I'm an Analog Applications Engineer, with the Precision Linear Group at Texas Instruments.

Transcription

1 PETE SEMIG: Hello, my name is Pete Semig. I'm an Analog Applications Engineer, with the Precision Linear Group at Texas Instruments. This presentation is entitled understanding current sensing applications and how to choose the right device. There are four main components to this presentation. They include fundamentals, devices, accuracy, and PCB layout and troubleshooting. There are four main goals of this presentation. First of, all at the end of this presentation you should be able to select a current sensing topology, and understand topics such as direct versus indirect current sensing, and high versus low-side side sensing. Secondly, you should be able to select the appropriate device for your current sensing topology. This includes the four main devices that are used for direct current sensing, which are: operational amplifiers, difference amplifiers, instrumentation amplifiers, and current shunt monitors. Thirdly, how do we calculate the accuracy of a solution? We will look at main sources of error, which include specifications such as initial input offset voltage, common-mode rejection ratio, and power supply rejection ratio. And finally, upon completing a design, if there's an error, how do we troubleshoot it? In other words, is the device out of specification? Or is there something else at play here? There are two main types of current sensing: indirect and direct. The majority of this presentation focuses on direct current sensing. However, I do have a couple of slides depicting what indirect current sensing is and what topologies and what sensors you can use in these applications.

2 Indirect current sensing is based on Ampere's and Faraday's laws. Essentially, a current through a wire will induce a magnetic field. A sensor that can measure the strength of this field can ultimately translate the strength of that field into a voltage, which you can then measure and get an idea of the relationship between the output voltage and the current running through the conductor. This type of sensing is noninvasive, meaning that the current sensing circuitry is completely isolated, and not in contact with the system bus or the system load. Therefore, this type of sensing is inherently isolated. Additionally, there is no power lost by the system because again, they are completely separate. And this type of sensing is recommended when you have very high currents, such as greater than 100 amps, very high voltage, 100s or 500 volts common-mode, you have very dynamic load currents, and when isolation is absolutely required. Now these are more guidelines than they are rules. You can certainly have direct current sensing for high load currents and high voltages. But just as a general guide, when you have very high load currents, we would certainly recommend you consider indirect current sensing. A couple of the solutions that TI has with regard to indirect current sensing include the following; first of all, if you have a hall effect sensor, you can interface a hall effect sensor directly with an analog-to-digital converter, or ADC. In this figure here, we interface a hall effect sensor with the ADS1208. Other devices that can be used for this type of topology include the ADS1205, 7861, 8361, et cetera.

3 As a matter of fact there's a good application note that I recommend you take a look at. It is SLAA286 and can be obtained at Another indirect sensing sensor that you may be interested in is the VAC sensor, or v-a-c sensor. This sensor is produced and specifically utilized with the TI's DRV401. TI's DRV401 is a custom solution for v-a-c sensors. You combine this with a SAR ADC, for example the ADS8509, and you have a complete indirect current sensing solution. In this particular case, the benefit of developing a device such as the DRV401, for a specific sensor, allows the overall solution to be less sensitive to temperature drift and initial offset than a hall effect sensor solution. Now for the majority of the presentation, or for the rest of the presentation, we will be discussing direct current sensing. Indirect current sensing we said was based on Ampere's and Faraday's Laws. Direct sensing, however, is based on Ohm's Law. In this type of sensing, we have a current shunt resistor, labeled R shunt, placed in series, with the load. Depending on the load current, a voltage will be generated across R shunt, that we will call V shunt. And that differential voltage can then be sensed, and amplified, by a differential amplifier. This topology, unlike indirect sensing, is invasive in that the shunt resistor is physically in series with the bus and the load, meaning that there will be a voltage drop generated, and power will be lost by the overall system that is being monitored. In this topology, the sensing circuitry, the differential amplifier in this case, is not isolated from the system load, unlike indirect current sensing.

4 However, this is a very common method for determining the current through a system load. It is highly recommended for smaller load currents, say less than 100 amps. It is also recommended when the system can tolerate power loss because there will be powered dissipated by the shunt resistor. It is also recommended for low voltage applications. And by low voltage, I'm actually referring to the common-mode voltage. Also when the load current is not very dynamic. So this methodology finds its place in DC, or low frequency applications. And also when isolation is not required. However, it should be said that you can use this topology and isolate as well. Now what do I mean by differential input amplifier? Differential input amplifiers are devices, or circuits, that amplify a differential signal while suppressing common-mode signals. An ideal differential amplifier will reject all common-mode signals and only amplify the differential signal. A differential input amplifier, by that definition, now includes: devices such as instrumentation amplifiers, current shunt monitors, difference amplifiers, and even the common operational amplifier. Notice the operational amplifier, by the strict definition of a differential input amplifier, does in fact have a differential input signal and a single-ended ended output signal. And its ideal common-mode rejection ratio is infinite. However, due to the large open-loop gain, operational amplifiers will require negative feedback to rein in that open-loop gain and become usable. But, by itself, it is strictly a differential input amplifier.

