4G↔5G in Practice: What CSL Multi-network IoT SIMs Mean for Vehicles, Hospitals and Buildings

Most real-world 4G/5G capable fleets spend time on both types of network. Even with 5G coverage improving, many mobile devices pass through cells, buildings and load conditions that trigger a 4G↔5G radio access transition. Depending on the network state, this may involve inter-RAT handover, idle-mode reselection or, on Non-Standalone (NSA) networks, the addition or removal of the 5G NR secondary carrier.

That transition alters the performance profile of devices. For example, when moving from 5G down to 4G:

• Latency: at UK median level, the difference between 5G and 4G response times is small, often only a few milliseconds. In the conditions that commonly trigger a change, such as weak signal, loaded cells, indoor locations or session recovery, the application-level impact can be materially larger, and jitter (packet delay variation) can widen.
• Uplink headroom is usually greater on 5G than 4G, so a move to 4G is felt most on cameras, telemetry bursts and OTA updates.
• Setup and short transfers are typically faster on 5G, particularly on Standalone (SA) networks, but it is important to understand that IoT devices will not land on 5G every time.

For applications that require voice, push-to-talk, video or control in near real time, these shifts show up as stutter, lag, or longer mouth-to-ear (voice) and glass-to-glass (video) delays, unless you plan your devices and applications for them.

The remedy is to:

• design for both 4G and 5G performance envelopes where possible
• use adaptive bitrate and elastic jitter buffers for media
• prioritise critical flows over background jobs
• keep safety-critical decisions local where appropriate
• measure latency, jitter and packet loss so buffers and SLAs are evidence-based, using p95/p99 values rather than medians alone.

Packet loss should also be measured alongside latency and jitter, because late or dropped packets often determine whether a transition is merely measurable or actually visible to the user.

In short, dual-capable 4G/5G devices succeed when they are prepared for both technologies and have access to multiple permitted networks where available.

If you work on vehicle systems, hospital technology, building management or critical infrastructure, this article shows what 5G↔4G switching means for latency, jitter and uplink, and how that translates to IoT devices in everyday use.

CSL’s multi-network IoT SIMs and rSIMs help by broadening access to permitted networks at each location, subject to coverage, device band support, roaming agreements, SIM provisioning and steering policy. That increases the likelihood of usable 5G coverage, and rSIM adds profile-level resilience when connectivity on the active profile is impaired. These mechanisms improve availability rather than guaranteeing session continuity: active sessions may still need router- or application-level recovery. The 4G↔5G transition itself is managed by the device and the serving network, so the practical task is to measure the latency and jitter envelope on both and size buffers accordingly.

Explore our 5G-ready hardware: 5G IoT Router

Quick definitions for the acronyms used throughout this article:

• PDV: packet delay variation (jitter)
• RAT: Radio Access Technology, the radio generation in use (for example 4G LTE or 5G NR)
• SA: Standalone 5G. 5G NR radio connected to a dedicated 5G core network, operating independently of 4G.
• NSA: Non-Standalone 5G. 5G NR radio added alongside a 4G LTE anchor, with both served by the 4G core network (EPC). The device keeps an LTE connection and gains 5G capacity on top.
• EN-DC: E-UTRAN New Radio Dual Connectivity, the NSA arrangement in which a device holds an LTE anchor connection with 5G NR added as a secondary carrier.
• UL: uplink (device to network)
• DL: downlink (network to device)

Understanding the effects of 5G↔4G transitions means going beyond theoretical limits and placing devices in the wider cellular ecosystem: how core network architecture, radio access infrastructure and operator policy interact in practice. Real-world performance depends on how network and service providers operate across the mobile landscape, which is why this article works from measured data rather than headline specifications.

What UK 5G and 4G actually deliver today: latency, speeds and how often devices switch.

