Repeater software · SvxReflector compatible · free, GPL

A repeater that puts only voices on the talkgroup

TailChaser is repeater software for FM repeaters and simplex nodes on a reflector talkgroup, on a Raspberry Pi with a sound card and a radio. It listens for a voice before it passes an over on, brings every station to the same level, and sits on a talkgroup next to SvxLink nodes without either noticing the difference.

Runs onRaspberry Pi OSDebianSA818 or any radio
on0ora-simplex:8080
The status page of ON0ORA's simplex node, TailChaser-Simplex on 145.450 MHz: an S-meter at S9+20, the reflector link on talkgroup 1745, a voice going to the reflector, three overs passed as speech and one 0.4 second noise burst blocked, and two overs heard from the reflector and transmitted
The status page of ON0ORA's simplex node on 145.450 MHz, with a station on the input drawn in.

One source tree, five small programs

Pick the one that matches the site. Their configuration files look alike, and each has only what its site needs.

TailChaser-Simplex

tailchaser-simplex

A simplex node: one radio, one frequency.

  • Receiver
  • Transmitter
  • Reflector link

TailChaser-Duplex

tailchaser-duplex

A repeater in one box.

  • Receiver
  • Transmitter
  • Repeater logic
  • Reflector link

TailChaser-Receiver

tailchaser-receiver

The receive site of a split repeater.

  • Receiver
  • Serves it to the transmit site

TailChaser-Transmitter

tailchaser-transmitter

The transmit site of a split repeater.

  • Transmitter
  • Repeater logic
  • Reflector link
  • Its receiver over the network

TailChaser-SDRReceiver

tailchaser-sdrreceiver

An extra receiver, anywhere. No radio: it listens on a web SDR.

  • OpenWebRX, KiwiSDR, Web-888 or UberSDR
  • Serves it to a repeater or node, which votes

The path of an over, in tailchaser-duplex

Received

  1. Radio
  2. Sound card
  3. Gate
  4. Normaliser
  5. Opus
  6. Reflector

Transmitted

  1. Reflector
  2. Opus
  3. Queue
  4. Shaper
  5. Sound card
  6. Radio

What passes the gate also goes straight to the repeater's own transmitter. The transmit side adds the TX delay, the hang time, the roger beep, the time-out and the Morse identification. On a split site the same path is cut in two: tailchaser-receiver has the gate and the normaliser and serves its audio over RemoteTrx (TCP port 5210) to tailchaser-transmitter, which repeats it and does everything on the transmit side. tailchaser-sdrreceiver is a receive site without a radio: the same gate and normaliser, on the audio of a web SDR.

On the air at ON0ORA

A split repeater: the receiver on 431.200 MHz at one site, the transmitter on 438.800 MHz at another, on talkgroup 8. Each program serves its own status page.

on0ora-rx:8080
The status page of ON0ORA's receive site, TailChaser-Receiver on 431.200 MHz with CTCSS 131.8: connected to the repeater, squelch closed, and all 1800 squelch openings kept off the repeater, 1478 as noise and 322 as dead carrier, each of them 0.02 seconds long in the table below
The receive site, as it was on 20 November 2026: 1800 squelch openings, none of them a voice, all kept off the repeater.
on0ora-tx:8080
The status page of ON0ORA's transmit site, TailChaser-Transmitter on 438.800 MHz: the receiver reports a signal of 78 of 100 over RemoteTrx, the transmitter is on the air for a station on the input, the reflector link is on talkgroup 8, and overs from ON0DEN and ON0UHF were transmitted
The transmit site, with a station on the input drawn in: the signal is the receiver's, over RemoteTrx.
[gate] mode = on

The talkgroup only hears a site when somebody is actually talking

The radio's squelch says “there is a carrier”, not “there is a voice”. The gate listens first: audio is kept in a look-back buffer, and the over is passed on, from its first word, once a voice is heard on it. Dead carriers, kerchunks, noise bursts and 1750 Hz tones stay local.

