
The Wireless Channel
Characterization of the wireless channel and its impact on
digitally modulated signals.
From the physics of propagation to multi-path fading
channels.
Statistical characterization of wireless channels:
Doppler spectrum,
Delay spread
Coherence time
Coherence bandwidth
Simulating multi-path, fading channels in MATLAB.
Lumped-parameter models:
discrete-time equivalent channel.
Path loss models, link budgets, shadowing.



Outline

Part III: Learning Objectives

Pathloss and Link Budget

From Physical Propagation to Multi-Path Fading

Statistical Characterization of Channels



Learning Objectives

Understand models describing the nature of typical
wireless communication channels.
The origin of multi-path and fading.
Concise characterization of multi-path and fading in both
the time and frequency domain.
Doppler spectrum and time-coherence
Multi-path delay spread and frequency coherence
Appreciate the impact of wireless channels on transmitted
signals.
Distortion from multi-path: frequency-selective fading and
inter-symbol interference.
The consequences of time-varying channels.



Outline







Path Loss

Path loss LP relates the received signal power Pr to the
transmitted signal power Pt :

Gr · Gt
Pr = Pt · ,
LP

where Gt and Gr are antenna gains.
Path loss is very important for cell and frequency planning
or range predictions.
Not needed when designing signal sets, receiver, etc.



Path Loss

Path loss modeling is “more an art than a science.”
Standard approach: ﬁt model to empirical data.
Parameters of model:
d - distance between transmitter and receiver,
fc - carrier frequency,
hb , hm - antenna heights,
Terrain type, building density, . . ..



Example: Free Space Propagation
In free space, path loss LP is given by Friis’s formula:
2 2
4πd 4πfc d
LP = = .
λc c

Path loss increases proportional to the square of distance d
and frequency fc .
In dB:
c
LP (dB ) = −20 log10 ( ) + 20 log10 (fc ) + 20 log10 (d ).
4π

Example: fc = 1GHz and d = 1km

LP (dB ) = −146 dB + 180 dB + 60 dB = 94 dB.



Example: Two-Ray Channel
Antenna heights: hb and hm .
Two propagation paths:
1. direct path, free space propagation,
2. reﬂected path, free space with perfect reﬂection.
Depending on distance d, the signals received along the
two paths will add constructively or destructively.
Path loss:
2 2
1 4πfc d 1
LP = · · .
4 c sin( 2πchd hm )
fc
b

For d hb hm , path loss is approximately equal to:
2
d2
LP ≈
hb hm
Path loss proportional to d 4 is typical for urban
environment.©2009, B.-P. Paris Wireless Communications 10


Okumura-Hata Model for Urban Area
Okumura and Hata derived empirical path loss models
from extensive path loss measurements.
Models differ between urban, suburban, and open areas,
large, medium, and small cities, etc.
Illustrative example: Model for Urban area (small or
medium city)

LP (dB ) = A + B log10 (d ),

where
A = 69.55 + 26.16 log10 (fc ) − 13.82 log10 (hb ) − a(hm )
B = 44.9 − 6.55 log10 (hb )
a(hm ) = (1.1 log10 (fc ) − 0.7) · hm − (1.56 log10 (fc ) − 0.8)



Signal and Noise Power
Received Signal Power:
Gr · Gt
Pr = Pt ·
,
LP · LR
where LR is implementation loss, typically 2-3 dB.
(Thermal) Noise Power:
PN = kT0 · BW · F , where

k - Boltzmann’s constant (1.38 · 10−23 Ws/K),
T0 - temperature in K (typical room temperature,
T0 = 290 K),
⇒ kT0 = 4 · 10−21 W/Hz = 4 · 10−18 mW/Hz =
−174 dBm/Hz,
BW - signal bandwidth,
F - noise ﬁgure, ﬁgure of merit for receiver (typical value:
5dB).


Signal-to-Noise Ratio

The ratio of received signal power and noise power is
denoted by SNR.
From the above, SNR equals:

Pt Gr · Gt
SNR = .
kT0 · BW · F · LP · LR

SNR increases with transmitted power Pt and antenna
gains.
SNR decreases with bandwidth BW , noise ﬁgure F , and
path loss LP .



Es /N0

For the symbol error rate performance of communications
system the ratio of signal energy Es and noise power
spectral density N0 is more relevant than SNR.
Pr
Since Es = Pr · Ts = Rs and N0 = kT0 · F = PN /BW , it
follows that
Es B
= SNR · W ,
N0 Rs
where Ts and Rs denote the symbol period and symbol
rate, respectively.



