This document proves that the space of bandlimited continuous time signals is a Hilbert space, and that the sinc functions form an orthonormal basis for this space. It shows that the inner product between any two sinc basis functions is 1 if they are the same function and 0 if they are different. This establishes that the sinc functions form an orthonormal basis, which means the space of bandlimited signals is a Hilbert space that allows representing any signal using this basis.
This document proves that the space of bandlimited continuous time signals is a Hilbert space, and that the sinc functions form an orthonormal basis for this space. It shows that the inner product between any two sinc basis functions is 1 if they are the same function and 0 if they are different. This establishes that the sinc functions form an orthonormal basis, which means the space of bandlimited signals is a Hilbert space that allows representing any signal using this basis.
This document proves that the space of bandlimited continuous time signals is a Hilbert space, and that the sinc functions form an orthonormal basis for this space. It shows that the inner product between any two sinc basis functions is 1 if they are the same function and 0 if they are different. This establishes that the sinc functions form an orthonormal basis, which means the space of bandlimited signals is a Hilbert space that allows representing any signal using this basis.
We have just seen that given a discrete time signal
with finite energy, we can choose an interpolation
interval T_s and transform this discrete time signal into a continuous time signal also with finite energy. The continuous time signal will be F_s band-limited where F_s is equal to 1 over T_s. Now the question is, is this mapping invertible? Namely, given an F_s bandlimited continuous time signal, can we map that to a discrete time signal with no loss of information? The answer is yes. To prove this result, which will lead naturally to the most general form of the sampling theorem, we will use once again the power of vector space. Indeed, we will show that the space of F_s bandlimited function is a Hilbert space, and we will find a suitable basis for this space. To prove this for a while, let's lighten the notation and choose T_s equal to 1. This will entail no loss of generality and will have the added advantage that F_s will be equal to 1 as well. So the claims are that the space of one bandlimited function, again, one hertz bandlimited function is a Hilbert space, the functions sinc of t minus n with n belonging to the set of integers, form a basis for the space, and if a signal belongs to the space of one bandlimited functions, the set of values for the function computed all integer values for the argument, which is a countable set, is a sufficient representation for the function itself. In other words, we can reconstruct x of t from this countable set of samples. So let's recall the three ingredients that are necessary to have a Hilbert space. We need the vector space and clearly the space of one bandlimited functions is a vector space, because linear combinations of bandlimited functions will always yield bandlimited functions because of the linearity of the Fourier Transform. The second ingredient is having an inner product, and the inner product in the space of one band-limited function is the standard inner product in L_2 of R, and the third ingredient is completeness, which as you remember is kind of a technical notion, and in this case unfortunately it's hard to prove, so we will just assume it for the rest of this lecture. What is important is that we will be able to find an orthonormal basis for this space. Before we proceed, we said that the inner product is going to be the standard inner product in L_2 of R, let's review the definition. The inner product between two elements in L_2 of R is just the integral from minus infinity to plus infinity of the product but doing the conjugate of the first vector and the second vector. Remember that we can always define the convolution between two continuous time signals in terms of the inner product, this will be useful later. So the convolution of x and y is just the inner product between the conjugate of x of tau and y of t minus tau, where tau is a dummy variable. Let's find this basis for the space of one band-limited functions, the claim is that the basis is given by the family phi n of t where n is the ordinality of the vector in the basis. Phi n of t is defined as sinc of t minus n, n is an integer. How do we prove that this is a basis? Well, we take the inner product between any two elements of the basis and we verify if this basis is orthogonal. If that is the case, we're done. So let's take this inner product because of the definition of phi of n of t as sinc of t minus n, we can write the following, phi n of t is equal to phi 0 of t minus n, and similarly phi m of t is equal to phi 0 of t minus m. Now, these are sinc functions, so they are symmetric, and we can change the sign of the argument with no change in the expression. Let's replace now the definition of the inner product and we have the integral over the entire real axis of sinc of t minus n times sinc of m minus t in dt. Now we perform a change of variable and we have sinc of tau times sinc of m minus n minus tau into tau, which we recognize as the convolution of a sinc with itself computed in m minus n. Now, the Fourier Transform of this sinc is the rect, and since we have chosen conveniently T_s equal to F_s equal to 1, we don't have any normalization factors in this. So to compute the convolution of the sinc with itself in m minus n, we compute the inverse Fourier Transform of the product of the Fourier Transform of the sincs, which are rects, and then we compute this inverse Fourier Transform in m minus n. So we have the formula for the inverse Fourier Transform, the integral over the entire frequency axis of the Fourier Transform of the sinc times the Fourier Transform of the sinc. So we have rect square of f, and then where we need to Fourier Transform, so e to the j 2pi f m minus n df. So the rect is zero everywhere except in the minus one half interval. This changes the integration limits and the rect over that interval is equal to 1, so we're left with the integral between minus one and a half, one and a half of this phase term. Now, with the change of variable, omega equal to 2pi f, we have that this integral is equal to this, 1 over 2pi integral between minus pi and pi of e to the j omega m minus n the d mega. We have seen this plenty of time is the integral of a pure phase factor around the unit circle. So we know that it's going to be equal to 1 if m is equal to n and therefore the argument of the phase is 0 and 0 otherwise. So we have now found that the inner product between basis vector number n and basis vector number m is equal to 1 if m is equal to n, and 0 otherwise, so the set of vectors is actually orthonormal and therefore, it is a basis for the space of one bandlimited functions.
Communication Engineering Prof. Surendra Prasad Department of Electrical Engineering Indian Institute of Technology, Delhi Lecture - 4 The Hilbert Transform
Biomedical Signal Processing Prof. Sudipta Mukhopadhyay Department of Electrical and Electronics Communication Engineering Indian Institute of Technology, Lecture - 51 Tutorial - II (Contd.)