Information and Signal Theory by Anders Gjendemsjø, et al - HTML preview

PLEASE NOTE: This is an HTML preview only and some elements such as links or page numbers may be incorrect.
Download the book in PDF, ePub for a complete version.

reconstruction formula, so that we can recover x( t) from x s ( n).

Proof part II - Signal reconstruction

index-69_1.jpg

index-69_2.jpg

index-69_3.jpg

index-69_4.jpg

index-69_5.jpg

index-69_6.jpg

index-69_7.jpg

index-69_8.jpg

Proof part II - Signal reconstruction

For a bandlimited signal the inverse fourier transform is

In the interval we

are integrating we have:

. Substituting this relation into Equation we get

Using the DTFT relation for X s ( Ω T s ) we have Interchanging integration and summation (under the assumption of

convergence) leads to

Finally we perform the integration and arrive

at the important reconstruction formula

(Thanks to R.Loos for pointing

out an error in the proof.)

Summary

Spectrum sampled signal

Reconstruction formula

Go to Introduction; Illustrations; Matlab Example; Hold operation; Aliasing applet; System

view; Exercises ?

4.3. Illustrations*

In this module we illustrate the processes involved in sampling and reconstruction. To see how all

these processes work together as a whole, take a look at the system view. In Sampling and

reconstruction with Matlab we provide a Matlab script for download. The matlab script shows the process of sampling and reconstruction live.

Basic examples

Example 4.1.

To sample an analog signal with 3000 Hz as the highest frequency component requires

sampling at 6000 Hz or above.

Example 4.2.

The sampling theorem can also be applied in two dimensions, i.e. for image analysis. A 2D

sampling theorem has a simple physical interpretation in image analysis: Choose the sampling

interval such that it is less than or equal to half of the smallest interesting detail in the image.

The process of sampling

We start off with an analog signal. This can for example be the sound coming from your stereo at

home or your friend talking.

The signal is then sampled uniformly. Uniform sampling implies that we sample every

T s seconds. In Figure 4.2 we see an analog signal. The analog signal has been sampled at times t = n T s .

Figure 4.2.

index-71_1.jpg

index-71_2.jpg

index-71_3.jpg

index-71_4.jpg

Analog signal, samples are marked with dots.

In signal processing it is often more convenient and easier to work in the frequency domain. So

let's look at at the signal in frequency domain, Figure 4.3. For illustration purposes we take the frequency content of the signal as a triangle. (If you Fourier transform the signal in Figure 4.2 you will not get such a nice triangle.)

Figure 4.3.

The spectrum X( ⅈΩ) .

Notice that the signal in Figure 4.3 is bandlimited. We can see that the signal is bandlimited because X( ⅈΩ) is zero outside the interval [ – Ω g , Ω g ] . Equivalentely we can state that the signal has no angular frequencies above Ω g , corresponding to no frequencies above

.

Now let's take a look at the sampled signal in the frequency domain. While proving the sampling theorem we found the the spectrum of the sampled signal consists of a sum of shifted versions of

the analog spectrum. Mathematically this is described by the following equation:

index-72_1.jpg

index-72_2.jpg

index-72_3.jpg

Sampling fast enough

In Figure 4.4 we show the result of sampling x( t) according to the sampling theorem. This means that when sampling the signal in Figure 4.2/Figure 4.3 we use F s ≥ 2 F g . Observe in Figure 4.4

that we have the same spectrum as in Figure 4.3 for Ω ∈ [-Ω g , Ω g ] , except for the scaling factor

. This is a consequence of the sampling frequency. As mentioned in the proof the spectrum of the sampled signal is periodic with period

.

Figure 4.4.

The spectrum X s . Sampling frequency is OK.

So now we are, according to the sample theorem, able to reconstruct the original signal exactly.

How we can do this will be explored further down under reconstruction. But first we will take a look at what happens when we sample too slowly.

Sampling too slowly

If we sample x( t) too slowly, that is F s < 2 F g , we will get overlap between the repeated spectra, see Figure 4.5. According to Equation the resulting spectra is the sum of these. This overlap gives rise to the concept of aliasing.

Aliasing

If the sampling frequency is less than twice the highest frequency component, then

frequencies in the original signal that are above half the sampling rate will be

"aliased" and will appear in the resulting signal as lower frequencies.

