OPTIMAL RECEIVERS

FOR AWGN CHANNEL

by
Dr Debashis Adhikari

Reference: 1. Digital Communication by John Proakis, 4th Edition , McGrawHill

2. Digital Communication by John Proakis & Salehi, 5th Edition , McGrawHill

Dr D Adhikari, School of Electrical Engineering

TRANSMISSION PATH
Digital Modulator

Baseband /
Channel Encoder Mapper Bandpass
modulation

Note:
• We have not included multiplexing from other users, power amplifier, up-converter in the block diagram.
• The Digital Modulation block includes GSOP for signal constellation.
• The original message is embedded in the output of channel encoder (we have added (channel encoder) and
removed bits (source encoder) in the process)
• Each bit is of equal duration, Tb

Dean, School of Electrical Engineering

TRANSMISSION PATH
Step 1 - Mapper

Incoming bit sequence Convert bits into symbols by Bits are mapped into
from output of Channel grouping (eg 1symb=2bits symbols where
encoder (say) i.e., 2m = M, m=2, M=4) Si Є {S1, S2, S3, S4}, i
= 1 to M=4

100110010111000101010 01 00 11 00 10 11 10 00 10 10 10 S2 S1 S4 S1 S3 S4 S 3 S1 S3 S3 S3

Note:
• We have considered m = 2 thereby number of symbols M = 4
• [S1 S2 S3 S4 ] Ξ [00 01 10 11]

Dr D Adhikari, School of Electrical Engineering

 TRANSMISSION PATH
Step 2 (Mapping symbols to Signals onto Signal Space as using Grahm Schmidt Orthogonalization
Procedure)
∅1 ∅2 (t)

si1 𝑠2 𝑠1 𝑠11 , 𝑠12

𝑠12
∅2
𝑁
si= s 𝑠𝑖𝑗 𝜑𝑗 = 𝑠𝑖 𝑡
i2 𝑠11 𝑠11 ∅1
𝑗=1
∅𝑁
where i = 1 to M
𝑠3 𝑠4
Set of signals for M=4, N=2
siN 𝑠1 𝑡 = 𝑠11 𝜑1 𝑡 + 𝑠12 𝜑2 𝑡
𝑠2 𝑡 = 𝑠21 𝜑1 𝑡 + 𝑠22 𝜑2 𝑡
𝑠3 𝑡 = 𝑠31 𝜑1 𝑡 + 𝑠32 𝜑2 𝑡 Constellation Diagram
𝑠4 𝑡 = 𝑠41 𝜑1 𝑡 + 𝑠42 𝜑2 𝑡 e.g: QPSK constellation

Dr D Adhikari, School of Electrical Engineering

 Receiver
Transmitted 𝑠𝑖 𝑡 Channel
𝑟 𝑡
signal 𝑠𝑖 𝑡 DEMODULATOR DETECTOR 𝑠𝑖 𝑡
Estimated
signal
𝑛 𝑡 𝑠𝑖 𝑡
• Transmitter sends digital information by use of M signal waveforms {𝑠𝑖 𝑡 , i =1,2,.., M, 0 ≤ 𝑡 ≤ 𝑇}
• Received signal in the interval 0 ≤ 𝑡 ≤ 𝑇 is
𝑟 𝑡 = 𝑠𝑖 𝑡 + 𝑛 𝑡
𝑁0
where the power spectral density of noise is W/Hz
2

OBJECTIVE: To design a receiver, based on 𝑟 𝑡 that is optimum in the sense that it minimizes the
probability of making an error

Demodulator types: (a) Correlation demodulator (b) Matched filter demodulator

Dr D Adhikari, School of Electrical Engineering

CORRELATION DEMODULATOR Suppose we had transmitted the signal 𝑠𝑖 𝑡 which is not known
by the receiver. For the receiver signal 𝑟 𝑡 is being received.
∅1 (𝑡) Hence at the output of kth integrator will be (kth branch of the
𝑇 correlator):
𝑟1
𝑑𝑡 𝑇 𝑇 𝑇 𝑇

. 0 𝑟 𝑡 𝜑𝑘 𝑡 𝑑𝑡 = [𝑠𝑖 𝑡 + 𝑛 𝑡 ]𝜑𝑘 𝑡 𝑑𝑡 = 𝑠𝑖 𝑡 𝜑𝑘 𝑡 𝑑𝑡 + 𝑛 𝑡 𝜑𝑘 𝑡 𝑑𝑡
0 0 0 0
.
∅𝑘 (𝑡)
𝑇