A note on the data first. Ofcom’s Mobile Matters 2025 is a benchmark of consumer Android smartphones on UK public networks, using crowdsourced data provided by Opensignal and collected between October 2024 and March 2025; its methodology excludes MVNO and roaming subscribers from operator results. It should therefore be treated as directional for IoT design rather than a direct prediction of multi-network IoT SIM performance, because IoT devices, antennas, roaming arrangements, modules and application paths can produce materially different results.

Coverage and 5G SA availability continue to change, so the performance figures here should be treated as a measured benchmark rather than a permanent coverage statement. Deployment coverage should be checked against the latest Ofcom Connected Nations data when planning a live estate.

With that caveat, the UK picture is as follows:

• Standalone 5G (SA) showed around 15% lower latency than Non-Standalone 5G (NSA) and around 45% faster short file downloads; 70% of SA download speed measurements were at 100 Mbit/s or higher, compared with 46% on NSA.
• Median response times were close between generations: roughly 15 to 21 ms on 5G and 18 to 23 ms on 4G across operators. The structural 5G advantage shows more clearly in file transfer times than in median latency.
• 5G SA’s data connection success rate (95.9%) was slightly below NSA’s (97.6%), although slightly above 4G’s (95.7%). Uploads were mixed: fewer very slow results on SA (10% of tests under 2 Mbit/s versus 18% on NSA), but NSA had a marginally higher share at 20 Mbit/s or above (30% versus 28%).
• 5G SA accounted for only 2% of all 5G connection attempts during the measurement period, so its benefits are not yet widely experienced, although rollout is expanding.
• Only around 28% of connection attempts were on 5G during the study period, meaning most 5G devices will still spend meaningful time on 4G LTE.

Independent measurement providers such as Ookla have reported broadly similar UK patterns, although results vary by operator, geography, device mix, spectrum holdings and methodology. For example, Ookla Speedtest Intelligence data for the first half of 2025 reported UK 5G median downloads ranging from roughly 100 Mbit/s to 230 Mbit/s across operators, with median uploads typically in the teens of Mbit/s.

What that measurement data means in plain terms for IoT devices

• Median measured latency at UK level is around 18 ms on 5G and 21 ms on 4G, ranging from 15 to 23 ms across operators and technologies. Note that Ofcom’s latency figure is half the round-trip time to a nearby content delivery network test server; it does not represent the full application path through VPNs, private APNs, brokers and application servers, so allow additional budget for those.
• On downloads, 47% of 5G measurements exceeded 100 Mbit/s against around 10% on 4G. On uploads, 30% of 5G tests reached 20 Mbit/s or higher versus 14% on 4G, while 18% of 5G uploads and 33% of 4G uploads fell below 2 Mbit/s. 5G usually offers more uplink headroom, but the distribution is wide, so sustained video upload should be engineered conservatively.
• Time on 5G will be well below 100%, so your fleet will regularly move between 5G and 4G LTE (Ofcom, Mobile Matters 2025). Design against measured p95/p99 latency and jitter rather than median values alone.

How a device decides to use 5G or fall back to 4G, and what that switch does to performance.

CSL’s multi-network IoT SIMs allow devices to register on permitted partner networks (PLMNs), subject to coverage, device band support, roaming agreements, SIM provisioning and steering policy. Once registered, the device and that network determine which Radio Access Technology (RAT) is used, based on coverage, signal quality and device capability; the SIM does not control the 4G↔5G transition within a serving network. 5G access while roaming also depends on the home and visited operators’ 5G roaming agreement, SIM and profile provisioning, steering policy, APN/DNN configuration, device capability and the visited network enabling the relevant 5G roaming service; where these conditions are not met, devices use 4G. If a serving network becomes unavailable, the device can reselect or register on another permitted network, but active sessions will usually need to re-establish at application or router level.

When 5G New Radio (NR) signal and load conditions are strong, devices can remain on 5G; when they are not, 4G LTE becomes the active radio connection. On NSA networks, what appears as losing 5G is often the removal of the NR secondary carrier while the LTE anchor continues, which is typically less disruptive than a full inter-RAT handover.