After an over that passed, the channel is in conversation for a while and the next over passes at once.

on
only overs with a voice go to the talkgroup, from their first word
monitor
everything passes, and the log says what would have been blocked
off
a plain squelch
rx: 4.72 s passed — voice after 180 ms, stream delay 180 ms
    [speech, voiced 2140 ms, level -27 dBFS, rssi 121] -> TG 8rx: 0.42 s BLOCKED — no voice
    [noise, voiced 0 ms of 380 ms loud, level -31 dBFS, peak -24 dBFS]net: ON4ABC on TG 8, 12.3 s — transmittedtx: off after 13.6 s
journalctl -u tailchaser-simplex -f: every opening, and what the gate made of it.
99.6 %
of 32 128 squelch openings in 36 days were shorter than 2 s: noise, not people
0
false passes on eight kinds of noise, dead carrier and tone, cut into openings of 0.5, 1 and 3 s in the lab
114 / 114
three-second overs detected as speech at every noise level, from clean down to 0 dB above the noise
≈ 0.3 s
to recognise a clean voice (267 and 288 ms for the two lab voices); the look-back buffer still sends the first word
1

Level is only a floor

Below −50 dBFS a frame is silence. Loudness is never evidence of speech: squelch noise is louder than a voice, and a strong dead carrier is quieter than a soft one.

2

Pitch

Every 20 ms the last 40 ms of audio is searched for a pitch of 75 to 400 Hz, the rate of the vocal cords. Voiced speech averages 0.8 on that scale, squelch noise 0.25.

3

Not a single sine

A 1750 Hz tone or a whistle is periodic too. Spectral purity and the shape of the autocorrelation tell a sine from a voice, also when it is buried in noise.

4

Not steady

Across the voiced frames the pitch or the level has to move, as it does within a syllable. Mains hum and steady tones do not.

Minimum duration, audio level and RSSI each fail on one of the three things to keep out. Positive evidence of a voice is the only test that covers them all. The numbers come from 36 days of the ON0ORA receiver's log and from a lab corpus of recorded speech in modelled squelch noise; one case did get through there, a steady 400 Hz tone only 6 dB above heavy noise. The method, the measurements and what the lab does not show are in DETECTION.md.

What they all have in common

And what a site needs around it: the login, the identification, the squelch and the hardware.

Every site equally loud

Stations arrive at very different levels. Before received audio goes anywhere, each over is brought to one speech level. The gain follows the loud parts of speech, drops fast and rises slowly, so pauses are not pumped up.

A transmitter that cannot get stuck

The PTT line is pinned inactive before anything else is set up, a time-out limits every transmission, and the daemon exits when the sound card stops. Whatever happens, the service releases PTT and the amplifier line afterwards.

One kind of configuration file

/etc/tailchaser/<role>.conf: the same sections with the same meaning in every program, each with only the sections for the parts it has. A file with its REQUIRED lines still commented out is refused, so a box that was not set up does not transmit.

A status page in every program

On port 8080: the links, an S-meter, the last overs with what the gate made of each, and what was transmitted. On the same port, an admin portal and an API with keys.

Certificate login, by itself

On its first connection the node makes a key and sends a signing request. Once the reflector's sysop has signed it, the node logs in on its own and renews in time. Passwords still work for reflectors that use them.

The reflector is checked too

Its CA is kept from the first connection, and from then on the reflector has to show a certificate from that CA made out to its name. A node that holds a certificate refuses to fall back to the unencrypted protocol.

Talkgroups, with priorities

A home talkgroup and monitored ones in SvxLink notation. When somebody talks on a monitored talkgroup and the node is idle, it switches there. DTMF 91<talkgroup># selects one, as on an SvxLink node.

Morse on the clock

The callsign goes out in Morse every hour, every quarter, or whatever the licence asks. On a repeater it is sent over what is on the air, 6 dB quieter while somebody is talking.

A voice, from your own text

Spoken announcements with %CALL, %TIME and %DATE in them, turned into speech once a day by an online voice service (an OpenAI-compatible one, or ElevenLabs) and played in a quiet moment, or under the conversation.

A tone squelch in software

For a receiver without a CTCSS decoder, or one that gives nothing but audio. It finds the tone in about 0.3 s and holds the audio back by as much, so the first word is kept and the noise tail is not passed on.

