Signals and Systems: DR Tania Stathaki

Signals and Systems
Lecture 12
DR TANIA STATHAKI
READER (ASSOCIATE PROFFESOR) IN SIGNAL PROCESSING
IMPERIAL COLLEGE LONDON
Signal transmission through LTI systems
• We have seen previously that if 𝑥(𝑡) and 𝑦(𝑡) are input and output of a
LTI system with impulse response ℎ(𝑡), then:
𝑌(𝜔) = 𝐻 𝜔 𝑋(𝜔)
• We can, therefore, perform LTI system analysis with Fourier transform
in a way similar to that of Laplace transform.
• However, FT is more restrictive than Laplace transform because the
system must be stable and 𝑥(𝑡) must itself be Fourier transformable.
• Laplace transform can be used to analyse stable and unstable systems,
and applies to signals that grow exponentially.
• As already mentioned, if a system is stable, it can be shown that the
frequency response of the system 𝐻(𝑗𝜔) is just the Fourier transform of
ℎ(𝑡) (i.e., 𝐻(𝜔)):
𝐻 𝜔 = 𝐻(𝑠)
𝑠=𝑗𝜔
Distortionless transmission
• In certain types of systems we require the input to pass through the

system without distortion. For example:
 Signal transmission over a communication channel.
 Amplifying systems.
• Distortionless transmission implies that for a specific frequency band

the output is the same as the input apart from:
 A constant multiplicative factor.
 A delay.
• Therefore, if 𝑥(𝑡) is the input and 𝑦(𝑡) is the output, distortionless

transmission implies that:
𝑦 𝑡 = 𝐺0 𝑥(𝑡 − 𝑡𝑑 )
Distortionless transmission cont.
• Distortionless transmission of an input 𝑥(𝑡) implies that:
𝑦 𝑡 = 𝐺0 𝑥(𝑡 − 𝑡𝑑 )
• Taking the Fourier transform of the above yields:
𝑌 𝜔 = 𝐺0 𝑋(𝜔)𝑒 −𝑗𝜔𝑡𝑑
• Knowing that:
𝑌 𝜔 = 𝐻(𝜔)𝑋(𝜔)
we can write that the transfer function of a distortionless system is:
𝐻 𝜔 = 𝐺0 𝑒 −𝑗𝜔𝑡𝑑
 𝐻 𝜔 = 𝐺0 amplitude response must be a constant
 ∠𝐻 𝜔 = −𝜔𝑡𝑑 phase response must be a linear function of 𝜔 with
slope −𝑡𝑑 which passes through the origin
Group delay
• In order to assess phase linearity we can find the slope of ∠𝐻 𝜔 as a

function of frequency and see whether it is constant. We define:
𝑑
𝑡𝑔 𝜔 = − ∠𝐻 𝜔
𝑑𝜔
• 𝑡𝑔 𝜔 is called group delay or envelope delay.
• Note that a phase response given by ∠𝐻 𝜔 = 𝜙0 − 𝜔𝑡𝑑 also has a
constant group delay. From now on we can write 𝑡𝑑 = 𝑡𝑔 .
• Therefore, the condition for phase linearity by testing whether the
group delay is constant is more relaxed.
• Human ears are sensitive to amplitude distortion, but not phase
distortion.
• Human eyes are sensitive to phase distortion, but not so much to
amplitude distortion (recall the experiment where we have combined
the amplitude of one image and the phase of another).
Bandpass systems and group delay
• For lowpass systems, the phase must be linear over the band of interest
and also must pass through the origin.
 Recall that phase is an odd function. Therefore, if it doesn’t pass
through the origin, it will have a jump at the origin; this means that the
group delay will be a Dirac function.
 Infinite group delay means that the input takes infinite time to arrive at
the output, i.e., it doesn’t practically get through.
• For bandpass systems, the phase must be linear over the band of interest
but does not have to pass through the origin.
• Consider the following bandpass LTI system.
• The pass band is of width 2𝑊 centred at 𝜔𝑐 .

