SAMPLING

Process in A/D Conversion

 Sampling Theorem
Basic principle: An absolutely band limited waveform cant be absolutely time limited & An
absolutely time limited waveform cant be absolutely band limited.
■ Statement: A continuous time signal can be Nyquist’s theorem deals with
represented in its samples and can be recovered the maximum signalling rate over a
back when sampling frequency fs is greater than channel of given bandwidth. Shannon’s
or equal to the twice the highest frequency theorem deals with
component of message signal. i. e. the reconstruction of a signal from a
finite number of samples.

■ Proof: Consider a continuous time signal x(t). The

spectrum of x(t) is a band limited to fm Hz i.e. the
spectrum of x(t) is zero for |ω|>ωm.
■ Sampling of input signal x(t) can be obtained by
multiplying x(t) with an impulse train δ(t) of period
Ts. The output of multiplier is a discrete signal
called sampled signal which is represented with
y(t) in the following diagrams:
■ Note:Here, you can observe that the sampled
signal takes the period of impulse.
■ The process of sampling can be explained by the following mathematical
expression:
Fourier series expansion is normally
used only for periodic signal.
A periodic signal is one that repeats
the sequence of values exactly after a
fixed length of time, known as the
period.

a0:Total area in one time period(periodic signal X(t) )

and divide it by fundamental time period T 0.
an:X(t)multiplied by cosnwt under total area in one time
period divided by T0/2.
Similarly calculate bn
■ To reconstruct x(t), you must recover input signal spectrum X(ω) from sampled
signal spectrum Y(ω), which is possible when there is no overlapping between
the cycles of Y(ω).
■ Possibility of sampled frequency spectrum with different conditions is given by
the following diagrams: Aliasing Effect:
The overlapped region in case of under sampling
represents aliasing effect, which can be removed by
i)Pre-alias filter must be used to limit band of
frequency of the required signal fm Hz.
ii)Sampling frequency fs must be selected such
that sampling frequency is greater than twice the
maximum modulating signal frequency.
(considering fs >2fm )

Aliasing:https
://www.youtube.com/watch?v=v7qjeUFxVwQ
Three types of sampling techniques:
■ Impulse sampling.
■ Natural sampling.
■ Flat Top sampling.

https://www.youtube.com/watch?v=tZR7h
LJx6Ms
Flat Top Sampling
■ During transmission, noise is introduced at top of the
transmission pulse which can be easily removed if the pulse is in
the form of flat top.
■ Here, the top of the samples are flat i.e. they have constant
amplitude.
■ Hence, it is called as flat top sampling or practical sampling. Flat
top sampling makes use of sample and hold circuit.
■ Theoretically, the sampled signal can be obtained by convolution
of rectangular pulse p(t) with ideally sampled signal say yδ(t) as
shown in the diagram:
Nyquist Rate
It is the minimum sampling rate at which signal can be
converted into samples and can be recovered back
without distortion.
Nyquist rate fS = 2fm hz
Nyquist interval = 1/ fS = 1/ 2fm seconds.
 Sampling Process
■ Let g δ(t) denote the ideal sampled signal
g δ(t) = δ(t- ) --(1)
Where - Nyquist Sampling Period
= 1/ Sampling rate
Note: G(f) *
=
g δ(t) --(2)
Equation(2) states that the process of uniformly sampling a continuous time signal of finite
energy results in a periodic spectrum with a period equal to the sampling rate.
Or we may apply Fourier transform on (1) to obtain
G δ(f) = (-j2π) -----(3)
This relation is called discrete time fourier transform.
Equation (3) represent the complex fourier series representation of the periodic frequency
function
G δ(f) with the sequence of samples ,defining the coefficient of the expansion.
■ Suppose the signal g(t) is strictly band limited ,with no frequency components higher than W hertz.
i.e., G(f)=0 for |f|