Any radio, and the SA818 by itself

An SA818 module is programmed by the daemon over its serial port: frequency, tones, squelch, filters. type = none is for any other radio, of which only the audio and the squelch and PTT lines are used.

More than one receiver

A repeater or a node takes other receivers besides its own and votes: the first to hear a station has the over, and a clearly better one takes it. One of them can be a web SDR, with no radio at all.

Listening in

With [status] monitor = true the status page gets a Listen card under the signal meter: the audio of the receiver, squelch open or not, and of what goes to the transmitter, each with a level meter, a spectrum and a mute button. For finding out why something sounds the way it does. Off unless you turn it on.

The station, on the reflector

With its login the node sends its node information, as SvxLink does: place, sysop and position, the frequency, the CTCSS tone and which tone opens the home talkgroup. You add what it cannot know: power, antenna and height.

An amplifier, sequenced

A second output switches a power amplifier just before PTT and releases it just after. With a delay it is a sequencer: the exciter first, the amplifier some milliseconds later, and off again before the exciter.

The repeater roles

tailchaser-duplex and tailchaser-transmitter are repeaters. This is how they behave on the air.

On the air

  • What a station says on the input is transmitted again and sent to the talkgroup; what is said on the talkgroup is transmitted.
  • A station on the input wins. A talkgroup over that arrives while somebody is on the input is not transmitted, and the log and the status page say so.
  • The carrier comes up 300 ms before the audio and stays up 3 s after the last of it. Nothing is lost: the audio waits.
  • Which repeater is coming from the network is sent in fast Morse before it is heard, at 34 wpm: ON0OST becomes OST. Only when it is another repeater than the one announced last, and other stations get none: the roger beep already says “network”. The speech waits in a buffer meanwhile.
  • One level on the transmitter. Talkgroup audio is levelled like a station on the input, announcements and Morse have their own fixed levels. What is left to set per site is the deviation, once.
  • The time-out counts for as long as the transmitter is busy without a pause. After it the carrier drops until whoever was talking has stopped.

A roger beep that says where from

“blop”
after a station on the input
“bleep blop”
after an over from the talkgroup

Morse, and a voice

The callsign in Morse on the clock, and announcements from your own text. The two clocks are independent: this is Morse every quarter, and the voice on the hour right after it.

[ident]interval_min = 15[announce]enabled      = trueinterval_min = 60with_ident   = truetext         = "This is %CALL, the repeater of Opwijk. It is %TIME."

A split repeater

The receiver on one site, the transmitter on another. The transmitter connects to the receiver and reconnects by itself. The link carries uncompressed audio by default, fine on a LAN, or Opus for a slow link.

tailchaser-receiver speaks SvxLink's RemoteTrx protocol, so it also serves an SvxLink repeater with a TYPE=Net receiver.

receive site · receiver.conf
[uplink]listen_port = 5210auth_key    = a-long-shared-secret[radio]freq_rx     = 431.2000ctcss_rx    = 131.8
transmit site · transmitter.conf
[reflector]domain      = be.svx.linkcallsign    = ON0XXXdefault_tg  = 8[receiver]host        = on0xxx-rx.example.orgauth_key    = a-long-shared-secret[radio]freq_tx     = 438.8000ctcss_tx    = 131.8[ident]callsign    = ON0XXX

More receivers: voting, and a receiver on a web SDR

A repeater or a node can listen with more than one receiver. Besides its own it takes up to three or four others: a tailchaser-receiver on another hill, or a receiver without a radio.

A station, on the air
The receiver in the box, or the receive siteA web SDR → tailchaser-sdrreceiver → TCP port 5211
The voter: repeated, and to the talkgroup

TailChaser-SDRReceivertailchaser-sdrreceiver

A receiver without a radio. It listens to one FM channel on a web SDR and serves what it hears the way a tailchaser-receiver does. It needs no hardware and can run on the same box as the repeater, or anywhere else.