𝑟 𝑡 = 𝑠𝑖1 𝜑1 𝑡 + 𝑠𝑖2 𝜑2 𝑡 +. . +𝑠𝑖𝑘 𝜑𝑘 𝑡 + ⋯ 𝑠𝑖𝑁 𝜑𝑁 𝑡 𝜑𝑘 𝑡 𝑑𝑡 + 𝑛𝑘

. 𝑇
𝑟𝑘 0
𝑑𝑡 𝑟𝑘 = 𝑠𝑖𝑘 + 𝑛𝑘
. 0
Similarly from branches 1, 2 ..N we will get scalar outputs as 𝑟1 ,
∅𝑁.(𝑡) 𝑟2 ,.. 𝑟𝑁 . Hence at the output of correlator demodulator we get
.
𝑇
𝑟𝑁 𝑇
𝑑𝑡 𝐫 = 𝑟1 𝑟2 … . . 𝑟𝑁 𝑇

𝑠𝑖𝑘 = 𝑠𝑖 𝜑𝑘 𝑑𝑡
0 Sampling at • Hence this is called a vector
0
𝑡=𝑇 receiver.
• Guess why you sample the output of
integrator

Dr D Adhikari, School of Electrical Engineering

CORRELATION DEMODULATOR • Note that the signal 𝑠𝑖 𝑡 was represented by a vector in the
signal space as 𝐬𝑖 with components 𝑠𝑖𝑘 for k = 1, 2, .. N.
∅1 (𝑡) • The values of 𝑠𝑖𝑘 depends on which of the M signals was
𝑇 transmitted.
𝑟1 • Also note that the signals {𝑠𝑖 𝑡 }are deterministic while the
𝑑𝑡
. 0 noise 𝑛 𝑡 is random having a Gaussian pdf.
• Hence the correlator outputs {𝑟𝑘 } will be Gaussian random
.
∅𝑘 (𝑡)
variables, with a conditional probability that the ith signal
𝑟 𝑡
. 𝑇
𝑟𝑘
was transmitted.
• This implies that the random variable 𝑟𝑘 has the following
𝑑𝑡
. 0
statistical parameters:
(a) Mean:𝐸 𝑟𝑘 = 𝐸 𝑠𝑖𝑘 + 𝑛𝑘 = 𝐸 𝑠𝑖𝑘 + 𝐸 𝑛𝑘 = 𝑠𝑖𝑘 + 0 = 𝑠𝑖𝑘
∅𝑁.(𝑡)
.
𝑇
𝑟𝑁 1
(b) Variance: 𝜎𝑟2 = 𝜎𝑛2 = 𝑁0
𝑑𝑡 2
0 Sampling at 1 𝑟𝑘 −𝑠𝑖𝑘 2
𝑡=𝑇 (c) Cond. pdf: 𝑝 𝑟𝑘 𝑠𝑖𝑘 = exp −
𝜋𝑁0 𝑁0

Dr D Adhikari, School of Electrical Engineering

CORRELATION DEMODULATOR Noting that all the correlator outputs {𝑟𝑘 } , k = 1, 2, .. N , are
statistically independent, we may write the joint conditional pdf as the
∅1 (𝑡) product of all the individual conditional probabilities:
𝑇
𝑟1 𝑝 𝑟 𝑠𝑖 = 𝑝 𝑟1 𝑠𝑖1 .. 𝑝 𝑟𝑘 𝑠𝑖𝑘 …𝑝 𝑟𝑁 𝑠𝑖𝑁 = 𝑁
𝑘=1 𝑝 𝑟𝑘 𝑠𝑖𝑘
𝑑𝑡
. 0 1 𝑟1 −𝑠𝑖1 2 1 𝑟𝑁 −𝑠𝑖𝑁 2
= exp − 𝑁0
X…X exp − 𝑁0
𝜋𝑁0 𝜋𝑁0
.
∅𝑘 (𝑡)
𝑟 𝑡 𝑁
. 𝑇
𝑟𝑘 =
1
𝜋𝑁0
exp −
𝑟1 −𝑠𝑖1 2
𝑁0
….. exp −
𝑟𝑁 −𝑠𝑖𝑁 2
𝑁0
𝑑𝑡
. 0
=
𝑁 𝑁
1 𝑟1 −𝑠𝑖1 2 +⋯+ 𝑟𝑁 −𝑠𝑖𝑁 2 1 𝑁 𝑟𝑘 −𝑠𝑖𝑘 2
exp − exp −
∅𝑁.(𝑡)
= 𝑘=1
𝜋𝑁0 𝑁0 𝜋𝑁0 𝑁0

