The Short Answer
The noise floor is the background of random electrical noise present in every receiver, the level a wanted signal has to clear before it can be recovered. Most of it is thermal noise, the unavoidable jitter of electrons at any temperature above absolute zero, and at room temperature that noise has a power density of −174 dBm per hertz. Widen the channel and the floor rises 3 dB for every doubling of bandwidth; add the receiver’s own noise figure and it rises further. So the noise floor of a real receiver is:
Noise floor (dBm) = −174 + 10 · log₁₀(B) + NF
where B is the noise bandwidth in hertz and NF the noise figure in dB. A signal is usable when enough energy can be extracted from it relative to the noise, usually expressed as a required signal to noise ratio (SNR). The minimum signal a receiver can decode, its sensitivity, is then the noise floor plus the SNR the chosen modulation needs. Everything below expands those three ideas and ends with a worked example that produces a real sensitivity figure.
Where −174 dBm/Hz Comes From
The number every RF engineer eventually memorises is −174 dBm/Hz, the thermal noise power density at room temperature. It is not a property of any particular radio, it is set by physics and temperature alone. Thermal noise power in a bandwidth is given by the Johnson and Nyquist relation:
N = k · T · B
where k is Boltzmann’s constant (1.38 × 10⁻²³ J/K), T the absolute temperature in kelvin, and B the bandwidth in hertz. Take one hertz of bandwidth at the standard reference temperature of 290 K (about 17 °C, the IEEE reference for noise work):
N0 = k · T0 = 1.38 × 10⁻²³ × 290
= 4.0 × 10⁻²¹ W/Hz
= 4.0 × 10⁻¹⁸ mW/Hz
10 · log₁₀(4.0 × 10⁻¹⁸) ≈ −174 dBm/Hz
So in every one hertz slice of spectrum, at ordinary temperatures, there is about −174 dBm of thermal noise waiting for you, before the radio adds anything of its own. That single figure anchors every noise floor, sensitivity and link budget calculation in the band.
A real channel is far wider than one hertz, and a real receiver is noisier than an ideal one, so two terms get added to that −174 dBm/Hz baseline. The first is bandwidth. Because thermal noise is spread evenly across frequency, the total noise in a channel grows directly with how much spectrum the channel occupies, expressed in decibels as 10 · log₁₀(B). The second is the receiver’s noise figure, covered in the next section. Together:
Noise floor (dBm) = −174 + 10 · log₁₀(B) + NF
Here B is strictly the equivalent noise bandwidth, the width of an ideal brick wall filter that would pass the same noise power as the real receiver. It runs a little wider than the 3 dB occupied bandwidth, so the channel width is a fair first pass but not exact, a point the mistakes section returns to.
The bandwidth term is worth feeling in your bones, because it rises 3 dB for every doubling of bandwidth. A wide, high throughput channel therefore starts life with a higher noise floor than a narrow one over the same path, which is one reason a fast link is harder to close than a slow one. This table is the thermal floor (NF = 0 dB) for common channel widths:
| Bandwidth | 10 · log₁₀(B) | Thermal noise floor |
|---|
| 1 Hz | 0 dB | −174 dBm |
| 1 kHz | 30 dB | −144 dBm |
| 12.5 kHz (narrowband voice) | 41 dB | −133 dBm |
| 25 kHz (LMR channel) | 44 dB | −130 dBm |
| 200 kHz (GSM) | 53 dB | −121 dBm |
| 1 MHz | 60 dB | −114 dBm |
| 5 MHz | 67 dB | −107 dBm |
| 20 MHz (Wi-Fi / LTE) | 73 dB | −101 dBm |
| 100 MHz (5G NR) | 80 dB | −94 dBm |
Read off the channel and you have the floor before the receiver adds its own noise. A 25 kHz land mobile channel sits at about −130 dBm, a 20 MHz Wi-Fi channel at about −101 dBm. The wider channel carries far more data, but it does so on a noise floor nearly 30 dB higher, so it needs a correspondingly stronger signal to work.
