10/3/03 EC2500MPF/FallFY04 42
II.B Baseband Transmission
Digital Baseband ReceptionMatched filter
DefinitionImplementation (correlator)
Application to digital baseband signalM-ary Baseband ReceptionBrief Review of Probability and Noise Concepts
Basic pdf definitionsWhite noiseNarrowband noise
Bit Error RateError probabilityThreshold definitionApplication to the binary matched filter detectorExamples
M-ary baseband PerformanceApplication to the CD Format
Speech rangeD/A converter issuesOversamplingNoise shaping & Dither effects
(Reception & Applications)
10/3/03 EC2500MPF/FallFY04 43
10) Digital Baseband Reception
• Goal: to recover s(t) from potentially noisy received signal
– use a filter to decrease the effect of noise
• Generic filter does not take advantage of known signal shape transmitted
better result obtained when using that information
“matched filter”
s(t)
�V
�VTb
t…
compareto zero
12 bn T� �
�� �� �
10/3/03 EC2500MPF/FallFY04 44
• Matched Filter
� Definition: a matched filter is a linear filter which minimizes the output signal to noise ratio (SNR) � at time T, where ��is definedas:
� �
� �
� � � �
� � � �
2020
22
2
j fT
n
s Tn t
S f H f e df
S f H f df
�
�
�
��
��
��
��
�
�
�
�
� Goal: find H(f) which minimizes �
� Proof: Use Schwartz’s inequality which states:
equality holds only when A(n) = KB*(x)K real constant
� � � � � � � �2
2 2A x B x dx A x dx B x dx�� � �
H (f)s(t) + n(t) s0(t) + n0(t)T
10/3/03 EC2500MPF/FallFY04 45
� � � �
� � � �
22
2
j fT
n
S f H f e df
S f H f df
�
�
��
��
��
��
�
�
�
� �
� �� � � �
� �
� �� � � �
22
2
222 2
j fTn
n
j fTn
n
S fdf H f S f e df
S f
S fdf H f S f e df
S f
�
�
�� ��
�� ��
�� ��
�� ��
� �
� �
� �
� �
� �
� �� � � �
� � � �
22
2
nn
n
S f dfH f S f df
S f
S f H f df�
��
��
��
��
�
� �
� �
�
� � � �� �
� �� � � �
22
2 2j fT j fTn
n
S fS f H f e df H f S f e df
S f� �
�� �
BA
� �
� �
2
n
S f dfS f
���
��
� � �
� �
� �� � � �* 2j fT
nn
S fKH f S f e
S f��
�
� �� �
� �
*2j fT
n
S fH f K e
S f��
� �
�max obtained when
10/3/03 EC2500MPF/FallFY04 46
� Example:
When n(t) is white noise �
� � � �
� � � �
* 2j fTH f KS f e
h t Ks T t
��� �
�
� �
� � 2n nS f ��
1
Tt
s(t)
t
h(t)
10/3/03 EC2500MPF/FallFY04 47
� Matched Filter Implementation (correlator)
� � � � � �� � � �
� � � �� � � �
� � � �� � � �
y t s t n t h t
s n h t d
s n s d
� � � �
� � �
��
��
��
��
� � �
� � �
� �
�
�
�
� � � �h Ks T� �� �
H (f)s(t) + n(t)
Ty(t)
�T
10/3/03 EC2500MPF/FallFY04 48
• Matched Filter Detector for Digital Baseband
...
Tb
s(t)
� �0
.bT
dt�y(T)s(t)
10/3/03 EC2500MPF/FallFY04 49
10/3/03 EC2500MPF/FallFY04 50
10/3/03 EC2500MPF/FallFY04 51
• M-ary Baseband Reception
– We can transmit more than two symbols
– Example: 4-ary baseband communications
– How to apply binary result to M-ary case ?