.
𝑇
𝑟𝑁 Thus the joint conditional probability of having received the signal
𝑑𝑡 𝑟 𝑡 given that 𝑠𝑖 𝑡 was transmitted (also called the likelihood
0 Sampling at function) is
𝑁
𝑡 = 𝑇𝑠 1 1
𝑝 𝑟 𝑠𝑖 = exp − 𝑟 − 𝑠𝑖 2
𝜋𝑁0 𝑁0

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR

• At the output of the correlator demodulator we have obtained scalar voltages 𝑟𝑘 , k = 1, 2, .. N from each
𝑇
correlator branch which is represented as 𝐫 = 𝑟1 𝑟2 … . . 𝑟𝑁 (same output is obtained from a matched
filter demodulator)
• In other words we have obtained all the basis coefficients 𝑠𝑖𝑘 , k = 1, 2, .. N , for the transmitted signal 𝑠𝑖 𝑡 .
However, these coefficients are corrupted by noise 𝑛𝑘 , and hence there are chances of these 𝑠𝑖𝑘 to take
erroneous values, resulting in the incorrect detection of 𝑠𝑖 𝑡 [since 𝑠𝑖 𝑡 = 𝑠𝑖1 𝜑1 𝑡 + ⋯ + 𝑠𝑖𝑁 𝜑𝑁 𝑡 ].

Objective:
To design a signal detector that makes a decision on the transmitted signal in each signal interval (𝟎 ≤
𝒕 ≤ 𝑻 ), based on the observation of the vector 𝐫 = 𝒓𝟏 𝒓𝟐 … . . 𝒓𝑵 𝑻 , such that the probability of correct
decision is maximized (equivalently minimizing the probability of error).

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR
• Since in our hand we only have the measured vector 𝐫 = 𝑟1 𝑟2 … . . 𝑟𝑁 𝑇 , our decision rule required to be formulated for
the detector, should be conditioned on 𝐫.
• We define the posterior probability as : 𝑷( 𝒔𝒊𝒈𝒏𝒂𝒍 𝒔𝒊 𝒘𝒂𝒔 𝒕𝒓𝒂𝒏𝒔𝒎𝒊𝒕𝒕𝒆𝒅 𝒓 = 𝑃 𝑠𝑖 𝑟 , 𝑖 = 1,2, … 𝑀 .
• For maximizing the probability of correct decision, we need to maximize 𝑃 𝑠𝑖 𝑟 . This decision criterion is called the
Maximum a posteriori probability (MAP) criterion.
𝑝 𝑟 𝑠𝑖 𝑃 𝑠𝑖
𝑃 𝑠𝑖 𝑟 = (By Bayes rule)
𝑝 𝑟

where 𝑝 𝑟 𝑠𝑖 = Prior conditional probability of receiving signal vector 𝒓 given that signal 𝒔𝒊 was transmitted.
𝑃 𝑠𝑖 = Apriori probability of the ith signal being transmitted from the set of M signals (=1/M if equiprobable)
𝑀
𝑝 𝑟 = 𝑖=1 𝑝 𝑟 𝑠𝑖 𝑃 𝑠𝑖
• Note that 𝑃 𝑠𝑖 is only a number and does not affect our calculation and 𝑝 𝑟 is independent of which signal is being
transmitted.
• Hence maximizing 𝑃 𝑠𝑖 𝑟 boils down to maximizing 𝑝 𝑟 𝑠𝑖 , i.e., maximizing the likelihood function.

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR
𝑠𝑖 Channel 𝑟
𝑝 𝑟 𝑠𝑖

• As seen in the previous slide, maximizing 𝑃 𝑠𝑖 𝑟 boils down to maximizing 𝑝 𝑟 𝑠𝑖 , i.e., maximizing the
likelihood function. In other words, to compute 𝑃 𝑠𝑖 𝑟 we need to know 𝑝 𝑟 𝑠𝑖 for 𝑖 = 1,2, … 𝑀 .
• The decision criterion based on maximum 𝑝 𝑟 𝑠𝑖 over M signals is called Maximum-likelihood (ML) criterion
• For ease of mathematical understanding we consider a binary signal which we can generalized later for an M-ary
signal.
• Therefore transmitted signal in the symbol interval 0 ≤ 𝑡 ≤ 𝑇 will be (i =1, 2, i.e., M = 2)

𝑠1 𝑡 , 0 ≤ 𝑡 ≤ 𝑇, for a binary 1
𝑠𝑖 𝑡 =
𝑠2 𝑡 , 0 ≤ 𝑡 ≤ 𝑇, for a binary 0

• The received signal at the input of (demodulator) receiver is 𝑟 𝑡 = 𝑠𝑖 𝑡 + 𝑛 𝑡

• As discussed earlier, sampled values at the output of demodulator 𝑟 = 𝑠𝑖𝑘 + 𝑛𝑘 = 𝑎𝑖 + 𝑛𝑘 , 𝑖 = 1,2

