Designing a Multi-Channel 4-20mA Analog Input Module
This training helps to understand relevant parameters of a 4-20mA analog input module. These parameters include surge protection, handling miss-wiring, broken wire detection, as well as isolation or protection of the module against over-current at power and signal inputs. The training will provide a selection guide to proper components, configuration of the ADC, power tree options, and finally show a possible solution.
Resources
Hello. Welcome to designing a multichannel 4-20 milliamp pair analog input module. My name is Lars Lotzenburger, and I'm a systems engineer in the factory automation and control team in Texas Instruments. This is the agenda. First, we have a look at the purpose the 4-20 milliamp pair analog input module. You And then we try to design such a module, which includes requirements, A-to-D converter selection, burden resistor selection, ADC setup, errors, protection, and finally power.
So what's the purpose? We need a 4-20 milliamp pair input module to actually measure current, of course. The current is actually dictated by the sensor transmitter. You can actually measure temperature, pressure, whatever you like. The 24-volt power supply supplies the sensor transmitter. And the sensor transmitter itself outputs the right site by the 4-20 milliamp pair, which is then measured over the burden resistor and sent by A-to-D converter.
Requirements of our input module are we want to do an 8-channel 4-20 milliamp pair analog input module. We want to do it group isolated, which means only one group, so all input channels share one ground. Then we have a solution of 16 bits, which is 15-bit plus 1 sign bit. Then we want to reject interference frequencies of 16.67 hertz, 50, 60, or 400 hertz.
Our update rate per channel should be at least 60 milliseconds for the 16 hertz, 200 milliseconds, 166 milliseconds, and 25 millisecond for the 400 hertz. The entire module power dissipation should be less than one watt. And in terms of accuracy, you want to have 1.2% full scale at 25 degree, and a 0.2% full scale error over the temperature range.
But before we start, before we really dive into it, one thing I need to mention, the 4-20 milliamp pair are not really 4-20 milliamp pair. So we have to comply with the NAMUR NE43 recommendation, which actually specifies 3.6 to 21 million amp. So you can see here in the picture below that this is an extended range.
So we have the normal range of 4-20 milliamp pair. And then we have an extended working range from 3.8 to 20.5 milliamp. And then we have this colored, rose-ish zone here, which actually is the area up to 21, which actually leads then to failure information. The same at the lower end coming from 3.8 to 3.6, we have the failure information there.
So the heart of our module is the LD converter. No doubt about it. OK, but which one do we take, delta-sigma or SAR converter? For this, we need to check what we really want to do. We need precision at the DC level because our 4-20 milliamp pair loop is just a few hertz of bandwidth.
We want to reject our famous frequencies we mentioned a couple of times before, which is really asking for SINC filter. And so SINC filters are made for that. And finally, we may want to gain up the small analog input voltage. I will tell you in a minute why you want to do this. So it would be nice if you have already a gain stage. And all these three bullets are greatly covered by a delta-sigma converter. So we take a delta-sigma converter.
And the next thing is, OK, once we decided for that, what about the sampling, the architecture? Do we want to do a multiplex or maybe some simultaneous sampling. Because we have a group isolated, this means we can do this in one device.
If you would have it channel-to-channel isolated, we would have shared grounds. We could not use a multichannel because each channel needs to be sampled and would have its own ground. But that is not the case here, so we could actually use it. You could actually run the module with one A-to-D converter.
The second one is the frequency. So what does it mean? Rejection of the frequency, it gives the maximum data rate this means. If I want to do 400 hertz rejection, I need 400 samples per seconds, which actually it converts into 2.5 millisecond conversion time. So I would have at least 2.5 millisecond per channel for the conversion for the 400 hertz, which is my highest frequency.
But for this frequency, for this rejection frequency, I need a required update rate of 25 millisecond as seen in the requirements before. This means we have a update rate per channel of 25 millisecond, which is 10 times higher than the conversion time. Which means, hey, this is easy. So we can use a multiplexing architecture to fulfill our needs.
So now that we put all our needs together, we actually go to dataconverter.ti.com and put all the requirements in our filter. So a solution of 16 bit plus input channels of 8 plus separate. I put here 1 kiloseconds per second. We actually need 400 as seen before. But we increase it to 1 kilosecond per second, just to make sure. We need the operating temperature of 125 degree or more and then we sort for price.