5 Before we move on, we need to define something called the common-mode voltage. In the figure, up and to the left, you'll see that the common-mode voltage is equal to the voltage at the positive terminal, or non-inverting terminal, plus the voltage at the negative terminal, or inverting terminal of an op amp, divided by 2. This translates to the average of the input voltages. That is the textbook definition, if you will, of common-mode voltage. For current sensing, it actually becomes beneficial to look at an alternate definition of common-mode voltage, in the lower right. In that figure, you'll see that we differentiate between common-mode voltage and an input differential voltage. If you perform Kirchhoff's Voltage Law around the input of the circuit in the lower right hand corner, you'll find ultimately that the common-mode voltage cancels out, and the output of the differential amplifier is equal to the differential amplifier's gain, A sub dm, times the input differential voltage. In this case, in other words in current sensing, that input differential voltage is typically going to be the shunt voltage across our shunt resistor. There are two places that you can put a shunt resistor for direct current sensing. The first, and preferred method, is the low-side sensing topology. In this scenario, we place the shunt resistor between the system load and ground. What is advantageous about this set up is that the common-mode voltage, as we just defined, is roughly equal to 0 volts. You'll notice the inverting terminal is connected to ground. And therefore, the common-mode is roughly equal to 0. This method is very straightforward.

6 It is also inexpensive, because as you'll see soon, we can actually use a plain old operational amplifier for this type of sensing. The cons however, include difficulty when detecting load shorts, because if you were to-- here we go, here's the cursor-- if you were to short the bus voltage here, straight to ground, you'll see that the shunt resistor is not in the circuit anymore. OK? So therefore, you can't detect load shorts to ground. It is also single-ended measurement. And we'll see why that can be a difficulty shortly. And also, if you look at the voltage that the system load is based on-- right here, the reference voltage-- the system load sees a ground potential, if you will, equal to V shunt. So therefore, the system is operating on a potential. Instead of exactly 0 volts, it's now V shunt. This becomes a difficult situation when you have systems that are perhaps communicating, where one is referenced off of V shunt and the other one is referenced literally to ground. You could have communications issues when one system is pedestaled and the other one is not. The second place that you can put a shunt resistor is on the high-side of the load. This is defined as being in series with the load, or in between the bus voltage and the system load. In this scenario, the pros include that you can, in fact, detect load shorts to ground. In other words, if we were to actually place the shunt resistor right next to the bus voltage--

7 in this case here if you can see my cursor, right in there, which is where it generally is placed-- any short from the bus to ground will actually include the shunt resistor. Secondly, this topology monitors the current directly from the source. However, one of the drawbacks to this topology is that the common-mode voltage is now roughly equal to the bus voltage. So how do you deal with systems where, say the bus voltage is 48 volts? We will see that you can't use things like basic operational amplifiers. So this solution may not be as inexpensive as low-side sensing with a basic operational amplifier. Directionality refers to the flow of the current, whether or not the system is sinking current or sourcing current. There are devices that can sense current in both directions, which are called bi-directional. And there are devices that can sense current in one direction, which are called uni-directional. Sometimes you can combine two uni-directional devices into a bi-directional solution. But this is an important part of the system that should be known when determining which device to use. And finally output. Operational amplifiers, difference amplifiers, and instrumentation amplifiers all typically have voltage outputs. They can readily be connected to analog-to-digital converters, digitized, quantified, and measured. However, current shunt monitors are a little more complex. They can have three different outputs. First of all, you'll see in the upper right hand corner the INA138/168 device actually has a current output. You'll notice Rl connected from pin 1 to ground.

8 Well the output is actually a current and one forced through that load resistance, Rl, that becomes a voltage. The advantage here is that Rl can be changed. In other words, you can select Rl and ultimately select the gain of the overall device, as you can see from the equation for Vo. One of the drawbacks of this topology is that it does require another resistor on the output to set the gain. And also, you'll notice on the inside of the device, these five kilohm resistors, right here, those devices, or those resistors, have to be trimmed to an absolute value. If they're not trimmed to an absolute value, as opposed to a ratio metric value, then the gain will be incorrect, because the gain is dependent on those values and Rl. In the lower left corner, you'll see the INA219. This device is a digital output device. You'll notice it also is a bit more complex than the rest of these in that it has a programmable gain amplifier on the front end. It has an analog-to-digital converter built in. It has registers that keep track of the power current and voltage. And finally, it has an I2C interface, which you can directly connect to a microcontroller, to read the current. And also these devices, such as the INA219, can also detect, or keep track of, the bus voltage, as well as the current to the load. And then finally, the third type of output in the lower right hand corner, as illustrated by the INA193, is