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Maximum Likelihood Detector)
• Figure shows the conditional pdfs when we receive 𝑟 for having
transmitted a ‘1’(𝑝 𝑟 𝑠1 ) or having transmitted a ‘0’ (𝑝 𝑟 𝑠0 ), which
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 can be mathematically expressed as

2
1 𝑟 − 𝑎1
𝑝 𝑟 𝑠1 = exp − 𝑙𝑖𝑘𝑒𝑙𝑖ℎ𝑜𝑜𝑑 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑜𝑓𝑠1
𝐼1 𝜋𝑁0 𝑁0

1 𝑟 − 𝑎0 2
𝑝 𝑟 𝑠0 = exp − 𝑙𝑖𝑘𝑒𝑙𝑖ℎ𝑜𝑜𝑑 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑜𝑓𝑠0
𝜋𝑁0 𝑁0
𝐼2

• Here in the figure, 𝛾 represents the threshold voltage value, which

𝑎0 𝛾 𝑎1 𝑟 𝑇𝑠
signifies that if 𝑟 𝑇𝑠 > 𝛾, then 𝑠1 (i.e., 1) was transmitted and vice-
versa.
𝑟𝑎 𝑇𝑠 • We define hypothesis for this decision function as

𝐻1, , 𝑟 𝑇𝑠 > 𝛾
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝐻2, , 𝑟 𝑇𝑠 < 𝛾

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Maximum Likelihood Detector)
𝐻1, , 𝑟 𝑇𝑠 > 𝛾
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 𝐻2, , 𝑟 𝑇𝑠 < 𝛾

From our previous discussion we can write the above in terms of po as

𝐻 , 𝑃 𝑠1 𝑟 > 𝑃 𝑠0 𝑟
𝐼1 𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = 1
𝐻2 , 𝑃 𝑠1 𝑟 < 𝑃 𝑠0 𝑟
Using Bayes rule, we can write the above as
𝐻 , 𝑃 𝑠1 𝑝 𝑟 𝑠1 > 𝑃 𝑠0 𝑝 𝑟 𝑠0
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = 1
𝐻2 , 𝑃 𝑠1 𝑝 𝑟 𝑠1 < 𝑃 𝑠0 𝑝 𝑟 𝑠0
𝐼2
By re-arranging:

𝑟 𝑇𝑠 𝑝 𝑟 𝑠1 𝑃 𝑠0 Maximum
𝑎0 𝛾 𝑎1 𝐻1 , >
𝑝 𝑟 𝑠0 𝑃 𝑠1 A priori
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑟𝑎 𝑇𝑠 𝑝 𝑟 𝑠1 𝑃 𝑠0 (MAP)
𝐻2 , < Detector
𝑝 𝑟 𝑠0 𝑃 𝑠1

Where as discussed earlier 𝑃 𝑠0 and 𝑃 𝑠1 are the priori probabilities

(i.e probability of transmitting a 0 or 1)

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Maximum Likelihood Detector)
Note that if the priori are equiprobable, as is usually the case,
1
𝑃 𝑠0 = 𝑃 𝑠1 = 2 (probability of transmitting ‘0’ or ‘1’ is ½) the
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1
previous equation becomes:
𝑝 𝑟 𝑠1 Maximum
𝐻1 , >1
𝐼1 𝑝 𝑟 𝑠0 Likelihood
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑝 𝑟 𝑠1 (ML)
𝐻2 , <1
𝑝 𝑟 𝑠0 Detector

𝐼2 Substituting the likelihood values 𝑝 𝑟 𝑠1 and 𝑝 𝑟 𝑠0 we obtain a

likelihood ratio as
𝑟 𝑇𝑠 1 𝑟 − 𝑎1 2
𝑎0 𝛾 𝑎1 exp −
𝜋𝑁0 𝑁0 𝑟 − 𝑎0 2 − 𝑟 − 𝑎1 2
Λ 𝑟 = = exp
𝑟𝑎 𝑇𝑠 1 𝑟 − 𝑎0 2 𝑁0
exp − 𝑁0
𝜋𝑁0
2𝑟 𝑎1 − 𝑎0 𝑎12 − 𝑎02
= exp −
𝑁0 𝑁0

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Maximum Likelihood Detector)
Substituting the likelihood ratio into the MAP detector equation we obtain
2𝑟 𝑎1 − 𝑎0 𝑎12 − 𝑎02 𝑃 𝑠0
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 𝐻1 , exp − >
𝑁0 𝑁0 𝑃 𝑠1
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
2𝑟 𝑎1 − 𝑎0 𝑎12 − 𝑎02 𝑃 𝑠0
𝐻2 , exp − <
𝑁0 𝑁0 𝑃 𝑠1
𝐼1
Taking logarithm on both sides (Log Likelihood Ratio (LLR))