And then we come up with the ADS114S08. This is a quite new A-to-D converter, which actually has all we need for our application now. It has 12 single-ended inputs. So this works. We need 8. It has a very low noise. It has a gain amplifier in it. It has a data rate from 2.5 samples per second up to 4 kilo samples per second. Perfect.
And it has internal reference, internal clock. So we don't need to take care about all this other stuff. So this is all inside. Although you could, of course, also have an external reference if you find the internal is not good enough. Although, with 10 ppm over temperature, it's very good reference.
Now we have the A-to-D converter. Now you want to connect a burden resistor to it. So the burden resistor measures actually the current, or let's say translates the current, into a voltage. So the important thing is we have an 16-bit converter here. The ADS114S08 is a 16-bit converter. And we want to have 16-bit output. So for that, we would need the full range of the inputs.
So this means ideally, we have an input to the A-to-D converter of 2.5 volts. So 2.5 volts, quite easy calculation. Our burden, 2.5 volt divided by 21 milliamp is our maximum current you would expect. We end up with 119 ohm, so this is like an E-series. Translated, this would be a 180 ohm. So the voltage drop, however, is quite high.
The customers today for such a module try to differentiate by a very low voltage drop. Imagine you have your 24 volt as seen before, and you have your sensor transmitter, and you have your input. So and the drop over the analog input module is dictated, among others. Protection is also dictated by the burden resistor. And the smaller the burden resister gets, that's to lower is the voltage drop. And is more voltage is available for the entire loop, right?
The sensor transmitter needs a certain voltage. You have a line. You have a drop over the lines, over the wires, so you would like to maintain as much voltage as possible for the loop and not wasting, let's put it this way, erasing voltages inside the module. So what can we do here? We can actually use our built-in PGA and use a smaller resister.
So for example, if you take the gain of 4, we can reduce the burden from 118 ohm down to 29.4 ohm. This actually leads to a voltage drop of about 617 millivolt, which is four times less compared to the 2.5 volt with the 118 ohm. And apart from that, we also have a much low power dissipation. With the 118 ohm, we had 52 milliwatt per channel, while you have now with the smaller burden resistor only 13 milliwatt.
This means with eight channels, you would waste, or you would power dissipate, or you would dissipate already 400 milliwatts for the entire module just for burden resistor. While for the smaller one, it's only about 100 milliwatt. So this is much smaller. You don't get the temperature increase so high inside the module. Yeah, this has a lot of advantages.
And then of course, once we have now decided for the resistor value, we have to make sure that this one has to be very precise. So we see later that the absolute precision and the drift over temperature has a quite a big impact on the overall system error. That's why we have to buy not just an off-the-shelf 1% burden resistor, but we would take here at least 0.01% and 10 ppm per degree resistor. And is this still like affordable. There are much more expensive resistors, but then it can get hard to justify this price. So this is actually a $0.50 resistor at 5 kilo units, so at a very or quite a good performance level.
So then we have our A-to-D converter. We have our burden resistor. How do you connect this burden resistor now to our A-to-D converter? I can think of three options here. The first option is we have a bipolar operation, which means we have a plus/minus 2.5 volt. So you have to know the ADS114S08 can be run in a unipolar mode or in the bipolar.
In the bipolar, you can actually supply plus/minus 2.5 volts, while in a unipolar mode, you can actually connect ground to AVSS. And AVDD can be in the range of 2.7 to 5 volts. So for the first one, we have plus/minus 2.5 is set. The good thing, is we have high PGA gains. And I will tell in a minute why it is. We have a lower system voltage drop of 0.617 volts. And the negative one, OK, power 3 gets a bit more expensive because you need a negative voltage.
Then we have the unipolar operation. We see here ground and the unipolar voltage, simple power tree. You just need one voltage. Same voltage drop, but the problem is, the burden is at the negative edge of the input range. So here is the ground. And here is the ground. So this means, the burden really hangs on one edge of the analog input range. This increases actually our input leakage current. And at the end, no gain is possible.
