DMR repeater roaming

A user radio on a wide-area system of conventional DMR repeaters (Tier II, repeater mode) keeps a list of repeater sites that carry the same talkgroups. While idle it watches how well it hears the site it is on and, when the repeaters send idle beacons, surveys the other sites in the beacon window. When its site gets weak it moves to the strongest one it can use, from the survey data when that is fresh, else after a search. The user can lock the radio to the site it is on, or ask for a search at once. Switch: CONFIG_DMR_ROAMING, a child of CONFIG_DMR, on by default (docs/build-config.md). Roaming itself is off until the user turns it on for a zone.

1. Legal and standards position

ETSI TS 102 361-1..4 define no roaming procedure for Tier II conventional operation: a repeater neither announces its neighbors nor tells a radio to move. Tier III (trunking, TS 102 361-4) has its own registration and hunting procedures, which do not apply to conventional repeaters. Roaming between conventional repeaters is therefore a radio-side feature, and this one uses only:

• the received signal strength the RF chip reports (hal_radio_t.status),
• the presence of the repeater's outbound: base station (BS) sourced sync with the expected color code, which the call stack already tracks for its own repeater handling (dmr_call_view_t.repeater_up),
• optionally the standard BS downlink activation CSBK (BS_Dwn_Act, TS 102 361-2), the same burst the radio sends before every transmission through a repeater that is not up, to wake a silent site for a measurement.

It does not implement, imitate or interoperate with any vendor's multi-site signaling: no Motorola IP Site Connect, Capacity Plus or Linked Capacity Plus site beacons, rest channel or site announcements, no Hytera XPT, and no proprietary CSBK or data opcodes. Picking the strongest of several programmed channels by signal strength, and waking a repeater with the standard activation burst, are generic radio behavior that predates those systems (scan with priority, "strongest signal" selection in analog trunked and conventional radios).

Beacons. Many conventional repeaters can key their standard outbound on their own for a few seconds every interval while idle ("roaming beacon", "BS beacon"; networked repeaters are often set to do it at the same time). The radio treats that as what it is on air: a standard ETSI BS-sourced outbound (BS sync, CACH, idle or any other burst) with the site's color code. It measures it as it measures traffic (section 3) and learns when it comes (section 4.8). It decodes nothing proprietary from it: no site or neighbor lists, no rest channel, no vendor CSBK or data. Listening to a repeater at the time it is known to transmit is generic receiver behavior, as is a periodic look at other programmed channels (scan with priority). The survey (section 4.9) and the periodic search (section 4.2) only listen. Our own repeater roles can send such a beacon (section 7): idle bursts on the outbound, nothing else.

2. The roam list (sites)

A site is a channel in the current zone that:

• is DMR, not RX-only, and a repeater channel (TX frequency set and different from RX, not "force DMO"),
• has its RX frequency inside one of the board's bands,
• talks to the same talk target as the channel the user picked (the resolved contact ID and call type; two channels with no contact both talk to the default talkgroup 9),
• and, when any channel of the zone has the open-firmware Roaming flag set (section 5.2), has that flag set too.

The sites are the first ROAM_SITES_MAX (16) such channels in zone order. Roaming is active when the zone is a roaming zone (section 5.1), the radio is on one of its sites (not the VFO, not an FM channel, not a channel that fails the rules above), and the zone has at least two sites.

The channel the user picked is the home site. Roaming moves the radio between sites without changing home; a channel, zone or VFO change by the user (keys, Find, the debug link, a CPS write that moves the channel) makes the new channel home and restarts the minimum dwell. The roaming moves themselves are remembered like any channel change, so the radio powers up on the site it was on.

