Mcnaeeee Three Noise and Distortion in Microwave Systems The effect of noise is one of the most important considerations when evaluating the perfor- mance of wireless systems because noise ultimately determines the threshold for the minimum signal level that can be reliably detected by a receiver. Noise is a random process associated with a variety of sources, including thermal noise generated by RF components and devices, noise generated by the atmosphere and interstellar radiation, and man-made interference. Noise is omnipresent in RF and microwave systems, with noise power being introduced through the receive antenna from the external environment, as well as generated internally by the receiver circuitry. In our later study of modulation methods, we will see that parameters such as signal- to-noise ratio, bit error rates, dynamic range, and the minimum detectable signal level are all directly dependent on noise effects. Our objective in this chapter is to present a quantitative overview of noise and its character- ization in RF and microwave systems. Since noise is a random process, we begin with a review of random variables and associated techniques for the mathematical treatment of noise and its effects. Next we discuss the physical basis and a model for thermal noise sources, followed by an application to basic threshold detection of binary signals in the presence of noise. The noise power generated by passive and active RF components and devices can be characterized equivalently by either noise temperature or noise figure, and these parameters are discussed in Section 3.4, followed by the propagation and accumulation of noise power through a cascade of two-port networks. A more detailed treatment of the noise figure of general passive networks is given in Section 3.5. Finally, we consider the problem of dynamic range and signal distortion in general nonlinear systems. These effects are important for large signal levels in mixers and amplifiers, and can thus be viewed as complementary to the effect of noise, which is an issue for small signal levels. 688.1 Review of Random Processes 69 REVIEW OF RANDOM PROCESSES In this section we review some basic principles, definitions, and techniques of random processes that we will need in our study of noise and its effects in wireless communications systems. These include basic probability, random variables, probability density functions, cumulative distribution functions, autocorrelation, power spectral density, and expected values. We assume the reader has had a beginning course in random variables, and so will not require a full exposition of the subject. References [1]-[3] should be useful for a more thorongh discussion of the required concepts Probability and Random Variables Probability is the likelihood of the occurrence of a particular event, and is written as P (event). The probability of an event is a numerical value between zero and unity, where zero implies the event will never occur, and unity implies the event will always occur Probability events may include the occurrence of an equality, such as P(r = 5), of events related to a range of values, such as P(x < 5} In contrast to the actual terminology, a random variable is neither random nora variable, but is a function that maps sample values from a random event or process into real numbers. Random variables may be used for both discrete and continuous processes. Examples of discrete processes include tossing coins and dice, counting pedestrians crossing a street, and the occurrence of errors in the transmission of data. Continuous random variables can be used for modeling smoothly varying real quantities such as temperature, noise voltage, and received signal amplitude or phase. We will be primarily concerned with continuous random variables. Consider a continuous random variable X, representing a random process with real continuous sample values x, where —co < x < oo. Since the random variable X may assume any one of an uncountably infinite number of values, the probability that X is, ‘exactly equal to a specific value, x9, must be zero. Thus, P{X = xo} = 0. On the other hand, the probability that X is less than a specific value of x may be greater than zero: 0 < P(X < x} < L. In the limit as x — 00, P(X = xo} — 1, as the event becomes a certainty. We will not adopt any particular notation for random variables in this book, as it should be clear from the context which variables are random and which are deterministic. In most cases, the only random variables we will encounter will be associated with noise voltages, and typically denoted as va(*), or m(t) The Cumulative Distribution Function ‘The cumulative distribution function (CDF), F(x), of the random variable X is defined as the probability that X is less than or equal to a particular value, x. Thus Fy(x) = P(X Sx). G1) It can be shown that the cumulative distribution function satisfies the following prop- erties (1) Fx@) 20 (3.2a) (2) Fx(e0) (3.2b) (3) Fx(-20) =0 (32c) (4) Px(n) S Feo) ifr so (3.24)70 Chapter 3: Noise and Distortion in Microwave Systems ‘The last property is a statement that the cumulative distribution function is a monotonic (nondecreasing) function. The definition in (3.1) shows that the result of (3.2¢) is equivalent to the statement that Play gon) P(x =). (.15a) tt J=EV= For the case of a continuous random variable this becomes 20) = ete = fee faanax 6.156) ‘The result of (3.15b) can be used to find higher-order statistical averages, such as the nth ‘moment of the random variable, X: ? zor) = [ x" f.Ga)de G16) ‘The variance, 0, of the random variable X is found by calculating the second moment of X after subtracting the mean of X: te 9= [oa hends = E(x? — 2k +37 GD The root-mean-square (rms) value of the distribution is o, the square root of the variance. If 4 particular zero-mean random voltage is represented by the random variable x, the power delivered to a 1 @ load by this voltage source will be equal to the variance of x ‘The expected value of a function of two random variables involves the joint PDF: BED = lets =f e6x y0fila, ddedy. G.18) This result can be applied to the product of two random variables, x and y, by letting the function g(x, y) = xy. For the special case of independent random variables the joint PDF is the product of the individual PDFs by (3.11), so (3.18) reduces to EXxy] [sherds [™ sordy = Bnet) G9) Autocorrelation and Power Spectral Density An important characteristic of both deterministic and random signals is how rapidly their sample values vary with time. This characteristic can be quantified with the autocor- relation function, defined for a complex deterministic signal, x(1), as the time average of the product of the conjugate of x(t) and a time-shifted version, x(t + 1): roy [ xOx(t + r)at. 620) ‘It can be shown that R(0) > R(r), and R(t) = R(—r). Also, R(O) is the normalized energy of the signal, For stationary random processes, such as noise processes, the autocorrelation function is defined as R(x) = Ele"(Ox(t +O} (3.21) Because of the relation between the time variation of a signal and its frequency spectrum, we can also characterize the variation of random signals by examining the spectra of the9.1. Review of Random Processes 73 autocorrelation function in the frequency domain. For stationary random processes, the power spectral density (PSD), 5(w), is defined as the Fourier transform of the autocorrelation. function: Slo) = f Rye" dr. (3.224) ‘The inverse transform can be used to find the autocorrelation from a known PSD: 1 = Rr) = x! Swe!" dea (3.22b) It buco For a noise voltage, the power spectral density represents the noise power density in the spectral (frequency) domain, assuming a 1 & load resistor. If u(t) represents the noise voltage, the power delivered to a 1 © load can be found as PL = VO) = lv} = RO) = L Siw)do = f SiQn frag W, G23) where 5,(o) is the PSD of v(t). The last equality follows from a change of variable with @ = 2nf. Writing this integral in terms of (in Hz) is convenient because $,(«) has dimension W/Hz, and therefore appears as a power density relative to frequency in Hertz. EXAMPLE 3.1 OPERATIONS WITH RANDOM VARIABLES: ») Consider a sinusoidal voltage source, Vo cos yt, which is randomly sampled in time to form a random process v(?)= Vo cos 6, where 0 = wot is a random variable representing the sample time. Assume @ is uniformly distributed over the interval 0.<0 <2r, since the cosine function is periodic with period 2. Find the mean of the sample voltages, the average power delivered to. I 2 load, the autocorrelation function of the random process v(t), and the power spectral density, Solution ‘The PDF for the random variable @ is fo(@) = 3, for 0 <0 < 2. Then we can calculate the average voltage as _ a Lp WO = Evo) f (0) fol0) d0 = zf cos d0 =0. 5 ax Jy The average power delivered to a 1 $2 load is given by the variance of u(r): P.=80= Ew) =[{" (8) fol) = w f cos? 6d = Me I, an ty 2 ‘The autocorrelation can be calculated using (3.21): Rel) Elu(t)u(t + 1) = Vg Elcos wot cos wo(t + t)} ve pe v2 pe = =f cos 9 cos(8 + wnt) dd = =f [cos wz + cos(20 + wor)] do 2n Jy ax bo v2 oy cos wor Note that R,(0) = V2/2, which is the variance of v(t). The power spectral density is found using (3.22a): = ie store [ator mara 'f [7 feta reteonte ° = Pion) +8 +00)74 3.2 Chapter 3: Noise and Distortion in Microwave Systems This result shows that power is concentrated at w= cng and its image at ~ap. The ‘otal power can also be calculated by integrating the PSD over frequency, as in 3.23) 2 =a Sw + ap)idw = 2 e an n= 5 [_ Swdo= FE [000 - a This result agrees with the earlier result obtained as the variance using the PDF. ° — THERMAL NOISE Thermal noise, also known as Nyquist, or Jobnson,, noise, is caused by the random motion of charge carriers, and is the most prevalent type of noise encountered in RF and microwave systems. Thermal noise is generated in any passive