Digital Band-pass Modulation

PROF. MICHAEL TSAI

2011/11/10
Band-pass Signal 𝑔 𝑡
𝑎 𝑡
Representation
2𝜋𝑓𝑐 𝑡 + 𝜙 𝑡
• General form:

𝒈 𝒕 = 𝒂 𝒕 𝒄𝒐𝒔 𝟐𝝅𝒇𝒄 𝒕 + 𝝓 𝒕

Envelope Phase

• Envelope is always non-negative, or we can switch

the phase by 180 degree
• This is called the canonical representation of a band-
pass signal
Band-pass Signal
Representation
• 𝒈 𝒕 = 𝒂 𝒕 𝒄𝒐𝒔 𝟐𝝅𝒇𝒄 𝒕 + 𝝓 𝒕 can be re-arranged into
• 𝒈 𝒕 = 𝒈𝑰 𝒕 𝒄𝒐𝒔 𝟐𝝅𝒇𝒄 𝒕 − 𝒈𝑸 𝒕 𝒔𝒊𝒏 𝟐𝝅𝒇𝒄 𝒕
• 𝒈𝑰 𝒕 = 𝒂 𝒕 𝒄𝒐𝒔 𝝓 𝒕 and 𝒈𝑸 𝒕 = 𝒂 𝒕 𝒔𝒊𝒏 𝝓 𝒕
• 𝒈𝑰 𝒕 and 𝒈𝑸 𝒕 are called inphase and quadrature
components of the signal g(t), respectively
𝒈𝑸 𝒕
• Then 𝒂 𝒕 = 𝒈𝟐𝑰 𝒕 + 𝒈𝟐𝑸 𝒕 and 𝝓 𝒕 = 𝒕𝒂𝒏−𝟏
𝒈𝑰 𝒕
Band-pass Signal 𝑔 𝑡
𝑎 𝑡
Representation
𝜙 𝑡
• We can also represent g(t) as

𝒈 𝒕 = 𝑹𝒆 𝒈 𝒕 𝒆𝒙𝒑 𝒋𝟐𝝅𝒇𝒄 𝒕
• 𝒈 𝒕 = 𝒈𝑰 𝒕 + 𝒋𝒈𝑸 𝒕
• 𝒈 𝒕 is called the complex envelope of the band-pass
signal.
• This is to remove the annoying 𝒆𝒙𝒑 𝒋𝟐𝝅𝒇𝒄 𝒕 in the
analysis.
Sinusoidal Functions’
Fourier Transform
• Complex exponential function
• 𝐹 exp 𝑗2𝜋𝑓𝑐 𝑡 = 𝛿(𝑓 − 𝑓𝑐 ).
• Sinusoidal functions:
1
• cos 2𝜋𝑓𝑐 𝑡 = exp 𝑗2𝜋𝑓𝑐 𝑡 + exp −𝑗2𝜋𝑓𝑐 𝑡
2
1
• 𝐹 cos 2𝜋𝑓𝑐 𝑡 = 𝛿 𝑓 − 𝑓𝑐 + 𝛿 𝑓 + 𝑓𝑐
2
1
• sin 2𝜋𝑓𝑐 𝑡 = 2 exp 𝑗2𝜋𝑓𝑐 𝑡 − exp −𝑗2𝜋𝑓𝑐 𝑡
1
• 𝐹 sin 2𝜋𝑓𝑐 𝑡 = 2 𝛿 𝑓 − 𝑓𝑐 − 𝛿 𝑓 + 𝑓𝑐

𝐺(𝑓) 𝐺(𝑓) 𝐺(𝑓)

𝑓𝑐 𝑓𝑐 𝑓𝑐
−𝑓𝑐 −𝑓𝑐
f f f

5
 𝑔 𝑡 = 𝑔𝐼 𝑡 cos 2𝜋𝑓𝑐 𝑡 + 𝑔𝑄 𝑡 sin 2𝜋𝑓𝑐 𝑡

Band-pass Signal
Transmitter
𝑔𝐼 (𝑡)
×

Maps each bit into

𝑔𝐼 𝑡 and 𝑔𝑄 𝑡
∼ cos⁡(2𝜋𝑓𝑐 𝑡)

90 +
Message Signal Band-pass
Source Encoder
degree
shift
Σ Signal g(t)
+

sin⁡(2𝜋𝑓𝑐 𝑡)

