Encontre seu próximo livro favorito

Torne'se membro hoje e leia gratuitamente por 30 dias.
Electronic Systems

Electronic Systems

Ler amostra

Electronic Systems

1,190 página
9 horas
Lançado em:
Aug 26, 2019


The book deepens the understanding of important concepts and elements necessary to properly design an electronic system by exploiting analog, mixed-signal and digital components. The book provides tools to analyze and develop electronic boards and systems, by focusing on:

  • noise in electronic components and circuits;
  • operational amplifier performance;
  • frequency compensation of OpAmp stages;
  • advanced INA, ISO, Current feedback/mode, and OTA amplifiers;
  • Sample&Hold sampling circuits;
  • analog mux, digital potentiometers and universal active filters;
  • standard and advanced DAC and ADC converters;
  • under- and over-sampling;
  • Sigma-Delta modulators.

Many actual circuits and exercises are provided at the end of each Chapter and also in three specific Chapters focused on examples of analog and mixed-signal electronic systems employing OpAmps, S&H, DAC and ADC converters. Most exercises are fully solved, with detailed step-by-step stage design and electronic schematics analysis. The book targets an audience interested in hardware and firmware design of electronic circuits and systems for acquisition, conditioning and conversion of analog and digital signals.
Lançado em:
Aug 26, 2019

Sobre o autor

Relacionado a Electronic Systems

Livros relacionados
Artigos relacionados

Amostra do Livro

Electronic Systems - Franco Zappa



Electronic noise

Songs from the Wood, Jethro Tull

After a brief dissertation about electronic noise and on the ways to compute power spectra and transfer functions of noise generators, several kinds of noise sources in electronics devices will be analyzed. In particular, the concepts of Noise Figure and Optimal Resistance will be introduced and studied in depth. Finally, OpAmps made by BJTs and MOSFETs will be compared.


In electronics, noise is defined as the random fluctuation associated to the deterministic amplitude of every electrical signal. When designing an analog electronic circuit (e.g. an amplifier), it is very important to determine its noise performance as, whenever this quantity became comparable or even higher than the input signal, the information brought by the latter could be corrupted or get completely lost. Noise also impacts other aspects of the electronic system, like the component selection and the signal conditioning partition among blocks.

In general, every electric signal x(t) can be described as the superposition of the useful electric signal, s(t), carrying useful information, the noise n(t), randomly fluctuating in time, plus potential electromagnetic disturbances d(t), caused by interferences with other circuits or devices. So the trend of a generic electric variable (voltage or current) follows this model:

x(t) = s(t) + n(t) + d(t)

While the source of disturbances, typically being specific circuits or components, can be identified and potentially attenuated by means of shielding and filtering, electronic noise is due to the random statistical movements of the charge carriers in the electronic devices and is therefore ineradicable. Examples of noisy electrical waveforms are shown in Fig. 1.1.

Fig. 1.1: Examples of electrical waveforms affected by electronic noise (left) and typical Gaussian distribution (right).

Being a random variable, in order to study the behavior of electronic noise it is necessary to apply statistical concepts. In most cases, noise comes as the superposition of many elementary (microscopic) processes, each one, generally, described by a Poisson stochastic process. Thus, it is possible to apply the Central Limit Theorem, which implies that the macroscopic process can be described by a Gaussian process, totally characterized by its first-and second-order moments, i.e. mean and variance. By definition, noise must always be characterized by nil mean (otherwise its mean could be considered as an offset):

In the case of a Poisson process, only one moment is required to describe it: the first order moment (i.e. the noise variance, or mean square value). The Poisson distribution describes single elementary processes (usually at the microscopic level). The Poisson distribution is:

Similarly to the procedure used for ordinary signals, also noise can be studied in frequency domain: suppose to send a noisy signal n(t) through a bandpass filter, which cuts out any frequency outside an interval with bandwidth Δf centered around a specific frequency f0, as shown in Fig. 1.2. Let us measure the mean square value of the signal obtained after the filter. Considering a very narrow Δf, the output of the bandpass filter is a sine wave at the frequency f0: the mean square value is proportional to the square of the amplitude at that harmonic component, i.e. to the electrical power at that frequency. The ratio between the mean square value and the filter bandwidth Δf is called power spectrum:

with units of measurement V²/Hz or A²/Hz depending on the nature of the noise (voltage or current).

Depending on the shape of the power spectrum, we can distinguish between white noises (when the spectrum density is constant in frequency), pink/red/brown noises (when the low-frequency components are more intense than those at higher frequencies) and blue/violet noises (when high-frequency components are stronger).

Fig. 1.2: Block diagram for the power spectrum measurement.

Fig. 1.3: Noise Transfer Function in a linear network.

The integral power spectrum is the integral of the mean square values of each harmonic components, representing the noise mean square value, i.e. the noise power:

Now, let us study the output of an electric network with Transfer Function T(jω) and a noise generator at the input, which is characterized by a power spectrum Sin(f) (Fig. 1.3). The linear network transfers at the output all the input harmonic components modifying their amplitude and phase. Thus, a generic harmonic component of the noise at frequency f is transferred to the output with the amplitude modified by |T(jω)|. Thus, if its mean square value at the input is Sin(f)df, the output has a mean square value equal to:

In conclusion, the linear network transfers the power spectrum of the input noise (voltage square or current square) to the output, amplified by the square module of the (voltage or current, respectively) transfer function. Since noise is a stochastic contribution, we are not interested in its phase.