Es /N0

Thus, Es /N0 is given by:

Es Pt Gr · Gt
= .
N0 kT0 · Rs · F · LP · LR
in dB:
E
( Ns )(dB ) = Pt (dBm) + Gt (dB ) + Gr (dB )
0
−(kT0 )(dBm/Hz ) − Rs(dBHz ) − F(dB ) − LR (dB ) .



Outline







Multi-path Propagation
The transmitted signal
propagates from the
transmitter to the receiver
along many different paths.
These paths have different
path attenuation ak ,
path delay τk , TX RX
phase shift φk ,
angle of arrival θk .
For simplicity, we assume
a 2-D model, so that the
angle of arrival is the
azimuth.
In 3-D models, the
elevation angle of arrival
is an additional parameter.



Channel Impulse Response

From the above parameters, one can easily determine the
channel’s (baseband equivalent) impulse response.
Impulse Response:
K
h (t ) = ∑ ak · ejφ k · e−j2πfc τk · δ(t − τk )
k =1

Note that the delays τk contribute to the phase shifts φk .



Received Signal

Ignoring noise for a moment, the received signal is the
convolution of the transmitted signal s (t ) and the impulse
response
K
R (t ) = s (t ) ∗ h (t ) = ∑ ak · ejφ k · e−j2πfc τk · s (t − τk ).
k =1

The received signal consists of multiple
scaled (by ak · ejφk · e −j2πfc τk ),
delayed (by τk )
copies of the transmitted signal.



Channel Frequency Response
Similarly, one can compute the frequency response of the
channel.
Direct Fourier transformation of the expression for the
impulse response yields
K
H (f ) = ∑ ak · ejφ k · e−j2πfc τk · e−j2πf τk .
k =1

For any given frequency f , the frequency response is a sum
of complex numbers.
When these terms add destructively, the frequency
response is very small or even zero at that frequency.
These nulls in the channel’s frequency response are typical
for wireless communications and are refered to as
frequency-selective fading.


Frequency Response in One Line of MATLAB

The Frequency response
K
H (f ) = ∑ ak · ejφ k · e−j2πfc τk · e−j2πf τk .
k =1

can be computed in MATLAB via the one-liner
14 HH = PropData.Field.*exp(-j*2*pi*fc*tau) * exp(-j*2*pi*tau’*ff);

Note that tau’*ff is an inner product; it produces a matrix
(with K rows and as many columns as ff).
Similarly, the product preceding the second complex
exponential is an inner product; it generates the sum in the
expression above.



Example: Ray Tracing
1300

1250

1200

1150
Transmitter

1100

1050
y (m)

1000

950 Receiver

900

850

800

750
450 500 550 600 650 700 750 800 850 900 950
x (m)

Figure: All propagation paths between the transmitter and receiver in
the indicated located were determined through ray tracing.



Impulse Response
−5
x 10
4

3

Attenuation
2

1

0
0.8 1 1.2 1.4 1.6 1.8 2
Delay (µs)

4

2
Phase Shift/π

0

−2

−4
0.8 1 1.2 1.4 1.6 1.8 2
Delay (µs)

Figure: (Baseband equivalent) Impulse response shows attenuation,
delay, and phase for each of the paths between receiver and
transmitter.



Frequency Response
−78

−80

−82

−84
|Frequency Response| (dB)

−86

−88

−90

−92

−94

−96

−98
−5 −4 −3 −2 −1 0 1 2 3 4 5
Frequency (MHz)

Figure: (Baseband equivalent) Frequency response for a multi-path
channel is characterized by deep “notches”.



Implications of Multi-path

Multi-path leads to signal distortion.
The received signal “looks different” from the transmitted
signal.
This is true, in particular, for wide-band signals.
Multi-path propagation is equivalent to undesired filtering
with a linear filter.
The impulse response of this undesired filter is the impulse
response h(t ) of the channel.
The effects of multi-path can be described in terms of both
time-domain and frequency-domain concepts.
In either case, it is useful to distinguish between
narrow-band and wide-band signals.



Example: Transmission of a Linearly Modulated Signal
Transmission of a linearly modulated signal through the
above channel is simulated.
BPSK,
(full response) raised-cosine pulse.
Symbol period is varied; the following values are
considered
Ts = 30µs ( bandwidth approximately 60 KHz)
Ts = 3µs ( bandwidth approximately 600 KHz)
Ts = 0.3µs ( bandwidth approximately 6 MHz)
For each case, the transmitted and (suitably scaled)
received signal is plotted.
Look for distortion.
Note that the received signal is complex valued; real and
imaginary part are plotted.