■ Suppose we choose the sampling period T s = and thereby putting Ts = in eqn (3) yields
G δ(f) = (- ) ---(4)
From eqn (2),we can express the fourier transform of g δ(t) as
G δ(f) = G (f) + --(5)
Hence under the following two conditions :
i) G(f)=0 for |f|
ii) =2W
From eqn(5) we find that G (f) = () G δ(f) , -W <f<W ---(6)
Sub (4) in eqn(6) we may also write
G (f) = () (- ) , -W <f<W -------(7)
g(t) is uniquely determined by g () for - <n < or g () contains all the information of g(t).
Reconstruction:
To reconstruct g(t) from g (), we may have
g(t) =
=()exp( - )
=() )] df --(8)
= ()
= () , - <t < ---(9)
Note: Sin )/ = ()
(9) is an interpolation formula of g(t )
Sampling Theorem for strictly band -limited signals
1) a signal which islimited to, - <f <W can be completely described by()
2) The signal can be completely recovered from ()
Nyquist rate =2W
Nyquist interval =1/2W
 ■ When the signal is not band -limited (under sampling) aliasing occurs.
■ To avoid aliasing, we may limit the signal bandwidth or have higher
sampling rate.

(a) Spectrum of a signal. (b) Spectrum of an undersampled version of the signal exhibiting the
aliasing phenomenon.
■ Two corrective measures to combat the effects of aliasing:
– Prior to sampling, a low pass anti aliasing filter is used to attenuate those high
frequency components of the signal that are not essential to the information
being conveyed by the signal.
– The filtered signal is sampled at a rate slightly higher than the Nyquist rate.

(a) Anti-alias filtered spectrum of an information-bearing signal. (b) Spectrum of

instantaneously sampled version of the signal, assuming the use of a sampling rate greater
than the Nyquist rate. (c) Magnitude response of reconstruction filter.
Thus from the above figure we observe that:
-The reconstruction filter is low pass with a passband extending from –w to w,which is
itself determined by the ant-aliasing filter.
-The filter has a transition band extending from w to -w , where is the sampling rate.
 Pulse Amplitude Modulation
■ In PAM,the amplitude of regularly spaced pulses are varied in proportion to the
corresponding sample pulses of a continuous message signal.
■ Somewhat similar to natural sampling in which periodic train of rectangular pulses
are considered to be sampling signal but the top pulses remains as
flat top.
■ Message signal m(t);PAM signal s(t)
Generation of PAM signal:
1)Instantaneous sampling of the message signal m(t)
every Ts seconds,where the sampling rate fs = 1/ Ts
is chosen in accordance with the sampling theorem.
2)Lengthening the duration of each sample so
Flat top sample,representing an analog signal
obtained to some constant value T.
*Two operations are called as “sample and hold”.
Natural Sampling:
■ The FET is the switch used as a sampling gate.
■ When the FET is on, the analog voltage is shorted to
ground; when off, the FET is essentially open, so that
the analog signal sample appears at the output.
■ Op-amp 1 is a noninverting amplifier that isolates the
analog input channel from the switching function.
■ Op-amp 2 is a high input-impedance voltage follower
capable of driving low-impedance loads (high
“fanout”).
■ The resistor R is used to limit the output current of
opamp 1 when the FET is “on” and provides a voltage PAM-natural & flat top sampling
division with rd of the FET. (rd, the drain-to-source
resistance, is low but not zero)
■ The most common technique for sampling voice in
PCM systems is to a sample-and-hold circuit.
Flat top sampling;

 The instantaneous amplitude of the analog

(voice) signal is held as a constant charge on
a capacitor for the duration of the sampling
period Ts.
 This technique is useful for holding the
sample constant while other processing is
taking place, but it alters the frequency
spectrum and introduces an error, called
aperture error, resulting in an inability to
recover exactly the original analog signal.
 The amount of error depends on how much
the analog changes during the holding time,
called aperture time. Sample& Hold circuit and flat top
 To estimate the maximum voltage error sampling
possible, determine the maximum slope of
the analog signal and multiply it by the
aperture time ∆T.
Mathematical derivation for the generation of PAM: =
*S(t)- sequence of flat top pulses
= -nTs)h(t- )d (Apply eqn 3)