To avoid instability at cell edges, modern devices apply hysteresis and dwell timers so they do not bounce between cells at the boundary. In well-engineered same-network mobility, the radio transition may be brief; however, cross-PLMN registration, roaming recovery, profile switching and application session recovery can take longer, so designs should assume that active sessions may need to reconnect and should measure recovery time in the target estate. The practical effect of a 5G to 4G change is therefore:

• Latency typically rises versus the same location on 5G. At the median the difference is small, around 3 ms at UK level, but congestion and weak coverage widen the tails, and it is the tails that real-time applications feel; measure p95/p99 for your own routes and sites.
• Uplink (UL) headroom often drops into single to low double-digit Mbit/s, which you will notice first on camera and video traffic, bulk telemetry and Over-the-Air (OTA) updates.
• Jitter (packet delay variation) tends to increase under LTE load, which can disturb real-time streams unless buffered or prioritised. Ofcom’s 2025 measurements confirm the structural 5G advantages on response time and short-file performance, with SA strongest.

In simple terms, jitter is delay variation. For real-time and interactive applications, either keep it small or add a buffer and accept extra delay.

Jitter is formally packet delay variation (PDV): the variability of one-way packet delay across a flow.

The IETF defines IP packet delay variation in RFC 3393 and discusses the applicability of delay-variation metrics in RFC 5481. ITU-T Y.1540 and Y.1541 define related IP performance parameters and network performance objectives. Together, these standards explain how delay variation is measured, interpreted and budgeted, especially where de-jitter buffers are used for real-time media.

For voice, the ITU-T G.107 E-model and G.114 show how delay and loss (often exacerbated by jitter) degrade conversational quality.

In short: stable latency matters as much as low latency for real-time control, voice and video.

What jitter looks like in the field: bursts of queuing or radio scheduling variance produce frame-time variation. For example:

• If your de-jitter buffer is small, you will see choppy audio, video stutter or control loop spikes.
• If it is large, you will add delay, hurting interactivity.
• The right buffer is application-specific, but the governing principle from the ITU and IETF is universal: either bound the variation, keeping jitter within a tight window so the application sees a near-constant delay, or budget for it by adding a jitter buffer and accepting the extra latency.

The practical effects of 5G↔4G switching for vehicles, hospitals, buildings and critical infrastructure.

1) Vehicle system providers (telematics, ADAS offload, V2X support)

What a 5G to 4G change does to in-vehicle uploads, video and remote assistance.

• The reality today: in favourable urban 5G conditions, uplink may support 10 to 20 Mbit/s for 1080p streams and rich telemetry, but this should not be treated as a guaranteed service envelope: around half of UK 5G upload measurements fell below 10 Mbit/s. On corridors with coverage gaps, devices drop to LTE, where uplink headroom typically falls further and latency and jitter rise. That shifts your effective camera count and bitrate, and the stability of driver-assist teleoperation or remote monitoring. Ofcom’s UK measurements substantiate the variability of 5G exposure; results vary by location and load.
• Why jitter matters here: cooperative or tele-assisted manoeuvres are latency- and variation-sensitive. Automotive groups such as 5GAA document service-level requirements ranging from around 10 ms for some cooperative manoeuvres to 100 ms or more for basic safety messaging, depending on the use case (5GAA, C-V2X Use Cases and Service Level Requirements). These targets depend on architecture, QoS, edge placement, radio conditions and application design, and are not generic guarantees from public cellular. Public 4G and 5G are well suited to telematics, video upload, monitoring and non-safety-critical assistance; safety-critical vehicle control should remain local unless the communications service has been engineered and validated against the required latency, reliability and fail-safe targets.

2) Hospital systems (clinical mobility, imaging carts, in-hospital communications)

How 4G↔5G transitions affect consultations and imaging across hospital campuses.

