Wireless Intercom and Talkback
Overview
Wireless intercom has run on DECT for two decades. Clear-Com FreeSpeak II and Riedel Bolero both work at 1.9 GHz, with beltpacks roaming between transceivers: each transceiver a small cell, each roam a handoff. It works well, and its limits are structural: coverage reaches wherever transceivers were cabled to, capacity is bounded by the RF channel plan and the base station's beltpack count, and every transceiver needs a cable run back to a base or splitter.
Moving the beltpacks onto the site's private 5G network changes the RF layer without changing the operating model. The production still works in partylines and talk keys, IFB and remote mic kill, answerback and call signal. Underneath, the Onyx RUs already carrying cameras and production data now carry the intercom too, and because those RUs form a single shared cell, the beltpacks do not hand off at all.
This is not theoretical. Clear-Com FreeSpeak Cell is a shipping beltpack (the FSC-BP-C) running over 4G LTE and 5G, with dual SIM so a production can hold a private network as primary and a public carrier as fallback, registering to an Arcadia Central Station or an Eclipse HX matrix through the E-IPA-HX card.
How It Works
Reference architecture
What each component does
Beltpack
The operator's endpoint. Captures mic audio, encodes it, and holds an open full-duplex path. Five to nine assignable talk/listen keys with per-key crosspoint level control. Carries a SIM: dual SIM on FreeSpeak Cell.
Onyx RU
The radio cell. Runs the O-RAN 7.2x low-PHY and provides the air interface the beltpack attaches to. Multiple RUs are deployed for coverage.
FHM
Aggregates all RUs into one logical cell and distributes IEEE 1588v2 PTP and SyncE timing so every RU is frame-aligned. This is what removes handoff.
Onyx Edge · DU / CU
High-PHY, MAC, RLC in the DU; RRC and PDCP in the CU. Because every RU is one cell, there is no handover decision to make.
Onyx Edge · 5G Core
Terminates the PDU session and enforces the QoS flow. Each beltpack's bearer is 5QI 1 GBR, reserved bidirectionally.
Local breakout
Puts intercom audio onto the production LAN on site. Audio never traverses a WAN or a public network.
Intercom matrix
Arcadia Central Station, or Eclipse HX with the E-IPA-HX card. Routes and mixes: for each destination it sums the sources that destination's keys have open, at that destination's levels. Applies IFB ducking, remote mic kill, and call signal.
Plant interfaces
4-wire analogue to wired panels; AES67, Dante, or SMPTE ST 2110-30 to the venue audio network.
The audio path, hop by hop
One operator speaks; another hears them. This is every hop in between.
1
Headset → Beltpack
XLR-4M / XLR-5F analogue
Mic audio captured, gain applied, encoded to a low-delay voice codec
2
Beltpack → Onyx RU
5G NR Uu, uplink
Encoded audio as RTP inside the beltpack's PDU session, on its 5QI 1 GBR bearer. The scheduler grants uplink resource continuously: the path is open whether or not a key is pressed
3
Onyx RU → FHM
eCPRI, O-RAN 7.2x fronthaul
RU completes the low-PHY. FHM aggregates every RU into one logical cell and holds PTP timing
4
FHM → Onyx Edge DU
Aggregated fronthaul
High-PHY, MAC, RLC. No handover decision exists: the beltpack has not crossed a cell boundary because there isn't one
5
DU → CU → UPF
F1 / N3, internal to Onyx Edge
RRC and PDCP at the CU; the UPF terminates the session and polices the QoS flow
6
UPF → Production LAN
N6 local breakout
Audio exits onto the venue LAN, on site
7
LAN → Matrix
IP / RTP
The matrix receives one uplink stream per beltpack
8
Inside the matrix
n/a
Routing and mixing. For each destination, the matrix sums the sources whose keys that destination has open, at that destination's crosspoint levels, then applies IFB ducking, RMK, and call signal
9
Matrix → LAN → UPF
IP / RTP
One personalized downlink mix per beltpack: not a copy of a shared bus
10
UPF → CU → DU → FHM → RU
Reverse of 5→3
Downlink on the same reserved GBR bearer
11
Onyx RU → Beltpack
5G NR Uu, downlink
Beltpack receives its mix
12
Beltpack → Headset
Jitter buffer, decode, D/A
The operator hears the talker
Branches off hop 8. The matrix is also where the wireless domain meets everything else: 4-wire analogue out to wired panels, AES67 / Dante / ST 2110-30 to the venue audio network, and (in a hybrid estate) DECT transceivers still serving FreeSpeak II or Bolero beltpacks alongside the cellular ones.