And then, the enhanced unipolar operation, which actually adds up some, must add some diodes here. And these diodes are used to shift up the current mode of the input signal. So we have a simple power tree. Here, we can do PGA up to 4 again, which is nice. We have for 3 a reverse current protection. But we have a really high system voltage because we have to consider the burden resistor drop and two diode drops. And this hertz over temperature, it can be up to 2.7 volt, which is really high compared to 0.6 volts.
And the major drawback is here. We need two analog inputs per input channel because we cannot really tie all the burdens on the negative together. Here, we have ground. So all the grounds for all the burdens go to one point, here as well. But here, it is not possible because between ground and our negative, there are these two diodes. And this means we would need two inputs per channel, which actually means we would need a second ADS114S08. And this, of course, is costly.
That's why we vote for option number one. This is bipolar power supply, shared ground among all the burden resistors. And the positive thing is really our current mode is near the mid input range. And this enables high PGA gains. And here, actually, I put all the resistors in a table you can use for certain gains. Not all make sense, for sure. But these are the options for the gain stages. So we can have from 119 what we measured previously before, down to a 3.72.
There are even more gain stages, 64 and 128, but these are not considered here. You see the power dissipation. We are in a good range with the 13 millivolt here. We have the maximum drop. This is not the burden drop, but it is the drop accepted by the analog input of 0.625 volt, which is not really much, but higher than the 0.614 from the burden.
Actually now, to understand how our input looks like, we have an nice calculation tool. And you can find it on the product folder from the ADS114S08. Here, you can actually among other stuff, like register maps, SINC filter, and so on, you can also check your analog input. For this, you actually put your PGA gain, which is 4 in our case; the supply voltage, which is plus/minus 2.5 in our case; and the reference voltage, which is 2.5. This is the internet reference voltage.
Then you see here the input voltage. So this is also what you type in. So since our burden is connected to ground, it's a 0 volt. And our 29.4 ohm is here, 0.617 volts. And this gives us this range, from 0 to 0.6 volt. These dashed lines are dictated by our power supply, plus/minus 2.5 volt.
Because the PGA needs about 150 millivolt on both sides as a headroom, our signal, which is going to the A to D converter can be between these red lines. But since we are in the middle here of our input range, we can greatly gain by 4, and we are still fine here. And this is what I meant with if your analog input range sticks to one edge, for example, for the 0 volts. Well, imagine you have here 0 volt, and here you have the 2.7 to 5 volt for the unipolar. Then you could not gain, because you will always violate the output of the PGA, because you would go negative.
So you have to stay as much as you can inside. You have the ground mode should be like really inside analog input range. And then we can gain to 4 or even to 8. For 8, it would go again a bit higher here. So we would use a smaller resistor for that.
Then we check now the analog stuff, the AFE. Everything is fine. So now, we talk about the A-to-D converter setup. A data [INAUDIBLE] converter can be quite feature rich. And so this data converter, data [INAUDIBLE] converter, is. Let's talk about the data rate. I said for the 20 milliamp pair loop, we have just a few hertz of changing. Here is a very small bandwidth. That's why you don't need much conversion time. But we want to actually reject frequencies.
So if you connect your PSE, for example, to main's grid in Europe or China, you want to suppress the 50 hertz. In US, this is 60 hertz. 400 hertz for the aircraft, so the spacecrafts, submarines, and so on. And for the railways, we see 16.67 hertz. How can we actually suppress that? This is not like that you can choose any data rate and you get the suppression. No. We have to actually be in line with the SINC filter.
So the SINC filter is like this. You actually want to remove 50 hertz, for example, so if you take a data rate of 50 samples per second, the first notch comes at 50, the second at 100, 150, and so on. So if you would have, for example, a rejection of 60 hertz, then you would choose a data rate of 60 hertz. Of course, you can also use data rate of 10 hertz or 20 hertz, because the notches come every multiple of. No problem.
And yes, on one specialty, on the 20 samples per second there is the possibility to reject 50 and 60 hertz at the same time. So what did we learn? So the maximum data rate, basically your frequency you want to suppress, you want to reject. And you can have also a multiple of this frequency lower.
So then we talk about the filter. We have inside the ADS114S08 two filter schemes, one SINC-free filter and one low-latency filter. The SINC-free filter is a very good filter in terms of noise. It's better performant than the low-latency filter because it filters more, has deeper filter curves. But you need three cycles of settling until you actually can use the data.