No real receiver is silent. Its amplifiers, mixers and filters each add noise of their own, and the noise figure (NF) measures how much. It is the number of decibels by which the receiver’s output noise exceeds that of a perfect, noiseless receiver at 290 K.
A rough sense of scale: an ideal receiver has NF = 0 dB, a good low noise amplifier sits at 0.5 to 1.5 dB, and an ordinary front end at 3 to 6 dB.
Noise figure has a linear twin, the noise factor F, and an equivalent noise temperature Te, which are the forms you actually compute with:
F = 10^(NF / 10) and Te = T0 · (F − 1)
The noise temperature form is handy because temperatures add: a receiver with a 290 K noise temperature doubles the effective noise, which is F = 2, or NF = 3 dB. The conversion runs both ways, and the noise figure converter handles it.
That same noise temperature also rolls up into the G/T figure of merit, the antenna gain divided by the system noise temperature, which is how a satellite earth station packs its whole receive performance into one number.
| Front end | Typical NF |
|---|
| Cryogenically cooled LNA | 0.2 to 0.5 dB |
| Satellite / low noise LNA | 0.5 to 1.0 dB |
| Good receiver front end | 1.5 to 3 dB |
| Ordinary receiver | 3 to 6 dB |
| Lossy front end (long feed before the LNA) | 6 dB and up |
The First Stage Sets the Tone
In a chain of stages, the noise figures do not simply add. The Friis formula for cascaded noise factor shows why the front end dominates:
F_total = F₁ + (F₂ − 1) / G₁ + (F₃ − 1) / (G₁ · G₂) + …
Each later stage’s noise contribution is divided by the gain ahead of it, so a high gain, low noise first stage (a good LNA, mounted right at the antenna) hides the noise of everything behind it. The corollary is the costly mistake: put a long, lossy feed run before the LNA and that loss adds straight onto the system noise figure, decibel for decibel, because there is no gain ahead of it to mask it. The Friis transmission calculator covers the free space link side of Friis’s work, and the cascade builders in the noise floor calculator and noise figure converter walk a multi stage front end stage by stage so you can see which component is setting the floor.
SNR: How Far Above the Floor You Need to Be
Knowing the noise floor only tells you half the story. What decides whether a signal is recoverable is how much of its energy you can extract relative to the noise, and the everyday measure of that is the signal to noise ratio:
SNR (dB) = Signal power (dBm) − Noise power (dBm)
How much SNR you need depends entirely on the modulation. A rugged scheme like BPSK packs one bit per symbol and survives close to the noise; a dense scheme like 256-QAM packs eight bits per symbol but needs a far cleaner signal to tell its closely spaced constellation points apart. That trade, throughput against required SNR, is the central lever in every adaptive radio.
Representative received SNR for each modulation, for practical receivers with the coding typical of that mode. Read these as engineering ballparks, not specifications:
| Modulation | Bits per symbol | Representative SNR needed |
|---|
| BPSK | 1 | ~ 4 to 7 dB |
| QPSK | 2 | ~ 7 to 11 dB |
| 16-QAM | 4 | ~ 13 to 17 dB |
| 64-QAM | 6 | ~ 19 to 23 dB |
| 256-QAM | 8 | ~ 27 to 31 dB |
These figures are indicative. The real threshold for a given modulation depends on the forward error correction coding, the target bit error rate, and implementation loss. An uncoded link needs several decibels more SNR than the values above, while a strongly coded modern system such as LTE or DVB-S2 reaches the same modulation at a few decibels less, which is why published performance curves never quite agree with a simple table. Treat these as the order of magnitude, not a specification, and read the exact threshold off the radio’s datasheet.
Two close relatives turn up alongside SNR and cause needless confusion. CNR (carrier to noise ratio) is the same quantity measured on the carrier before demodulation, numerically the same as SNR at the receiver input. Eb/N0 (energy per bit to noise density) normalises for data rate so different modulations can be compared fairly, and relates to SNR by Eb/N0 = SNR + 10 · log₁₀(B / R), where R is the data rate in bits per second. For sizing a link, SNR against the modulation threshold is usually the figure you want.