9 that of a voltage output. You will notice, in contrast to the INA138 in the upper right hand corner, that the 193 has Rl integrated into the device. So right there is Rl. This will fix the gain of the overall device thereby not requiring an external resistor. And also you'll notice a buffer at the output, allowing you to drive the signal. So those are the three different types of outputs for current shunt monitors: current output, digital output, and voltage output. Our next section we will discuss devices. The operational amplifier, as we discussed a bit earlier, is really only going to be good for low-side sensing. The reason why is it has a very large open-loop gain, and it must have feedback. So the input is essentially single-ended. It is referenced theoretically to ground, but it's not a differential input in the sense that an instrumentation amplifier, or difference amplifier, or a current shunt monitor, are differential inputs. Therefore, this op amp is really only useful on the low-side. Also since this is really a single-ended measurement, you have to beware of PCB layout parasitics. In this case here, the ideal situation is that we see that the current through the load goes through R shunt, generating a voltage, V shunt. It is then amplified by the op amp by a gain of 1 plus Rf over Rg. However, If we were to have some sort of a parasitic impedance to ground, such as a PCB parasitic, such as a pad, or a trace, that current, I load, will in fact generate a parasitic voltage, V sub p. Now the output of the amplifier is going to be the summation of V shunt and Vp, times the gain,

10 1 plus Rf over Rg, thereby introducing error. But nonetheless, on the low-side, op amp is certainly an option for sensing current. It is also relatively inexpensive in that it just requires an operation amplifier. It's generally advisable when selecting an op amp for the low-side to use a device that has a rail-to-rail input and a rail-to-rail output. Ideally, the rail-to-rail input would go below and above the supply rails, such as the OPA320. That is a device that I like to recommend for this sort of situation. The next device is that of a difference amplifier. As you see here the INA149 contains an op amp, along with four precision trimmed resistors, R1, R2, R3, and R4. In this case here, you see that we already have our negative feedback. And the output is going to equal the gain, which is set by the ratio of R2 to R1 times the difference between N plus and N minus. This type of device can be used for either high or low-side current sensing. It can tolerate very large common-mode voltages. For example, the INA149 has a common-mode voltage tolerance of up to plus and minus 275 volts. And that's with plus and minus 15 volt supplies. So the common-mode voltage seen at the inputs can greatly exceed the voltage that is supplied to the device, unlike operational amplifiers. The reason why the device can tolerate such large common-mode voltages is due to the resistive divider at the input pins, created by R1, R2, R3, and R4.

11 The problem is that the resistive elements at the inputs actually load down the system resistively. So depending on the common-mode and differential-mode impedance, the difference amplifier can place a burden on the bus voltage. So therefore it's advisable to ensure that the system impedance is significantly smaller than the difference amplifiers' input impedances. So using the INA459 for an example, you'll see in the chart at the bottom the differential input impedance is 800 kilohms and the common-mode impedance is 200 kilohms. So you would want your system load to have an impedance much, much less than those so that the difference amplifier will not place a significant load on the system. Here are two schematics depicting how to use a difference amplifier on both the high and the low-side. In the upper left hand corner, you'll see that the voltage is taken directly from V shunt, fed into the difference amplifier, and the output is equal to the shunt voltage times the differential gain. Now with difference amplifiers, because R1, R2, and R3 and R4 have to be trimmed, ratiometrically, so that the gain error is significantly small, the gain will be fixed. Typically let's say the gain is 1. Therefore if you need to add gain, you could buffer the output with an op amp, or perhaps select a different difference amplifier with a larger gain. In the lower right hand corner, you'll see the shunt resistor is now on the low-side of the system. And if you connect the inputs of the difference amplifier directly across the shunt resistor you'll see that any parasitic impedance to ground is no longer a factor when measuring the load current. Essentially we've taken that parasitic voltage V sub p out of the equation.

12 And the output of the difference amplifier's equal to V shunt times the differential mode gain. Now the difference amplifier, as I mentioned, has a fixed gain. Also we mentioned that the different amplifier places a load on the system due to the on board resistors. If we were to take a typical difference amplifier and buffer the inputs, such as shown in this picture, we would solve the issue of burdening the system because the inputs are now high impedance. Also, in this configuration, which is a typical three op amp instrumentation amplifier configuration, we have the ability to change the gain by using an external resistor, R sub g. So this picture depicts a three op amp instrumentation amplifier, which is made of a difference amplifier on the output, as you see right here, R1, R2, R3, and R4. This is the same topology as the previous slide. And then we have our two buffers. This first one here that's buffering minus N. This one here that is buffering plus N. And then here is our gain setting resistor, R sub g. So again, the pros to this is that the input impedance is extremely large because we're looking at the non-inverting terminal of a buffer. We can also change the gain with an external resistor. However, unlike the difference amplifier, instrumentation amplifier input common-mode range is limited by the supply voltage. This is a scenario similar to the operational amplifier. Therefore, it's usually used for low-side sensing. However, we have seen it used for high-side sensing depending on the common-mode voltage.

