# From high voltage to differential ADC

#### Jack Burton

Joined Feb 11, 2019
3
Good Morning everyone.
I'm using an ADC from Linear Tecnology, LTC2344-16 , and i have to measure an isolated voltage throug an analog isolator TLP7820 and a FDA,like in figure below:
I draw the schematic with the goal to be "multi-purpose" so the input voltage divider can be configure as "balanced", with central gnd or single ended .. anyway the analog isolator is used in its "optimal linear range" 0-200mv as described in the application note. the isolator has a fixed gain of 8,2 v/v, so its output is in the range 0-1,64v. I used then FDA LTC1992 to amplify the differential signal to match the the input range of the ADC. Since i want to use the configuration indicated in the table (0-2,5v) i choose the resistors to configure a gain of 1,51 (2,5/1,64). This is the simulation i did with the DiffAmp Calculator from analog Devices:

In my analog chain I chose to get to the entrance of the ADC with an analog range of +-2.5v but I think I had badly interpreted the datasheet of the LTC2344 I wanna use the external Vref configuration, in order to use the 0-2.5 or +-2.5V input range (negative range it should be usefull to use the circuit to measure a current using a shunt resistor, changing some parts in the input stage)

In configuration indicated by the arrow, it speaks of analogue range of +2.5V but in reality the "full range" can be +-5v (ie VDD).So could I also use Vref 2.5 as VOCM, as long as I stay in a 2.5V differential range? in other words i wanna ask : is the Vref voltage used as reference for sampling each single diff signal (Vin+ and Vin- respect to the GND) or is it used just as reference for the diff signal (Vin+ - Vin-) so i should use also the mid voltage of LTC1992 as Vocm even if it is not related to vref ?

Can this configuration guarantee the best match between the sensor signals and the ADC input ? (use all the full analog input range, with max accuracy) Am I forgetting something?
Thank you all for the support. I hope this thread can also be useful for other users with the same needs

#### Oleksandr Fatiuk

Joined Jan 3, 2019
17
Hi Jack,
Very well presented and detailed work. Well done!
I wonder what is the application: you mentioned some sensor, but not clear why isolation is required.

The overall design seems thought through and logical. I would add common mode capacitors at the input dividers (you have only differential one of 330pF - pretty small value for DC voltage measurements, but it depends on impedance of the divider) - to make sure the common mode range is not affected by noise.

Something to consider: the ADC is reasonably high resolution (16-bit), but overall tolerance may end up quite high. Input divider, then opto isolator with Gain tolerance 0.5% (but can be 3% - depending on rank) + temperature drift.
You provided options for 3 different reference voltage sources, which is good. One of them seems a bit basic (1% if I read the PN correctly). Another one you have, REF2025, is much better. ADC has integrated reference which maybe as good. I don't know the requirements, so maybe Ok, but I would look into it.

Your question about Vocm: this is essentially to create a virtual ground. The OpAmp is differential, but common mode range is only positive (single rail operation), so you need to create a virtual ground. For that reason the OpAmp provides a divider (pin Vmid). You need just to connect Vmid to Vocm and add a decoupling capacitor, no need for external voltage (datasheet has some example circuits illustrating this), unless you want it to be something different to VCC/2. Either way, you probably would need to add some decoupling close to Vocm pin.

I don't want to throw a wrench in a wheel, but I better ask now than later: to achieve opto isolation for analog signal you can go at least two different ways - one presented here, another one - to isolate digital signals and keep ADC on the "left" side of the barrier. The reason to ask is that analog opto isolation is usually not that accurate and generally more expensive (more accuracy - more cost). This particular one seems to have specific application - to measure voltage across shunt resistor in motor drivers (hence 200mV input range and Gain). I understand that each channel will be isolated from another (I assume so, may be not?), that would mean using single channel ADCs instead of 4-channel one. It would be a big change, that's why I prefer to ask it first thing, as we here don't know all the details and requirements of your project.
Here are some benefits of the alternative architecture (if it was an option here):
• Simpler analog circuitry: divider straight into ADC (or via simple buffer if required, can be single ended as shown in datasheet for LTC2344). This would eliminate a few sources of error in the chain. Having a buffer could be good, as you mentioned current measurement (gain could be added then).
• Smaller number of components and potentially cheaper outcome (digital optocouplers could be much cheaper, especially if there is no need for high speeds as sampling rate is quite low)
• ADC itself could be cheaper too. You would need 4 of them, but they potentially could be cheaper than single specialized one (this depends, may or may not be the case). If channels are isolated - maybe differential input is not needed. If so - again some simplification and cost reduction.
• Reducing the area where analog signal is still analog reduces chances to add noise to it. Converting it into digital form as soon as possible helps with that.
I didn't want to criticize the current design - it will work fine. You have plenty of provisions so there is some flexibility to play with. Just wanted to explore the full scope based on my limited knowledge about the project.

