Small Scale Fading in Radio Propagation

16:332:546 Wireless Communication Technologies Spring 2005 Suhas Mathur (suh as@winlab.rutgers.edu )

Department of Electrical Engineering, Ru tgers University, Piscataway, NJ 08904

Abstract - One of the many impairments inherently present in any wireless communication system, that must be recognised and often effectively mitigated for a system to function well, is fading. Fading itself has been studied and classified into a number of different types. Here we present a detailed mathematical analysis and some useful models for capturing the effect of small scale fading. Further we discuss the types of fading as per the behaviour of the wireless channel with respect to the transmit signal.

I. RAYLEIGH FADING

Figure 1:

Small scale fading is a characteristic of radio propagation resulting from the presence reflectors and scatt erers that cause multiple versions of the transmitted signal to arrive at the receiver, each distorted in amplitude, phase and angle of arrival. Consider the situation shown in Fig. 1 wherein a mobile receiver (mobile station or MS) is assumed to be travelling along the positive x axis with a velocity v m/s. The figure shows one of the many waves arriving at the mobile station. Let us call this the nth incoming wave. Let it be incident at an angle n (t), where the dependence on t stems from the fact that the receiver is not stationary. The motion of the MS produces a Doppler shift in the received frequency as compared to the carrier frequency. This doppler offset is given by:

The figure shows a mobile station moving alo ng the positive x-axis moving at a velocity of v m/s and the nth incoming wav e at an angle of n (t).

fD,n (t) = fm cos(n (t))

(1)

where fm = maximum Doppler frequency = v/, being the wavelength of the radiowave. Waves arriving from the direction of motion cause a positive doppler shift, while those coming from the opposite diection cause a negative doppler shift. We wish to derive a mathematical framework to characterize the effects of small scale fading. Consider the transmit bandpass signal: s(t) = Re{u(t).ej2fc t } (2)

Taught by Dr. Narayan Mandayam, Rutgers University.

where u(t) is the complex baseband equaivalent of the bandp ass transmit signal. If N waves arrive at 1

the MS, the received bandpass signal can be writt en as: x( t) = Re{r(t)ej2fc t } with X r(t) = n (t).e where
n=1 N j2n (t)

shall see later. Further continuing our modeling of the received signal, we can neglect the baseband modulating signal for narrowband signals (i.e. signals in which the baseband signal bandwidth is very small compared to the carrier frequency, which is true of most communication systems) and consider the unmodulated carrier alone.
N X n=1 N X jn (t) jn (t)

(3)

u(t n (t))

(4)

n (t) = (fc + fD,n (t))n (t) fD,n (t).t (5) is the phase associated with the nth wave. The above expression for r(t) looks like the output of a linear time-varying system. Therefore the channel can be modeled as a linear filter with a time varying impulse response given by: c(, t) = (t)e
N X

r(t) =

n (t)e

(7)

x( t) = Re{ (t)e

} ej2fc t

(8)

n=1

= rI (t) cos(2fc t) rQ (t) sin(2fc t) n

jn (t)

( n (t))

(6) where

n=1

c(, t) is the channel response at time t to an input at time t . Typically the quantity fc + fD,n (t) is large. This means that a small change in delay n (t) causes a large change in the phase n (t). The delays themselves are random. This implies that the phases of the incoming waves are random. The n (t)s are not very different from one another, i.e. the n (t)s do not change much over a small time scale. Therefore the received signal is a sum of a large number of waves with random phases. The random phases imply that sometime these waves add constructively producing a received signal with large amplitude, while at other times they add destructively, resulting in a very low amplitude. This precise effect is termed small-scale fading, and the time scale at which the resulting fluctuation of amplitude occurs is of the order of one wave-cycle of the carrier frequency. The range of amplitude variation that can result can be upto 60 to 70 dB. Small scale fading is therefore pramarily due to the random variations in phase n (t) and also because of the doppler frequency fD,n (t). The effect of fading is even more important at higher data rates, as we

r (t) = I

N X n n=1 N X n=1

(t) cos(2f t)
c

(9)

rQ (t) =

n (t) sin(2fc t)