Bandpass systems and group delay cont.
• Within the pass band and for 𝜔 ≥ 0 the phase can be described as
∠𝐻 𝜔 = 𝜙0 − 𝜔𝑡𝑔
• The phase is always an odd function, and therefore,
∠𝐻 −𝜔 = −∠𝐻 𝜔 = −(𝜙0 −𝜔𝑡𝑔 ) = −𝜙0 + 𝜔𝑡𝑔
• We can write:
𝜙0 − 𝜔𝑡𝑔 𝜔≥0
∠𝐻 𝜔 =
−𝜙0 − 𝜔𝑡𝑔 𝜔<0
• For a distortionless system we have 𝐻 𝜔 = 𝐺𝑜 𝑒 𝑗(𝜙0−𝜔𝑡𝑔) , 𝜔 ≥ 0.
• Consider the distortionless system 𝐻 𝜔 = 𝐺𝑜 𝑒 𝑗(𝜙0−𝜔𝑡𝑔) , 𝜔 ≥ 0.

• Consider the bandpass modulated signal 𝑧 𝑡 = 𝑥 𝑡 cos𝜔𝑐 𝑡 centred at 𝜔𝑐
where 𝑥(𝑡) is a lowpass signal with bandwidth 𝑊.
 cos𝜔𝑐 𝑡 is the carrier of 𝑧(𝑡)
 𝑥(𝑡) is the envelope of 𝑧(𝑡)
• Consider now the input 𝑧 𝑡 = 𝑥 𝑡 𝑒 𝑗𝜔𝑐 𝑡 with 𝑍 𝜔 = 𝑋(𝜔 − 𝜔𝑐 ).
• The corresponding output is:
𝑌 𝜔 = 𝐻(𝜔) 𝑍 𝜔 = 𝐻(𝜔)𝑋(𝜔 − 𝜔𝑐 )
𝑌 𝜔 = 𝐺0 𝑋 𝜔 − 𝜔𝑐 𝑒 𝑗(𝜙0−𝜔𝑡𝑔) = 𝐺0 𝑒 𝑗𝜙0 𝑋(𝜔 − 𝜔𝑐 )𝑒 −𝑗𝜔𝑡𝑔
• We use the properties:
 If 𝑥 𝑡 ⇔ 𝑋 𝜔 then:
𝑥 𝑡 − 𝑡0 ⇔ 𝑋 𝜔 𝑒 −𝑗𝜔𝑡0 and 𝑥 𝑡 𝑒 𝑗𝜔0𝑡 ⇔ 𝑋 𝜔 − 𝜔0 .
• We obtain: 𝑦 𝑡 = 𝐺0 𝑒 𝑗𝜙0 𝑥 𝑡 − 𝑡𝑔 𝑒 𝑗𝜔𝑐 (𝑡−𝑡𝑔) = 𝐺0 𝑥 𝑡 − 𝑡𝑔 𝑒 𝑗[𝜔𝑐 𝑡−𝑡𝑔 +𝜙0 ]
• Consider the distortionless system 𝐻 𝜔 = 𝐺𝑜 𝑒 𝑗(𝜙0−𝜔𝑡𝑔) , 𝜔 ≥ 0.
• We showed that for the input 𝑧 𝑡 = 𝑥 𝑡 𝑒 𝑗𝜔𝑐 𝑡 the output is:

𝑦 𝑡 = 𝐺0 𝑥 𝑡 − 𝑡𝑔 𝑒 𝑗[𝜔𝑐 𝑡−𝑡𝑔 +𝜙0]
• For the input 𝑧 𝑡 = 𝑥 𝑡 cos𝜔𝑐 𝑡 = Re{𝑧 𝑡 } the output is

𝑦 𝑡 = Re 𝑦 𝑡 = Re 𝐺0 𝑥 𝑡 − 𝑡𝑔 𝑒 𝑗[𝜔𝑐 𝑡−𝑡𝑔 +𝜙0]
= 𝐺0 𝑥 𝑡 − 𝑡𝑔 cos[𝜔𝑐 𝑡 − 𝑡𝑔 + 𝜙0 ]
 The output envelope 𝑥 𝑡 − 𝑡𝑔 remains undistorted.
 The output carrier acquires an extra phase 𝜙0 .
 In a modulation system the transmission is considered distortionless if
the envelope 𝑥(𝑡) remains undistorted. This is because the signal
information is contained solely in the envelope.
 Therefore, the above type of transmission is considered distortionless.
Example
• A signal 𝑧(𝑡) shown below is given by 𝑥 𝑡 cos𝜔𝑐 𝑡 where 𝜔𝑐 = 2000𝜋. The

pulse 𝑥(𝑡) is a lowpass pulse of duration 0.1sec and has a bandwidth of
about 10𝐻𝑧 . This signal is passed through a filter whose frequency
response is shown below. Find and sketch the filter output 𝑦(𝑡).
• 𝑧(𝑡) is a narrow band signal with bandwidth of 20𝐻𝑧 centred around

