Skip to Content
Electronic test instruments on a laboratory workbench
Technical Analysis

LTC6655 and LTC6655LN Noise: Compare the Bandwidth

By SupplyICs Editorial
Table of Contents

Two reference-noise numbers cannot be compared until their units and bandwidths match. A low-frequency peak-to-peak result, a spectral-density value and an integrated RMS result describe different measurements.

LTC6655LN adds a noise-reduction connection that can change the wideband-noise tradeoff. The purchasing question is whether that filtering meets the instrument’s bandwidth, startup and leakage requirements in the intended circuit.

Which noise quantities belong in the same comparison?

Compare spectral density with spectral density at a stated frequency, or integrated noise over the same bandwidth and with the same filtering. Keep low-frequency peak-to-peak results separate unless a justified model connects them to the required metric.

The LTC6655/LTC6655LN datasheet provides the applicable noise curves and conditions. Read the voltage option and circuit used for each curve; a value from one configuration should not become a family-wide acceptance limit.

For white noise with flat density e and a filter’s equivalent noise bandwidth B, a first estimate is Vrms = e × √B. A 50 nV/√Hz density over a 1 kHz equivalent noise bandwidth produces approximately 1.58 µV RMS. The same density over 10 kHz produces 5 µV RMS. The change follows the square root of bandwidth, not its linear ratio.

A real reference can include low-frequency noise and non-flat regions, requiring integration of its actual spectrum. The white-noise calculation is an illustration of units, not a prediction for either device.

What does the LTC6655LN NR capacitor change?

The NR capacitor forms a low-pass function ahead of the integrated output buffer, reducing wideband noise while introducing its own settling and leakage considerations. It is a different arrangement from placing a large series resistance between the reference output and an ADC.

ADI’s reference-noise article explains that architecture and gives voltage-dependent examples. For the 2.5 V option, its illustrative table associates 10 µF at NR with a cutoff near 53 Hz and 100 µF with about 5.3 Hz. Those are design inputs with component and device variation, not a guaranteed system bandwidth.

Using a single-pole approximation, 5.3 Hz corresponds to a time constant of about 30 ms. A full-scale step settling to 0.01% would take about 9.21 time constants, or roughly 276 ms. This is a simplified filter estimate; actual startup depends on the device, charging conditions and the rest of the circuit.

That tradeoff matters in a battery-powered instrument that wakes briefly to measure. A filter that is excellent during continuous operation may consume most of the wake interval before the first acceptable sample.

Which capacitor characteristics affect the NR node?

Use a low-leakage, stable dielectric and verify effective capacitance at the applied voltage. A high-leakage capacitor can create a DC error; a strongly voltage-dependent dielectric can change the filter cutoff and settling. Check the manufacturer’s recommended value and voltage rating rather than scaling the capacitance for a lower cutoff without rechecking startup.

How should leakage and converter loading be evaluated?

Evaluate NR-capacitor leakage as a possible DC-error contributor and evaluate converter loading at the output as a separate dynamic problem. A lower measured noise floor does not compensate for an unacceptable DC shift or conversion disturbance.

Compare capacitor leakage and its temperature behavior, effective capacitance, dielectric effects and startup history. Use the reference’s stated network conditions to calculate the possible error rather than treating the capacitor as ideal. Place the sensitive node according to the manufacturer’s layout guidance and control contamination around it.

At the output, check the ADC’s transient current requirement. The SAR ADC reference-settling guide addresses this REF-node load. Do not assume that a quiet NR node proves the output can support the converter’s charge pulses.

Translate the noise target into a purchasing decision

Define a measurable requirement before choosing between LTC6655 ordering options: integrated output noise over the application band, allowed startup delay, DC-error allocation and temperature conditions. Then identify the voltage, grade, package, NR components and output circuit that meet it.

Keep three comparisons distinct. The reference error budget allocates system accuracy; the LS8/MS8 package comparison addresses environmental stability; this noise comparison specifies bandwidth and filtering. Together they support an exact component choice without turning every desirable characteristic into an unsupported “best reference” claim.

A supplier comparison should preserve the approved full ordering code and the associated circuit revision. If an offered alternate changes LN status or package, return it to the relevant engineering comparison before treating its price as equivalent.

Frequently Asked Questions (FAQ)

Can peak-to-peak noise be converted to RMS with one universal factor?

No. The relationship depends on the noise distribution, observation interval and measurement bandwidth. Use the manufacturer’s stated test conditions or a justified statistical model.

Does an NR capacitor replace the output capacitor?

No. The noise-reduction pin and the output node have different functions. Follow the selected device’s datasheet requirements for both connections.

Does selecting an LN ordering option settle the MS8 versus LS8 decision?

No. Noise-reduction architecture and package stability are separate choices. Preserve voltage, grade, package and packing in the full ordering code.

Share:

Need Electronic Components?

Our team specializes in sourcing hard-to-find, EOL, and obsolete components with full traceability. Get a personalized quote within 24 hours.