10/3/03 EC2500MPF/FallFY04 52
11) Brief Review of Probability and Noise Concepts
• Probability distribution function F (x) =
• Probability density function f (x)
• Expected value E (x) =
• Variance 2x� �
• Basic pdfs
(1) Uniform pdf
f (x)
x
f (x) =
10/3/03 EC2500MPF/FallFY04 53
(2) Gaussian pdf
f (x)
x
f (x) =
10/3/03 EC2500MPF/FallFY04 54
x(n) is random with a constant PSD
• White Noise
Gx(f)
f
N0/2
• Narrowband Noise
– most communication systems contain bandpass filters
– white noise gets transformed into BP noise
– when noise band is small compared to center frequency fc
BP noise called narrowband noise
expressed as:
� � � � � � � � � �
� � 2
cos 2 sin 2
Re c
c c
j f t
w t x t f t y t f t
r t e �
� �� �
� �� � �
complex function
Quadraticnoisecomponents
10/3/03 EC2500MPF/FallFY04 55
� �
out
2V
P
E V n
�
�
�� �� �
�
w(n) v(n)BPF
� �20, wN �
–fL fL fH–fHf
H(f)white noise
10/3/03 EC2500MPF/FallFY04 56
12) Bit Error Rate
• Evaluate errors made during transmission of hits
• Errors occur when:
receive “1” when “0” is sentreceive “0” when “1” is sent
• How to decide if you receive a “1” or “0” when transmission is noisy ??
detection theory
1
PM PFA
1
0 0Send Receive
10/3/03 EC2500MPF/FallFY04 57
r(t)=s(t)+Kw(t)
N(0,�w2=1)
10/3/03 EC2500MPF/FallFY04 58
• f (w) =
• f (r|s0) =
• f (r|s1) =
For which range of “r” do you decide you sent a
“1” (s1) “0” (s0)
�0
r
• Assume you have additive white noise distortion atthe receiver
r(t)=si(t)+w(t), si(t)=s0 , s1
How to pick �0 ?
10/3/03 EC2500MPF/FallFY04 59
• Goal: To quantify likelihood of making an error in assigning bit values at the
receiver.
• Statistics of r(n)
– mean of r(n) ?
– r(n) deterministic ?random ?
Assume transmission is noisy
� � � � � �r n s n w n� �
receivedsignal
transmission noise(random Gaussian)
signalsent
� �20, wN �0
1
ss
���
10/3/03 EC2500MPF/FallFY04 60
• Notation
H1: receive a “1” (s1)
H0: receive a “0” (s0)
� �
� �1 1
0 0
P H s
P H s
�
�
• Correct decisions:
� �
� �
1 0
0 1
P H s
P H s
�
�
• Incorrect decisions:
eP �
• Overall probability of error:
• Pe when there is equal probability of sending s0 & s1
10/3/03 EC2500MPF/FallFY04 61
2
2
/ 2
0
1 1( ) erfc22 2
2erfc( ) 1 1 erf ( )
u
xx
u
xQ x e du
x e du x
�
�
�
�
�
� �� � � �
� �
� � � �
�
�
• Definitions: Q, erf, & erfc functions
• Assume s0=-V & s1=+V
0
1 0 0 1 012 ( | ) ( | ) ( | )2eP P H s P H s f r s dr
�
�
� � � �
10/3/03 EC2500MPF/FallFY04 62
BER for single sample detector
10/3/03 EC2500MPF/FallFY04 63
• How to select the threshold �0 ?
