Resistor Noise Coefficient Testing: How to Measure What the Datasheet Won't Tell You

Every resistor generates noise. It is not a defect. It is physics. The thermal agitation of electrons inside the resistive material creates a voltage fluctuation that appears across the terminals even with zero current flowing. For most circuits, this noise is irrelevant. For precision amplifiers, sensor interfaces, and audio front-ends, it can be the difference between a clean signal and a useless one.

The problem is that noise is not a single number. It varies with frequency, temperature, DC bias, and the resistor's construction. Measuring it properly requires understanding what you are actually looking for and using the right method to capture it.

Why Resistor Noise Is More Complicated Than the Datasheet Suggests

The Noise You Can Calculate vs. The Noise You Actually Get

Thermal noise follows a clean formula. The open-circuit noise voltage is:

Vn = √(4kTRB)

where k is Boltzmann's constant, T is temperature in Kelvin, R is resistance, and B is bandwidth. This is Johnson-Nyquist noise, and it is white — meaning it has equal power at every frequency within the band.

But real resistors also produce excess noise that this formula does not capture. Carbon composition resistors are notoriously noisy. Thick-film resistors have 1/f noise that dominates at low frequencies. Thin-film metal resistors are quieter but still produce popcorn noise from microscopic defects in the film. The datasheet might give you a noise index in dB or a noise voltage in microvolts per volt of DC bias. That number is measured under specific conditions. Your conditions are different.

What "Noise Index" Actually Means

The noise index (NI) is defined as the noise voltage in microvolts per volt of applied DC voltage, measured in a 1-decade bandwidth (typically 10 Hz to 100 Hz or 100 Hz to 1 kHz). It is expressed in decibels:

NI (dB) = 20 × log(Vn / Vdc)

A low-noise metal film resistor might have an NI of -40 dB. A carbon composition resistor could be -10 dB or worse. That 30 dB difference is a factor of 30 in noise voltage. In a preamplifier with 60 dB gain, that difference becomes audible.

The catch is that the noise index only covers the 1/f region. It says nothing about the white noise floor above 1 kHz, where thermal noise dominates. You need both numbers to make a real design decision.

The Three Noise Types You Will Encounter

Thermal Noise: The Floor You Cannot Eliminate

This is the baseline. Every resistor at any temperature above absolute zero produces it. The only way to reduce thermal noise is to lower the resistance, lower the temperature, or narrow the bandwidth. You cannot eliminate it. You can only manage it.

For a 10-kohm resistor at room temperature (290 K) with a 10 kHz bandwidth, the thermal noise voltage is about 1.3 microvolts RMS. That is your floor. Anything above that is excess noise from the resistor's construction.

1/f Noise: The Low-Frequency Killer

Also called flicker noise, this rises as frequency drops. Below 100 Hz, 1/f noise often dominates over thermal noise. The spectral density follows:

S(f) = K / f^α

where K is a constant and α is close to 1. This means the noise power doubles every time you halve the frequency. For DC-coupled amplifiers and sensor interfaces, 1/f noise is usually the limiting factor, not thermal noise.

Metal film and thin-film resistors have low 1/f noise. Carbon and thick-film resistors have high 1/f noise. Wirewound resistors have almost none because the resistive element is a bulk metal, not a thin film with grain boundaries.

Popcorn Noise: The Random Spikes Nobody Warns You About

Popcorn noise (also called burst noise) appears as sudden steps in voltage or current. It sounds like popcorn popping through a speaker — hence the name. The cause is contamination or defects in the resistive film that create temporary micro-shorts. The noise is non-Gaussian and impossible to predict.

You will not see popcorn noise in a spectrum analyzer. It shows up as random spikes in the time domain. The only way to catch it is to look at the time-domain waveform for extended periods. A resistor that looks clean in the frequency domain can still ruin a low-frequency measurement because of popcorn bursts.

How to Actually Measure Resistor Noise

The Gain Method: The Most Practical Approach

This is the method most labs use because it does not require a noise figure analyzer. You build a low-noise amplifier with a known gain (typically 100 to 1000), apply a DC bias to the resistor under test, and measure the output noise with a spectrum analyzer or a true-RMS voltmeter.