2
Transmitted
Real(Received)
1.5 Imag(Received)

1

0.5
Amplitude

0

−0.5

−1

−1.5

−2
0 50 100 150 200 250 300
Time (µs)

Figure: Transmitted and received signal; Ts = 30µs. No distortion is
evident.



2
Transmitted
Real(Received)
1.5 Imag(Received)

1

0.5
Amplitude

0

−0.5

−1

−1.5

−2
0 5 10 15 20 25 30 35
Time (µs)

Figure: Transmitted and received signal; Ts = 3µs. Some distortion is
visible near the symbol boundaries.



2
Transmitted
Real(Received)
1.5 Imag(Received)

1

Amplitude 0.5

0

−0.5

−1

−1.5

−2
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
Time (µs)

Figure: Transmitted and received signal; Ts = 0.3µs. Distortion is
clearly visible and spans multiple symbol periods.



Eye Diagrams for Visualizing Distortion
An eye diagram is a simple but useful tool for quickly
gaining an appreciation for the amount of distortion present
in a received signal.
An eye diagram is obtained by plotting many segments of
the received signal on top of each other.
The segments span two symbol periods.
This can be accomplished in MATLAB via the command
plot( tt(1:2*fsT), real(reshape(Received(1:Ns*fsT), 2*fsT, [ ])))

Ns - number of symbols; should be large (e.g., 1000),
Received - vector of received samples.
The reshape command turns the vector into a matrix with
2*fsT rows, and
the plot command plots each column of the resulting matrix
individually.



Eye Diagram without Distortion
0.5

Amplitude
0

−0.5
0 10 20 30 40 50 60
Time (µs)

1

0.5
Amplitude

0

−0.5

−1
0 10 20 30 40 50 60
Time (µs)

Figure: Eye diagram for received signal; Ts = 30µs. No distortion:
“the eye is fully open”.



Eye Diagram with Distortion
2

1
Amplitude

0

−1

−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Time (µs)

2

1
Amplitude

0

−1

−2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Time (µs)

Figure: Eye diagram for received signal; Ts = 0.3µs. Signiﬁcant
distortion: “the eye is partially open”.



Inter-Symbol Interference

The distortion described above is referred to as
inter-symbol interference (ISI).
As the name implies, the undesired ﬁltering by the channel
causes energy to be spread from one transmitted symbol
across several adjacent symbols.
This interference makes detection mored difﬁcult and must
be compensated for at the receiver.
Devices that perform this compensation are called
equalizers.



Inter-Symbol Interference
Question: Under what conditions does ISI occur?
Answer: depends on the channel and the symbol rate.
The difference between the longest and the shortest delay
of the channel is called the delay spread Td of the channel.
The delay spread indicates the length of the impulse
response of the channel.
Consequently, a transmitted symbol of length Ts will be
spread out by the channel.
When received, its length will be the symbol period plus the
delay spread, Ts + Td .
Rule of thumb:
if the delay spread is much smaller than the symbol period
(Td Ts ), then ISI is negligible.
If delay is similar to or greater than the symbol period, then
ISI must be compensated at the receiver.



Frequency-Domain Perspective
It is interesting to compare the bandwidth of the transmitted
signals to the frequency response of the channel.
In particular, the bandwidth of the transmitted signal relative
to variations in the frequency response is important.
The bandwidth over which the channel’s frequency
response remains approximately constant is called the
coherence bandwidth.
When the frequency response of the channel remains
approximately constant over the bandwidth of the
transmitted signal, the channel is said to be ﬂat fading.
Conversely, if the channel’s frequency response varies
signiﬁcantly over the bandwidth of the signal, the channel
is called a frequency-selective fading channel.



Example: Narrow-Band Signal
−75

−80

−85

−90

−95

−100

−5 −4 −3 −2 −1 0 1 2 3 4
Frequency (MHz)

Figure: Frequency Response of Channel and bandwidth of signal;
Ts = 30µs, Bandwidth ≈ 60 KHz; the channel’s frequency response
is approximately constant over the bandwidth of the signal.



Example: Wide-Band Signal
−75

−80

−85

−90

−95

−100

−5 −4 −3 −2 −1 0 1 2 3 4
Frequency (MHz)

Figure: Frequency Response of Channel and bandwidth of signal;
Ts = 0.3µs, Bandwidth ≈ 6 MHz; the channel’s frequency response
varies significantly over the bandwidth of the channel.