2𝑟 𝑎1 − 𝑎0 𝑎12 − 𝑎02 𝑃 𝑠0
𝐻1 , − > ln
𝐼2 𝑁0 𝑁0 𝑃 𝑠1
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
2𝑟 𝑎1 − 𝑎0 𝑎12 − 𝑎02 𝑃 𝑠0
𝐻2 , − > ln
𝑎0 𝛾 𝑎1 𝑟 𝑇𝑠 𝑁0 𝑁0 𝑃 𝑠1
𝑁0 𝑃 𝑠0 𝑎12 − 𝑎02
𝐻1 , 𝑟 > ln + 𝑁0
𝑟𝑎 𝑇𝑠 2 𝑎1 − 𝑎0 𝑃 𝑠1 2. 𝑁0 𝑎1 − 𝑎0
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑁0 𝑃 𝑠0 𝑎12 − 𝑎02
𝐻2 , 𝑟 < ln + 𝑁0
2 𝑎1 − 𝑎0 𝑃 𝑠1 2. 𝑁0 𝑎1 − 𝑎0

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Maximum Likelihood Detector)
𝑁0 𝑃 𝑠0 𝑎12 − 𝑎02
𝐻1 , 𝑟 > ln + 𝑁0
2 𝑎1 − 𝑎0 𝑃 𝑠1 2. 𝑁0 𝑎1 − 𝑎0
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑁0 𝑃 𝑠0 𝑎12 − 𝑎02
𝐻2 , 𝑟 < ln + 𝑁0
2 𝑎1 − 𝑎0 𝑃 𝑠1 2. 𝑁0 𝑎1 − 𝑎0

𝐼1 𝑁0 𝑃 𝑠0 𝑎12 − 𝑎02
𝐻1 , 𝑟 > ln + =𝛾
2 𝑎1 − 𝑎0 𝑃 𝑠1 𝑁0
𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑁0 𝑃 𝑠0 𝑎12 − 𝑎02
𝐻2 , 𝑟 < ln + =𝛾
2 𝑎1 − 𝑎0 𝑃 𝑠1 𝑁0
𝐼2
𝛾 is known as the MAP threshold
𝑟 𝑇𝑠 1
𝑎0 𝛾 𝑎1 If we take equiprobable signals then 𝑃 𝑠0 = 𝑃 𝑠1 = 2 therefore above
equation becomes
𝑟𝑎 𝑇𝑠 𝛾0 is known as the ML 𝑎1 + 𝑎2
𝐻1 , 𝑟 > = 𝛾0
threshold 𝐷𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = 2
𝑎1 + 𝑎2
𝐻2 , 𝑟 < = 𝛾0
ML DETECTOR 2

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Maximum Likelihood Detector)
Example: Bipolar pulse waveforms, 𝑠𝑖 𝑡 , 𝑖 = 1,2 , of amplitude ± 1V are received in the presence of AWGN
that has a variance of 0.1 V2. Determine the optimum detection threshold 𝛾that a MAP detector would select, if
the a priori probabilities are (a) P(𝑠1 ) = 0.5 (b) P(𝑠1 ) = 0.7.

1
Solution: We have 𝑎1 = +1V and 𝑎0 = -1V, 𝜎𝑛2 = 𝑁0 = 0.1V2 , i.e, 𝑁0 = 0.2V2
2

𝑁0 𝑃 𝑠0 𝑎12 −𝑎02
Threshold of MAP detector is 𝛾= ln +
2 𝑎1 −𝑎0 𝑃 𝑠1 𝑁0

0.2 𝑃 𝑠0 0.1 𝑃 𝑠0
= ln +0 = ln
2(1− −1 ) 𝑃 𝑠1 2 𝑃 𝑠1
For (a) P(𝑠1 ) = 0.5; P(𝑠0 ) = 1-0.5=0.5; 𝛾 = 0 V
𝑃 𝑠0 0.3
(b) P(𝑠1 ) = 0.7; P(𝑠0 ) = 0.3 : 𝛾 = 0.05 ln =0.05ln =-0.04 V
𝑃 𝑠1 0.7

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)