13 In this slide, we show both scenarios. In the upper right hand corner is a more typical scenario, where we see the low-side shunt resistor, this signal being fed into the instrumentation amplifier, differentially. And then the output is equal to V shunt times the differential mode gain of the instrumentation amplifier. Now again, don't forget we can change this gain by changing this resistor right here. This set-up also takes care of the issue of parasitic impedances to ground, because this is a differential measurement right across the shunt. A little more complex is where we have scenarios where we have on the high-side resistor, R shunt and V shunt, this voltage being fed into the instrumentation amplifier, and the output is equal to V shunt times the differential mode gain. However, in this scenario we must make sure that the supply voltage, VIA plus and VIA minus, are set such that the input common-mode of the instrumentation amplifier includes this bus voltage. So V bus must be within the common-mode range, given the instrumentation amplifier supply voltages. One such scenario where this would work is if V bus, for example, was 5 volts and you provided the instrumentation amplifier with plus and minus 15 volts. And then finally current shunt monitors. A current shunt monitor is an differential amplifier that has unique input-staged topologies. For example, in the lower right hand corner, you see a common-based topology. This allows the device to have input common-mode voltage range, well outside of the supply voltages. And also has a very large input impedance.

14 So it's kind of the best of all the worlds of the previous devices that we discussed. However, the devices typically only have a maximum common-mode voltage of say, 80 volts. So a current shunt monitor has the high-input impedance, can handle common-mode voltages outside of the range, but only up to voltages such as 80 volts. To where as a difference amplifier can have common-mode voltages, like the INA149, up to plus and minus 275 volts. However, it will load the system and have a fixed gain. Input and output voltage range is a very important specification to look at when deciding which device will work with a particular current sensing solution. So you must always ensure that the desired output and input voltage ranges are within the data sheet specifications. In particular, we see this becoming a problem with instrumentation amplifiers, especially traditional three op amp instrumentation amplifiers. For example, if we take a look at the INA826 data sheet, can we use it for low-side current sensing measurement? In other words, the common-mode voltage is equal to 0. If we look at the data sheet, we'll find a graph, as shown below, entitled input common-mode voltage versus output voltage. In this case here, we have a single supply of 5 volts with a gain of 100. And the red and blue lines correspond to reference voltages of 2 and 1/2 volts and 0 volts respectively. So if we look at the y-axis, and we find-- which it represents the common-mode voltage-- we find 0 volts common-mode.

15 And we see where it intersects the blue line, assuming that our preference is to have the reference grounded, we'll see that the output voltage, which is the x-axis, can only swing from just above 0 volts to about 1.2 volts, which is definitely not using the full linear range of the device. You also may be paying a penalty if you're feeding this into a data converter because you're not getting the full resolution. Alternately, if we were to operate the device with a dual supply-- in this case here in the lower right hand corner, we see that Vs is equal to plus and minus 5 volts-- and we look at the common-mode voltage equal to 0 volts, you'll see now that the output voltage can swing from negative 5 volts to positive 5 volts, a much larger output swing than the single supply case. So be wary of the input common-mode range and the output voltage swing when selecting any current sensing topology. There are also some unique topologies such as the INA326. This device is very interesting in that it's an instrumentation amplifier, and it actually has a rail-to-rail input and output capability, while using a single supply. That can come in very handy for low-side sensing measurements. The INA326 can have an input voltage range from V minus, which single supply is 0, minus 20 millivolts, to 100 millivolts above V plus. The output can swing to within 75 millivolts over temperature. And the typical spec is just five millivolts. However, these devices have only one kilohertz of bandwidth. So therefore, they're mainly useful in DC or low frequency current sensing.

16 The way that the INA326 gains this advantage, with respect to the input and output voltage ranges, is that it has a unique current-based topology internal to the device, as opposed to a voltage-based topology. So in summary, here's a slide that depicts which device you may want to look at first when selecting your solution, based on common-mode voltage and your approximate load current. If you have a small load current, and you'll notice that current shunt monitors can be used for any situation, either small or large load currents and any common-mode voltage, both low and high. Op amps, however, are only used when the common-mode voltage is equal to 0 volts. So you'll see right here, the op amps, right there and right there. And they can be used for both small and large load currents. Difference amplifiers, we typically recommend them for large load currents because that means that the system load impedance is small, thereby negating the effects of the input common-mode impedance. And difference amplifiers, we see can have a common-mode voltage anywhere from 0 to-- as we saw with the INA149-- plus or minus 275 volts. Instrumentation amplifiers, we've seen cases where they're actually used when the common-mode voltage is between that and the device. So that's why we added instrumentation amplifiers here. So that's a brief discussion of the four different types of current sensing devices and where you'll typically see them in solutions. The next section is on accuracy. Accuracy can be defined in one of two ways. The worst case definition of accuracy is where you simply add up all possible error sources and subtract it from