1.2.1      Shot noise (Poisson)

Shot noise is a source of white noise, mainly present in diodes and in BJTs. It is associated to the DC current flowing through the component, and its power spectrum can be approximated as:

Thus, a DC current flowing in every P-N junction has an associated noise whose amplitude is proportional to the square root of the current itself, as shown in Fig. 1.4. An important figure to keep in mind is the 4pA/√Hz introduced by a 50μA DC current.

Starting from the variance σ² (square value), given the system bandwidth Δf it is easy to infer the root mean square (rms) value:

that is the noise amplitude can be either positive or negative, with an amplitude limited between ±3σ for the 99.7% of the time.

1.2.2      Thermal noise (Johnson)

A second source of noise, discovered by John Johnson and often referred to as thermal noise, is instead caused by the random thermal motion of electrons within the electronic component (e.g. within a resistor); this random motion doesn’t depend on the presence of a DC current, since the typical transport velocity of electrons in a conductor is much lower than the thermal velocity of electrons. Due to its nature, this phenomenon is of course extremely dependent on the absolute temperature, T. As for Poisson noise, thermal noise can be considered white, i.e. its noise spectral density is independent from the frequency. More accurate models, however, show that thermal noise spectral density actually starts to drop at very high frequencies (above about 10THz), outside the frequency range of typical electronic circuits and devices.

Fig. 1.4: Shot noise versus current I (left) and small signals equivalent circuit of a diode, including the shot noise (right).

Fig. 1.5: Thermal noise versus resistance value (left) and thermal noise of a resistance representation , series and parallel (right).

Fig. 1.6: Flicker noise spectral density versus frequency (left) and noise corner frequency (right).

Fig. 1.7: Flicker noise trend versus frequency (left) and versus time (right).

Fig. 1.8: Pop-corn noise spectrum with a disturbance at 50 Hz (left),trend versus time(center) and amplitudes distribution (right)

It is possible to demonstrate that, in a resistor R, the thermal noise can be modeled by a voltage generator in series to its terminal, or –equivalently– by a current generator in parallel to the resistor itself, as shown in Fig. 1.5, whose values can be computed as:

1.2.3     Flicker noise (1/f noise)

Flicker noise is typical of all active devices and also of some discrete passive elements, such as carbon resistors. Its amplitude is very dependent on the nature of the phenomena taking place inside the device: for instance, it has a strong impact on MOSFET transistors, while being almost negligible for BJT transistors, in which it is mainly caused by traps with contaminations and crystalline defects in the emitter-base depletion region, whose energy is concentrated at low frequencies.

Flicker noise can be modeled with a parallel generator and is always associated to a current flow; moreover, differently from the two kinds of electronic noise described up to now, flicker noise has a very distinctive frequency-dependent behavior, shown in Fig. 1.6:

where Δf is a small bandwidth around the frequency f; I is a DC current; a and b are constant values that depend on the device, typically around one (in general, in the range 0.5-2); h is a constant value typical of the device.

Since its spectral distribution is substantially inversely proportional to the frequency (see Fig. 1.7), it is often called 1/f noise or pink noise, and its contribution is particularly important at low frequencies. As shown in Fig. 1.6, its amplitude is often quoted in terms of noise corner frequency, i.e. the frequency at which the 1/f noise intersects the white noise; above this frequency, the 1/f component can typically be neglected. Depending on the specific device considered, the noise corner frequency could range from few Hz or kHz (typical for BJTs or JFETs) up to MHz or even GHz (typical for MOSFETs). The amplitude distribution of the flicker noise is usually not Gaussian.

Note that if we consider a frequency interval with boundaries f1 and f2, the total noise power in such interval only depends on the ratio between f1 and f2, and not on their absolute values:

For instance, the flicker noise power is the same if we consider an interval between 0.01 Hz and 0.1 Hz and a second interval between 100 Hz and 1000 Hz. This is very different with respect to the case of white noise, which instead depends on the absolute width of the considered interval (in our example, the white noise power would be a factor 10⁴ higher in the latter case than in the former).

An important observation: mathematically, the 1/f noise would result in an infinite spectral density at 0Hz, thus resulting in infinite noise power, which is absurd. This paradox can be easily solved by considering that the minimum frequency is determined by the observation time. For instance, let us consider an amplifier with 1 kHz bandwidth and assume to observe the behavior for 1 day, i.e. to have an inferior limit set by:

the observed bandwidth spans across 8 decades. If, instead, we observe it for 100 days the bandwidth would span across 10 decades (since the inferior limit would be 10-7 Hz), its spectral density approaching infinity at low frequency, but only giving a power increase of 25% with respect to the former case.

Fig. 1.9: Burst noise spectral density versus frequency (left) and different burst noise sources combined with 1/f noise (right).

