Post Go back to editing

LT6658 with voltage divider on output

Category: Hardware
Product Number: LT6658

Is it ok to add a voltage divider on one of the outputs of a LT6658 voltage reference? This would be a 3.3v chip and I need 2.5v for an external reference to an AD7124 and feed the multiple thermistors. Doing it this way I can get away with only 5v source to LT6658. I don't know if the voltage out would still be considered stable reference. The other output would be the 3.3v supply to the AD7124.

The other option is to try and find a 2.5v version of the chip then boost one of the outputs like the datasheet shows but then I would need minimum 5.8v to get the 3.3v needed for AD7124 supply. I do have 12v available so that is an option if I can get one.

I tried to run in LTspice to see if there is any issues but I am not familiar enough with the software. I just see 2.48v out.

  • Hi Engrjhnsn,

    The AD7124 MIGHT be able to tolerate the 3.3k || 10k impedance of this circuit - note that it has reference buffers that can be enabled / disabled. Since this is mainly an AD7124 question, I'm moving it to the  Precision ADCs forum.

    But a couple of things to note - unless you pay particular attention to the ratio drift w/ respect to temperature of your external resistors, it is likely that they will completely dominate the overall drift of the OUT3 node.  See the LT5400 datasheet and application notes for some insight into this subject.

    Also - why not use the AD7124's internal 2.5V reference?

    Also - depending on how your thermistors are configured, the measurement is often ratiometric such that reference accuracy and drift cancel out of the equation. Circuit Note 545 goes into great detail on measuring thermistors with the AD7124 specifically.


  • Mark,

    The resistors I have are .1% and 25ppm so I was hoping they would be good enough especially as you pointed out with the ratiometric measurement. The CN0545 is exactly what I am following but with the AD7124-8PMDZ board. This will be interfaced to a Raspberry Pi.

    I wasn't using the internal reference because with worse case using 1K resistor for my thermistor divider would be 2.5mA per channel x8 so 20mA. The internal can only handle 10mA.

    I will see if I can figure out the impedance you pointed out but if I can't then I will look into another reference chip or wait to get a 2.5v version. I did find the LT5400 when looking at other datasheets so I will revisit that one.

  • Interesting you bring up that lab because I am planning on using that write up for some development work once my new PI comes in. I like the ADI-KUIPER-LINUX idea with all the software built in. 

    To be clear what I was referring to with the different voltages is that if I am doing an external reference it is recommended to use 2.5v. It can go to 3.2v if using 3.3v for AVdd. It would be much easier if one supply can be used for both reference and supply.

    You are correct that with the 3.3v supply the internal refout will be 2.5v but due to current demand I can't use the internal reference, hence my problem and needing two supplies. Although with the LT3040 you mentioned I could use the internal reference and buffer through it to supply the thermistors.

    Just realized you are the author of that lab. That's awesome. Good work there. I'm currently using a PI with the CN0391 and using Python to import a compiled C library to communicate to it rather than use Pyadi.

  • Re: Kuiper Linux, keep an eye out, there should be a new image releasing shortly at . 

    If you're talking directly to the SPI port from your C library this likely won't affect you, but there may have been some updates to the AD7124 driver and / or pyadi-iio interface. Pyadi-iio is easy to update, but the driver would need to be recompiled.

    Re: using the same 3.3V supply for both ADC and sensor excitation - it should  be pretty straightforward to do a reality check on whether supply noise is affecting your readings. Try powering the ADC/sensors from a quiet, linear (non-switching) supply, observe the noise in both the individual channel readings (voltage across the reference resistor, and voltage across the sense resistor) and see how they compare to the datasheet typicals. And note that for noise measurements, you should replace the thermistor with a fixed resistor with some nominal value, so that ambient temperature fluctuations don't show up in your noise measurement.

    Next, run the readings through the ADC reading to temperature calculation, and see what the noise is in "degrees C RMS" (Or degrees F, if you prefer :) )

    Then - run the same measurements with your actual 3.3V supply (like the 3.3V rail from the Raspberry Pi?)

    Good luck!


  • Thank you for the heads up on the updates coming to Kuiper. I'll keep an eye out for it.

    I figured out where I was going wrong with the external reference voltage not being able to do 3.3v. Following the cn0545 lab and ad7124_temperature_measurement_demo they enable all the buffers for REFIN and AIN. This is fine for both internal and external at 2.5v but not above 3.2v if using 3.3v for supply. I kept getting this error when using the online EVAL tool. Once I disabled buffers everything is good.

    Do you see any issue with not buffering these inputs?

    Once up and running I'll follow your tutorial for noise and do some test.


  • Re: buffered reference -  , any thoughts?

    But looking at the datasheet - reference current in unbuffered mode is only 12 microamps, so the only reason you'd ever need to enable it is if you're deriving the reference from a high-impedance source.

    Side note - occasionally you'll see circuits that derive a reference voltage from a current sense resistor. While this works in theory and saves you a bit of math, in practice it's almost always better to driver the ADC's reference pins with a quiet, low-noise source.


  • Hi All,

    Apologies, I missed this thread and thanks for tagging me Mark. 

