LTC1043 with missing supplies

[mrjam]

Hi everyone,

I am designing a daughter board to occasionally calibrate a bipolar DAC.

For the task I chose the LTC2400 ADC, which will work mainly in ratiometric mode, so at the inputs I have the DAC reference voltages and the DAC analog output.

Via some relays, the input signals are connected alternatively to the S1 and S3 pins of the two halves of the LTC1043, so as to invert the negative voltages for the (positive) VREF and VIN inputs of the ADC, or pass the positive voltages directly.

The card is powered by two external batteries, using a cheap jack cable. Therefore it may happen that the unpowered card is connected to the working DAC.

In your opinion, is there a risk of damaging the LTC1043 inputs? Have you ever ruined an LTC1043?

Could 1k series resistors be enough to limit the current through the protection diodes to a safe range, without degrading the LTC1043's DC performance?

PS. The relays are latching type, so I can't rely on them to disconnect the inputs. In any case, I'd avoid adding an extra layer of relays.

Best regards
Roberto

[Andreas]

In your opinion, is there a risk of damaging the LTC1043 inputs?

Have you ever ruined an LTC1043?

Could 1k series resistors be enough to limit the current through the protection diodes to a safe range, without degrading the LTC1043's DC performance?

a)  Datasheet says: input voltages have to be within Supply Voltages (with 0.3V over voltage)
b) no not up to now. (but my voltage references have current limiting and the cirquit behind LTC1043 uses less than 2 mA (static) in case of back powering)
c) specifiy degrade by how many %?

I am using external protection to protect my voltage references from being degraded. (Input voltage is clamped to VDD-3V)

with best regards

Andreas

[mrjam]

c) specifiy degrade by how many %?

Thanks Andreas for your kind reply.

The DAC is 24 bits resolution. I would like to get 20 bit linearity, but I know it is nearly impossible using ICs. Maybe 19 bits can be achieved.

Best regards
Roberto

[MiDi]

LTC1043 is sensitive to latch-up:

Input Pins, SCR Sensitivity
An internal 60Ω resistor is connected in series with the
input of the switches (Pins 5, 6, 7, 8, 13, 14, 15, 18) and
it is included in the R ON specification. When the input
voltage exceeds the power supply by a diode drop, current
will flow into the input pin(s). The LTC1043 will not latch
until the input current reaches 2mA–3mA.

[Andreas]

Made a short measurement of a LTC1043 as 2:1 divider with input connected to VRef (a bit over 5000mV)  of a LTC2400 and 0R 100R 1K and 10K

result at 0R the voltage is somewhat lower as with 100R (obviously the switching noise of the LTC1043 influences the reference voltage more than with 100R)
At 1K the voltage decreases slightly. 1K is about the resistance of the LTC1043 switches and the switching frequency is optimized to the 1uF capacitors for the LTC1043.
At 10K the voltage decreases significantly ~39 ppm so is far away from your 1 ppm target.

Note that the LTC2400 noise (0.3 ppm RMS = 9uVpp @ 5V) needs several values in average to get down to 1 ppm.
And finally you will need to calibrate out INL to get linearity of LTC2400 to ~1 ppm

with best regards

Andreas

[mrjam]

Thanks MiDi, I was aware of that part of the datasheet, so I consider it safe to keep the current at the input pins less than 2mA.

Andreas, wow! Thanks for running this experiment for me. The LTC1043 seems to work best with low source impedances, less than about 500 ohms.
The circuit you posted is very interesting, unfortunately having to measure also the output of a bipolar DAC I cannot use a single mosfet switch. A fault-protected analog switch IC is needed, such as the TMUX7411F, which is quite expensive.
Thinking of a similar implementation, I drew the simplified schematic that I attached (decoupling and Vin circuit are missing). Basically the input voltage to the unpowered LTC1043 would provide, through the protection diodes, the current needed for its operation (0.4mA max in the operating temperature range), but would not propagate through the power pins to the remaining circuits thanks to the action of the two mosfet switches.
The following unpowered buffer is protected by the 4.7k ohm resistor, which limits the input current to less than 0.8mA when the input voltage is 4V (worst case in my scenario). This resistor is already part of the low-leakage limiter circuitry needed to protect the LTC2400 from negative voltages that may occur during input switch operations.
So the total current at an input pin of the LTC1043 could be 1.3mA in the worst case.

I would be very grateful to know your opinion about it.

To reduce the influence of noise I plan to average 256 ADC readings for each measurement.

Best regards
Roberto

[Andreas]

Hello,
should work. (although the diagram is hard to read with only pin-numbers and without internal cirquitry).

I assume that VRef is ~+/-2V in this case. (you write the maximum Voltage is 4V; and at the output the difference of both voltages is shifted above ground).

Q1 and Q2: you only use the body diode of the FETs. I would use a 1N4148 or BAT48 in this case.
Q4 is not really necessary when VREF_DIR is always > 0V and VREF_INV always <0V.
ADA4522 is also my favorite (does not give temperature dependant interference with LTC1043 and LTC2400 switching frequencies)
I typically replace C1 by a short and leave R3 unpopulated. (is only optional in my cirquit diagram).

with best regards

Andreas

[Andreas]

Hi Andreas,

I have to choose the most suitable 4.096V voltage reference for a LTC2400-based ADC intended for calibrating bipolar DACs to obtain the greatest possible precision.

The measurements will be:
1 - absolute type - to measure the offset and the positive and negative full-scale outputs of the DACs, so I need a local Vref for the ADC;
2 - ratiometric type - to measure the weight of individual DAC bits, in both the positive and negative quadrants, with respect to the DACs reference voltages, which are set at plus and minus 4.1V (negative reference voltage will be inverted by an LTC1043 before ADC input).

Each measure will last approximately 1 minute for low noise. I think that Vref long-term performance is not a primary requirement for my project.

The choice to use a 4.096V reference depends on the DAC's Vref maximum ratings: +-4.7V. I can't get to 5V, so the closest standard value is 4.1V. Then comes the calibration of the LTC2400: it will be performed using the local vref as a reference, and considering that the parabolic curve of the INL error changes with the reference voltage, it is important that the local Vref has the same value of the DAC's Vref, to ensure accuracy and consistency of both types of measurement (absolute and ratiometric).

Good candidates seem to be the LTC6655 and the MAX6241/6341. What reference would you use?
I would be very grateful if you could give me some advice based on your experience.

Best regards
Roberto

Hello,

with the cirquit above you have no choice. (Maximum 6V supplies.)
In this case you have to use a low drop voltage reference like LTC6655.

I would usually prefer a buried zener device (like the MAX6xxx) but these need minimum (8-)10V supply voltage.

For the LTC6655:
- the LS8 package has lower thermal hysteresis than the MSOP package.
  (this is no issue when used in a thermal controlled environment).
- LTC6655 needs a lot of supply current and gets warm.
  so use a thermal shield against air drafts.

By the way:
with +/- 4.1V the LTC1043 in cirquit above will output 8.2V at the output pin 5 (clamped by ESD diodes to supply voltage).

with best regards

Andreas