Five years into 5G—success stories abound and more to come

Qualcomm SVP Durga Malladi looks at 5G success stories to date, and discusses the AI-enabled future of telecoms

In an wide-ranging interview ahead of Mobile World Congress with RCR Wireless News, Qualcomm SVP and GM of Technology Planning and Edge Solutions Durga Malladi looked back at the first five years of 5G, bringing us up to present, then ahead at what’s to come with the advent of 5G-Advanced and the continued evolution of use cases and services. 

“Where are we going with 5G as we enter the mid-point of a typical G cycle?” he asked. “As we enter 2024, it’s been five years give or take.” He recalled the classic 5G triangle we’re all familiar with–enhanced mobile broadband (eMBB), ultra reliable low latency communications (URLLC) and massive machine-type communications (mMTC). “It’s a good time to actually take stock of where we are as an industry. What are some of the emerging trendlines we didn’t predict…in 2019, but they actually happened to latch on pretty quickly as a part of what we’ve done with 5G.” 

The beginning phase of 5G, Malladi said, was all about expanding coverage, increasing network capacity and passing on to end users the benefits of higher data rates “which we kind of take for granted these days.” But, he said, “The cost per bit delivered to the end customer has come down dramatically with 5G—huge plus to network operators.” And this will continue, he said, noting the ongoing advancements of mix-and-match carrier aggregation in both sub-6 GHz and millimeter wave frequencies as well as with FDD and TDD bands. 

The 5G FWA “sleeper hit story”

“That being said,” Malladi continued, “one specific industry which has continuously done well, and I would argue that even though it existed even before with 4G, it really got legs with 5G, is fixed wireless access [FWA].” Indeed, FWA has been a bright spot for US operators bringing the competition to entrenched cable home and enterprise broadband providers, and in developing markets has brought fiber-like speeds but wirelessly. 

He called FWA “one of the success stories of 5G. I believe we should embrace that…I think it’s almost like a sleeper hit story. And there’s more to come with the inclusion of 5G Reduced Capability (RedCap) platforms like Qualcomm’s Snapdragon X35 in consumer premise equipment. The idea here is that while fiber-like speeds are in demand in some markets, the lower data rates offered by RedCap will be a good fit in many markets and with many consumers. “There has been a true democratization, in that sense, of fixed wireless offerings across all tiers,” Malladi said. 

Read this blog for more information on Qualcomm’s 5G FWA Platform Gen 3 and the newest Qualcomm 5G Fixed Wireless Access Ultra Gen 3 Platform which is tailored for millimeter wave operators seeking a lower power profile, smaller footprint and reduce system cost. 

Malladi also examined the increasing penetration of artificial intelligence (AI) into modem-RF platforms which Qualcomm introduced with its X70 platform with dedicated AI accelerators. “This is not necessarily about using something specifically from standards, 3GPP standards, because that we expect will happen as a part of the general evolution of 5G-Advanced and eventually going into 6G, but it’s about increasing the efficiency of the platform in of itself,” he said, calling out improved user experience, advantageous power consumption, improved use of network capacity and more advanced positioning. “The whole idea of smart platforms which rely upon AI techniques to improve the overall efficiency, this has become mainstream…We are driving this. Operators love the idea.” 

Hybrid AI and the outlook for edge inferencing

Switching over to how networks are evolving, Malladi called out developments in network hardware/software disaggregation and open fronthaul interfaces specified through the O-RAN Alliance. He said he expects “tons of news” on this topic next week at Mobile World Congress in Barcelona. “I believe we have reached a point in time where we are no longer talking about the feasibility of O-RAN or the technical capabilities of O-RAN…This is the year for commercial rollout announcements…As Qualcomm, we have been a strong proponent of continuing down that path because we believe it is a natural step to take as we go towards a virtualization of the network and, second, incorporation of additional capabilities in the network that go beyond what we have done so far with the traditional way of vertical silos.” 

With both O-RAN and more broadly with the distribution of radio access network (RAN) architectures, Malladi noted the opportunity to infuse AI more deeply into radio systems. “We also expect to see AI to start making a difference in terms of network management and orchestration.” This sort of bluesky work has segued into more practical applications with major implications for RAN efficiency and optimization. “It’s a trendline as well as we make the networks more and more cloud native, clearly the capabilities of the network go beyond just RAN….We see this as a pretty big natural step.”” 

This blog goes into detail around Qualcomm’s approach to open and/or virtualized RAN processing for intelligent edge systems as well as the role of AI in the larger context of a hybrid AI architecture that utlizes datacenters, edge computing and on-device AI capabilities.

From Malladi: ““We are diving AI as close as possible to the edge… Let’s go all the way to the [distributed unit]—there is a case for, at the DU in a network, you have the ability to now run not just RAN workloads, so not just your MAC, your scheduler and Layer 2 processing, but you also have the opportunity to run AI workloads.” Referencing the capabilities for on-device generative AI in the context of handsets, he continued, “There are other categories of devices going beyond handsets which might not be able to run these large models…at that point in time the idea of running generative AI inference all the way at the edge is a new angle. Back in 2019, we didn’t talk about it like that; we talked about in a generic concept of edge compute. This is far more specific. This is simply saying, if I am interested in offloading my processing as an edge device…then, in a simplistic sense, the inference will run right at the edge, at the DU, then I will have all the information cutting down the latency dramatically. So there is a an opportunity  for edge AI to make its presence in the DU servers…We believe that journey has just about started right now.” 

“I think we redefine the typical G cycle with 5G”

Back to the 5G triangle, Malladi recapped the success of eMBB, talked up RedCap as the trajectory for mMTC, and said URLLC is a work in progress with significant advancements in latency and the gradual insertion of reliability. And looking at the “typical G cycle” which are about 10 years give or take, and in the context of the slow roll from 5G Non-standalone to 5G Standalone, and with 5G-Advanced in the offing, Malladi said, “We designed 5G so that it was like an easier transition from 4G. It was not like you are either on 4G or 5G. It’s not surprising that standalone came in a little later.” In fact, the ability to exploit new 5G core technologies “is just in the beginning now. It is by design…Is this a typical G cycle? I think we redefine the typical G cycle with 5G.”