A random variable (rv) is unknown and unpredictable beforehand, that is, it has a random value, but its value is known completely once it occurs. A rv may be continuous or discrete. For example, the noise voltage generated by an electronic amplifier is random and has a continuous amplitude. On the other hand, coin flipping with outcomes head (H) and tail (T) is discrete. One does not know the outcome before flipping a coin. For each experiment, a value is assigned for each of the possible outcomes in the experiment. For example, the rv may be assumed to be X(s) = 1 for the outcome s = H and −1 for s = T. However, once the coin is flipped, the outcome (H or T) and hence X(s) is known. In the sequel we denote the rv simply by X but not as X(s).
A rv is characterized by its probability density function (pdf) or cumulative distribution function (cdf), which are interrelated. The cdf, FX(x), of a rv X is defined by
which specifies the probability that the rv X is less than or equal to a real number x. The pdf of a rv X is defined as the derivative of the cdf:
Conversely, the cdf is defined by the integral of the pdf:
Based on (F.1)–(F.3), a rv has the following properties:
Figure F.1 shows the pdf and cdf for coin flipping, based on the assumption that the probability of head and tail are equally likely, that is P(H) = P(T) = 1/2. The pdf and the cdf may then be written as
A rv is characterized by its moments. The n‐th moment of a rv X is defined as the expectation of Xn:
where integration by parts is used to obtain the last expression. The mean (expected) value mX of a rv X is given by its first moment:
In some applications we encounter functions of rv’s. For example, the mean value of Y = g(X) may be determined as follows:
Central moments of X may be determined using Y = (X−mx)n:
The variance of X is given by
When X is discrete or of a mixed type, the pdf contains impulses at the points of discontinuity of FX(x). In such cases, the discrete part of fX(x) may be expressed as
where the rv X is assumed to be discontinuous at N points, x1, x2,…, xN. For example, in case of coin flipping, two outcomes may be represented as x1 for head and x2 for tail. Then, P(X = x1) = p ≤ 1 shows the probability of head, while P(X = x2)= 1−p denotes the probability of tail. Mean value and the variance of a discrete rv is found by inserting (F.10) into (F.6) and (F.9), respectively:
When the events are equally likely, that is, , (F.11) simplifies to
The median is closely related to the mean value of a rv. In statistics and probability theory, the median is the value of the random number separating the higher half of a data sample from the lower half. The median of N samples can be found by arranging all the observations from lowest value to highest value and picking the middle one. If there is an even number of observations, then there is no single middle value; the median is then usually defined as the mean of the two middle values. The median coincides with the mean value if the rv has a symmetrical pdf.
Now consider two rv’s X and Y, each of which may be continuous, discrete or mixed. The probability that (X,Y) take values in the rectangle shown in Figure F.2 is given by
The joint cdf of the rv’s X and Y may be obtained by inserting and into (F.13):
where
Joint pdf and joint cdf are related to each other as follows:
The marginal pdf’s are found as follows:
Joint moments of two rv’s X and Y may be obtained from their joint pdf :
Joint central moments are defined as
The covariance of X and Y is given by
Correlation coefficient between X and Y is defined by
If the experiments result in mutually exclusive outcomes, then the probability of an outcome in one experiment is statistically independent of the outcome in any other experiment. Then, the joint pdf’s (cdf’s) may be written as the product of the pdf’s (cdf’s) corresponding to each outcome:
Two rv’s are said to be uncorrelated with each other if the correlation coefficient is identically equal to zero:
Two rv’s X and Y are perfectly correlated with each other if Y = a X where a is a constant. This implies
Hence, the correlation coefficient ρ varies between −1 and +1. Note that when X and Y are statistically independent, they are also uncorrelated. If X and Y are uncorrelated, then (F.23) holds but they are not necessarily statistically independent.
Two rv’s are said to be orthogonal when
Hence, X and Y are orthogonal when they are uncorrelated, and mX and/or mY is equal to zero.
The conditional pdf fX|Y (x|y) of the rv X for a given deterministic value y of the rv Y has the following properties:
The marginal pdf of X may be determined from its conditional pdf as follows:
Generalization of the above to more than two rv’s is straightforward.
MGF of a rv X, which is defined as
reduces to the Fourier transform of fX(x) for s = j2πf. Therefore, the MGF may be used to exploit the advantages of the Fourier analysis.
For example, when the pdf of the sum of n independent rv’s is required, one first determines the MGF of the sum by multipling the MGF’s of the individual rv’s by using the linearity of (F.34):
where i.i.d. stands for independent and identically distributed. The next step would be to determine the resulting pdf by taking the inverse of the MGF thus found in (F.35). However, it may not always be easy and/or not necessary to take the inverse, since some desired performance parameters, such as bit error probability, can be obtained directly from the MGF.
MGF of Y = g(X) may be obtained as
The moments of a rv X can be directly obtained from its MGF:
As an alternative to the MGF, the characteristic function of a rv X is defined by
One may easily observe that (F.34) and (F.38) are the same if s = jv.
