Linear regulator characteristics over time, temperature (LM3x7 vs. 78xx/79xx)

[jolshefsky]

I have a tool that makes measurements of a linear variable differential transformer (LVDT). (In short its primary is driven at a fixed frequency and amplitude, and the secondary's output is proportional to the position of the movable core.) The signal is generated relative to the regulated power supply (e.g. if the supply voltage increases, the excitation amplitude increases) and the measurement is done as an absolute read of the secondary voltage. Thus, the circuit is sensitive to changes in the regulated supply.

It currently uses an LM317 and LM337 that are trimmed to supply +5V and -5V respectively. I am interested in switching to a 7805 and 7905 instead so I can do away with the trimming.

I understand I can get better absolute accuracy with the LM3x7 regulators—I can trim them to within hundredths of a volt and they'll hold that value. The absolute accuracy isn't particularly important: it's the consistency over time and temperature.

These tools get used in a variety of human environments and are specified accurate from 5C to 40C. Current draw is about ±30mA, so heat sinking is not a factor. Two factors that are at play are drift over time, and temperature sensitivity.

As best I can tell from the data sheets, both regulators are about the same with regards to temperature. Is this true?

The LM78xx and LM79xx specify an absolute accuracy of about ±5%. But will it change over time? In other words, if I'm getting 5.07V from a 7805, a year from now will it be 5.07V or some other value between 4.8V and 5.2V?

TI's LM317 data sheet specifies a typical value of ±0.3%/1K hours, so after a year, that could be as much as 3% or so, and after 10 years, 30% which seems downright absurd.

My guess is the regulators are pretty stable over time: more so than the data sheets imply.

(And yeah, the RIGHT answer is to either use a precision reference to drive the primary of the LVDT or make a ratiometric calculation.)

[SeanB]

The 7805 will drift with both time, temperature and which company made it. Good units from a reputable manufacturer ( probably will be hard to find) will drift less. Have you considered keeping the 317/337 and just using a set of precision resistors along with a small trim network to set the voltage. Simple to do by having a fixed pair to set the voltage slightly low with a small area on the bottom leg with a few parallel low value resistors  which you can remove to raise the output voltage on test, or a series chain of low value resistors that are then bridged out with zero ohm jumpers on test to do a final tolerance trim.

[David Hess]

Some regulators are better than others in regards to absolute accuracy and drift.  Linear Technology makes higher performance alternatives to the LM317 and LM7805 series for instance.

http://www.linear.com/product/LT1086

The solution as you have identified is to use a precision voltage regulator circuit.  The common and superior alternative though is to make a ratiometric measurement instead of an absolute measurement.  Then the excitation voltage accuracy becomes irrelevant.

[David Hess]

TI's LM317 data sheet specifies a typical value of ±0.3%/1K hours, so after a year, that could be as much as 3% or so, and after 10 years, 30% which seems downright absurd.

My guess is the regulators are pretty stable over time: more so than the data sheets imply.

The long term stability specification only applies over the time period listed.  The drift do to aging will decrease over time and after 1000 hours, most of the drift has occurred.

[jolshefsky]

Thanks to everyone for their input. I designed the board to accept different kinds of regulators as a way to test them. Unfortunately, I had the pins swapped in my LM337 model (doubly unfortunate: pins 2 & 3, not pins 1 & 3 for which I could just install it backward) so I'm forced to resort to an alternative. I did some voltage measurements before I sent one out (with the 7x05's) and will get back to the forum when I get some data when it's returned for calibration some years from now.

The common and superior alternative though is to make a ratiometric measurement instead of an absolute measurement.  Then the excitation voltage accuracy becomes irrelevant.

That's on the list for the project list for the drawing board for the next design. 

[Richard Crowley]

Why is your circuit so sensitive to absolute power rail voltage?  Something doesn't sound right.

[jlmoon]

Why is your circuit so sensitive to absolute power rail voltage?  Something doesn't sound right.

Richard,
Maybe not quite enough headroom or reserve on pre reg rail?

JLM

[David Hess]

Why is your circuit so sensitive to absolute power rail voltage?  Something doesn't sound right.

Richard,
Maybe not quite enough headroom or reserve on pre reg rail?

JLM

He already explained this.  The drive signal for the LVDT uses the regulator output as a reference so the relatively poor accuracy of the regulator compared to a true reference limits the accuracy of the system.

Usually either the drive signal is generated closed loop and compared to a reference in which case the accuracy of the drive signal directly affects the accuracy of the output or the drive signal is sampled to make a ratiometric measurement.

[Richard Crowley]

He already explained this.  The drive signal for the LVDT uses the regulator output as a reference so the relatively poor accuracy of the regulator compared to a true reference limits the accuracy of the system.

Usually either the drive signal is generated closed loop and compared to a reference in which case the accuracy of the drive signal directly affects the accuracy of the output or the drive signal is sampled to make a ratiometric measurement.

Wow. That is just poor system design. I wonder if the circuit can be modified as Mr. Hess suggests to eliminate this unfortunate condition.