Digital Modulation and Detection: Wha Sook Jeon
Digital Modulation and Detection: Wha Sook Jeon
Digital Modulation and Detection: Wha Sook Jeon
2
Signal and System Model (1)
System sends K = log 2 M bits every T seconds
Each bit sequence of length K comprises a message mi ={b1, … , bK}
The message i has the probability pi of being selected for transmission
( ∑iM=1 pi = 1 )
Each message is mapped to a unique analog signal si (t), is transmitted
over the channel during the time interval [0, T], and has energy
T
ESi = ∫ si2 (t ) dt , i = 1, , M
0
The transmitted signal is sent through an AWGN channel (a white
Gaussian noise process n(t) of power spectral density N0/2)
Received signal r(t)=s(t)+n(t)
The receiver should determine the best estimate of the transmitted
signal and outputs the best estimate of the transmitted message
mˆ = {bˆ1 , , bˆK }
3
Signal and System Model (2)
Goal of the receiver design
− minimizing the probability of message estimation error
M
Pe = ∑ p (mˆ ≠ mi | mi sent) p (mi sent)
i =1
4
Geometric Representation of Signals (1)
By representing the signals geometrically, we can solve for the
optimal receiver design in AWGN based on a minimum distance
criterion
Basis function representation
− Any set of M real energy signals S = {s1 (t ), , sM (t )} can be
represented as a liner combination of N (≤ M) real orthonormal
basis functions {φ1 (t ),, φ N (t )} according to Gram-Schmidt
orthogonalization procedure
N
: Modulation
si (t ) = ∑ sijφ j (t ), 0≤t <T
j =1
0
0 if i ≠ j
− Real coefficient representing the projection of si(t) onto φ j (t )
T
sij = ∫ si (t )φ j (t )dt : Demodulation
0
5
Geometric Representation of Signals (2)
Basis set of linear pathband modulation technique
− φ (t ) = 2 cos(2πf t ), φ2 (t ) =
2
sin( 2πf c t )
1 c
T T
− With fcT >> 1 for orthonormality
Transmitted signal
2 2
si (t ) = si1 cos(2πf c t ) + si 2 sin( 2πf c t )
T T
The basis set can include a bandpass pulse-shaping filter g(t) to
improve the spectral characteristics of the transmitted signal
si (t ) = si1 g (t ) cos(2πf c t ) + si 2 g (t ) sin(2πf c t )
The pulse shape g(t) must maintain the orthonormal property of basis
functions
T T
∫ g (t ) cos (2πf c t )dt = 1, ∫ g 2 (t ) cos(2πf c t ) sin( 2πf c t )dt = 0
2 2
0 0
6
Geometric Representation of Signals (3)
Signal space representation
− Signal constellation point of the signal si(t)
■ s i = ( si1 , siN ) ∈ R N
■ the vector of coefficients in the basis representation of si(t),
that is, si (t ) = ∑ sijφ j (t )
N
j =1
■ One-to-one correspondence
between si(t) and si
− The distance between two signal
constellation points si and sk
N
si − sk = ∑ (s
j =1
ij − skj ) 2
T
= ∫
0
( si (t ) − sk (t )) 2 dt
8
Receiver Structure and Sufficient Statistics (2)
In this receiver structure
− r (t ) = si (t ) + n(t )
= ∑ j =1 sijφ j (t ) + ∑ j =1 n jφ j (t ) + nr (t )
N N
= ∑ j =1 ( sij + n j )φ j (t ) + nr (t ) = ∑ j =1 rjφ j (t ) + nr (t )
N N
T T
− sij = ∫0 si (t )φ j (t )dt , n j = ∫0 n(t )φ j (t )dt
− nr(t) is the “remainder” noise which is orthogonal to signal
space
9
Receiver Structure and Sufficient Statistics (3)
(r1 , r2 , nr )
r = (r1 , r2 )
[r (t ) = r1φ1 (t ) + r1φ1 (t ) + nrφnr (t )]
10
Receiver Structure and Sufficient Statistics (3)
Goal of receiver design is to minimize the error probability
Pe = p (mˆ ≠ mi | r (t )) = 1 − p (mˆ = mi | r (t ))
Maximizing
p ( si sent | r (t )) = p (( si1 , , siN ) sent | (ri1 , , riN ) , nr (t ))
p (( si1 , , siN ) sent, (ri1 , , riN ) , nr (t ))
=
p ((ri1 , , riN ) , nr (t ))
p (( si1 , , siN ) sent, (ri1 , , riN )) p (nr (t ))
=
p ((ri1 , , riN )) p (nr (t ))
= p (( si1 , , siN ) sent | (ri1 , , riN ))
r =(r1, …, rN) is a sufficient statistic for r(t) in optimal detection
of transmitted message
11
Decision Region
Optimal receiver selects m ˆ = mi corresponding to constellation si
that satisfies p( si sent | r ) ≥ p( s j sent | r ) for all j ≠ i
Design region: Z i = {r : p( si sent | r ) > p( s j sent | r ) ∀ j ≠ i}
12
Maximum Likelihood Decision Criteria (1)
Optimal receiver selects m ˆ = mi corresponding to constellation si
that maximizes
p (r | si ) p ( si )
p ( si | r ) =
p(r )
p (r | si ) p ( si )
arg max = arg max p (r | si ) p ( si )
si p(r ) si
Likelihood function: L( si ) = p (r | si )
A maximum likelihood receiver outputs m ˆ = micorresponding to
constellation si that maximizes L(si)
13
Maximum Likelihood Decision Criteria (2)
Conditional distribution of r
− Since n(t) is a Gaussian random process, r(t)=si(t)+n(t) is also a
Gaussian random process and n(t) has a zero mean.