1.2.4      Burst noise (pop-corn noise)

The burst noise is another type of low-frequency noise typical of some integrated circuits or discrete transistors, which owes its name to the bursts on few discrete levels (two or more levels) that appear when watching this signal at an oscilloscope, as shown in Fig. 1.8 (where some bursts appear superimposed to a seemingly-white noise). The sources of this noise are not completely known yet, although it has been proved that it is associated to the presence of heavy metal ion contaminations and that it is linked to the simultaneous release of many electrons from different traps. In particular, golden-doped devices present a very high burst noise.

The repetition frequency of the bursts is usually in the audio band (some kilohertz or lower), causing a typical popping when transmitted with a speaker. The spectral density of the burst noise is:

where k1 is a constant value distinctive of each device, I is a DC current, c a constant between 0.5 and 2 and fc is a characteristic frequency associated to a specific noise process. The spectrum is shown in Fig. 1.9; at higher frequencies the noise spectrum decreases as 1/f². The burst noise processes often occur with different time constants, causing the presence of many humps in the spectrum. Furthermore, often the burst noise combines with the flicker noise resulting in the spectrum shown in Fig. 1.9. The amplitude distribution is not a Gaussian. Fig. 1.8 reports the burst noise probability density, highlighting the non-Gaussianity of such process, which results in the superposition of two (or more) Gaussian probability densities.


Typically, when speaking about noise the most important characteristic is its total power, or rms amplitude.

In principle, it is easy to compute the total noise power by integrating its spectrum density across the bandwidths of interest Δf. However, real circuits don’t have a defined bandwidth with ideally-steep limits, but instead complex transfer functions with one or multiple poles. If the equivalent input noise density is independent from the frequency (white noise), it is possible to simplify the computations using the concept of noise equivalent bandwidth.

Consider an amplifier with constant noise spectral density /Δf=Sio, and a single-pole transfer function with voltage gain Av(f). The spectral density of the output voltage noise S0(f)=/Δf is the product of the input voltage noise spectral density and the square of the voltage gain, decreasing by 40dB/dec beyond the pole. The total output voltage noise is obtained by integrating the output spectral density:

We can express this result as the product of the circuit DC gain Av0 and introducing the definition of noise equivalent bandwidth. By comparing the two equations, we obtain:

Considering a single pole transfer function, given by:

where fpole is the -3dB frequency, and solving the integral, we find the following noise bandwidth, which is a general solution valid for every single pole circuit:

Note that this equivalent bandwidth is broader than the -3dB one, by a factor 1.57, thus a circuit with single pole transfer function causes a noise equivalent to a clean frequency cut at the frequency Δf=1.57fpole. For circuits with more than one pole, the noise equivalent bandwidth decreases, ideally reaching the -3dB bandwidth when the circuit has an infinite number of poles. For instance a transfer function with two conjugate poles at 45° has a noise bandwidth that is only the 11% broader than the -3dB bandwidth (Δf=1.11·fpole), whereas a circuit with a coincident double pole has an equivalent noise bandwidth equal to Δf=1.22·fpole.

Thus the noise bandwidth doesn’t coincide with the amplifier bandwidth, because we are interested in the power transfer function, and not in the voltage one.

Let’s analyze a special case: the simplest noisy circuit, with limited bandwidth, i.e. the resistor with a capacitor in parallel. Consider the resistor equivalent parallel noise shown in in Fig. 1.10 and proceed as above with the normal computations. The transfer functions is:

Fig. 1.10: Circuit for the equivalent noise bandwidth computation.

Fig. 1.11: Linear electric network with its input noise generators.

the overall output noise voltage (mean square value) is:

This last equation corresponds to multiplying the 4kTR spectral noise by a term

The output noise is therefore independent from the resistance, thus the overall output noise voltage only depends on capacitance and temperature, despite the resistor being the noisy component. The noise bandwidth of the circuit in Fig. 1.10 is equal to fn=1.57·15000=23562Hz, considering that the output voltage loses 3dB at 15 kHz.


Since every active electronic component introduces its own sources of white and non-white noise, when dealing with complex circuits it is often useful to model the noise of each single sub-circuit (i.e. of every generic linear circuit) with a couple of equivalent noise generators, one for modelling current noise sources and one for voltage ones, as shown

Sv(f) this is the equivalent power spectrum which, applied to the circuit input and in absence of other noise sources, causes the same output noise spectrum when the input terminals are shorted.

A similar definition applies to the equivalent current noise generator. In this case, the noise spectrum at the real network output, S∞( f consequently, we find:

The definition of equivalent input noise generators allows to compute the effects caused by all the noise sources introduced by each electronic component. In the general case, the network can be sourced by a generator with generic impedance RS, as in the case of Fig. 1.12. The noise spectrum at the output terminals can be evaluated referring to Fig. 1.12 bottom, in which both noise equivalent generators have their own transfer function to the output, giving two separate noise contributions which must be quadratically summed to compute the total output noise power:

Observing this equation, we note that with low source resistances, Rs<>Rin, the input current noise dominates. Thus, when selecting the source impedance (which of course gives its own contribution to the total output noise), we must pay attention to the values of the two noise generators.

