4.6 - DTFT formalism
-
0:01 - 0:04In the preceding sub module, we saw conditions under which
-
0:04 - 0:08a sequence actually has a finite-magnitude Fourier Transform,
-
0:08 - 0:11namely, the sequence has to be absolutely summable.
-
0:11 - 0:12We also saw that it can be extended
-
0:12 - 0:15to square-summable sequences.
-
0:15 - 0:16However in real life,
-
0:16 - 0:19many sequences are neither absolutely nor square-summable.
-
0:19 - 0:23For example, cosine of ω naught, times n
-
0:23 - 0:25is an infinite-length sequence
-
0:25 - 0:27that is neither absolutely nor square-summable.
-
0:27 - 0:31Therefore it doesn't have, in the strict sense, a DTFT,
-
0:31 - 0:33as we have seen before.
-
0:33 - 0:36To be able to deal with sequences like this,
-
0:36 - 0:38we need to introduce a new character in support.(?)
-
0:38 - 0:42This is the δ function or δ distribution,
-
0:42 - 0:46or also called the δ generalized function.
-
0:46 - 0:48As you can see, this is a more subtle object
-
0:48 - 0:49than what we are used too.
-
0:49 - 0:53It's actually a function that this zero everywhere,
-
0:53 - 0:55except at the origin where it's infinity.
-
0:55 - 0:56Doesn't sound like a nice function,
-
0:56 - 0:59but is actually is a very useful mathematical abstraction
-
0:59 - 1:03and a very powerful object to have in the toolbox.
-
1:03 - 1:05So we're going to introduce this δ function
-
1:05 - 1:09and then, once we understand it intuitively,
-
1:09 - 1:10we can use it to calculate, for example,
-
1:10 - 1:14the DTFT of this cosine of ω naught times n
-
1:14 - 1:19which will have the spectrum that contains Dirac δ's.
-
1:19 - 1:20So the good news is that
-
1:20 - 1:24now we have this subtle Dirac δ function that will allow us
-
1:24 - 1:27to take Discrete Time Fourier Transforms
-
1:27 - 1:30off a whole set of new functions and sequences
-
1:30 - 1:32that we didn't know how to handle before,
-
1:32 - 1:34and to look at their spectrum.
-
1:36 - 1:41Module 4.6: Discrete Time Fourier Transform formalism.
-
1:41 - 1:44The overview is simply that we are going to introduce
-
1:44 - 1:47the Dirac δ functional, or generalized function.
-
1:47 - 1:53And we can use this Dirac δ to study the spectras of sequences
-
1:53 - 1:57for which we didn't know so far how to compute a DTFT.
-
1:57 - 1:59We'll do this through a number of examples.
-
2:01 - 2:04The tricky thing with the δ functional
-
2:04 - 2:07is that we cannot describe the function itself,
-
2:07 - 2:09we can only describe it indirectly,
-
2:09 - 2:12through what's called the sifting property.
-
2:12 - 2:17So for example here we have the integral between δ of t-s.
-
2:17 - 2:19With the function f of t,
-
2:19 - 2:22the result of the integral is that it takes out,
-
2:22 - 2:27it weeds out (?) the value of f of t, at location s.
-
2:27 - 2:32And this will be true for all functions -- or for all well-behaved functions --
-
2:32 - 2:35of t belonging to the set of real numbers.
-
2:37 - 2:41The intuition for this Dirac δ function
-
2:41 - 2:45is really through thinking about a family of functions.
-
2:45 - 2:49We index them by k here, or k of t.
-
2:49 - 2:54And k is a positive integer, t is a real number,
-
2:54 - 2:58and the support of these functions is inversely proportional to k
-
2:58 - 3:00and have a constant area.
-
3:00 - 3:02We're going to specifically look at an example.
-
3:04 - 3:07The example is the rect function.
-
3:07 - 3:08We have seen this before.
-
3:08 - 3:10It's a function that is piecewise constant.
-
3:10 - 3:14It's zero inside an interval minus a half to one half
-
3:14 - 3:17where it is actually equal to 1.