𝑔𝑄 (𝑡)
×
Assumption
• The channel is linear: flat-fading channel.
• 𝐵𝑐 > 𝐵𝑠
• Negligible distortion to 𝑔(𝑡)
• The received signal s(t) is perturbed by AWGN
𝑁0
• noise w(t) ~𝑁 0, 2
𝑁0
• 2
is the PSD of the noise and also its variance (since it’s white)
 AWGN Channel

Channel path loss

or attenuation Add  “additive”

Band-pass Channel + Received Signal plus Noise

Signal g(t) 𝐴𝑐 Σ 𝑠 𝑡 + 𝑤(𝑡)
+

White Gaussian Noise 𝑤 𝑡

𝑥 𝑡 = 𝑠 𝑡 + 𝑤 𝑡 = 𝐴𝑐 𝑔 𝑡 + 𝑤(𝑡)
 Band-pass Signal
Receiver
Mixer
Low-pass 1
× 𝐴 𝑔 𝑡 + 𝑛𝐼 𝑡
Filter 2 𝑐 𝐼
Filters out out-
of-band signals
and noises ∼ cos⁡(2𝜋𝑓𝑐 𝑡)

Received 90
Signal Band-pass degree Signal Message
plus Filter shift Detector Sink
Noise
𝑥 𝑡 = 𝑠 𝑡 + 𝑛(𝑡)
sin⁡(2𝜋𝑓𝑐 𝑡)
1
𝐴 𝑔 𝑡 + 𝑛𝑄 𝑡
2 𝑐 𝑄
𝑔𝑄 (𝑡) Low-pass
×
Filter
 Band-pass
Filter

Band-pass Filter
• The band-pass filter at the frontend filters out out-of-band
signals and noises
1. Signal s(t) is within the band  not affected
2. White noise w(t) becomes narrowband noise n(t)
• Much smaller since now we only include noises within the band
• Still “white over the bandwidth of the signal”
3. Other signal (out-of-band) is filtered out

Other signals

s(t)

Noise

𝑓
𝐵 𝑓𝑐 𝐵
𝑓𝑐 − 𝑓𝑐 +
2 2
 ×

Up-conversion (TX) ∼ cos⁡(2𝜋𝑓𝑐 𝑡)

In time domain

𝐴 𝑡 exp 𝑗𝜃 𝑡 × Ac cos 2𝜋𝑓𝑐 𝑡

In frequency domain Convolution

∗

𝑓 𝑓
𝑓 −𝑓𝑐 𝑓𝑐 −𝑓𝑐 𝑓𝑐
 × Low-pass
Filter

Down-conversion (RX) ∼ cos⁡(2𝜋𝑓𝑐 𝑡)

In time domain

s t A′c cos⁡(2𝜋𝑓𝑐 𝑡 + 𝜙) × Ac cos 2𝜋𝑓𝑐 𝑡

Low-pass
In frequency domain Convolution Filter
∗

𝑓 𝑓 𝑓
−𝑓𝑐 𝑓𝑐 −𝑓𝑐 𝑓𝑐 −2𝑓𝑐 2𝑓𝑐
 Signal
Detector

Signal Detector
• The signal detector:
• Observes complex representation of the received signal,
𝒈𝑰 𝒕 + 𝒏𝑰 𝒕 + 𝒋[𝒈𝑸 𝒕 + 𝒏𝑸 𝒕 ],
• For a duration of T seconds (symbol/bit period)
• And the make its best estimate of the corresponding transmitted
signal 𝒈𝑰 𝒕 + 𝒋𝒈𝑸 𝒕
• 𝒈𝑰 𝒕 + 𝒋𝒈𝑸 𝒕  bit stream
Time synchronization
• To simplify, we assume we have time synchronization
between the TX and the RX
Where does each symbol start and end?