Fig. 1.12: Noisy network with source resistance (top), not-noisy network with the respective noise sources (bottom).

Fig. 1.13: Signal and noise power at the input and at the output of a circuit.


In order to characterize the noise performance of a two-pole network, whose generic representation is shown in Fig. 1.13, we can define a noise factor F, which expresses the degradation of the signal-to-noise ratio introduced by the network:

Even thought this method is only limited to cases where the input source is resistive, its exploitation is a very powerful tool, especially for telecommunications systems.

Referring to Fig. 1.13, where S represents the signal power and N the noise power, we have as only source of input noise (Ni) the source resistor RS. At the output, instead, we find a noise power N0 given by all circuit components and by the source resistor. We obtain:

Considering an ideal noise-less amplifier, the output noise is only due to the input source resistor, and thus, if the circuit power gain is G, the output signal S0 and the output noise N0 are given by S0 = GSi and N0 = GNi. Replacing these equations in the definition of noise figure, we obtain in this case F=1, i.e. 0 dB.

Another useful definition of F comes from the original one:

Another figure of merit often used to characterize the noise performance of a network is the noise figure NF, defined in the same way as the noise factor but expressed in dB: NF = 10·log10 F. The noise figure is usually defined for a small bandwidth Δf around a frequency f with Δf<Fig. 1.14, to evaluate the effects of each circuit parameter on the noise figure: a generic circuit, with input impedance zi and voltage gain G=vo/vx, is driven by a source resistor RS and feeds a load RL. The source resistor introduces a thermal noise , while the circuit noise is represented by its equivalent (uncorrelated) generators and . In order to compare the noise introduced by the circuit to that of the source resistor, we can consider the Thevenin equivalent at the input, at the left of the proper circuit (i.e. without considering Zi).

Fig. 1.14: Representation of the equivalent input noise, for the noise figure computation.

Using the definition of noise figure and replacing the values we already found, we obtain:

where is the overall input noise. This equation provides the spot noise factor, in the (pretty much realistic) hypothesis that the correlation between and is negligible. Observe that F is independent from all the circuit parameters (G, RL, zi) except for the equivalent input noise generators and for the source resistance RS: therefore, the noise factor of different circuits must always be compared at the same RS. Moreover, for small values of RS the noise is dominated by the > generator, whereas the > generator is dominant for large RS ; in the between, there is an optimal value of RS which minimizes the noise factor. This optimal resistance can be computed by differentiating the previous equation:

The optimal noise figure is:

thus the smaller the product ·, the smaller the NFopt.

Let us suppose to have the two trends shown in Fig. 1.15, associated to two different devices A and B. From a superficial analysis of the two graphs, the device A could appear to be less noisy than B, since its NF is lower. Nevertheless, this comparison strongly depends on the source resistance value: indeed, in case the source resistance is equal to RS*, the device B is much better than A! Therefore, comparing two devices only basing on their NFopt doesn’t make any sense, as its circuit must be evaluated in its actual operating conditions. In case the source resistance is large (RS>RS,opt) and a selection has to be performed between two amplifiers with different equivalent noise generators, the choice should fall on the one with the smaller parallel generator. The opposite applies in case of small source resistance (RSFinally, it is possible to compute the total input noise voltage for a generic two-port network (see Fig. 1.14) as:

Fig. 1.15: Comparison between two different trends of NF as a function of R s .

Let us now make a practical example: consider the operational amplifier LM394, whose noise trends are represented in Fig. 1.16, with bi-logarithmic axes. The NF (highlighted area in the figure) is the difference between overall input noise and source resistor noise (in log scale). From this figure we can observe that RSopt, i.e. the point at which the equivalent voltage and current generators have the same value, is equal to 15 kΩ. The total noise coincides with the OpAmp equivalent voltage noise for low source resistances, being then dominated by the source resistor noise beyond 200 Ω and by the equivalent current generator beyond 10MΩ.

whereas the optimum noise figure is:

For matched circuits (source resistance equal to Thevenin input resistance), the smallest possible NF is 3dB, unless particular techniques are used for further noise suppression; the NF for a practical low-noise circuit can range between one and few dBs.

Fig. 1.16: Equivalent input voltage noise of the OpAmp LM394.


Let us now analyze in detail the sources of noise in electronic components, starting from the most widespread electronic component: the resistor.

In addition, there exists a secondary excess noise, whose intensity in general decreases at higher frequencies strongly dependent on the kind of resistor (e.g., metal film, wirewound, metal oxide, carbon film, etc.), in its internal structure (both macroscopic and microscopic), on the nature and characteristics, and also on the fabrication process (for this reason we said that it is process dependent). Its nature is not fully understood, even though in general its contribution has been reduced (but never completely eradicated) thanks to technological improvements.

Fig. 1.17: Complete equivalent circuit for small signals of a diode with noise sources.