• The reality today: on hospital campuses, public 5G latency is routinely in the tens of milliseconds, and short file downloads benefit materially from SA where available. When the connection moves to LTE, or in dense buildings, transfer times and uplink throughput can degrade, affecting high-bitrate imaging pushes, while voice and telepresence remain workable with appropriate buffering. Ofcom’s UK dataset shows SA’s material advantage on short downloads and response time, exactly the metrics you feel during session setup and transfer bursts; SA use is still a small share of 5G connections, so plan for NSA-level performance as the norm.
• Why jitter matters here: for voice and video in remote clinical collaboration, jitter forces larger de-jitter buffers to avoid stutter; that either adds delay or causes late packets to be discarded, which the E-model captures as delay and packet-loss impairment to conversational quality. In imaging workflows, jitter mainly affects throughput smoothness, where sustained, steady-rate delivery matters more than peak bitrate. Public 5G suits mobile connectivity, carts, telepresence and non-critical data flows; regulated or safety-critical clinical workflows require site-specific risk assessment, redundancy and validated fallback before relying on any public network.

3) Building Management Systems (BMS): HVAC, lifts, metering, access

What switching means for lift and HVAC telemetry, access video and firmware updates.

• The reality today: most BMS control and telemetry tolerates latency in the tens to hundreds of milliseconds. Falling back from 5G to LTE rarely breaks the application, but you will notice 4G↔5G transitions on video intercom, mobile access and bulk firmware updates (lower uplink, more jitter). UK-wide measurements show 5G time-share is still limited, so mixed 5G and 4G operation is normal.
• Why jitter matters here: for control loops with seconds-level update intervals, jitter is usually absorbed. But for event-driven access control and video, jitter drives how large your buffers must be to keep interactions responsive.

4) Critical infrastructure and industrial control

Where public cellular fits for cyber-physical and control applications.

For cyber-physical control applications in industrial and CNI environments, public cellular should normally be treated as a connectivity layer rather than the sole control-plane dependency. 3GPP TS 22.104 sets out service requirements for cyber-physical control applications in vertical domains, including factories and electric power systems, with latency and delay-variation bounds that best-effort public networks are not engineered to guarantee. Where bounded latency, delay variation or reliability targets are required, the service should be engineered and validated using private, managed or QoS-controlled architecture, with public 4G and 5G providing wide-area reach, telemetry and backup paths rather than assumed deterministic control.

• For vehicles, 5G-capable cellular modules with LTE fallback preserve functionality across coverage changes. Antenna quality, MIMO (Multiple-Input, Multiple-Output) and uplink features such as UL carrier aggregation will matter far more in practice than theoretical maximums.
• For BMS controllers, LTE Cat-1 or Cat-4 remains adequate unless you run video or heavy remote access. For long-life battery sensors, consider Low-Power Wide-Area (LPWA) profiles such as LTE-M (LTE for Machines) or NB-IoT (Narrowband IoT) for multi-year life, noting that LTE-M and NB-IoT coverage and roaming support vary by operator and territory.

These are industry-standard views reflected in 3GPP specifications and industry alliance guidance.

The device traits that matter and the two biggest cost drivers.

For device budgets, two line items dominate:

1. Data volume, especially uplink: 5G makes video and high-rate telemetry feasible, and costs track volume, not the “5G” label itself. UK measurements put median uplink on public 5G in the teens of Mbit/s, reaching the low tens on some networks (Ookla, H1 2025), which encourages richer feeds unless you compress, filter or schedule them.
2. Hardware capability: 5G-capable modules and routers cost more than LTE-only, and dual-modem designs add again. Whether that is justified depends on your measurable gains: faster transfers, more uplink headroom, better tail-latency behaviour, or a more stable service envelope under your tested deployment conditions.

How to set success criteria by linking latency and jitter to voice, video and control quality.