Why a zone and not a new list: the open-firmware codeplug layout that qdmr and the open-firmware CPS write (open-firmware-cps.md 7) has no roaming list or roaming zone record, and the native codeplug (codeplug-native.md) is a draft. A zone whose channels share their talkgroups is what a user of a multi-site system already programs, so it is the natural roam list; the per-channel Roaming flag the open-firmware layout does have narrows it when a zone mixes sites with other channels. When the native codeplug is built, the zone record gets a "roaming zone" flag and the threshold values (section 10, decision 1), and the settings store bits become its default.

3. What the radio measures

• Site heard: dmr_call_view()->repeater_up: a burst with BS sync and the channel's color code within the last 300 ms (DMR_CALL_REPEATER_HOLD_MS). Bursts with another color code are dropped by the call stack before they count, so a co-channel repeater with another color code is never taken for ours.

• Signal: hal_radio_t.status().rssi_dbm, read every poll (25 ms) while the site is heard. The board picks the source: the receiver's RSSI, not the DMR modem's.

• Current site average: an exponential average of those readings (weight 1/8 per reading), restarted on every site change. Readings taken while the radio isn't hearing the site don't count, so a silent repeater keeps its last average until it is too old (ROAM_FRESH_MS, 60 s): then the site's signal is unknown.

• Candidate measurement: during a search, the strongest reading taken while the candidate is heard, after the settle time.

• Survey measurement: during a survey visit (section 4.9), the mean of up to ROAM_SURVEY_READINGS (4) readings taken while the site is heard, after ROAM_SURVEY_SETTLE_MS (420 ms: the call stack's 300 ms hold of the last site's outbound, then about 120 ms for the PLL, the modem's sync and the receiver's RSSI filter). The site's survey average is the mean of its last two survey measurements; a site missed in two visits made while the beacons were on the air ("live": its own site or another one was heard in that window) counts as not heard.

• Traffic: a call or data on our slot (RS_RX, or the call stack not idle) while a site is heard. A keyup that carries traffic is never taken for a beacon.

Why 420 ms per visit and not less: all sites share the color code, and the call stack keeps a repeater "up" for 300 ms after its last burst, so the first 300 ms after a retune can't tell the new site from the old one. A visit could be about 250 ms if the stack forgot the last site on a retune, but then a PTT that ends a visit would always go through a repeater wake-up on the way back (section 4.9). A board's receiver RSSI and modem sync acquisition should need well under 100 ms on a continuous BS outbound (a burst every 30 ms); four 25 ms readings follow.

A conventional repeater only transmits while it carries traffic (plus its hang time), when it beacons, or when something wakes it. Passive measurement therefore only sees sites that are busy or beaconing; section 4.5 covers silent ones.

4. Behavior

4.1 State machine

4.2 Triggers

A trigger is latched: a weak call that ends with the repeater's hang time still starts its search once the radio is idle and nothing blocks. The latch clears when the average climbs back to the threshold (for weak) or when the search runs. On an idle system the current site is heard only in its beacons; its average stays known for two intervals plus the duration (at least ROAM_FRESH_MS), and the weak test is also made between readings, so a weak idle site latches weak after its first weak beacon plus ROAM_WEAK_MS.

With the survey on and beacons set, weak and manual first look at the survey data (section 4.10); tx_fail, check and stale always search.

4.3 Never roam while

The block list (roam.blocked), checked before a search or a survey round starts, and during a survey visit (any reason ends the visit at once):

A request from the user (any key action that reaches the radio core: PTT, channel or zone change, monitor, a temporary change...) during a search or a survey visit ends it at once: the radio goes back to the site it started from (the origin) before the request is handled, so PTT always transmits on the site the user saw before the search. A call that plays on a candidate during the search ends the search early on that candidate if it qualifies (section 4.6), so the call keeps playing; otherwise the search goes on.

4.4 The search

1. The visit list is every site except the origin, in zone order; the origin is visited too when its signal is unknown (section 3).
2. For each: tune (the radio core's normal channel change; the call stack is reconfigured for the site's color code), wait ROAM_SETTLE_MS (400 ms: longer than the 300 ms the call stack keeps the last site's outbound alive), then listen ROAM_LISTEN_MS (1.5 s), or ROAM_PROBE_MS (2.5 s) when a wake probe is sent (section 4.5). Readings are taken as in section 3.
3. After the last visit the radio chooses (section 4.6) and tunes the choice, or the origin.

