Most markets still rely on 5G Non-Standalone (NSA) as the main approach for commercial 5G. This setup allows operators to activate new radio (NR) capacity over an existing LTE core, avoiding a complete overhaul to a standalone (SA) core. However, NSA’s architecture, based on 3GPP’s EN-DC option, introduces a structural asymmetry: downlink throughput increases rapidly, while uplink performance lags significantly behind, often more than what marketing claims suggest.
This piece explains what truly influences DL and UL performance in 5G NSA, highlighting the key network-side and device-side factors that can help close the gap. It draws on reliable sources like ITU-R, 3GPP, and GSMA, rather than vendor claims.
The theoretical ceiling: what ITU-R and 3GPP actually define
Starting with the standard baseline helps avoid industry misquotes and ensures clarity. ITU-R defines the minimum technical performance requirements for IMT-2020 (5G) — sets peak data rate targets of 20 Gbit/s downlink and 10 Gbit/s uplink, measured under ideal conditions with all assignable radio resources allocated to a single device. Peak spectral efficiency requirements are set separately at 30 bit/s/Hz downlink and 15 bit/s/Hz uplink.
Two key points from the standards matter more than the headline figures.
- These are laboratory ceilings, not deployment targets. As ITU-R and industry commentary have both noted, the 20 Gbps/10 Gbps figures have not been realized in any production 5G network under typical conditions — they describe what a candidate radio interface technology must be theoretically capable of, evaluated in a controlled test environment, not what a live cell delivers to a user in traffic.
- The DL/UL gap is by design. The 2:1 ratio between downlink and uplink peak targets reflects the asymmetric traffic profile 5G was designed around (content-heavy downlink, lighter uplink), not a technical afterthought. That asymmetry compounds at the deployment level through TDD frame-structure choices, described below.
At the UE side, the actual achievable rate for any given device and deployment is governed by 3GPP TS 38.306 (NR UE Radio Access Capability parameters), which specifies that a UE’s maximum DL and UL data rate is computed per band or band combination from three inputs: the UE’s supported maximum MIMO layers, its maximum modulation order, and the number of physical resource blocks corresponding to the allocated bandwidth.
Critically, for multi-RAT dual connectivity configurations — which is what EN-DC (5G NSA) is — TS 38.306 defines the approximate maximum data rate as the sum of the NR leg and the E-UTRA (LTE) leg. This is the standards-level confirmation of a point that gets lost in a lot of operator PR: an NSA “5G speed” figure is very often an LTE-anchor-plus-NR-secondary combined number, not a pure NR carrier result.
Why uplink lags: the TDD slot-ratio constraint
The single largest structural factor behind weak 5G NSA uplink performance isn’t RF — it’s how the TDD frame is configured.
Most 5G NSA mid-band deployments (n78, n77, n41) use TDD, where a single frequency carries both downlink and uplink, separated in time rather than frequency. The NGMN Alliance’s 5G TDD Uplink White Paper states plainly that with currently deployed TDD DL/UL configurations, uplink capacity is not sufficient to satisfy the requirements of a number of vertical-industry use cases, even though the NR specification itself allows highly flexible TDD frame structures where the ratio of uplink to downlink slots can be configured almost arbitrarily within the frame periodicity.
In practice, operators default to heavily downlink-weighted patterns. Many commercial 5G TDD deployments use downlink-heavy configurations—often around 70–80% of resources—reflecting the historically asymmetric nature of mobile traffic. However, the exact DL/UL ratio varies by operator, spectrum band, and network configuration.
This is how we can understand a surprising discovery from recent throughput-modeling research: on a 100MHz mid-band channel with a common 7:2 DL: UL slot ratio, a high-end device with 2-layer uplink transmission can theoretically reach up to ~286 Mbps uplink, but average real-world uplink throughput measured in the wild was as low as ~10 Mbps, degrading further on typical single-antenna devices.
The difference between what the slot ratio allows and what devices or RF conditions actually provide can be quite large. It’s helpful to look into this first before blaming coverage issues or device hardware.
