Sampling and Aliasing

Any continuous time signal can be sampled and processed in the sampled-data domain
A/D converters look at signals at only discrete points in time Computer models of continuous systems must operate at discrete time intervals Even analog filters can use continuous or sampled time signal representations

Multiple continuous time signals can yield exactly the same sampled data signal aliasing
Lets look at samples taken at 1sec intervals of several sinusoidal waveforms
A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 1

Sampling Sinewaves
T = 1s fs = 1/T = 1MHz fx = 101kHz
voltage

time

v(t) = sin [2(101000)t]

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 2

Sampling Sinewaves
T = 1s fs = 1MHz fx = 899kHz voltage

time

v(t) = - sin [2(899000)t]

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 3

Sampling Sinewaves
T = 1s fs = 1MHz fx = 1101kHz voltage

time

v(t) = sin [2(1101000)t]

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 4

Aliasing
Multiple continuous time signals can produce identical series of sampled voltages The translation of signals from NfSfIN down to fIN is called aliasing
Sampling theorem: fs > 2fB

If aliasing occurs, no signal processing operation downstream of the sampling process can recover the original continuous time signal
If you dont like it, sample faster!
A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 5

Time vs Frequency Domain

fB

T multiply convolve

fS=1/T

Time
A/D DSP

Frequency
2002 B. Boser 6

EECS 247 Lecture 8: Sampling

Nomenclature
Continuous time signal x(t) Sampling interval T Sampling frequency fs = 1/T Sampled signal x(kT) = x(k)

x(t)

x(kT) x(k)

T time
A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 7

Sampled Signal Properties

Its obvious from the preceding slides that the frequency content of a continuous signal can be grossly distorted by sampling What do we know about the relationships between x(t) and x(k)?
Well assume that x(t) is stationary; that is, its mean is constant and its autocorrelation R(t1,t2) depends only on the time difference t1-t2

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 8

Sampled Signal Properties

If x(t) is stationary, then

(Athanasios Papoulis, Signal Analysis, 1977, Section 9.4.):

x(k) is also stationary Sampling doesnt change the mean: E{x(k)}=E{x(t)}

E{} is the expectation operator

Sampling doesnt change energy: E{[x(k)]2}= E{[x(t)]2}

Continuous time energy will show up someplace in the frequency domain after sampling
Aliasing may move continuous signal frequency components to wrong frequencies
EECS 247 Lecture 8: Sampling 2002 B. Boser 9

A/D DSP

Anti-Aliasing Filter
fs
time time

Aliasing

fB Anti-Aliasing Filter Filtered Spectrum fB

A/D DSP

frequency

time

fs

frequency

fs

time

EECS 247 Lecture 8: Sampling

2002 B. Boser 10

Sampled Signals
Sampled data signals are valid only at sampling instances In an actual circuit, the signal transitions to the new value between sampling instances The value between sampling instances is insignificant and often erroneous (e.g. amplifier slewing distortion)
A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 11

Zero-Order Hold
1 0.8 0.6

0.4 0.2 Amplitude

Reconstructs CT signal from SD signal Frequency response?

-0.2 -0.4

-0.6 -0.8 sampled data after ZOH -1 0 0.5 1 1.5 Time 2 2.5 3 x 10 3.5
-5

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 12

Zero-Order Hold
Step response Laplace transform

1 sTd =
Td time

e sTd sTd

Td 0

1 e sTd sTd

sin (2fTd ) jfTd e 2fTd

2002 B. Boser 13

A/D DSP

EECS 247 Lecture 8: Sampling

Spectrum of Reconstructed Signal

1 0.9 sampled signal sinc zoh signal

0.8

0.7

Magnitude

Reconstructed signal

SD signal: periodic spectrum

0.6

0.5

0.4 0.3

0.2

Sinc Td = T = 1/fs

0.1

0.5

1.5 Frequency

2.5 x 10

3
6

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 14

Spectrum of Reconstructed Signal

1 0.9 sampled signal sinc zoh signal 0.8

0.7

Magnitude

Reconstructed signal

SD signal: periodic spectrum

0.6

0.5

0.4 0.3

0.2

Sinc Td = 0.5T

0.1

0.5

1.5 Frequency

2.5 x 10

3
6

A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 15

Summary
Quantization in Time: Continuous time (CT) vs Sampled Data (SD) Sampling theorem: fs > 2fB IF sampling theorem is met, CT signal can be recovered from SD signal without loss of information ZOH reconstructs CT from SD signal
A/D DSP

EECS 247 Lecture 8: Sampling

2002 B. Boser 16