Noise Optimization in Sensor Signal Conditioning Circuit Part I
Noise Optimization in Sensor Signal Conditioning Circuit Part I
Noise Optimization in Sensor Signal Conditioning Circuit Part I
in the world, and welcome to todays webinar, Noise Optimization in Sensor Signal Conditioning Circuits. This is part one, presented by Analog Devices. Ill be your moderator today. before we begin. Im Gary Cocker.
You may submit questions for the presenter at Just type your question into the
ask a question area on the left-hand side of your interface, and then click the submit button. You may enlarge the slide
window at any time by clicking on the enlarge slide button located below the presentation window, and the slide will advance automatically throughout the event. If youre
experiencing any technical problems, please visit our webcast help guide by clicking on the help link below the video window. In addition, you can contact our technical support Now,
helpline, which is also located in the webcast help guide. onto the presentation, Noise Optimization in Sensor Signal Conditioning Circuits," part one. Noise is a term that is
broadly used in signal path design, but in many cases there is a level of confusion as to what it exactly means and how to deal with it. Today, in this first of a two-part series, we will
explore the topic in detail and get insight and answers on the sometimes intimidating subject of noise. Making this
presentation today is Reza Moghimi, applications engineering manager in Analog Devices Precision Amplifier Group. welcome, and over to you. Reza,
Reza Moghimi: Thanks, Gary. Lets start. There have been many papers written
on the topic of noise and, as you said, still there is -- the term is used broadly with some levels of confusion. yelling is asked not to make too much noise. A person
engineers think of DC errors, such as offset voltage and bias current as noise, but others refer AC parameters such as current noise and voltage noise densities as noise. We will establish a
definition of noise from an electrical engineering point of view. But for now, generally speaking, we can say that noise is
any unwanted, undesirable signal that affects the quality of the useful information.
Slide 2: So I know that my audience today comes from a very diverse background. Some of you come from a digital background and are
not familiar with the analog concepts, and some of you are very
getting something out of this and the next session, which is scheduled for the month of February. If I do not get to all
questions or give satisfactory answers to your questions, or dont know the answer to your questions, I promise you that ADI has enough experts in this field that I can consult with and get back to you, so please do not hesitate to send us email. So
after explaining why design for low noise and defining noise and its types, I will introduce a general formula to calculate noise of a signal condition circuit. Next, I will introduce a low
noise design process, and in my February session I will explain deeper into this low noise design process and share with you dos and donts, and a number of ways to use and optimize common signal conditioning circuits.
Slide 3: So lets see what changes have come about and why, more than ever, every system designer needs to know about low noise design. So, why low noise?
Slide 4: Lets look at the types of applications that are in the mainstream these days and need low noise signal conditioning. typical signal chain is shown over here. A
many sensor types, and the rest of the circuit is a single conditioning path. Making high-resolution measurements A
parameter Im sure youre all familiar with, called an S signal to noise ratio, gives a good idea as to how much noise we have in a system. noise? So the question is, what are the major sources of
themselves, or a pickup or radiated noise, as it is called? Remember that sensors have their own noise and they are supposed to detect small signals which cannot be distorted. So to attain
best noise floor, designers need to understand component levels noise sources, pick the best signal processing architecture, and prevent external noise sources that might interfere with the application circuit.
Slide 5: Why low noise signal conditioning? Popular applications have It used to be
common to have +/- 22V, and now its pretty common to have +/0.9 V or +/- 1.5 V, and more applications are requiring higher precision and accuracy. As an example, car industry has moved Lower supply
operations, combined with higher accuracy requirements -meaning the number of bits -- has made measurements of
for 5 V and 10 V full-scale data acquisition systems, as it is shown here, and we used to say that these were very small and we needed low noise signal conditioning. For example, we used to
point out, for a 14-bit system, where the full scale was 5 V, the LSB size was 305 V. But remember that this is the amount
of error that we are allowed after signal conditioning circuits. The situation is worse when we look at the output of the sensors, as I showed you in the previous slide -- signal chain. Sensors have, or produce, very small signal.