Cheers!

#### Jack Burton

Joined Feb 11, 2019
3
Hi Oleksandr,
i really appreciated your reply, and sorry for my late reply but i was away from home for the past few days..

I think it's better to start first from these two points to better explain my choices...

Hi Jack,
Very well presented and detailed work. Well done!
I wonder what is the application: you mentioned some sensor, but not clear why isolation is required.
....
I don't want to throw a wrench in a wheel, but I better ask now than later: to achieve opto isolation for analog signal you can go at least two different ways - one presented here, another one - to isolate digital signals and keep ADC on the "left" side of the barrier. The reason to ask is that analog opto isolation is usually not that accurate and generally more expensive (more accuracy - more cost). This particular one seems to have specific application - to measure voltage across shunt resistor in motor drivers (hence 200mV input range and Gain). I understand that each channel will be isolated from another (I assume so, may be not?), that would mean using single channel ADCs instead of 4-channel one. It would be a big change, that's why I prefer to ask it first thing, as we here don't know all the details and requirements of your project.
Here are some benefits of the alternative architecture (if it was an option here):
• Simpler analog circuitry: divider straight into ADC (or via simple buffer if required, can be single ended as shown in datasheet for LTC2344). This would eliminate a few sources of error in the chain. Having a buffer could be good, as you mentioned current measurement (gain could be added then).
• Smaller number of components and potentially cheaper outcome (digital optocouplers could be much cheaper, especially if there is no need for high speeds as sampling rate is quite low)
• ADC itself could be cheaper too. You would need 4 of them, but they potentially could be cheaper than single specialized one (this depends, may or may not be the case). If channels are isolated - maybe differential input is not needed. If so - again some simplification and cost reduction.
• Reducing the area where analog signal is still analog reduces chances to add noise to it. Converting it into digital form as soon as possible helps with that.
I didn't want to criticize the current design - it will work fine. You have plenty of provisions so there is some flexibility to play with. Just wanted to explore the full scope based on my limited knowledge about the project.
...
This project inherits another previous project requested by my customer, and some of their requirements were about the ADC IC, becuase they already developed in the past the VHDL driver to acquire data with the onboard Intel Cyclone V FPGA.
The board must be able to acquire 6x current and 12 x Analog channels from a very noisy environment, those signals come from Batteries and electrical generators inside a turbine. They didn't know if all the signals has same references, and so they specified the requirements of 12 isolated channels (isolated each other and from PCB logic).
I agree with you about the best way to isolate the signals, and surely it were to put the adc in the "field" portion of the pcb, and use digital isolators, but the requirements of the LTC2344-16 and all isolated channels led me to that choice
I think I have at least one advantage: I have the least number of components in the dangerous part of the pcb exposed to noise and potential damage...
In order to simplify the development of the vhdl in charge to them, i decided would have been a good idea to use the same adc to acquire both 12 isolated Voltages and 6x Currents.. (in my fist post i spoke only about the current measure..)
but...
I had a critical design review with my customer in the past week and they choose to change the strategy for the measurements of current...
We've found some chips from allegro with digital output: ACS37800KMACTR-015B5-SPI

I don't like how they provide isolation: in the datasheet they say that the VINN pin must alway be connected with GND, and the isolation will be provided by high values of resistors.....
Probably i will not use the Voltage measurement section of the ic but they don't specify if the Vinn to GND connection must be done altough i don't use that pin ..
So i want to isolate the digital bus and my final design will have 6 x this :

