CHAPTERTEN NOISE 10.1 INTRODUCTION The sensitivity of communications systems is limited by noise. The broadest def- inition of noise as “everything except the desired signal” is most emphatically not what we will use here, however, because it does not separate, say, artificial noise sources (¢.g., 60-Hz power-line hum) from more fundamental (and therefore irre~ ducible) sources of noise that we discuss in this chapter. ‘That these fundamental noise sources exist was widely appreciated only after the invention of the vacuum tube amplifier, when engineers finally had access to enough gain to make these noise sources noticeable. It became obvious that simply cascading more amplifiers eventually produces no further improvement in sensitivity because a mysterious noise exists that is amplified along with the signal. In audio systems, this, noise is recognizable as a continuous hiss while, in video, the noise manifests itself as the characteristic “snow” of analog TV systems, The noise sources remained mysterious until H. Nyquist, J. B. Johnson and W. Schottky! published a series of papers that explained where the noise comes from and how much of it to expect. We now turn to an examination of the noise sources they identified. 10.2 THERMAL NOISE Johnson? was the first to report careful measurements of noise in resistors, and his colleague Nyquist? explained them as a consequence of Brownian motion: thermally ' This name is frequently misspelled in English-language pul \. IL is not all that uncommon to see "Shotkey;” “Shottkey,” or “Schottkey:” While we're at it, “Schmitt,” as in the trigger, is also misspelled quite often, with common incorrect renderings being “Shmit” and “Schmidt.” 2 “Thermal Agitation of Electricity in Conductors,” Phys. Rev.. V. 32, July 1928, pp. 97-109, 3 Thermal Agitation of Electric Charge in Conductors,” Phys. Rev. v. 32, July 1928, pp. 110-13. 243,24d CHAPTER 10 NOISE agitated charge carriers in a conductor constitute a randomly varying current that gives rise to a random voltage (via Ohm’s law). In honor of these fellows, thermal noise is often called Johnson noise or, less frequently, Nyquist noise. Because the noise process is random, one cannot identify a specific value of volt- age at a particular time (in fact, the amplitude has a Gaussian distribution), and the only recourse is to characterize the noise with statistical measures, such as the mean- square or root-mean-square values, Because of the thermal origin, we would expect a dependence on the absolute temperature. It turns out that thermal noise power is exactly proportional to 7 (the astute might even guess that it is proportional to AT), Specifically, a quantity called the available noise power is given by Pwa =kTAf, dy where & is Boltzmann’s constant (about 1.38 x 10-% J/K), 7’ is the absolute temper- ature in kelvin, and Af is the noise bandwidth in hertz (equivalent brickwall band- width) over which the measurement is made. We will clarify shortly what is meant by the terms “available noise power” and “noise bandwidth,” but for now simply note that the noise source is very broadband (infinitely so, in fact, in the simplified picture presented here*), so that the total noise power depends on the measurement bandwidth. Note that this equation tells us that the available noise power has a spectral density that is independent of frequency, and that the total power thus grows with bandwidth without limit. This is a bit of a lie, but it is true enough for all bandwidths of interest to the electrical engineer. With Eqn. 1, we can compute that the available noise power over a 1-Hz. band- width is about 4 x 10°?! W (or —174 dBm)‘ at room temperature. Further note that the constancy of the noise density implies that the thermal noise power is the same over any given absolute bandwidth, Therefore the noise power in the interval between | MHz and 2 MHz is the same as between 1 GHz and 1.001 GHz. Be- cause of this constancy, thermal noise is often described as “white,” by analogy with white light. However, the analogy is not exact, since white light consists of constant energy per wavelength whereas white noise has constant energy per herts. ‘The term “available noise power” is simply the maximum power that can be deliv- ered to a load. Recall that the condition for maximum power transfer (for a resistive network) is equality of the load and source resistances. This suggests the use of the network shown in Figure 10.1 to compute the available noise power. + The implication that the available power grows without bound as the bandwidth approaches infinity should hint at the need to modify this formula, We will take care of this detail shortly. 5 Recall that the reference level for 0 dBm is one milliwatt10.2. THERMAL NOISE 245 FIGURE 10.1. Network for computing the thermal noise of a resistor. ‘The model of the noisy resistor is enclosed within the dashed box, and is here shown as a noise voltage generator in series with the resistor itself. The power de- livered by this noisy resistor to another resistor of equal value is by definition the available noise power: 2 Pra =kTAS = —, (2) wa = RTS = 3 2) where @ is the open-circuit rms noise voltage generated by the resistor R over the bandwidth Af ata given temperature. The mean-square open-circuit noise voltage is therefore e2 =AKTRA. 8) ‘A couple of useful rules of thumb emerge from plugging some numbers into this last equation. Atroom temperature, a 1-k@ resistor generates about 4 nV of rms noise over a bandwidth of 1 Hz, while a 50-@ resistor generates a bit over 0.9 nV (some people just call it 1 nV, since it’s easier to remember) of rms noise over that same bandwidth. These values are perhaps among the very few worth committing to mem- ‘ory. Just remember that the rms voltage is proportional to the square root of the band- \width (and resistance) when scaling these numbers to any bandwidth or resistance. Tn many cases, the noise is specified in terms of the spectral density rather than total value, and is found simply by dividing the mean-square noise value by Af (or the rms noise value by the square root of Af). Thus, a 1-k&2 resistor has an rms noise spectral density of about 4 nV/V/Hz (funny units, yes) or a mean-square noise density of approximately 1.6 x 10~'” V?/ Hz. For thermal noise, the spectral density is a constant that depends only on temperature (and Boltzmann’s constant), and is independent of frequency.° In the more general case, the density may vary with fre- quency, and one then uses the term “spot noise density” to underscore that the stated density applies only at some specified spot in the spectrum. We've seen that every physical resistor has a noise source associated with it. In the Thévenin representation of Eqn. 1, there is a voltage source in series with the re- - sistor, Alternatively, we may construct a Norton equivalent model in which a noise * See footnote 4, however.