4G/5G Backup Internet Line: CGNAT, Antennas and Signal Quality

How are VPN and inbound access affected when a mobile backup line takes over? Assess which jobs the line is suitable for through CGNAT, APN, signal quality and failover testing.
Technical summary
Technology
LTE, 5G NR, CGNAT (RFC 6598), APN, IPsec NAT-T, MIMO antenna
Topic
Designing and operating a cellular connection as a backup WAN
Level
Intermediate
Prerequisites
  • Cellular modem or router with a modem
  • Existing primary line and failover configuration
In this article8 sections

The wired line dropped, the 4G backup took over, users are getting out to the internet; but the tunnel to headquarters won’t come up. This is the picture we see most often on cellular backup lines. At locations where a second wired line can’t be brought in or is expensive, 4G/5G is the fastest backup WAN option to commission. But it isn’t a wireless copy of the wired line: it has its own constraints on addressing, inbound connections, performance fluctuation and data quota. Below we explain how to handle these constraints in design and operation. For general failover and load-balancing logic, see the failover with a dual internet line article.

The role and limits of a cellular backup line

What’s expected of a cellular line is that business-critical traffic continues when the primary line is cut, not that it carries the primary line’s entire load. This distinction determines the rest of the enterprise network design.

Topic Wired line Cellular line
Capacity Fixed by contract Varies with cell load and signal
Latency Stable Can be higher and fluctuating
Public IPv4 address Usually available Mostly behind CGNAT
Pricing Fixed Can be quota- or usage-based
Physical independence May share the same duct or exchange Independent of wired infrastructure

The last row is the real strength of a cellular line. If two wired lines come from the same building entry or the same duct, a single excavator cuts both; a cellular line doesn’t share this common failure point. On the other hand, in a regional power outage base stations can be affected too, and the carrier’s base station power backup time isn’t under your control. A cellular line is independent. It isn’t unlimited.

CGNAT: the effect on inbound connections and VPN

Mobile carriers mostly share IPv4 addresses behind Carrier-Grade NAT (CGNAT). The router’s cellular interface gets an address either from the RFC 1918 private range or from the 100.64.0.0/10 shared address range reserved for this purpose by RFC 6598; on the way out to the internet, the carrier’s NAT device translates this address into a public address shared with many subscribers. The consequences in the field:

  • There are no inbound connections: Port forwarding can’t be done; a server or recorder at the branch can’t be reached directly from the internet.
  • In IPsec, the branch must be the initiator: Headquarters can’t know the branch’s address in advance, and carrier NAT doesn’t forward an incoming IKE request to the branch; headquarters can’t initiate the tunnel. The headquarters side is configured to accept dynamic peers, and the branch identity is specified with an FQDN or user-defined identity instead of an IP address. If the branch’s peer address is written statically at headquarters, the tunnel never comes up on the backup line; this is the most common cause of the picture in the introduction.
  • NAT-T is needed when there’s NAT in between: When IKE detects NAT in the path, it encapsulates traffic on UDP 4500. UDP 500 and 4500 must be open on the headquarters firewall.
  • The NAT mapping must be kept alive: The mapping in the carrier NAT table is deleted if there’s no traffic for a certain time. NAT keepalive sends small packets while there’s no traffic, to delay this mapping from timing out. DPD, on the other hand, checks whether the far end responds and, if not, triggers re-establishment of the tunnel; it doesn’t prevent the drop, it makes it noticed. An IPsec session initiated outbound by the device behind CGNAT and supporting NAT traversal can be established; connections initiated toward this device from outside don’t reach it without a public IPv4 address or a private APN.
  • The public IP changes: Access rules at headquarters based on source IP, or IP allowlists at third-party services, don’t work over the backup line. Learn this on installation day, not on outage day.

The details of tunnel design, especially route-based tunnels and dynamic routing over the tunnel, are in the site-to-site IPsec VPN design article. Some mobile carriers also assign IPv6 addresses; but whether inbound connections over IPv6 are allowed depends on carrier policy, so verify it in advance.

APN, static IP and private APN options

The APN (Access Point Name) determines which network the cellular device connects to in the carrier core. For an enterprise backup line there are three options:

  1. Public internet APN: The default; with most carriers it’s behind CGNAT, so verify the situation with the carrier. In setups where the branch initiates the tunnel, it’s often sufficient.
  2. Static public IP: Many carriers offer a static public IPv4 address for enterprise lines with a separate APN or an add-on service. If inbound connections or IP-based allowlists are needed, this solves the problem.
  3. Private APN: Carries cellular traffic, without going out to the internet, to the carrier’s enterprise network service or to a private circuit connected to your headquarters. It’s worth considering if there’s cellular use at many locations and traffic should never go out to the internet.

Which option is offered under which conditions varies by carrier. At the quotation stage, ask in writing about the addressing type, the inbound traffic policy, and whether the service is defined per SIM card or per account. A verbal answer doesn’t hold up on outage day.

Signal quality: interpreting RSRP, RSRQ and SINR

The bars on a phone screen aren’t enough for an installation decision. LTE modems report three basic values:

  • RSRP (Reference Signal Received Power): The received power of the reference signal, in dBm. The basic reporting range defined in 3GPP TS 36.133 is from -140 to -44 dBm; the closer the value gets to zero, the stronger the signal.
  • RSRQ (Reference Signal Received Quality): The ratio of signal power to total received power, in dB; it’s affected by interference and cell load.
  • SINR: The ratio of signal to interference and noise. It’s the value that most directly determines the achievable modulation, and therefore the speed.