-
3:17 - 3:21From the rect function, we can derive a localizing family,
-
3:21 - 3:25r k of t, by simply scaling the rect function
-
3:25 - 3:28both in its height and its width.
-
3:28 - 3:35So it will be nonzero over an interval going from -1/2k to 1/2k.
-
3:35 - 3:39So the support is of width 1/ k.
-
3:39 - 3:40Since it's of height k,
-
3:40 - 3:44the area is exactly equal to 1, independently of k.
-
3:47 - 3:52So let's look at the few instantiations of this rect function,
-
3:52 - 3:55scaled in height and in width.
-
3:55 - 3:57So k is equal to 1,
-
3:57 - 3:58k is equal to 5,
-
3:58 - 4:00k is equal to 15,
-
4:00 - 4:02k is equal to 40.
-
4:02 - 4:05You can see it becomes higher and higher, obviously,
-
4:05 - 4:07but it's narrower and narrower,
-
4:07 - 4:10and its integral is always exactly equal to 1.
-
4:12 - 4:16Now, let's see these localizing functions at work.
-
4:16 - 4:19So remember, there are rk of t.
-
4:19 - 4:23We take an integral of rk of t times f of t.
-
4:23 - 4:26And this is equal to k times the integral
-
4:26 - 4:30over an interval of length 1/k,
-
4:30 - 4:33and the mean value theorem says that on this interval,
-
4:33 - 4:36there is a value of f, namely f of γ,
-
4:36 - 4:39there exists at least one value f of γ
-
4:39 - 4:41which is equal to this integral.
-
4:41 - 4:43And so, as k goes to infinity,
-
4:43 - 4:45the interval becomes narrower and narrower,
-
4:45 - 4:49and therefore, the limit of the integral is equal to f of 0.
-
4:49 - 4:54Okay, so we have seen how taking the limit of the localizing functions,
-
4:54 - 4:58extracts the value of f of t at the origin f0.
-
4:58 - 5:02Now, in this case we have assumed that a function is continuous.
-
5:02 - 5:05Otherwise, it's obviously more technical.
-
5:06 - 5:11So, in the end, the Dirac δ functional is simply a shorthand,
-
5:11 - 5:14instead of always writing the limiting process
-
5:14 - 5:19-- so, for example, our integral on the previous page here, which is limit as k goes to infinity --
-
5:19 - 5:26we simply write the integral between δ of t minus s times f of t.
-
5:26 - 5:30This integral is going to be equal to f of s,
-
5:30 - 5:32as if we have taken the limit:
-
5:32 - 5:35so it's a shorthand of this limiting process
-
5:35 - 5:37that we have seen on the previous slide.
-
5:39 - 5:40If we have one Dirac,
-
5:40 - 5:42we can also look at a sequence of Dirac:
-
5:42 - 5:44there is one, in particular,
-
5:44 - 5:46that is used very often in signal processing.
-
5:46 - 5:50It's δ of ω tilde.
-
5:50 - 5:54This is equal to 2π times an infinite sum of Diracs,
-
5:54 - 5:56located at multiples of 2π.
-
5:59 - 6:01Let's look at the graphical representation.
-
6:01 - 6:06So you have Diracs at zero, at plus minus 2π,
-
6:06 - 6:08at plus minus 4π, etcetera.
-
6:11 - 6:15Now this function, δ tilde of ω,
-
6:15 - 6:18this sequence of Diracs, spaced 2πi apart,
-
6:18 - 6:20is actually very useful.
-
6:20 - 6:22So let's put it to good use.
-
6:22 - 6:26So take the inverse DTFT of δ tilde.
-
6:26 - 6:30So it's the integral over one period, minus π to π,
-
6:30 - 6:33of δ tilde against (?) e to the j ω n.
-
6:33 - 6:36This, of course, involves only a single Dirac at the origin,
-
6:36 - 6:41because that's the only Dirac in the sequence that is inside the interval.
-
6:41 - 6:45This is equal to the evaluation of e to the jωn:
-
6:45 - 6:49at ω = 0, this is exactly equal to 1.
-
6:49 - 6:53Therefore the inverse DTFT of δ tilde
-
6:53 - 6:57is equal to the sequence x(n)O is equal to 1.