You can see in the picture down below here. So SINC-free filter, you have one sampling, second sampling, third sampling. And then you get a data ready signal, which actually indicates there is a sample available. So this means every time you switch a channel, for example, and this is what we do 8 times, right? We have 8 input channels, so we switch them. So you would need to rate 3 samples until you get a valid side. This takes too long.
That's why we have also a low-latency filter. This settles after the first sample. OK, we have to add a very small delay for the first sample. But this is like almost negligible, or it's negligible. And that's why we take this one. So since we do MUX, we do the low-latency filter.
For the conversion mode, we have a single-shot mode. This means you send the stop command. You wait until all the filters settle. You get the data ready. And you can get the data out of it, out of A-to-D converter, and you're done. Where in the continuous mode, you start and this guy samples, samples, samples all the time, pushes out data until you send the stop command.
You may think, OK, we a single shot, because we want to do a single shot, switch to the next channel, single shot one sample, switch to the next sample, and so on. Well, this is right, but we have a nice feature inside the ADS114S08 which works as follows. You'll run in the continuous mode and when you write to a MUX register-- for example, because you want to switch the input channel-- and you write to it, it automatically stops the ongoing conversion.
It adds a delay for the analog input signal to settle. And then it restarts a conversion with the new channel. So this way, you don't need to send stop command on all the time, as you would do in a single-shot mode. And this actually speeds up your bandwidth. This eases your communication. You just simply save one communication per sample. So we would go for the continuous mode.
Then the PGA, of course, it's enabled. We want to gain up by 4. And the good side effect, the positive side effect here is, when we enable the PGA, we have a higher input impedance, because the signal sees the input not of the A-to-D converter, but of the PGA.
Then we have a nice feature which is called chopping. So this chopping means we take two samples taken in with reverse polarity to the average. So we can actually cancel the offset. But this, of course, as I said, takes two samples, so it takes time again. So this is not used.
You can still perform offset calibration at the beginning. It doesn't help for the drift, but for the initial offset hour. But you will see in the later slide that the offset is not our biggest concern. And that's why this is not worth to turn on the chop.
And then finally, clock. The internal clock is used. This what we do by design. You can always use an external clock, but no need to do. Same is for VREF. We have a good internal reference here. And unless you have somewhere in the system precise voltage, lower drift, you can use it, no problem. But it this is for our example here, this is.
So here is just an overview of the data rates. This is all famous data rates-- 16.67 samples, 50, 60, and 400 again with the low-latency filter and the chop off, and this is actually what you want to go. Then you have a 61 millisecond conversion time per channel. So if you, say I want to send it this channel, it takes you 61 milliseconds until you have the [INAUDIBLE] available.
If your seed is for all 8 input channels, so you have to divide that, right? So and this means every single channel will be updated every 2 hertz. So it goes round robin principle, first channel, second channel, third channel, up to 8 channels, the again, first channel. So and each of those channels, so for example channel 1, gets updated every 2 hertz. This, of course, raises a bit for the higher data rates-- 5.9, 7.3, and 47 finally for the 400 samples per second. So and this is well in our specification.
So here we see the accuracy our contributions contribute us. So we have a lot of errors, right? We have offset error. We have an offset drift. We have a gain error. We have a gain drift. We have an INL error. And we have like the overall error, which is the like square and the root out of it of all these errors. Here.
You see already the offset error of the drift is really lower compared to the gain error and the gain drift. So that's why we don't really need to offset or compensate, to offset or compensate here our system. So we have we end up with an overall error for the A-to-D converter of 209 ppm.
Then we have the VREF. The VREF, the reference, has a gain error of 500 and a drift of 680. So it's much higher. And the burden resistor is even higher. We have a 100 ppm. This is our 0.01%. And then we have 1,250 offset. And this is actually not at 85 as written here on this. This is more at 125. So at the end over 125 Kelvin we have 0.15% of error overall. And at 25, so removing all drifts, we end up with even 0.05%. So we are also in the limits.
The precision error contributors, what is precision? Precision is a repeatability. So if I have a constant voltage or a constant current, in this case, and if I sample 100 times, how big is my deviation? How big is my standard deviation at the end? So this is basically noise.