(10)

r(t) = rI (t) + rQ (t)

(11)

rI (t) and rQ (t) are respectively the in-phase and the quadrature-phase components of the complex base- band equivalent of the received signal. Now we invoke the Central Limit Theorem for large N . This makes rI (t) and rQ (t) independent gaussian random pro- cesses. Further, assuming all the random processes involved are WSS, we have: fD,n (t) = fD,n n (t) = n (12)

(13)

n (t) = n We also assume that x(t) is WSS. xx ( ) = E{x(t).x(t + )} = rI rI ( ). cos(2fc t) rQ rI ( ). sin(2fc t) Now, rI rI ( ) = E{rI (t)r I (t + )}
N NX X = E{{ i cos(i t)}.{ j cos(j (t + ))}} i=1 j=1

(14) = (15)

p E {cos(2fm cos())} 2

If the 2-D isotropic scatt ering assumption is used in the above analysis (i.e. the incoming angle is uniformly distributed over (, )) then the above is called the Clarkes Model. Using the uniform distribution for in the above, we get: p 1 . 2 2 Z
+

(16)

rI rI ( ) =

cos(2fm . cos())d (23)

which with a change of variable gives us: Z p 1 + = . cos(2fm . sin())d 2 |

0

(24)

{ z
J0 (2 fm )

We can assume the j s are independent because delays and doppler shifts are independent from path to path. n (t) = U (, ) (17) On evaluating the expectations, we get: rI rI ( ) = where p E {cos(2fD,n )} 2
N

p .J 0(2f m ) 2 where J0 (.) is the Bessel function of the zeroth order and first kind.1 =

Similarly, using the uniform pdf for in the expres(18) sion for cross correlation of the in-phase and quadrature phase components of r(t) gives: (19) rI r Q = 0 (25)

We are now in a position to talk about the PSD of which is the total average received power from all rI (t): SrI rI (f ) = F {rI rI ( )} (26) multip ath components. Now, in the expression above (18), we have ( 1 p | f | < fm fD,n = fm cos(n ) (20) 4fm 1(f /fm )2 = 0 otherwise Therefore, we have the auto correlation function 1 of the in-phase component rI (t): The Bessel functions of the first kind Jn (x) are defined rI rI ( ) = p E {cos(2fm cos()} 2
2 as the solutions to the Bessel differential equation: x

p 1X = E 2 i 2 2 i=1

(21)

x + (x2 n2 )y = 0. The bessel function Jn (x) can also be defined in terms of the contour integral: Jn (x) = e(x/2)( t1/t) tn1dt where the contour encloses the origin and is traversed in a counter-clockwise direction. For the special case of n = 0 a closed form expression due to Frobenius is J (x) = P K(4x ) 1 2 (1) (k!)2 2 or the integral J0 (x) = o k=0
1 1 2j

dy dx

d2 y dx2

Going through a similar series of steps for the cross-correlation function between the in-phase and quadrature-phase component, we get: rI rQ ( ) = E {rI (t)rQ (t + )} (22)

ejx cos() d

( = F

rI rI ( ).ej2fc t + rI rI ( ).e 2

j2fc t

Note that rI rI ( ) = rI ( ), and so for real rI rI (t), rI rI ( ) = rI rI ( ). Thus we have: Sxx (f ) = F rI rI ( ).e
j2fc t

+ rI rI ( ).e 2

j2fc t

(31) 1 {SrI rI (f fc ) + SrI rI (f fc )} 2 (32)

Figure 2:

Bessel fun ction of the zeroth or de r and the first type. This is the shape of the autocorr elation fun ction rI rI ( ) of the in-phase component of the com lex baseb and equivalent of the p rece ived signal.

