Field Guide · term

Also known as: voice_taps, wideband DDC taps, virtual voice tuners

Wideband voice taps are per-call digital down-converter channels carved out of one SDR’s IQ capture, so a single dongle can decode a trunked control channel and record the voice grants it issues. On a trunked system the control channel and each voice call sit on different frequencies, so one tuner cannot do both jobs by retuning — either you add a second SDR with role: voice, or you channelize: give the dongle role: wideband, point center_freq_hz at the middle of the site’s frequency cluster, and set voice_taps: N. Each grant whose frequency lands inside the capture window then gets its own on-demand DDC tap on the same IQ stream, no extra hardware needed (#379).

The window math

A tap can only be created where there is capture to tap. The validated rule is:

  • Every tapped frequency must fall within center_freq_hz ± sample_rate/2, with a 5% guard band at each edge (the channelizer and DDC filters need clean skirts).
  • voice_taps: 0 disables the feature; there is no hard upper bound, but CPU scales roughly linearly per tap and the daemon warns above 16.
  • Grants that fall outside the window spill over to a physical role: voice SDR when one is configured, and otherwise drop with the standard no voice device available log.

So a 2.4 MS/s RTL-SDR capture gives a usable span of about ±1.1 MHz around center — plan the center frequency around where the voice channels actually are, not just the control channel.

Misleading messages and the bugs behind them

Symptom Looks like Actually Fix / check
voice pool full but no actives Pool exhausted by traffic The pool was empty — the message fired only when the daemon had zero voice devices (a control-only config), the exact opposite of “full” (#379) Add voice_taps to the wideband entry or a role: voice SDR; current builds emit an actionable warning instead
Wideband-only setup drops every grant as “no voice SDR” Grants out of window Init ordering: the virtual voice tuners were created after the voice pool was built, so the pool had zero devices; a physical voice SDR would have masked it entirely (#422) Fixed; on old builds, adding any physical voice SDR hid the bug
composer: p25p2 voice chain started trellis=0 Config problem on your side The wideband path never forwarded the p25_phase2_* FEC options to the decoder, even though the config read back correctly through the API (#882) Fixed; trellis=1 in that line is the confirmation the options arrived
Wideband P25 voice garbled on a CQPSK site Weak signal The wideband path stamped no demod mode onto its grants, so voice always ran the C4FM chain regardless of the setting (#882, #935) Fixed; set the per-channel p25_phase1_demod_mode override where needed

The middle two rows are the same lesson twice: the wideband engine is a second, parallel pipeline, and “the config is accepted and echoed back by the API” does not prove the value ever reached the decoder. The observable tell is always a decoder-side log line, not the config surface.

Per-channel demod mode

Each channels: entry on a wideband dongle can carry its own p25_phase1_demod_mode: override, inheriting the system default when absent. The override is keyed by frequency, not by RFSS/site identity, for a structural reason: the symbol-recovery path must be chosen before the control channel locks (it is what lets it lock), but the RFSS and site numbers are only known after that control channel decodes — a chicken-and-egg the config resolves by using the channel frequency, the one place site identity exists at config time (#935). See P25 demod-mode selection for how to choose the value.

  • signalling_taps: N allocates signalling-only taps that harvest P25 Phase 2 talker aliases off granted traffic channels without following the call as voice — useful where voice taps rarely win a grant.
  • A multi-tap wideband dongle usually wants gain: "auto"; a single fixed gain cannot serve sites of differing strength, and overload has its own signature — see SDR gain and overload.

Provenance

  • #379 — the backwards “voice pool full” message and the one-SDR-can’t-do-both explanation that motivated voice_taps.
  • #422 — wideband-only topology dropped every grant because the virtual voice tuners were built after the pool.
  • #882 — wideband pipeline silently dropped the P25 Phase 2 FEC options; the trellis=0 tell.
  • #935 — per-channel demod-mode override and why it is keyed by frequency rather than site.

See also

From the blog