Home › Topics › VoIP & Call-Quality Monitoring › Phones That Won't Register

Phones That Won't Register: Provisioning, DHCP Options, and Dependencies


Monday, ten past eight. Half the phones on the third floor say "Registering..." and the other half show nothing at all. The PBX is up. The switches are up. Nothing in the phone system changed over the weekend, and yet the phones have opinions.


A desk phone does not simply switch on. It climbs a ladder of dependencies: power, link, VLAN, address, options, time, provisioning, registration. It can fall off at any rung. Each rung belongs to a different system, and often to a different team. "The phone won't register" describes the top rung. The fault is usually several rungs down, in a DHCP scope or a certificate that has nothing to do with telephony.


This guide maps that ladder as a dependency chain. It shows what to monitor at each link. It teaches you to read the phone's own signals. It treats mass deregistration as the incident signal it is. And it gives you two troubleshooting routines: one for a single stubborn phone and one for a whole site. It ends with a boot-chain checklist you can turn into monitoring this week.


In this guide

  • The boot chain as a dependency map
  • Monitoring each link in the chain
  • Reading the phone's own signals
  • Mass deregistration as an incident signal
  • Troubleshooting a single phone
  • Troubleshooting a whole site
  • The boot-chain checklist
  • Where to go next

The boot chain as a dependency map


Here is what happens between plugging in a phone and hearing a dial tone. The details vary by phone and phone system, but the order almost never does, and each step depends on every step before it.


  • 1. Power. Most desk phones draw power from the switch port over PoE (Power over Ethernet). The switch must have budget left and must agree with the phone about how much it may draw. No power, no phone, and nothing further in the chain even starts.
  • 2. Link. The phone and the switch negotiate an Ethernet link. A cable fault or a duplex mismatch here produces a phone that is powered but never gets an address, or gets one and then sounds terrible.
  • 3. VLAN assignment. The phone needs to land in the voice VLAN, the separate virtual network reserved for phones. Two mechanisms are common. With discovery, the switch tells the phone which VLAN to use through LLDP-MED or a vendor equivalent. LLDP-MED is a standard extension of the Link Layer Discovery Protocol for media devices. With the DHCP method, the phone takes a temporary lease in the data VLAN. It reads a DHCP option that names the voice VLAN, and restarts its network stack there. If neither works, the phone stays in the data VLAN, where it may register but will get none of the voice treatment.
  • 4. DHCP lease and options. In the voice VLAN the phone asks for an address. The scope must have free leases. The relay on that VLAN's router must forward the request to the DHCP server. And the reply must carry the options phones need. These are gateway, DNS, an NTP server for time, and a pointer to the provisioning server. That pointer is commonly the standard boot-server-name option (option 66), or a vendor-specific option keyed to the phone's vendor class.
  • 5. Time and names. The phone sets its clock from NTP and resolves the provisioning server and the phone system by name, often through DNS SRV records. Time sounds optional; it is not. A phone whose clock is decades wrong will reject every certificate it meets as not yet valid.
  • 6. Provisioning. The phone contacts the provisioning server over TFTP, HTTP, or HTTPS and downloads its configuration. This is typically a file named after the phone's MAC address. The phone also downloads firmware if the server says its current version is wrong. A firmware update means another reboot, which restarts the chain from the top.
  • 7. Registration. With credentials from its configuration, the phone sends a SIP REGISTER to the PBX or hosted service. It usually gets challenged for authentication, answers, and receives an OK with an expiry. It must repeat this before the expiry runs out, forever.
  • 8. Ready. The extension lights up. Everything from here on is call quality, which is the rest of this topic.

LinkDepends onSymptom when it failsWhere to look
PowerSwitch PoE budget, port PoE state, cablePhone dark, or reboots under loadSwitch PoE status per port and per switch
LinkCable, port negotiationPhone lit, "network error," or terrible audioPort status, speed, duplex, error counters
VLANLLDP-MED or the DHCP VLAN option; port configurationAddress from the wrong subnet; registers but no QoSSwitch neighbor table; port voice-VLAN setting
DHCPScope leases, relay, options"Obtaining address" forever; wrong provisioning serverDHCP server leases and scope; relay configuration
Time and namesNTP server, DNS server, SRV recordsCertificate errors; "server not found"Phone's clock; DNS answers from the phone's resolver
ProvisioningServer up, file present, certificate validStuck "provisioning"; old configuration; boot loop after firmwareProvisioning server logs and certificate
RegistrationPBX or hosted service up, credentials, TLS certificate, path open"Registering...", "no service"PBX registration log; firewall for remote phones

Monitoring each link in the chain


The chain suggests its own monitoring: one health check per link, plus one outcome metric at the top that tells you whether the whole thing worked.


Registration state: the outcome metric

The PBX knows exactly which phones are registered, and it is the single most valuable number in voice monitoring. Collect the count of registered phones, per site, every minute, and the last-registration time per phone. Compare the count with the number of phones you expect at that site. A floor with sixty phones and forty-one registered has a problem even if nobody has called yet. With a hosted phone system you may not get a registrar view. So use the provider's portal where it exists and keep one physical phone per site whose registration you check directly.