Excess noise is ultimately generated when current flows through a discontinuous conductor; for instance, the strong excess noise of a carbon resistor can be explained by its structure made by carbon particles mixed with specific binders and baked together, that introduce conductivity variations, thus causing micro-arcs between the carbon particles and creating jumps in the current flow and thus excess noise. Instead, wire-wound or metallic film resistors have a much more continuous conductivity, therefore introducing almost negligible excess noise. In Fig. 1.18, the total noise generated by a resistor is sketched with respect to frequency: thermal noise is dominant high frequency (>10kHz) whereas excess noise (in this case with a 1/f dependency) dominates at low frequency. Excess noise can be conveniently described be the parallel generator:

in which γ is typically very close to 2 and α to 1. The constant H, which determines the noise intensity, depends on the resistor type and on the fabrication process; as a first approximation, it doesn’t depend on resistor value (even though a slight trend increasing with R can be often identified). The noise corner frequency is obtained when the contributions gets equal:

In order to express the excess noise amount in a resistor, we can introduce the noise index NI, defined as the resistor noise mean square value expressed in μV per each Volt of DC drop across the resistor in one decade of frequency. By applying this definition, we find an equation independent from IDC and R:

should be expressed in μV² per frequency decade. The low excess noise resistors have NI≤20dB.

Fig. 1.18: Total noise of a resistor (left) and noise index for some typology of resistor (right).


Differently from resistors, which mainly introduce thermal noise, a diode mainly introduces shot and flicker (1/f) noise; its equivalent noise circuit is represented in Fig. 1.17 (where rd=1/gm is not a physical resistance, thus it doesn’t introduce any noise). Both flicker and shot noise can be represented by means of current generators in parallel to rd, leading to the following expression for equivalent noise generators:

The shot noise can also be expressed as:

In addition, a secondary noise source associated to a diode is the thermal noise due to its series resistance rS (caused by the non-nil resistivity of the semiconductor); even though this noise source is often very negligible with respect to shot noise, it is worth to spend a few words discussing its contribution.

Note that increasing the diode bias current brings to an increased thermal noise. This trend is due to the fact that increasing the biasing the electric fields become significant also in the ohmic regions and the present carriers are not anymore in balance with the reticle at the temperature T.

Fig. 1.19: Computation of the current noise in a biased junction (left) and an equivalent circuit for small signals (right) with the noise generators.

The average kinetic energy, E, of the charges in the resistive paths can by formally written as E=3/2kTe, with Te>T thus it is not surprising if also the thermal noise of these charges is 4kTe/Re and it increases at higher biasing. Let’s consider rs=100 Ω Id equal to 0.01 mA, 0.1 mA, 1 mA, 10 mA obtaining the following white noise:

thus with rs=100Ω, in order to minimize the overall noise, we will chose the lowest current possible even if it causes an increase of (1/gm)² and an increase of the flicker noise.

Note that both shot noise (2qIDΔf) and small signal impedance (1/gm = kT/qID) depend on ID: increasing ID has the double effect of reducing the diode small-signal impedance (beneficial effect) and of increasing its noise contribution (detrimental effect). The equivalent noise generator is equal to :

from which we can notice that increasing ID has a global beneficial effect to the (voltage) noise, due to the square power associated to it.

Let us now consider the network in Fig. 1.19 (right) and analyze the current power spectrum associated to iout. We can consider the equivalent small signal circuit (Fig. 1.19 left) which also schematizes the two noise equivalent generators, associated to both the resistor R (thermal noise 4kT/R) and to the junction (shot noise 2qI).

The noise current that flows between power supply and ground, iout, is the superposition of two independent contributes; its spectral density equals:

If 1/gm>> R, the dominant contribution is 2qI; conversely, if 1/gm<< R, the contribution 4kT/R becomes dominant. At low currents, i.e. when the junction resistor is much higher than the resistor R, the current noise spectrum has an intensity equal to 2qI; when the bias current increases, the series resistance dominates the overall noise and thus the spectrum intensity tends to the constant value 4kT/R.


Ideally, both capacitors and inductors are completely noise-less components, as they don’t introduce neither current- nor voltage-noise. However, the model for real inductors and capacitors include series resistors (ESR), that introduce their own thermal noise. Typically, however, their values are so small (in the order of few ohms for electrolytic capacitors, and even less for other components) that their contribution is negligible. The dual discussion can be applied to equivalent parallel resistors (EPR), typically close to infinite.


The noise in a bjt is mainly caused by four separate and independent contributions.

In the active region on a bipolar transistor, the minority carriers entering the collector-base depletion region are accelerated from the electric field and collected by the collector. The time needed by the carriers to reach the collector-base junction, either by diffusion or by conduction, is random; therefore, the collector current consists in a series of random pulses, which originate a shot noise <i²C> superimposed to the collector-emitter current IC. Similarly, the base current IB is due to the recombination in the base and in the base-emitter depletion region and also to the carrier injection from base to emitter; all this processes are random and independent, thus originating a shot noise <i²b> superimposed to the base current IB. The spreading base resistor rb is the resistor of the weakly doped base region between the external base contact and the active base region (see Fig. 1.20). This is a physical resistor, thus exhibiting thermal noise <v²b>. Also the collector resistance rC, in series to the collector, introduces thermal noise, but, being in series with a high impedance node (the collector itself introduces an ideally infinite resistance), its noise is negligible and usually it is not included in the model. Note that the resistors rπ=β/gm and ro are dummy, equivalent resistors, introduced only in the linear equivalent circuit for small signals of the BJT, and, thus, they don’t introduce thermal noise.

