Delivering Value Through Innovation Noise and Noise Parameters Introduction The subject of noise is vast and volumes have been devoted to it. Therefore, it is important that we restrict our coverage of the subject and concentrate on the more essential basics. The goal of this article is to present a clear picture of important measurement parameters that relate to random noise. This article will concentrate to the study of random noise of the types generated by resistors, diodes, active components etc. This is essential for gaining good understanding of noise measures such as “Noise Factor”. We will concentrate only on the properties of random noise and its measurement. Only the amplitude characteristics and measures of noise will be considered in this article. We will not consider noise types such as impulsive interference and unwanted CW radiation from signal sources, nor will we cover the processing of noise by electronic circuits, such as the way in which an FM system improves Signal/Noise Ratio. The subject of RF carrier random Phase Jitter and its related parameters will be covered in Frequency Stability and Frequency Stability Parameters, later in this series of articles. The Nature of Random Noise In simplistic terms, random noise in an electronic circuit is any voltage or current whose instantaneous value is unpredictable. And yet, some characteristics of random noise can be predicted to some degree. Consider pure thermal noise as generated by an ideal resistor. By looking at the longterm statistical behavior of this noise, we find that there are some predictable characteristics. For instance, the long-term mean value is small compared to the largest instantaneous value and with time it is asymptotic to zero. This is a predictable characteristic of pure thermal noise. The long-term r.m.s. value of the noise is asymptotic to a finite value. This is another predictable characteristic of pure thermal noise. Now consider a symmetrical band of possible instantaneous noise values that is centered on the zero long-term mean value and has a width equal to twice the longterm r.m.s. value. © 2011 LTX-Credence Corporation All rights reserved. All trademarks are the property of their respective owners www.ltxc.com 1 Noise and Noise Parameters Counter-intuitively, in the long-term, the instantaneous value of the noise will lay in this band for 68% of the time. Similarly, for a band twice this wide the value will be in this band for 95% of the time. The value for three times is 99.5%. This predictable statistical characteristic of pure thermal noise is called a Gaussian distribution, and noise of this type is called “Gaussian noise”. Pure thermal noise has no correlation with itself. If we continuously compare the waveform for the noise with that at a fixed earlier time, then there will be no common characteristics. We have already said that random noise is unpredictable. However, we have also said that random noise has predictable characteristics. In fact, random (unpredictable) noise can have correlation with itself. This form of correlation is known as “auto-correlation” and is yet another characteristic of pure thermal noise is that it has no correlation with itself. HINT: If the RF engineer has problems with understanding auto-correlation, then the correlation of a sinusoidal signal with itself is well worth considering. Yet another predictable characteristic of pure thermal noise exists. We must look to the frequency domain representation of the noise. Clearly, random noise is not periodic. Another way to look at this is to consider that its period is infinite. This is tantamount to saying that the Fourier components of the noise are contiguous: the frequency spectrum is continuous. Furthermore, this spectrum is uniform to frequencies of around 10,000GHz, after which it falls off as predicted by quantum theory. If the long-term r.m.s. noise value is measured by an instrument with a fixed bandwidth, then the same value would be obtained at all frequencies in this band. Noise with this characteristic is called “white noise” by analogy with white light, which has a “flat” response for optical frequencies. Given this analogy, pure thermal noise is called “Gaussian white noise”. In summary, we have described five predictable characteristic of pure thermal noise whose instantaneous value is itself unpredictable. Up to this point we have only considered pure thermal noise which, from an electronic system viewpoint, can be considered to have an infinite frequency spectrum. Real systems have finite bandwidths and therefore, we must consider what happens when we pass Gaussian white noise through such systems. The type of noise that appears at the output of these system is called “Colored noise”. What happens when we pass Gaussian white noise through a low-pass filter? The output noise will be colored, that is, it will be band-limited. However, it is surprising to discover that the output noise is correlated. The action of the filter is to impose correlation on uncorrelated noise. This fact is easier to understand when we consider what happens when we pass Gaussian white noise through a bandpass