• Based on the previous computations MAP detector or ML detector has decided in favor of a symbol, say ‘0’
or ‘1’.
• However, due to presence of AWGN, we cannot say with certainty that the decision taken was correct.
• We introduce a figure of merit called the probability of error with the help of which we can compute a
statistical average of the error performance.
• In this discussion we consider the case of binary signal to continue with our binary ML detector.
• We will denote 𝑃 0 𝑠1 = Probability that a ‘0’ has been detected when a ‘1’ was sent
𝑃 1 𝑠0 = Probability that a ‘1’ has been detected when a ‘0’ was sent

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)
There are two ways by which we can commit a decision error
i. 𝑠1 𝑡 is sent but the received signal 𝑟 𝑡 is measured below the
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 threshold 𝛾.
ii. 𝑠0 𝑡 is sent but the received signal 𝑟 𝑡 is measured above the
threshold 𝛾.
𝐼1 Situation (i) (probabilities) can be expressed mathematically as
𝛾0
𝑃 𝑒 𝑠1 = 𝑃 𝐻2 𝑠1 = −∞
𝑝 𝑟 𝑠1 𝑑𝑟 (shown shaded in fig.)

𝐼2 Situation (ii) (probabilities) can be expressed mathematically as

∞
𝑎0 𝛾 𝑎1 𝑟 𝑇𝑠 𝑃 𝑒 𝑠0 = 𝑃 𝐻1 𝑠0 = 𝑝 𝑟 𝑠0 𝑑𝑟
𝛾0
𝑟𝑎 𝑇𝑠 The probability of error (Pbe) is the sum of the above two probabilities:
2 2

𝑃𝑏𝑒 = 𝑃 𝑒, 𝑠𝑖 = 𝑃 𝑒 𝑠𝑖 𝑃 𝑠𝑖 = 𝑃 0 𝑠1 𝑃 𝑠1 + 𝑃 1 𝑠0 𝑃 𝑠1
𝑖=1 𝑖=1

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)
• For equiprobable cases, 𝑃 𝑠1 = 𝑃 𝑠0 = ½ we have from the
previous equation
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 𝑃𝑏𝑒 = 𝑃 0 𝑠1 𝑃 𝑠1 + 𝑃 1 𝑠0 𝑃 𝑠1
1 1
= 𝑃 0 𝑠1 + 𝑃 1 𝑠0
2 2
• Noting from figure that 𝑝 𝑟 𝑠0 and 𝑝 𝑟 𝑠1 are symmetrical, we
𝐼1 may write from above
𝑃𝑏𝑒 = 𝑃 0 𝑠1 = 𝑃 1 𝑠0

• Hence for binary detection, the probability of a bit error is

𝐼2 numerically equal to the area under the ‘tail’ of either of the
likelihood functions 𝑝 𝑟 𝑠0 or 𝑝 𝑟 𝑠1 , falling on the incorrect side
of the threshold (shaded region for 𝑝 𝑟 𝑠1 )
𝑎0 𝛾 𝑎1 𝑟 𝑇𝑠
• Keeping into consideration the standard error function, we compute
the following area:
𝑟𝑎 𝑇𝑠

• 𝑃𝑏𝑒 =

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)
• Keeping into consideration the standard error function, we compute
𝑝 𝑟 𝑠0 𝑝 𝑟 𝑠1 the following area (tail of 𝑝 𝑟 𝑠0 from threshold to infinity, which is
the detection error):

𝐼1 ∞ ∞ 2
1 𝑟 − 𝑎0
𝑃𝑏𝑒 = 𝑃 𝑟 𝑠0 𝑑𝑟 = exp − 𝑑𝑟
𝛾0
𝑎1 +𝑎0 𝜋𝑁0 𝑁0
2
𝑟−𝑎0
𝐼2 • Substituting 𝑢 = , then 2 𝑁0 𝑑𝑢 = 𝑑𝑟
2 𝑁0

𝑟 𝑇𝑠 𝑎1 +𝑎0 2
𝑎0 𝛾 𝑎1 • When 𝑟 = , 𝑢= 𝑎1 − 𝑎0 ; 𝑟 = ∞, 𝑢 = ∞
2 𝑁0

𝑟𝑎 𝑇𝑠
∞
1 𝑢2 2
𝑃𝑏𝑒 = exp − 𝑑𝑢 = 𝑄 𝑎 − 𝑎0
2
𝑎 −𝑎 2𝜋 2 𝑁0 1
𝑁0 1 0

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)
• Note that the Q-function is a decreasing function of x. Hence to
minimize 𝑃𝑏𝑒 we need to increase the argument of Q(x).

2
𝑃𝑏𝑒 = 𝑄 𝑎1 − 𝑎0
𝑁0

• Hence to minimize 𝑃𝑏𝑒 , we need to increase (𝑎1 − 𝑎0 ) at the out

put of correlator demodulator.