Provisioning server availability

Check that the provisioning server answers, from the voice VLAN if you can. Make an HTTP or HTTPS request for a small known file, or a TFTP fetch of a test file, on a schedule. Check its certificate expiry weeks ahead. An expired certificate on an HTTPS provisioning server means trouble for every phone that reboots from that day on. Each comes up with stale configuration, or does not come up at all.


DHCP scope health

Watch the free leases in each voice scope, the DHCP service itself, and the relay path. DHCP Basics for Monitoring explains how leases, scopes, and relays fit together. A voice scope that silently fills is a slow-motion outage. Existing phones renew happily until the day they cannot, then drop out one by one as leases expire. That looks nothing like a DHCP problem until you check.


The rest of the chain

PoE budget consumed versus available per switch, and any port in a power-denied state. The NTP server answering, and the DNS names phones use resolving from the resolver phones use. The PBX host itself: service state, CPU, memory, and the expiry of the certificate it presents to phones registering over TLS. None of these are voice checks; they are ordinary infrastructure checks aimed at the things phones depend on.


Tip: model the chain as dependencies in your monitoring, so that a dead DHCP server raises one alert about DHCP rather than a hundred alerts about phones. The registration count still drops, and it should still show on the dashboard. But the page that wakes you up should name the rung that actually broke.


Reading the phone's own signals


The phone narrates its own boot, if you know where to look. Its screen shows a status word for each stage: "Obtaining address," "Provisioning," "Registering," "No service." Its web page usually has a status section with the VLAN it landed in, its address and DHCP server, and the options it received. That section usually has the result of its last provisioning attempt, and the registration server and last error. And most phones can send syslog to a collector, which turns a hundred phones into a readable stream. A healthy boot reads like this:


ph-001a2b3c4d5e  LLDP-MED: voice VLAN 120 learned from switch port
ph-001a2b3c4d5e  DHCP: lease 10.120.4.57/22 from 10.120.0.1; option 66 = prov.example.internal
ph-001a2b3c4d5e  NTP: time set from 10.120.0.5
ph-001a2b3c4d5e  PROV: GET https://prov.example.internal/cfg/001a2b3c4d5e.cfg -> 200 OK
ph-001a2b3c4d5e  SIP: REGISTER sip:pbx.example.internal -> 401 -> auth -> 200 OK (expires 3600)

An unhealthy one tells you which rung failed, and often why:


ph-001a2b3c4d5e  LLDP-MED: no voice VLAN advertised; staying in native VLAN
ph-001a2b3c4d5e  DHCP: lease 10.10.8.201/22 from 10.10.0.1; no option 66
ph-001a2b3c4d5e  PROV: GET https://prov.example.internal/cfg/001a2b3c4d5e.cfg -> TLS error: certificate not yet valid
ph-001a2b3c4d5e  SIP: REGISTER sip:pbx.example.internal -> 403 Forbidden

Read the second log from the top, not the bottom. The phone never learned its voice VLAN, so it took a data-VLAN lease with no provisioning pointer. So it used a stale server name from memory. It hit a certificate error because its clock was never set (no NTP option either). Finally, it registered with old credentials that the PBX rejected. Four symptoms, one cause: the switch port lost its voice-VLAN configuration. This is why the chain matters. The bottom line of a log is the least useful one.


Mass deregistration as an incident signal


A sharp drop in the registered count is one of the clearest incident signals in IT. And its shape tells you where to look before anyone picks up a phone. Three families of cause cover nearly every case.


  • The PBX or registrar. A service restart, an overloaded registrar, a license limit, a bad configuration push, or the expiry of the certificate the PBX presents to phones over TLS. Signature: every phone everywhere, at the same moment, or every TLS phone at once.
  • The network. A switch rebooting takes its phones with it. A WAN outage takes a whole site. A DHCP outage or an exhausted scope produces a slow wave as leases expire over hours. A firewall change or session limit hits remote phones only. Signature: bounded by a switch, a site, a scope, or a path.
  • Certificates and time. An expired provisioning or registration certificate, or an NTP failure that lets clocks drift until certificates stop validating. Signature: only the phones that rebooted since the failure are affected, so the count trickles down over days. In that case, the phones that stayed up look fine.

Shape of the dropMost likely causeFirst check
Everything, instantlyPBX or registrar service, or a configuration pushPBX service state and logs
Every TLS phone, instantlyRegistration certificate expired or replacedCertificate dates on the PBX
One siteWAN link, site DHCP relay, site core switchWAN state; DHCP from that site
One switch or one floorSwitch reboot, power supply, PoE budget, VLAN changeSwitch uptime and PoE status
Slow wave over hoursDHCP server down or voice scope fullDHCP service and free leases
Trickle over days, rebooted phones onlyProvisioning certificate, NTP, or a bad configuration fileProvisioning server logs; phone clocks
One phone model onlyA firmware push or a model-specific configuration templateProvisioning server change history

The alert is straightforward: registered count for a site falling by more than a set share in a few minutes pages someone. And a count that stays below the expected number for an hour opens a ticket. Suppress the page when the site's WAN link is already known to be down, because then the alert is telling you something you know.