In 5G NR, similar measurements are reported as SS-RSRP, SS-RSRQ and SS-SINR. The table below is an approximate interpretation framework commonly used in the field; it isn’t a standard classification, and vendors may use different thresholds.

Assessment RSRP (dBm) RSRQ (dB) SINR (dB)
Very good -80 and above -10 and above 20 and above
Good -80 to -90 -10 to -15 13 to 20
Intermediate -90 to -100 -15 to -20 0 to 13
Weak Below -100 Below -20 Below 0

If strong RSRP and low SINR appear together, the problem isn’t signal strength but interference. Rather than raising the antenna, pointing a directional antenna at a single base station can be more effective. Record the values not at a single moment but at different times of day, including business hours; cell load changes during the day. A measurement taken on Sunday morning doesn’t describe Monday morning.

Router and external antenna placement

The cellular router is often put in the server room, inside a metal cabinet. In terms of signal, this is usually the worst spot in the building. Apply these rules for placement:

  • If possible, place the modem or antenna on the exterior wall or near a window facing the base station. Avoid basements and interior rooms.
  • If you use an external antenna, choose one that matches the MIMO configuration the modem supports (2×2 or 4×4); a single-element antenna eliminates the MIMO gain.
  • Loss increases as coaxial cable gets longer, and this loss grows as frequency rises. Rather than running a long cable, placing the modem near the antenna and carrying the data over Ethernet (a PoE-powered outdoor unit) gives a better result in most cases.
  • Don’t skip lightning protection and grounding for antennas mounted on the roof.
  • Measure the coverage of more than one carrier at the same spot. Choosing the same carrier as the primary line can cause both lines to drop together during a failure in the carrier core.

The modem’s diagnostic interface or CLI shows the connected cell ID and band. Forcing it onto a better band with band locking can help during testing; but if it’s left in place permanently, the connection can’t be established at all when that band becomes unavailable. Remove the lock once testing is over.

Data quota and traffic restriction

If the backup line is on a quota-based plan, all traffic flowing over the cellular line when the primary line is cut can use up the quota within hours. A remote backup job that starts at night consumes the quota by morning. Define in advance the policy to apply on failover to the backup line:

  • Remote backup, replication, operating system updates and video streaming are blocked or deferred on the cellular line.
  • Business-critical applications (ERP, voice, payment systems) are prioritized.
  • A total bandwidth limit is applied on the cellular line; a single user can’t use up the quota.

Health checks consume data too. Worked example: an ICMP echo packet carrying 56 bytes of data is 84 bytes at the IP layer; the request and reply together come to 168 bytes. One probe per second, 86,400 probes a day, consumes about 14.5 MB a day and about 435 MB in 30 days. Because the carrier’s billing layer may differ, the real value may come out slightly higher; if probes are sent to more than one target, the multiplier increases accordingly.

Periodic testing and installation checklist

The most common failure of a backup line is discovering it doesn’t work at the moment it’s needed. Continuous monitoring and planned testing should go together: the cellular interface status, signal values and reachability of a target over the backup line are added to the monitoring system; at least once a month, a full failover test is run by physically disconnecting the primary line. In the test, don’t just check that the connection comes up; verify that tunnels are re-established, DNS resolution works and critical applications open. An untested backup isn’t a backup. How to measure latency and loss is described in the WAN SLA measurement article.

  • CGNAT status has been determined; if inbound connections are needed, a static IP or private APN has been requested.
  • The IPsec tunnel is initiated by the branch; NAT-T, DPD and keepalive have been configured.
  • RSRP, RSRQ and SINR values have been measured and recorded at different times.
  • The router and antenna location were chosen based on measurement; the MIMO antenna and cable length are appropriate.
  • A different carrier from the primary line was chosen.
  • The traffic policy to apply on failover to the backup line has been defined and tested.
  • Data consumption of the quota and health checks has been calculated.
  • The modem, router and antenna unit are connected to the UPS.
  • The monthly failover test has been added to the calendar; results are being recorded.

Selecting cellular backup lines, carrier processes and monitoring fall within our Metro Ethernet and internet line management service. The primary source for the definition of the shared address range is RFC 6598.

Frequently asked questions

How do I tell whether the cellular line is behind CGNAT?

Compare the IPv4 address assigned to the router’s cellular interface with the public address shown by an IP lookup service over the same connection. If the interface address is in the 100.64.0.0/10 or RFC 1918 range, or the two addresses differ, the connection is behind carrier NAT.

Can a branch behind CGNAT build an IPsec tunnel to headquarters?

It can, but the branch must initiate the tunnel. The headquarters side is configured to accept dynamic (dial-up) peers, UDP 4500 is open for NAT-T, and the branch identity is defined with an FQDN or similar identity instead of an IP address. NAT keepalive delays the mapping in the carrier NAT table from timing out; DPD detects that the far end is unreachable. Whether the tunnel comes up and stays stable depends on the carrier’s NAT behavior; verify it in the failover test.

Signal bars are full but speed is low. Why?

Bars mostly show only signal strength (RSRP or RSSI). If interference is high, SINR and RSRQ stay low and speed drops; the cell’s load at that moment also determines speed. When deciding, look at RSRP, RSRQ and SINR together, and repeat tests at different times of day.

How often should we test the backup line?

At least once a month, run a planned failover test by physically disconnecting the primary line. In addition, continuously monitor the cellular interface status, signal values and a periodic reachability check over the backup line; problems then show up before you need the line.

Support

Contact us for support or a discovery session on this topic.

Contact form
[email protected]

Eretek Network

We build switching, routing, cabling and wireless networks with VLAN segmentation and redundant uplinks, and provide regular maintenance and fault support.

Knowledge Center

Other articles