We find experimentally that flicker and burst noise in a bipolar transistor can be modelled with an equivalent current generator in parallel to the internal base-emitter junction. These generators can therefore be combined with the current shot noise generator in a single equivalent <i²bT>. The small signal equivalent circuit with all noise sources is shown in Fig. 1.21. Since they are originated by different and independent physical mechanism, all the noise sources are independent from one another; their mean square values are:

In the followings, we will neglect burst noise to simplify computations. It has been experimentally observed that in silicon transistors the current noise1/f of the base resistor is not due to the overall rb (as for the germanium) but only due to a part of rb, thus the effective 1/f noise for the voltage generator results:

where rb' is the part of rb that generates 1/f noise. The remaining part of rb generates burst noise. The values of the parameters are: 10-15 < K1 < 10-12 V, with a varying between 1÷2.

The 1/f noise is originated by the trapping and release of the carriers in superficial states, typically caused by defects such as impurities and dislocations. Transistors with high β and low collector current exhibit fewer 1/f noise. Since the 1/f noise strongly depends on superficial properties, the passivation processing allows to protect transistors surface.

In order to give a quantitative estimation of BJT noise, let us first review some of the fundamental parameters of a transistor:

Fig. 1.20: Cross-section of a diffused npn transistor.

(the thermal voltage V T =kT/q equals 26 mV @ 300 K);

Fig. 1.21: Small signal equivalent circuit of a transistor highlighting its noise sources (top) and equivalent circuit with input equivalent noise generators (bottom).

Fig. 1.22: Voltage noise spectrum of the 2N4125 bipolar transistor.

The <vi²> value is computed by shorting the input of each circuit in Fig. 1.21; note that for computing the rms value, we only need to consider the magnitude of its spectral components and not its phase, since all noise generators are independent and have random phase (noise generators’ polarity has no influence on the circuit response). We obtain:

At intermediate frequencies the noise is white. Since the resistor rb is typically in the order of hundreds of ohms (it can only become in the order of kΩ in particularly fast transistors with sub-micron base region thickness), whereas rπ is in the order of kΩ, the former can normally be neglected, leading to the following simplified expression:

Fig. 1.22 shows a graph representing the input voltage noise spectrum of a bipolar transistor. As can be seen from the last equation and from the plot, at intermediate frequencies the equivalent noise voltage decreases for increased currents, because of the decreased 1/gm. At higher frequency, the collector noise current approximates to the constant value 2qI cr²b, since all other noise generators present between base and emitter are not effectively transferred to the collector (because of the pole at the frequency fT). Because of this pole, the square module of the transfer function between collector current and base-emitter voltage decreases with the inverse of the square of the frequency. Only a generator applied at the input with a spectrum that increases with the square of the frequency can perfectly compensate for the transfer efficiency decrease and can cause a white noise at the output. From this graph, we observe that increasing the collector current the noise decreases.

we leave the inputs of both the circuits in Fig. 1.21 open. Since the base is open, the voltage generators and have no effects; only the generators of base and collector current and of 1/f noise are effective. We obtain:

Since <ii²> is the dominant equivalent input noise generator in circuits where the transistor is fed by a high impedance source, reducing the bias current and maximizing β allows to achieve the best performance under these conditions. Note that the requirement of low bias currents to minimize <ii²> is in contrast with the requirement of high bias currents to minimize <ii²>.

Fig. 1.23 shows the spectral density <ii²>/Δf : the increase at low frequency is due to the flicker noise, whereas the increase at high frequency is due to the collector current noise referred to the input. This noise increases at high frequency because of the reduction in transistor current gain which translates into a worse transistors noise performance. The frequency fb can be computed by equaling the high frequency noise with the medium frequency noise, obtaining:

By replacing the values in Fig. 1.23, we find fb=50MHz. At the same way we can compute fa equaling the 1/f noise to the medium frequency noise:

with K1=320·10-18 we find fa=1kHz. Fig. 1.24 shows the input current noise of the transistor 2N4125.

We can observe that the cutoff frequency becomes progressively higher than when the bias current is increased. This is due to the increase of β with the component bias current. Furthermore, by increasing the collector current the noise increases (differently from <v²i> that decreases increasing the collector current).

Fig. 1.23: Spectral density of the equivalent input noise current for a bipolar transistor with I C =100μA, β=100, f T =500MHz.

Fig. 1.24: Current noise spectrum of the bipolar transistor 2N4125.

Fig. 1.25: Trend of the current and voltage noise generators of a BJT.

The graph in Fig. 1.25 compares how the equivalent voltage and current noises of the BJT vary as a function of the frequency and the collector current. From this graph we can observe that, when the collector current is increased, the current noise generator increases whereas the voltage noise generator decreases. The trend with frequency is the same: in both cases the equivalent noise generator is decreased when the frequency is increased.