− rj = sij + n j
µ r |s = E[rj | sij ] = E[ sij + n j | sij ] = sij
j i
14
Maximum Likelihood Decision Criteria (3)
j =1 ( N 0 ) N0 j =1
N0 j =1 N0
minimizing
15
Maximum Likelihood Decision Criteria (4)
si si
j =1
Decision Region
Z i = {r : r − si < r − s j ∀ j = 1, , M , j ≠ i}
16
Union Bound of Error Probability (1)
Aik: the event that r − sk < r − si given that the constellation point si
was sent
If the event Aik occurs, the constellation will be decoded in error.
M M
Pe (mi sent) = p Aik ≤ ∑ p (Aik )
k =1 k =1
k ≠i k ≠i
a one-dimensional
sk − si p ( Aik ) = p ( r − sk < r − si | si sent)
Gaussian r.v.
= p ( ( si + n) − sk < ( si + n) − si )
= p ( n − ( sk − si ) < n )
17
Union Bound of Error Probability (2)
The event Aik occurs if n > d ik 2 , where d ik = si − sk
d ik ∞ 1 − v2 d ik
p ( Aik ) = p n > = ∫d / 2 exp dv = Q
2 πN 0 N0 2N
ik
0
M d ik
Pe (mi sent) ≤ ∑ Q
2N
k =1
k ≠i
0
M
1 M M d ik
Pe = ∑ p (mi ) Pe (mi sent) ≤ ∑ ∑ Q
M i =1 2N
i =1 k =1
k ≠i
0
18
Approximation of Error Probability
The minimum distance of constellation: dmin
d min
− Pe ≤ ( M − 1)Q
: looser bound
2 N 0
The number of neighbors at the minimum distance: M d min
− d min
Pe ≈ M d min Q
2N
0
− (
In case of binary modulation (M=2): Pb = Q d min 2N 0 )
Gray code: mistaking a constellation point for one of its nearest
neighbors results in a single bit error
− P ≈ Pe
b
log 2 M
19
Amplitude and Phase Modulation
20
Amplitude and Phase Modulation (1)
Over the time interval Ts, K (log2M) bits are encoded in the amplitude
and/or phase of the transmitted signal
− Signal: s(t ) = sI (t ) cos(2π f ct ) − sQ (t ) sin(2π f ct )
− In signal space:
s (t ) = si1 (t )φ1 (t ) + si 2 (t )φ2 (t )
φ1 (t ) = g (t ) cos(2π f c t + φ0 )
φ2 (t ) = − g (t ) sin(2π f c t + φ0 )
There are three main types of amplitude/phase modulation:
− Pulse Amplitude Modulation (MPAM) : Uses amplitude only
− Phase Shift Keying (MPSK) : Uses phase only
− Quadrature Amplitude Modulation (MQAM)
21
Amplitude and Phase Modulation (2)
Amplitude/Phase
Modulator
22
Amplitude and Phase Modulation (3)
Amplitude/Phase
Demodulator
Coherent detection (φ = φ0 )
24
Pulse Amplitude Modulation (MPAM) (1)
25
Pulse Amplitude Modulation (MPAM) (2)
Over each symbol period, the MPAM signal associated with the ith
constellation has energy Es = T si2 (t )dt = T Ai2 g 2 (t ) cos 2 (2π f c t )dt = Ai2
∫ ∫
s s
i 0 0
M
1
− Average energy: Es =
M
∑A
i =1
i
2
(− ∞, Ai + d ) i = 1,
Z i = [ Ai − d , Ai + d ) 2 ≤ i ≤ M − 1,
[ A − d , ∞ ) i=M
i
26
Pulse Amplitude Modulation (MPAM) (3)
27
Phase Shift Keying (MPSK) (1)
Encodes information in the phase of the transmitted signal
Transmitted signal over one symbol time:
{
si (t ) = Re A g (t )e j 2π (i −1) / M e j 2πfct }
2π (i − 1)
= A g (t ) cos 2π f c t +
M
2π (i − 1) 2π (i − 1)
= A g (t ) cos cos 2 π f t − A g (t ) sin M sin 2π f c t
M
c
28
Phase Shift Keying (MPSK) (2)
29
Phase Shift Keying (MPSK) (3)
30
Phase Shift Keying (MPSK) (4)
31
Quadrature Amplitude Modulation (MQAM) (1)
32
Quadrature Amplitude Modulation (MQAM) (2)
33
Quadrature Amplitude Modulation (MQAM) (3)
Decision Regions
for 16-QAM
34
Differential Modulation (1)
Information in MPSK and MQAM signals is carried in the signal phase
