RF interference
| Value | |
|---|---|
model_family | rfi |
| Version | v1 |
| Predicts | The SNR (signal-to-noise ratio) a link loses to added noise (rfi_penalty_db), the chance of interference (p_interference) and what that chance rests on (p_interference_basis) |
| Time | One future instant per request |
| Entity | One link: a ground antenna and a satellite |
| Reads | Your network config, where the Sun and Moon sit, and optionally a noise channel in telemetry |
Overview
RF interference estimates how much noise is added to one of your links at a time you name, and what it costs in SNR. It looks at three things:
- The Sun and the Moon, from where they sit against your antenna's beam.
- Your measured noise, when your station reports it in telemetry.
- How often interference hits the link: its own recent record, and for GNSS L1 a public regional map.
The more of these the model has, the more it can tell you. A noise rise shows that noise went up, not who caused it.
Intended use
- Seeing environmental effects on a link, such as noise from the Sun and Moon.
- Watching for unexplained noise on links that report their system noise.
- Telling a noise rise from a weather fade:
unknown_noise_dbis measured noise with the weather taken out, to read beside the Weather attenuation forecast. - Recommended prediction horizon. Use it two ways. During a contact, ask for
ata few seconds ahead: with a fresh noise reading,p_interference(basismeasured) andunknown_noise_dbshow a disturbance as it happens. For contacts up to 3 days ahead, readp_interference(basislink_historyorregional) as the chance of interference from the link's record, andsun_noise_dbandmoon_noise_dbas the known noise on the pass. Further out, element sets grow stale, so where the Sun and Moon sit against the beam is less certain.
Inputs
| Parameter | Description |
|---|---|
ground_station_id | Required. A ground station in your network config. |
ground_antenna_id | Required, once. One of the station's antennas. |
satellite_id | Required. A satellite in your network config. |
at | Required. The future instant to price, in UTC. |
The model reads the link from your network config: the satellite's downlink radio, the antenna's antenna_diameter_m, and g_t_db_k on the antenna's downlink frequency range. Add sun_coupling to the antenna and half_power_beamwidth_deg to its frequency range if you know them; they sharpen the Sun and Moon terms.
To go beyond the Sun and Moon, send a noise channel in telemetry.
Outputs
| Field | Unit | Description |
|---|---|---|
rfi_penalty_db | dB | SNR lost to all added noise at at. |
p_interference | 0 to 1 | The chance of interference. What it is a chance of depends on p_interference_basis. null when the model could not assess it. See How to read p_interference. |
p_interference_basis | none | What p_interference was computed from: measured, link_history, regional or geometry. |
event_windows, assessed_windows | count | With link_history only: the 5 s windows that held a 2 dB loss, and the 5 s windows of measured noise the rate was taken over. |
sun_noise_db, moon_noise_db | dB | Each body's share of the added noise. |
unknown_noise_db | dB | Measured noise that the Sun, the Moon and the weather do not explain. null with no noise channel. |
assumptions | none | Each value the model filled in for you, with the value used. |
{
"ground_station_id": "gs-1",
"ground_antenna_id": "ant-1",
"satellite_id": "sat-1",
"target_at": "2027-03-06T17:00:00Z",
"value": {
"rfi_penalty_db": 14.6,
"p_interference": 1,
"p_interference_basis": "geometry",
"sun_noise_db": 14.6,
"moon_noise_db": 0.0,
"unknown_noise_db": null
}
}
One item from the response, with illustrative values. Fields outside value are left out.
How to read p_interference
p_interference is in beta. It is being validated against real interference, and its values can change between releases. We plan to add more interference maps and data sources, so more regions and bands get a regional estimate.
p_interference answers a different question for each p_interference_basis. Read the basis first.
p_interference_basis | Computed from | What p_interference means |
|---|---|---|
measured | A reading of your measured noise at most 5 s before at | 1 when the measured noise, or the Sun or Moon, costs the link 2 dB or more; 0 when nothing does. |
link_history | With no fresh reading: the link's measured noise over the 28 days before the nightly calibration or, when that holds no rate for the link, its last 24 h of noise channel, at least 20 windows of 5 s | The share of the link's measured 5 s windows that held a 2 dB loss, pulled toward the antenna's rate across its links when the nightly calibration has one. A background rate, not a reading. |
regional | For a GNSS link (1,559 to 1,610 MHz): a public regional map, the station's own record and its latest L1 signal | The chance of an L1 interference event in the hour of at. |
geometry | Only where the Sun and Moon sit against the beam | 1 when the Sun or Moon alone costs 2 dB or more; otherwise null, because nothing else was assessed. |
null means the model had nothing to judge from. That is not the same as a zero chance.