17 100, 100% accuracy meaning 0% error. And that is the absolute worst possible case. However we typically see a more probable estimation of accuracy by using the root-sum-square methodology, where you take each error term, square them all, add them together, take the square root of that sum, and then subtract it from 100. That is generally a more probable, or realistic, accuracy calculation. Now there are many sources of error in a current sensing application, the biggest of which is generally the input offset voltage, the initial input offset voltage, Vos. Secondly, we see common-mode rejection ratio and power supply rejection ratio as also being fairly large contributors to the overall error. In addition to those three, which we'll discuss in more detail later, we have error sources such as offset voltage drift, which is the change in offset voltage with respect to time. We see offset voltage shift, which is the change in-- I'm sorry, offset voltage drift is with respect to temperature-- and offset voltage shift, which is change in offset with respect to time. We also see gain error, which is referred to the output of the device. And we also see input offset current errors and the general shunt resistor tolerance error. All of these can add error to the measurement. So what is initial input offset voltage? Vos is defined as the DC voltage that must be applied between the input terminals to force the quiescent DC output voltage to 0, or some other level if specified. And it is typically the largest source of error.

18 Essentially the initial input offset voltage error is the voltage that is across the inputs due to mismatched input transistor pairs. When calculating error, we always calculate error with respect to the ideal voltage across the shunt resistor, or V shunt. So we take the ideal shunt resistance, and the ideal load current, and calculate V shunt. And use that in our error calculations. So let's take a look at an example of how initial input offset voltage can affect a measurement. Say we have offset voltage max of one millivolt, such as the INA170, the load current is 5 amps, nominally, and R shunt is equal to 1 milliohm. We can calculate the error as follows. The error due to Vos is going to equal the max offset voltage, which we said is 1 millivolt, divided by the ideal shunt voltage. The ideal shunt voltage is equal to the load current, 5 amps, times 1 milliohm, which is equal to 5 millivolts. So ultimately we see that we can have up to 20% error due to the initial input offset voltage in this design. One millivolt divided by 5 millivolts times 100 equals 20%. That is a fairly large source of error. So how would we decrease this error? We get this question very often. How do we decrease the error due to the initial input offset voltage. Well the initial input offset voltage is a specification of the device.

19 And basically, the designer does not have control over it, per se. If we look at the equation, in the very middle bottom of the page, right here, we see that Vos(max) divided by I load plus R shunt. So if we want this error to go down, we must either decrease the numerator or increase the denominator. So since we don't really have control over Vos, it's a specification of the device, the only way to decrease the error, in this case with respect to Vos is to select a device with better initial input offset voltage. An additional way to decrease the error is to look at the denominator. Well I load is typically fixed, right? That's a product of the actual system. However, we could change R shunt. Well in order to make the denominator bigger, we would have to make R shunt bigger. Now the problem is, is if you make R shunt larger, you will dissipate more power and be more invasive upon the system. So we could either decrease Vos(max) by selecting a new device, or we could increase V shunt, which is typically done by increasing R shunt. A third option, that I'm not going to go into detail in this presentation, but it is an option, is to calibrate. One could develop a circuit to input a known current, basically develop a current source. If you have a known shunt resistor value, you could then calibrate out the errors due to the initial input offset voltage and a few other errors. However, typically this adds quite a bit of complexity, both from a hardware and software perspective.

20 But it is an option. A second major source of error is common-mode rejection ratio. This is a measure of a device's ability to reject common-mode signals. As we said earlier an ideal differential amplifier rejects all common-mode signals. In reality, common-mode signals are always present, to some extent. And they can manifest themselves into differential signals, which will add to the error. Common-mode rejection ratio is typically specified in one of two ways: either in a linear scale or a logarithmic scale. The linear scale is typically microvolts per volt, in which case, if specified in this manner, the worst case is the maximum value. If listed as a logarithm, in decibels, or db, the worst case scenario is the minimum value. So in order to calculate the error due to common-mode rejection ratio we need three pieces of information: the worst case specification, which is given in the device's data sheet; the common-mode test condition from the data sheet, which we'll call Vcm-pds for product data sheet; and thirdly the system common-mode voltage, which we will label Vcm-sys, for system. That piece of information must come from the customer. So in this example, we will have the common-mode voltage of the system equal to 50 volts. We're given that from our customer. Again we will use the INA170. Again we will use R shunt equal to 1 milliohm. And I load equal to 5 amps, nominally. If we look at the data sheet, you'll see that V in plus, which represents the common-mode voltage upon which all of these specifications, unless otherwise noted, were determined, that is equal to 12 volts.