A search over 4 other sites takes about 7.6 s passive, 11.6 s with probes. The status strip says ROAM during it and the site name on Home follows the visits, as it does for a scan.

4.5 Silent repeaters: the wake probe (optional)

roam_probe = wake sends one BS_Dwn_Act CSBK on each visited site that is not already heard, through the call stack's data path (dmr_call_data_tx, which wakes the repeater with its own activation bursts and retries, up to DMR_CALL_WAKEUP_ATTEMPTS). A repeater in range answers by keying its outbound for its hang time, which the radio measures. Probes are only sent when the radio could transmit on that channel anyway: a DMR ID is set, TX inhibit is off, the battery is not critical, the TX frequency is in a TX band. They are counted (roam.probes).

Because the probe transmits and keys up every repeater in range, it is off by default (P7: no automatic transmission until the user turns it on). Tools > Roaming row 7 and debug link setting 23 set it. With probing off, roaming relies on beacons (sections 4.8 - 4.10) and traffic: on a system without beacons a repeater nobody uses is not found until it carries a call, but a failed transmission (tx_fail) still starts a search, and "Roam now" while the network is busy finds the others.

4.6 Choosing

Values: T = the threshold (roam_rssi_dbm, default -105 dBm), H = the hysteresis (roam_hyst_db, default 6 dB), O = the origin's signal (its average, or its measurement from this search; unknown = -200; for tx_fail the origin counts as unknown unless it was heard during the search).

• A candidate qualifies when it was heard and its measurement is at least T.
• B = the strongest qualifying candidate other than the origin.
• If home qualifies, is not the origin, is within H of B, and is at least O + H: go home (return to home wins ties).
• Else if B is at least O + H: go to B.
• Else stay on the origin.

Moving restarts the minimum dwell (ROAM_DWELL_MS, 30 s) and the average (seeded with the measurement). Staying starts the retry back-off for weak and stale: 30 s, doubled after each fruitless search up to 240 s, back to 30 s after a move or once the site is good again (average at least T + H).

4.7 Lock and roam now

• Lock to this site: state locked until unlocked, "roam now", or a channel change by the user. The site does not become home; the lock is RAM only.
• Roam now: a manual search at once (the block list still applies; blocked, it waits as a latched trigger and the UI says why).

4.8 Beacons and the beacon schedule

Settings (radio-wide, section 5.1; they must match the repeaters, which the radio can't read):

Beacon detection. The radio follows the keyups of the site it is tuned to (its own site while watching, the visited one during a visit). A keyup is a beacon when it carried no traffic and lasted about the beacon duration: from half of it to the duration + 1.5 s when its start was seen (at least ROAM_CLEAN_MS, 200 ms, of silence before it), or at most the duration + 0.5 s when the radio tuned in during it (its start is then its end minus the duration). The end is the last burst: "heard" falls ROAM_HOLD_MS (300 ms) after it.

Synced (the main case). A networked system keys all its repeaters' beacons in one shared window: the radio learns the window from its own site's beacons (it is there most of the time) and from any site heard beaconing. The first beacon gives a tentative window (win_conf 1, enough for a round); a beacon within ROAM_WIN_TOL_MS (1 s) of the schedule confirms it (win_conf 2, 3) and pulls it a quarter of the way toward itself (jitter); a site heard in the predicted window during a round confirms it too. A beacon off the schedule is held as an alternative, and a second one in step with that moves the schedule (the network restarted). The schedule is forgotten after ROAM_WIN_STALE (30) intervals without a matching beacon (crystals drift a few ms per minute). When its own site gives no beacon for two intervals plus the duration while it is weak or not heard (out of range), the radio makes one discovery visit: it listens to the site with the oldest data through a whole interval (at most interval + duration + 1.4 s), and learns the window from that site's beacon.

