5G

5G is the fifth generation of cellular network technology, the successor to 4G LTE, standardized by the 3GPP standards body and designed to meet the ITU-R IMT-2020 performance targets for higher data rates, lower latency and greater connection density. The radio interface, 5G New Radio (NR), was first specified in 3GPP Release 15 (non-standalone mode completed December 2017, standalone mode in 2018), and the earliest large-scale commercial launches followed in 2018 and 2019. For virtual and augmented reality, interest in 5G centers on two of its properties, very high downlink bandwidth and low end-to-end latency, which together make it possible to stream rendered graphics to a lightweight headset rather than rendering everything on the device.

Overview of 5G

5G is built around three broad usage scenarios defined for IMT-2020: enhanced mobile broadband (eMBB) for high-throughput consumer services, ultra-reliable low-latency communications (URLLC) for mission-critical control, and massive machine-type communications (mMTC) for large numbers of low-power devices. The first two are the ones that matter for Virtual Reality and Augmented reality: eMBB supplies the bandwidth needed to carry high-resolution stereo video, and URLLC supplies the bounded, reliable latency that interactive headsets require.

5G NR operates in two frequency ranges defined by 3GPP: FR1, the sub-6 GHz bands (roughly 410 MHz to 7,125 MHz), and FR2, the millimeter-wave (mmWave) bands (roughly 24.25 GHz to 71 GHz). mmWave offers the very wide channels and high capacity that streamed graphics benefit from, but its signals travel only short distances and are easily blocked by walls, windows and vegetation, so coverage is harder to guarantee than on sub-6 GHz and mmWave is mainly used in dense urban areas such as stadiums and city centers.

The headline IMT-2020 minimum performance targets, set by the ITU-R, are listed below. These are theoretical maxima for the technology family rather than figures any single user reaches in the field.

Metric	IMT-2020 target
Peak data rate (downlink)	20 Gbit/s
Peak data rate (uplink)	10 Gbit/s
User-experienced data rate (downlink)	100 Mbit/s
User-plane latency, eMBB	4 ms
User-plane latency, URLLC	1 ms
URLLC reliability	1 - 10-5 for a 32-byte packet within 1 ms

Why 5G matters for VR and AR

A head-mounted display is sensitive to a delay called motion-to-photon latency: the time from a user moving the head to the corresponding new pixels being lit on the screen. If this delay is too long the displayed scene lags behind real head motion, which breaks the sense of presence and can cause discomfort or nausea. Industry guidance, used by both Qualcomm and Ericsson, puts the upper bound for the full motion-to-photon path at about 20 milliseconds; above that, lag becomes perceptible and VR sickness becomes more likely. 3GPP adopted the same 20 ms motion-to-photon figure as the reference target in its own XR studies.

This is the constraint that makes 5G relevant. A self-contained standalone VR or AR glasses device is limited by the processing power, thermal budget and battery of a small wearable, which caps how detailed its graphics can be. Moving the heavy rendering off the device to a more powerful computer removes that cap, but only works if the network round trip is short and consistent enough to fit inside the latency budget. The wide bandwidth of eMBB carries the rendered frames, and the low, bounded latency associated with URLLC and 5G time-critical communication is what could keep the loop inside the budget; high latency, not bandwidth, is generally the binding limit.

Cloud and edge rendering

The main XR application of 5G is remote rendering, also called split or cloud rendering, in which a thin headset offloads expensive graphics work to a server and receives finished frames back over the air. Qualcomm describes the concept as "Boundless XR": essential, latency-sensitive work (tracking, pose generation and final display adjustment) stays on the device, while heavier rendering is augmented by an edge cloud reached over a low-latency, high-capacity 5G link.

In Qualcomm's split-rendering pipeline the device samples its sensors and generates a six-degrees-of-freedom (6DoF) head pose, which it sends uplink to the edge server. The server uses that pose to render the next frame, encodes it, and streams the encoded data back downlink. The headset decodes the frame and, using the most recent head pose, performs a final on-device adjustment known as asynchronous time warp to reproject the image to where the user is actually looking. That last on-device step is what keeps motion-to-photon latency within the roughly 20 ms budget even though the frame itself travelled to a server and back; Qualcomm notes the time-warp adjustment has to be done on the device for this reason. The edge server can sit on-premise (for example at a venue) or deeper in the network, and the closer it is, the lower the achievable round-trip latency. Locating compute at the operator's network edge in this way is referred to as multi-access edge computing (MEC).