    In terms of enabling the reference buffers, these buffers will require a 100mV headroom above AVSS and below AVDD supplies. So if you have AVDD=3.3V and AVSS=GND=0 then each REFIN pins can only accept 0.1V to 3.2V thus the tool flag this as an error since a 0V at REFIN- is outside range. But In general the AD7124 can accept reference voltage up to AVDD when buffers are disabled. 

    I'm trying to read the conversations above, and going back to the original question. I think having additional divider would affect your thermistor resistive divider, if in case you are trying the same config as CN0545. So if you want accurate measurements then these resistive dividers must also use high precision/accuracy and low tempco. And since you are driving the reference inputs with high resistor values then the reference buffers must be enabled as these may cause additional gain error. 

    Mark suggesting to use the internal reference and you are worried about the limitations of the output current. If you have the same config as CN0545 then you can just probably use a higher value of the sense resistor to limit the current.



  • Jellenie,

    Thanks for joining the conversation.

    As you pointed out that is where I wasn't fully understanding Mark with being able to use 3.3v for reference. My settings were wrong using the labs mentioned and enabling the buffers.

    I realize the divider on the output of the voltage source is a bad choice. I can't change the divider resistance based on the criteria I need so it is about 1K worst case if the thermistor is maxed out and close to 0 ohms. With that said worst case would be all 8 thermistors at 1K so if using 2.5v that would be 20mA and if using 3.3v 26.4mA and the reason I was looking at external reference.

    Yes I have high precision low tempco resistors for this and I think the input would be low impedance since there is only the 1K from divider plus the 1K inline on the board already. Unless I am missing something. 

    I guess with all this in mind would it be better to use 2.5v with or without buffers or 3.3v without buffers for the external reference?


  • Hi, 

    In general, since thermistor is a resistive sensor, we always tend to keep current through the sensor not too large to ensure that the power dissipation or self heating effect of the sensor will not affect the measurement results. So I think you need to balance the design whether you need to use an excitation voltage or an excitation current. For a 1k thermistor resistor it might be good also to consider another option of using an excitation current rather than an excitation voltage so that only small amount of current will flow through the sensor. The AD7124 have a built-in excitation current that you could use on your application. For a higher resistance values of thermistor (i.e. 10k or above) a voltage excitation may suit here and you can just use the internal reference as excitation voltage. I have here some article that you may find helpful when designing a resistive temperature sensing measurement. 

    How to Select and Design the Best RTD Temperature Sensing System | Analog Devices

    Thermistor-Based Temperature Sensing System—Part 1: Design Challenges and Circuit Configuration | Analog Devices



  • I saw those articles last week and look forward to the other parts of the thermistor one.

    I can look into the excitation current and see what kind of voltage it will create but its possible when doing test for low end accuracy at -17c it will be to high at 4v with 50uA.

    The thermistor itself is 10k but I am using a 1k Rsense to shift the peak sensitivity to the range I need. My working range is 70c to 150c. Most thermistors have better 10K tolerance then there 100k counterparts if they even have one or I would go that route. 

    The Rsense is rated 250mW and I think at peak I would see 6mW. The thermistor says it has a 100mW dissipation factor and it would see a peak of 1.4mW. Another test I need to add is how much they self heat.

    I have also used the thermistor configurator and error budget calculator and wish it had the option to change Rsense size.

    I need to go back through my spreadsheet to verify all of this cause now I am second guessing my math.

  • Hi 

    Thanks for your comments. With regards to the Rsense, the reason I designed it that way is to ensure that the output voltage of the Rsense and Rth at the base temp of 25C is always at the midpoint at nominal temp. But if you need to change it to meet your requirements then I see no harm. I just did not allow it in the tool as this would complicate things in terms of calculations for limitations. But we'll see for future revision, I'll take note of this as consideration. 

    With regards to your thermistor resistance, I thought you are using 1k thermistor that is why I recommended an excitation current. If it is 10k then a voltage divider is more appropriate. So I guess going back to your original question, it could be safer if you use 2.5V as if you use 3.3V you need to ensure that this 3.3V is very accurate and could not be higher than your supply voltage. I know voltage reference has also limited supply current so it is very unlikely to use it as both supply and voltage reference. 



  • Most of my post I have mentioned 1k, meaning that would be the minimum resistance in worse case if the thermistor is maxed out or shorted.

    If going back to the 2.5v I guess I can do 2v and it looses a small amount of p-p resolution. Right now I can use the LT6658 or LT1461 to handle the current demand. 

    What about buffering the inputs and refin1?

  • Most of my post I have mentioned 1k, meaning that would be the minimum resistance in worse case if the thermistor is maxed out or shorted.

    If going back to the 2.5v I guess I can do 2v and it looses a small amount of p-p resolution. Right now I can use the LT6658 or LT1461 to handle the current demand. 

    What about buffering the inputs and refin1?

  • Hi, 

    I think buffers must be required as the analog input is driven by a large resistance value. For the reference, maybe it is not required as the resistors in parallel could be small. But no harm if you want to enable the buffer at the positive input, and just disable it for the negative input so it can accept the 0V from your absolute reference IC.