In some applications, one may need to determine the pdf of a rv Y defined as a function of another rv X, that is, Y = g(X). If the mapping Y = g(X) from X to Y is one‐to‐one, then the determination of fY(y) is straightforward. However, when the mapping is not one‐to‐one, fY(y) is determined by using all roots of the function Y = g(X). The mean value of g(X) is given by (F.7).
Let Z and W denote two functions of rv’s X and Y:
If, for all z and w, z = g(x,y) and w = h(x,y), have a finite number of solutions {xi,yi}, and the determinant of the Jacobian matrix,
is nonzero, then the pdf fZ,W(w,z) is given by
Consider N rv’s Xk, k = 1,2,…, N, each with cdf F(x). These rv’s are ordered as follows:
The cdf of X(k) is denoted by F(k)(x). The cdf of the largest rv:
The cdf of the smallest rv:
The cdf of the kth rv:
where the last expression is obtained by inserting a = 1−b = F(x) into (D.6). The binomial coefficient
denotes the number of combinations of k in n. One may easily observe that (F.85) reduces to (F.83) for k = N and to (F.84) for k = 1. The pdf of the kth rv is found as follows:
Since last two terms cancel each other, the pdf of the kth rv reduces to
The pdf of the largest rv (k = N):
The pdf of the smallest rv (k = 1):
The pdf’s f(N)(x) and f(1)(x) reduce to (F.72) and (F.75), respectively, for N = 2 when X and Y are i.i.d. Figure F.9 shows the pdf and the cdf of kth rv for N = 5.
A continuous rv X is said to be uniformly distributed in [a,b] if it is equally likely in [a,b]:
The cdf is (see Figure F.10)
The mean and the variance of a uniformly distributed rv is given by
The MGF is given by
The pdf of Gaussian (normal) distribution is defined by
where and denote respectively the mean (expected) value and the variance of the rv X. The cdf is found to be
The Gaussıan Q(x) function
represents the area under the tail of a zero‐mean and unity variance Gaussian pdf (see (B.1)). Hence, it is a monotonically decreasing function of x.
The higher order moments of a Gaussian rv are given by
where the binomial coefficient is defined by (F.86).
The MGF of X is
In many areas of telecommunications, the signal levels are measured in dB. Consequently, the mean and the variance of signals are also expressed in dB. Log‐normal distribution is a Gaussian distribution for the random signals expressed in dB. For example, in a shadowing environment, the received signal power level at a distance r from the transmitter may be written as (see (11.123))
where Pm(r) denotes the mean received power level in dB at distance r and χ in dB denotes the shadow fading with a zero dB mean and a standard deviation σ in dB. Then, the pdf of the received signal at a given distance r has a normal distribution:
As long as P(r), Pm(r) and σ are in dB, (F.102) obeys all the rules for a normal distribution given by (F.95)–(F.99). Note that the above formula is valid for a given value of the distance r since (F.101) describes a random process in r (see Chapter 1, Section 1.3.2).
The pdf and the cdf of an exponentially distributed rv X are given by
The mean and the standard deviation of X are
Using (F.34), the MGF is found to be
If X is a Gaussian rv with , then the pdf of
is given by (F.46) and is said to have a central chi‐square distribution:
The MGF corresponding to (F.107) is found from (F.34) as
The statistics of the power sums of i.i.d. Gaussian rv’s with zero‐mean and variance
is found using the MGF approach:
The pdf corresponding to (F.110) is called as the central chi‐square pdf with n degrees of freedom and is determined using the Fourier transforms in Appendix C:
The Gamma function is defined by (D.108):
The moments:
The cdf corresponding to (F.111) is given by
In wireless communications, one often encounters the sum of the powers of m rv’s with central chi‐square distribution. The power P of a complex rv, , is given by , where Xr and Xi denote, respectively, the real and the imaginary parts of X. Therefore, the pdf and cdf of the sum are obtained by inserting m = n/2 into (F.111) and analytical evaluation of (F.114) using (D.49):
Here, γ(n, x) denotes the incomplete Gamma function (see (D.106) and (D.108)):
The pdf and the cdf given by (F.115) are shown in Figure F.11 for various values of n. The pdf and the cdf were observed to shift towards higher values of z as n increases, implying the increased likelihood of observing higher values of z.
When the Gaussian rv X has non‐zero mean with , Y = X2 has a non‐central chi‐square distribution with the pdf obtained by inserting a = 1 and b = 0 into (F.45):
The MGF is found to be:
The MGF of the power sums, , as defined by (F.109), of n i.i.d. Gaussian rv’s with non‐zero means , and identical variances is found from (F.118) as follows:
where χ denotes the noncentrality parameter. The pdf of the non‐central chi‐square distribution with n degrees of freedom) is found using (F.119): [2]
where Iα(x) is defined by (D.95). The moments are given by
When m = n/2 is an integer, the cdf may be expressed in terms of the Marcum’s Q function as follows:
The Marcum’s Q function is defined by (B.15):
The PDF of the envelope R of the power sums can be obtained from (F.120) via a variable transformation, z = r2:
The moments are given by
where 1F1(a; b; z) denotes the confluent hypergeometric function defined by (D.88):
The cdf of the envelope is determined from (F.122) using the variable transformation, z = r2:
If m = n/2 is an integer
The Gamma distribution is similar to central chi‐square distribution and is characterized by the following pdf:
where α does not have to be an integer. The cdf may be expressed in terms of the incomplete Gamma function (see (F.115)):
where Γ(α) and γ(α, z) denote respectively the Gamma function and the incomplete Gamma function (see (F.112) and (F.116). The mean and the variance of the Gamma distribution are given by
The MGF is found from (F.34) to be
If α is a positive integer, then Gamma distribution reduces to Erlang distribution. For α = 1 it reduces to exponential distribution. If α = m for m = 1,2,… and the Gamma distribution reduces to central chi‐square distribution (see (F.115)) [3].