cixcuit element that contains Joss, such as resistors, lossy transmission lines, and other lossy components. It can also be {generated by atmospheric attenuation and interstellar background radiation, which similarly involve random motion of thermally excited charges. Other sources of noise include shot noise, due to the random motion of charge cariers in electron tubes and solid-state devices, flicker noise, also occurring in solid-sate devices and vacuum tubes; plasma noise, caused by random motions of charged particles in an ionized gas or sparking electrical contacts; and quantum noise, resulting from the quantized nature of charge carriers and photons. Although these other types of noise differ from thermal noise in terms of their origin, their characteristics are similar enough that they can generally be treated in the same way as thermal noise. Noise Voltage and Power Figure 3.la shows a resistor of value R at temperature T degrees Kelvin (K). The electrons in the resistor are in random motion, with a kinetic energy that is proportional to the temperature, 7. These random motions produce small random voltage fluctuations across the terminals of the resistor, as illustrated in Figure 3.1b. The mean value of this voltage is zero, but its nonzero rms value in a narrow frequency bandwidth B is given by V_ = VaRTBR, (3.24) where k= 1.380 x 10 J/K is Boltzmann’s constant T is the temperature, in degrees Kelvin (K) B is the bandwidth, in Hz, Ris the resistance, in Q = ) 1 R} ut od FLL @ © FIGURE 3.4 (a) A resistor at temperature T produces the noise voltage v,(t). (b) The random hoise voltage generated by a resistor at temperature T.3.2 Thermal Noise 75 bandpass ter FIGURE 3.2 (2) The Thevenin equivalent circuit for anoisy esistor (b) Maximum power transfer — ‘of noise power from a noisy resistor to a load over a bandwidth B. ‘The result in (3.24) is known as the Rayleigh-Jeans approximation, and is valid for fre- quencies up through the microwave band [4] ‘The noisy resistor can be modeled using a Thevenin equivalent circuit as an ideal (noiseless) resistor with a voltage generator to represent the noise voltage, as shown in Figure 3.2a. The available noise power is defined as the maximum power that can be delivered from the noise source to a load resistor. As shown in Figure 3.2b, maximum power transfer occurs when the load is conjugately matched to the source. Then the available noise power can be calculated as vu\r1 _ v2 MeV lve 2 Pn a) tar deel 25) where Vp isthe rmsnoise voltage of the resistor. This is a fundamental result that isuseful ina ‘wide variety of problems involving noise. Note that the noise power decreases as the system bandwidth decrease. This implies that systems with smaller bandwidths collect less noise power. Also note that noise power decreases as temperature decreases, which implies that internally generated noise effects can be reduced by cooling a system to low temperatures. Finally, note that the noise power of (3.25) depends on absolute bandwidth, but not on the center frequency of the band. Since thermal noise power is independent of frequency, itis referred to as white noise, because of the analogy with white light and its makeup of all other visible light frequencies. It has been found experimentally, and verified by quantum mechanics, that thermal noise is independent of frequency for 0 < f < 1000 GHz. ‘The noise power of (3.25) can also be represented in terms of the power spectral density according to (3.23). Since the power given by (3.25) is independent of frequency, the power spectral density must also be independent of frequency, and so we have that, Py _ kT _ nto 2° 2 This is known as the two-sided power spectral density of thermal noise, meaning that the frequency range from —B to B (Hz) is included in the integration of (3.23). This is the conventional definition as used in communication systems work. The notation defined in (3.26), where no/2=KT /2 is the two-sided power spectral density for white noise, will be used throughout this book. (Note that mg is a constant, with the subscript ‘zero’. This should not be confused with the notation no(t), which we will often use to denote a noise output signal. The subscript for ths latter notation is ‘oh’, and will always be written as a function of time.) Since the power spectral density of thermal noise is constant with frequency, its auto- correlation must be a delta function according to (3.22b): Salo) = (3.26) 1 fn Ot Sop 2 By the central limit theorem, the probability density function of white noise is gaussian RO no. Bsr 3, ol 27)76 Chapter 3: Noise and Distortion in Microwave Systems re ev IR i I | hth Ra. / \ wy / FIGURE33 Circuit for Example 32 with zero mean: (3.28) where 6? is the variance of the gaussian noise. Thermal noise having a zero mean gaussian PDF is known as white gaussian noise. Since the variance of the sum of two independent random variables is the sum of the individual variances (see