Thermal noise floor

Why can no amplifier ever reveal a signal quieter than its own warmth?

Take a resistor, connect nothing to it, and put a sensitive voltmeter across its terminals. You will measure a voltage. Not a steady one — a restless hiss, with no source, no battery, no radio station behind it. Cool the resistor and the hiss gets quieter. Warm it and it gets louder.

The tempting response is that this is a defect of the instrument, or of the resistor, and that a better-made component would be silent. It is worth resisting that. If the hiss were a manufacturing flaw we would have engineered it away in the century since it was measured. That we have not — and that the amount of it can be predicted from temperature alone, without knowing anything else about the component — suggests we are looking at something closer to a law than a fault.

Ask first where a voltage could come from with nothing connected. A resistor is a lattice of atoms with charge carriers moving through it. Above absolute zero, those carriers have thermal energy: they jitter. At any instant slightly more of them happen to be at one end than the other, and that momentary imbalance is a voltage. The next instant it is a different voltage. The average over time is zero; the average of the square is not.

Now push on the consequences. If the jitter comes from thermal energy, its size should depend on temperature — and it does. This is what John Johnson measured at Bell Labs in the late 1920s and what Harry Nyquist explained theoretically almost immediately afterwards.

Here is Nyquist's result, and it is stranger than it first looks. The available noise power a resistor can deliver into a matched load is the Boltzmann constant times absolute temperature times bandwidth — and the resistance does not appear. A one-ohm resistor and a one-megohm resistor at the same temperature offer the same noise power. The voltage differs, because a larger resistance delivers the same power at a higher voltage, but the power available is set by temperature and bandwidth alone.

Why should that be? Because dissipation and fluctuation are two faces of the same thing. Anything that can absorb energy from a signal must also, at finite temperature, radiate energy back as noise — otherwise you could couple a hot resistor to a cold one and watch heat flow uphill forever. The noise is not incidental to the resistor's ability to dissipate; it is required by it. That relationship, later generalised as the fluctuation-dissipation theorem, is why this cannot be engineered away. A component that absorbed without fluctuating would be a perpetual motion machine.

Now bring in the amplifier. It contains resistances, junctions, channels — things that dissipate — so it must fluctuate too, and it adds that fluctuation to whatever arrives at its input before amplifying the sum. Both parts get multiplied by the gain, so gain never improves the ratio between them. Turning the volume up on a hiss produces a louder hiss.

And there is a further point that decides amplifier design. Because the first stage's noise gets amplified by everything downstream while later stages' noise does not, the noise contributed by the first stage dominates the whole chain. This is the content of the Friis formula, and it is why enormous care goes into one transistor at the front of a receiver and comparatively little into the rest.

So what levers actually remain? Only three appear in the expression. Lower the temperature — which is exactly why radio telescope and deep-space receivers run their front ends in cryostats at tens of kelvin or below. Narrow the bandwidth — accept a slower signal in exchange for less noise admitted. Or integrate for longer: average many independent samples, and because the signal adds coherently while the noise partly cancels, the ratio improves as the square root of the observing time.

That last one matters for honesty about what the floor is. It limits how much you can see in a single instant of a given bandwidth. It does not forbid detecting an arbitrarily faint steady source, if you are willing to spend time. Radio astronomy lives entirely in that gap. What no amount of patience buys you is a faint signal that is also brief — there, the floor really is final. And at very high frequencies a second, quantum limit set by the energy of individual photons takes over from the thermal one.

Three refinements.

First, the misconception: this is not a limitation of the amplifier's engineering. It comes with any device that dissipates, at any temperature above absolute zero, and better fabrication cannot remove it — only colder operation and narrower bandwidth move it.

Second, "warmth" in the question is meant literally, not as a metaphor for imperfection. The floor is proportional to absolute temperature, which is why the standard reference figure quoted in receiver design is stated at a specific room-ish temperature, and why engineers describe an amplifier's added noise as an equivalent temperature rather than a voltage. It is the natural unit.

Third, bandwidth is the underrated lever. Doubling the bandwidth admits twice the noise power for the same signal. Much of practical low-noise design is not about exotic devices at all but about refusing to listen to frequencies where the wanted signal is not.

• Noise figure and noise temperature as two ways of stating the same quantity.
• Why cryogenic front ends transformed radio astronomy and satellite reception.
• Where the quantum noise limit takes over from the thermal one, and what that means at optical frequencies.

Every term the collection defines is gathered in the glossary.