Thermal Noise Floor: Where -174 dBm/Hz Comes From

At 290 K, kT = 4.004e-21 W/Hz, which is -173.98 dBm/Hz and is usually rounded to -174 dBm/Hz. A receiver's input-referred floor is -174 dBm/Hz + 10 log10(B) + NF, so a 1 MHz bandwidth and a 3 dB NF give -111 dBm.

Engineer’s note: At room temperature (290 K), thermal noise of -174 dBm/Hz is a hard physical floor. If the noise floor in your receiver calculation comes out below kTB, it's either a math error or a Nobel Prize, and I know which one I'd bet on. Add 10 log of your bandwidth to -174 and keep your own system's real noise floor in your head.

For practitioners

Every link budget, sensitivity calculation and spectrum analyzer floor goes back to one number: -174 dBm/Hz. It is not an empirical constant or a rule of thumb. It is Boltzmann's constant multiplied by a reference temperature and converted to decibels. Once you see the derivation, you can rebuild the number for any bandwidth and noise figure, and you can tell when someone has applied it wrongly.

Available noise power: N = kTB

The electrons in any resistor above absolute zero move at random. That motion produces a fluctuating open-circuit voltage across the resistor's terminals. Connect the resistor to a conjugate-matched load and the largest power it can deliver, the available noise power, is:

N=kTBN = kTB

Here k is Boltzmann's constant (1.380649e-23 J/K), T is the resistor's physical temperature in kelvin and B is the bandwidth in hertz. Two features of this result matter in practice.

First, the resistance value cancels out. A 50 Ω resistor and a 75 Ω resistor at the same temperature each deliver the same available noise power into their own matched loads. The open-circuit noise voltage depends on R, but the available power does not.

Second, the power per hertz is flat. Doubling the bandwidth doubles the noise power. This makes thermal noise white, with a power spectral density of kT watts per hertz.

From kT to -174 dBm/Hz

At T = 290 K:

kT=1.380649×10−23×290=4.004×10−21 W/HzkT = 1.380649 \times 10^{-23} \times 290 = 4.004 \times 10^{-21}\ \text{W/Hz}

Convert to decibels relative to 1 W:

10log⁡10(4.004×10−21)=−203.98 dBW/Hz10\log_{10}\left(4.004 \times 10^{-21}\right) = -203.98\ \text{dBW/Hz}

1 W is 1000 mW, so converting from dBW to dBm adds 30 dB:

PdBm=PdBW+30P_{\text{dBm}} = P_{\text{dBW}} + 30

That gives -173.98 dBm/Hz. Everyone rounds it to -174 dBm/Hz, and the 0.02 dB difference is far below any measurement uncertainty you will meet on a bench.

Boltzmann constantk = 1.380649e-23 J/K×TemperatureT = 290 KNoise densitykT = 4.004e-21 W/Hz10 log10Relative to 1 W-203.98 dBW/Hz+30 dB(W to mW)Relative to 1 mW-173.98 dBm/Hz≈ -174 dBm/HzN = kTBRayleigh-Jeans limitValid for hf ≪ kTkT/h ≈ 6 THz at 290 K
Step-by-step derivation of the -174 dBm/Hz thermal noise density from kT at 290 K. Concept drawing.

Why 290 K, and where the formula holds

The value 290 K is a convention. It is close to room temperature and gives a round kT. It is also the reference temperature at which noise figure is defined. Noise factor F is the ratio of input SNR to output SNR when the source is at 290 K, and NF = 10 log10(F). The equivalent noise temperature of a stage follows from the same reference: Te = 290 × (F − 1). This shared reference is why -174 dBm/Hz and NF can be added in decibels without a correction term.

N = kTB is the Rayleigh-Jeans, or low-frequency, limit of Planck's radiation law. It is valid when hf ≪ kT. At 290 K, kT/h is about 6 THz, so the flat kT density holds well into the millimeter-wave range. For handset bands, Wi-Fi and most microwave links, quantum corrections can be ignored.

If the source is not at 290 K, the density scales linearly with temperature. In decibels the shift is 10 log10(T / 290) relative to -174 dBm/Hz.

Adding bandwidth and noise figure

A real receiver does not look at 1 Hz. It integrates noise over its noise-equivalent bandwidth B: the width of an ideal rectangular filter that passes the same noise power as the actual filter. Because the density is flat, the total noise in decibels is the density plus 10 log10(B). The receiver also adds its own noise, and NF expresses that contribution referred to the input. The input-referred floor is:

Nfloor=−174 dBm/Hz+10log⁡10(B)+NFN_{\text{floor}} = -174\ \text{dBm/Hz} + 10\log_{10}(B) + \text{NF}

This floor is referred to the input. It is the noise level a signal at the antenna port has to compete with. It is not the noise power you would measure at the output, which also includes the receiver's gain.

Worked example: 1 MHz bandwidth, 3 dB noise figure

Take a receiver with a 1 MHz noise-equivalent bandwidth and a 3 dB noise figure, fed from a 290 K source.

  1. Thermal density: -174 dBm/Hz.
  2. Bandwidth term: 10 log10(1,000,000) = 60 dB. The thermal noise in 1 MHz is -174 + 60 = -114 dBm. In linear terms, kTB = 4.004e-21 × 1e6 = 4.004e-15 W.
  3. Noise figure: -114 + 3 = -111 dBm.