• Symbol boundary needs to be same for TX and RX

• In practice, a timing recovery circuit is required
Coherent & non-coherent
• Sometimes, the receiver is phase-locked to the
transmitter
• That means, the ∼ in TX and in RX generate
𝒄𝒐𝒔(𝟐𝝅𝒇𝒄 𝒕) with no phase difference.
• RX looks at the received signal to lock onto TX’s carrier
• When that happens, we say
• The receiver is a coherent receiver, carrying out coherent detection
• Otherwise, we say
• The receiver is a non-coherent receiver, carrying out non-coherent
detection
Basic forms of digital modulation
Amplitude Shift Keying

Frequency Shift Keying

Phase Shift Keying

Keying == Switching
(Binary) Amplitude Shift
Keying (BASK)

• Fixed Amplitude/fixed frequency for a duration of 𝑻𝒃 to

represent “1”
• No transmission to represent “0”
• Or, more formally,
• 𝑠1 𝑡 = 𝐴𝑐 cos(2𝜋𝑓𝑐 𝑡)
for a duration of 𝑇𝑏
• 𝑠0 𝑡 = 0
(Binary) Phase Shift
Keying (BPSK)

• Same amplitude, same frequency

• Send the original carrier to represent “1”
• Send an inverted carrier (phase difference 180 degrees) to
represent “0”
• Or, more formally,
• 𝑠1 𝑡 = 𝐴𝑐 cos(2𝜋𝑓𝑐 𝑡)
• 𝑠0 𝑡 = 𝐴𝑐 cos 2𝜋𝑓𝑐 𝑡 + 𝜋 = −𝐴𝑐 cos(2𝜋𝑓0 𝑡)
(Binary) Phase Shift
Keying (BPSK)
(Binary) Frequency Shift
Keying (BFSK)
𝑇𝑏

• Same amplitude
• Send a carrier at 𝒇𝟏 to represent “1”
• Send a carrier at 𝒇𝟎 to represent “0”
• Or, more formally,
• 𝑠1 𝑡 = 𝐴𝑐 cos(2𝜋𝑓1 𝑡)
for a duration of 𝑇𝑏
• 𝑠0 𝑡 = 𝐴𝑐 cos(2𝜋𝑓0 𝑡)
(Binary) Frequency
Shift Keying (BFSK)
• Usually we have 𝒇𝟏 = 𝒇𝒄 + 𝚫𝐟, 𝐟𝟎 = 𝐟𝐜 − 𝚫𝐟
• 𝑠1 𝑡 = 𝐴𝑐 cos[2𝜋 𝑓𝑐 + Δ𝑓 𝑡]
• 𝑠0 𝑡 = 𝐴𝑐 cos[2𝜋 𝑓𝑐 − Δ𝑓 𝑡]
• Then,
• 𝑠1 𝑡 = 𝑅𝑒 𝐴𝑐 exp 𝑗2𝜋 𝑓𝑐 + Δf 𝑡 = 𝑔 𝑡 exp 𝑗2𝜋𝑓𝑐 𝑡
• 𝑠0 𝑡 = 𝑅𝑒 𝐴𝑐 exp 𝑗2𝜋 𝑓𝑐 − Δf 𝑡 = 𝑔 𝑡 exp 𝑗2𝜋𝑓𝑐 𝑡
• So, Q
• For “1”, 𝑔 𝑡 = 𝑔𝐼 𝑡 + 𝑗𝑔𝑄 𝑡 = 𝐴𝑐 exp⁡[−𝑗2𝜋Δ𝑓𝑡]
• For “0”,𝑔 𝑡 = 𝑔𝐼 𝑡 + 𝑗𝑔𝑄 𝑡 = 𝐴𝑐 exp⁡[+𝑗2𝜋Δ𝑓𝑡]

I
Coherent Detection of
FSK and PSK signals
𝟏
• Since 𝒇𝒄 is large compared to (symbol rate, or bit
𝑻𝒃
rate), we can say that the same signal energy 𝑬𝒃 is
transmitted in a bit interval 𝑻𝒃 :

𝑇𝑏 𝑇𝑏
𝐸𝑏 = 𝑠02 𝑡 𝑑𝑡 = 𝑠12 𝑡 𝑑𝑡
0 0

𝐴2𝑐 𝑇𝑏
=
2
 Two-path correlation
receiver (general case)

Correlator: see how similar 𝑥 𝑡 and 𝑠1 (𝑡) are

𝑠1 (𝑡)
x(t): received signal 𝑇𝑏
× 𝑑𝑡 Choose 1 if 𝑙 > 0
0 +
𝑙
𝑥(𝑡) Detection
Σ device
𝑇𝑏 −
× 𝑑𝑡 Otherwise, choose 0
0

𝑠0 (𝑡)

