Several themes have begun to emerge in the 6G community, providing some clues as to what the next generation might bring.
Universities around the world now experiment with candidate 6G technologies. Many companies are participating in research, too. One such company, which was instrumental in 5G mmWave research is NI (formerly National Instruments). 5G Technology World spoke with NI’s David Hall and Charles Schroeder. NI recently joined the NextG Alliance and became an industrial affiliate of NYU Wireless.
“Odd number Gs introduce new capabilities and the even number Gs get it right,” said Hall. That’s not to say that 5G, and by default 3GPP, got it wrong. As with the introduction of any technology, engineers immediately see ways to improve and enhance it. While 6G is alive in the research community, it’s still undefined. One thing is clear, researchers want 6G development to be based on use cases, not simple specs such as higher data rates.
“5G was about connecting devices, said Schroeder. Enhanced mobile broadband (eMBB) came first and it’s paying the bill for research. We all want to stream AR/VR, which is data intensive.” Figure 1 shows a representation of use cases and features that 6G might address.
Hall and Schroeder noted four areas of 6G research coming into focus:
- Terahertz frequencies. Well, not really 1 THz, but 100 GHz to 350 GHz.
- Joint communications and RF sensing. Indeed, IEEE has already organized a conference around the topic.
- Evolution of MIMO. Cooperative MIMO is one such enhancement.
- AI and ML in the network. Networks will need to adapt to changing conditions and reduce energy consumption.
The sub-terahertz frequencies offer tremendous bandwidth compared to today’s 5G, even with mmWave service. The 77 GHz to 79 GHz band, used for automotive radar, offer twice the spectrum allocated for all of 5G. Higher frequencies such as 95 GHz to 350 GHz show promise for 6G. Such frequencies might bring us holograms as a use case.
The silicon and packaging needed to make those frequencies viable aren’t available yet. “Semiconductor companies need to work on processes and packaging now,” said Schroeder. In addition, the test-and-measurement industry must keep up with developments in semiconductor technology, added Hall. There’s still work to do at mmWave frequencies, which will enhance 5G.
Joint communications and RF sensing
“If 5G connects everything, joint communications and RF sensing will help systems understand the world,” said Schroeder. “If a base station could understand its surroundings, it could plot better signal paths to users.” Going further, Schroeder asked “What if we could put smart surfaces on buildings, which could lead to better path planning?”
Using spectrum for both communications and sensing sounds great, but is there enough spectrum to go around? Could communications and sensing coexist? “Accurate radar needs wide bandwidth but has poor spectral efficiency,” replied Schroeder. “We need to work on using slices of spectrum for more than one purpose.” We’re beginning to see that with dynamic spectrum sharing between 5G and LTE, but that must expand to other forms of spectral use.
While opinions have converged over terahertz frequencies, the future of MIMO brings about a divergence of opinions. “MIMO and beamforming must evolve,” said Schroeder. Why? Because of power. Higher frequencies mean smaller antennas can fit in a given space. That results in tighter beams but increased energy consumption. Schroeder argues that we need to figure out MIMO/beamforming from a power perspective. Furthermore, the wireless industry needs more research into circuit size, spacing, interference, and so on.
Cooperative MIMO could help. Currently, base stations operate independently. A device connects to one and only one base station at a time. With cooperative MIMO, base stations could share connections. Thus, just enough signal power coming from more than one base station could reduce overall energy consumption. Different power signatures among base stations could reduce energy consumption.
Artificial intelligence and machine learning will have applications throughout telecom, from the radio to the fiber and copper network. AI has proven itself useful for calculating coefficients for RF physical-layer signal processing to optimize the signal chain. Emerging use cases in the network are gaining momentum, though Schroeder noted that “People use the terms AL and ML as a way to get funding for their research.”
“Machine learning could turn test and measurement on its head,” added Hall. “In traditional test, you design f(x), put in a stimulus, and measure the result. If it meets specs, it passes the test.” With machine-learned systems, you put data in but don’t describe f(x). You achieve a reasonable result by adding more data. While noting that automotive engineers test using scenarios, not specs, Hall asked, “What conditions do you put in to know of the result is good?”
If engineers use AI/ML to test one part of the network, say a radio, how do you see the downstream implications of and ML-trained radio? The test industry has not yet to figure out the downstream implications.
If the user doesn’t complain, then the product must work. Sounds simple, but what do you when you must pass industry or regulatory requirements? How does ML help you? That’s one of the unknowns about using ML-trained systems. Hall asked, “What does your data sheet look like if you used AI/ML to develop your product?” If a data sheet is a contract, then what product or network characteristics do you use to select a network technology and prove if the product works properly?
We don’t yet know.