Rayleigh distribution is frequently used to characterize signals propagating through multipath fading channels. It is a special case of the central chi‐square pdf with two degrees of freedom, Z = X12 + X22, where X1 and X2 are zero‐mean statistically independent Gaussian rv’s, each with variance σ2. Z is often used to represent the power of a complex Gaussian rv, X1 + jX2. The pdf and the cdf of Z are found by inserting m = 1 into (F.115):
We define a new rv R, which denotes the envelope of this complex‐Gaussian rv:
The envelope R is characterized by the Rayleigh pdf:
The moments of the envelope are given by
The cdf of the envelope is found to be
The variation of the pdf and the cdf given by (F.135) and (F.137) are plotted in Figure F.12 for σ2 = 1.
The Rician pdf characterizes the statistics of the envelope of a narrow‐band signal with non‐zero mean corrupted by additive narrowband Gaussian noise. Rician distribution is a special case of the non‐central chi‐square pdf with n = 2 degrees of freedom, that is, Z = X12 + X22. If X1 and X2 denote statistically independent Gaussian rv’s with E[X1] = m1, E[X2] = m2, var[X1] = var[X2] = σ2, then the pdf is a non‐central chi‐square pdf with two degrees of freedom (see (F.120) and (F.122)):
The pdf of the envelope R is Rician and is given by (F.124) with n = 2:
where In(x) denotes the modified Bessel function of the first kind of order n (see Figure D.2). Note that I0(0) = 1, In(0) = 0 for n > 0 and In(x) is a monotonically increasing function of x.
The moments of R are given by
where the Rice factor K denotes the ratio of the signal power to the noise power, 2σ2. In a fading environment, the Rice factor shows the ratio of the signal power received from the line‐of‐sight (LOS) to the received signal power due to diffused scattering.
The cdf of R is obtained by inserting m = 1 into (F.128):
where the Marcum‐Q function is defined by (F.123) and (B.15).
Nakagami‐m distribution is a flexible pdf which is widely used to characterize signals propagating through multipath fading channels. The pdf and the cdf of the envelope of a signal with Nakagami‐m distribution are given by
where m denotes the so‐called fading figure (parameter). Figure F.13 shows the variation of the pdf and cdf of the Nakagami‐m distribution for various values of the fading parameter m. The Nakagami‐m distribution reduces to one‐sided Gaussian distribution for m = 1/2 and Rayleigh distribution for m = 1. In the limiting case as m → ∞, (F.144) reduces to a delta function, hence representing a deterministic signal:
The moments of the envelope are found to be
The PDF of Y = R2 is obtained by a variable transformation:
The cdf of Y is found using (D.49):
Comparing with (F.115), one may observe that the power of a Nakagami‐m distributed rv has a central chi‐square distribution. The MGF is given by
Let the rv k denote the number of events during a time interval where t1 and t2 are arbitrary times with t2 ≥ t1. The events may represent telephone calls, e‐mails, accidents, earthquakes, departure/arrival of airplanes, deposits/withdraws from an account, births/deaths, arrivals of customers in a bank/supermarket etc. The events are assumed to be independent. The average number of events per unit time is defined by the arrival rate λ in arrivals/s.
The probability of k events during a time interval τ is given by
For example, the probability that no customers arrive to a bank during τ = 1 minutes is given by if one customer arrives on the average per 10 minutes, that is, λ = 0.1 arrivals/minute. Also note .
The pdf of the number of events during τ is given by
The corresponding cdf is found, by integrating (F.154):
If the time intervals of any two events do not overlap, then the corresponding rv’s are independent. [3][4] The mean value and the variance of k in [0, t] is then given by
Using (F.34), the MGF is found to be
The probability that k out of n events occur with a probability p and n−k events occur with a probability 1−p is given by
where 0 < p < 1. The binomial coefficient, defined by (F.86), denotes the number of combinations of k events within n. For example, in an experiment, we flip a coin n = 3 times where p denotes the probability of head (H) and 1−p is the probability of tail (T). Heads can occur k = 0,1,2 or 3 times in n = 3 trials. For k = 2, one has combinations of heads, namely, HHT, HTH, THH.
The binomial pdf is defined by
Integrating (F.171), one gets the cdf of the binomial distribution
where u(x) denotes the unit step function. The mean the variance are given by [3]
The characteristic function is