− MPSK and MQAM require coherent detection
− Phase recovery mechanism required in receiver
− Coherent demodulation
■ makes receiver complex
■ is hard in rapidly fading channel
■ is more susceptible to phase drift of the carrier
The principle of differential modulation is to use the previous symbol as
a phase reference for current symbol for avoiding the need for a coherent
phase modulation
− Differential BPSK (DPSK)
■ 0-bit: no change in phase, 1-bit: a phase change of π
− Differential QPSK (DQPSK)
■ 00: no change in phase, 01: a phase change of π/2
■ 10: a phase change of - π/2, 11: a phase change of π
35
Differential Modulation (2)
Phase Comparator
− Transmitted signal: s(k ) = Ae j (θ ( k )+φ0 )
− Received signal at time k: r (k ) = r1 (k ) + jr2 (k ) = Ae j (θ ( k )+φ −φ ) + n(k ) 0
37
Modulation with Quadrature Offset (1)
38
Modulation with Quadrature Offset (2)
Modulator with
quadrature offset
39
Pulse Shaping (1)
40
Pulse Shaping (2)
Raised Cosine Pulse
− These pulses are designed in the frequency domain
1− β
Ts 0≤ f ≤
2Ts
P( f ) =
Ts 1 − sin π Ts
f −
1
1− β
≤ f ≤
1+ β
2 β 2Ts 2Ts 2Ts
sin π t / Ts cos βπ t / Ts
p (t ) = ×
π t / Ts 1 − 4 β 2t 2 / Ts2
41
Pulse Shaping (3)
P(f)
42
Pulse Shaping (4)
p(t)
43
Error Probability of Digital Modulation over
AWGN Channel
44
Signal-to-Noise Power Ratio (SNR)
In an AWGN channel
− j 2πf t
Modulated (transmitted) signal: s (t ) = Re{u (t )e c }
− Received signal: r (t ) = s (t ) + n(t )
− n(t): a white Gaussian random process with mean zero and power spectral
density N0/2
SNR
− Ratio of the received signal power Pr to the power of the noise within the
bandwidth of the transmitted signal
− P
SNR = r =
Es
=
Eb
N 0 B N 0 BTs N 0 BTb
− In system with interference
Pr
■ SINR =
N 0 B + PI
45
Bit/Symbol Errors
For pulse shaping with Ts=1/B (e.g., raised cosine pulse with β=1), SNR=Es/N0
For general pulse, Ts= k/B and SNR = Es N 0 ×1 k
Define
− SNR per symbol: γ s = Es N 0
− SNR per bit: γ b = Eb N 0
We are interested in bit error probability Pb as a function of γ b
Approach
− First, compute the symbol error probability Ps as a function of γ s
− Then, obtain bit error probability as a function of SNR per bit using assumptions.
■ The symbol energy is divided equally among all bits,
■ Gray encoding is used
− These assumptions for M-ary signaling lead to the approximations
γs and
γb ≈ Pb ≈
Ps
log 2 M log 2 M
46
Error Probability for BPSK
2 Eb
Pb = Ps = Q
(
= Q 2γ b
)
N0
47
Error probability for QPSK
The QPSK system is equivalent to the system consisting of BPSK modulation on
both the in-phase and quadrature components of the signal.
The bit error probability on each component is the same as for BPSK
− Pb = Q( 2γ b )
[
The symbol error probability is Ps = 1 − 1 − Q( 2γ b ) ]
2
Since the transmitted symbol energy Es is split between each branch, the signal
energy per branch is ( Es 2)
The symbol error probability is the probability that either branch has a error
P = 1 − [1 − Q ( γ )]
− 2
s s
For the same Eb N 0 and therefore the same average probability of bit error, QPSK
system transmits data at twice the bit rate of a BPSK system for the same channel
bandwidth.
