The evolution of cellular networks follows a fairly predictable cycle of R&D, standardization, application of standards to new equipment, deployment of that equipment, and repeat. But Qualcomm’s Tingfang Ji, Senior Director of engineering, is studying a future where neural networks could potentially replace the approximately18-month air interface upgrade process allowing a kind of natural evolutionary improvement arc.
Speaking to RCR Wireless News, Ji noted the important role of machine learning in many areas including telecommunications, enhancing the interface between device and network and within the network infrastructure. But, “The idea I’m trying to bring up is to say that this type of air interface is updated every 18 months and you only have the next level of feature or performance every 18 months. Is it possible to make the interface more dynamic? As neural network technology evolves, the air interface itself evolves.”
He continued: “As long as you have a pair of networks that understand each other, you don’t have to fix it. We know neural networks evolve very fast…every couple of months you see new breakthroughs, some new improvement. If you can leave the air interface more dynamic, less defined, potentially the 5G–or 6G–network performance can just evolve naturally with the neural network technology without waiting for the air interface to update every 18 months.”
While this concept is relatively nascent, machine learning can have a very real and significant near-term impact on the way operators plan, deploy and optimize networks. With network topology planning, for example, machine learning tools can be applied to transition from a model-based approach to a data-based approach.
As Qualcomm Vice President of Engineering John Smee put it, “The ability to have data and have that machine learning train on that data is something that can transform networks…From the operator standpoint, it can increase that cost efficiency, enabling those more effective deployments where you putting just the right infrastructure at just the right spot.”
5G precise positioning and the road to full duplex
Other areas of focus that speak to the goal of optimizing network performance and efficiency, improving user experience, and supporting expanded feature set, is Qualcomm’s work on both precise positioning and the move from sub-band half duplex and ultimately to full duplex.
Precise positioning is key for 5G applications like autonomous driving, smart agriculture, precision robotics, and more. However, dense 5G network deployments in urban areas–where precise positioning could be of huge benefit to the aforementioned applications–the built environment hinders the performance of existing GPS systems.
Ji and his colleagues are working on technology that would run positioning signals from 5G base stations. Operators could “essentially have a new constellation,” he said, with much better resolution that can complement GPS.
In a telecommunications system, duplexing at a high-level refers to bi-directional communications. Half duplex allows for that bi-directional transmission but not at the same time. Full duplex would enable a simultaneous flow of data in both directions. Ji characterized sub-band half duplex as a bridge to full duplex. In unpaired TDD spectrum, the downlink and uplink slots would be right next to each other in the time domain, which would significantly reduce interference. Subband half-duplex takes it a step further, allocating a portion of the TDD bandwidth for uplink and another for downlink in the frequency domain, so that crosslink interference can be mitigated. “The benefit of this is actually, because the channel is separated, now you don’t see other cell interference anymore.”
Watch the RCR Interview: Wide-area 5G here, for more content, including demos and live interviews, from Qualcomm’s recent Advanced 5G Research Demonstrations for MWC ’21 click here.
The post Can neural networks enable natural network evolution without cyclical air interface updates? appeared first on RCR Wireless News.