Where the latency goes
Mouth-to-ear is a chain. The 5G segment is a minority of it, which is the point most often missed.
Headset capture, beltpack encode
5–15 ms
5G uplink: scheduling and transmission
5–15 ms
Fronthaul and Onyx Edge processing
1–3 ms
Matrix routing and mix
2–10 ms
5G downlink
5–15 ms
Beltpack jitter buffer and decode
10–30 ms
Total mouth-to-ear
~30–90 ms typical, < 150 ms maximum
The two terms worth tuning are the jitter buffer and the scheduler. Everything else is largely fixed by the equipment.
Two architectural models
The intercom engine's architecture determines what the 5G network is asked to carry, and the two models differ more than they first appear.
Matrix-based: Clear-Com Arcadia and Eclipse HX, Riedel Artist. Mixing happens centrally, so each beltpack is exactly one uplink stream and one downlink mix. N beltpacks means 2N unicast streams. This maps directly onto 5G unicast bearers and scales linearly.
Decentralized: Green-Go. There is no central matrix and no single point of failure; devices are peers, and each subscribes directly to the channels it listens to, addressing 32 channels plus a program channel and a direct user channel. Architecturally this removes the matrix as a failure domain. But it is multicast, and 5G does not carry IP multicast natively without Multicast-Broadcast Services. Absent MBS, each multicast stream is replicated as unicast to every subscribed listener, and downlink load multiplies with listener count. Resolve it deliberately: an MBS-capable configuration, or a multicast-to-unicast gateway hosted on the Onyx Edge · and size the downlink to match.
What changes relative to DECT
RF layer
1.9 GHz DECT
5G NR: n77, n78, n48
What a cell is
Each transceiver
All RUs together, as one shared cell
Mobility
Handoff between transceivers
No handoff: there is no cell boundary
Extending coverage
Cable another transceiver and splitter
Beltpack works wherever the network already reaches
Capacity ceiling
RF channel plan and base station beltpack count
GBR capacity of the carrier
Infrastructure
Dedicated transceivers, splitters, cabling
Shares RUs with cameras and production data
Spectrum
License-exempt 1.9 GHz
Licensed or CBRS, coordinated
Fallback path
None
Dual SIM to a public carrier
Network Requirements
Per-beltpack bearer, uplink
64–150 kbps depending on codec
Per-beltpack bearer, downlink mix
64–150 kbps
Per-beltpack total, symmetric
~150–300 kbps including overhead
Aggregate, 100-beltpack production
15–30 Mbps symmetric
Aggregate, 300-beltpack production
45–90 Mbps symmetric
Concurrency model
Every registered beltpack, always: full-duplex paths are open continuously
One-way network latency
< 30 ms
Mouth-to-ear latency, end to end
< 100 ms target, < 150 ms maximum
Jitter
< 10 ms
Packet error rate
10-3
Availability during show window
99.999%
Primary QoS
5QI 1: GBR, conversational voice, 100 ms PDB, priority 20
Priority / emergency paths
5QI 65: GBR, 75 ms PDB, priority 7
Signaling QoS
5QI 69: non-GBR, 60 ms PDB, priority 5
TDD pattern
Symmetric, e.g. DDSUU; avoid heavily uplink-weighted patterns on the intercom carrier
Bands
Any Onyx-supported band; may share the production carrier
The concurrency row is the one that catches people out. Push-to-talk is sized for concurrent talkers: typically 5–10% of the fleet. Full-duplex partyline is sized for every beltpack, continuously. A 200-beltpack production is 200 open bidirectional GBR paths, not twelve.