Troubleshooting a single phone


One phone will not register while its neighbors are fine. Walk the chain from the bottom, and stop at the first rung that fails:


  • 1. Read the screen. The status word tells you roughly where it stopped.
  • 2. Power. Is the phone lit? Does the switch show the port delivering power, and how much? A port in a denied or faulted power state is the answer already.
  • 3. Link. Is the port up, at the expected speed and duplex, with no growing error counters?
  • 4. VLAN. Does the switch's neighbor table show the phone, and is the port configured with the voice VLAN? Which VLAN did the phone actually land in?
  • 5. Address. Does the phone hold a lease in the voice scope, and does the DHCP server show it? Did the lease carry the provisioning and NTP options?
  • 6. Time and names. Is the clock right? Do the provisioning and PBX names resolve from the phone's resolver?
  • 7. Provisioning. Do the server logs show the phone's request, and with what result? Does a configuration file exist for its MAC address?
  • 8. Registration. Does the PBX log show the REGISTER arriving, and how did it answer? An authentication failure means credentials; "forbidden" often means the phone was locked out after repeated bad attempts. No request at all means the phone never got that far, or a firewall is in the way.
  • 9. Swap. A known-good phone on this port, and this phone on a known-good port. Whichever combination fails tells you whether the fault travels with the phone or stays with the port.

Two habits make this fast. Keep a copy of the syslog stream for each phone, so that steps three through eight are a single search by MAC address. And write down which rung it was. After a few months you will know that your building's phone problems are, say, mostly cables and mostly on the second floor. That is a maintenance plan.


Troubleshooting a whole site


When every phone at a site is affected, the single-phone routine is the wrong tool; the fault is in something they share. Work through the shared dependencies from the outside in:


  • The site's WAN link, if the PBX or the hosted service is on the other side of it. If the link is down, everything else is a symptom.
  • DHCP for the site. Is the relay forwarding? Is the server reachable from the site? Has the voice scope run dry? Phones that still have leases will look fine for a while, which makes this one sneaky.
  • Power. A failed power supply in a redundant switch can halve the PoE budget without taking the switch down. And the phones the switch chose to shed are the ones calling you.
  • The voice VLAN and its uplinks. A trunk link that stopped carrying the voice VLAN after a change leaves phones with link but no path to anything.
  • Time and names from the site. Do the site's phones reach an NTP server and a DNS server that knows the internal names?
  • Provisioning from the site. Fetch the test file from a machine on the site's voice VLAN. A server that answers from head office and not from the branch is a routing or firewall problem.
  • What changed? Firmware pushed, certificate renewed, scope edited, switch upgraded, firewall rule added. Sites rarely break spontaneously.

The scope question settles it quickly. If every site is affected, the cause is central — the PBX, a certificate, the provisioning server, central DHCP or DNS. If one site is affected, the cause is that site's network or its path to the center. Rarely both, and when it is both, start central.


The boot-chain checklist


LinkHealth checkAlert when
Registration (outcome)Registered count per site versus expected; per-phone last registrationCount drops sharply in minutes; stays below expected for an hour
PoE powerConsumed versus available per switch; per-port power stateHeadroom nearly gone; any port power-denied
LinkPhone ports up; error countersErrors growing on a phone port; repeated flaps
VLANVoice VLAN present on phone ports and uplinks; LLDP-MED advertisedAfter any switch change, by check; on phones landing in the data scope
DHCPService up; free leases per voice scope; relay reachableFree leases below a margin; service down; no leases issued in business hours
TimeNTP server answering from each siteNo answer, or the phones' clocks drifting
NamesProvisioning and PBX names resolve from the phones' resolverResolution fails or returns unexpected addresses
ProvisioningTest-file fetch from each site; certificate expiryFetch fails twice; certificate within a few weeks of expiry
PBX or registrarService, CPU, memory, disk; registration certificate expiryService down; certificate within a few weeks of expiry
Path for remote phonesFirewall session count; WAN state per siteSession table near its limit; link down (and suppress the rest)

Where to go next


A phone that will not register is the last rung of a ladder. And the ladder is the whole lesson: power, link, VLAN, DHCP and its options, time and names, provisioning, registration. Monitor a link at each rung and the registered count at the top. Read the phone's log from the first failure rather than the last error. And treat a drop in the registered count as an incident whose shape names the cause. One stubborn phone gets the bottom-up walk; a whole site gets the shared-dependency walk.


Once the phones are up, the remaining questions are about how the calls sound. What Makes VoIP Different is the foundation. The article Building a Call-Quality Monitoring Program makes quality routine. The rest of the guides live in VoIP & Call-Quality Monitoring.


Put this into practice with SysPeep Monitor

Almost every rung of the boot chain is ordinary infrastructure. That includes switches and their PoE budgets, the DHCP server, the DNS and NTP servers, the provisioning host, and the PBX machine. SysPeep Monitor watches all of them over SNMP, WMI, and similar collectors. Its dependency rules let you model the chain so that a dead DHCP server raises one alert instead of a floor full of phone alerts. The free trial installs on a Windows machine in minutes.

Download Free TrialLearn About SysPeep