Let us consider the value of the NFopt for medium frequencies:

Fig. 1.26: Optimum noise figure as a function of I C .

Fig. 1.27: Optimum source resistor as a function of I C .

Fig. 1.28: Representation of the circuit noise by means of a single input voltage noise generator (a) Original circuit; (b) Equivalent circuit.

Fig. 1.29: NF as a function of I C (left) and Rs (right).

As we see, it depends on base resistance, emitter resistance and transistor β. In order to reduce the NFopt, the transistor β should be increased, or rb or re should be reduced. Fig. 1.26 shows the trend of Fopt as a function of IC for different values of β and rb.

We can observe that increasing IC also increases the NFopt. Thus the optimum is working with low collector currents. The value of the optimum source resistor is determined with the following equation:

In Fig. 1.27 RSopt is represented as a function of IC for different values of rb and β. From the figure we can observe that a decrease in the collector current causes an increase in RSopt, whereas the transistor β and the base resistance have only a weak effect on RSopt. In the limit case in which the base resistance rb is negligible, RSopt is linearly dependent on the collector current: decreasing IC the RSopt increases in accordance with the equation RSopt≈r√β.

Once the input noise generators are computed, we can easily determine the noise behavior of the transistor with any source impedance. Consider the circuit in Fig. 1.28.

The total equivalent input noise results:

Let us now analyze the trend of NF as a function of the collector current, varying the source resistor. The minimum points of these graphs correspond respectively to the optimum collector current value to have the minimum NF for a fixed source resistor, and to the value of the optimum source resistor to have the minimum NF for a fixed bias current. By modifying the collector current and source resistor values, we can find the best NF. In order to clarify this graphs let us introduce a numerical example. Consider a bias current Ic=100μA: from Fig. 1.29 left, we observe that in order to have the best NF (thus the lowest, i.e. low noise) we should use Rs=10kΩ. With this resistor the NF is equal to about 0.8dB, thus very close to the optimum value of 0.2dB. Selecting instead Rs=100kΩ the NF would be equal to 3.2dB, i.e. the circuit is very noisy being the NF very far from its minimum of about 0.1dB. Let’s now analyze the overall input noise.


For Rs=10kΩ we obtain:

For Rs=100kΩ we obtain:

In the former case, the selected Rs being very close to its optimal value (i.e. 6.7kΩ), the input noise is lower and the contribution of the current and voltage generator of the BJT to the overall input noise is almost the same. In the second latter case, the source resistance being very far from the ideal one, the overall input noise is higher and the contribution of the current generator of the BJT to the overall input noise is higher than the contribution of the voltage generator. A similar example is shown in Fig. 1.28b. With a fixed source resistor we have to select a bias current that minimize the NF, if Rs=10kΩ we observe that the optimum current value is 100μA, as already shown.

In conclusion, in order to have the minimum noise, we need to select both source resistor and bias current so that the operating point is close to the minimum of the curves in Fig. 1.29. In both graphs, at the left of the minimum the voltage noise is dominant, whereas at the right the current noise dominates. We analyze now the NF variation as a function of frequency.

From this graph, we see that the NF translates to right when the frequency is increased, as the noise generators values decrease (we transition from 1/f to white noise). The voltage equivalent generator dominates on the right of the minimum point, whereas the current generator dominates on the left. For such high resistor values the current generator dominates, thus for this reason only the right part moves with the frequency.

Another way to represent on a graph the NF is the one shown in Fig. 1.30, in which per each frequency we represent the lines with constant NF. The main limitation of this representation is due to the fact that, in order to represent very many frequencies, many graphs of this kind are required.

From Fig. 1.31, we observe that with Rs=1kΩ it is better to use an Ic between 200μA and 300μA, which correspond to NF=1dB, thus the lowest value among the reported ones. With higher Rs it is better to use lower Ic, in order to reduce the parallel contribution <i²in> at the expense to increase the series contribution <v²in>. The signal source influences the total system noise, which in turn depends on the transistor bias point, thus we can select the Ic that minimizes the NF.

Fig. 1.30: Lines with the same NF, for different working frequencies of the transistor 2N4250.

Fig. 1.31: Overall rms voltage noise for the LM194.


In a FET transistor, the drain current is modulated by modifying the resistivity of the resistive channel connecting source and drain. Since the material that constitutes the channel is resistive, it introduces thermal noise: this is the dominant noise contribution for both JFET and MOSFETs transistors, and can be represented with a current noise generator connected between source and drain. Also the flicker noise in FET transistors is represented with a current generator connected between drain and source; thus, the two noise generators can be joined in a single noise generator <i²d>. The other noise source <i²g> is the shot noise caused by the gate leakage current and is usually very negligible, only becoming significant with very large source impedance. Fig. 1.32 (top) represents the equivalent circuit for small signals of a FET, with the respective noise generators.