Express the PAM signal as

s(t) = --(1)
where – sampling period
- sample value of m(t) obtained at time t=
*h(t) –standard rectangular pulse of unit amplitude and duration
is given by

h(t) = 1 , 0<t<T
, t=0,t=T ---(2)
0 , otherwise

By definition ,the instantaneously sampled version of m(t) is

m δ(t) = --(3)
(a)Rectangular Pulse h(t) (b) spectrum h(f),made up of
where - time shifted delta function. the magnitude |H(f)| and phase arg [H(f)]
*Convolve m(t) with the pulse h(t) then,
Using the shifting property of the delta function,
m δ(t) = --(4) m δ(t)*h(t) = --(5)
■ From eqn(1)&(5), M δ(f) =
PAM signal s(t) is mathematically equivalent to
Where fs -Sampling rate
m δ(t) ,the instantaneously sampled version of m(t)
and the pulse h(t) is Substitute eqn (8) into (7) yields
S(f) =
s(t) = m δ(t) * h(t) --(6)
Take fourier transform on both sides of eqn(6) Reconstruction:

Note:Convolving two time function is transformed

into the multiplication of the respective fourier
transform
S(f) = M δ(f)H(f) --(7)
For reconstruction pass s(t) through a low pass filter where assume that
Where S(f) = F[s(t)] the message is limited to bandwidth W and the sampling rate f s is
larger than the Nyquist rate 2W.
M δ(f) = F[m δ(t)]
From eqn (9) the spectrum of the resulting filter output is M(f)H(f).
∞
H(f) = F[h(t)] This output is equivalent to the original message m(t) through another
∑ G (f −m 𝑓 𝑠) low pass filter of frequency response H(f).
Adapt eqn g δ(t) 𝑚=−∞ From eqn (2) we note that the FT of the rectangular pulse h(t) is
To eqn (7) ,then the fourier transform M
δ(f) is related to the fourier transform M(f) of the --(10)
original message m(t) is given by
■ Thus by using flat top samples to generate a PAM signal,we introduced amplitude
distortion as well as a delay of T/2.
■ The distortion caused by the use of PAM to transmit an analog information bearing
signal is called as ‘Aperture Effect’ .
■ This effect is similar to the variation in transmission with frequency that is caused
by the finite size of the scanning aperture in television.
■ How to avoid this aperture effect?
It can be corrected by connecting an equalizer in cascade with the low pass
reconstruction filter.
Function: Equalizer has the effect of decreasing the in-band loss of the
reconstruction filter as the frequency increases in such a manner as to compensate
for the aperture effect.
 Quantization
■ process of constraining an input from a continuous or otherwise
large set of values (such as the real numbers) to a discrete set (such as the
integers).
Midtread
Uniform
Quantizati
Midriser
on
Non
Uniform

Uniform Quantization
• quantization step size remains same throughout the dynamic
range of the signal (i.e.,uniformly spaced)
• SNR varies with the input signal amplitude.

Non Uniform Quantization

• quantization step size varies with the amplitude of the input
signal
• SNR can be maintained constant.
Quantized&Non Quantized Waveform
The Mid-Rise type is so called because the origin lies in the
middle of a raising part of the stair-case like graph. The
quantization levels in this type are even in number.

The Mid-tread type is so called because the origin lies in the

middle of a tread of the stair-case like graph. The quantization
levels in this type are odd in number.

