Raising Upload Performance for Mobile AI with FDD Massive MIMO

There’s renewed interest in ending the twin 5G digital divides — between rural and urban 5G, and between indoor and outdoor. 5G has tended to offer fast speeds only where higher-frequency spectrum bands reach, but this is starting to change, provided operators take advantage of progress in mobile technology. The arrival of AI and new device types is making this more pressing.

Network vendors are now offering new frequency-division duplex (FDD) equipment that makes it easier for operators to deploy multiple FDD bands in one unit. At the same time the multiple-input and multiple-output (MIMO) technologies that drive the performance of time-division duplex (TDD) networks are increasingly arriving on FDD too, with increased numbers of antenna elements that help boost FDD band capacity.

This enables operators to improve upload speeds everywhere, which supports the rise of mobile AI and extended reality. The release of new spectrum and reuse of existing bands as operators switch off 2G and 3G mean there are more low-frequency bands available for 5G that benefit from new FDD radios and antennas.

Mobile AI’s Potential to Change Mobile Data Usage

Unusually, there is a consensus among the large mobile network suppliers about the importance of uplink for supporting mobile users’ needs. AI is one of the key drivers increasing the importance of uplink performance. Ericsson indicates that although uplink traffic accounts for about 10% of overall data traffic, for AI the proportion is already much higher, with 26% of AI traffic being uplink.

Huawei reports that for 17 operators in 12 countries, growth in mobile uplink data traffic is on average 1.59 times faster than growth in downlink traffic. Both Nokia and Huawei argue that a consistent 20 Mbps upload speed everywhere is essential to support emerging mobile uses, for example, extended reality, connected cameras, visual intelligence and robotics. The latter is what Nvidia calls “physical AI”.

High frequency 5G spectrum bands using TDD are not great at supporting wide coverage and good upload performance. Many early 5G deployments relied on these bands, which are typically around 3.5 GHz. This type of spectrum boosted network capacity and overall speed enormously, but these bands do not penetrate well inside buildings nor are they suited to rural deployment, which led to unpredictable 5G speeds for users and applications.

The other issue for AI is that mobile speeds are highly asymmetric, with typical upload speeds that are considerably slower than download speeds. This means that anything that relies on uplink suffers. Ookla reports that in August 2025, mobile devices with modern chipsets saw upload speeds that were just 14% of mobile download speeds. This compares with a considerably more symmetric fixed experience, where uplink speeds are slightly more than half of typical download speeds. Fixed upload speeds will rise further as fibre-to-the-premises broadband becomes more widespread.

For users of smartphones and AI smart glasses, this weak mobile uplink experience leads to a wildly inconsistent and unpredictable experience. For AI that runs on the device — typically smaller models that fit in a smartphone’s memory — the response will be quick. But for requests that are sent to cloud AI because they need to be processed by a large model or they relate to global knowledge or data that only exists off the device, response will depend on network latency, network uplink speed as well as the speed of the cloud AI to respond. For the user, it will not always be clear which request will be handled on-device and which will be sent to the cloud, leading to dissatisfaction with the varying response times.

There are a few reasons for this increased importance of uplink for AI data traffic. Much of the AI uplink traffic will be visual information captured by still or video cameras that enable AI to understand the world around it. Today, that visual information mostly comes from smartphones or cellular-connected traffic and security cameras. But in the future, smart glasses, vehicles, drones and robots will all create significant amounts of visual data for cloud AI to analyse.

An early example of AI’s creation of large amounts of uplink data traffic is the Live AI feature for the Ray-Ban Meta smart glasses that is available in select countries. When the wearer triggers Live AI, the camera and microphones continuously record, with Meta analysing what the glasses can see and hear. The value to wearers is that when they ask a question about their surroundings, the glasses can respond extremely quickly because the AI-powered analysis has already been completed.

The Role of New FDD Spectrum and Equipment in Boosting Uplink Performance

Sub-3 GHz FDD bands offer better rural coverage and indoor reach as well as adding capacity. Some of this is new to mobile, in other markets operators are refarming low-frequency bands for 5G when they switch-off 2G and 3G networks. New 700 MHz spectrum is on the way in Kenya, Indonesia, Turkiye and Morocco. Similarly, there is new 600 MHz capacity in Saudi Arabia and the United Arab Emirates, with Mexico at a planning stage. Across the EU, there are efforts to use 1.4 GHz for 5G.

To create a consistent network experience for users, wherever they are, operators need to utilize all of this low-band FDD spectrum. No single band has sufficient spectrum capacity on its own. Instead, it’s essential to use carrier aggregation across multiple FDD carriers to meet user demands.

By increasing the proportion of data traffic carried on FDD bands, operators can also improve revenue from fixed wireless access (FWA) services — an important new revenue stream. In urban areas boosting FDD capacity takes load off TDD bands, meaning there is more TDD capacity to support FWA. It also makes self-installation of FWA customer premises equipment easier because of the improved in-building reach of FDD bands. Where only FDD exists, for example in rural areas, offering deploying 5G on more FDD bands on all sites makes FWA more viable.

Massive MIMO Arrives on FDD Sub-1 GHz

As well as the new multiband FDD radio units, the other important innovation is the arrival of massive MIMO on FDD. For example, Huawei’s triple low-band FDD GigaGreen units now offer up to 16T16R antenna elements. Innovations in the antenna design make it easier to deploy on existing sites. New materials used in place of metal reduce weight by 70%, claims Huawei. Also, the company says it has reduced the width of the antenna array by 70%. The units support all radio access technologies, enabling a straightforward upgrade path from whatever mix is currently in use to 5G.

The range of this unit is also longer than older 2T2R LTE units. Huawei reports that the range for delivering 20 Mbps uplink increases from 500 m to 1 km. There are similar improvements in indoor performance.

For higher capacity, Huawei also has a 32T32R unit that supports a combination of 2.6 GHz, 2.1 GHz and 1.8 GHz. This reduces the overall weight by 42% and power consumption by 10%, according to Huawei. This band combination is suited to operators that have existing 3G or 4G deployments on those bands. Uplink should be boosted about fivefold compared with an older 4G deployment using 4R.

Mobile AI Is Not the Only Reason to Focus on FDD Now

There is still a degree of uncertainty about the extent to which mobile AI and new device types will boost uplink data usage. But those are not the only reasons for operators to act now. Notably, these FDD innovations can remove the pain points associated with current 5G experience indoors and outside cities, helping operators to differentiate from rivals while also building the foundations for further 5G-Advanced services.