Slide 6: So, as an example, imagine a real-world signal that generates signals of 30 mV maximum, full-scale. bit system is 3.5 The half an LSB for a 12-
error or noise from the amplifier, would invalidate the measurement. Eliminating noise is very critical since this
Slides 7, 8, 9: Something else that we need to know when we have a driver for an ADC is the signal to noise ratio of an ADC can degrade as the result of the goodness of the amplifier driving it. Designing
low-noise signal conditioning is critical, since the noise generated by an ADC driver needs to be kept as low as possible in order to avoid worsening the signal to noise ratio. In this
slide, I show how the SNR gets degraded if you pick a wrong amplifier. As an example, when using the AD7671, which is the
16-bit, 20 V RMS noise and 10 mHz of bandwidth, if you pick this ADC lets see how we can degrade its SNR by picking different amplifiers. Shows us amplifiers that have different noise
specs, and we see the SNR loss as a result of the amplified noise. So lets get a better understanding of what noise is,
Slide 10: I explained that sensor signals are small, the need for resolution has gone up, power supply voltages have shrunk, and in order to make meaningful signal conditioning we need to learn about low noise design. It is critical to optimize the signal
conditioning circuitry so we get rid of the noise and measure the real signal. So lets see how we define noise and what are
Slide 11: So what is noise? One can define noise to fall into two
or inherent.
They can be periodic, intermittent, or random, and system designers can reduce the effect of these through a number of ways. Thermal agitations of electrons and random
generations/recombinations of electron-hole pairs are examples of inherent noise, which IC manufacturers have tried to reduce by better processes or better circuit design techniques.
Slide 12: And intrinsic noise is what I will be talking about today. We
will cover the extrinsic noise and how to deal with it in the future.
Slide 13: So I mentioned earlier, in loose terms, that noise is any unwanted signal. engineering term? So what is really noise in the electrical Lets define it. Engineers define noise as a
random process due to quantum fluctuations inherent in all resistors and semiconductor devices, specifically P-N junctions that create voltages and currents in any application. an instantaneous, has an instantaneous value, and is unpredictable, its possible to predict it in terms of probabilities, which we will talk about. Most noise sources are When Noise is
it comes to defining noise, we use terms such as peak to peak, RMS, and we show noise in the peak to peak and, many times, in spectral noise density graphs.
Slide 14: So here is how noise looks like over a 0.1 to 10 Hz, and over a much wider frequency of 10 Hz to 10 kHz, as it is shown on the right-hand figure. It is difficult to mathematically
characterize amplifier noise at low frequencies due to 1/f, temperature, and aging drifts, and possibly even popcorn. This
is why it is easier for us to just show 0.1 to 10 Hz photographs in our datasheets, and move on. Peak to peak is really only
Slide 15: So here is a noise signal, and unlike AC signals, whose power is concentrated at just one frequency, noise power is spread all over the frequency spectrum. unpredictable, but possible. of probabilities. Instantaneous value of noise is It is possible to predict in terms
Slides 16, 17, 18: Derivative of noise power, which is the mean square voltage of
the voltage noise, the frequency is called the noise power density, and is denoted as En2 or E2 in the case of the voltage noise. When specifying noise, as we can see, we should always
Slide 19: So lets see what RMS means. mathematically. Here is how we define RMS
Slide 20: Lets say that we have a pure sine wave. To get the RMS, we
have to square the signal, and once we do we get a signal like this.
Slide 21: We have to do the average. We have done the squaring. Again, its the root mean square. We have to do the mean.
Slide 22: And this is how it looks like after averaging. root of the signal, thats what happens. If you take the
So a sine wave in a
root, RMS fashion, is just going to look like as it is shown on the bottom figure on this slide.
Slide 23: So the graph on the left shows a peak-to-peak noise value of a part over the broadband frequencies. It is very difficult to
read peak-to-peak values accurately and consistently, as I mentioned, from the graph on the left. When noise power density
is plotted versus frequency, it provides a visual indication of how power is distributed over a frequency. In ICs, as you can
see on the right-hand figure, the two most common forms of power density distributions are what is called the 1/f and white noise. The flat part of the graph is called the white noise, The quantities
of the voltage noise and the current noise are noise spectral densities and expressed in nV per root Hz and peak A per root Hz. In some cases, A per root Hz per root Hz. The noise
spectral density shows the noise energy at a given frequency, while an RMS gives an RMS value over a given bandwidth or time interval.
