rpitx is a general radio frequency transmitter for Raspberry Pi which doesn't require any other hardware unless filter to avoid intererence. It can handle frequencies from 5 KHz up to 1500 MHz.
Rpitx is a software made for educational on RF system. It has not been tested for compliance with regulations governing transmission of radio signals. You are responsible for using your Raspberry Pi legally.
A forum is available : https://groups.io/g/rpitx
_Created by Evariste Courjaud F5OEO. See Licence for using it.
Assuming a Raspbian Lite installation (raspios-bookworm) : https://www.raspberrypi.org/downloads/raspbian/
Be sure to have git package installed :
sudo apt-get update
sudo apt-get install gitYou can now clone the repository. A script (install.sh) is there for easy installation. You could inspect it and make steps manualy in case of any doubt. You can note that /boot/config.txt should be prompt to be modified during the installation. If it is not accepted, rpitx will be unstable.
This fork keeps the DSP/transmitter dependencies (csdr, librpitx, ft8_lib) as git submodules, so clone with --recursive:
git clone --recursive git@github.com:specture724/rpitx.git
cd rpitx
./install.shMake a reboot in order to use rpitx in a stable state. That's it !
sudo reboot
| Raspberry Model | Status |
|---|---|
| Pizero | OK |
| PizeroW | OK |
| PiA+ | OK |
| PiB | Partial |
| PiB+ | OK |
| P2B | OK |
| Pi3B | OK |
| Pi3B+ | OK |
| Pi4 | In beta mode |
| Pi5 | OK (RP1 backends, see below) |
On the Raspberry Pi 5 the output pin depends on the frequency - see the Pi 5 section below - but on every other model, plug a wire on GPIO 4, means Pin 7 of the GPIO header (header P1). This acts as the antenna. The optimal length of the wire depends the frequency you want to transmit on, but it works with a few centimeters for local testing.
The Raspberry Pi 5 uses the RP1 south bridge, which drops the BCM2835 PLL/PWM/PCM DMA paths the original rpitx relies on. This fork adds RP1 backends so the most useful modes transmit again.
Run the tools with sudo (they need /dev/pio0 and write the PCIe ASPM
policy so GPIO reads stay fast). A reboot after install is recommended.
Everything except DVB-S2. The tools do not choose a backend themselves - a factory in librpitx picks one from the requested carrier:
| Backend | Used for | Carrier range | Timing |
|---|---|---|---|
| RP1 PIO + its DMA | FM/AM/IQ sample streams | up to 25 MHz | sample-exact, DMA paced |
pll_video + CPU |
everything above that, and all OOK/FSK bursts | up to 1.6 GHz | CPU paced, measured 1.4 ppm over a 12.6 s FT8 frame |
| PIO bit-serial + NCO | DVB-S (PSK) | 25 MHz, UHF on a harmonic | DMA paced, carrier exact to 0.02 Hz |
| Tool | Status | Notes |
|---|---|---|
tune |
works | two bands, see below |
rpitx -m RF/RFA/IQ/IQFLOAT |
works | PIO under 25 MHz, pll_video above |
piofm, pio_fsk |
works | standalone PIO tools, HF only |
morse, sendook |
works | carrier keyed by the pad's OD bit |
pocsag, pift8, corel8 |
works | FSK by rewriting fbdiv_frac |
piopera |
works | OOK envelope |
pisstv, pirtty, pifsq, pichirp, foxhunt |
works | FM, runs of equal samples are coalesced |
freedv, pifmrds |
works | high sample rates, still paced correctly |
sendiq, spectrumpaint |
works | polar (frequency + keyed envelope) |
dvbrf -m dvbs |
works | QPSK by phase-accumulator synthesis on the PIO |
dvbrf -m dvbs2 |
not ported | its encoder is 32-bit ARM assembly with no C equivalent |
Two things behave differently from the BCM2835 original:
- Amplitude is one bit. The BCM path varies the pad drive strength per sample to get 8 amplitude levels. Neither RP1 path can do that, so AM is an on/off envelope and SSB is polar FM plus keying. Voice is intelligible but distorted.