Sxx (f ) =

Now we shall make use of the knowledge that that r(t) = rI (t) + j.rQ (t) is a complex Gaussian process for large N . Therefore the envelope z(t) = |r(t)| = q 2 2 rI (t) + rQ (t) has a Rayleigh distribution2 x x2 /22; x 0 (33) .e 2 where E{z 2 } = p = 2 2 = average power. Thus we have the probability density function of the recd. signal given by: Pz (x) = Pz (x) =
x2 /p; x 0 x .e p /2

Having obtained the PSD of rI (t), we can now proceed to derive the PSD x(t) as follows: r(t) = rI (t) + jrQ (t) (27)

rr ( ) = E {r (t).r(t + )}

(28)

(34)

= rI rI ( ) + jrI rQ ( ) Therefore:

Further:

rr ( ) = rI rI ( ) xx ( ) = Re rr ( ).ej2fc t
j2f = Re rI rI ( ).e c t

The above is called Rayleigh fading and is derived from Clarkes fading model, wherein the PSD of the received signal has the U-shape shown above. Rayleigh fading is generally applicable when there is no line-of-sight component. This is a good model for cellular mobile radio. Also note that the squared envelope |r(t)|2 is exponentially distributed at any time t: (29) Pz 2 (x) = 1 .ex/p ; x 0 p (35)

S
x

= F Re rI rI ( ).e

j2fc t

(30)

It is kno wn tha t the ran do m var iab le obt ain ed by fin din g the squ are roo t of the su m of the squ are s of tw o ind epe nde nt gau ssi an ran do m var iab les has a Ra yle igh dis tri but ion

simple AWGN model with no fading. The avegrage power is given by E[z 2 ] = p = s2 + 2 2 . Also: s2 = K p K +1 , 2 2 = p K +1 (37)

III. NAK AGAMI FADIN G MODEL

Figure 3:

Power spectral density of the rece ived signal, Sx (f x ). This is call ed the U-shaped PS D characteristic of Rayleigh fading modeled by the clarkes model

The Nakagami Fading model is a purely emperical model and is not based on results derived from physical consideration of radio propagation. It uses a chisquare distrbution with m degrees of freedom. The distribution of the received signals envelope is given by: pz (x) = 2mm x2m1 .e mx2 /p 1 ; m (m)m 2 p (38)

II. RICIAN FADING

We now consider the situation that arises when there is a line-of-sight component in the received signal. This is common in microcellular systems. The probability dstribution for the envelope of the received signal is then given by: x (x2 +s2 ) n xs ; x 0 .e 22 .I0 2 2

pz (x) =

(36)

where s2 = 2 cos2 0 + 2 sin2 0 = 2 = non cen0 0 0 trality parameter. It denotes the power in the lineof-sight component. 3 I (.) is the modified bessel function of the zeroth order 4 .
s The quantity K = 22 is called the Rice factor. Note 2 that sett ing K = 0 transforms this model into 2 dx the dx Rayleigh fading model discussed in the preceeding section and sett ing K = would transform it into a
3 0 denotes the amplitude gain of the zeroth wave, (ref: previous section) which in this case is the line-of-sight compo- nent. 4 The modified Bessel functions I (x) are defined as the son
2

where p = E[z 2 ] = average power, and m is a model parameter. By varying the value of this parameter, the model can capture various distributions. For m = 1 the model converges to the Rayleigh fading model, sett ing m = 1/2 makes it a one sided gaussian distribution, while sett ing m = transforms it into a no-fading model. Finally, the Rician dis- tribution can be approximated though the Nakagami model using the followingrelationships: m2 m K = ;m 1 m m2 m or,
dy 2 2

(39)

lutions to the modified Bessel differential equation: x2

(K + 1)2

d y

x (x + n )y = 0 and can be expressed in terms of the Bessel functions as: In (x) = (j )n Jn (j x)

(40) 2K + 1 The Nakagami model is favoured because it has a closed form analytical expression. All the small-scale fading models considered above assume that all frequencies in the transmitted signal are affected similarly by the channel, i.e. by the fading. This is called flat-fading or frequency nonselective fading.

m=

Figure 4:

The power de lay prof ile gi ves the the power rece ived as a fun ction of time whe n an impu lse is tra nsmitt ed over the wireless chann el.