That normalisation is also why some systems decode happily at a negative SNR. Spread spectrum and heavily coded waveforms spend bandwidth to buy processing gain, lifting the decoded signal back above the noise even when the raw channel SNR sits below 0 dB. A GPS signal arrives tens of decibels under the noise floor in its own bandwidth and is still recovered, because what has to clear the noise is the energy per bit, not the instantaneous signal power. The integrated noise floor is the reference the system works against, not an absolute wall every signal must top.
Putting It Together: Receiver Sensitivity
Now the three pieces combine into the single number a datasheet quotes and a link budget reads off: the receiver sensitivity, sometimes called the minimum detectable signal (MDS). It is the weakest signal the receiver can still decode at the target error rate, and it is nothing more than the noise floor plus the SNR the modulation demands:
Sensitivity (dBm) = −174 + 10 · log₁₀(B) + NF + SNR_required + L_impl
The implementation loss L_impl is a dB or two of honesty for a real demodulator that never quite hits the textbook threshold. This one equation is the bridge between this article and the RF link budget: the link budget predicts the received power arriving at the antenna, and the sensitivity it compares that power against is exactly the figure this formula produces. The noise floor is the receiver side of the link budget.
Worked Example
Take a microwave point to point radio running a 30 MHz channel with a 4 dB noise figure, configured for 64-QAM, which needs roughly 22 dB of SNR, with 2 dB of implementation loss. Build the sensitivity up one term at a time.
First the thermal noise in the channel:
Thermal noise = −174 + 10 · log₁₀(30,000,000)
= −174 + 74.8
= −99.2 dBm
Add the 4 dB noise figure to get the receiver noise floor:
Noise floor = −99.2 + 4 = −95.2 dBm
Then add the SNR the modulation needs and the implementation loss to reach sensitivity:
Sensitivity = −95.2 + 22 + 2 = −71.2 dBm
So this radio needs about −71 dBm to hold 64-QAM. That is precisely the kind of figure a link budget treats as a fixed datasheet value, and now you can see where it comes from and what moves it.
What moves it is instructive. Drop the radio down to QPSK, which needs only about 9 dB of SNR, and the sensitivity becomes −95.2 + 9 + 2 = −84.2 dBm, roughly 13 dB more sensitive. That is the whole logic of adaptive modulation: when a fade eats into the link margin, the radio steps down to a tougher modulation that lives closer to the noise floor, trading throughput to keep the link up rather than dropping it. The size of that fade is the fade margin from the link budget, and the noise floor is what sets the bottom of the range it has to work in.
What Raises the Noise Floor in the Field
The −174 dBm/Hz figure is the thermal floor, the best you could ever do. In service the effective floor is often higher, and it pays to know why:
- Bandwidth. The biggest lever, at 3 dB per doubling. Widening a channel for throughput raises its noise floor and costs sensitivity.
- Receiver noise figure. Set mostly by the first active stage, and degraded by any loss ahead of the LNA.
- Physical temperature. The floor scales with absolute temperature, which is why deep space and radio astronomy receivers are cryogenically cooled.
- External and man made noise. Below a few hundred megahertz especially, the noise arriving on the antenna from the environment can exceed the receiver’s own thermal noise, lifting the effective floor well above kTB.
- Interference. A real band is rarely quiet. Other emitters, intermodulation and spurious products all add to the noise the receiver sees, so the practical floor in a congested environment is set by interference, not thermal noise. This is the receiver side of why RF interference is getting worse: every decibel the noise floor rises is a decibel of sensitivity, range or throughput lost.
Common Mistakes
- Using the occupied bandwidth instead of the noise bandwidth. The noise that matters is integrated over the equivalent noise bandwidth of the receiver filter, which is a little wider than the 3 dB occupied bandwidth. Using the wrong one biases the floor by a fraction of a dB to a dB.
- Assuming an ideal receiver. Leaving the noise figure out understates the floor and overstates sensitivity. There is no NF = 0 dB radio.