• Native mobile voice: behaviour depends on what the device and serving network support. On NSA networks, voice generally remains on VoLTE because the device is anchored to LTE. On SA networks, devices use VoNR where supported, or fall back to LTE and VoLTE (EPS fallback) where it is not. Use 114 as the planning reference for one-way delay: 0 to 150 ms is generally acceptable, 150 to 400 ms is acceptable with increasing care, and above 400 ms is generally unsuitable for normal conversational use. G.107 (the E-model) estimates conversational quality from delay and packet-loss impairments; jitter enters indirectly, through the playout-buffer delay it forces and any late packets discarded.
• Push-to-talk, telepresence and app-based voice: these are usually IP application flows over the data connection rather than IMS voice services, so engineer them separately, with adaptive jitter buffers, reconnection logic, QoS marking where supported, and local fallback procedures.
• A worked example: when the radio connection changes (for example, 5G to 4G), jitter often rises. To keep audio smooth, the application increases its jitter buffer, holding audio a little longer before playing it, which adds delay to the path. The result is increased mouth-to-ear delay: people start talking over each other and pauses feel awkward. If radio changes push you to bigger buffers, you pay in conversational responsiveness.
• Live video and remote viewing: smoothness depends on bounded PDV (jitter). If 5G to LTE switching increases jitter, either raise the buffer, increasing glass-to-glass (camera-to-screen) delay, or drop bitrate and frame rate on the fly. Industry measurements and operator-agnostic analyses show why: upload capacity and response time are often the bottlenecks on public cellular networks.

Why CSL’s multi-network IoT SIMs make 4G↔5G transitions easier to design for.

• Single-network SIMs make service availability dependent on one operator’s coverage, capacity and operational availability at that location. CSL multi-network roaming gives each site more than one permitted network option, subject to roaming agreements and coverage, reducing downtime when a single operator fails. Within whichever network is in use, the 4G↔5G transition is handled by the device and operator, and a change of network requires sessions to re-establish, so the gain is availability at site level rather than uninterrupted sessions. Multi-network access also widens the pool of usable 5G coverage.
• Operator-agnostic performance reality: the UK’s measured data (Ofcom) shows SA’s technical edge, limited but rising time on 5G, and median latency of around 18 ms on 5G and 21 ms on 4G. The medians are close; what separates the experience is exposure (time spent on each technology) and behaviour at the tails, which you should measure for your own estate. Your 5G-capable fleets will experience both RATs daily, and how switching is planned for determines jitter budgets, upload feasibility and session reliability.

Sector depth: for vehicles, automotive bodies such as 5GAA document service-level needs from around 10 ms to over 100 ms depending on the cooperative function; for industrial and CNI applications, cyber-physical control requirements such as those in 3GPP TS 22.104 generally call for engineered private or managed networks rather than best-effort public cellular.

With CSL’s multi-network IoT SIMs and network policy expertise, 4G↔5G transitions become events you can detect, measure and design around rather than assumptions left unmanaged. If your workload is uplink-heavy or jitter-sensitive, we can help you understand and quantify the expected latency and PDV envelope on public 5G versus LTE for your routes and sites, measured at the percentiles that matter, so that control traffic can be prioritised and protected even while video is streaming.

• Ofcom, Mobile Matters 2025, published July 2025; crowdsourced data provided by Opensignal, collected October 2024 to March 2025.
• Ofcom, Connected Nations reports, latest edition; UK fixed and mobile coverage data.
• Ookla Speedtest Intelligence, UK 5G performance for the first half of 2025, reported by ISPreview, July 2025.
• IETF, RFC 3393: IP Packet Delay Variation Metric for IP Performance Metrics (IPPM).
• IETF, RFC 5481: Packet Delay Variation Applicability Statement.
• ITU-T, 107: The E-model, a computational model for use in transmission planning.
• ITU-T, 114: One-way transmission time.
• ITU-T, 1540: IP packet transfer and availability performance parameters.
• ITU-T, 1541: Network performance objectives for IP-based services.
• 3GPP, TS 22.104: Service requirements for cyber-physical control applications in vertical domains.
• 5GAA, C-V2X Use Cases and Service Level Requirements, Volume I, version 2.0.

Note: this article draws on published measurements and guidance from Ofcom, Ookla, 3GPP, 5GAA, the IETF and ITU-T, as listed in the references above. Public-network benchmark figures are directional for IoT design; performance for specific deployments should be validated by measurement.