The difference between “it works” and “it works under load” comes down to a single interface: Rx in EPS, N5 in the 5G system. It is the only interface through which the voice domain can tell the transport domain that a call is in progress. Without it, the radio network knows nothing about the call – and treats it accordingly. Speech becomes another best-effort IP flow, competing with every video stream in the same cell, at the exact moment most of your subscribers are trying to talk.
Rx and N5 are not a QoS optimisation you add later
They are the precondition for speech being treated any differently from background traffic. If the interface is missing, disabled, or sitting on a vendor roadmap, you do not have VoLTE. You have SIP on a mobile data connection – and a PoC that will not survive the first busy hour.
Two Paths, Two Jobs
After attach, a VoLTE device establishes a default bearer to the IMS APN. That bearer carries SIP signalling only and typically uses QCI 5: non-GBR, priority level 1, packet delay budget 100 ms, packet error loss rate 10−6 (TS 23.203, table 6.1.7). High priority, very low loss tolerance, but no guaranteed bitrate – signalling is bursty and does not need one.
The speech path does not run over that bearer. It needs QCI 1: GBR, priority level 2, PDB 100 ms, PELR 10−2. A guaranteed bitrate, in exchange for a comparatively high tolerable packet loss rate – RTP carrying AMR or EVS handles occasional loss far better than it handles delay. The 5G system applies the same model under different names: 5QI 5 for IMS signalling, 5QI 1 for conversational voice (TS 23.501, table 5.7.4-1). Instead of separate bearers there are QoS flows within the same PDU session, identified by a QFI and mapped onto data radio bearers by the SDAP layer.
| Signalling | Conversational voice | |
|---|---|---|
| EPS / 5GS | QCI 5 / 5QI 5 | QCI 1 / 5QI 1 |
| Resource type | Non-GBR | GBR |
| Priority | 1 | 2 |
| Packet delay budget | 100 ms | 100 ms |
| Packet error loss rate | 10−6 | 10−2 |
| Who establishes it | Default bearer / IMS PDU session | The network, triggered by the AF |
What matters in both generations: the device does not establish that second path itself. It cannot. Establishment is network-initiated – and the trigger sits in the IMS. The P-CSCF, acting as Application Function, is the only SIP element that can look at an SDP offer and tell the policy layer: a call is starting, these are the flows, this is the bitrate, reserve the radio for speech.
The Procedure in EPS
Towards the policy layer, the P-CSCF acts as an Application Function. When an SDP offer passes through, it translates the media description – codec, bandwidth, IP flows, ports – into a media component description and sends it as an AA-Request over Rx to the PCRF (TS 29.214).
The PCRF authorises the requested resources against the subscriber profile, derives PCC rules including QoS parameters, and pushes them over Gx to the PCEF in the P-GW (TS 29.212, with mapping rules in TS 29.213). The P-GW issues a Create Bearer Request that travels via S-GW and MME down to E-RAB setup in the eNodeB and reaches the device as an RRC Connection Reconfiguration. Only then does a dedicated bearer with QCI 1, GBR and MBR exist.
The confirmation travels back the same way. The PCRF informs the P-CSCF over Rx that resources were successfully allocated. That is exactly what the signalling is waiting for – and, in a standards-compliant IMS, exactly what preconditions are gated on before the phone is allowed to ring.
The Procedure in 5GC
Architecturally different, functionally identical. Acting as an AF, the P-CSCF invokes Npcf_PolicyAuthorization_Create over N5 – HTTP/2 instead of Diameter (TS 29.514). The PCF evaluates the request against policy and updates the session management policy in the SMF over N7 (TS 23.503). The SMF configures the UPF over N4 and initiates a PDU Session Modification over N1/N2, resulting in a dedicated QoS flow with 5QI 1. SDAP maps that QFI onto a data radio bearer. Same job, new names, same dependency on the AF.
What Happens Without Rx or N5
The call does not fail. That is what makes it treacherous. It goes through – but over the default IMS bearer, non-GBR, competing for the same radio resources as every video stream in the cell. The RAN is not malicious. It is simply uninformed.
No scheduler priority
The eNodeB or gNB prioritises based on the QoS parameters of the bearer. Without a dedicated bearer, the scheduler sees no reason to move RTP packets ahead of anything else. At low cell load nobody notices. At high load speech quality collapses – precisely when most subscribers are making calls.
No admission control
GBR bearers are checked against available radio resources. If the cell is full, the bearer is rejected and the call is cleanly refused. Without that mechanism every call is accepted and all calls in the cell degrade together. You do not get a busy tone. You get a network that sounds busy.
Preconditions break
GSMA IR.92 relies on RFC 3312: the device signals in SDP that resource reservation is a prerequisite for session establishment. Call setup pauses at 183 Session Progress until the bearer is in place. Without Rx that confirmation never arrives and the session times out. The usual workaround is to disable preconditions. Then the phone rings before any media path exists, and the beginning of the conversation is clipped.