Per site (secondary). Repeaters that beacon on their own clocks, each with its own phase: the radio keeps each site's last beacon start (phase_ms) and visits one site just before its next beacon (section 4.9). A site with no known phase gets a discovery visit (up to an interval + duration + 1.4 s). Phases also last ROAM_WIN_STALE intervals and are refreshed by every visit that hears the site on time.

No beacons (roam_beacon_s 0) or survey off: no survey; roaming works as in sections 4.1 - 4.7 (traffic, the optional wake probe, the optional periodic search).

4.9 The survey

While watching (not locked), with beacons set and the survey on, and nothing on the block list:

• Synced: in each window, ROAM_WIN_LEAD_MS (400 ms) after its start (the own site's beacon has begun, its start is seen and refines the schedule), a round visits as many other sites as fit before the window's end minus ROAM_WIN_TAIL_MS (300 ms), oldest data first (never measured first), so the list rotates across windows. Each visit lasts until four readings or ROAM_SURVEY_VISIT_MS (420 + 250 ms). With the defaults (4 s), six sites fit in a window, so a 16-site zone is covered in three windows. Outside the window the radio stays on its site: nothing else is on the air.
• Per site: one visit at a time, to the site with the oldest data, tuned ROAM_SURVEY_SETTLE_MS + ROAM_CLEAN_MS + half the tolerance before its expected beacon (so its start is seen and refines its phase), until four readings or the tolerance + 250 ms past the expected start.

Bounds:

• Time away: a round is at most the window minus 700 ms (3.3 s with the defaults). After a round or a visit the next one waits nine times as long as it was away (ROAM_SURVEY_DUTY, 10 %): with 60 s / 4 s that never skips a window; with a 10 s interval it skips two. A discovery visit (about an interval) is followed by about nine intervals without a survey.
• Traffic on the own site: no round within ROAM_SURVEY_HOLD_MS (10 s) after traffic there, doubled per busy period up to 160 s (busy_score, 0 - 4); each quiet round lowers the score by one. The radio can't hear its own site while it is away: a call that starts on it during a round is joined late, at most one round (3.3 s) late. In the synced case the own site is idle in the window (it is beaconing), so little is missed. A call heard on a visited site ends the round at once (on a networked system it is likely carried on the own site too).
• PTT: ends a visit at once (roam_abort, counted in roam.cuts); the radio retunes to its site and the transmission starts there. The call stack still holds the visited site's outbound as "up" for up to 300 ms, so with the own site beaconing in the window the voice starts without a wake-up: the extra delay is the retune (the host test measures PTT-to-voice within 100 ms of PTT on the own site without a survey). After a visit to a silent site longer than the hold, the call stack wakes the repeater first, as after any channel change.
• Screens: during a survey visit the screens and the debug link's core view keep showing the site the radio returns to (radio_sm_shown_view, applied when the view is published), so Home doesn't flicker once a minute. roam.state=survey and roam.visit say where the radio is.

Results: each visit updates the site's survey average and its time (surv.<i> on the debug link). Survey data is fresh for roam_fresh_ms(): synced, (windows needed to cover the zone x windows between rounds + 2) intervals; per site, (other sites + 2) intervals; never less than ROAM_FRESH_MS.