Web SDRs
OpenWebRX and OpenWebRX+, KiwiSDR, Web-888 and UberSDR. One connection, narrow FM on [sdr] freq: for a repeater that is its input frequency.
A good guest
It is one of the SDR's listeners, all day: ask the owner. It never changes what listeners share. On an OpenWebRX with several devices it moves its own connection to the device that is on the band, and waits when none is.
Squelch
A web SDR gives audio only, so the squelch is made from it: an FM receiver without a carrier hisses, a carrier quiets it. [ctcss] detect adds the tone, so it opens for the same stations as the repeater.
Signal strength
From the same hiss: full quieting is 100. That is what the repeater votes with.
Then
The same as a receive site: the gate (a voice, not a carrier), levelling, DTMF, the RemoteTrx link on port 5211 and a status page on 8081.

Tested, and not yet

OpenWebRX is tested on a live receiver. KiwiSDR, Web-888 and UberSDR are written from their source code and tested against stand-ins only, not against the real thing yet.

Voting

In tailchaser-transmitter, -duplex and -simplex

  • The receiver that hears a station first has the over.
  • Another one takes it when it hears the station clearly better for a while. That can be turned off.
  • A receiver on the internet is late, a web SDR by a second or two. What it still brings of an over that has ended on another receiver is the same over, and is not repeated. Only when it goes on for longer than it was late is it a station the other receiver lost. How late each receiver is, is learnt as it goes.
  • On a simplex node the other receivers hear the node's own transmitter, late: they are not listened to during a transmission and for a while after it.

The status page and the API say which receiver has the over and how many each has had; the log says where every over came from.

the web SDR · sdrreceiver.conf
[sdr]type     = openwebrxurl      = https://sdr.example.orgfreq     = 431.200       # the repeater's input[ctcss]detect   = truetone     = 131.8[uplink]listen_port = 5211auth_key = a-long-shared-secret

Listen to the input and the output, in the browser

For setting a node up and for finding out why something sounds the way it does: is the hiss there, is the tone, what does the normaliser make of a quiet station, what does an announcement sound like before the radio.

The Listen card of the status page: Input and Output under each other, each switched on, with a mute button, a level meter with a peak mark reading -6 and -9 dBFS, and a spectrum from 0 to 8 kHz
The Listen card, under the signal meter of the status page of a transmit site.
Input
The audio of the receiver, all of it, squelch open or not. On a transmit site it is the receiver that has the over.
Output
What goes to the transmitter, before the radio. A receive site and the SDR receiver have an input only.
A meter and a spectrum
Each has a level meter in dBFS with a peak mark and a spectrum of 0 to 8 kHz. Muted, they keep running.
Off unless turned on
[status] monitor = true. Whoever can open the status page can then listen, four listeners at most.

The streams are websockets, /monitor/rx and /monitor/tx: 16-bit mono PCM at 16 kHz, so a script can record them too.

Calibrate the audio without test equipment

Getting a repeater to sound like the others on the talkgroup normally takes a deviation meter or a service monitor. TailChaser needs neither: a web SDR that hears the transmitter is enough. It comes down to two things, and only one of them is yours to set.

1

Every source has one level

Nothing to set. Whatever is transmitted is brought to the same level before it reaches the modulator.

A station on the input
normalised to one target, per over
Audio from the talkgroup
normalised to the same target, looking 100 ms ahead
Spoken announcements
their own fixed level
Morse, roger beep, test tones
their own fixed level

A quiet station, a loud one, an SvxLink node that sends 10 dB too little: on the air they are alike, and there is no gain per source to tune.

2

The deviation: set once, measured in Hz

What is left is how far full level swings the transmitter. devcal.py measures that on a web SDR and says what to change.

$ python3 tools/devcal.py --sdr http://your.websdr.example/ \    --profile 70cm --freq 438.800 --ctcss 131.8 --max-dev 2500 \    --trigger "ssh node sudo systemctl kill -s USR1 tailchaser-transmitter"
carrier            +1030 Hz off (transmitter and SDR together)
CTCSS 131.8 Hz     266 Hz deviation
test tone 1000 Hz  625 Hz deviation at -20 dBFS
full level (0 dBFS) would be 6250 Hz — wanted 2500 Hz
change the transmit level by -8.0 dB: [tx] level_db -8.0 from what it is now

Set [tx] level_db, run it again, and it says “that is right”. That is the whole alignment.