Addressing the uplink challenge
Improving 5G uplink performance requires more than simply adding spectrum. The first lever operators can pull is the TDD downlink/uplink slot ratio. Because NR supports flexible slot configurations, operators can allocate more uplink resources on carriers or cells serving uplink-intensive scenarios such as stadiums, live-broadcast venues, enterprise campuses and fixed wireless access hubs. This approach has already been demonstrated in practice: Ericsson, stc and Qualcomm aggregated four 100MHz mmWave component carriers using an uplink-heavy TDD configuration alongside NR-DC to achieve multi-Gbps uplink speeds for live media and broadcasting. The key lesson is that, in some scenarios, the limiting factor is not necessarily the amount of spectrum available, but how much of that spectrum is allocated to uplink transmission.
The second lever is Massive MIMO and beamforming. While these technologies are often associated with improving downlink capacity, they can also significantly enhance uplink performance through spatial processing gains at the base station. In TDD networks, channel reciprocity provides an additional advantage: the network can estimate channel conditions from uplink reference signals and reuse that information for downlink precoding, improving the antenna system’s overall efficiency.
Operators can also use carrier aggregation across FDD and TDD spectrum to overcome TDD’s uplink limitations. Unlike TDD, where uplink and downlink share the same spectrum resources over time, FDD provides dedicated paired uplink spectrum. Combining an FDD carrier with a high-capacity TDD carrier can therefore provide a more balanced traffic architecture, particularly for uplink-intensive services. This principle is also relevant to NSA deployments, where the LTE anchor can carry a meaningful share of uplink traffic while NR delivers much of the additional capacity.
MediaTek also explains that R17 Tx Switching lets a device combine 10MHz FDD + 80MHz TDD for uplink by dynamically switching its transmit chain between the two carriers instead of splitting power across both. Compared to older approaches on the same spectrum, it delivers 277 Mbps peak UL—about 80% higher than R15 carrier aggregation (154 Mbps) and 18% higher than R16 Tx Switching (234 Mbps) —making it the strongest of the compared methods for narrowing 5G’s uplink gap.
Ultimately, however, these technical optimizations need a long-term spectrum strategy to support them. GSMA’s Vision 2030 analysis recommends planning for up to 2 GHz of mid-band spectrum per market by 2030, reflecting the growing role of mid-band frequencies in delivering 5G capacity and economic value. For operators, this provides an important benchmark: current allocations of around 100MHz per operator may be sufficient to launch 5G, but supporting future capacity growth—including increasingly demanding uplink applications—will require a deeper, more strategic spectrum pipeline.
What device-side improvements contribute
Uplink MIMO layer support. Qualcomm’s technical material on switched uplink notes the practical limitation of many devices to a single uplink layer (1L) on TDD bands versus 2-layer (2L) uplink achievable nearer the cell center — meaning the UL MIMO ceiling is often set by device antenna configuration and power budget as much as by network configuration.
Uplink transmit switching between FDD and TDD bands. Ericsson’s Technology Review describes a technique where a 2-transmit-chain device dynamically switches its transmit chains between a TDD band (during the TDD carrier’s uplink slot) and an FDD band (during the TDD carrier’s downlink slot), giving devices a way to sustain higher uplink throughput despite the TDD frame’s DL-heavy slot allocation — though Ericsson notes explicitly that this benefit, like most UL enhancement techniques, is strongest near the cell center and decreases as the UE moves toward the cell edge.
Modulation and power headroom at the UE. Because uplink transmit power is device-limited in a way downlink (base-station) transmit power isn’t, achievable uplink modulation order and MIMO rank both degrade faster with distance/SINR than their downlink counterparts — which is why UL performance is more sensitive to indoor/outdoor placement and cell-edge positioning than DL is.
Conclusion
The growing importance of uplink-intensive applications means that operators can no longer optimize 5G networks primarily around downlink demand. While additional spectrum will remain essential, operators can also address the uplink challenge through smarter use of existing resources—by adopting more flexible TDD configurations, strengthening uplink performance with Massive MIMO and beamforming, and combining FDD and TDD spectrum to create a more balanced capacity architecture.
For operators, the strategic priority is clear: future 5G performance will depend not only on how much spectrum they own, but also on how effectively they configure, combine, and evolve it to meet changing traffic demands in both directions.