Correlator: see how similar 𝑥 𝑡 and 𝑠0 (𝑡) are

Coherent Detection
𝑵𝟎
• 𝒘(𝒕): AWGN, 𝑵 𝟎,
𝟐
• 𝑯𝟎 : 𝒙 𝒕 = 𝒔𝟎 𝒕 + 𝒘(𝒕)
• 𝑯𝟏 : 𝒙 𝒕 = 𝒔𝟏 𝒕 + 𝒘(𝒕)
• Receiver output:
𝑇𝑏
𝑙= 𝑥 𝑡 𝑠1 𝑡 − 𝑠0 𝑡 𝑑𝑡
0
• Decision level: 0
• If 𝑙 is larger than 1, than 𝑥(𝑡) is “more similar” to 𝑠1 (𝑡)
• If 𝑙 is smaller than 1, than 𝑥(𝑡) is “more similar” to 𝑠0 (𝑡)
 𝑇𝑏
𝑙= 𝑥 𝑡 𝑠1 𝑡 − 𝑠0 𝑡 𝑑𝑡
0

Coherent Detection
• 𝑯𝟏 :
𝑇𝑏 𝑇𝑏
𝑙= 𝑠1 𝑡 𝑠1 𝑡 − 𝑠0 𝑡 𝑑𝑡 − 𝑤 𝑡 𝑠1 𝑡 − 𝑠0 𝑡 𝑑𝑡
0 0

• Since the noise w(t) is zero-mean, L: the random variable whose value is 𝑙
𝑇𝑏
𝐸 𝐿 𝐻1 = 𝑠1 𝑡 𝑠1 𝑡 − 𝑠0 𝑡 𝑑𝑡 = 𝐸𝑏 (1 − 𝜌)
0

• 𝝆: the correlation coefficient of the signals 𝒔𝟎 (𝒕) and

𝒔𝟏 𝒕
𝑇𝑏 𝑇𝑏
0
𝑠0 𝑡 𝑠1 𝑡 𝑑𝑡 1
𝜌= 1 = 𝑠0 𝑡 𝑠1 𝑡 𝑑𝑡⁡
𝑇𝑏 2 𝑇𝑏 2 2
𝐸𝑏 0
𝑠0 𝑡 𝑑𝑡 𝑠1 𝑡 𝑑𝑡
0 0 0≤𝜌≤1
Coherent Detection
• Similarly,
𝐸 𝐿 𝐻0 = −𝐸𝑏 1 − 𝜌

• L’s variance is the same for 𝑯𝟏 and 𝑯𝟎 . Since 𝒔𝟏 (𝒕) and

𝒔𝟎 (𝒕) is deterministic given the transmitted bit, we have

𝑉𝑎𝑟 𝐿 = E L − E L 2

𝑇𝑏 𝑇𝑏
=E 𝑤 𝑡 𝑤 𝑢 𝑠1 𝑡 − 𝑠0 𝑡 𝑠1 𝑢 − 𝑠0 𝑢 𝑑𝑡 𝑑𝑢
0 0

𝑇𝑏 𝑇𝑏
= 𝑬𝒘 𝒕𝒘 𝒖 𝑠1 𝑡 − 𝑠0 𝑡 𝑠1 𝑢 − 𝑠0 𝑢 𝑑𝑡 𝑑𝑢
0 0
𝑇𝑏 𝑇𝑏
= 𝛿(𝑡 − 𝑢) 𝑠1 𝑡 − 𝑠0 𝑡 𝑠1 𝑢 − 𝑠0 𝑢 𝑑𝑡 𝑑𝑢
0 0
 𝑇𝑏 𝑇𝑏
= 𝛿(𝑡 − 𝑢) 𝑠1 𝑡 − 𝑠0 𝑡 𝑠1 𝑢 − 𝑠0 𝑢 𝑑𝑡 𝑑𝑢
0 0

𝑇𝑏
𝑁0 2 𝑑𝑡
= 𝑠 𝑡 − 𝑠0 𝑡 = 𝑁0 𝐸𝑏 (1 − 𝜌)
0 2 1

• Therefore, we know that L conditioned on 𝑯𝟎 is a Gaussian

distributed random variable: 𝑵 𝑬𝒃 𝟏 − 𝝆 , 𝑵𝟎 𝑬𝒃 𝟏 − 𝝆
Q Function
• Q function is defined over the CDF of Gaussian
distribution 𝑵(𝟎, 𝟏)
CDF of 𝑵(𝟎, 𝟏)
∞
1 𝑢2
𝑄 𝑥 = exp − 𝑑𝑢 = 1 − Φ(𝑥)
2𝜋 𝑥 2

f(u)
N(0,1)’s PDF

x u
Integration (Area under the curve) 
 Bit Error Rate
𝑳|𝑯𝟏 ~𝑵 𝑬𝒃 𝟏 − 𝝆 , 𝑵𝟎 𝑬𝒃 𝟏 − 𝝆

𝝇𝟐 ~𝑵𝟎 𝑬𝒃 𝟏 − 𝝆

0 𝑬𝒃 𝟏 − 𝝆
Integration (Area under the curve)
Divide u by 𝑁0 𝐸𝑏 1 − 𝜌
How to express this area with Q function?