Maximum likelihood detector approach
Minimize the overall probability of error Pe
1 0 0 0 1 1( | ) ( ) ( | ) ( )eP P H s P s P H s P s� �
�0
r
20 1 0
01 1 0
In2
ws s PP s s
��
� ��� � � �
��
(More details in EC4570…)
10/3/03 EC2500MPF/FallFY04 64
• Application to Binary Matched Filter Detector
• Need statistics on y = y1 – y0
• pdf of y: ?
s1(t)
s0(t)
r(t)
y1
y0
y
Tb
compare tothreshold�
� �0
bTdt���
�
�
�
� �0
bTdt��
Is y random or deterministic?
r(t)=si(t)+w(t)
10/3/03 EC2500MPF/FallFY04 65
� � � �
� � � �� � � � � � � � � �
� � � � � � � �
1 10
1 1 00
1 10 0
or
b
b
b b
T
T
i i
T T
i
y r t s t dt
s t w t s t dt s t s t s t
s t s t dt w t s t dt
�
� � �
� �
�
�
� ��������
� � � �
0
1 0
y
y y t y t
�
� �
�
10/3/03 EC2500MPF/FallFY04 66
10/3/03 EC2500MPF/FallFY04 67
10/3/03 EC2500MPF/FallFY04 68
Bit error rate for matched filter detector
0 10
2 20 1
0 0
( ) ( )
1 ( ) ( )2
b
b b
T
T T
s t s t dt
E
E s t dt s t dt
� �
� �� �� �� �
� �
�
� �
10/3/03 EC2500MPF/FallFY04 69
• How to compute the threshold��0 ?
Recall for simple detector, the threshold was selected asthe mid point between the two means for basicproblem.How can we apply the result here ?
E[y|s0]=
E[y|s1]=
10/3/03 EC2500MPF/FallFY04 70
• Example
• Design a matched filter detector for the two signals.
• Find Pe when the additive white noise has apower P = 10–3 w/Hz.
T = 8 10–3 s
T
–1t
s0(t)
+1
Tt
s1(t)
+1
10/3/03 EC2500MPF/FallFY04 71
10/3/03 EC2500MPF/FallFY04 72
10/3/03 EC2500MPF/FallFY04 73
• Example
• Design a matched filter detector for the two signals.
• Find Pe when the additive white noise has a powerPe = 0.1 w/Hz.
t1
–1
s0(t)
+1 0.5 1t
s1(t)
+1
–1
sin(2�t)
10/3/03 EC2500MPF/FallFY04 74
10/3/03 EC2500MPF/FallFY04 75
• M-ary Baseband Performance
Assume we send M = 4 different levels(Bi= 0, A, 2A, 3A, i=0,…3)
r(t) ?� �0
bTdt��
Assume additive Gaussian noise r(t)=s(t)+n(t)
10/3/03 EC2500MPF/FallFY04 76
• P(error in receiving B2)=
10/3/03 EC2500MPF/FallFY04 77
2 2
20 0
P(error in receiving B ) 22 4
s sA T A TQ erfc
N N
� � � �� � � �� � �� � � �� � � �
• P(error in receiving B1)=
• P(error in receiving B0)=
• P(error in receiving B3)=
• Assume each error may occur as likely as theothers
Pe=
10/3/03 EC2500MPF/FallFY04 78
• Assume we send M different levels, computethe overall probability of error becomes:
10/3/03 EC2500MPF/FallFY04 79
� Dynamic range for audio signals Ref [3]
13) Application: Compact Disk
10/3/03 EC2500MPF/FallFY04 80
� Signal reproduction in CD Player
Digitaloversampling
filterX4
NoiseshaperH(f)Digital
input16 bits
44.1 KHz
x(n)y(n)
28 bits176.1 KHz
14 bits
14 bitsD/A LPF
–20 KHzf
xa(n)
20 KHz
88.2 KHzf
x(n)
44.1 KHz0
10/3/03 EC2500MPF/FallFY04 81
• Why use oversampling ?
• First, assume we don’t use oversampling.
x(r)z(t) y(t)LPF
Hr(j�)14 bitsD/A
001
anal
og o
utpu
t
010 011 100 101 110 111000
7654321
0
Input/outputrelationship fora unipolar D/A(3 bits)converter.