-
6:57 - 6:58But that's a miracle because we didn't know
-
6:58 - 7:00how to take the forward DTFT
-
7:00 - 7:04of the constant signal x(n) is equal to 1,
-
7:04 - 7:06so now we actually know what it is.
-
7:08 - 7:10In conclusion,
-
7:10 - 7:15the Discrete Time Fourier Transform of the constant sequence
-
7:15 - 7:17is δ tilde of ω,
-
7:17 - 7:21the sequence of Diracs spaced 2π apart.
-
7:23 - 7:26Now this looks like we played a trick on you
-
7:26 - 7:28because we started with a constant sequence,
-
7:28 - 7:30we looked at the DTFT,
-
7:30 - 7:33and we were confused what the result would be.
-
7:33 - 7:35So let us do a sanity check.
-
7:35 - 7:38Let's just calculate the partial sums of the DTFT.
-
7:38 - 7:44So sk of ω is the sum between minus k and k, of e to the -jωn.
-
7:46 - 7:50We plot the magnitude of sk of ω
-
7:50 - 7:51for k is equal to 5,
-
7:51 - 7:58k is equal to 10, 15, 30, 40,
-
7:58 - 8:00and we start to see a Dirac emerging:
-
8:00 - 8:02one spike at the origin.
-
8:04 - 8:09The partial sum DTFT looks like a family of localizing functions.
-
8:09 - 8:16So sk of ω will, as k grows, converge to δ tilde of ω.
-
8:19 - 8:22Now we can do the exact same exercise
-
8:22 - 8:26on δ tilde of ω minus ω naught
-
8:26 - 8:28by the same argument as before.
-
8:28 - 8:32We will sift out the Dirac that is inside the interval, minus π to π.
-
8:32 - 8:35And this one will lead to a complex exponential,
-
8:35 - 8:38e to the jω naught times n.
-
8:38 - 8:41From this, we have DTFT pairs.
-
8:41 - 8:45The DTFT of 1 is, as we have seen, δ tilde of ω,
-
8:45 - 8:48the DTFT of e to the jω naught times n
-
8:48 - 8:52is equal to δ tilde, ω minus ω naught.
-
8:52 - 8:57More interestingly is the DTFT of cos ω naught times n:
-
8:57 - 9:00Well, using Euler's formulas, we know that the cosine
-
9:00 - 9:03is a combination of 2 complex exponential.
-
9:03 - 9:05Each one will generate a δ tilde,
-
9:05 - 9:07one shifted to ω naught,
-
9:07 - 9:10the other one shifted to minus ω naught.
-
9:10 - 9:14Finally, the DTFT of sine ω naught n
-
9:14 - 9:16is equal to something very similar,
-
9:16 - 9:20there is a scaling factor and multiplication by minus j
-
9:20 - 9:22and then two δ tildes again
-
9:22 - 9:24shifted to ω naught and minus ω naught.
-
9:26 - 9:29It is time to summarize, and to see the parallels
-
9:29 - 9:33between the DFT, which acts on CN,
-
9:33 - 9:37and the DTFT that acts on C infinity.
-
9:37 - 9:38Please note before starting
-
9:38 - 9:43that the δ over sequences has square brackets,
-
9:43 - 9:44as we have seen.
-
9:44 - 9:48And the δ over the real line has round brackets.
-
9:48 - 9:49So these are very different elements.
-
9:49 - 9:52δ in discrete time is a very simple sequence
-
9:52 - 9:54with a single nonzero element,
-
9:54 - 9:59δ over the continuous line is a distribution
-
9:59 - 10:00or a generalized function:
-
10:00 - 10:03as we have seen, it's a Dirac δ function.
-
10:03 - 10:05With this preliminaries,
-
10:05 - 10:08let's look now at the left side, and properties on C N.
-
10:08 - 10:12So we had verified that complex exponential, of length N,
-
10:12 - 10:16with frequencies that are multiples of 2π/N,
-
10:16 - 10:17are orthogonal to each other:
-
10:17 - 10:19that's formula A,
-
10:19 - 10:24this is written as also DFT of the complex exponential
-
10:24 - 10:26of a frequency 2π/N times h,
-
10:26 - 10:32gives a transform that is equal to N times a δ at the location h.