𝑓𝑐 = 𝜔𝑐 /2𝜋 = 1𝑘𝐻𝑧.
• The gain at the centre frequency of 1𝑘𝐻𝑧 is 2.
2.4𝜋−0.4𝜋
• The group delay is: 𝑡𝑔 = = 10−3 . It can be found by drawing the
2000𝜋
tangent at 𝜔𝑐 .
• The intercept along the vertical axis by the tangent is 𝜙0 = −0.4𝜋.
Example cont.
• Based on the above analysis the output of the system is:
𝑦 𝑡 = 𝐺0 𝑥 𝑡 − 𝑡𝑔 cos[𝜔𝑐 𝑡 − 𝑡𝑔 + 𝜙0 ]
= 2𝑥 𝑡 − 10−3 cos[2000𝜋 𝑡 − 10−3 − 0.4𝜋]
Signal energy: Parseval’s theorem
• The energy of a signal 𝑥(𝑡) can be derived either in time or in frequency

domain:
∞ ∞
1
𝐸𝑥 = 𝑥(𝑡) 2 𝑑𝑡 = 𝑋(𝜔) 2 𝑑𝜔
𝑡=−∞ 2𝜋 𝜔=−∞
Proof
∞ ∞ ∞
∗
1
𝐸𝑥 = 𝑥(𝑡)𝑥 (𝑡)𝑑𝑡 = 𝑥(𝑡) 𝑋 ∗ 𝜔 𝑒 −𝑗𝜔𝑡 𝑑𝜔 𝑑𝑡
𝑡=−∞ 𝑡=−∞ 2𝜋 𝜔=−∞
1 ∞ ∗ ∞ −𝑗𝜔𝑡 𝑑𝑡 𝑑𝜔
= 𝑋 𝜔 𝑥(𝑡)𝑒
2𝜋 𝜔=−∞ 𝑡=−∞
1 ∞ ∗ 1 ∞ 2
= 𝑋 𝜔 𝑋(𝜔)𝑑𝜔 = 𝑋(𝜔) 𝑑𝜔
2𝜋 𝜔=−∞ 2𝜋 𝜔=−∞
Signal energy: Parseval’s theorem cont.
• The total energy of a signal is the area under the curve 𝑋(𝜔) 2 divided by
2𝜋.
• The energy over a small frequency band Δ𝜔, where Δ𝜔 → 0 is:

1 2 2 Δ𝜔
Δ𝐸𝑥 = 𝑋(𝜔) Δ𝜔 = 𝑋(𝜔) Δ𝑓, Δ𝑓 = 𝐻𝑧
2𝜋 2𝜋
• The function 𝑋(𝜔) 2 is the energy spectral density (per unit bandwidth in
𝐻𝑧).
Energy spectral density of a real signal
• If 𝑥(𝑡) is a real signal, then 𝑋(𝜔) and 𝑋(−𝜔) are conjugate.
2 2
• In that case 𝑋(𝜔) is even, since 𝑋(𝜔) = 𝑋 𝜔 𝑋 ∗ 𝜔 = 𝑋 𝜔 𝑋 −𝜔 .
• Therefore,
1 ∞ 1 ∞
𝐸𝑥 = 𝑋(𝜔) 2 𝑑𝜔 = 2 𝑋(𝜔) 2 𝑑𝜔
2𝜋 𝜔=−∞ ∞ 2𝜋 𝜔=0
1
= 𝑋(𝜔) 2 𝑑𝜔
𝜋 𝜔=0
• Consequently, in a real signal the energy contributed by all spectral

components between 𝜔1 and 𝜔2 is:
1 𝜔2
Δ𝐸𝑥 = 𝑋(𝜔) 2 𝑑𝜔
𝜋 𝜔1
Windowing and its effect
• Extracting a segment of a signal in time is the same as multiplying the