The consequence of aliasing is that we cannot recover the original signal, so aliasing has to be

avoided. Sampling too slowly will produce a sequence x s ( n) that could have orginated from a number of signals. So there is no chance of recovering the original signal. To learn more about

aliasing, take a look at this module. (Includes an applet for demonstration!)

index-73_1.jpg

Figure 4.5.

The spectrum X s . Sampling frequency is too low.

To avoid aliasing we have to sample fast enough. But if we can't sample fast enough (possibly due

to costs) we can include an Anti-Aliasing filter. This will not able us to get an exact reconstruction

but can still be a good solution.

Anti-Aliasing filter

Typically a low-pass filter that is applied before sampling to ensure that no

components with frequencies greater than half the sample frequency remain.

Example 4.3.

The stagecoach effect

In older western movies you can observe aliasing on a stagecoach when it starts to roll. At

first the spokes appear to turn forward, but as the stagecoach increase its speed the spokes

appear to turn backward. This comes from the fact that the sampling rate, here the number of

frames per second, is too low. We can view each frame as a sample of an image that is

changing continuously in time. (Applet illustrating the stagecoach effect) Reconstruction

Given the signal in Figure 4.4 we want to recover the original signal, but the question is how?

When there is no overlapping in the spectrum, the spectral component given by k = 0 (see

Equation),is equal to the spectrum of the analog signal. This offers an oppurtunity to use a simple reconstruction process. Remember what you have learned about filtering. What we want is to

change signal in Figure 4.4 into that of Figure 4.3. To achieve this we have to remove all the extra components generated in the sampling process. To remove the extra components we apply an ideal

analog low-pass filter as shown in Figure 4.6 As we see the ideal filter is rectangular in the

index-74_1.jpg

frequency domain. A rectangle in the frequency domain corresponds to a sinc function in time domain (and vice versa).

Figure 4.6.

H( ⅈΩ) The ideal reconstruction filter.

Then we have reconstructed the original spectrum, and as we know if two signals are identical in

the frequency domain, they are also identical in the time domain. End of reconstruction.

Conclusions

The Shannon sampling theorem requires that the input signal prior to sampling is band-limited to

at most half the sampling frequency. Under this condition the samples give an exact signal

representation. It is truly remarkable that such a broad and useful class signals can be represented

that easily!

We also looked into the problem of reconstructing the signals form its samples. Again the

simplicity of the principle is striking: linear filtering by an ideal low-pass filter will do the job.

However, the ideal filter is impossible to create, but that is another story...

Go to? Introduction; Proof; Illustrations; Matlab Example; Aliasing applet; Hold operation;

System view; Exercises

4.4. Sampling and reconstruction with Matlab*

Matlab files

Samprecon.m

Introduction; Proof; Illustrations; Aliasing applet; Hold operation; System view; Exercises

4.5. Aliasing Applet*

index-75_1.jpg

index-75_2.jpg

The applet is courtesy of the Digital Signal Processing tutorial at freeuk.com,

http://www.dsptutor.freeuk.com/. You can also have a look at the Light Wheel applet.

Introduction

In this module we shall look at sampling a sinusoidal signal. According to the sampling theorem, a sinusoidal signal can be exactly reconstructed from values sampled at discrete, uniform intervals

as long as the signal frequency is less than half the sampling frequency. Any component of a

sampled signal with a frequency above this limit, often referred to as the folding frequency, is

subject to aliasing.

The applet is based on a fixed sampling rate of F s = 8000 samples per second (one sample every

0.125 milliseconds, i.e

).

Instructions

Set the frequency of the sinusoidal signal, in Hz, in the "Input frequency" box, i.e choose an f in the following signal: sin(2 π f t) . When you click the "Plot" button, with "Input signal" checked, the input signal is plotted against time.

The "Grid" checkbox toggles on and off vertical gridlines indicating the instants at which the

signal is sampled. The "Sample points", representing the sampled values of the input signal, can

also be toggled.

Finally, the "Alias frequency" checkbox (visible only when aliasing occurs) controls the plotting of the "reconstructed" sinusoidal signal, with f = f alias .

Overview of the process

When using the applet it is important to have an understanding of where the different signals

occur in a sampling system.

Figure 4.7.

Ideal sampling process

index-76_1.jpg

index-76_2.jpg

index-76_3.jpg

Relating the applet signals to the figure we get

Input signal = x( t) = sin(2 π f t) , where f is the input frequency chosen by the user.

The sampled signal =

.

The reconstructed signal = , is shown as the original signal if sampling is done fast enough, or

as the aliased signal if sampling is too slow.

( h( t) is an ideal reconstruction filter).