4.10 Moving with the survey data

A weak or manual trigger that is ready (dwell, back-off and block list as in 4.2 and 4.3), with beacons set and the survey on:

1. Among the other sites with fresh data, B = the strongest that reaches the threshold T; home wins instead when it is fresh, reaches T, is within H of B and H better than the current site (as in 4.6).
2. If B is at least O + H (O = the current site's average, unknown = -200): a confirm visit to B when it is next on the air: now if it is in its beacon with time left, else just before its next beacon (the shared window, or its own phase), else now (traffic only). The visit listens up to ROAM_SURVEY_SETTLE_MS + ROAM_CONFIRM_MS (800 ms) or eight readings. B's mean reaching T and O + H: move (the radio is already there, roam.roams + 1, a confirm counted in roam.confirms). Else stay and back off (4.6); B's survey data takes the new measurement.
3. Else, if every other site's data is fresh: nothing is better; weak stays and backs off without a search, manual runs the full search (it also catches traffic).
4. Else, stale data with a known window: the stale sites are surveyed first, in every window regardless of the duty cycle, and step 1 runs again when they are fresh. If that takes longer than the freshness time, a full search.
5. No schedule: the full search (4.4).

A weak site on a synced system therefore moves within about one interval of the weak trigger (the next window), without visiting sites that are known to be worse, and without the full search's 1.9 s per site, which on an idle system would hear nothing.

4.11 Defaults: what turning roaming on for a zone does

Roaming for a zone (Tools > Roaming row 1) turns on: watching the site, the weak trigger, and the synced survey with 60 s / 4 s beacons. The survey only listens, so P7 allows it on by default; 60 s / 4 s is what networked systems commonly use. On a system without beacons it costs nothing: no beacon is learned, so no round runs (a discovery visit happens only when the own site is weak or not heard). The wake probe stays off: it transmits and keys every repeater in range. The periodic search (roam_check_s) stays off: the survey does its job without the 1.9 s per site. If the repeaters beacon on another interval, or not in step, the user sets rows 4 - 6.

5. Storage

5.1 Settings store (settings-store.md 3)

Appended to settings_t, layout 1:

The last four are appended after reserved5 (payload 168 -> 176 bytes, then reserved6[3]). A record written before them is shorter: the loader keeps the defaults for the missing bytes (settings-store.md 3), so an upgraded radio gets the survey on with 60 s / 4 s and the repeater beacon off.

The values reach the radio core as radio_settings_t.roam_*. The zone bits are written by the radio core when the user turns roaming on or off for the current zone. They follow the zone index, so a CPS write that reorders or deletes zones moves the flags with the indices (section 10, decision 2).

5.2 Open-firmware codeplug

The open-firmware channel record's flag 1 bit 0 "Roaming" (open-firmware-cps.md 7, offset 0x26) is read into channel_t.roam. It is never written by the radio (the encoder keeps the bit as read). When any channel of a roaming zone has it, only those channels are sites. Whether qdmr exposes this bit for the open-firmware layout is open (section 10, decision 3).

6. Interfaces

6.1 Radio core

• radio_req_roam(cmd) (RREQ_ROAM): RROAM_ZONE_OFF 0, RROAM_ZONE_ON 1, RROAM_ZONE_TOGGLE 2, RROAM_NOW 3, RROAM_LOCK 4, RROAM_UNLOCK 5, RROAM_LOCK_TOGGLE 6.
• radio_view_t.roam (radio_roam_view_t): zone flag, state, trigger, block reason, site count, current and home index, the current average, counters, and per site its channel number and last measurement; for the survey the visit kind, the schedule's confidence and next window, the pending confirm, the survey counters, and per site its survey average, its age, its beacon phase and whether it is fresh.
• radio_sm_shown_view(): the view the screens get during a survey visit (the site the radio returns to).
• The engine is core/radio/roam.c (pure logic: sites, averages, triggers, the search sequence, the choice; no hardware and no codeplug, like scan.c); core/radio/radio_sm.c builds the site list, tunes, measures and applies the block list. The scan engine is not reused for the search: it stops on a signal, roaming has to measure every site and compare. What is shared is the radio core's tuning path (set_channel + tune), the scan's settle idea and the call stack's repeater tracking.