Slide 24: It is always good to know the peak-to-peak noise value. Because
noise is random, there is always a probability the voltage could exceed the peak-to-peak value. This probability is shown here,
in terms of a new term, crest factor, which we define as the ratio of the peak-to-peak value over the RMS value of the noise.
10
RMS values are easy to measure repeatedly and are the most usual form for presenting noise data. One can use the table above to
estimate the probabilities of exceeding peak values, given the RMS values. A very common number to convert from an RMS to a The
probability of a peak-to-peak value exceeding 6.6 times the RMS noise value is 0.1%.
Slide 25: It was mentioned earlier that noise spectral density of an IC is composed of white noise and 1/f noise. I referred to these two But there are other
contributors to IC noise, and these are popcorn noise, shot noise, and avalanche noise. one by one. We cover all of these parameters
components such as resistors, capacitors, and inductors, that are commonly used in system designs and these elements have their own noise sources for noises. So lets get a better
Slide 26: So the white noise mentioned is the flat part of the noises spectral density, as it is shown over here.
11
Slide 27: It is also called the broadband noise, and the voltage noise density is constant over a frequency.
Slide 28: And this is how we show it mathematically in the RMS form.
Slide 29: Remember I mentioned that you need to have the bandwidth specified, and the f2 minus f1 is the bandwidth that we have in mind in this example.
Slide 30: If f1 is a lot smaller than, lets say, ten times of the f2 -lets say we pick f1 to be 100 Hz and the f2 to be 10 mHz or something, then the f1 is really irrelevant.
Slide 31: Then we can approximate the noise to be just the En times the square root of the f2 frequency. Remember, this is the noise
floor of the systems and a limiting factor for system resolution. white noise. This is what is called the broadband noise, or the
12
Slide 32: Another noise term that we define, or we mention, is the pink noise -- also called the flicker, or 1/f noise. At low
frequencies, as is shown over here, noise goes up inversely proportional to the frequency. term. And thats why the term -- 1/f
caused due to traps, which, when current flows, capture and release charge carriers randomly, therefore causing random fluctuations in current itself. In the BJTs, it is caused by
contamination and imperfect surface conditions at the baseemitter junction of a transistor. In CMOS, it is mostly
associated with extra-electron energy states at the boundary between silicon and silicon dioxide. The 1/f corner frequency,
which is a figure of merit, is a frequency above which the amplitude of noise is relatively flat and independent of frequency.
Slide 33: To measure, or to calculate the 1/f noise, the equations are shown over here, as you can see. Please note that the corner
frequency for a voltage noise is different, or might be different, than the corner frequency of a current noise. I
mentioned earlier that the unit for voltage noise is nV/root Hz, and for a current is sometimes A per root Hz.
13
Slide 34: One characteristic of a 1/f noise is that the power content in each decade is equal, and this is what is shown in this slide. As you see, we have bandwidths that are apart from each other by one decade. I showed you the formula earlier, and squaring both
sides of that equation gives 1/f noise power, which is proportional to the log ratio of the bandwidth, regardless of the band location. So if you pick 200 Hz and 20 Hz is a factor
Slide 35: So here is -- some interesting facts, which I like to share with you. White noise has equal energy per frequency.
Slide 36: And we mentioned that the f2, when it is very large compared to f1, RMS noise is set by the f2.
Slide 38:
14
RMS noise is set, I showed you in the equation, by the ratio of the f2 to f1.
Slide 40: And that is, I ask you which do you think sounds louder?
Slide 42: Why do they call it pink noise? I give you the answers for two
of them, and you can give me an answer for another one, for the third one, later, because I dont know the answer to that question myself. White noise sounds louder. It has more energy
Our ears are much more White light has all White light
frequencies with equal energy from each frequency. has equal energy per frequency.
15
Slide 43: To give you a feel as to how broadband noise and 1/f noise look like -- we actually get a lot of questions related to noise all the time, and one of the things that I usually ask my customers to do is look at the noise on the scope and interpret the results, or if they dont understand they can send us a picture. So here are those two references that you can use. The upper
figure is the broadband noise, and the one shown below it is the 1/f noise. The broadband noise is the furry shape, and 1/f Thats how you can say it.