filter. If the bandpass filter has a center frequency of f and a bandwidth of 1Hz, then we would expect to see a sinusoidal signal of frequency f at the output of the filter. In reality the amplitude and phase of the signal change randomly at a rate determined by the filter bandwidth. For very small bandwidths, the rate of change will be small and, over a www.ltxc.com 2 LTX-Credence Delivering Value Through Innovation Noise and Noise Parameters relatively short period, the filter output approximates to a pure sine wave. A pure sinusoidal signal has correlation with itself: the smaller the filter bandwidth, the greater the correlation. The amplitude of the filter output varies with what is known as a “Rayleigh distribution”. Noise Bandwidth and Noise Density Consider again pure thermal noise as generated by an ideal resistor. As the frequency spectrum for pure thermal noise is “flat“, it is to be expected that the long-term noise power observed in a given frequency band is proportional to bandwidth. The term for bandwidth, in this context, is “noise bandwidth” and is that provided by a hypothetical filter with infinitely steep response transitions between the pass-band and stop-bands. The subject of “Noise Power” discussed in greater depth in Thermal (Johnson) Noise. The term noise bandwidth can be applied not only to white noise but, also, to any form of colored noise. For colored noise, the power will not be proportional to bandwidth. There is a defined noise bandwidth for a real circuit such as a bandpass filter, even though is does not have infinitely steep transitions. The noise bandwidth of a bandpass filter is the bandwidth of an idealized hypothetical filter with identical effect on white noise power reduction. The noise power in a bandwidth of just 1Hz is specifically known as “Noise Power Spectral Density”. This is a very common measure of random noise characteristics. Some Examples of Random Noise Thermal noise (also known as “Johnson noise”, after an early researcher) has already been mentioned and results from the motion of electrons within a conductor. The degree of motion and hence, noise magnitude is dependent upon absolute temperature. Pure thermal noise is independent of frequency and has a Gaussian distribution. Shot Noise results from the random passage of individual charge carriers across a potential barrier. It was first observed in thermionic devices but is also present when current flows across a semiconductor junction. Pure shot noise is independent of frequency and has a Gaussian distribution. Partition noise is peculiar to thermionic devices and results from random fluctuations in the way in which current is divided between the various electrodes. The physical mechanisms for thermal, shot and partition noises are thought to be fully understood. When a DC current is passed through a resistor or semiconductor, noise in excess of that which is expected results. For a resistor we would expect to observe just thermal noise. For a semiconductor we would expect a combination thermal noise and shot noise, both of which are independent of frequency. Excess or “flicker noise” differs from thermal, shot and partition noise in that it is inversely proportional to frequency. For this reason it is sometimes referred to as “pink noise”, where the light analogy is again used, and it applies to any noise that has a falling frequency spectrum, irrespective of LTX-Credence Delivering Value Through Innovation www.ltxc.com 3 Noise and Noise Parameters its origins. The mechanism for excess noise is still not fully understood. The magnitude of excess noise in a resistor or semiconductor depends on the DC current. For resistors it also depends on the component type, being very small for some metal film resistors. For semiconductors, the excess noise is dependent upon device material, type and design. It can even vary between devices that are nominally identical. Thermal (Johnson) Noise A knowledge of thermal (or “Johnson”) noise is fundamental to the study of noise measurement and to the design of low noise systems and circuits. Thermal noise is the result of random electron motion in conductors. The noise from an ideal resistor is fully predictable. It is also predictable for good quality resistors that do not suffer from excess noise. It is even predictable for any resistor type over the normal range of frequencies experienced in RF engineering. Because of its predictability, pure thermal noise is used as a reference standard for noise measurement. Real circuits and device models abound with resistors and, therefore, a knowledge of thermal noise is of paramount importance in circuit design. Research has found that the r.m.s. noise voltage VN across a resistor of value R, due to electron thermal activity alone, is given by: where k = Boltzmann’s constant = 1.380662 x 10-23 JK-1 T = Absolute Temperature = 298.15K for 25°C B = Bandwidth in Hz We now invoke Thévenin’s (pronounced “tay-venin’s”) theorem to explain this phenomenon. Thévenin’s theorem states that the equivalent circuit for any two-terminal linear system consists of a voltage generator in series with a resistor. The