The tools

devcal.py
The deviation of your transmitter in Hz, measured on any OpenWebRX receiver that hears it, and what to change to get it right. It listens only: it never switches the SDR's band profile for anybody else.
sdrmon.py
The other question, how you sound next to the neighbours: it tunes your repeater and reference repeaters in turns and prints speech level, peaks, bass, mid and treble balance and the CTCSS tone. Its numbers compare stations on the same receiver; they are not kHz.
Level in the log
Every over is logged with its level, for talkgroup audio and for a station on the input: one number to compare the two paths with.
ctcsstest
On a recording of the receiver: whether the tone squelch would open, the tone's level and margin, and the two numbers the noise signal meter needs.
agctest, gatetest, vadtest
What the normaliser and the gate would have done with real overs, before you change a live repeater.
Status page
Live: receive level, tone level and margin, noise level, normaliser gain and the S-meter.

Exact, without a calibrated receiver

The SDR listens to the carrier in SSB. There an FM signal is a row of tones: the carrier and, one tone frequency apart, its sidebands. How strong the sidebands are next to the carrier depends on the deviation alone (Bessel functions), not on the SDR's gain, its AGC or how strong the signal is. The test transmission sends 1000 Hz at a known level with receiver and talkgroup muted, so nothing else is in it.

Three signals put something on the air, each in a quiet moment

USR1
Test tones: 1000 Hz, then 700 + 1900 Hz, with receiver and talkgroup muted meanwhile
USR2
The next spoken announcement, now
RTMIN+1
This repeater's call in fast Morse, as other repeaters send it

An admin portal, and an API with keys

On the same port as the status page: the configuration in a browser, and a way for a display, a knob on the desk or a script to read what a node is doing and to block it from far away.

on0ora-simplex:8080/admin
The admin portal of TailChaser-Simplex: receiver and transmitter each open with a Block button, a Restart button, the configuration file /etc/tailchaser/simplex.conf in a text field with Check and save, and the API keys, a control key for a VFO knob in the shack and a read key for a status display, with the curl commands for the API below
The admin portal, as the program serves it, with an example configuration and example keys.

The portal http://node:8080/admin

Log in
One user, admin. On a new node the password is admin; the first login has to change it before anything else opens, and that first login works from a private address only: the local network, a VPN.
Configuration
The node's configuration file in the browser. Check and save lets the program itself check it first; what it refuses is not written, and the file before is kept.
Restart
At the next quiet moment. That is when a saved configuration takes effect.
Block
The receiver: nothing it hears is repeated or sent to the talkgroup. The transmitter: it is not keyed, whatever comes in. A block stays, also over a restart or a power cut, until it is released.
API keys
For other programs and devices. read keys see the status, control keys can also block. A key is shown once; only its hash is kept.

The API

# everything the status page shows$ curl -H "X-API-Key: KEY" http://node:8080/api/v1/status# what is blocked: {"rx":false,"tx":false}$ curl -H "X-API-Key: KEY" http://node:8080/api/v1/block# block the transmitter (a control key)$ curl -H "X-API-Key: KEY" -d "tx=1" http://node:8080/api/v1/block

Every answer is JSON, and the answer to a block is the state after the call. HTTP/1.0 is enough, with no TLS and no chunked encoding, so a microcontroller can speak it. The whole of it is in API.md.

A split repeater has one address: the transmit site

It fetches its receive site's status and passes it on in its own as remote_receiver, and blocking the receiver there blocks what comes from the receive site. The signal strength of the receiver that has the over is in gate.rssi, so a display reads the same field on every program, and voter says which receiver that is.

It is plain HTTP: keep it on the local network or a VPN, or put a TLS proxy in front of it. [status] admin = false turns the portal and the API off.

Speaks the API

Your repeaters, on a knob on the desk

RF.Guru's VFO Knob has a TailChaser firmware: a remote for TailChaser repeaters, over the API above. It watches them, plays what they transmit and blocks their receiver or transmitter.

  • A repeater a detent. Up to eight, every one read all the time, so a turn shows the next at once. One that becomes busy comes up by itself.

  • Who is talking. The talkgroup's talker while the repeater sends the over, LOCAL for a voice on its input, CARRIER for a signal with no voice, or who spoke last and how long ago.

