is rxy[l] is defined as

rxy [l ] =

n =

x[n]y[n l ] = x[n] y[ l ]

for all integer l. The cross correlation sequence can sometimes help to measure similarities between two signals. Its very similar to convolution, unless the indices changes from l n to n l. Autocorrelation:

rxx [l ] =

n =

x[n]x[n l ]

Properties Consider the following non-negative expression:

= a 2 rxx [0] + 2arxy [l ] + ryy [0] 0

n =

(ax[n] + y[n l ])2 = a 2 x 2 [n] + 2a x[n]y[n l ] + y 2 [n l ]

n = n = n =

rxx [0] rxy [l ] a That is, [a 1] 0 rxy [l ] ryy [0] 1

for all a

rxx [0] rxy [l ] Thus, the matrix is positive semidefinite. rxy [l ] ryy [0]
Its determinate is nonnegative.

The determinant is rxx[0]ryy[0] rxy2[l] 0. Properties rxx[0]ryy[0] rxy2[l] rxx2[0] rxy2[l] Normalized cross correlation and autocorrelation:

rxx [l ] xx [l ] = rxx [0]

xy [l ] =

rxx [0]ryy [0]

rxy [l ]

The properties imply that |xx[0]|1 and |yy[0]|1. The DTFT of the autocorrelation signal rxx[l] is the squared magnitude of the DTFT of x[n], i.e., |X(ejw)|2.
Correlation is useful in random signal processing

DFT and DTFT A closer look

We discuss the DTFT-IDTFT pair (I means inverse) for a discrete-time function given by

and The pair and their properties and applications are elegant, but from practical point of view, we see some limitations; eg. the input signal is usually aperiodic and may be finite in length.

Example of a finite-length x(n) and its DTFT X(ejw).

A finite-length signal

Its magnitude spectrum

Its phase spectrum

The function X(ejw) is continuous in w, and the integration is not suitable for computation by a digital computer. We can discretize the frequency variable and find discrete values for X(ejwk), where wk are equally sampled whthin [-, ]. Discrete-time Fourier Series (DFS) Let x(n) (nZ) be a finite-length sequence, with the length being N; i.e., x(n) = 0 for n < 0 and n > N. Consider a periodic expansion of x(n):

x p (n + KN ) = x(n ), n = 0,1,..., n 1, K is any integer

xp(n) is periodic, so it can be represented as a Fourier series:

To find the coefficients Xp(k) (with respect to a discrete periodic signal), we use the following summation, instead of integration: First, multiply both sides by e-jmw0k, and sum over n from n=0 to n=N-1:

By interchanging the order of summation, we get

Noting that

pf. When n=m, the summation reduces to N When nm, by applying the geometric-sequence formula,

r N +1 r M rn = , r 1 M r 1 n=
N

we have

e j ( 2 / N )k ( n m ) =
n =0

N 1

N m 1 n '= m

e j (2 / N )kn '

e j ( 2 / N )k ( N m ) e j ( 2 / N )k ( m ) = e j ( 2 / N )k 1

e j (2k )+ j (2 / N )k ( m ) e j (2 / N )k ( m ) = =0 j ( 2 / N )k e 1

Since there is only one nonzero term,

= Xp(k)N The final result is

The following pairs then form the DFS

Relation between DTFT and DFS for finite-length sequences We notice that

DFS coeficient DTFT spectrum

In other words, when the DTFT of the finite length sequence x(n) is evaluated at the discrete frequency wk = (2/N)k, (which is the kth sample when the frequency range [0, 2] is divided into N equally spaced points) and dividing by N, we get the Fourier series coefficients Xp(k) .

A finitelength signal

Its magnitude spectrum (sampled)

Its phase spectrum (sampled)

To simplify the notation, let us denote WN = e j (2 / N )n The DFS-IDFS (I means inverse) can be rewritten as (W=WN)

Discrete Fourier Transform (DFT) Consider both the signal and the spectrum only within one period (N-point signals both in time and frequency domains)
IDFT (inverse DFT) DFT

Relation between DFT and DTFT: The frequency coefficients of DFT is the N-point uniform samples of DTFT with [0, 2]. The two equations DFT and IDFT give us a numerical algorithm to obtain the frequency response at least at the N discrete frequencies. By choosing a large value N, we get a fairly good idea of the frequency response for x(n), which is a function of the continuous variable w. Question: Can we reconstruct the DTFT spectrum (continuous in w) from the DFT? Since the N-length signal can be exactly recovered from both the DFT coefficients and the DTFT spectrum, we expect that the DTFT spectrum (that is within [0, 2]) can be exactly reconstructed by the DCT coefficients.

Reconstruct DTFT from DFT

(when the sequence is finite-length) By substituting the inverse DFT into the x(n), we have

a geometric sequence

By applying the geometric-sequence formula

So

The reconstruction formula

The function sin(x)/x is referred to as the sinc function. The sinc function appears in DSP as for interpolation.