- A tool killed with SIGKILL cannot clean up, so it may leave the
carrier running and
pll_videoborrowed. Ctrl-C is fine.
dvbsenco8.s (energy dispersal, RS(204,188), convolutional interleaver) was
rewritten in C and checked bit-identical against the original assembly over
2000 packets of varied payloads, so the outer coding is unchanged - the
assembly only existed to hit 15 us/packet on a 700 MHz Pi 1.
The modulator streams the carrier itself rather than a sample stream. The BCM2835 path clocks a serialiser at carrier*phases and rotates a repeating pattern to step the phase; on the RP1 the PIO clock divider has only 8 fractional bits, and that error is multiplied by the harmonic, so instead the serialiser runs at an exact integer division of the 200 MHz PIO clock and the carrier comes from a 32-bit phase accumulator. Frequency is then exact to ~0.02 Hz and phase steps stay exact for any number of phases.
It is limited by DMA throughput, not the PIO: measured 5.89 Mword/s (188 Mbit/s) with 32 KB buffers, so the serial rate is held at 100 Mbit/s and the carrier ceiling is 25 MHz. UHF is reached on an odd harmonic, as it is on the BCM2835. A 20 second run at 434 MHz / 250 kSym/s showed no underruns.
DVB-S2 is a different matter: its encoder (BCH + LDPC for every code rate) is
2700 lines of 32-bit ARM assembly with no C equivalent in the tree, so on
64-bit builds -m dvbs2 reports that it is unavailable.
Anything still using the BCM2835 DMA path now stops with a clear message instead of writing into unrelated RP1 registers.
tune picks the synthesiser from the frequency you ask for, and the
output pin changes with it:
| Frequency | Synthesised by | Output pin |
|---|---|---|
| exact divide of 200 MHz (100, 50, 25, 10 MHz ...) | GP0 divider off pll_sys |
GPIO4 (header pin 7) |
| everything else, up to 1.6 GHz | pll_video PLL -> GP2 |
GPIO6 (header pin 31) |
GP0 is a clock divider, not a PLL: asking it for a non-integer ratio
makes it dither between two divisors and the output becomes a comb of
spurs instead of a carrier (200/1.3333 for 150 MHz was a bad case). So
GP0 is only used when pll_sys divides exactly; anything else goes to
pll_video, which is a real PLL and can be followed by a large integer
divider.
The high band works the way the BCM2835 port does: a fractional-N PLL is
retuned to the carrier. RP1's pll_video is idle unless a DPI/DSI panel
is attached, so rpitx borrows it and hands it back on exit. Its 24-bit
feedback fraction gives ~1-3 Hz resolution anywhere in the band (the
BCM2835 has 20 bits). Measured with RP1's on-chip frequency counter, the
VCO is exact from 600 MHz to 2 GHz; 1.6 GHz is used as a conservative
ceiling.
When the VCO can be an exact multiple of the 50 MHz crystal and an
exact integer multiple of your carrier, the PLL runs integer-N with its
delta-sigma modulator off and the carrier is clean. Otherwise it runs
fractional-N and you will see DSM spurs either side of the carrier.
tune says which one it picked:
(fbdiv 30+0/2^24, integer-N: no DSM spurs) <- clean
(fbdiv 26+671089/2^24, fractional-N: expect ...) <- spurs
Integer-N needs carrier x R = 50 MHz x n for some integer R that
factors into the dividers. Useful clean spots: 145, 150, 200, 250, 300,
400, 425, 433.333333, 450, 500, 1500 MHz. 434.000 MHz cannot be
integer-N from a 50 MHz reference, so it will always have spurs - when
that happens rpitx prints the nearest frequency that can be done
integer-N.
-p <ppm> now works on the Pi 5 and corrects the RP1 crystal in both
bands. It is off unless you pass it: rpitx also reads an NTP-derived ppm,
but on the Pi 5 that disciplines the system clock (a BCM2712
oscillator) and says nothing about RP1's 50 MHz crystal, so it is not
applied to the synthesiser.
Calibrate against a signal of known frequency, not against an uncalibrated RTL-SDR - a dongle without a TCXO is routinely 50-100 ppm off, which looks like 30-100 kHz of "error" at UHF.
Two caveats:
- Borrowing
pll_videodisables DPI/DSI display output while a transmit tool is running. It is restored on exit - but a tool killed with SIGKILL cannot restore it, andtune -kdeliberately leaves the carrier (and the PLL) running. - Above 200 MHz you must move your wire from GPIO4 to GPIO6. rpitx prints a reminder when it switches bands.