𝑵(𝟎, 𝟏)
Shift left by 𝑬𝒃 𝟏 − 𝝆 𝑵 𝟎, 𝑵𝟎 𝑬𝒃 𝟏 − 𝝆

0
−𝑬𝒃 𝟏 − 𝝆
0
−𝑬𝒃 𝟏 − 𝝆 𝑵𝟎 𝑬 𝒃 𝟏 − 𝝆
Bit Error Rate

𝑵(𝟎, 𝟏) 𝑵(𝟎, 𝟏)

0 0
−𝑬𝒃 𝟏 − 𝝆 𝑬𝒃 𝟏 − 𝝆
𝑵𝟎 𝑬 𝒃 𝟏 − 𝝆 𝑵𝟎 𝑬𝒃 𝟏 − 𝝆

𝐸𝑏 1 − 𝜌
𝑃𝑒 = 𝑄
𝑁0
For BPSK, 𝜌 = −1 For BFSK, 𝜌 = 0

2𝐸𝑏 𝐸𝑏
𝑃𝑒 = 𝑄 𝑃𝑒 = 𝑄
𝑁0 𝑁0
Signal Space - BPSK

sin⁡(2𝜋𝑓𝑐 𝑡)
Quadrature

Noise 𝑠1 𝑡 = 𝐴𝑐 cos⁡(2𝜋𝑓𝑐 𝑡)
𝑠0 𝑡 = −𝐴𝑐 cos⁡(2𝜋𝑓𝑐 𝑡)
Bit 0 Bit 1
Inphase
Energy 𝐴
−𝐴𝑐 𝑐 cos⁡(2𝜋𝑓𝑐 𝑡)
Signal Space - QPSK
𝜋
𝑠11 𝑡 = 𝐴𝑐 cos 2𝜋𝑓𝑐 𝑡 +
sin⁡(2𝜋𝑓𝑐 𝑡) 4
Quadrature 3𝜋
Bit 01 Bit 11 𝑠01 𝑡 = 𝐴𝑐 cos 2𝜋𝑓𝑐 𝑡 +
𝐴𝑐 4
2 𝐴𝑐 5𝜋
𝑠00 𝑡 = 𝐴𝑐 cos 2𝜋𝑓𝑐 𝑡 +
4
−𝐴𝑐 𝐴𝑐
𝜙 7𝜋
2 2 Inphase 𝑠10 𝑡 = 𝐴𝑐 cos 2𝜋𝑓𝑐 𝑡 +
cos⁡(2𝜋𝑓𝑐 𝑡) 4

−𝐴𝑐
Bit 00 2 Bit 10
 Quadrature
M-ary Modulation
4-PSK

8-PSK Quadrature

Inphase
𝐴𝑐

𝜙 Inphase

Increasing M would increase the data rate

(given the same signal bandwidth)
M-ary Modulation

16-QAM Quadrature

Inphase
M-PAM BER versus SNR

𝐸𝑏
𝑆𝑁𝑅 =
𝜍𝑛2

1 3 × 𝑆𝑁𝑅
𝐵𝐸𝑅 ≤ 𝑃𝑒 ≤ 2 × 1 − ×𝑄
𝑀 𝑀2 − 1
M-QAM BER versus SNR

1 3 × 𝑆𝑁𝑅
𝐵𝐸𝑅 ≤ 𝑃𝑒 ≤ 4 × 1 − ×𝑄
𝑀 𝑀−1

1 3 × 𝑆𝑁𝑅
𝐵𝐸𝑅 ≤ 𝑃𝑒 ≤ 4 × 1 − ×𝑄
2𝑀 𝑀
31 × 32 − 1
M-PSK BER versus SNR

𝜋
𝐵𝐸𝑅 ≤ 𝑃𝑒 ≤ 2 × 𝑄 2 × 𝑆𝑁𝑅 × sin
𝑀