output of the generator is equal to the open-circuit voltage of the system. The value of the resistor is equal to the resistance between the two terminals. www.ltxc.com 4 LTX-Credence Delivering Value Through Innovation Noise and Noise Parameters If we apply Thévenin’s theorem to the noisy resistor, we obtain the equivalent circuit shown in Figure 1. It is important to note that the series resistor is noiseless. If noise was present in this resistor, then it would have been included twice. Figure 1. Noise Model for a Resistor The above model can be used as a model for resistors in a circuit or device model. The noiseless resistor is “seen” by all other signal/noise voltages and currents. What happens when we connect a noiseless resistive load to a noisy resistor? Applying the noiseless model from Figure 1, we can carry out a simple analysis. The equivalent circuit for resistor and load is shown in Figure 2. Figure 2. Noise Model with Noiseless Resistive Load LTX-Credence Delivering Value Through Innovation www.ltxc.com 5 Noise and Noise Parameters The noise power delivered to the load is given by: What happens if we make RL equal to R? This is the condition for obtaining maximum available noise power PNAV from the resistor. For this condition we have: At first sight, this expression might seem amazing—the maximum available noise power PNAV is independent of resistor value R. However, it will be realized that the noise power is a result of electron mobility, which is only dependent upon absolute temperature and not on resistor physical size or electrical value. Note that the noise voltage for a resistor is, however, dependent on resistor value R. The maximum available noise power kTB is a very convenient reference standard for noise measurement. Note that when considering noise power spectral density, that is noise power in a 1Hz bandwidth, the expression for maximum available noise power simplifies to kT. Signal/Noise Ratio How do we compare the noise performances of two electronic systems? A low frequency analogue engineer’s answer to this question would probably be “by comparing both Signal/Noise Ratios”. Signal/Noise Ratio is normally written as “S/N” ratio but, strictly speaking, should be (S+N)/N ratio. However, N is always small so the two ratios are nearly equal. The low frequency analogue engineer’s answer is fine as long as the rest of the world uses the same signal levels when making comparisons between systems. Even RF engineers use S/N ratio to check the behavior of an individual system but not for comparing two systems. Noise Factor and Noise Figure Consider a two-port linear system. This can be a system where the input and output frequencies are the same, such as an amplifier. It can also be a system where the input and output frequencies are different, such as a mixer. However, linearity, over the signal levels of interest, is a prime requirement. www.ltxc.com 6 LTX-Credence Delivering Value Through Innovation Noise and Noise Parameters To get around the fundamental problems associated with a single S/N ratio, we will compare the S/N ratios at the input and output. Thus we can define a hypothetical noise measurement factor (NMF) as: where SI = input signal SO = output signal NI = input noise NO = output noise Here, the signals and noises are not defined, but all have the same dimensions. The RF engineer may have already seen that the definition of NMF is independent of signal level. For different levels of SI, SO will change pro-rata if the system is truly linear. To put this another way, the system gain G = SO/SI is a constant over the signal levels of interest. Thus we have: where G = gain As SI and SO are not defined, so then is the gain G. Thus, we have defined a hypothetical noise measurement factor NMF that is independent of signal levels. It is described in terms of NI, NO, and G, none of which have been defined. The definition of NMF forms the basis for the definitions of noise factor and Noise Figure. We will proceed to the full definitions of “noise factor” and “noise figure”, that includes those systems having any input impedance. LTX-Credence Delivering Value Through Innovation www.ltxc.com 7 Noise and Noise Parameters Consider the two-port system shown in Figure 3. Here, both the source and load resistors are considered to be noiseless. Figure 3. A Two-Port System Pertaining to Noise Factor and Noise Figure To covert our definition of NMF into something more meaningful, we must define the quantities SI, SO, NI and NO. We can do this in terms of voltage, current or power as long as the dimensions for each are the same. However, it turns out best if we work solely with power. The RF engineer will, probably, have already realized that the dimensions for the signals and noises should be power. After all, noise magnitude can only be defined in terms of r.m.s. voltage/current and, hence, power. Thus we have: where PSI = input signal power PSO = output signal power PNI = input noise power PNO = output noise power GP = power gain We will now introduce the full definitions for all the parameters. www.ltxc.com 8 LTX-Credence Delivering Value Through Innovation Noise