Frequency-Selective Fading and ISI

Frequency-selective fading and ISI are dual concepts.
ISI is a time-domain characterization for significant
distortion.
Frequency-selective fading captures the same idea in the
frequency domain.
Wide-band signals experience ISI and
frequency-selective fading.
Such signals require an equalizer in the receiver.
Wide-band signals provide built-in diversity.
Not the entire signal will be subject to fading.
Narrow-band signals experience flat fading (no ISI).
Simple receiver; no equalizer required.
Entire signal may be in a deep fade; no diversity.



Time-Varying Channel

Beyond multi-path propagation, a second characteristic of
many wireless communication channels is their time
variability.
The channel is time-varying primarily because users are
mobile.
As mobile users change their position, the characteristics
of each propagation path changes correspondingly.
Consider the impact a change in position has on
path gain,
path delay.
Will see that angle of arrival θk for k -th path is a factor.



Path-Changes Induced by Mobility
Mobile moves by ∆d from old position to new position.
distance: |∆d |
angle: ∠∆d = δ
Angle between k -th ray and ∆d is denoted ψk = θk − δ.
Length of k -th path increases by |∆d | cos(ψk ).

k -th ray k -th ray

|∆d | sin(ψk )
|∆d | cos(ψk )
ψk
∆d
Old Position New Position



Impact of Change in Path Length
We conclude that the length of each path changes by
|∆d | cos(ψk ), where
ψk denotes the angle between the direction of the mobile
and the k -th incoming ray.
Question: how large is a typical distance |∆d | between
the old and new position is?
The distance depends on
the velocity v of the mobile, and
the time-scale ∆T of interest.
In many modern communication system, the transmission
of a frame of symbols takes on the order of 1 to 10 ms.
Typical velocities in mobile systems range from pedestrian
speeds (≈ 1m/s) to vehicle speeds of 150km/h( ≈ 40m/s).
Distances of interest |∆d | range from 1mm to 400mm.



Question: What is the impact of this change in path length
on the parameters of each path?
We denote the length of the path to the old position by dk .
Clearly, dk = c · τk , where c denotes the speed of light.
Typically, dk is much larger than |∆d |.
Path gain ak : Assume that path gain ak decays inversely
−
proportional with the square of the distance, ak ∼ dk 2 .
Then, the relative change in path gain is proportional to
(|∆d |/dk )2 (e.g., |∆d | = 0.1m and dk = 100m, then path
gain changes by approximately 0.0001%).
Conclusion: The change in path gain is generally small
enough to be negligible.




Delay τk : By similar arguments, the delay for the k -th path
changes by at most |∆d |/c.
The relative change in delay is |∆d |/dk (e.g., 0.1% with the
values above.)
Question: Is this change in delay also negligible?



Relating Delay Changes to Phase Changes

Recall: the impulse response of the multi-path channel is
K
h (t ) = ∑ ak · ejφ k · e−j2πfc τk · δ(t − τk )
k =1

Note that the delays, and thus any delay changes, are
multiplied by the carrier frequency fc to produce phase
shifts.



Relating Delay Changes to Phase Changes

Consequently, the phase change arising from the
movement of the mobile is

∆φk = −2πfc /c |∆d | cos(ψk ) = −2π |∆d |/λc cos(ψk ),

where
λc = c/fc - denotes the wave-length at the carrier
frequency (e.g., at fc = 1GHz, λc ≈ 0.3m),
ψk - angle between direction of mobile and k -th arriving
path.
Conclusion: These phase changes are significant and
lead to changes in the channel properties over short
time-scales (fast fading).



Illustration
To quantify these effects, compute the phase change over
a time interval ∆T = 1ms as a function of velocity.
Assume ψk = 0, and, thus, cos(ψk ) = 1.
fc = 1GHz.
v (m/s) |∆d | (mm) ∆φ (degrees) Comment
1 1 1.2 Pedestrian; negligible
phase change.
10 10 12 Residential area vehi-
cle speed.
100 100 120 High-way speed;
phase change signifi-
cant.
1000 1000 1200 High-speed train or
low-flying aircraft;
receiver must track
phase changes.



Doppler Shift and Doppler Spread
If a mobile is moving at a constant velocity v , then the
distance between an old position and the new position is a
function of time, |∆d | = vt.
Consequently, the phase change for the k -th path is

∆φk (t ) = −2πv /λc cos(ψk )t = −2πv /c · fc cos(ψk )t.