  • The repeater's ear. Its received signal on the S-meter, in S-points on TailChaser's own scale; TX in red while it is keyed.

  • Listen. What the repeater shown transmits, on the jack or a Bluetooth headset or speaker. That needs [status] monitor = true on the node.

  • Block RX or TX. The repeater's receiver or transmitter blocked, or opened again, with a tap each. The knob itself never transmits.

  • From anywhere. On the LAN, over the internet, or over your own Tailscale tailnet, which the knob joins as a device of its own.

The VFO Knob with a talker on the talkgroup, on the air: the call, the talkgroup and the repeater's signal on the S-meterThe VFO Knob with the repeater's receiver and transmitter, each blocked or opened with a tap

What it needs. An API key for each repeater, made in its admin portal: a read key watches, a control key may also block. To hear one, [status] monitor = true in its configuration.

Install it

Four steps on a Raspberry Pi. A box that was not set up does not transmit.

  1. 1

    Clone it on the Pi

    Raspberry Pi OS or Debian, Bookworm or Trixie, with a WM8960 Audio HAT and the SA818 on /dev/serial0. Other sound cards and radios work too.

  2. 2

    Run bootstrap.sh with the role

    simplex, duplex, receiver, transmitter or sdrreceiver. It installs what is needed, builds the program and sets up its service.

  3. 3

    Fill in the REQUIRED lines

    The reflector, the callsign, the talkgroup and the frequencies. Until they are filled in the program refuses to start: no restart loop, no carrier.

  4. 4

    Reboot, and open port 8080

    Once the reflector's sysop has signed the node's request it logs in by itself, and the status page shows what it hears, what it passes on and what it kept off the talkgroup.

On the Pi

$ git clone https://github.com/Guru-RF/TailChaser && cd TailChaser$ sudo ./bootstrap.sh simplex          # or duplex | receiver | transmitter | sdrreceiver$ sudo nano /etc/tailchaser/simplex.conf   # the lines marked REQUIRED$ sudo reboot

On a box that already runs the program, bootstrap.sh <role> is an update: rebuild, install, check the configuration, restart. By hand it is make && sudo make install ROLE=simplex, and tailchaser-<role> -t -c FILE checks a configuration file.

Coming from SvxLink

One command takes the settings, the reflector key and the certificate over, stops the old program and starts the new one. If the new one does not log in on the reflector within a minute, the old one is put back.

svxsimplexd (SVXSimplexRepeater)
sudo ./bootstrap.sh simplex upgrade
svxrxd (SVXRepeaterRX)
sudo ./bootstrap.sh receiver upgrade
SvxLink: SimplexLogic + ReflectorLogic
sudo ./bootstrap.sh simplex replace-svxlink
SvxLink's remotetrx
sudo ./bootstrap.sh receiver replace-remotetrx
Back to what ran before
sudo ./bootstrap.sh <role> rollback

Not there yet

What TailChaser does not do today, said plainly.

Three of the four web SDRs, on the air

tailchaser-sdrreceiver is tested on a live OpenWebRX receiver. KiwiSDR, Web-888 and UberSDR are written from their source code and tested against stand-ins only.

A carrier squelch for a radio

A radio without a squelch line needs a CTCSS tone on its input. A noise squelch in software is there for a web SDR, not yet for the input of a sound card.

Taking over an SvxLink repeater

A simplex node and a remote receiver are taken over from SvxLink by one command. A repeater (RepeaterLogic) is not: write its configuration from the example and copy the key and certificate.

Modules

No EchoLink, parrot, voice mail, APRS, scripting or sound packs, and no DTMF command trees beyond talkgroup selection. SvxLink has those.