Aliasing demo applet

(This media type is not supported in this reader. Click to open media in browser.)

4.6. Hold operation*

Any practical reconstruction system must input finite length pulses into the reconstruction filter.

The reason is that we need nonzero energy in the nonzero pulses.

Introduction

The operation performed to produce these pulses is called hold. Using the hold-operation we get

pulses with a predefined length and height proportional to the input to the digital-to-analog

converter. By means of the hold operation we get nonzero pulses with energy.

Figure 4.8.

Output signal from the hold device

As we have made changes relative to the ideal reconstruction, we need to look at the output signal the reconstruction filter will give us. Quite obviously the output will not be the original

signal. So, is it still useful?

index-77_1.jpg

index-77_2.jpg

index-77_3.jpg

Analysis

As before, and as will be the situation later, using the frequency domain simplifies the analysis.

To model the hold operation we use convolution with a delta function and a square pulse. The square pulse has unit height and duration τ . The duration τ is the holding time, i.e. how long we hold the incoming value. For the pulses not to overlap we must choose τ < T s . The convolution

can be seen as a filtering operation, using the square pulse as the impulse response. If we fourier

transform the square pulse we obtain the frequency response of the filter, which is a sinc

function.

Figure 4.9 shows the frequency response of the analog square pulse filter. We have plotted the frequency response for τ = T s and

.

Figure 4.9.

Frequency response of the analog square filter as a function of digital frequency f.

From the figure we can make the following observations

The signal will be attenuated more and more towards the band edge, f = 0.5

For τ = T s the maximum attenuation is 3 dB at f = 0.5.

For

the maximum attenuation is 0.82 dB at f = 0.5.

The distortion is a result of linear operations and can thus be compensated for by using a filter

with opposite frequency response in the passband, f ∈ [ – 0.5, 0.5] . The compensation will not be

exact, but we can make the approximation as accurate as we wish. The compensation can be made

in the reconstruction filter or after the reconstruction by using a separate analog filter. One can

also predistort the signal in a digital filter before reconstruction. Where to put the compensator

and it's quality are cost considerations.

index-78_1.jpg

index-78_2.jpg

index-78_3.jpg

index-78_4.jpg

index-78_5.jpg

index-78_6.jpg

Go to? Introduction; Proof; Illustrations; Aliasing applet; System view; Exercises

4.7. Systems view of sampling and reconstruction*

Ideal reconstruction system

Figure 4.10 shows the ideal reconstruction system based on the results of the Sampling theorem

proof.

Figure 4.10 consists of a sampling device which produces a time-discrete sequence x s ( n). The reconstruction filter, h( t), is an ideal analog sinc filter, with

. We can't apply the time-

discrete sequence x s ( n) directly to the analog filter h( t). To solve this problem we turn the sequence into an analog signal using delta functions. Thus we write

.

Figure 4.10.

Ideal reconstruction system

But when will the system produce an output

? According to the sampling theorem we

have

when the sampling frequency, F s , is at least twice the highest frequency component

of x( t).

Ideal system including anti-aliasing

To be sure that the reconstructed signal is free of aliasing it is customary to apply a lowpass filter,

an anti-aliasing filter, before sampling as shown in Figure 4.11.

Figure 4.11.

Ideal reconstruction system with anti-aliasing filter

index-79_1.jpg

index-79_2.jpg

index-79_3.jpg

Again we ask the question of when the system will produce an output

? If the signal is

entirely confined within the passband of the lowpass filter we will get perfect reconstruction if

F s is high enough.

But if the anti-aliasing filter removes the "higher" frequencies, (which in fact is the job of the

anti-aliasing filter), we will never be able to exactly reconstruct the original signal, s( t). If we sample fast enough we can reconstruct x( t), which in most cases is satisfying.

The reconstructed signal,

, will not have aliased frequencies. This is essential for further use

of the signal.

Reconstruction with hold operation

To make our reconstruction system realizable there are many things to look into. Among them are

the fact that any practical reconstruction system must input finite length pulses into the

reconstruction filter. This can be accomplished by the hold operation. To alleviate the distortion caused by the hold opeator we apply the output from the hold device to a compensator. The

compensation can be as accurate as we wish, this is cost and application consideration.

Figure 4.12.

More practical reconstruction system with a hold component

By the use of the hold component the reconstruction will not be exact, but as mentioned above we

can get as close as we want.

Introduction; Proof; Illustrations; Matlab example; Hold operation;