• If we consider 𝐸𝑑 to be the energy of the difference signal [𝑠1 𝑡 -

𝑠0 𝑡 ], then

𝑇𝑠 2 𝑑𝑡
• 𝐸𝑑 = 0
𝑠1 𝑡 −𝑠0 𝑡 = 𝑎1 − 𝑎0 2 𝑇𝑠

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)
SNR of the correlator demodulator is

𝑆 𝑎1 −𝑎0 2 𝑎1 −𝑎0 2 2𝐸𝑑

= = 1 =
𝑁 𝑇𝑠 𝜎𝑛2 𝑁 𝑁0
2 0

Therefore , probability of error is given as,

2 2𝐸𝑑
𝑃𝑏𝑒 = 𝑄 𝑎1 − 𝑎0 =𝑄 =𝑄 𝑆𝑁𝑅
𝑁0 𝑁0

Dr D Adhikari, School of Electrical Engineering

OPTIMAL DETECTOR (Error Probability)
Example: Signal vectors 𝒔1 and 𝒔0 are used to transmit information in a communication system. The received signal is
𝑁0
defined by 𝑟 𝑡 = 𝑠𝑖 𝑡 + 𝑛 𝑡 , 𝑖 = 0,1 and 𝑛 𝑡 is AWGN defined as 𝒩 0, . Calculate the average probability of
2

error when 𝑁0 =3T/4. For binary signaling 𝑑12 = 𝐸𝑏

- 3𝑇 3𝑇
𝝋1
𝒔0 𝒔1
𝑑12

Solution: The inter-symbol distance is 𝑑12= 2 3𝑇 = 𝐸𝑏 . Hence the probability of error will be given by :

2𝐸𝑑 2
𝑑12 14.6𝑇
𝑃𝑏𝑒 = 𝑄 =𝑄 =𝑄 =𝑄 28 = 𝑄 5.29 = 5.79 X 10-8
𝑁0 2𝑁0 3𝑡

Dr D Adhikari, School of Electrical Engineering

Assignment:

1. Submit the derivation of a Matched Filter Demodulator. (by 30 Mar20 by mail).

2. Study the Band-pass modulation techniques before I take the next webinar.

Thank You!!
Dr D Adhikari, School of Electrical Engineering
Last version available at www.eng.tau.ac.il/∼jo/teaching

Q function and error function

We first note that
Z ∞ Z ∞ r
−x2 √ − ax2
2 2π
e dx = π ; e dx =
−∞ −∞ a

For our needs in Digital Communication course, we define:

Z ∞
1 x2
e−
∆
Q(α)= √ 2 dx
2π α

The Q(·) function is monotonically decreasing. Some features:

1
Q(−∞) = 1 ; Q(0) = ; Q(∞) = 0 ; Q(−x) = 1 − Q(x)
2

Known bounds (valid for x > 0):

µ ¶
1 1 2 1 2
√ 1 − 2 e−x /2 < Q(x) < √ e−x /2
2πx x 2πx
1 −x2 /2
Q(x) ≤ e
2

Matlab does not have a build-in function for Q(·). Instead, we use its erf function:
Z α
2 2
e−x dx
∆
erf(α)= √
π 0

Note that erf function is defined over [0, ∞) only, and

erf(0) = 0 ; erf(∞) = 1

The relations between the two functions are

µ ¶ √
1 1 α
Q(α) = − erf √ ; erf(α) = 1 − 2Q( 2α)
2 2 2

If we have a normal variable X ∼ N (µ, σ 2 ), the probability that X > x is

µ ¶
x−µ
Pr{X > x} = Q
σ

Now, if we want to know the probability of X to be away from its expectation µ by at least a
(either to the left or to the right) we have:
µ ¶
a
Pr{X > µ + a} = Pr{X < µ − a} = Q
σ

The probability to be away from the center where we don’t matter in which direction is 2 · Q( σa ).