[ ( )]
− Ps = 1 − 1 − Q γ s
2
( ) (
≈ 2Q γ s = 2Q 2γ b )
( )
Ps ≈ 2Q γ s
48
Error Probability for MPSK
Es
φ1 d min = 2 Es sin(π M ) ,
M d min = 2
(
Ps = 2Q 2γ s sin(π M ) )
49
Error Probability for MPAM
The constellation for MPAM is Ai = (2i − 1 − M )d , i = 1,2,..., M
Since each of the M-2 inner constellation points has two nearest neighbor at distance
2d, Ps ( si ) = p ( n > d ), i = 2,..., M − 1
For outer constellation points, there is only one nearest neighbor.
M
The average energy per symbol for MPAM is Es = ∑ Ai 2 = 1 ( M 2 − 1)d 2
1
M i =1 3
The symbol error probability Ps in terms of the average energy as
M −2
( ) ( )
M
1 2
Ps =
M
∑ Ps (si ) =
i =1 M
× 2 × Q 2d 2N0 +
M
× Q 2d 2N0
2( M − 1) 6γ
Ps = Q s
M M −1
2
50
Error Probability for MQAM
MQAM system can be viewed as two MPAM systems with signal constellations of
size L = M transmitted over the in-phase and quadrature signal components, each
with half the energy of the original MQAM system.
The constellation points in the in-phase and quadrature branches take values
Ai = (2i − 1 − L)d , i = 1,2,..., L
The symbol error probability for each branch is
2( M − 1) 3γ s
Ps ,branch = Q
M M −1
The probability of symbol error the for MQAM system is Ps = 1 − (1 − Ps ,branch ) 2
If we take a conservative approach and set the number of nearest neighbors to be
four,
3γ
Ps ≈ 4 × Q s
M −1
51
Summary in Error Probability for Coherent Modulation (1)
Many of the exact or approximation values for Ps derived for coherent modulation
are in the following form:
(
Ps (γ s ) ≈ α M Q β M γ s )
− αM : the number of nearest neighbors at the minimum distance dmin
− βM : a constant that relates the minimum distance to average symbol energy
Performance specifications are generally most concerned with the bit error rate as a
function of the bit energy.
Pb (γ b ) ≈ αˆ M Q βˆM γ b
− With Gray coding and high SNR, αˆ M = α M log 2 M and βˆM = β M × log 2 M
52
Summary in Error Probability for Coherent Modulation (2)
Modulation Ps (γ s ) Pb (γ b )
BPSK (
Pb = Q 2γ b )
QPSK ( )
Ps ≈ 2Q γ s (
Pb ≈ Q 2γ b )
MPAM 2(M − 1) 6γ s
2(M − 1) 6γ b log 2 M
Ps = Q Pb ≈ Q
M −1 M 2 −1
2
M M log 2 M
π 2 π
MPSK Ps ≈ 2Q 2γ s sin Pb ≈ Q 2γ b log 2 M sin
M log 2 M M
MQAM
3γ s 4 3γ b log 2 M
Ps ≈ 4Q Pb ≈ Q
M − 1 log 2 M M −1
53
Flat Fading Channel
Outage Probability
Average Probability of Error
54
Performance Criteria (1)
55
Performance Criteria (2)
When the fading coherence time is on the order of a symbol time (Ts ≈ Tc )
− The signal fading level is roughly constant over a symbol period
− The error correction coding techniques can recover from a few bit errors
− An average error probability is a reasonably good figure
When the signal fading is changing slowly (Ts << Tc )
− A deep fade affects simultaneously many symbols
− Large error bursts that cannot be corrected for with coding of reasonable
complexity
− Outage probability
− When the channel is modeled as a combination of fast and slow fading (e.g.,
log-normal shadowing with fast Rayleigh fading), outage and average error
probability is often combined
When Ts >> Tc , the fading will be averaged out by the matched
filter in the demodulator
− For very fast fading, performance is the same as in AWGN
56
Outage Probability (1)
Ts
γs Outage
γ0
out s 0
0
γs
57
Outage Probability (2)
− Average SNR
−γ0
■
γs =
ln(1 − Pout )
58
Average Error Probability
The averaged probability of error is computed by integrating the error
probability in AWGN over the fading distributions.
∞
Ps = ∫ Ps (γ ) pγ s (γ )dγ
( )
0
− An error probability in AWGN with SNR γ : Ps (γ ) ≈ α M Q β M γ s
In Rayleigh fading,
α M
( )γ
∞
1 0.5β M γ s ≈ αM
Ps = ∫ α M Q β M γ ⋅ e −γ / γ s
dγ = 1−
2 1 + 0.5β M γ s 2β M γ s
0 s
− BPSK: Pb (γ b ) = Q( 2γ b )
1 γb
Pb = 1 −
2 1+ γ b
59
Average Pb for MQAM in Rayleigh Fading and AWGN
60