SoftLoRa

Decode LoRa packets from radio recordings and live SDR streams, in pure Python.

Point it at an IQ recording and it finds the packets, corrects the frequency and timing errors introduced by the radio link, and gives you back the payload. No GNU Radio or C++ needed — just numpy and scipy.

It was built for satellite downlinks, where signals are weak and Doppler-shifted, and is tested on synthetic data, over-the-air captures and real satellite passes.

Developed for Google Summer of Code 2026 with LibreCube.

Install

pip install softlora

Python 3.9+. The decoder needs only numpy and scipy.

Receiving live off an SDR additionally needs the SoapySDR bindings, which are a system package rather than a pip one — see Receive from an SDR.

Decode a recording

from softlora import LoRaDecoder

decoder = LoRaDecoder(sf=10, bw=125_000, fs=125_000, fc=437e6)

for packet in decoder.decode_file("recording.wav"):
    print(packet.payload_text, packet.crc_valid)

The four arguments describe the signal you are decoding:

.wav, .cfile, .dat and .bin files are supported. A recording may hold several packets, so you always get a list back.

Decode a live stream

Feed IQ chunks as they arrive from an SDR. Packets are returned as soon as they are complete, even when one spans two chunks:

decoder = LoRaDecoder(sf=10, bw=125_000, fs=250_000, fc=437e6)

while receiving:
    for packet in decoder.decode_stream(read_iq_from_sdr(8192)):
        print(packet.payload_text)

for packet in decoder.flush():      # decode whatever is left in the buffer
    print(packet)

Receive from an SDR

One ready-to-run receiver, in examples/, for three common radios -- set RADIO to the one you have:

python examples/sdr_live.py                   # RTL-SDR, Airspy or HackRF

It is a thin script over softlora.sdr.LiveReceiver, which owns the radio, the reader thread and the frequency correction:

from softlora.sdr import LiveReceiver, format_packet

with LiveReceiver(436.13125e6, sf=8, radio='airspy') as rx:
    for packet in rx:
        print(format_packet(packet))

436.13125 MHz  sf=8  bw=125k  fs=1000k (8x bw, decim 3)  offset=0k   Ctrl-C to stop

[16:04:38] t=  22.43s  snr=+11.7dB  cfo=     -8Hz
  len=15  cr=4/5  crc=on
  hex  736f66746c6f726120626561636f6e
  text 'softlora beacon'

It tunes off-channel to dodge the radio's DC spike, decimates to a rate the decoder likes, and folds each packet's residual frequency error back into the software shift — so the receiver calibrates itself out of its own crystal error while it runs. RECORD keeps the IQ for offline replay.

That capture is the bundled examples/arduino/lora_beacon sketch, which beacons one LoRa packet every 3 s from an STM32 + SX1262 at exactly these defaults — a transmitter to test against when you have only one radio.

softlora.sdr needs SoapySDR, which is not on PyPI:

sudo apt install soapysdr-tools python3-soapysdr
sudo apt install soapysdr-module-rtlsdr    # or -airspy, or -hackrf
python3 -m venv --system-site-packages .venv

Reading the result

Every decode path returns Packet objects. The fields you will usually want:

packet.payload_bytes   # the data
packet.payload_text    # the same data decoded as UTF-8
packet.crc_valid       # True when the payload passed its checksum
packet.snr_est         # signal-to-noise estimate in dB
packet.ok              # a packet was found and demodulated

ok and crc_valid answer different questions. ok=True, crc_valid=False means a packet arrived but was corrupted on the way. Filter on packet.crc_valid is True when you only want trustworthy payloads.

A LoRa packet

Each packet starts with a preamble of plain upchirps, then a sync word, then a 2.25-symbol start-of-frame delimiter (SFD) made of downchirps, and finally the header, payload and CRC. All four regions are visible below.

A real LoRa packet received from the Polytech Universe-3 (PU-3) satellite (SF8, 62.5 kHz bandwidth). The payload is truncated for display.

Going further

References

[1] M. Xhonneux, O. Afisiadis, D. Bol, and J. Louveaux, "A Low-Complexity LoRa Synchronization Algorithm Robust to Sampling Time Offsets," IEEE Internet of Things Journal, 2021. arXiv:1912.11344

[2] J. Tapparel, O. Afisiadis, P. Mayoraz, A. Balatsoukas-Stimming, and A. Burg, "An Open-Source LoRa Physical Layer Prototype on GNU Radio," SPAWC, 2020.

License

MIT — see LICENSE and CONTRIBUTORS.txt.