So we have the current loop noise. We have passives, like the burden resistor, which is noisy, and also the [INAUDIBLE] and the capacitor. We have the PGA and the A-to-D converter, which is noisy, thermal noise, quantization noise for the A-to-D converter. We have a VREF. We have the power supply, which is not the cleanest in the world, most likely. We have a clock jitter. We have layout, like noise caused by layout. And we have also the signal bandwidth.
So we filter. You want to get rid of all this noise in a certain way that we actually keep the signal, but remove all the high-frequency noise. All these noise sources actually dictate our repeatability. But this, we'll take not too far, and that's why I just mentioned this. I'm just talking about accuracy here in this presentation. Precision is a different story.
So now that we kind of covered the A-to-D converter, let's talk about the signal input protection. So the signal input protection is at the input of the terminal, of the pins where the, actually, the 4-20 milliamp pair loop comes in. We can actually, we try to protect against ESD, EFT, and surge. Our quite new TVS3300 TVS diode can actually clamp the high energy at 1 kilovolt and 42 ohm, which means 24 amps you can clamp or this TVS diode can clamp, as you can see here, to almost 40 volt, so at 30, even to 35. So we asked for 24 only, but even at 35, we have a clamping voltage less than 40 volts.
And this actually brings us directly to our series resistor of the anti-aliasing filter. The anti-aliasing filter has two purposes here. The first purpose is to protect our input. So the specification of the ADS114S08 says do not put more than 20 milliamp pair into the analog input pins. This means the protection diode cannot handle more than 10 milliamps.
So we assume that we have never more than 40 volts in case of a surge event. And 54 kiloohm, we actually are safe. So we are at 10 milliamps. So the anti-aliasing filter, of course, the resistors to cover with this capacitor for the anti-aliasing filter. And here, we have to make sure that we remove as much noise as possible by putting the 3 dB corner frequency mode to lower frequencies. But it cannot be too low, because then our signal would take too long to settle when we switch.
So each input, as you know, each input channel has its anti-aliasing filter. And if you switch around, then we need to settle. And if the [INAUDIBLE] sequence of our anti-aliasing filter is too high, it doesn't settle and we cannot meet our requirements. So it has to be somewhere in the middle. And this is really application specific. If you, for example, need less update rate, you can have a higher [INAUDIBLE] sequence.
But this does not really help us against miswiring. So we talked just about surges, right? But miswiring means that the installation guy connected the 24-volt power supply directly to our input. What will happen? The 24 volt, like the 24 volt, virtually unlimited current will flow over the 29.4 ohm resistor over our burden resistor. And the burden resistor would need to handle like tens of watts.
But normally, they, like this one, just can survive 100 milliwatts. So what can we do here? The TVS diode will not help because it's clamping only at 33 volt and not at 24. So we need some kind of PTC or some other protection scheme which actually prevents our burden resistor from dying the heat death.
So the input power protection, like the input power to 24 volt which is coming in, if you use the field power, you can actually use the TPS2660. This has an input voltage of 55 volts. So if you put there a TVS diode, you can clamp this to 40, 50 volts. So there is no problem, because this device can handle up to 55 volt. We have an adjustable current limit, including in monitor for the current. We have under voltage lock out and over voltage cut off.
We have an output slew rate control. And we also have reverse current blocking, of course. This is against reverse polarity. We have over temperature reporting. And then we can during a fault event, we can either auto retry or shut off until it's manually.
Signal isolation, of course, we need signal insulation as well. So we have our PSE side, which is kind of I would call it the cold side. And we have the hot side, which is the side where the A-to-D converter sits. And we need four SPI channel as minimum. And they need to be isolated. So we use the ISO 7741 from Texas Instruments, for example.
And if you need more, for example, you want to control the A-to-D converter stop, and which has dedicated stop, and so you can send the command. This is what we did. And you can have the dedicated stop in. But this is not required here, so we just save this isolation channel.
The same for the data ready. We know, and this is written in the specification, the data sheet of the ADS114S08. We know when our data is available. And on top of that, we have also a bit which says new data available or not. So we really know when new data arrive, so we don't need this data ready pin. So this also not required here.
On top of that, we have for the ADS114S08, we have 12 input channels. And for the 12 input channels, we have also, we have some GPIOs left. And these GPIO can be freely programmed. And you can actually control some LEDs or something as you wish. So you have some GPIO capability on the hot side.