6.2 UI (docs/ux)

• Home: in DMR chips, Roam after the repeater chip when roaming is active (Lock when locked, amber Roam? when the current site is below the threshold). The status strip shows ROAM (amber) during a search, as it shows SCAN.
• Tools > Roaming (Hub 6, last row: 8, or 7 in a build without diagnostics, which keeps 7): 1 "Roam in this zone" On/Off (the zone flag), 2 "Roam now" (goes Home, where ROAM shows the search), 3 "Lock to this site" On/Off, and a body with the state: the site (its position in the list and the channel name), its signal, or why roaming is not available here ("Turn it on for this zone", "This channel is not a repeater site", "Needs two sites in the zone with the same talkgroup", not in the VFO). Rows 2 and 3 are dimmed while roaming is off here. Radio-wide rows (all builds with roaming): 4 "Site survey" Synced / Per site / Off, 5 "Beacon interval" (60 s, 2 min, 5 min, 10 min, None, 10 s, 15 s, 20 s, 30 s; Enter steps through them), 6 "Beacon length" (4, 5, 6, 8, 10, 1, 2, 3 s), 7 "Wake probe" On/Off. The body adds "Beacons: surveying" (schedule known) or "Beacons: learning". The hint reads "1-7 pick Back". Roaming is not in Quick settings: the nine tiles are taken (information-architecture.md 5) and roaming is set per zone, not changed per call.
• Toasts: "Roaming on for ", "Roaming off", "Moved to " (after any move, the radio core's short scan hold beep when beeps are on), "Searching sites", "Locked to ", "Roaming again" (unlocked), "Turn roaming on first" / "Can't roam on this channel" (rows 2 and 3 while roaming is off here), "Survey: ", "Beacons every " / "No beacons", "Beacon length s", "Wake probe on" / "Wake probe off".
• During a survey visit the screens keep showing the site the radio returns to (section 4.9); no chip changes.

6.3 Debug link

View section 12 roam (docs/spec/debuglink.md 3.7): roam.built roam.zone_on roam.state roam.why roam.pending roam.blocked roam.weak roam.sites roam.site roam.home roam.channel roam.rssi_dbm roam.searches roam.roams roam.stays roam.aborts roam.probes roam.triggers roam.visit roam.win_conf roam.win_next_ms roam.confirm roam.surveys roam.visits roam.cuts roam.beacons roam.confirms set.roam_rssi_dbm set.roam_hyst_db set.roam_probe set.roam_check_s set.roam_beacon_s set.roam_beacon_len_s set.roam_survey set.rpt_beacon, site.<i>=<channel>:<rssi> for each site (rssi 0 = not measured) and surv.<i>=<dbm>:<age_s>:<phase_ms>:<fresh> (dbm 0 = never surveyed, age -1 = never, phase -1 = unknown, fresh 0/1). roam.visit is the site being visited (-1 at home); roam.win_next_ms is the time to the next window start (-1 unknown). RADIO op 26 cmd u8 = radio_req_roam; settings 21 roam_rssi_dbm (i16 as u16, two's complement), 22 roam_hyst_db, 23 roam_probe, 24 roam_check_s, 25 roam_beacon_s, 26 roam_beacon_len_s, 27 roam_survey, 28 rpt_beacon through op 24.

7. Our repeaters' beacons

The SFR runner (docs/spec/sfr.md: MMDVM SFR today; Standalone SFR's repeater roles are not built) can send the beacon itself: with rpt_beacon on (debug link setting 28, off by default by P7), the runner keys its outbound with idle bursts for roam_beacon_len_s every roam_beacon_s while it carries nothing (sfrRunnerSetBeacon, counted in sfr.tx_beacons in the debug link's sfr section). Traffic always has priority: a beacon due during a call is skipped, and a call that starts during a beacon takes the outbound over. The beacon is standard idle bursts with the repeater's color code, nothing else.

The radio keeps no network time, so its beacon starts at its own phase (the first one when the role starts) and is not in a network's shared window: a radio surveying such a system in the synced mode learns the window from the other repeaters and finds ours only through traffic or a discovery visit; the per-site mode handles it. Aligning it would need a time source: the hotspot host's clock (the MMDVM host knows the network time) could send the phase later. MMDVM duplex hosts send their own beacons (the host's setting), not through this setting.