The resistor's noise voltage is the measured output noise divided by the amplifier gain. But you also need to subtract the amplifier's own noise contribution. This is done by shorting the input and measuring the output noise with no resistor connected. The resistor noise is then:

Vn_resistor = √(Vn_total² - Vn_amp²)

This subtraction only works if the amplifier noise is at least 3 dB below the resistor noise. If the amplifier is too noisy, you cannot separate the two. Use a low-noise op-amp with input voltage noise below 1 nV/√Hz for resistors above 1 kohm. For low-value resistors, you need a current-noise-optimized amplifier because the resistor's thermal noise current (In = √(4kTB/R)) becomes the limiting factor.

The Direct Spectrum Method

If you have a spectrum analyzer with a low-noise preamplifier, you can measure the resistor noise directly without building a custom amplifier. Connect the resistor to the analyzer input through a DC block, apply the bias through a bias tee, and measure the noise floor.

The challenge is dynamic range. The analyzer's own noise floor must be well below the resistor's noise. For a 1-kohm resistor at room temperature, the noise density is about 4 nV/√Hz. Most spectrum analyzers have a noise floor around 10 to 20 nV/√Hz, which means you cannot measure a 1-kohm resistor directly. You need a resistor of at least 10 kohms to get a clean measurement, or you need to use the gain method to amplify the signal first.

The Cross-Correlation Method

This is the gold standard for ultra-low-noise measurements. You use two identical amplifiers, each connected to the same resistor. The resistor noise is correlated between the two channels. The amplifier noise is not. By cross-correlating the two outputs, the uncorrelated amplifier noise cancels out, leaving only the resistor noise.

This method can push the effective noise floor down by 20 to 30 dB. It is what national metrology labs use to characterize primary standard resistors. For most engineers, it is overkill. But if you are trying to measure the noise of a 100-ohm resistor or a thin-film part with an NI below -50 dB, this is the only method that gives you trustworthy data.

Building a Test Setup That Does Not Lie to You

Shielding Is Not Optional

Resistor noise at the nanovolt level picks up every electromagnetic source in the room. Switching power supplies, digital clocks, fluorescent lights — all of these couple into your test leads and show up as noise. Enclose the entire setup in a metal box. Use double-shielded cables. Ground the shield at one end only to avoid ground loops.

The test leads themselves are a noise source. Every centimeter of unshielded wire acts as an antenna. Keep leads under 5 cm. Use twisted pairs for the signal path. If you are measuring below 100 nV/√Hz, even the thermal EMF from dissimilar metal junctions in your connectors can swamp the signal. Use gold-plated connectors and keep all junctions at the same temperature.

Temperature Control Matters

Thermal noise scales with the square root of temperature. A 10-degree change in ambient temperature shifts the noise floor by about 1.6 percent. That sounds small, but when you are comparing two resistors with noise levels that differ by 5 percent, the temperature drift alone can flip your conclusion.

Run the test in a temperature-controlled environment. Let the resistor soak for at least 15 minutes before measuring. The resistor body and the test leads need to reach the same temperature. If you touch the resistor with your fingers, you are heating it by 2 to 3 degrees and invalidating the measurement.

Reading the Data: What Numbers Actually Tell You

Noise Density vs. Integrated Noise

Noise density (nV/√Hz) tells you the noise at a specific frequency. Integrated noise (nV RMS over a given bandwidth) tells you the total noise your circuit will see. Both matter, but for different reasons.

If you are designing an audio preamp with a 20 kHz bandwidth, you care about integrated noise from 20 Hz to 20 kHz. If you are designing a DC sensor interface, you care about noise density at 0.1 Hz because that is where 1/f noise dominates. Do not compare noise density numbers between resistors unless the bandwidth is the same. It is meaningless otherwise.

When the Datasheet Is Wrong

Datasheet noise numbers are typically measured on a sample of parts from one production lot. The actual lot you receive can be significantly noisier, especially for carbon and thick-film types. If your application is noise-critical, test the actual parts you are using. A 30-minute gain method measurement on ten random samples from the lot will tell you more than any datasheet.

For metal film resistors, the lot-to-lot variation is small. For carbon composition, it can be enormous. One sample might measure -30 dB NI while another from the same reel measures -15 dB. That is a factor of 5 in noise voltage. The datasheet shows the best case. Your board gets the average case, and sometimes the worst case.