The high band is not carrier-only: the modulated modes reach it too, by
keying or retuning pll_video from the CPU instead of streaming samples
through the PIO. Only the standalone piofm/pio_fsk tools are PIO-only
and therefore still capped at ~25 MHz.
The PIO waveform is a variable-period square wave: carrier =
PIO_CLK / (2 * (P+3)) with P >= 1, so the ceiling is ~25 MHz and the
period is quantized in integer half-cycles (fine below ~1 MHz, coarse at
HF). The sample rate must stay at or below the carrier (X >= 2).
# VFO: 10 MHz carrier (low band, wire on GPIO4)
sudo ./tune -f 10000000
# VFO: 434 MHz carrier (high band, wire on GPIO6)
sudo ./tune -f 434000000
# FM tone: 20 kHz carrier, +/-5 kHz deviation, 1 kHz tone, 2 s
sudo ./piofm -f 20000 -d 5000 -r 8000 -t 1000 -n 2
# rpitx FM from an RF-format samplerf file (frequency deviation samples)
sudo ./rpitx -i file.samplerf -m RF -f 20000 -s 8000
# rpitx AM from an RF-format samplerf file (amplitude samples)
sudo ./rpitx -i file.samplerf -m RFA -f 100000 -s 8000Wire: GPIO4 (pin 7 of the 40-pin header) with a short wire antenna; for
SDR testing add an attenuator before the receiver input. The PIO square
wave has strong odd harmonics, so a 10 MHz carrier is receivable at
50/70/90 MHz - pi5_sdr_test.sh automates this.
easytest is the easiest way to start and see some demonstration. All transmission are made on free ISM band (434MHZ). To launch it, go to rpitx folder and launch easytest.sh :
cd rpitx
./easytest.shChoose your choice with arrows and enter to start it.Don't forget, some test are made in loop, you have to press CTRL^C to exit and back to menu.
Easy way to monitor what you are doing is by using a SDR software and a SDR receiver like a rtl-sdr one and set the frequency to 434MHZ.
A simple carrier generated at 434MHZ.
A carrier which move around 434MHZ.
A picture is displayed on the waterfall on your SDR. Note that you should make some tweaks in order to obtain contrast and correct size depending on your reception and SDR software you use.
Spectrum painting of your face using the raspicam for fun !
Broadcast FM with RDS. You should receive it with your SDR. This is the modulation that you should hear on your classical FM Radio receiver, but at this time, the frequency is too high.
This is the classical Hamradio analog voice modulation. Use your SDR in USB mode.
This is a picture transmission mode using audio modulation (USB mode). You need an extra software to decode and display it (qsstv,msstv...). This demo uses the Martin1 mode of sstv.
This is a mode used by pagers. You need an extra software to decode. Set your SDR in NBFM mode.
This is state of the art opensource digital modulation. You need Freedv for demodulation.
This a beacon mode which sound like Morse. You need opera in mode 0.5 to decode.
rtlmenu allows to use rtl-sdr receiver dongle and rpitx together. This combine receiver and transmission for experimenting. To launch it, go to rpitx folder and launch rtlmenu.sh :
./rtlmenu.shYou have first to set receiver frequency and gain of rtl-sdr. Warning about gain, you should ensure that you have enough gain to receive the signal but not to strong which could saturate it and will not be usefull by rpitx.
Choose your choice with arrows and enter to start it.Don't forget, some test are made in loop, you have to press CTRL^C to exit and back to menu.
A typical application, is to replay a signal. Picture above shows a replay of a signal from a RF remote switch. So first, record few seconds of signal, CTRL^C for stop recording. Then replay it with play.
We can also live transmitting a received band frequency. Here the input frequency is a FM broadcast station which is retransmit on 434MHZ.
We assume that input frequency is tuned on FM station. It is demodulated and modulate to SSB on 434MHZ. SSB is not HiFi, so prefere to choose a talk radio, music sounds like bit weird !
rpitx is a generic RF transmitter. There is a lot of modulation to do with it and also documentation to make all that easy to contribute. This will be the next step ! Feel free to inspect scripts, change parameters (frequencies, audio input, pictures...).
All rights of the original authors reserved. I try to include all licences and authors in sourcecode. Need to write all references in this section.