Columnists

Filter your voltage reference for low-noise performance

By Ron Mancini -- EDN, 11/25/2004

The short-term variation in output voltage from a voltage reference is noise. Reference-voltage noise occurs in two frequency ranges: 0.1 to 10 Hz for short-term, peak-to-peak drift, and 10 Hz to 1 kHz for wideband noise. Expressing noise in parts per million is popular because the noise voltage is usually proportional to the reference voltage, thereby keeping the parts-per-million value relatively constant. Bandgap-voltage references have noise voltages ranging from 3 to 16 ppm, but buried-zener voltage references are quieter, with noise voltages ranging from 0.1 to 0.5 ppm. Noise decreases with increased reference current, but increasing the reference current is not an option for most references. Thus, the path to improved noise performance is an external noise filter. Filters are effective noise reducers: A reduction in the noise bandwidth of 100-to-1 results in a noise reduction of 10-to-1 (Reference 1).

The circuit in Figure 1a shows a typical voltage-reference filter in which the load current flows through R₁, causing a voltage drop across R₁. R₁C₁ provides the filtering function, but the loss of regulation that the voltage drop across R₁ causes demands the addition of a buffer (Figure 1b). The buffer reduces the current through R₁ to 1 nA maximum. When you can accomplish the noise filtering with R₁ of 50Ω and a ceramic capacitor, the voltage drop across R₁ is 0.05 µV, and you can ignore it. Use the circuit in Figure 1 when C₁ is a ceramic capacitor and if the low-frequency breakpoint is adequate for the desired noise performance. The buffer contributes noise to the reference voltage, so configure it as a lowpass filter with a breakpoint that eliminates its internally generated noise.

When the filter's –3-dB breakpoint has to be low, this circuit configuration has some problems. Increasing R₁ has limitations because its voltage drop limits the reference-voltage accuracy. You can't use a ceramic capacitor because of its poor volumetric efficiency, so select a tantalum or an aluminum-electrolytic capacitor. These capacitor types have appreciable current leakage that is a function of their operating voltage and temperature. Thus, the capacitor-leakage current, I_CL, flows through R₁, causing the dreaded voltage drop that ruins the regulation.

The circuit in Figure 2 solves the resistor problem. Enclosing R₁ in the feedback loop minimizes the effect of the voltage drop across R₁ by dividing it by the op-amp gain of approximately 134 dB. Thus, the effect of the capacitor-leakage currents is negligible. The voltage drop across R₁ subtracts from the output-voltage swing, but if the load current is less than the op-amp output-current capability of 10 mA, the voltage drop is less than 0.5V. The voltage drop across R₂ that the op-amp bias current causes is critical because it adds to the voltage-reference error. When R₂ is 2 kΩ, its voltage drop is 2 µV (1-nA input-bias current); if this voltage drop is too large, reduce the value of R₂.

The filter is at the op-amp output, so a lowpass filter with a –3-dB breakpoint of 1/2πR₁C₁ reduces both the voltage-reference and the op-amp noise. The R₂C₂ network ensures stability by adding a zero into the bode plot. The R₂C₂ network can cause noise-gain peaking, so you should maintain the R₂C₂=2R₁C₁ relationship to minimize amplifier-noise-gain peaking. C_1A can be an aluminum-electrolytic capacitor that has a low self-resonant frequency; hence, C_1B, a ceramic-dielectric capacitor, parallels C_1A to keep the total reactance low. This filter can drive high-capacitance loads (consider them part of C₁) if you retain the peaking relationship.

Author Information

Ron Mancini is staff scientist at Texas Instruments. You can reach him at 1-352-569-9401, rmancini@ti.com.