Offloading does not have to be all-or-nothing. Ericsson describes a spectrum of split points, from a low-offload mode where the device does most of the work, through a mid-offload mode where rendering happens at the edge while tracking stays local, to a high-offload mode where the device sends only raw sensor data and the network does everything else. Ericsson reports that the deeper offload modes cut device energy use substantially, by roughly threefold, fourfold and more than sevenfold respectively, which is what allows a smaller and lighter headset; the trade-off is a heavier dependence on the network. How much of the latency budget the transport network consumes depends on where the rendering server sits: Ericsson puts transport latency at roughly 1 to 5 milliseconds for an edge site close to the user, rising to around 5 to 20 milliseconds for a regional data center.

Several products and trials have demonstrated the approach. Qualcomm unveiled what it called the first 5G-enabled XR reference design in February 2020, pairing the Snapdragon XR2 platform with the Snapdragon X55 5G modem, and ran over-the-air Boundless XR trials with Ericsson beginning in 2020 to optimize the airlink for XR traffic. Separately, NVIDIA announced its CloudXR platform in October 2019, a software development kit that streams cloud- or edge-rendered VR and AR (including SteamVR and OpenVR content) over 5G to client devices that do not need high-end graphics hardware of their own.

Network slicing and time-critical communication

Because XR competes for the same radio resources as ordinary mobile traffic, 5G provides ways to protect it. Network slicing lets an operator create multiple logical networks on one physical infrastructure, each with its own isolation, resources and quality of service; 3GPP defines a slice as a logical partition created on demand to serve a particular service category. A dedicated low-latency slice can carry XR traffic separately from best-effort broadband, which Ericsson frames under the broader heading of time-critical communication: delivering data within a stated latency bound (from about 1 ms to tens of ms) at a chosen reliability level (around 99 to 99.999 percent), while admission control and resource partitioning shield other services from the demands of XR.

3GPP has progressively added XR-aware features to 5G NR. Release 16 (frozen in 2020) introduced the URLLC enhancements and reduced latency that interactive media depend on, and 3GPP's XR study (TR 26.928) characterizes the traffic profiles of split-rendered and other XR architectures, including their downlink rates, latencies and packet-error tolerances.

Limitations and realistic assessment

Despite years of demonstrations, cloud-rendered XR over 5G remains largely nascent rather than mainstream, and the constraints are practical rather than theoretical.

• Latency, not bandwidth, is the hard limit. The ITU latency targets are unloaded best cases for the radio link alone, whereas the full motion-to-photon path also includes uplink of the pose, processing and rendering on the server, encoding, downlink, and decoding. Fitting that entire loop within about 20 ms is demanding, and analysts note that streaming XR has long been held back more by latency than by bandwidth. On-device reprojection (time warp) hides some of this delay but cannot eliminate the dependence on a consistently short round trip.
• Coverage and the mmWave trade-off. The capacity that streamed graphics want is most available on mmWave, whose signals are short-range and easily blocked by walls, windows and vegetation, so high-end cloud XR tends to work only in well-covered, densified or indoor deployments such as venues rather than over wide-area macro coverage.
• Capacity per cell. Tightening the latency target or raising the guaranteed bitrate for XR reduces how many users a single cell can serve simultaneously, a direct capacity-versus-quality trade-off.
• Edge proximity is as important as the air interface. Because the server must be physically close to keep the round trip short, the availability of edge computing nodes near users can matter as much as the 5G connection itself; vendor claims of imperceptible latency are best treated with caution.

In short, 5G genuinely enables remote-rendered XR and removes the on-device compute ceiling, but the experience is gated by real-world latency, coverage and edge placement, so as of the mid-2020s it is an emerging capability concentrated in controlled environments rather than a ubiquitous one.

See also

• Cloud gaming
• Augmented reality
• Standalone VR
• Qualcomm Snapdragon XR2
• Motion-to-photon latency