The phase is a linear function of t.
Hence, along this path the signal experiences a frequency
shift fd ,k = v /c · fc · cos(ψk ) = v /λc · cos(ψk ).
This frequency shift is called Doppler shift.
Each path experiences a different Doppler shift.
Angles of arrival θk are different.
Consequently, instead of a single Doppler shift a number of
shifts create a Doppler Spectrum.


Illustration: Time-Varying Frequency Response

−70

−80

−90

−100

−110

−120

−130
200
150 5
100
0
50

Time (ms) 0 −5
Frequency (MHz)

Figure: Time-varying Frequency Response for Ray-Tracing Data;
velocity v = 10m/s, fc = 1GHz, maximum Doppler frequency
≈ 33Hz.



Illustration: Time-varying Response to a Sinusoidal
Input
−80

Magnitude (dB)
−100

−120

−140
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

10

0
Phase/π

−10

−20

−30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure: Response of channel to sinusoidal input signal; base-band
equivalent input signal s (t ) = 1, velocity v = 10m/s, fc = 1GHz,
maximum Doppler frequency ≈ 33Hz.



Doppler Spread and Coherence Time
The time over which the channel remains approximately
constant is called the coherence time of the channel.
Coherence time and Doppler spectrum are dual
characterizations of the time-varying channel.
Doppler spectrum provides frequency-domain
interpretation:
It indicates the range of frequency shifts induced by the
time-varying channel.
Frequency shifts due to Doppler range from −fd to fd , where
fd = v /c · fc .
The coherence time Tc of the channel provides a
time-domain characterization:
It indicates how long the channel can be assumed to be
approximately constant.
Maximum Doppler shift fd and coherence time Tc are
related to each through an inverse relationship Tc ≈ 1/fd .


System Considerations
The time-varying nature of the channel must be accounted
for in the design of the system.
Transmissions are shorter than the coherence time:
Many systems are designed to use frames that are shorter
than the coherence time.
Example: GSM TDMA structure employs time-slots of
duration 4.6ms.
Consequence: During each time-slot, channel may be
treated as constant.
From one time-slot to the next, channel varies signiﬁcantly;
this provides opportunities for diversity.
Transmission are longer than the coherence time:
Channel variations must be tracked by receiver.
Example: use recent symbol decisions to estimate current
channel impulse response.



Illustration: Time-varying Channel and TDMA
−80

−90

−100
Magnitude (dB)

−110

−120

−130

−140

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Time (s)

Figure: Time varying channel response and TDMA time-slots;
time-slot duration 4.6ms, 8 TDMA users, velocity v = 10m/s,
fc = 1GHz, maximum Doppler frequency ≈ 33Hz.



Summary
Illustrated by means of a concrete example the two main
impairments from a mobile, wireless channel.
Multi-path propagation,
Doppler spread due to time-varying channel.
Multi-path propagation induces ISI if the symbol duration
exceeds the delay spread of the channel.
In frequency-domain terms, frequency-selective fading
occurs if the signal bandwidth exceeds the coherence
band-width of the channel.
Doppler Spreading results from time-variations of the
channel due to mobility.
The maximum Doppler shift fd = v /c · fc is proportional to
the speed of the mobile.
In time-domain terms, the channel remains approximately
constant over the coherence-time of the channel.



Outline







Statistical Characterization of Channel
We have looked at the characterization of a concrete
realization of a mobile, wire-less channel.
For different locations, the properties of the channel will
likely be very different.
Objective: develop statistical models that capture the
salient features of the wireless channel for areas of
interest.
Models must capture multi-path and time-varying nature of
channel.
Approach: Models reflect correlations of the time-varying
channel impulse response or frequency response.
Time-varying descriptions of channel are functions of two
parameters:
Time t when channel is measured,
Frequency f or delay τ.



Power Delay Profile
The impulse response of a wireless channel is
time-varying, h(t, τ ).
The parameter t indicates when the channel is used,
The parameter τ reflects time since the input was applied
(delay).
Time-varying convolution: r (t ) = h(t, τ ) · s (t − τ )dτ.
The power-delay profile measures the average power in
the impulse response over delay τ.
Thought experiment: Send impulse through channel at
time t0 and measure response h(t0 , τ ).
Repeat K times, measuring h(tk , τ ).
Power delay profile:
K
1
|h(tk , τ )|2 .
K + 1 k∑
Ψh (τ ) =
=0



Power Delay Profile

The power delay profile captures the statistics of the
multi-path effects of the channel.
The underlying, physical model assumes a large number
of propagation paths:
each path has a an associated delay τ,
the gain for a path is modeled as a complex Gaussian
random variable with second moment equal to Ψh (τ ).
If the mean of the path loss is zero, the path is said to be
Rayleigh fading.
Otherwise, it is Ricean.
The channel gains associated with different delays are
assumed to be uncorrelated.