- Adding cascade noise figures directly. Stage noise figures do not add in dB; use the Friis formula, where the first stage gain divides down everything behind it.
- Putting loss before the LNA. Feed loss ahead of the first amplifier adds straight onto the system noise figure. Mount the LNA at the antenna, not at the bottom of the tower.
- Confusing SNR, CNR and Eb/N0. They are related but not interchangeable; check which one a threshold is quoted against before you compare it to your number.
- Treating the thermal floor as the real floor. Below VHF, and in any congested band, external noise and interference can dominate, so the live floor is higher than −174 + 10·log₁₀(B) + NF.
Working It Out in noIM₃
The noise floor calculator is the main workstation for this: it computes the thermal floor kTB, the system noise floor with the noise figure, the noise density per hertz, the equivalent noise temperature and the G/T figure of merit, and it has a Friis cascade mode for up to twenty stages and a sensitivity mode that produces the minimum detectable signal directly. Standards presets seed the bandwidth and required SNR for common systems from TETRA and GSM through LTE, 5G NR, Wi-Fi 6 and GPS, so you can sanity check a datasheet figure in seconds.
When the question is the other way around, the SNR calculator takes a received power and returns the SNR, CNR and Eb/N0 with a margin verdict against the modulation threshold, or in reverse mode inverts the whole thing to tell you the receiver sensitivity needed to hit a target SNR. The noise figure converter handles the NF, noise temperature, Friis cascade and Y-factor measurement side, and the link budget calculator carries the sensitivity you have just derived through path loss to a received level and a fade margin, closing the loop between the transmit side and the noise floor.
Frequently Asked Questions
What is the noise floor in RF? The noise floor is the level of background noise present in a receiver, mostly thermal noise, that a wanted signal has to rise above to be detected and decoded. It is calculated as −174 dBm/Hz plus the channel bandwidth in decibels (10·log₁₀ of the bandwidth in hertz) plus the receiver noise figure. In conventional receivers, signals below the noise floor are generally not recoverable directly, although spread spectrum and coded systems can operate below the raw channel noise floor through processing gain.
What does −174 dBm/Hz mean? It is the thermal noise power density in one hertz of bandwidth at the standard reference temperature of 290 K, derived from kTB. It is the irreducible noise floor before any bandwidth or receiver noise is added, so every noise floor and sensitivity figure in the band is built up from it.
How do I calculate receiver sensitivity? Add four terms: the thermal floor of −174 dBm/Hz, the bandwidth as 10·log₁₀(B) in hertz, the receiver noise figure, and the signal to noise ratio the modulation needs (plus a small implementation loss). The result is the minimum signal power, in dBm, that the receiver can still decode at the target error rate.
What is a good SNR? It depends on the modulation. A rugged scheme like BPSK or QPSK can work with single digit SNR, while 64-QAM needs around 20 dB and 256-QAM closer to 30 dB. A higher SNR carries more data but demands a stronger signal, which is why adaptive radios drop to a tougher modulation when the SNR falls.
Does a wider channel have a higher noise floor? Yes. The noise floor rises 3 dB for every doubling of bandwidth, because thermal noise is spread evenly across frequency. A 20 MHz channel has a noise floor 13 dB higher than a 1 MHz channel over the same path (10 · log₁₀ of 20 is 13.0 dB), so it needs a correspondingly stronger signal to close.
What is the difference between noise figure and noise floor? The noise floor is an absolute power level in dBm, the total noise in a given bandwidth. The noise figure is a relative figure in dB, how much noise the receiver adds on top of an ideal one. The noise figure is one of the terms that sets the noise floor; it is not the floor itself.
Key Takeaway
The noise floor is the whole reason sensitivity exists. Start at −174 dBm/Hz, add the channel bandwidth and the receiver noise figure to get the floor, then add the SNR the modulation needs to get the sensitivity. That sensitivity is the number a link budget compares the received signal against, so the noise floor is where a link’s reach is ultimately decided. Quieten the front end, keep the channel no wider than it needs to be, and put the LNA at the antenna, and you win back the decibels that turn into range.