Slide 44: So now the question is, what happens when we have spectral noise density and not the peak-to-peak graph in a data sheet? This
has happened sometimes in some of the data sheets that we have done over the past 40 or 50 years. Here, I show you a formula
Slide 45: Assuming you are given the graph on the left, you can use this formula. You can find out the corner frequency and, given the
two frequencies for the f1 and f2 of 0.1 and 10 Hz, you can find out what the peak-to-peak value would be based on the noise spectral density that you have. If you go through the
16
calculation, as I have here, you get the peak-to-peak noise to be 218 nV.
Slide 46: And here is the graph, which is in the data sheet, showing the peak-to-peak noise to be 200 nV. have a good idea. They come pretty close, and we
Slide 47: Another noise term is called, or noise type, is called the popcorn noise, also called the burst noise. It causes the step It is
caused by transistors jumping erratically between two values of beta. In the early days, popcorn noise was a serious issue that
resulted in random discrete offset shifts in a timescale of a few tens of milliseconds. Today, although popcorn noise can
still occasionally occur during manufacturing, the phenomenon is sufficiently well understood and parts are scrapped during limited testing. Popcorn noise is a part of the 1/f noise and Its purely a
days and not as big of a deal these days, although every once in a while you have some issues. We have done extensive in this
area and we have experts who have done great amount of work who
17
Slide 48: The noise that is called the shot noise, also called a [Schottky?] noise, occurs whenever a current passes through PN junctions and there are many junctions, as you know, in our parts. Barrier crossing is purely random and the DC current
thats observed is the sum of many random elementary current pulses. So this is the current noise, and it has a uniform Its a part of white noise. And remember, white
power density.
the value of a shot noise is given over here, which is the square root of 2 qI delta f -- I in the case of an amplifier is the Ib that you are talking about. There is a handy number that
you can use if I is given in the peak amps, which, for our J FET parts and CMOS parts, the Ibs are in the [pento arm?], you can use the simplified equation to quickly find out what is the amount of shot noise based on that current.
Slide 49: I mentioned that this is when the current passes through a PN junction, so here I have used -- I give you an example of two diodes operating at two different currents -- at one micron and 1 nA, and I have calculated the noise based on those currents
18
and the signal to noise -- the signal being the current that I forced through the diodes and the noise generated as a result of those currents -- the signal to noise ratio, as you can see, is 65 dB and 35 dB.
Slide 50: Another noise type is the avalanche noise, and it is found in PN junctions in reverse breakdown modes. We dont really have this I just
in our parts and I have the slide here to be complete. have a couple of bullets on these. much time on this.
Slide 51: The thermal noise, also called Johnson noise, found in all resistors, even a resistor sitting in a drawer in the lab somewhere. This noise is generated as a result of temperature Its the
thermal agitation of electrons in resistors that cause random movement of charge, causing a voltage to appear, and thermal noise is part of the white noise, or the broadband noise, that we just talked about earlier. The equation to find the thermal
noise is given right in the middle of this slide, which is the square root of 4 kTR times the bandwidth that we are working on,
19
thermal noise is, of course, to pick a small resistor -- as it is shown over here, the smaller the resistor the less noise. If
we can control the temperature, if we can cool the temperature that will be smaller, and as one of my friends, Mr. James Bryant, says, there isnt anything you can do with the Boltman constant because hes dead and that constant is fixed. So,
remember, doubling the resistance increases the noise by 3 dB because its under the square root term. resistance equals to doubling the noise. So four times the
Slide 52: One of my characteristics is that I always want to make sure that we all understand what we are talking about. question to you: So here is a
Slide 53: Is it A, B, C, or D?
Slide 54: I give you a hint: root Hz of noise. Remember that the 1 k resistor has 4 nV per
20
Slide 55: So, knowing that information and using this equation, by looking at this you will find out that the answer is C, which is 12 nV. And this way you can quickly look at your circuits and your components that you have picked and find out what the noise of a resistor is right off the top of your head. So remember, the 1
k resistor has 4 nV per root Hz and then from that point on its just the square root of -- you use that if you want to pick the 10 -- square root of 10 is almost 3, 3 times 4 is 12.
Slide 56: And if you cannot remember the equation or you dont want to be bothered, I am giving you a graph here that plots the resistor noise based on the resistor value and, as you see, as the resistor value goes up its noise up. Its very critical to
notice things, because in the next session, when we go and design the signal conditioning circuit, we need to account for all the errors, or all the noise sources that come into picture.