Figure 5:

Since r ece ived power can only be me asured on a discrete time scale, we can only have a discrete power de lay profile, which ind icates the power rece ived at discrete instants of time whe n an impu lse is tra nsmitt ed on the wireless chann el.

The RMS delay spread is a way of quantifying the multipath nature of the channel. It is of the order Consider only WSSUS (Wide Sense Stationary of s in outdoor situation and of the order of ns in Uncorrelated Scatt ering). Recall that the channel indoor situations. Note that the absolute transmit response is given by c(t, ) and represents power level does not affect the definition of the response of the channel at time t to an input and impulse . Instead the above two definitions only at time t depend . on the relative amplitudes of multip ath components. Definition: The power delay profile or multipath in- As against the power delay profile shown above, in tensity profile is defined as: reality we can only have a discrete power delay profile. Corresponding to this discrete delay profile, we 1 c ( ) = E[c(t, )c (t, )] (41) have the following definitions: 2 P It gives the average power at the channel output as k P (k )k (44) = P a function of time delay. k P (k ) Definition: Average delay is defined as: R 0 c ( )d = R c ( )d 0 = (42) where P 2 2 = k P (k ) k P k P (k ) (46) q 2 ()2 (45)

IV. FREQUENCY SELECT IVE FADIN G CHANN ELS

Definition: RMS Delay spread is defined as:

sR =

0 ( R 0

T )2 c ( )d c ( )d

(43)

V. CHAR ACTE RIZATION CHANN ELS

OF

FADIN G

Fading radio channels have been classified in two ways. 5 The first type of classification discusses whether the fading is flat (frequency non-selective) or frequency selective, while the second classification is based on the rate at which the wireless channel is changing (or in other words, the rate of change of the impulse response of channel), i.e. whether the fading is fast or slow. In connection with these characterizations of fading channels, it is useful to note the following quantities: Cohe rence bandwidth : Coherence bandwidth is a statistical measure of the range of frequencies over which the channel can be considered flat (i.e. frequency non-selective, or in other words a channel which passes all spectral components with equal gain and phase). It may also be defined as the range of frequencies over which any two frequency components have a strong potential for amplitude correlation. It has been shown that: Bc 1

defined as a measure of spectral broadening caused by the time-rate of change of the channel (related to the doppler frequency). 6 . The coherence time is a statistical measure of the time duration over which two received signals have a strong potential for amplitude correlation. Thus if the inverse bandwidth of the basebad signal is greater than the coherence time of the channel then the channel changes during transmission of he baseband message. This will cause a distortion at the receiver. It is shown that: 1 Tc BD (48)

If the coherence time is defined as the duration of time over which the time correlation function is > 0.5, then: s 9 Tc (49) 2 16fm where fm is the maximum doppler frequency = v/.

Ex ample - Consider a vehicle travelling at 60 mi. (47) per hour and communicating with a stationary base station using a carrier frquency fc = 900 Mhz. This where is the RMS delay spread. Also, if we define would give a channel coherence time of Tc 6.77 the coherence bandwidth as that bandwidth over msec. Therefore if the symbol rate of transmission which the frequency correlationfunction is above is greater than 150 samples per second then the 0.5 (i.e. the normalized cross-correlation coefficient fading nature of the channel doesnt really affect the 1 > 0.5 for all frquencies) then Bc 5 . Note that transmitted signal being received by the receiver in a if the signal bandwidth is > BC , then the different Channels in Mobile Digital Communication Systems: Part I: Caharac- terization, by Bernard Sklar. frequency components in the signal will not be faded the same way. The channel then appears to be frequency-selective to the transmitted signal. Dopp ler spread and Cohe rence time : While and Bc describe the time dispersive nature of the channel in an area local to the receiver, they do not offer any information about the time-variations of the channel due to relative motion between the trans- mitter and the receiver. The doppler spread BD ,
5 An excellent treat ment of the characterization of fading channels is found in an article in the Sept. 1997 issue of the IEEE communications magaz ine: Rayleigh Fading

harmful way. For a smaller symbol rate, the symbol width is so large that the channel changes (symbol duration > Tc ) within a single symbol. Flat fading : If a channel has a constant response for a bandwidth > the transmitted signal bandwidth, then the channel is said to be a flat fading channel. The conditions for a flat fading channel are: Bs Ts Bc Tc (50) (51)