Explore the noise terms
Drag the terms to see how added noise sets rfi_penalty_db, and how a fresh noise reading turns null into an answer.
- Sun and Moon
- 0.04 dB
- Unexplained
- null
- p_interference_basis
- geometry
Every term is added noise in kelvin over one system noise temperature. 2 dB is reached when the added noise is 58% of the system noise.
Noise channel
A noise channel is your station's measured receiver noise, sent in telemetry at least once every 5 s.
| Telemetry field | Description |
|---|---|
system_noise_temp_k | System noise temperature, in K. |
noise_power_dbm | Noise power in a fixed bandwidth, in dBm, if you have no noise temperature. |
noise_diode_on | true while the calibration diode is on. |
elevation_deg, azimuth_deg | Where the antenna points. |
On Operator and above, the model learns each antenna's normal noise in clear sky, so it can see interference that lasts a whole contact. On Explore, it compares each reading with the rest of the same contact.
For a GNSS link, send C/N0 on L1 and L5 for each satellite instead.
Request forecasts
curl --fail-with-body -sS -G \
-H "x-api-key: $CONSTELLATION_API_KEY" \
"https://api.constellation.space/predictions" \
--data-urlencode "model_family=rfi" \
--data-urlencode "ground_station_id=gs-1" \
--data-urlencode "ground_antenna_id=ant-1" \
--data-urlencode "satellite_id=sat-1" \
--data-urlencode "at=$(date -u -v+1H +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || date -u -d '+1 hour' +%Y-%m-%dT%H:%M:%SZ)"
Each request names one link and one future instant. A request needs a key with predictions:read; a test key is refused with 400 sandbox_model_family_unsupported. From an AI agent, call get_predictions on the MCP server with the same parameters.
Example: a GNSS station
A GNSS reference station in the Baltic, with its GNSS satellites in your network config on L1 and L5. Its telemetry sends C/N0 on both bands for each satellite, as records that name each radio_id:
curl --fail-with-body -sS -G \
-H "x-api-key: $CONSTELLATION_API_KEY" \
"https://api.constellation.space/predictions" \
--data-urlencode "model_family=rfi" \
--data-urlencode "ground_station_id=riga-gnss" \
--data-urlencode "ground_antenna_id=riga-gnss-ant" \
--data-urlencode "satellite_id=gps-g05" \
--data-urlencode "at=$(date -u -v+1H +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || date -u -d '+1 hour' +%Y-%m-%dT%H:%M:%SZ)"
{
"ground_station_id": "riga-gnss",
"ground_antenna_id": "riga-gnss-ant",
"satellite_id": "gps-g05",
"target_at": "2027-03-10T15:30:00Z",
"value": {
"rfi_penalty_db": 2.4,
"p_interference": 0.22,
"p_interference_basis": "regional",
"sun_noise_db": 0.0,
"moon_noise_db": 0.0,
"unknown_noise_db": null
}
}
Read it as: about a 1 in 5 chance of an L1 interference event at this station in the 15:00 hour, with an expected loss of 2.4 dB. Ask again each hour: the chance moves with the station's latest L1 signal and the region's recent activity.
Declines
The API declines the link for the same reasons as Weather attenuation, for example not_in_network_config or below_min_elevation, and for:
| Reason | Cause |
|---|---|
band_mismatch | The antenna receives none of the satellite's downlink radios. |
optical_link | The radio is optical, where radio noise does not apply. |
prediction_unavailable | The link could not be priced this time. Retry. |
Limitations and known failure modes
p_interferenceis in beta. Use it to flag and rank links, and check it against your own outcomes.- Unmeasured interference. Without a noise channel, the model cannot see interference as it happens, so
p_interferenceisnullrather than a guess. - Receiver faults that raise system noise read as interference.
- Short events, a few seconds long, read lower than they were.
- Other satellites are not named as a source. Their interference still shows in your measured noise and the link's record, in
unknown_noise_db.
Changelog
| Version | Change |
|---|---|
| v1 | First release. Added noise from the Sun, the Moon and your measured noise, p_interference and what it rests on (p_interference_basis), and a regional chance for GNSS links. |
Related
- RF interference monitoring: where interference comes from and how to spot it today
- Weather attenuation: the weather the noise channel takes out
- Link SNR: the forecast the penalty comes off