Example
1 2

0.9 1.5
|h(τ)|2

0.8 1

0.7 0.5
Power Delay Profile

0.6
0
0 2 4 6
0.5 Delay τ (µs)

1
0.4

0.3 0.5
Phase of h(τ)

0.2 0

0.1 −0.5

0 −1
0 2 4 6 0 2 4 6
Delay τ (µs) Delay τ (µs)

Figure: Power Delay Profile and Channel Impulse Response; the
power delay profile (left) equals Ψh (τ ) = exp(−τ/Th ) with Th = 1µs;
realization of magnitude and phase of impulse response (left).



RMS Delay Spread
From a systems perspective, the extent (spread) of the
delays is most signiﬁcant.
The length of the impulse response of the channel
determines how much ISI will be introduced by the channel.
The spread of delays is measured concisely by the RMS
delay spread Td :
∞ ∞
2 (n ) (n )
Td = Ψh (τ )τ 2 dτ − ( Ψh (τ )τdτ )2 ,
0 0

where ∞
(n )
Ψh = Ψh / Ψh (τ )dτ.
0
Example: For Ψh (τ ) = exp(−τ/Th ), RMS delay spread
equals Th .
In urban environments, typical delay spreads are a few µs.


Frequency Coherence Function
The Fourier transform of the Power Delay Spread Ψh (τ ) is
called the Frequency Coherence Function ΨH (∆f )

Ψh (τ ) ↔ ΨH (∆f ).

The frequency coherence function measures the
correlation of the channel’s frequency response.
Thought Experiment: Transmit two sinusoidal signal of
frequencies f1 and f2 , such that f1 − f2 = ∆f .
The gain each of these signals experiences is H (t, f1 ) and
H (t, f2 ), respectively.
Repeat the experiment many times and average the
products H (t, f1 ) · H ∗ (t, f2 ).
ΨH (∆f ) indicates how similar the gain is that two sinusoids
separated by ∆f experience.



Coherence Bandwidth
The width of the main lobe of the frequency coherence
function is the coherence bandwidth Bc of the channel.
Two signals with frequencies separated by less than the
coherence bandwidth will experience very similar gains.
Because of the Fourier transform relationship between the
power delay proﬁle and the frequency coherence function:
1
Bc ≈ .
Td
Example: Fourier transform of Ψh (τ ) = exp(−τ/Th )
Th
ΨH (∆f ) = ;
1 + j2π∆fTh
the 3-dB bandwidth of ΨH (∆f ) is Bc = 1/(2π · Th ).
For urban channels, coherence bandwidth is a few 100KHz.


Time Coherence
The time-coherence function ΨH (∆t ) captures the
time-varying nature of the channel.
Thought experiment: Transmit a sinusoidal signal of
frequency f through the channel and measure the output at
times t1 and t1 + ∆t.
The gains the signal experiences are H (t1 , f ) and
H (t1 + ∆t, f ), respectively.
Repeat experiment and average the products
H (tk , f ) · H ∗ (tk + ∆t, f ).
Time coherence function measures, how quickly the gain
of the channel varies.
The width of the time coherence function is called the
coherence-time Tc of the channel.
The channel remains approximately constant over the
coherence time of the channel.



Example: Isotropic Scatterer
Old location: H (t1 , f = 0) = ak · exp(−j2πfc τk ).
At new location: the gain ak is unchanged; phase changes
by fd cos(ψk )∆t:
H (t1 + ∆t, f = 0) = ak · exp(−j2π (fc τk + fd cos(ψk )∆t )).

k -th ray k -th ray

|∆d | sin(ψk )
|∆d | cos(ψk )
ψk
∆d
Old Position New Position



Example: Isotropic Scatterer

The average of H (t1 , 0) · H ∗ (t1 + ∆t, 0) yields the
time-coherence function.
Assume that the angle of arrival ψk is uniformly distributed.