Ideal D/A Converter
t
x1(t)
t
z(t)
Practical D/A Converter
(p. 331)
x(n)x1(t) z(t)Hz(f)
hz(t)
tT
1
10/3/03 EC2500MPF/FallFY04 82
� D/A Converter Output Expression
� � � � � �
� � � � � �
1
1
k
z
x t x k t kT
z t x t h t
�� �
� �
�
� �
� �
� �
0
2
1 1
sin 22
T j tz
j T
j T
H e dt
ej
Te
�
�
�
�
�
�
�
�
�
�
�
� ��
� �� � �
� �
�
�s 2�s 3�s
�
|X1(j�)|
�s 2�s 3�s
�
|Z(j�)|
�s 2�s 3�s
�
|Hz(j�)|
� � � � � �1 zZ j X j H� � ��
2 /s T� ��
10/3/03 EC2500MPF/FallFY04 83
• Need for LPF filter ?
– to smooth out output steps
– to undo distortion added by D/A converter
�s 2�s
�
|Z(j�)|
�
|Hr(j�)|
�
Hr(j�)
10/3/03 EC2500MPF/FallFY04 84
• Oversampling in the Time Domain
Digitaloversampling
filterX4
NoiseshaperHs(f)16 bits
44.1 KHz
x(n)y(n)
176.1 KHz
14 bitsD/A LPF
a(n)
input to upsampler by 4
output to upsampler by 4
output of FIR filter
n�4321
010
n210
n�
001 010 � 001 000 000 000 010
x(n)y2(n)
y(n)� 4 FIR filter
10/3/03 EC2500MPF/FallFY04 85
f
Xa(f)
(KHz)
X(f)
f(KHz)132.3 176.488.244.10
Y1(f)
f(KHz)176.40
f
Y(f)
(KHz)
f
A(f)
(KHz)
without oversampling
with oversampling
Y1(f) after FIR filter
after analog LPF
� Advantages of Oversampling
10/3/03 EC2500MPF/FallFY04 86
• Example:
Assume1) you have an analog signal which spans [0 20KHz],2) the D/A converter has a sampling frequency fs=176.4KHz.Determine the characteristics (order and cutoff frequency) forthe anti-imaging Butterworth type filter which satisfy thefollowing specifications:
1) Image frequencies must be attenuated by at least 40dB2) Signal components may be altered by a maximum of0.5dB
� �1/ 22
1( )1 / n
c
H ff f
�� ��� �
10/3/03 EC2500MPF/FallFY04 87
10/3/03 EC2500MPF/FallFY04 88
– Goal: to decrease noise in the audio band
� Noise Shaping Filterno
ise
leve
l
noise shapingcharacteristic
noise levelwithout shaping
KHz
f20 88
10/3/03 EC2500MPF/FallFY04 89
�Dither
Figure 2.121 - Coding of Dithered SignalFigure 2.121 - Coding of Dithered Signal
Figure 2.122 - Fourier Transform of Dithered SignalFigure 2.122 - Fourier Transform of Dithered Signal
10/3/03 EC2500MPF/FallFY04 90
Poh1man Fig. 6-27 An example of noise shaping showing a 1kHz sinewave with -90 dB amplitude; measurements are madewith a 16 kHz lowpass filter.A. Original 20 bit recording.B. Truncated 16 bit signal.C. Dithered 16 bit signal.D. Noise shaping preserves information in lower 4 bits.
Ref [4,5]
�Dither & noise shaping effects
10/3/03 EC2500MPF/FallFY04 91
CD Section References:
[1] S. Mitra, Digital Signal Processing, McGraw-Hill, 1998.[2] K. Pohlmann, A. Red, Compact Disk Handbook, 2, 1992.[3] H. Nakajima and H. Ogawa, Compact Disk Technology, IOS Press, 1992.[4] B. Evans, Real Time Digital Signal Processing Lab, Fall 2003 (lecture 10 notes).[5] K. Pohlmann, Principles of Digital Audio, McGraw-Hill, 1995