Both the mid-rise and mid-tread type of uniform quantizers are

symmetric about the origin.
 Uniform Quantization
 Applicable when the signal is in a finite range (f min ,f )
max

 The entire data range is divided into L equal intervals of

length Q (known as quantization interval of quantization
step size)
)|/L

 Q=
Interval I is mapped to the middle value of this interval

 We store/send only the index of quantized value

https://forms.office.com/Pages/ResponsePage.aspx?
id=eRkUvfC31k6UfaARuMD2DuIpPVhOWHRIjEGkR7xCtwNUQk9OMjgxVUZUT1RBS0FWRzBSMjVBSjFBNS4
u
Example

Note:
Q=(f max-f min)/L
Q=(1.5 - ( - 1.5 ) ) /4
=3/4
=0.75
Quantization interval
starts from -1.5 to 1.5
Intervals = (1.5+0.75)/2
=2.25/2
=1.125
 Quantization noise/error and SNR
■ Assume quantization error is uniformly Step1:Find out the noise power. So find mean
distributed over a single step size Mean:
( Δ )interval µ = E[Q]
Note: = )dq
= 1/ ; µ = 0
The pdf of uniformly distributed Variance:
random variable ‘x’ is denoted by σq2 =E[(q- µ)2] (since µ = 0)
=E[q2 ]
f x(X) = b- ;a<x<b
a = )dq ; σq2 = 2/12
0 ;else Where σq2 =quantization noise variance or
Note:In uniform quantizer the quantization power
interval always lies between – Δ/2 to +Δ/2 Assume ‘m’ bits are used to represent a
sample
The pdf of quantization error is given by Number of quantization levels (L)
Δ L=2m m= log2L
f q(Q) = ; – Δ/2 <q<+Δ/2
where m-no.of bits/sample
0 ;else Number of quantization levels
q-Random variable
Q-value the random variable takes L= 2 |x max| =2|x max|
Now we need to find SNR for uniform
L (or 2m )
■ Substitute in variance expression SNRo = 3P*2 2m
|xmax|2
σq2 = 2/12
= 2||xmax| 2
Inference:
12
2m It is clear that the o/p SNR increases exponentially(because
2 2m) with the
increase in the number of bits per sample.

Maximizing the SNR is nothing but minimizing the quantization noise

=(4 |x max| ) 2
power.
So if we represent each level by more number of bits then SNR
2 2m *12 increases automatically decreases the quantization noise power.
σq 2 = 1 |x max| 2 2-2m
3

Step2: To find average input signal power

Let us assume that average input signal (x(t)) power to be P .

The output SNR of an uniform quantizer

(SNR) = P
o
Example: Consider the previous SNR o equation
Consider a sinusoidal signal
SNR o = 2 2m

X(t) = Vm cos (2πfmt)

This sinusoidal signal is sampled and the samples are uniformly = 2 2m

quantized.
SNR o =2 2m

Now find the SNR at the o/p of the uniform quantizer.

Consider signal over one fundamental period.
SNR o (db) = 10log10 ( 22m )
The average signal power
= 10log10()
P= +10log10(22m )
= = 1.7609
+20mlog102
= dt =
Split the integrals into two.Integral means finding area for the 1.7609+6.02m
full period. SNR o (db)

First integral means from dt . Inference:

Second integral means from fmdt It is clear that each additional
bit of quantization increases the o/p SNR
= (T) by 6db.
 Non Uniform Quantization
Problems with uniform quantization:

-Only optimal for uniformly distributed signal

-Real audio signals(speech and music) are more concentrated near zeros
-Human ear is more sensitive to quantization errors at small values

Solution:
-Using non uniform quantization
*Quantization interval is smaller near zero.

Uniform
Compressor Expander
Quantizer

Model of non uniform quantizer

The type of quantization in which the quantization levels are unequal and
mostly the relation between them is logarithmic, is termed as a Non-uniform
Quantization.
Companding of Speech signal:

Compander = Compressor + Expander

-In Non - Uniform Quantizer the step size varies.

-The use of a non – uniform quantizer is equivalent to passing the baseband signal through a compressor and
then applying the compressed signal to a uniform quantizer.
-The resultant signal is then transmitted.
-At the receiver, a device with a characteristic complementary to the compressor called Expander is used to
restore the signal samples to their correct relative level.
-The Compressor and expander take together constitute a Compander.