Specifications

Programs
tailchaser-simplex, -duplex, -receiver, -transmitter and -sdrreceiver, from one source tree
Language
About 14 000 lines of C; no interpreter in the daemon
Runs on
Raspberry Pi OS and Debian, Bookworm or Trixie; one systemd unit per program
Reflector
SvxReflector protocol version 3 (TLS with certificates, encrypted audio) and version 2 (password); Opus audio
Split site
SvxLink's RemoteTrx protocol on TCP port 5210; uncompressed audio (256 kbit/s) or Opus
Receivers
The one in the box or at the receive site, and up to three or four others over RemoteTrx, with voting; in simplex, duplex and transmitter
Web SDR
tailchaser-sdrreceiver on OpenWebRX and OpenWebRX+ (tested live), KiwiSDR, Web-888 and UberSDR (stand-ins only); squelch and signal strength from the hiss; serves on TCP port 5211, status page on 8081
Radio
SA818 module, programmed over its serial port, or any radio by audio and GPIO lines
Sound
ALSA; the mixer of a WM8960 Audio HAT is set by the program, other cards with alsamixer
Squelch
The radio's squelch line, a CTCSS tone found in the audio, or both
Gate
Voice detector on 20 ms frames, 300–3000 Hz, with a look-back buffer; on, monitor or off
Normaliser
One speech level for every over: target −20 dB, up to 30 dB of gain, on by default
Transmit
TX delay 300 ms, hang 3 s, time-out, optional software pre-emphasis and limiter
Identification
Morse on the clock; spoken announcements through an OpenAI-compatible service or ElevenLabs
Status
Web page and /status.json on port 8080, read-only
Listening in
Off by default. A Listen card on the status page: input and output, each with a level meter in dBFS, a spectrum of 0 to 8 kHz and a mute button. Websockets /monitor/rx and /monitor/tx: 16-bit mono PCM at 16 kHz, four listeners at most
Admin and API
/admin and /api/v1 on the same port, plain HTTP: one user, API keys with read or control rights; can be turned off
Builds with
libasound2, libopus, libgpiod (1.x or 2.x) and libssl
Licence
GNU General Public License, version 3 or later

Free software

GPL-3.0-or-later

Anyone may run it and work on it. Whoever distributes it, changed or not, ships the source with it under the same licence and keeps the names of those who made it.

CsystemdOpusOpenSSL

Standing on shoulders

  • SvxLink / SvxReflector, Tobias Blomberg SM0SVX and contributors: the reflector and RemoteTrx protocols TailChaser speaks. No SvxLink code is included.
  • Opus (Xiph.Org), OpenSSL, ALSA, libgpiod: the codec, the TLS and the hardware underneath
  • The sysops and stations of ON0ORA and ON0DEN: who ran it on the air first, keyed up on request and said how it sounded
  • RF.Guru (Joeri Van Dooren ON6URE): the author

Where it comes from

TailChaser is the merge of SVXSimplexRepeater (svxsimplexd) and SVXRepeaterRX (svxrxd). Both can be upgraded in place.