No bearer-loss notification
If the dedicated bearer is lost mid-call, the PCRF reports this over Rx to the P-CSCF, which can terminate the session in a controlled way. Without that return path, sessions remain up in the IMS while the subscriber hears silence. In the statistics they appear as successful calls.
No charging correlation
Rx carries the binding between the IMS session and the IP-CAN session – the access-network charging identifier on one side, the IMS charging identifier on the other. Without that correlation, CDRs from the IMS and the packet core cannot be reconciled. For billing that may be tolerable; for audits and disputes it is not.
No ARP for emergency calls
Emergency calls need an allocation and retention priority with pre-emption capability so they can displace existing bearers in a congested cell. That priority is set through the PCC rule. No PCC, no priority – the emergency call competes with everything else. That is not an edge case. It is a public-safety requirement hiding inside a policy interface.
The 5G Special Case: No N5, No EPS Fallback
Operators running 5G SA with voice via EPS fallback should pay particular attention to one point: fallback is not triggered by someone dialling a number. It is triggered by the attempt to establish a QoS flow with 5QI 1. The NG-RAN recognises that it cannot serve conversational voice and initiates handover or redirection to E-UTRAN (TS 23.502, clause 4.13.6.1). With N26 the transition is a prepared inter-system handover; without it, the UE is redirected and has to find an LTE cell first – the difference shows up directly in post-dial delay.
The trigger for that attempt is the P-CSCF’s N5 request. Without N5 there is no request for a 5QI 1 flow, and therefore no fallback. The session stays in the 5G network without anyone reserving dedicated resources for speech. VoNR never starts. EPS fallback never starts. The subscriber is on 5G, in a call, on best effort.
Native VoNR has the same dependency
When NG-RAN can serve conversational voice, N5 is still how 5QI 1 appears. EPS fallback and VoNR are two outcomes of the same AF request. Disable N5 and you disable both.
The PoC That Proves Nothing
This is the procurement failure mode we see most often, and it is almost never recognised as one. A proof of concept in which speech runs over the default bearer demonstrates that signalling works. It says nothing about whether the service will hold up in production. The lab is an idle cell. Production is Friday at 17:45, a stadium emptying, everyone calling a ride at once.
If the policy interface only becomes available in a future packet-core release, no dependable voice service is possible until then – and neither is a proof of concept that proves anything. Roadmap is not a deliverable. An IOT report from a loaded cell is.
What This Means for Procurement
Operators sourcing IMS and packet core separately – now the normal case – should settle three questions early, before the commercial phase begins:
- Is Rx or N5 included in the packet core deliverable, and is it enabled? Not every product variant ships with the policy layer active. “We have a PCRF” is not the same as “Rx is up, mapped, and used for every IMS session.”
- How is interoperability evidenced? Rx is well specified, but the interpretation of the media component description and the QoS mapping differ between implementations. A reference setup or an IOT report is worth more than a line in a datasheet.
- What is the roadmap? If the interface only becomes available in a future release, plan the voice launch for that date – not for the day SIP first registers. And do not sign a PoC that cannot exercise it.
How volte.io Puts Rx and N5 to Work
Policy control is one of those areas where the standards are unambiguous and the deployments are not. Our platform treats the AF session as part of call setup, not as an optional extra:
P-CSCF as a real AF
Diameter Rx toward a PCRF, HTTP/2 N5 toward a PCF. The same P-CSCF that terminates preconditions is the element that requests the bearer – and it waits for the confirmation, not for the AAR to leave the building.
PCRF in the platform, or yours
Run policy with us, or point us at the PCRF/PCF you already operate. The mapping of SDP to media component descriptions is the interop surface that actually matters, and it is one we exercise continuously rather than documenting once.
Failure is a call result, not silence
If GBR cannot be reserved, the call is refused. If the dedicated bearer is lost, the IMS session is torn down. Your KPIs then describe the network you actually have, not a graveyard of “successful” calls with no audio.
Verified under load, not on an idle cell
Continuous verification with real devices is how we catch the class of failure this article is about: the call that works in the lab and collapses when the scheduler has a choice.
Rx in EPS. N5 in 5G. Not optional. Not a nice-to-have.
Real networks carry more than one perfect call. Dedicated bearers, QoS flows, preconditions, charging correlation, emergency ARP, and EPS fallback all hang off the same AF request. Ship that interface with the IMS – or do not claim the service.
References: TS 23.203 (PCC architecture, EPS), TS 29.212 (Gx), TS 29.213 (PCC signalling flows and QoS mapping), TS 29.214 (Rx), TS 23.501 (5G system architecture, QoS model), TS 23.502 (procedures, including EPS fallback), TS 23.503 (PCC framework for 5GS), TS 29.514 (Npcf_PolicyAuthorization), TS 23.228 (IMS), TS 26.114 (IMS multimedia telephony), GSMA IR.92, RFC 3312.