-
10:32 - 10:37The inverse discrete Fourier transform of a δ at location h
-
10:37 - 10:43is equal to a complex exponential at frequency 2π/n times h.
-
10:43 - 10:47And this is written under d as explicitly
-
10:47 - 10:51as a summation of the δ with respect to the complex exponential.
-
10:51 - 10:51Same result
-
10:52 - 10:56Now let's more to the right hand side.
-
10:56 - 11:00There the inversion formula we have seen of Dirac tilde
-
11:00 - 11:02which is a sequence of Diracs,
-
11:02 - 11:04is integral over one period
-
11:04 - 11:07and if the Dirac sequence is shifted to σ,
-
11:07 - 11:12it will generate a complex exponential e to the jσn.
-
11:12 - 11:17So that's under C', it's the IDTFT of δ tilde.
-
11:17 - 11:23Then under B', the DTFT of e to the jσn
-
11:23 - 11:26is δ tilde shifted by σ,
-
11:26 - 11:28and this leads to A',
-
11:28 - 11:31where we simply write that the inner product
-
11:31 - 11:35-- so that's the expression the DTFT between two complex exponentials --
-
11:35 - 11:38leads to δ tilde shifted to σ.
-
11:39 - 11:44So now we see these formal parallels between these two transforms:
-
11:44 - 11:47one on CN as a DFT, which is very simple,
-
11:47 - 11:50it's elementary linear algebra,
-
11:50 - 11:52and the other one, the DTFT
-
11:52 - 11:56which acts on infinite sequences, which is more subtle,
-
11:56 - 11:58and which leads, in the cases we have seen here,
-
11:58 - 12:02to generalized functions (?), sequences of Diracs.
-
12:02 - 12:06Yet they act very similarly: the same intuitions about frequencies,
-
12:06 - 12:11about complex exponentials, about Diracs appear in both cases.
-
12:11 - 12:13And that's really the message we want to convey here.
-
12:16 - 12:17Let us now conclude.
-
12:17 - 12:20The final result we have seen shows
-
12:20 - 12:26a formal orthogonality of the "basis" of e to the jωn,
-
12:26 - 12:29the sequence of infinite-length complex exponentials.
-
12:29 - 12:32So if you're careful enough, everything works just nice.
-
12:32 - 12:37But remember the δ notation presumes that you took a limiting operation.
-
12:37 - 12:41And it's just a lazy shorthand to express something
-
12:41 - 12:44that always implies a limit. A δ in the spectrum,
-
12:44 - 12:50so if we have an x of e to the jω, and there is a δ function there,
-
12:50 - 12:52indicates that the signal has infinite energy,
-
12:52 - 12:57So we are not within our comfortable Hilbert space framework of l2 of Z.
-
12:57 - 13:00But the δ notation proves to be very useful
-
13:00 - 13:02in derivation of certain results,
-
13:02 - 13:04for example, the modulation theorems
-
13:04 - 13:07that we are going to see later in this module.
-
13:07 - 13:08Yet, there is a word of care.
-
13:08 - 13:10As a friend of mine says,
-
13:10 - 13:14when you use δ's it's like walking on the edge of a precipice.
-
13:14 - 13:17If you put the foot on the wrong side
-
13:17 - 13:19you might actually die.
-
13:19 - 13:21With this word of caution, let's move on.
- Title:
- 4.6 - DTFT formalism
- Description:
-
See "Chapter 4 Fourier Analysis" in "Signal Processing for Communications" by Paolo Prandoni and Martin Vetterli, available from http://www.sp4comm.org/ , and the slides for the entire module 4 of the Digital Signal Processing Coursera course, in https://spark-public.s3.amazonaws.com/dsp/slides/module4-0.pdf .
And if you are registered for this Coursera course, see https://class.coursera.org/dsp-001/wiki/view?page=week3
- Video Language:
- English
Claude Almansi edited English subtitles for 4.6 - DTFT formalism | ||
Claude Almansi added a translation |