21 If we look in the input section, we see common-mode rejection is listed as a minimum value of 100 db. So as we said earlier, we're going to use the minimum value, since it was listed in db or decibels. So now we know the minimum value, or the worst case scenario of the common-mode rejection, and we know the Vcm-pds. The customer gave us Vcm system, now we can calculate the error. Before we do that, we will have to convert this 100dB into a linear scale. And we can do that as follows. It's the reciprocal of 10 to CMRR db min over 20, which ultimately comes to 10 microvolts per volt. So for every volt above or below the 12 volts as specified in the data sheet, we will induce 10 microvolts of error, referred to the input. You can think of this as an additional offset voltage. So now that we have the CMRR spec in a linear scale, we can calculate the error contribution as follows. First, we calculate the difference in voltage between the system common-mode voltage and the common-mode voltage that the device was specified at. So the system common-mode voltage from the customer is 50 volts. And the device was specified at 12 volts. So the difference is 38 volts. Those 38 volts are going to generate an error due to the 10 microvolt per volt CMRR spec. So if we take the 38 volts, multiply by the 10 microvolts per volt, that'll give us the input-referred offset voltage, due to the CMRR.

22 If we then divide that by our ideal shunt voltage of 5 millivolts, multiply by 100 to convert it into a percentage, we see that the CMRR spec in this design can add 7.6% error. This is in addition to the 20% error that we saw due to the initial input offset voltage of the device. So let me ask this again. How do we minimize the error? Again, if we look at the equation's numerator and denominator, we'll see that we either select a device with a better CMRR specification, or we increase the shunt resistor. So in other words, we either decrease the numerator or increase the denominator. One option that may come up is changing the system common-mode voltage. However, that is typically specified by the design and typically not an option. And then finally, you could also look into calibrating the system as mentioned before. Similar to common-mode rejection ratio, power supply rejection ratio can also introduce an offset voltage error. Power supply rejection ratio is the measure in the change of Vos created by a change in the power supply. So many times we specify devices given a particular power supply voltage; however when used in a real system, the customer may not have those rails available and wish to use a different set of rails. That can introduce power supply rejection ratio issues. So this calculation is very similar to CMRR in both specification and calculation. And we similarly need three pieces of information to calculate the error. First you need the worst case spec, from the data sheet, supply voltage test condition, also from the data sheet, and then finally the supply voltage that is actually being supplied by the customer, VS-sys. So in this case here, the system is powering the device with 30 volts.

23 Again, we'll use the INA170 and an R shunt of 1 milliohm, a load current of 5 amps, nominally, and we'll see from the data sheet that for this particular device, the power supply rejection ratio is given in a linear scale. So therefore, we'll use the maximum value of 10 microvolts per volt. We can also find Vs-pds at the very top of the electrical characteristics table as 5 volts. So similar to CMRR, we take the difference between the customer's supply voltage and the specification in the data sheet. So 30 minus 5, which is 25. And then we multiply that by the linear representation of the PSRR, which is 10 microvolts per volt, divided by the ideal shunt voltage, multiply it by 100, and we come up with 5%. So we saw 20% error due to initial Vos, 7.6% error due to CMRR, and 5% error due to PSRR. And similarly to CMRR, how do we reduce the error? Well you could look at calibration, first of all. Secondly, you could look at increasing the shunt resistor. Or thirdly, you could look at new devices that have a better PSRR specification. There are many other errors that we listed. These include gain error, shunt resistor tolerance, Vos drift and shift. Gain error is usually listed in the data sheet already as a percentage and is pretty straightforward to calculate, or to include in our accuracy calculation, because it's already given in a percentage. The shunt resistor tolerance is also given as a percentage and can be selected by the designer. However, we've seen issues where shunt resistors with poor temperature drift can introduce error.

24 So be cognizant of the temperature conditions of your system and how that can affect the resistance of your shunt resistor. Vos drift is also typically given in the data sheet and specified in microvolts per degree C. So for every degree C above or below 25C, for example, you will introduce a certain number of microvolts of offset voltage. That again needs to be compared to the shunt voltage and calculated as error. And then finally Vos shift. This topic comes up quite frequently for it is a measure of the change of Vos due to time. This specification is usually not specified and most likely never guaranteed. The rule of thumb that you could use when trying to include this in a very detailed analysis of the error is that over a 10 year period of time, the Vos specification, the initial Vos specification in the data sheet, may shift no more than the Vos max specification. And that's over a 10 year period of time. If you wanted to look at it month-to-month, that gets very, very difficult and actually can't really be determined, for we see Vos shift, most of the shifting takes place in the first few months of operation and tapers off later. Also it's important to understand that this is in addition to a device's initial input offset voltage specification. So therefore, if you wanted to calculate an absolute worst case scenario, you would take the Vos(max) specification and double it. For example when we calculated error due to Vos earlier, we found it to be 20%? If you were to double that, that would be the absolute worst case over a 10 year period of time.