This version compiled on April 6, 2006

E&CE 411, Spring 2009, Table of Q Function 1

Table 1: Values of Q(x) for 0 ≤ x ≤ 9

x Q(x) x Q(x) x Q(x) x Q(x)
0.00 0.5 2.30 0.010724 4.55 2.6823×10−6 6.80 5.231×10−12
0.05 0.48006 2.35 0.0093867 4.60 2.1125×10−6 6.85 3.6925×10−12
0.10 0.46017 2.40 0.0081975 4.65 1.6597×10−6 6.90 2.6001×10−12
0.15 0.44038 2.45 0.0071428 4.70 1.3008×10−6 6.95 1.8264×10−12
0.20 0.42074 2.50 0.0062097 4.75 1.0171×10−6 7.00 1.2798×10−12
0.25 0.40129 2.55 0.0053861 4.80 7.9333×10−7 7.05 8.9459×10−13
0.30 0.38209 2.60 0.0046612 4.85 6.1731×10−7 7.10 6.2378×10−13
0.35 0.36317 2.65 0.0040246 4.90 4.7918×10−7 7.15 4.3389×10−13
0.40 0.34458 2.70 0.003467 4.95 3.7107×10−7 7.20 3.0106×10−13
0.45 0.32636 2.75 0.0029798 5.00 2.8665×10−7 7.25 2.0839×10−13
0.50 0.30854 2.80 0.0025551 5.05 2.2091×10−7 7.30 1.4388×10−13
0.55 0.29116 2.85 0.002186 5.10 1.6983×10−7 7.35 9.9103×10−14
0.60 0.27425 2.90 0.0018658 5.15 1.3024×10−7 7.40 6.8092×10−14
0.65 0.25785 2.95 0.0015889 5.20 9.9644×10−8 7.45 4.667×10−14
0.70 0.24196 3.00 0.0013499 5.25 7.605×10−8 7.50 3.1909×10−14
0.75 0.22663 3.05 0.0011442 5.30 5.7901×10−8 7.55 2.1763×10−14
0.80 0.21186 3.10 0.0009676 5.35 4.3977×10−8 7.60 1.4807×10−14
0.85 0.19766 3.15 0.00081635 5.40 3.332×10−8 7.65 1.0049×10−14
0.90 0.18406 3.20 0.00068714 5.45 2.5185×10−8 7.70 6.8033×10−15
0.95 0.17106 3.25 0.00057703 5.50 1.899×10−8 7.75 4.5946×10−15
1.00 0.15866 3.30 0.00048342 5.55 1.4283×10−8 7.80 3.0954×10−15
1.05 0.14686 3.35 0.00040406 5.60 1.0718×10−8 7.85 2.0802×10−15
1.10 0.13567 3.40 0.00033693 5.65 8.0224×10−9 7.90 1.3945×10−15
1.15 0.12507 3.45 0.00028029 5.70 5.9904×10−9 7.95 9.3256×10−16
1.20 0.11507 3.50 0.00023263 5.75 4.4622×10−9 8.00 6.221×10−16
1.25 0.10565 3.55 0.00019262 5.80 3.3157×10−9 8.05 4.1397×10−16
1.30 0.0968 3.60 0.00015911 5.85 2.4579×10−9 8.10 2.748×10−16
1.35 0.088508 3.65 0.00013112 5.90 1.8175×10−9 8.15 1.8196×10−16
1.40 0.080757 3.70 0.0001078 5.95 1.3407×10−9 8.20 1.2019×10−16
1.45 0.073529 3.75 8.8417×10−5 6.00 9.8659×10−10 8.25 7.9197×10−17
1.50 0.066807 3.80 7.2348×10−5 6.05 7.2423×10−10 8.30 5.2056×10−17
1.55 0.060571 3.85 5.9059×10−5 6.10 5.3034×10−10 8.35 3.4131×10−17
1.60 0.054799 3.90 4.8096×10−5 6.15 3.8741×10−10 8.40 2.2324×10−17
1.65 0.049471 3.95 3.9076×10−5 6.20 2.8232×10−10 8.45 1.4565×10−17
1.70 0.044565 4.00 3.1671×10−5 6.25 2.0523×10−10 8.50 9.4795×10−18
1.75 0.040059 4.05 2.5609×10−5 6.30 1.4882×10−10 8.55 6.1544×10−18
1.80 0.03593 4.10 2.0658×10−5 6.35 1.0766×10−10 8.60 3.9858×10−18
1.85 0.032157 4.15 1.6624×10−5 6.40 7.7688×10−11 8.65 2.575×10−18
1.90 0.028717 4.20 1.3346×10−5 6.45 5.5925×10−11 8.70 1.6594×10−18
1.95 0.025588 4.25 1.0689×10−5 6.50 4.016×10−11 8.75 1.0668×10−18
2.00 0.02275 4.30 8.5399×10−6 6.55 2.8769×10−11 8.80 6.8408×10−19
2.05 0.020182 4.35 6.8069×10−6 6.60 2.0558×10−11 8.85 4.376×10−19
2.10 0.017864 4.40 5.4125×10−6 6.65 1.4655×10−11 8.90 2.7923×10−19
2.15 0.015778 4.45 4.2935×10−6 6.70 1.0421×10−11 8.95 1.7774×10−19
2.20 0.013903 4.50 3.3977×10−6 6.75 7.3923×10−12 9.00 1.1286×10−19
2.25 0.012224