and Noise Parameters Starting with the output of the system, we can simply define PSO and PNO as follows: where VSO and VNO are r.m.s. quantities PSO and PNO are just the signal and noise powers in the load resistance. In the context of this discussion we will use the concept of “maximum available source power” to define input signal level. This leads to the definition of “Transducer Power Gain”, GPT. The concept of maximum available source power and transducer power gain (GPT) must be used for both signal and noise. This is convenient as we have already defined maximum available noise power PNAV for a resistor. By defining PSAV and PNAV as the maximum available signal and noise powers at the input and renaming NMF as “Noise Factor F”, we have: where PSAV = maximum available signal power PSO = signal power in load resistance PNAV = maximum available noise power PNO = noise power in load resistance GPT = transducer power gain What interpretation can we place on this expression? LTX-Credence Delivering Value Through Innovation www.ltxc.com 9 Noise and Noise Parameters The quantity PNO/GPT is the output noise power referred to the input using transducer power gain. Noise factor is the ratio of the input referred output noise power to the available input noise power. This is a very precise and solid definition of a noise defining parameter. The value of PNAV is the defined quantity “kTB”, which is independent of resistor value. For our final definition of F, we have: For this definition to be sound, PNO must also be defined in the bandwidth B. Also, note that for a linear system, PNO is proportional to the Absolute Temperature T and, therefore, is mutually canceling with PNAV. This assumes that the complete system and source are at the same temperature. HINT: Strictly speaking, the system in Figure 3 should include a noise source for the output load resistor. This was deliberately omitted for the sake of clarity. However, the effect of the omission is normally negligible as the output noise PNO is, in most cases, much larger than the noise generated by the resistor. Demonstrative Exercise To bring home the writers point that the definition of noise factor must apply to systems having any input impedance, the following simple exercise is suggested. What is the noise factor of an amplifier having a real input resistance of magnitude R0 and no other internal noise sources? It should be pointed out that the answer is not unity. Noise Figure {NF) is simply noise factor expressed in dBs. Thus we have: NF = 10 log F Noise Factors for Cascaded Systems Consider two cascaded systems. When considering noise relative to the overall signal, the noise contribution from the second system will be small if the gain of the first is large. This because an increase in gain increases the signal plus noise from the first system, so that the noise from the second system is a relatively smaller part of the overall signal. The overall noise factor of two cascaded systems is dependent not only on the individual noise factors for each www.ltxc.com 10 LTX-Credence Delivering Value Through Innovation Noise and Noise Parameters system but also on the gain of the first system. This argument can be extended to any number of systems. For most well designed systems, the overall noise factor is arranged to be close to that of the first system. The expression for the overall noise factor (F) of a cascaded system is given here without proof.: where FN = noise factor of nth system GN = Transducer Power Gain of nth system Equivalent Noise Temperature and Equivalent Noise Resistance The RF engineer must be made aware of two commonly used noise parameters. Consider the available noise power from a system such as an antenna. The actual noise power from the antenna PA will be higher than expected from a straight calculation of thermal noise from a resistive source at absolute temperature (T). One way to express this increased noise is by assuming a hypothetical temperature for the source. Thus we have: where TE = Equivalent Noise Temperature PA = actual available noise power from the antenna The concept of “Equivalent Noise Temperature” can be extended to other noisy systems including those containing non-thermal noise sources. Equivalent Noise Resistance is another technique used to express the total noise in a system. Equivalent noise resistance (RE) of a noise generator is the value of resistance that gives the same noise voltage as the generator at the standard temperature. This definition can also be expressed in terms of noise current. Equivalent noise resistances can be introduced into a circuit for noise representational purposes. However, they must be introduced in such a way that they do not affect other signals and noises. LTX-Credence Delivering Value Through Innovation www.ltxc.com 11 Noise and Noise Parameters In Conclusion... In this article we have discussed concepts of noise parameters in general and random noise in an RF system in particular. The series continues with a discussion of phase, group, and envelop delay. LTX-Credence Corporation 825 University Avenue Norwood, MA 02062-2643. +1-781-461-1000 www.ltxc.com