IBM device paves the way for 5G to reach full potential
5G, the next evolution of wireless communication standards, is already here – but it’s not possible to use it all the time just yet. A portable device and software stack developed by IBM that works with the millimeter-wave band of 5G can change that. The development is a huge step towards enabling 5G-smartphone users to continuously enjoy ultra-high data rates of 1.5 Gbps, and telecommunication service providers to have a much greater overall network throughput.
To have 5G today, it’s not enough to just buy a 5G-compatible mobile device. Your provider must support the new standard, and you have to be in the parts of those cities that have 5G coverage. And even if you do all that, as soon as you turn a corner, chances are you’ll lose it again – and you’re back to slower 4G.
As coverage expands, 5G-enabled IoT and personal devices will crowd urban spaces. This may result in network congestion due to the interference between devices. To enable multiple connections to coexist, earlier wireless technologies separated connections into two domains: time and frequency.
But millimeter-wave 5G allows separation of wireless links in a third domain – space – enabling the protection, or ‘shading’ of a device from unwanted signals (even those present at the same time and at the same frequency). Making the most out of this third domain is key to realizing the full throughput potential of 5G. This is why we have developed a programmable spatial filter to block all but one of the in-band signals that permeate our wireless space, thus steadily keeping the desired 5G connection.
Looking for a candle amidst bright city lights
Transmitting and receiving radio waves in all directions is how cellular communication has always worked, from 1G of the 1980s to today’s much faster 4G. Cellphones and base stations communicate much as light bulbs provide illumination– radiating all around, with no or limited directional capabilities. A mobile phone transmits wireless signals everywhere, with the cellular base station catching just a fraction of the energy the phone sends. And when the phone is receiving, the base station radiates in all directions too, to your entire neighborhood. The result: you receive just a tiny bit of that radiation, and everyone around you receives unwanted interference. It’s like talking in a crowded bar – there’s no way to communicate with only one person without others around hearing you.
Millimeter-wave 5G technology, though, is different. 5G-enabled devices and towers send energy in a specific direction, like a flashlight directs light. And it’s possible to control the beam electronically, with algorithms pointing it exactly where it needs to point. When receiving, 5G phones should be able to block out unwanted signals, electronically shading the device – a process called nulling.
In 2017, together with Ericsson, we developed award-winning chips and antenna-in-package designs (28-GHz phased array antenna module, an industry first) – 5G phased array hardware that produces beams and shades. Ericsson has since integrated the system in its latest base stations, currently on the market. And Verizon is deploying such directional base stations in the US.
Still, these systems don’t block the unwanted interference perfectly. It’s similar to looking up and shading your face, say, only on the left side. There are typically just a few dozens of different options for beam directions or shading settings to sift through to pick the right one – typically 32 or 64 – not enough to achieve a perfect result. It’s like being in the city at night and looking for a candle far away amidst all the bright lights nearby. You’d have to go through about 10180 options to account for every single shading scenario – more than the number of atoms in the universe.
IBM SDPAR to the rescue
We wanted to enable 5G research to go further. Enter SDPAR – a software defined phased array radio, a small portable device that can emulate a 5G-enabled base station or smartphone for research purposes. It weighs just 2kg, consumes less than 100W and can be powered with a small 12V adapter similar to those used for laptops. The device is easy to reconfigure and able to work with more than 10180 beam settings –moving the beam in thousands of directions in a fraction of a second. It can also evaluate link quality in each of these directions, and automatically choose beam settings that provide good connectivity. It needs only two data connections, for beam control and data input/output, and can be operated from a standard laptop.
We have shown that it’s possible to create a very large number of shading options with ease through a high-level application program interface (API). SDPAR enables the development and real-world evaluation of algorithms to navigate through this vast configuration space. An example is shown in the following video: in a matter of seconds, the device determines the best directions to shade, rapidly creating specific shading and brightening of areas.
The device can be used for other tests in future, for example to explore and develop 5G mm-wave beamforming and beam-steering algorithms and hybrid beamforming algorithms (paper # Mo4A-2 in https://rfic-ieee.org/technical-program/technical-sessions?date=2020-06-22), perform 28-GHz over the air testing as well as 28-GHz channel characterization, and develop and test custom digital basebands. And it’s possible to use SDPAR beyond 5G, too – for example, to explore and develop radar algorithms for sensing and imaging (paper # Mo3A-1 in https://rfic-ieee.org/technical-program/technical-sessions?date=2020-06-22).
But looking in one direction and trying to shade everything else out is only one way of configuring the array. We can shade better, in specific directions where bright interferers are located. The same applies to forming a beam. SDPAR can access 10180 different settings for beams and shades, creating complex shading patterns – to navigate this vast catalog of choices, we aim to use machine learning-based algorithms to quickly learn a scene and arrive at the best solution. Such AI-assisted beamforming techniques will enable us to take another leap toward fully harnessing the power of millimeter wave 5G – and allow smartphone users to consistently enjoy extremely fast data rates.
The new algorithms and AI-based techniques for 5G that can be developed with SDPAR are key examples of the emerging era of ‘acting on data at the source’ enabled by IBM’s solutions for 5G and edge computing.
We set up an experiment at the IBM T. J. Watson Research Center to show how SDPAR and beamforming algorithms can overcome communication impediments in an interference-limited scenario.
We use three SDPARs. one configured as transmitter, another one as a receiver, and a third one as an interferer. First, the interferer was turned off so that we could measure received energy (Fig. 1, left) and link quality (Fig. 1, right) for multiple choices of receiver beam direction.
We later set up the interferer to shine a bright signal towards the receiver, making it difficult for the receiver to decode the transmitted data. We found that if we shade all directions but one, the energy comes primarily from the left of the center where the transmitter is radiating – shown by the region of brighter dots that represent higher received energy in Fig. 1. The link quality in different receiver pointing directions is shown on the right side of Fig. 1. The link was good not only when the receiver was pointed in the direction of the radiating transmitter but also when pointed in other directions; the receiver was able to decode the data from the signal bouncing off the walls and ceiling and reaching the receiver. In this test, we only looked in 70 directions to emulate how the best 5G solutions today might work.
If we turn on the interferer, the situation changes. We now see energy (Fig. 2, left) coming both from the transmitter located towards the left and from the interferer located towards the right. The interferer is a lot stronger than the transmitter and overpowers the receiver. This is clear when we look at the link quality (Fig. 2, right), which is poor in all 70 directions that the SDPAR was configured to try in this case. Unfortunately, current 5G solutions are still vulnerable to some of the interference challenges they were designed to overcome.
Next, we relied on the large control space of our SDPAR to evaluate the link quality in thousands of directions and try to see if there are some directions that allow us to form a link even in the presence of a strong interferer. The results for more than 20,000 directions are shown in Fig. 3.
On the right side of Fig. 3, one can see a few directions where we are able to achieve good link quality. These directions were completely missed earlier in the search through only 70 catalog options.
Is it possible to find the best direction without having to search through every single option? Our early algorithms look promising (Fig. 4) – we use an optimizer to find a direction, among tens of thousands, that achieves a good link by looking in as few directions as possible.