Problem 3.5), and the variance is equivalent to power delivered to a 1 © load, the noise powers generated by two independent noise sources add in a common load, This is in contrast to the case of deterministic sources, where voltages add. Note that (3.27) is not completely consistent with (3.28), since (3.27) indicates that R(0), the variance of white noise, is infinite while (3.28) implies a finite variance. This problem arises because of the mathematical assumption that white noise has a constant power spectral density, and therefore infinite power. In fact, as we discussed earlier, thermal noise has a constant PSD only over a finite, but very wide, frequency band. We can resolve this issue if we understand our use of the concept of white noise to actually mean a bandlimited PSD having a finite frequency range, but broader than the system bandwidth with which we are working, ‘Twonoisy resistors, Ry and Ro, at temperature T, are shown in Figure 3.3. Calculate the available noise power from these sources by considering the individual noise power from each resistor separately. Next, consider the resistors as equivalent to a single resistor of value Ry + R, and verify that the same available noise power is obtained. Assume a bandwidth B for the system. >) aNmEE_catcuamNoRSoNE re -») Solution ‘The equivalent noise voltage from each resistor is found from (3.24): Vor = V4RTBR, Veo = V4RTBR, For maximum power transfer, the load resistance should be Ry + Rp. Then the noise power delivered to the load from each noise source is Pp -( mo KTBRy UND) RR RR (2 L KBR: 2) RFR RFR Px3.3 3.3 Noise In Linear Systems 77 So the total available noise power is Py = Pat + Ppa = HTB. Considering the two resistors as a single resistor of value Ri + R2, with a load resistance of Ry + Ro, gives an available noise power of P, = kTB, in agreement with the first result ° NOISE IN LINEAR SYSTEMS In a wireless radio receiver, both desired signals and undesired noise pass through various stages, such as RF amplifiers, filters, and mixers. These functions generally alter the statistical properties of the noise, and so it is useful to study these effects by considering the general case of transmission of noise through a linear system. We then consider some important special cases, such as filters and integrators, and the nonlinear situation where noise undergoes frequency conversion by mixing. Autocorrelation and Power Spectral Density in Linear Systems In the case of deterministic signals, we can find the response of a linear time-invariant system to an input excitation in the time domain by using convolution with the impulse response of the system, or in the frequency domain by using the transfer function of the system, Similar results apply to wide-sense stationary random processes, in terms of either the autocorrelation function or the power spectral density. Consider the linear time-invariant system shown in Figure 3.4, where the input random, process, x(t), has an autocorrelation R,(c) and power spectral density 5,(«), and the output random process, y(t), has an autocorrelation R,(r) and power spectral density S,(«). Ifthe impulse response of the system is k(t), we can calculate the output response as w= J more —wau, (3290) Similarly, a time-shifted version of y(t) is y(t+r) f. A(v)x(t +t — v)dv. (3.29b) So the autocorrelation of y(t) can be found as Rye) = ElyOy(e + of" f A(u)h(o)E (x(t ~ w)x(t +r — v)} dudv f f A(u)h(w)R,(e + u — v)dudv. (3.30) mo L_2) “Sey” [Hwy | Rye, Stoo FIGURE 3.4 A linear system with an impulse response h(t) and transfer function H(o). The input is a random process x(, having autocorrelation R,(c) and PSD S,(w). The output random process is y(0, having autocorrelation R,(t) and PSD S,(0).78 Chapter 3: Noise and Distortion in Microwave Systems AC) SS ~~ x Ee xz AG) HN) HD) ef f A f Lewpes Harpe aandpa FIGURE 3.5 System symbols and frequency responses for low-pass, high-pass, and band-pass filters. ‘This result shows that the autocorrelation of the output is given by the double convolution of the autocorrelation of the input with the impulse response: R,(t) = h(t) @ h(—t) ® y(t), We can derive the equivalent result in terms of power spectral density by taking the Fourier transform of both sides of (3.30), in view of (3.22a): £ Rya)e I" dt = f. f. A(udh(v) f RAt + u— vyeM dr du dv. Now perform a change of variable toa = r +u ~ v, so that da = dr. Then we obtain f- Rye ide = f hue f. heuer ft Re(@eP™ dadudy. Since H(w) is the Fourier transform of h(¢) Ho) = f hear, G31) the above simplifies to the following important result: Sy(w) = |H(@)PS.(@). 