Slide 57: So now the question is, whats the sum of two noise sources? How do we add up the two noise sources? If the noise sources
are uncorrelated -- as I mentioned, the majority of the noise sources are recombined in a root-sum-square as it is shown over
21
here.
Slide 58: This means that adding tow noise sources that have the same energy only increases the overall noise by 1.4 or 3 dB.
Slide 59: Thats the square root of 2, remember that. sources added up have a 1.4 factor. Two equal noise
Slide 60: And this works for audio, as well. Two people talking the same
volume about two totally different things only add 3 cB of sound pressure levels to the party, and its not going to be 2 times as much. So remember the RSS or the root-sum-square fashion for
Slide 61: So, let me just say, go back to a signal conditioning circuit thats interfacing a sensor and give you a handy formula that you can always use to calculate the two different noise of your signal conditioning circuit.
Slide 62:
22
So what we have done so far is that -- I have explained that all ICs have some inherent noise within them. In case of an
amplifier, those noise sources can be modeled as zero impedance voltage generator en noise source in series with the input, and infinite impedance current sources parallel with the input, as I have shown in.
Slide 63: Do not worry too much about the direction of the arrows here -I just drew something. Each of these terms vary with frequency,
as I have shown, and the type of amplifiers that you have picked for your signal conditioning. I mentioned that the voltage
noise density unit is nV per root Hz and the current noise density is pA or A per root Hz. Both of these sources can be
treated as uncorrelated noise sources and the amplifier here that I have shown, I am treating it as a noiseless amplifier. This is an ideal amplifier that I wish we had, which we dont have yet. So now if you put the ideal amplifier and its noise
sources there, we come up with the signal conditioning as it is shown here with a sensor, which I show a sensor on the left-hand side of the figure with its noise resistance and resistor, as we mentioned, has a noise. All the resistors around the amplifiers So now
23
called, the refer to inputs, we can use the equation that I show on the bottom of the slide. In general form, these are all
noise sources that I could think of in a typical signal conditioning circuit. Total output noise referred to input is
given by resistor noises, op amp noise, which are current and voltage noise sources. Its not going to be -- there are some And we will talk
terms here that need a bit of explanation. about these things later on.
here is the term called noise gain as it is defined, what is called the bandwidth, or the effective bandwidth, as I explain to you later on what that is. So if you want to look at the
noise on the output, you have to know the referred to input noise and multiply that by the noise gain and not really the signal gain. this. Ill explain to you what this is and how we define
Slide 64: So what is really the noise gain? How do we define this? This
is how much the noise and whatever garbage and extra useless information is gained up by. Regardless of the configuration,
whether the amplifier is an inverting or non-inverting configuration, the noise gain, as I showed you earlier, is equal to 1 plus R2 over R1. Here, I show two circuits. The one on The one on
24
In an
inverting configuration, as I state here, the signal gain is minus R3 over R4; the noise gain is 1 plus R3 over R4. In the
left-hand side, where we have a non-inverting configuration, as you all know, the signal gain is 1 plus R2/R1 and the noise is 1 plus R2 over R1. So in both configurations the noise gain is 1
plus the feedback resistor over the R1, or actually R4 in this case. Sorry -- there is an error in my drawing here. Instead
Slide 65: And I mentioned the term called noise effective bandwidth. White noise is passed as if the filters were a brick wall type, but with a cutoff frequency of 1.57 times as large. The 0.57
accounts for the transmitted noise above the cutoff frequency as a consequence of gradual roll-off.
Slide 66: We can think of the amplifier as a single-pole low-pass filter, which is a good approximation, and that -- you can use the equations to come up with the noise of a first-order low-pass filter, meaning you have to multiply the noise by the square root of 1.57 times that corner frequency.
25
Slide 67: And if you want to design a higher order of filters, and just figure out the noise effective bandwidth, here is the equation that you can use.
Slide 68: The N is the order of the filter and once you use this -- I give you an example here, that if N is equal to 1, as it is the first order filter that I just explained in the previous slide, the factor is 1.57 and as you go higher and higher in the order you get closer to the brick wall and the bandwidth is narrowed, and less noise is passed through, which is a desirable feature, but, again, there is a price to pay to go to a higher-order filter. You get rid of the noise, but there is a price to pay. how life is in the amplifier world, and in our world. always a tradeoff that you have to make. Thats Theres
Slide 69: I say here that the high-order filter is good and as you can see it can make an improvement to our signal to noise ratio.