Questions, answered

Is TailChaser a fork of SvxLink?
No, and it contains none of its code. It speaks SvxLink's protocols, SvxReflector and RemoteTrx, so the two mix freely on a talkgroup and on a split site. What differs is the scope and the defaults: TailChaser does one job, an FM repeater or simplex node on a reflector talkgroup.
Can it sit on a talkgroup next to SvxLink nodes?
Yes. The reflector link is the SvxReflector protocol, version 3 with certificates and encrypted audio or version 2 with a password, with Opus audio, talkgroups and monitored talkgroups with priorities. A TailChaser node sits on a talkgroup next to SvxLink nodes and neither notices the difference. And tailchaser-receiver serves a tailchaser-transmitter or an SvxLink repeater with a TYPE=Net receiver.
Does the gate cut off the first word?
No. Audio is kept in a look-back buffer while the detector listens, and an over that has a voice on it is passed on from its first word. The price is that the first over on an idle channel reaches the talkgroup up to the look-back time behind. After an over that passed, the channel is in conversation for hold_s and the next over passes at once. In heavy noise a detection can come later than the buffer is long: at 0 dB above the noise the lab voices lost 0.15 to 0.3 s of their start.
How do I try the gate without taking a risk?
Set [gate] mode = monitor. Everything passes as with a plain squelch, and the log and the status page say for every opening what the gate would have done. Run it for a day and read the page. A node that is taken over from SvxLink starts in monitor mode for that reason.
What hardware does it need?
A Raspberry Pi with a sound card and a radio. The installer assumes a WM8960 Audio HAT and an SA818 module on /dev/serial0; the first node on the air, ON0ORA simplex, is a Pi Zero 2 W with exactly that. Other sound cards work, with their levels set in alsamixer, and [radio] type = none takes any radio that has its audio on the sound card.
My receiver has no squelch line. Will it work?
With a CTCSS tone on the input, yes: [ctcss] detect = true finds the tone in the audio and opens on that alone. Such a set has no RSSI either, so [siglev] mode = noise makes a signal meter out of the receiver's hiss above the speech band. A noise squelch for a radio's input without tone is not there yet.
Can a repeater have more than one receiver?
Yes. tailchaser-transmitter, -duplex and -simplex take up to three or four receivers besides their own ([receiver2] and on) and vote. The receiver that hears a station first has the over, and another takes it when it hears the station clearly better for a while. The status page says which receiver has the over, and the log says where every over came from.
Can I add a receiver without a radio?
Yes: tailchaser-sdrreceiver listens to one FM channel on a web SDR, for a repeater its input frequency, and serves it like a receive site. It needs no hardware and runs on the same box as the repeater or anywhere else. It is one of the SDR's listeners, all day, so ask the owner first. OpenWebRX is tested on a live receiver; KiwiSDR, Web-888 and UberSDR are tested against stand-ins only.
A web SDR is a second or two late. Is everything repeated twice?
No. What a late receiver still brings of an over that has ended on another receiver is the same over, and it is not repeated. Only when it goes on for longer than it was late is it a station the other receiver lost, and then it gets the over. How late each receiver is, is learnt as it goes. On a simplex node the other receivers hear the node's own transmitter, late: they are not listened to during a transmission and for a while after it.
Can I hear what the node hears?
With [status] monitor = true, yes: a Listen card appears under the signal meter of the status page, with Input and Output under each other: the receiver, squelch open or not, and what goes to the transmitter. Each is off until it is switched on, and has a level meter in dBFS with a peak mark, a spectrum of 0 to 8 kHz and a mute button; muted, the meter and the spectrum keep running. A receiver and the SDR receiver have an input only. It is off unless it is turned on, because whoever can open the status page can then listen, four listeners at most. The streams are websockets, /monitor/rx and /monitor/tx, so a script can record them too.
Can I go back to SvxLink?
Yes. Taking over is one command, and nothing of the old setup is removed. If the new program does not come up, or does not log in on the reflector within a minute, the old one is put back by itself, and bootstrap.sh <role> rollback does it on request. An SvxLink repeater (RepeaterLogic) has no automatic take-over yet: write its configuration from the example and copy the node's key and certificate.
Does it do EchoLink, a parrot or voice mail?
No. Those are SvxLink modules, and SvxLink is the right program for them. TailChaser identifies in Morse, can say announcements you write as text, and takes DTMF for talkgroup selection only.
Can I switch a repeater off from far away?
Yes. The admin portal at /admin blocks the receiver or the transmitter, and so does the API with a control key, from a script or a small device on the desk. A blocked transmitter is not keyed, whatever comes in: no repeating, no talkgroup audio, no identification, no announcement. The block is still there after a restart or a power cut, and the log says which key set it and from where. A split repeater has one address for this, its transmit site.
Is the admin portal safe to put on the internet?
Not as it is. It is plain HTTP: keep it on the local network or a VPN, or put a TLS proxy in front of it. The first login, with the password every new node starts with, only works from a private address, and [status] admin = false turns the portal and the API off altogether.
Can I test it without a radio?
Yes. make NOGPIO=1 check on a desktop runs the real programs with WAV files as squelch openings and a recording as the transmitter, against stand-ins for the reflector, the SA818 and the repeater, in about three minutes. make interop puts two nodes through a real svxreflector in Docker.
What does it cost?
Nothing. TailChaser is free software under the GNU General Public License, version 3 or later: anyone may run it and work on it, and whoever passes it on ships the source with it under the same licence. There is no warranty.

Every setting and command in the README, how the gate decides in DETECTION.md.

Running it on your repeater?

Tell us how it sounds. Bugs, ideas and recordings of overs the gate got wrong go to GitHub; add your name to AUTHORS with your first contribution.