25 So if we were to now put it all together, where we calculated initial offset voltage, CMRR error, PSRR error, and I'm actually using the equation for accuracy as opposed to error. So you see the 100 minus the summation. You see the absolute worst case accuracy, the worst case accuracy is 67.4%. A more realistic, which is the root-sum-square approach, gives us 78% accuracy. Now we've done all of these calculations with a given load current of 5 amps. Now this is not typically the scenario. Typically, we see customers who want to monitor a current over a particular range. So my question is what happens if our load current varies from 2 amps to 6 amps? We had done all these calculations based solely on 5 amps. Well as the load current decreases, our shunt voltage decreases. And as we saw earlier in our calculations of error, that is in the denominator. So as the denominator gets smaller, our error increases. So therefore, our worst case error will occur when the load current is at its minimum. So all of this accuracy would go down at 2 amps as opposed to what we calculated at 5 amps. So whenever you're calculating accuracy, always use the minimum load current. Now the final section, we're going to discuss PCB layout and troubleshooting. Let's revisit our accuracy table. So we had listed all these different types of errors that can be introduced into a system. And we notice that almost all of them are referred to the input, except for the gain error, which is given as a percentage in the data sheet. These are all specifications. They're all given in the data sheet.

26 So when increasing accuracy by looking at better devices, you can inherently get more accuracy. The question becomes are they under the control of the designer? Well, not really, because they're actually specifications of the device. The designer can select a different device, for example, but can't necessarily do anything to mitigate errors due to these specifications. So none of them are really under the designer control except, I'm going to introduce a new error called offset voltage due to PCB layout, or V OS-PCB in the lower row. This error can be exacerbated by a PCB layout. In addition, the input bias current has been added to this as an error for input bias currents to the device, along with a non-ideal PCB layout, can introduce a differential voltage at the input thereby introducing error. And I wanted to note that Ib is defined as the average of the currents into the two input terminals, with the output at a specified level. So in order to understand the effects of PCB layout on accuracy, let's go through a quick example design. The requirements given for this design: our common-mode voltage of 70 volts; and available supply for the differential amplifier of only 5 volts; the load current range, so now we're given a range as opposed to a nominal value, the range will go from 5 amps to 10 amps. It's a uni-directional measurement. And the customer requests that it's as accurate as possible, or we need high accuracy. So given these specifications, first of all I would not recommend using an op amp because the common-mode voltage is very, very high. It's up to 70 volts.

27 Similarly, you can't use an instrumentation amplifier. And since the request is for high accuracy, I would prefer to look at current shunts monitors over difference amplifiers because of the input impedance of the difference amplifiers. Also our current shunt monitors, basically they have a higher input impedance than the difference amplifiers. And we also have current shunt monitors whose common-mode voltage can reach 70 volts. Also there was no mention of bandwidth, so we'll assume a DC measurement. The device that I would recommend looking at would therefore be the INA282. This device has a common-mode voltage range from negative 14 volts to 80 volts, so it includes the 70 volts spec from the requirement. The initial input offset voltage is 70 microvolts. The supply voltage, in other words, the device can operate from a supply voltage of 2.7 to 18 volts. And it's also a bi-directional device. So first we need to size the shunt resistor. This question actually comes up quite frequently. So in order to size the shunt resistor properly, we need to understand what the range of the load current is, which we already given, 5 to 10 amps, and what the available supply voltage is, because the available supply voltage will restrict our output voltage swing and our input common-mode voltage range. Not in this case, because we have a current shunt monitor, but it certainly should be looked at if we were looking at an instrumentation amplifier. So in this example, we have a load current from 5 amps to 10 amps, a supply voltage of 5

28 volts. So the strategy is to obtain the output voltage range from the data sheet, given the supply voltage of 5 volts, refer this voltage back to the input by dividing by the device's gain. So current shunt monitors typically have, except for current output, voltage output current shunt monitors will have a fixed gain. So we can refer the output range back to the input by dividing by the gain. And then we can use the load range min and max to find the range of acceptable shunt resistor values. So the input and output range of the INA282 are as follows. The output can swing from 40 millivolts within ground, or from within 40 millivolts of ground, to 400 millivolts of Vs. And since Vs is equal to 5 volts, therefore V out can swing from 40 millivolts to 4.6 volts. If we refer this back to the input of the device by dividing by the 282's gain, which is 50 volts per volt, we'll see that the input can operate from 800 microvolts to 92 millivolts. That would be the linear operating range of the device for this set-up. This allows us to calculate the minimum and maximum values of the shunt resistance. So to calculate these, we use Ohm's Law, the V in max divided by I load max is equal to R shunt max. So we take 92 millivolts divided by the 10 amps equals 9.2 milliohms. So that's the maximum shunt value. Anything greater than that is going to give us too large of a voltage such that will saturate the output. And then similarly, we can calculate this on the low-side of the load, and calculate V in min divided by I load min. And that equals 0.16 milliohms. The trade-off here is that you want a large shunt resistor to minimize errors, but also the smaller the shunt resistor, the less power it'll dissipate.

