Link Budget¶

TRANSMIT POWER · PATH LOSS · SENSITIVITY · MARGIN

A link works when more signal arrives at the receiving end than the chip needs. Everything in between is adding and subtracting decibels. This chapter sets up that sum, with firmware values where they exist and with explicitly marked assumptions where they do not.

Note

Source. This page has been verified against the firmware itself: MeshCore v1.16.0, commit 03b6ef4, 28 July 2026 — the root platformio.ini, src/helpers/radiolib/RadioLibWrappers.cpp and the LORA_TX_POWER flags in variants/. Every figure on this page is recomputed by tools/link-budget.py.

Bar chart of the link budget: transmit power plus antenna gain minus cable
loss minus path loss, down to the receiver sensitivity, with the remaining
margin

Warning

Two input values do not come from the firmware repo and are marked below with °: the noise figure of the receive chain and the required SNR per spreading factor. They are constants at the top of tools/link-budget.py and have not been verified against a datasheet. Every figure that follows from them carries the same uncertainty.

What the firmware fixes¶

Three values live in the root platformio.ini and apply to every board that does not override them:

Flag	Value
`LORA_FREQ`	869.618 MHz
`LORA_BW`	62.5 kHz
`LORA_SF`	8

Note

LORA_SF=8 is the compile-time default from platformio.ini, not the setting the Dutch network runs on. The radio parameters are overridden by the node configuration after flashing; the Netherlands preset sets SF7 with CR5 — see Getting Started. The sum further down this page uses the firmware default SF8 (−130.0 dBm). At SF7 the sensitivity is −127.5 dBm and the budget comes out 2.5 dB lower.

Transmit power is fixed per board. Counted across variants/*/platformio.ini, active lines only:

`LORA_TX_POWER`	Lines
22 dBm	93
20 dBm	13
19 dBm	1
10 dBm	1
9 dBm	4
8 dBm	1
7 dBm	2

That is 115 active lines across 76 variant directories. More lines than directories, because a variant file can hold several [env:…] sections that each set their own flag. The low values are not frugal boards but boards with an external power amplifier: 7 dBm at the chip becomes well over 27 dBm at the connector. See Antenna.

The noise floor¶

At the bottom of the sum sits the noise level. Thermal noise is −174 dBm per hertz at room temperature; over 62.5 kHz that is −126.0 dBm. The noise figure of the receive chain is added to that:

Item	Value
thermal noise over 62.5 kHz	−126.0 dBm
noise figure receive chain `°`	6.0 dB
receiver noise floor	−120.0 dBm

Note

That the computed floor lands exactly on the −120 dBm at which the firmware clamps its own measurement is a coincidence with this noise figure, not a proof. The noise figure is an assumption; change it to 5 or 7 dB and the equality is gone. The clamp itself is in the firmware — see The LoRa Transceiver.

Sensitivity per spreading factor¶

LoRa receives below the noise floor. How far below depends on the spreading factor:

SF	Required SNR `°`	Sensitivity
7	−7.5 dB	−127.5 dBm
8	−10.0 dB	−130.0 dBm
9	−12.5 dB	−132.5 dBm
10	−15.0 dB	−135.0 dBm
11	−17.5 dB	−137.5 dBm
12	−20.0 dB	−140.0 dBm

Each SF step buys 2.5 dB of sensitivity and doubles the airtime. The effect of that airtime on the duty cycle is discussed in Regulations & Duty Cycle.

The sum¶

Take a node at 22 dBm, a half-wave dipole of 2.15 dBi on both ends and 1 dB of cable loss:

  transmit power chip        +22.00 dBm
  antenna gain TX             +2.15 dBi
  cable loss TX                −1.00 dB
  ------------------------------------
  e.i.r.p.                   +23.15 dBm

  antenna gain RX             +2.15 dBi
  cable loss RX                −1.00 dB
  sensitivity at SF8        −130.00 dBm
  ------------------------------------
  budget                     154.30 dB

That budget may be spent on path loss. In free space the path loss is 32.44 + 20·log(f in MHz) + 20·log(d in km):

Distance	Free-space loss
100 m	71.2 dB
1 km	91.2 dB
5 km	105.2 dB
10 km	111.2 dB
50 km	125.2 dB

A budget of 154 dB would reach more than a thousand kilometres in free space. That number is correct and useless at the same time: free space does not exist at ground level. The earth curves away, buildings and trees get in the way, and the Fresnel zone touches the ground long before the budget runs out. Free-space loss is an upper bound, not a prediction. Why practice falls so far short is in Higher and stronger isn't always better.

What a dB is worth¶

More useful than an absolute distance is the ratio. Every 6 dB doubles the distance in free space, every 6 dB less halves it:

Change	Factor on distance
−6 dB	× 0.50
−3 dB	× 0.71
−1 dB	× 0.89
+1 dB	× 1.12
+3 dB	× 1.41
+6 dB	× 2.00

That is where the practical value of this whole chapter sits. A bad connector with 3 dB of loss reduces your range by almost 30 percent. One SF step buys 2.5 dB and therefore a good 30 percent — but doubles the airtime. And an antenna one metre higher often beats both, because it removes the obstacle instead of trying to transmit through it.

Sources¶

Firmware, commit 03b6ef4 (v1.16.0, 28 July 2026):

platformio.ini — LORA_FREQ, LORA_BW and LORA_SF
src/helpers/radiolib/RadioLibWrappers.cpp — the measured noise floor and the −120 dBm lower bound
variants/heltec_v3/platformio.ini — LORA_TX_POWER=22

In this repository:

tools/link-budget.py — recomputes every figure on this page

Not from the firmware repo: the noise figure of the receive chain and the required SNR per spreading factor, both marked with °. The thermal noise floor of −174 dBm/Hz is not a datasheet value but follows from kT at room temperature.