This allows computation of the average (isotropic scatterer
assumption:

ΨH (∆t ) = |ak |2 · J0 (2πfd ∆t )



Time-Coherence Function for Isotropic Scatterer
1

0.5

ΨH(∆t)

0

−0.5
0 50 100 150 200 250 300
Time ∆t (ms)

Figure: Time-Coherence Function for Isotropic Scatterer; velocity
v = 10m/s, fc = 1GHz, maximum Doppler frequency fd ≈ 33Hz. First
zero at ∆t ≈ 0.4/fd .



Doppler Spread Function

The Fourier transform of the time coherence function
ΨH (∆t ) is the Doppler Spread Function Ψd (fd )

ΨH (∆t ) ↔ Ψd (fd ).

The Doppler spread function indicates the range of
frequencies observed at the output of the channel when
the input is a sinusoidal signal.
Maximum Doppler shift fd ,max = v /c · fc .
Thought experiment:
Send a sinusoidal signal of
The PSD of the received signal is the Doppler spread
function.



Doppler Spread Function for Isotropic Scatterer

Example: The Doppler spread function for the isotropic
scatterer is
|ak |2 1
Ψd (fd ) = for |f | < fd .
4πfd 1 − (f /fd )2



Doppler Spread Function for Isotropic Scatterer
7

6

5

4
Ψd(fd)

3

2

1

0
−40 −30 −20 −10 0 10 20 30 40
Doppler Frequency (Hz)

Figure: Doppler Spread Function for Isotropic Scatterer; velocity
v = 10m/s, fc = 1GHz, maximum Doppler frequency fd ≈ 33Hz. First
zero at ∆t ≈ 0.4/fd .



Simulation of Multi-Path Fading Channels

We would like to be able to simulate the effects of
time-varying, multi-path channels.
Approach:
The simulator operates in discrete-time; the sampling rate
is given by the sampling rate for the input signal.
The multi-path effects can be well modeled by an FIR
(tapped delay-line)filter.
The number of taps for the filter is given by the product of
delay spread and sampling rate.
Example: With a delay spread of 2µs and a sampling rate of
2MHz, four taps are required.
The taps should be random with a Gaussian distribution.
The magnitude of the tap weights should reflect the
power-delay profile.




Approach (cont’d):
The time-varying nature of the channel can be captured by
allowing the taps to be time-varying.
The time-variations should reflect the Doppler Spectrum.



The taps are modeled as
Gaussian random processes
with variances given by the power delay proﬁle, and
power spectral density given by the Doppler spectrum.

s [n ]
D D

a0 (t ) × a1 ( t ) × a2 (t ) ×

r [n ]
+ +



Channel Model Parameters

Concrete parameters for models of the above form have
been proposed by various standards bodies.
For example, the following table is an excerpt from a
document produced by the COST 259 study group.

Tap number Relative Time (µs) Relative Power (dB) Doppler Spectrum
1 0 -5.7 Class
2 0.217 -7.6 Class
3 0.512 -10.1 Class
.
. .
. .
. .
.
. . . .
20 2.140 -24.3 Class



Channel Model Parameters
The table provides a concise, statistical description of a
time-varying multi-path environment.
Each row corresponds to a path and is characterized by
the delay beyond the delay for the shortest path,
the average power of this path;
this parameter provides the variance of the Gaussian path
gain.
the Doppler spectrum for this path;
The notation Class denotes the classical Doppler spectrum
for the isotropic scatterer.
The delay and power column specify the power-delay
proﬁle.
The Doppler spectrum is given directly.
The Doppler frequency fd is an additional parameter.



Toolbox Function SimulateCOSTChannel

The result of our efforts will be a toolbox function for
simulating time-varying multi-path channels:
function OutSig = SimulateCOSTChannel( InSig, ChannelParams, fs)

Its input arguments are
% Inputs:
% InSig - baseband equivalent input signal
% ChannelParams - structure ChannelParams must have fields
11 % Delay - relative delay
% Power - relative power in dB
% Doppler - type of Dopller spectrum
% fd - max. Doppler shift
% fs - sampling rate



Discrete-Time Considerations
The delays in the above table assume a continuous time
axis; our time-varying FIR will operate in discrete time.
To convert the model to discrete-time:
Continuous-time is divided into consecutive “bins” of width
equal to the sampling period, 1/fs.
For all paths arriving in same “bin,” powers are added.
This approach reﬂects that paths arriving closer together
than the sampling period cannot be resolved;
their effect is combined in the receiver front-end.
The result is a reduced description of the multi-path
channel:
Power for each tap reﬂects the combined power of paths
arriving in the corresponding “bin”.
This power will be used to set the variance of the random
process for the corresponding tap.