The alternative is, of course, we can put a microcontroller at the hot side, so we have unlimited GPIOs. And we can then talk to the PSE side with a smaller, where less channel interface, like you UART, 12C. I think the bandwidth for the data generated by the ADS, this can be handled by the UART and as 12C at bandwidth.
If you use the isolator, you of course, have also a free voltage translator. So you have the 3.3 volt at the hot side for the digital of the A-to-D converter. And then you have some I call it VIO, but it can be like anything from I think 3 volt, or even 2.7, up to 5 volt. So you have a three-level translator here with this isolator as well. Therefore, in terms of isolation, if it's basic isolation or reinforced isolation, this is really up to the application. So if you, for example, have a safety module you would like to target, enforced isolation is most likely the choice.
So then let's talk about the DC/DC power supply. So we have here an input voltage from the field side, from the hot side, from 24 volt plus/minus, minus 15% plus 20%. Or we have a power supply the backplane, so from the PSE for 3.3 volt to 24 volt, something like this. So this varies from vendor to vendor. So this actually enables us to give us certain power options.
But before we go to these power options, let's have a look how much power do we actually need? So the power consumption overall is only 60 milliwatt. This includes the A-to-D converter with 2.54 milliwatt maximum and the ISO7741, which takes about 13.5 milliwatt at the hot side.
For the plus/minus 2.5 volt, this is what we need, what we agreed on for the analog. We take the LM27762, which is a charge, like a charge pump with LDOs, afterwards. So this means you get the 5 volt in here. And or let's say, you can have up from 2.75 to 5.5 volt in. And the output can be higher than 2.5 plus/minus. But we use, of course, the 2.5 plus/minus. I said two charge pumps would take care of the plus/minus. And the LDO, with the feedback, you can set to 2.5 volt plus/minus.
So for the non-isolated DC/DC power supply, so we can think of the 24 volts from the hot side. And here, we need protection. If you don't have any issues, I just mentioned before, we would need to protect against surges. And that's why it's a good idea to have always a high V in for the first stage.
So for LM5165, we can have up to 65 volt input. And this is in DC/DC converter. And for the TPS7A4101, which is an LDO, we have a maximum V in of 50 watt. And we have given an LDO which can go up to 100 volts. Again, if you need, if you have the e-Fuse in place, you may use also a lower V in devices because you are protected by the e-Fuse already.
So in terms of efficiency, of course, the DC/DC is always better. So we have here an efficiency up to 80%, while on the LDO, we have only 14% because we have two. I mean, the current is really low, but the drop, the voltage drop, is really high with almost here of more than 20 points.
Another option would be we take the power from the backend side. But then we need to isolate the power, of course. So we can use this, we can do this with a push-pull converter with the SN6501 and a transformer. And then we actify it. And then we have an unregulated voltage coming out of the transformer. And this is unregulated because different current will cause-- like different currents needed by the application will actually cause a different voltage drop across the diodes.
And when the or if the VIO here changes, the output changes at the same level. And that's why we need this LDO. It can be quite a low-cost LDO. It doesn't need a high-input voltage. We just need the 3.3 volt for the digital part, which means for the isolator and for the digital part of the A-to-D converter.
And then the 3.3 volt will be feed into the charge pump and we have the plus/minus 2.5 volt. The overall efficiency is at 50%. And of course, we do not need a big protection here because the back plane power is much more controlled than the power coming from the outside, from the field. So this would be the second option.
The third option is another isolated DC/DC power supply. And this uses our brand new ISOW7841. This is actually an 4-channel isolator with also isolated power. So if you actually get the 3.3 volt on the PSE side in, you get 3.3 volt out, the current is about 100 milliamp, which is more than enough for us. And this 3.3 volt can directly be fit to the A-to-D converter and, of course, also to a charge pump.
So at the end, customers need to decide what is the here more urgent, which is this probably. What is the most cost-effective solution now? So efficiency here is also at around 50%.
So and this was actually the presentation of designing a 4-20 milliamp pair analog input module. I hope that you learned quite a bit during this presentation. Thanks a lot for your attendance. And I wish you a good day. Thanks a lot. Bye, bye.