While the Voltage meas section will be done as though initially :

I followed TLP7820 datasheet for the calculation of divider resistors and linearity error due to input impedance of TLP7820.
Anyway the question about the adc remains the same and since, as you said, using the analog isolator I introduce errors in gain tolerance, temperature drift, etc. at least I want to be sure to use the entire voltage range of the adc, or all the available bits..
The schematics you can look above are from a little prototype board i sketched to try and test these solutions, but i'm not sure i'll have time to build it and test it... at the moment it's just an idea and i wish..
The overall design seems thought through and logical. I would add common mode capacitors at the input dividers (you have only differential one of 330pF - pretty small value for DC voltage measurements, but it depends on impedance of the divider) - to make sure the common mode range is not affected by noise.
thaks for the suggestion, i'll put capacitors for common mode filterings too..

Something to consider: the ADC is reasonably high resolution (16-bit), but overall tolerance may end up quite high. Input divider, then opto isolator with Gain tolerance 0.5% (but can be 3% - depending on rank) + temperature drift.
You provided options for 3 different reference voltage sources, which is good. One of them seems a bit basic (1% if I read the PN correctly). Another one you have, REF2025, is much better. ADC has integrated reference which maybe as good. I don't know the requirements, so maybe Ok, but I would look into it.

Your question about Vocm: this is essentially to create a virtual ground. The OpAmp is differential, but common mode range is only positive (single rail operation), so you need to create a virtual ground. For that reason the OpAmp provides a divider (pin Vmid). You need just to connect Vmid to Vocm and add a decoupling capacitor, no need for external voltage (datasheet has some example circuits illustrating this), unless you want it to be something different to VCC/2. Either way, you probably would need to add some decoupling close to Vocm pin.

Cheers!
Ok, for the Vocm i know that i have to shift it to (at least) the half of the Vdiffpp to remain in the positive swing, since i have a single supply configuration... but my doubt were about the ADC input first..
I take for example the configuration that my customer already used in previous design:

then, i suppose i have to tune the FDA in order to amplify/attenuate the input signal to match these swing (0-2.5), but does this mean that only Vdiff has to swing from 0-2.5 ? Or also Vin+ and Vin- ha to swing between 0 and 2.5 ?
Looking at the datasheet of LTC2344 it says (pag. 3) :

So, this is why i provided differents type of Vocm on the LTC1992, the one i wanted to use is Vmid as i suppose the ADC can read a Differential signal in the analog input range of 0-Vref (2,5v) inside a full scale range of 0-Vdd (5v)...
Are my considerations correct ??
Is there some issue i'm not considering ? (for example is there distortion over the full scale range of the adc ? Is there an ideal range of input to use ??)

Regards !

#### Ian0

Joined Aug 7, 2020
8,942
If you do the voltage measurement with a transformer such as ZMPT101B, then there is no need for further isolation as the hall-effect current measurement is already isolated.
(although, I’d use a current transformer)

#### Oleksandr Fatiuk

Joined Jan 3, 2019
17
Hey Jack,
Understood. Work in progress with changing requirements, always fun.
After moving to Allegro ACS37800 the current and voltage are both measured by the same chip, so no need to external ADC. Your schematic with ACS37800 using digital isolator I think is a good solution. I agree, their isolation claims are confusing: current is isolated to 5kV, but voltage is connected via resistors. So using isolator seems like the way to go.
Now something to clarify: you still will have just voltage measurement channels and they still will be done with LTC2344, right?
For LTC2344 input and utilizing the full range:
• TLP7820 has output range from 0 to 2.5V (common mode). When Vin+=Vin- (Vdiff=0V) Vout=1.25V. For +-200mV input, the outputs are: Vout+ max=2.07V min=0.43. Vout- is the same range but opposite to Vin+, so Vdiff out=1.64V
• LTC2344: input common mode range 0-Vdd V (0-5V). Now, ADC is dealing with Vdiff, wich is scaled based on Vref. For example, you can have Vref=2.5V and Vdd=5V. You can have then Vin+=5V and Vin-=4V, then Vdiff in=1V. The table for softSpan is more of a code scaling choice. Here is the Vdiff input ranges:

Your SoftSpan is #5: 0...Vref. So the scaling is: if Vin+=Vin- (Vdiff in =0V), code is 0 and each of Vin+ and Vin- can be anything from 0 to 5V. For maximum output code Vdiff=Vref (2.5V). Example of input voltages for maximum code: Vin+=4V, Vin-=1.5V.