6 If the baseband signal frequency is much greater than the doppler spread BD then the effects of doppler spread are negligible

where Bs and Ts are the signal bandwidth and the Slow fading : In a slow fading channel, the chansymbol duration respectively. nel impulse response changes at a rate much slower than the transmitted baseband signal S(t). In the freFrequency selective fading : A channel is quency domain, this implies that the Doppler spread said to be frequency selective if the signal of the channel is much less than the bandwidth of the bandwidth is greater than the coherence bandwidth baseband signal. There fore, a signal undergoes slow of the chan- nel. In such a case, different frequency fading if: components of the transmit signal undergo fading to different ex- tents. For a frequency-selective fading situation: Bs > Bc Ts < Tc (52) (53)
Key Channe l Parameters and Time Scales Carri er frequenc y Commun ication band width Distance between Tx and Rx Velocity of mobile Do ppler shift for a path Time for change in path gain Time for change in path ph ase Coherenc e time Delay spread Coherenc e band width Symbol fc W d v fm = fc v/c d/v 1/(4fm ) Tc = 1/(BD ) Bc 1/2 Typical Value 1 GHz 1 MHz 1 km 64 km/h 50 Hz 1 min 5 ms 2.5 ms 1 s 500 kHz

Ts Bs

Tc BD

(57) (58)

The concept of pulse-shaping is used to control the transmit signal bandwidth. This is used in the degin of the transmit symbol such that given the required symbol rate of transmission, a pulse shape is designed so as to make the signal bandwidth fit within the coherence bandwidth of the signal. Ofcourse, this places an upper limit on the achievable symbol rate. OFDM attempts to solve this problem by breaking up the signal bandwidth into sub-carriers, each of which can be individually transmitted without the channel behaving in a frequency - selective manner. A common rule of thumb to characterize a channel as frequency selective is that if: > 0.1Ts (54)

Table 1: A summary of the physical parameters of the channel and the time scale of change of the key parameters in its discrete-time baseband model. (Taken from Fundamentals of Wireless Communication, David Tse, University of California Berkely, Promod Vishwanath, University of Illinios Urbana-champaign)

Fast fading : In a fast fading channel, the chan- nel impulse response changes rapidly within the sym- bol duration, i.e. the coherence time of the channel is smaller that the symbol period of the transmit- ted signal. Viewed in the frequency domain, signal distortion due to fast fading increases with increas- ing Doppler spread relative to the bandwidth of the transmitted signal. Therefore, a signal undergoes fast fading if: T s > Tc

It must be noted that the wireless channel is function of what is transmitted over it. In order to determine whether fading will affect communication on a wireless channel, we must compare the symbol duration of data transmission with the coherence time and the bandwidth of the baseband signal (fast / slow fading) with the coherence bandwidth of the channel (flat / frequency selective nature).

It should also be clear that when a channel is specified (55) as a fast or slow fading channel, it does not Bs < BD (56) specify whether the channel is flat fading or frequency selec- tive in nature. These are two where BD is the Doppler spread of the channel and independent classifica- tions. Fast and slow fading deal with the time rate of change of the channel with Tc is its coherence time. reference to the transmit- ted signal, whicle flat and

frequency-selective fading

deal with weather the relationship between the signal bandwidth and the range of frequencies over which the fading behaviour of the channel is uniform.

References
[1]
Wireless communication technologies, lecture notes, Spring 2005, Dr. Narayan Mandayam, Rutgers Univer- sity Rayleigh fading channels in mobile digital communication systems, Part I: Characterization, Bernard Sklar, IEEE Communications Magaz ine, Sept. 1997 Wireless Communications, Andrea Goldsmith, Stanford University 2004

[2]

[3] [4]

Fund amentals of Wireless Communication, David Tse, University of California Berkely, P romod Vishwanath, University of Illinios Urbana-champaign