Deploying It
The architecture above is the what. This is the sequence that gets it running.
Devices & Ecosystem
Clear-Com
FreeSpeak Cell: the FSC-BP-C beltpack over 4G LTE and 5G, dual SIM for private-primary and public-fallback. The most direct fit for this architecture today.
Arcadia Central Station and Eclipse HX with the E-IPA-HX card: the matrix the beltpacks register to.
FreeSpeak II (1.9 / 2.4 GHz) and FreeSpeak Edge (5 GHz): the DECT and 5 GHz estate, supported alongside cellular beltpacks in hybrid deployments.
Agent-IC and Station-IC: IP intercom clients for remote and occasional users, over the same bearer.
LQ series and HelixNet: IP linking and digital partyline for fixed positions and site-to-site.
Green-Go (Riedel)
Decentralized IP intercom: no central matrix, no single point of failure. Beltpacks address 32 channels plus a program channel and a direct user channel.
PoE-powered wired stations (802.3af) for fixed positions.
Dante Interface X: converts up to eight Dante or AES67 streams to and from the Green-Go network.
Riedel
Bolero: 1.9 GHz wireless beltpack, deployable integrated with an Artist matrix, in standalone link mode, or in Standalone 2110 (AES67) mode building an ST 2110 network with no matrix required.
Bolero Mini: the compact beltpack introduced at IBC 2025, with 5G filtering to resist desensitization from high-power cellular transmitters.
Also in this space
RTS (Bosch) and Pliant Technologies offer wireless intercom families with IP matrix integration; Unity Intercom and similar app-based systems run as IP clients over the same bearer. Where a production standardises on interfaces rather than vendors, the ones that matter are AES67, Dante, SMPTE ST 2110-30, 4-wire analogue, and SIP.
Design Considerations
Full-duplex partyline is not push-to-talk, and the difference is the whole capacity model. Industrial PTT is sized for concurrent talkers and uses floor control to arbitrate one speaker per group, see Connected Worker Push-to-Talk on 5G. Broadcast intercom holds an open bidirectional path per beltpack, continuously. Size GBR for the full registered fleet.
Resolve multicast before rehearsal, not during it. If the intercom engine is decentralized, the multicast question above is the single largest risk to the deployment, and it does not surface until enough listeners subscribe to the same channel.
Plan DECT and 5G coexistence in hybrid deployments. FreeSpeak II and Bolero at 1.9 GHz have no co-channel conflict with n77, n78, or n48. The risk is receiver desensitization: a high-power RU near a DECT beltpack can degrade its front end, which is why Bolero Mini ships with added 5G filtering. Plan physical separation between RUs and DECT transceivers, and validate the DECT system with 5G at full load rather than in isolation.
Decide the dual-SIM fallback policy before the show. Public fallback helps a beltpack that strays outside private coverage. It also silently moves production audio onto a congested public network with no QoS guarantee, during exactly the event that is congesting it. Decide whether it is permitted, and alert when a beltpack takes it.
Capacity claims are network claims. "Hundreds of beltpacks" is stated by vendors as network-dependent, and that dependency is the design work here: GBR capacity of the carrier, the TDD split, and the modulation available at the worst crew position.
Intercom does not go on the guest network. Its own subscriber group, its own guaranteed bearers, isolated from crew data and public access.
Battery logistics will cause more show faults than the network will. Plan charging, hot-swap, and spares as carefully as the RF, and monitor fleet battery state centrally.
Validation Checklist
Next steps
Size it. Take these requirements through Plan Your Network: capacity, coverage, architecture, and a preliminary bill of materials.
Compare it. Network Requirements at a Glance puts every workflow side by side, for sites carrying more than one.
Talk to us. For a validated design and a quote, contact GXC.
Related reading
Connected Worker Push-to-Talk on 5G: the push-to-talk model, and how it differs
Was this helpful?