Slide 70: So assume that I have used the circuit in the previous slide for
26
the signal conditioning that I showed you, where I showed you the noise calculation equation and all the noise sources. So
assume that I have used that circuit and I am using a 10 mHz amplifier in an inverting gain of 1000, using the resistor values of 100 k in the feedback and 100 ohm as a source, and what I have done here is I have plotted amplifier noise -meaning, if I picked different amplifiers that have different voltage noise densities, put it in this configuration, the amount of noise that I will get at the output is just going to look like this. So if you pick an amplifier that has a large
value for its voltage noise density, then the output noise will be higher even if it 10 mHz. picking the right component. So you need to make sure youre This is something that we talk You have to know how to go
about picking the right component, whether it is a capacitor, a resistor, an amplifier, or anything thats just going to be used for signal conditioning. and good practice. You have to exercise a good discipline
Slide 71: Here, just to be complete -- here we look at the noise gain of a second-order system. The shape of the noise gain is going to be
27
different, of course, because of the resistors because of the capacitors. At very low frequency, noise gain is a function of
the resistors, as I have shown here, and at higher frequency is the function of the capacitors. Capacitors do not generate
noise themselves, but the current noise of an amplifier drops across the capacitor and creates a voltage noise error.
suggestion is always to build in low-noise design ideas and concepts, rather than designing something and then trying to reduce the noise by shielding, layout, and other techniques that you try to figure out at the end of your design process. have to build it in. So what process do I suggest? So you
Slide 73: So here is the process that I suggest. I will say you have to
know which frequency range or which frequency band are you interested in? Are you working in the 1/f region? Is your
thing you need to understand -- where your signals, or whats the bandwidth of your signals. And then you have to design --
my suggestion is, for the best performance that you can find. Todays amplifiers have noise range of 0.9 nV to 60 nV.
28
Understanding about the input architecture and the amplifier helps you pick the right amplifiers. So you pick the right
amplifier, you pick the right target components, and you know the bandwidths. You design around that. Then you worry about
the other things that are non-noise requirements, like the input impedance or how much current you have to use in your system and what gain, and stuff like that. And if noise specs is not met
after going through this process, then you need to go back again and pick a different component, maybe a different amplifier, and go through this situation several times. So, its very
important to understand a little bit about the amplifier -- how it is put together, and my goal is to show you in the next few slides how noise sources inside of an op amp power and what benefits do you get if you go with a bipolar or a CMOS or a J FET amplifier.
Slide 74: So what are the noise sources in the bipolar amplifiers? If a
data sheet tells you that the noise is 3 nV, its not good enough. You need to know how these 3 nV is achieved, since, as
I mentioned earlier, there is always a tradeoff that has to be made and that tradeoff might affect your application.
Slide 75:
29
So here is the input structure of an amplifier on bipolar process. Thermal noise, shot noise, and 1/f noise are the three
noise sources that you need to worry about, or a designer needs to worry about, inside of the power, but you need to be aware of these things. And voltage noise, I mentioned earlier, is made
Slide 76: So if you want to know the complete story in what factors in bipolar transistors, or what elements in bipolar transistors, contribute to the voltage noise density or the voltage noise of a bipolar op amp, here is the complete equation. mentioned there is always a tradeoff. And I
get, say, a low-voltage noise density you need to use a very high beta, as it is shown in that equation, in the denominator. But that requires a light doping and very thin Rb on the input, and Gm and Ibs, as you know, are directly proportional to the IC, and therefore the current noises can go.
Slide 77: So what works to minimize the en is the opposite of what is good for low-current noise density, which represents the fundamental tradeoff in bipolar design.
30
Slide 78: There are a number of parts that have super beta or Ib cancellation circuits in them that introduce correlated noise. Ib cancelled ports -- not, still, as good as FET for bias currents, but they bring in a lot of good benefits. This is the
compensation for bias current error term and we have a number of parts, and I will talk briefly about the compensated parts, what to do with them, in the next session.
Slide 79: The noise in CMOS parts -- we have many, many CMOS, many, many J FETs, and many bipolar parts that we have designed over the past 40-50 years.