For the choice of Vref: default value is 4.096V, which can be adjusted (overdriven) in range of 2.5 to 5V. Currently your Gain delivers 2.5V max, so if you want full scale, 2.5V Vref could be used and you can adjust all of that with your Gain, so don't stress.
One thing about full range: I would recommend to have some margin in case your voltage goes a bit higher - at least you still can measure it. When you read 65535 you can't really tell if it is clamped and voltage is actually higher or is it really correct. I would allow 20% margin, at least 10%: say allow range of 40V, even if your expected nominal maximum is 36V. You still will have resolution 0.6mV, which is probably an overkill for your needs.

I hope this helps

Cheers!

#### Jack Burton

Joined Feb 11, 2019
3
Hey Jack,
Hi Oleksandr,
The things you wrote were very usefull, and confirmed my ideas about the design.

Understood. Work in progress with changing requirements, always fun.
The (almost) final design agreed with customer will be something like :

6x Current Meas. Channels 0-15A (previously it was 25A but they said that current probably will never exceed 5A (lol?))
12x Voltage Meas Channels 0-36V
36x GPIO for dicrete signals
Eth and USB for low speed communication with a PC with a software for data storage and visualization.

After moving to Allegro ACS37800 the current and voltage are both measured by the same chip, so no need to external ADC.
I'll use Allegro for Current Meas requirements, anyway i'll place some DNP resistor to give them the possibility to use also the Vmeas part of that IC.

Now something to clarify: you still will have just voltage measurement channels and they still will be done with LTC2344, right?
For LTC2344 input and utilizing the full range:
• TLP7820 has output range from 0 to 2.5V (common mode). When Vin+=Vin- (Vdiff=0V) Vout=1.25V. For +-200mV input, the outputs are: Vout+ max=2.07V min=0.43. Vout- is the same range but opposite to Vin+, so Vdiff out=1.64V
• LTC2344: input common mode range 0-Vdd V (0-5V). Now, ADC is dealing with Vdiff, wich is scaled based on Vref. For example, you can have Vref=2.5V and Vdd=5V. You can have then Vin+=5V and Vin-=4V, then Vdiff in=1V. The table for softSpan is more of a code scaling choice. Here is the Vdiff input ranges:
View attachment 300118
Your SoftSpan is #5: 0...Vref. So the scaling is: if Vin+=Vin- (Vdiff in =0V), code is 0 and each of Vin+ and Vin- can be anything from 0 to 5V. For maximum output code Vdiff=Vref (2.5V). Example of input voltages for maximum code: Vin+=4V, Vin-=1.5V.
ok as i thought its pretty useless to use Vbias from the REF2025 chip as Vocm on the FDA used to drive the ADC, because using a 2,5V as Vref on a 0-5V ADC gives me some "range" to move within.. I could use Vref also for Vocm..
One thing about full range: I would recommend to have some margin in case your voltage goes a bit higher - at least you still can measure it. When you read 65535 you can't really tell if it is clamped and voltage is actually higher or is it really correct. I would allow 20% margin, at least 10%: say allow range of 40V, even if your expected nominal maximum is 36V. You still will have resolution 0.6mV, which is probably an overkill for your needs.
i know this is a good practice, we already kept in mind this issue during the early meetings, in fact 36V is not the nominal value, but it's about 10% higher.. nominal value is about 32V..

I'll try to keep you updated of my results.
Probably i'll do a little test board with the current and voltage sensors and i'll do some measurements to evaluate precision and thermal performance with high currents on the PCB traces..
But first i need a week of holiday (the next !)
Thank you again for the support !
Cya !