Reduce voltage-reference output noise by half
Edited by Brad Thompson
Alfredo H Saab and Steve Logan, Maxim Integrated Products Inc, Sunnyvale, CA -- EDN, February 3, 2005
Reducing low-frequency (1/f) noise generated by an IC voltage reference can prove difficult. In theory, adding a lowpass filter to a reference's output reduces noise. In practice, a lowpass RC filter for suppression of noise frequencies below 10 Hz requires large values of series resistance and shunt capacitance. Unfortunately, a high-value series resistor introduces resistance errors and thermal noise, and a shunt capacitor's leakage resistance forms an unpredictable and unstable shunt path. Together, the two components form a noisy and temperature-dependent voltage divider that directly affects the reference's accuracy and long-term stability. In addition, pc-board surface contaminants can add yet another possible leakage path and error source.
You can stack multiple voltage references in series to reduce their 1/f noise. The references' dc outputs add linearly, and their uncorrelated internal noise sources add geometrically. For example, consider a stack of four voltage references, each comprising a dc reference source, VREF, in series with a random-noise generator, VNOISE. Adding four reference sources produces the following outputs: VREFTOTAL=4×VREF, and VNOISE-TOTAL=
=2×VNOISE. The original ratio of noise voltage to dc reference voltage thus divides in half.
Figure 1 illustrates a method of adding multiple references to produce a single, less noisy reference voltage. The resistors are parts of a highly stable metal-film network, and buffer amplifier IC5 offers low noise, low input-offset voltage, and low offset-temperature coefficients.
Table 1 and Table2 present the noise voltages that result from stacking four each of two types of 2.5V references. Each table shows the 0.1- to 10-Hz noise voltage for each of the four references, IC1 through IC4, and for the combination. Note that the dispersion in the ratios of rms to peak-to-peak values relates to subjectivity in the method of measuring the values. In addition to lower 0.1- to 10-Hz noise, the circuit also reduces long-term drift of the reference voltage.
Check out our
Best of Design Ideas
section!
-
While the concept of reducing noise (by adding multiple references) is good, the circuit shown is not the way to do it. In addition, the output voltage divider would produce 5V instead of 2.5V as stated.
One of the problems with the circuit is that the resistors will add an error due to their tolerance. The authors minimize this by calling for a high quality resistor network. Another limitation is that a power supply must be available which is over 10V, when only a 2.5V reference is being produced.
A simpler and cheaper solution can do the job better. Start with "n" grounded-referenced voltage reference chips. then simply combine all the reference outputs into a single node using a star network of resistors. The center of the star should be buffered with a voltage follower as in the original design. The resistor values are not critical since there is very little current flowing through them, though they should be nominally the same value. Since there is no voltage division, there is no error due to resistor tolerance or matching. The entire circuit can be run comforably from a single 5V supply or whatever the safe minimum is for a single reference and output buffer amplifier.
One can add a capacitor from the center of the star to ground to further reduce higher frequency noise.
George Pontis - 2005-10-2 09:52:00 PST -
If 1/F noise is a significant issue, go to a single bipolar voltage reference. There are a number of devices out there. (Search National, TI and Fairchild as a start) Although 1/F noise exists in bipolars as well, it is orders of magnitude smaller than in a CMOS device.
4 voltage references and an op-amp to reduce the noise of a single voltage reference.
Jerry Twomey - 2005-8-2 12:24:00 PST -
What am I missing? How do you put 10V across a voltage divider consisting of 2 equal resistors, and get 2.5V at their junction? That, to me, is more interesting than the low noise reference.
Harold Smith - 2005-8-2 12:17:00 PST