3.32) We will now demonstrate the utility of these results with several applications. Gaussian White Noise through an Ideal Low-pass Filter As we will see in Chapters 5,9, and 10, filters play an important role in wireless receivers and transmitters. The main function ofa filter is to provide frequency selectivity, by allowing a certain range of frequencies to pass, while blocking other frequencies. Figure 3.5 shows the symbols and associated idealized frequency responses for low-pass, high-pass, and bandpass filters. Here we examine the effect of an ideal low-pass filter on noise. Figure 3.6 shows white noise passing through a low-pass filter. The filter has a transfer function, H(f), as shown, with a cutoff frequeney of Af. Note that the transfer function is defined for both positive and negative frequency, since we will be using the two-sided power spectral density. Our usual notation will be to use lowercase letters, such as m(t) and. A) 10a f FIGURE 3.6 White noise passing through an ideal low-pass filter, and the transfer function of the filter,8.3 Noise In Linear Systems 79 no(t), for noise and signal voltages in the time domain, and capital letters, such as N; and Noy for average powers of noise and signals, Since the input noise is white, the two-sided power spectral density of the input noise is constant, as given in (3.26) Su(f) = calf). (3.33) Then from (3.32) the output power spectral density is given by 20 for > 2 forifl< af Su A= lAPPSa (A) =} 2 (3.34) 0 forlfl> af ‘The output noise power is then No = QAf)Sn(f) = Afro. (3.35) We see that the output noise power is proportional to the filter bandwidth Gaussian White Noise through an Ideal Integrator ‘As we will see in Chapter 9, integrators are critical components for the detection and demodulation of digital signals. Here we derive an expression for the output noise power from an integrator with white noise input; this result will be used later for the derivation of error probabilities for digital modulation in Chapter 9. Figure 3.7 shows a noise signal, n,(t), applied to the input of an ideal integrator. The ‘output noise signal is no(¢). The output of the integrator is the value of the integral, at time 1 = T, of the input signal. We need to find the average power of the output noise. The transfer function of the integration operation is, Ho) (3.36) where 7 is the integration interval time. Evaluating the magnitude squared of (3.36) gives (=e? ~ eT) _ 2-2eoswT IH(@)P = H@)H*(o} i z ” ° (3.37) seheT/A «sited _ yo(snnst o che aft )* since w = 2nf. If we assume white noise at the input, with S,,(w) = no/2, then the output noise power can be calculated using (3.23) and (3.32) to give no re pytgp = Mot? [ (sinafT\? No [Bian df= - ( yar a 2 dar 2 no [ (sinx nol fa [-( ) dea (3.38) (0) T] nett SOO" Fe FIGURE 3.7 White noise passing through an ideal integrator.80 Chapter 3: Noise and Distortion in Microwave Systems 0) cos(wyt +8) FIGURE 38 — White noise passing through 2 mixer with a local oscillator signal, cos(ot +8) ‘The integral was evaluated by using a change of variables, x= fT, with dx = xTdf, and a standard integral listed in Appendix B. Mixing of Noise One of the common functions of a receiver is to perform frequency conversion, by mixing a signal with a local oscillator to shift the original signal spectrum up or down in frequency. When noise coexists with the signal, the noise spectrum will also be shifted in frequency. While we will study mixers in detail in Chapter 7, here we idealize the function of mixing by considering it as a process of multiplication of the input signal by a local oscillator signal, as shown in Figure 3.8. We wish to find the average noise power of the output signal. ‘We assume that n(t) is a bandlimited white gaussian noise signal with variance, or average power, 0? = E{n"(t)}. The local oscillator signal is given by cos(oogt +4), where the phase, 6, is a random variable uniformly distributed on the interval 0 < @ <2, and is independent of n(t). The output of the idealized mixer is v(t) = n(t) cos (wot + 8). 339) ‘The average output power from the mixer can then be calculated as the variance of v() No = E{v(t)) = E{n?(t)cos’(wot + 0) = Eln?()) E(cos"(wot + 0)) op, 2 or (pt +8) dO = — 40) Fo [costes +040 = 5 G40) ‘This result shows that mixing reduces the average noise power by half. In this case, the factor of one-half is due to the ensemble averaging of the cos? wf term over the range of random phase, If we now consider a deterministic local oscillator signal of the form cos cxot (without a random phase), the variance of the output signal becomes E(v%(O) = E(n(0) cos? wot} = 0? cos? a G4L) The last result follows because the cos? wot term is unaffected by the expected value operator, since itis no longer a random variable. (In addition, v(t) is no longer stationary, and therefore does not have an autocorrelation function or power spectral density.) The variance of the output signal is now @ function of time, and represents the instantaneous power of the output signal. To find the time-average output power, we must take the time- average of the variance found in (3.41): 7 afm m=z f evoa= 2 f an hy since T =1/f =2z/ap. We see that the same average output power is obtained whether the averaging is over the ensemble phase variation, or over time. 0s" aptdt =F, 3.42)