8. Timings and constants

9. Tests

• Host (tests/host/roamTest.c): the engine alone (averages, weak latch, dwell, back-off, the choice with hysteresis, home preference, lock, abort) and through radio_sm.c against a fake board whose repeaters are frequencies with an RSSI and a color code, driven by the simulated modem (BS idle bursts while a site is "keyed", BS_Dwn_Act on air wakes it): no roam while a call plays, during PTT, in the reply window, with a temporary change, a manual target, hear all or monitor; weak site -> search -> the strongest qualifying site; hysteresis keeps the radio on a site that is not enough worse; return to home within the hysteresis; a candidate with the wrong color code is never chosen; PTT during a search returns to the origin first and transmits there; a failed wake-up on PTT starts a search; wake probing finds a silent site; lock and roam now; the zone flag is stored; CONFIG_DMR_ROAMING=0 builds. Survey (engine): window learning, jitter and moving the schedule, rotation oldest first, the 10 % duty, a weak site moving through a confirm visit within one window, stale data surveyed first then the full search, blocks and own-site traffic, a call on a visited site, the per-site mode and discovery, no survey without beacons. Survey (radio core, fake repeaters with a beacon period, length and phase): defaults and the stored fields, rounds in the window and the shown view, PTT during a visit transmits on the own site without a wake-up, own-site traffic holds the survey off. SFR (tests/host/hotspotSfrTest.c): the runner's idle beacon.
• UI flow (tests/host/uiFlowTest.c): Tools > Roaming toggles the zone flag in the press budget; "Roam now" asks the radio core.
• HIL (tests/hil/test_roaming.py, listed in tests/hil/README.md): radio only: the roam section, the settings round trip, the zone toggle and back, a passive roam now on the current zone (ends on one of its sites, nothing sent), lock and roam now ending it; with --allow-rf --allow-service-stop --roam-rx --roam-tx: the SDR bench runs bench_repeater.py as a second site and the radio, started on a dead site with the wake probe on, wakes it and moves there. A passive variant (the bench repeater carrying a bench call while the radio searches) needs the bench to transmit through its own repeater and is left for later. Beacons (-k beacon, bench_repeater.py --beacon --beacon-len, defaults from --roam-beacon 20 --roam-beacon-len 3): the survey settings round trip (radio only); the bench site as home beaconing, the radio learns the window (roam.win_conf, roam.beacons); the bench site as the other site, the radio surveys it in the window (surv.<i> fresh).

10. Open decisions

1. The roaming zone flag and the thresholds as per-zone fields of the native zone record (when it exists), instead of radio-wide settings plus zone index bits.
2. Zone bits keyed by index break when a CPS reorders zones. A name hash would survive reordering but not renaming.
3. Whether qdmr exposes the open-firmware Roaming bit; if not, a qdmr extension or a radio-side channel editor field.
4. Wake probing in the UI (Settings > Talk & listen) and whether a probe should use the lowest power level.
5. Whether a move should be announced by a beep, a voice prompt or nothing (today a short beep with key beeps on).
6. Talk target matching: by resolved contact ID (today) or also by listen list, for systems where sites carry the same talkgroups on different timeslots.
7. The survey costs about 5.8 KB of flash; a CONFIG_DMR_ROAMING_SURVEY sub-switch could drop it on tight radios.
8. Whether Home should show a survey visit at all (today it doesn't; the debug link does).
9. The "repeater up" from the CACH (slot toggle) is not color-code checked; on a system where a neighbor with another color code shares the frequency, a survey reading could count it.
10. Aligning our repeater's beacon with a network window (section 7) from the hotspot host's time.
11. A PTT after a visit to a silent site, once the 300 ms hold expired, goes through a wake-up on the own site.

View the source of this page