29 So there's the trade-off. In general, I recommend using the largest shunt resistor that your system can tolerate. Put succinctly, the R shunt value is directly related to both power and accuracy. And you typically want to minimize power and maximize accuracy. So therefore use the largest R shunt your system can tolerate. Using TINA-TI, I developed this schematic. And you'll see that I ultimately chose an R shunt value of 8 milliohms. I can sweep the load using a current source from 5 amps to 10 amps. And I set the common-mode voltage to 70 volts. You'll notice the INA282 is operating off a 5 volt supply. And I place the 10 killohm load and plotted load current versus output voltage. So as the load current is swept from 5 amps to 10 amps, we see that the output is linear from 2 volts to 4 volts. This can be fed into an ADC, digitized and measured. However, when we actually implement this example on a PCB, we will introduce parasitics. And these parasitics can significantly decrease our overall accuracy. The first two parasitics of note are Rpp and Rpn. These parasitics represent PCB traces that are in series with the inputs of the device. They can interact with the input bias currents. And if uneven, can introduce a differential voltage, thereby introducing error. The third, and perhaps most significant parasitic, is Rps, which is a parasitic trace in series with the load current. This parasitic can greatly affect our accuracy. As we'll see shortly. So ideally, we've been using V shunt as equaling I load times R shunt.

30 And ideally, the voltage across the shunt resistor is the exact same voltage as seen across the input pins of the device. However, once put in lay-out, this is not the case because of parasitics and PCB traces. So what we see is that, or what I do is I define a new voltage called V sense, where V shunt is the voltage across the shunt resistor, V sense is the voltage at the pins of the device. Now due to the parasitics due to the PCB layout, and the traces, V shunt is not equal to V sense. Having any imbalance there can introduce differential voltage and introduce error, as shown by this slide. V shunt and V sense are different. I can therefore define Vos due to PCB as the difference in the voltage seen across the shunt resistor, and the voltage measured at the pins of the device, or V sense. This error can then be calculated as V OS-PCB divided by the ideal shunt voltage, as we did previously, multiplied by 100. Now if you take this circuit in the lower left hand corner and use Kirchhoff's Voltage Law, and Kirchhoff's Current Law, and Ohm's Law, and you solve, ultimately for the voltage across the pins of the device, or V sense, you'll find that V sense equals V shunt. That's the ideal case, plus a bunch of error. Now the error is I shunt times Rps, where Rps is the parasitic in series with the shunt resistor, plus the quantity of Ibp, which is the bias current into the positive terminal, times Rpp, minus Ibn times Rpn. You'll notice that the box on the right, Ibp Rpp, minus Ibn Rpn, is actually a difference term.

31 So there is the possible case where those two terms, Ibp Rpp and Ibn Rpn, are equal. If they're equal, then their difference is zero. So the conclusion from that is that the input bias currents may or may not induce an error. However, any parasitic in series with the shunt, as represented by Rps, is additive to V shunt, as shown in the box in the middle. So therefore, the conclusion is that any parasitic resistance in series with the shunt will introduce an error. So how bad can Rpn and Rpp be? What we're going to find is that the parasitic impedances in series with the inputs generally aren't as significant as parasitics in series with the shunt. This first example, what we do is we calculate the approximate impedance of a trace that's 350mils in length and 10mils wide. And that turns out to be approximately 17 milliohms. If we set our Rpp and Rpn to 17 milliohms, we'll see that V shunt's-- well, so here's Rpn equals 17 milliohms. Here's Rpp equals 17 milliohms. We look at V shunt as being millivolts and that's different from V sense, which is millivolts. So these impedances do in fact introduce some sort of an error. If we put that in our error calculator, we see that comes to only micropercent, so it's not really that great. However there are some things that you can do that exacerbate this, and we'll talk about those shortly. If we were to have unbalanced traces, so in this case here we're going to have one trace 350mils long

32 and the other trace 275mils long, but both of the same width, we'll have Rpn equal to 13 milliohms, and Rpp equal to 17 milliohms. This actually creates a larger difference between V sense and V shunt. And we see we actually increased this error to 722 micropercent. Again, it's not by itself a large error, in and of itself. However, we notice that if there's any impedance we get some error. And if the impedances are imbalanced, we could exacerbate this error. So the observation's that they're generally not that significant, especially with comparison to Rps. Also, some things you can do to exacerbate this is by placing the device on opposite side of the board from the shunt resistor. We've seen cases where there's a shunt resistor on one side of a very thick board, and those input traces go through many layers of vias. Not having a well controlled PCB process can certainly exacerbate differential voltages and introduce this error, and make it even worse. Also, those vias can drift significantly with temperature and process. In addition, we would like to observe that trying to balance out the difference term, as noted earlier, the Ibp times Rpp, such that that term is equal to Ibn times Rpn, is not recommended. Ib can vary part-to-part, wafer-to-wafer, lot-to-lot, and it can also vary based on system parameters, such as common-mode voltage and power supply. So there's generally nothing you can really do, with regard to that equation. However, from a PCB layout perspective, we recommend that you make the traces as short as possible.