Slide 80: So when we look at the noise sources inside of a CMOS part, here is how it looks like. There are three noise associated, and
these are the gate leakage, which I show as the Ing 1/f noise, which I show as If, and the current noise, which is another shot noise term, which is shown as Ind.
31
graph.
be used to get better noise specs, but each have their implications on the approximation application level.
Slide 82: Flicker noise is inversely proportional as an example to the transistor WL, so to reduce the noise one has to use input stage transistors with large geometries. So here we see, as we use
larger geometries for W and L in the denominator the noise is just going to go down, but this has implications that you need to be aware of when we go to the application level.
Slide 83: Bigger transistor geometries are just going to have larger capacitances that come into the end applications, and you need to worry about this and compensate the amplifiers correctly to get the performance that you need to get out of it.
Slide 84: Here's an equation to measure the current spectral density -- is just showing you the equation for the corner frequencies. And
Slides 85-86:
32
We can, of course, operate the FET where gm is large. that to get lower noise.
We can do
Slides 88-89: Again, we get the noise down but we are consuming a lot of power, a lot of current, as it is shown over here, or, as I mentioned, we can use a large geometries to get the noise down. Each one of these we will talk about later on, as to what implication it may have in the final design.
Slide 91: We know where a corner frequency is, where the flat part and 1/f noise are equal to each other --
Slide 92: -- and if you recall the equations for the CMOS part you get the corner frequency if you use this equation. This, again, has
33
you there, that finding the right corner frequency, or pushing the lower corner frequency to very close to DC is very important.
Slide 93: And here is the noise in the J FET. Compared to bipolar Therefore, FET op amps
tend to have a higher voltage noise for similar operating conditions. We have some J FET parts that are very low noise,
also, but when we do a one-to-one comparison we can make a statement like that. And remember, at room temperature, the
current noise density, like the CMOS parts, is not a problem. It is negligible. Many times in the A per root Hz -- but one
drawback here is that it doubles for every 10 to 20 degrees and that may become a bit of a problem over temperature, if you need to operate over wide temperatures, like middle temperature ranges.
Slide 94: So, in a tabulated form, here I compare the bipolar, CMOS, and J FET input amplifiers for the processes that the amplifiers are on, and I give you an idea as to which amplifier process to pick if you have an interest in the voltage noise or a current noise, and this way make it easy for you.
34
Slide 95: So, in summary, what I have done is I have explained as to why it is important to understand the low-noise design. I built
some foundations and fundamentals on the noise definitions and the noise types that are available. I gave you a general
formula that can be used for an inverting, non-inverting, different sound amplifiers under many configurations, and how to calculate the noise of a typical signal conditioning circuit, and I very briefly introduced you to a low-noise design process. I will go through these things a bit further in my next talk, and I show you a lot of dos and don'ts and different ways of optimizing a typical circuit in my February talk. talk. Gary? This ends my
Gary Cocker: Thank you, Reza, for that very nice presentation. We are
running a bit long, so we're going to have to truncate our Q & A a bit. score. I'm going to ask you to please bear with us on that Reza, how does popcorn noise behave with time?
Reza Moghimi: Wow, that's a great question and, actually, we get that quite often. Occurrence of the popcorn is quite random. An amplifier
35
may exhibit several pops per second during one observation period, and then it remains popless for several minutes. We
don't really have a screening -- I don't think anyone has a screening -- a perfect screening, to remove all noisy poles, although we have a pretty good understanding that what causes it and how we can catch the parts with very high probability.
Gary Cocker: Reza, I'm afraid our production team has given me the time-out signal. That's all we have time for today. I wish we could get
of your submitted questions will be answered by Reza via email shortly after the conclusion of this broadcast. in a question, it is not in vain. to your email for that. today. So if you send So look
It will be answered.
And thank you for attending today's webinar, "Noise This was
Optimization in Sensor Signal Conditioning Circuits." part one, presented by Analog Devices.
part two, which is going to be scheduled for next month. hope you'll join us there. your Q & A answers.
additional information and documentation, please direct your attention to the Analog Devices resource page that's opened
36
by Analog Devices, Incorporated, which is solely responsible for its content. The individual speakers are solely responsible for On behalf of our panelist
today, Reza Moghimi, and our entire webinar production team, I'm Gary Cocker. Thank you for joining us, and have a great day.
END OF TRANSCRIPT
37