The noise generators have the following values:

where IG is the gate leakage current, ID is the drain bias current, K is a constant characteristic of each device, a is a constant value comprised between 0.5 and 2, and gm is the device transconductance in the operating point. The factor 2/3 multiplying the FET transconductance is due to the non-uniform transconductance, which varies across the conductive channel. If the FET is in the triode region, the transconductance cannot be defined; thus, the corresponding term for the drain current noise generator must be replaced with the inverse of the resistance in the FET ohmic region, equal to:

The input equivalent noise generators of a FET can be computed using the circuit in Fig. 1.32 (top). This circuit should be made equivalent to the one in Fig. 1.32 (bottom). The output noise is computed in each case with a short circuit load.

In order to find the input equivalent voltage noise generator, we need to short the input of each circuit in Fig. 1.32 and match the output noise currents. We find:

The equivalent voltage noise spectral density is represented in Fig. 1.33. Usually, in FETs the flicker noise is higher than in bipolar transistors, in particular for MOSFETs where this term can extend up to the MHz region.

Fig. 1.32: Small signal equivalent circuit of a FET with noise sources (top); representation of the two input noise generators (bottom).

Fig. 1.33: Spectral density of the input equivalent noise voltage of a FET.

The reason being the presence of electron energy levels in the Si-SiO2 interface, producing flicker noise with a spectral density that can even exceed the thermal noise contribution at low frequencies (lower than several kHz), for the most conditions of bias and components geometry. The dependence on bias conditions and on device geometry is very sensitive to the fabrication process.

In fact when the size of the transistor is larger, a higher number of superficial states are present under the gate, and an average effect occurs, thus reducing the overall noise. We can also observe that the input flicker noise is inversely proportional to the gate-oxide capacitance per area unity. From a physical point of view, this is reasonable, since the flicker noise can be considered to be originated from a random fluctuation during time of the superficial state charge; this generates a time-variant component in the threshold voltage, which is inversely proportional to the oxide capacitance.

Thus, for an MOS transistor we can conclude that:

A typical value of Kf is 3·10-12 V² pF. Usually, for CMOS technology the 1/f noise in p-MOS is lower than in n-MOS, i.e. Kp

Usually, the 1/f noise for a JFET is lower when the gate leakage current is reduced. However, a high leakage current is not necessarily related to a high noise, since it could be due to a superficial current. The input current noise generator can be computed keeping open the input of each circuit in Fig. 1.32 and equaling the output noises. We obtain:

At low frequencies the input current noise generator is dominated by the leakage gate current IG, which is very small (in the order of 10-12A or less). For this reason the FETs typically have much better noise performance compared to bipolar transistors when the source impedance is high. In these conditions, the input current noise generator dominates, and it is much smaller for a FET than for a bipolar transistor.

However we have to note that the input voltage noise generator of a bipolar transistor is typically lower than the one of a FET; thus, for small source impedances a bipolar transistor has better noise performance than a FET. The trend of the input equivalent noise generators for the MOSFET 2N3631 and the JFET 2N3821 are reported in Fig. 1.34.

As we can see, the current noise is apparently white at low frequencies, but, at a certain threshold frequency it starts to increase, because of the presence of parasitic capacitances and of the coupling capacitance CGD between input and output. At high frequencies, the noise generators and are not uncorrelated anymore. In the voltage noise, at low frequencies, we often observe some humps due to the fact that the trapping centers have different time constants.

Fig. 1.34: Input equivalent noise generators: MOSFET 2N3631 (left), JFET 2N3821 (right).

Fig. 1.35: Comparison among BJT, JFET and MOSFET.

In Fig. 1.35 we report on the same graph the current noise generators of BJT 2N4250, JFET 2N3821 and MOS 2N3631. From the voltage noise point of view, the BJT is the best device: indeed , beyond having the lowest white noise, it has a much lower flicker noise compared to JFETs and MOSFETs (mostly due to the generation and recombination centers). Concerning current noise, the best device is the MOSFET (noise due to the leakage current), whereas the worst one is the BJT. Thus, the selection of one or the other device is also related to the input signal source: indeed, if the source impedance is high the input current noise generator is dominant and thus it is preferable to use a MOSFET or a JFET (current noise much lower than in BJTs). Vice versa if the source impedance is small the input voltage noise generator is dominant and it is better to use a BJT. These considerations also have effects on the optimum source resistor and on the NF. Since a BJT has a lower voltage noise generator and a lower current noise generator compared to a JFET, it has a lower Rsopt than a JFET and a higher NF. This comparison is reported in Fig. 1.36.

Fig. 1.36: Comparison between the R s of a BJT and a FET.


Low noise amplifiers (LNAs) typically make use of an input stage with common emitter or a differential stage with resistive load. Since the input stage has both high voltage and current gain, the noise of the following stages is usually negligible and the resistive load scarcely contributes to the overall noise. Nevertheless, the good noise performance are achieved at the expense of other parameters, such as gain and bandwidth: for instance, by using an active load it would

Você chegou ao final desta amostra. Inscreva-se para ler mais!
Página 1 de 1


O que as pessoas pensam sobre Electronic Systems

0 avaliações / 0 Análises
O que você acha?
Classificação: 0 de 5 estrelas

Avaliações de leitores