The input-referred noise floor is -111 dBm. A signal at -111 dBm at the input gives an SNR of 0 dB. To reach sensitivity, add the SNR your demodulator needs on top of this floor.

kT at 290 K-174 dBm/Hz+10 log10(B)B = 1 MHz: +60 dBkTB in 1 MHz-114 dBm+NF = 3 dBreceiver noise floor-111 dBm (input-referred)Level rises left to right; vertical steps not to scale
Noise floor staircase for a 1 MHz, 3 dB NF receiver, following floor = -174 dBm/Hz + 10 log10(B) + NF. Concept drawing.
Every decibel of loss in front of the LNA adds almost one-for-one to the system noise figure, so it lifts this floor. A filter or switch ahead of the first gain stage counts in full. See Why Front-End Loss Before the LNA Costs Sensitivity.

Common misuses

  • Using the 3 dB bandwidth as B. The formula needs noise-equivalent bandwidth. For a steep channel filter the two are close. For gentle roll-offs they are not.
  • Adding NF to an output-referred level. -174 + 10 log10(B) + NF is an input-referred number. Gain appears only if you want the level at the output.
  • Applying NF with a non-290 K source. An antenna pointed at cold sky or a hot environment does not present 290 K. NF alone then does not predict the floor, and you need noise temperatures.
  • Treating instrument floors as thermal floors. A spectrum analyzer or VNA receiver has its own noise figure and resolution bandwidth. Its displayed floor follows the same formula with its own B and NF. The VNA dynamic range page covers what this means for stop-band measurements.

In plain words · for students and beginners

The thermal noise floor in plain words

An everyday picture

Think of a room full of people talking quietly. No one is saying anything useful, but the room is never silent. That background hum is like thermal noise. Electrons inside every warm object jiggle at random, and the jiggling makes a tiny, random electrical signal.

The picture breaks down in one place. Put more people in a room and it gets louder. Thermal noise power does not depend on how big the resistor is (a resistor is a part that resists current). It depends only on temperature and on how wide a slice of frequencies you listen to.

RT = 290 Knoise voltagesourceRwarm resistormatched loadavailable noise power N = kTBIn 1 Hz of bandwidth at 290 K:kT = 4.004e-21 W = -174 dBm
A resistor at 290 K delivers available noise power N = kTB into a matched load, which is 4.004e-21 W or -174 dBm in each hertz of bandwidth. Concept drawing.

The drawing shows a warm resistor connected to a matched load, meaning a load chosen so that the most power transfers. The power that arrives is N = kTB:

  • k is Boltzmann's constant, a fixed number from physics: 1.380649e-23.
  • T is temperature in kelvin. The standard value is 290 K, about room temperature.
  • B is bandwidth: how many hertz of frequency the receiver listens to.

The same example, step by step

  1. Multiply k by 290. You get 4.004e-21 watts in each hertz of bandwidth. That is an extremely small power.
  2. Convert to decibels so the number is easier to work with. In dBm (decibels relative to one milliwatt) it comes to about -174 dBm per hertz.
  3. Our receiver listens to 1 MHz, which is one million hertz. A million is 60 dB (10 × log10 of one million). Add it: -174 + 60 = -114 dBm.
  4. The receiver adds a little noise of its own. Noise figure (NF) tells you how much, in dB. Ours is 3 dB, so -114 + 3 = -111 dBm.

Any signal weaker than about -111 dBm at the receiver input is buried in the noise.

Mistakes beginners make

  • Forgetting the bandwidth. -174 dBm/Hz is per hertz. It is not the noise in your whole receiver. You must always add 10 log10(B).
  • Mixing up dBW and dBm. The same noise is -204 dBW/Hz (rounded) and -174 dBm/Hz. The two differ by 30 dB.
  • Thinking a bigger resistor makes more noise power. It makes a bigger noise voltage, but the power it can deliver into a matched load stays the same.
  • Forgetting the temperature. The -174 figure assumes 290 K. A colder or hotter source shifts it.

Remember

  • -174 dBm/Hz is just kT at 290 K written in dBm.
  • Floor = -174 dBm/Hz + 10 log10(B) + NF, referred to the receiver input.
  • 1 MHz adds 60 dB, so a 3 dB NF receiver in 1 MHz has a floor of -111 dBm.

Frequently asked questions

Why is the thermal noise floor -174 dBm/Hz?

At the 290 K reference temperature, kT = 1.380649e-23 J/K × 290 K = 4.004e-21 W/Hz. In decibels that is -203.98 dBW/Hz. Adding 30 dB to convert to milliwatts gives -173.98 dBm/Hz, which rounds to -174 dBm/Hz.

How do I calculate the noise floor for a given bandwidth?

Add 10 log10(B) to -174 dBm/Hz, with B in hertz. For 1 MHz that adds 60 dB, so the floor is -114 dBm before you include the receiver's noise figure.

How does noise figure change the receiver noise floor?

Noise figure in dB adds directly to the thermal floor when that floor is referred to the receiver input. A 3 dB NF in a 1 MHz noise bandwidth moves the floor from -114 dBm to -111 dBm.

Does N = kTB hold at millimeter-wave frequencies?

Yes, for typical work. N = kTB is the Rayleigh-Jeans limit of Planck's law and holds while hf ≪ kT. At 290 K, kT/h is about 6 THz, so the approximation is good well into the millimeter-wave range.

References and further reading