signal with a rectangular window:
Spectral spreading
Energy spread out
X * from 𝜔0 to width of
≈ 2𝜋/𝑇.
Leakage
Energy leaks out
from the mainlobe to
the sidelobes.
Windowing and its effect cont.
• Using the previous example as a basis to understand windowing effects

observe that:
 If 𝑥(𝑡) has two spectral components of frequencies which differ by less
4𝜋 2
than 𝑟𝑎𝑑/𝑠𝑒𝑐 ( 𝐻𝑧) they will be indistinguishable in the truncated
𝑇 𝑇
signal.
 The result is loss of spectral resolution.
Mainlobe and sidelobes of a rectangular window in dB
• Amplitude spectrum of a rectangular window in 𝑑𝐵.

Remedies for side effects of truncation
1. Make mainlobe width as narrow as possible; this implies as wide a

window as possible.
2. Avoid big discontinuity in the windowing function to reduce leakage
(i.e., high frequency sidelobes).
 1. and 2. above are incompatible – therefore a compromise is required.
Commonly used windows apart from the rectangular window are:
• Hamming windows
• Hanning windows
• Barlett windows
• Blackman windows
• Kaiser windows
 More on this topic in subsequent
Digital Signal Processing and
Filter Design courses.
Comparison of different windowing functions
Time domain vs frequency domain. Summary.
System’s response to Dirac System’s response to 𝒆𝒋𝝎𝒕 is

function (impulse response) 𝑯(𝝎)𝒆𝒋𝝎𝒕
𝒙(𝒕) as a sum of shifted impulse 𝒙(𝒕) as a sum of everlasting

components exponential components
𝒚(𝒕) as a sum of responses to 𝒚(𝒕) as a sum of responses to

impulse components exponential components

Signals and Systems: DR Tania Stathaki

Uploaded by

Copyright:

Available Formats

Signals and Systems: DR Tania Stathaki

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Signals and Systems: DR Tania Stathaki

Uploaded by

Copyright:

Available Formats

Signals and Systems

• In certain types of systems we require the input to pass through the

• Distortionless transmission implies that for a specific frequency band

• Therefore, if 𝑥(𝑡) is the input and 𝑦(𝑡) is the output, distortionless

• In order to assess phase linearity we can find the slope of ∠𝐻 𝜔 as a

• The pass band is of width 2𝑊 centred at 𝜔𝑐 .

• Consider the distortionless system 𝐻 𝜔 = 𝐺𝑜 𝑒 𝑗(𝜙0−𝜔𝑡𝑔) , 𝜔 ≥ 0.

• Consider the distortionless system 𝐻 𝜔 = 𝐺𝑜 𝑒 𝑗(𝜙0−𝜔𝑡𝑔) , 𝜔 ≥ 0.

• We showed that for the input 𝑧 𝑡 = 𝑥 𝑡 𝑒 𝑗𝜔𝑐 𝑡 the output is:

• For the input 𝑧 𝑡 = 𝑥 𝑡 cos𝜔𝑐 𝑡 = Re{𝑧 𝑡 } the output is

• A signal 𝑧(𝑡) shown below is given by 𝑥 𝑡 cos𝜔𝑐 𝑡 where 𝜔𝑐 = 2000𝜋. The

• 𝑧(𝑡) is a narrow band signal with bandwidth of 20𝐻𝑧 centred around

• Based on the above analysis the output of the system is:

• The energy of a signal 𝑥(𝑡) can be derived either in time or in frequency

• The energy over a small frequency band Δ𝜔, where Δ𝜔 → 0 is:

• If 𝑥(𝑡) is a real signal, then 𝑋(𝜔) and 𝑋(−𝜔) are conjugate.

• Consequently, in a real signal the energy contributed by all spectral

• Extracting a segment of a signal in time is the same as multiplying the

• Using the previous example as a basis to understand windowing effects

• Amplitude spectrum of a rectangular window in 𝑑𝐵.

1. Make mainlobe width as narrow as possible; this implies as wide a

System’s response to Dirac System’s response to 𝒆𝒋𝝎𝒕 is

𝒙(𝒕) as a sum of shifted impulse 𝒙(𝒕) as a sum of everlasting

𝒚(𝒕) as a sum of responses to 𝒚(𝒕) as a sum of responses to

You might also like