Converting to a Discrete-Time Model in MATLAB

%% convert powers to linear scale
Power_lin = dB2lin( ChannelParams.Power);

%% Bin the delays according to the sample rate
29 QDelay = floor( ChannelParams.Delay*fs );

% set surrogate delay for each bin, then sum up the power in each bin
Delays = ( ( 0:QDelay(end) ) + 0.5 ) / fs;
Powers = zeros( size(Delays) );
34 for kk = 1:length(Delays)
Powers( kk ) = sum( Power_lin( QDelay == kk-1 ) );
end



Generating Time-Varying Filter Taps
The time-varying taps of the FIR filter must be Gaussian
random processes with specified variance and power
spectral density.
To accomplish this, we proceed in two steps:
1. Create a filter to shape the power spectral density of the
random processes for the tap weights.
2. Create the random processes for the tap weights by
passing complex, white Gaussian noise through the filter.
Variance is adjusted in this step.
Generating the spectrum shaping filter:
% desired frequency response of filter:
HH = sqrt( ClassDoppler( ff, ChannelParams.fd ) );
% design filter with desired frequency response
77 hh = Persistent_firpm( NH-1, 0:1/(NH-1):1, HH );
hh = hh/norm(hh); % ensure filter has unit norm



Generating Time-Varying Filter Taps
The spectrum shaping filter is used to filter a complex
white noise process.
Care is taken to avoid transients at the beginning of the
output signal.
Also, filtering is performed at a lower rate with subsequent
interpolation to avoid numerical problems.
Recall that fd is quite small relative to fs .

% generate a white Gaussian random process
93 ww = sqrt( Powers( kk )/2)*...
( randn( 1, NSamples) + j*randn( 1, NSamples) );
% filter so that spectrum equals Doppler spectrum
ww = conv( ww, hh );
ww = ww( length( hh )+1:NSamples ).’;
98 % interpolate to a higher sampling rate
% ww = interp( ww, Down );
ww = interpft(ww, Down*length(ww));
% store time-varying filter taps for later use



Time-Varying Filtering
The final step in the simulator is filtering the input signal
with the time-varying filter taps.
MATLAB’s filtering functions conv or filter cannot be used
(directly) for this purpose.
The simulator breaks the input signal into short segments
for which the channel is nearly constant.
Each segment is filtered with a slightly different set of taps.

while ( Start < length(InSig) )
EndIn = min( Start+QDeltaH, length(InSig) );
EndOut = EndIn + length(Powers)-1;
118 OutSig(Start:EndOut) = OutSig(Start:EndOut) + ...
conv( Taps(kk,:), InSig(Start:EndIn) );

kk = kk+1;
Start = EndIn+1;



Testing SimulateCOSTChannel

A simple test for the channel simulator consists of
“transmitting” a baseband equivalent sinusoid.
%% Initialization
ChannelParameters = tux(); % COST model parameters
6 ChannelParameters.fd = 10; % Doppler frequency

fs = 1e5; % sampling rate
SigDur = 1; % duration of signal

11 %% generate input signal and simulate channel
tt = 0:1/fs:SigDur; % time axis
Sig = ones( size(tt) ); % baseband-equivalent carrier

Received = SimulateCOSTChannel(Sig, ChannelParameters, fs);



Testing SimulateCOSTChannel
1.8

1.6

1.4

1.2

Magnitude
1

0.8

0.6

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)

Figure: Simulated Response to a Sinusoidal Signal; fd = 10Hz,
baseband equivalent frequency f = 0.



Summary

Highlighted unique aspects of mobile, wireless channels:
time-varying, multi-path channels.
Statistical characterization of channels via
power-delay proﬁle (RMS delay spread),
frequency coherence function (coherence bandwidth),
time coherence function (coherence time), and
Doppler spread function (Doppler spread).
Relating channel parameters to system parameters:
signal bandwidth and coherence bandwidth,
frame duration and coherence time.
Channel simulator in MATLAB.



Where we are ...

Having characterized the nature of mobile, wireless
channels, we can now look for ways to overcome the
detrimental effects of the channel.
The importance of diversity to overcome fading.
Sources of diversity:
Time,
Frequency,
Space.
Equalizers for overcoming frequency-selective fading.
Equalizers also exploit freqeuncy diversity.


Wireless Channel Modeling - MATLAB Simulation Approach

Recommended

More Related Content

What's hot (20)

Similar to Wireless Channel Modeling - MATLAB Simulation Approach (20)

Wireless Channel Modeling - MATLAB Simulation Approach