The 3GPP standard for mobile communications only specifies the basic functionalities needed to make a device from any vendor (e.g., Apple, Samsung, Xiaomi) communicate with network infrastructure from any vendor (e.g., Ericsson, Nokia, Huawei). The exact implementation details (e.g., hardware and algorithms) can be selected freely—a setup selected to promote innovation and competition between vendors. Since the internal interfaces (blue cables in the picture) between radio units, lower-layer processor units, and higher-layer processor units are unspecified, a telecom operator must buy its entire radio access network (RAN) from a single vendor.

Single-vendor networks have several inherent benefits: jointly optimized hardware and software, ease of deployment, and clarity of who to blame when something fails. However, it also leads to vendor lock-in, where the operator has little bargaining power when buying new equipment and cannot benefit from new innovations made by other vendors.

Motivation for Open RAN

Today, there are only a few vendors capable of delivering a complete network, and recent geopolitical developments have reduced the viable options in each country even further. To create more options and market competition, the O-RAN Alliance has since 2018 worked to define what components a network must contain (e.g., radio unit, distributed unit, centralized unit) and specify open interfaces between them (e.g., fronthaul, midhaul, and APIs). This is called the Open RAN specification, and it is meant to enable new companies to enter the market and specialize in offering a limited number of components. If successful, this will allow operators to mix-and-match hardware and software from multiple vendors in the same network.

There has been much hype about Open RAN and the innovations it might bring over the past few years. Several people have even tried to convince me that future networks cannot work without it. Nevertheless, six years into the Open RAN era, there are barely any multi-vendor networks, and many people were surprised when AT&T decided to build a new nationwide Open RAN network that only consists of $14 billion of Ericsson equipment. This has been called “Single-vendor Open RAN” and has led to claims from independent experts that Open RAN is failing miserably [see: Light Reading, May 2024; telecoms.com, Aug 2024].

Success criteria for Open RAN

I think the Open RAN development mostly goes as intended, even if promises made in various press releases are yet to be fulfilled. The RAN technology is highly complex, so one cannot expect new vendors to be able to compete with the traditional giants immediately. A mix-and-match network using hardware and software from new vendors will likely not provide competitive performance any time soon. This is reminiscent of how Apple has tried for years to build high-performing 5G baseband chips for their phones but is still far behind Qualcomm.

The potential success of Open RAN should not be measured by the number of new multi-vendor networks being deployed in the next few years but by how the increased competition and intelligent network virtualization lead to cost reductions for the operators—both when deploying new networks and later when network components must be replaced due to hardware failures or upgraded with new functionalities. It will naturally take time for these potential benefits to materialize because most operators have already deployed traditional 4G and 5G networks and will not invest in new ”greenfield” deployments anytime soon. Perhaps the first large-scale multi-vendor deployments with hardware from untraditional vendors will take place in geographical regions currently lacking 5G connectivity and where cost savings are more important than top-of-the-line performance. In many other places, I believe Open RAN will not be a commercial reality until 6G networks are rolled out.

The O-RAN Alliance identifies four benefits of Open RAN: openness, interoperability, virtualization, and AI automation. Single-vendor Open RAN compliant networks will use the latter two benefits from day one to reduce hardware/operational costs and enable new services. The operators might also benefit from the former two in the long run, particularly for components that become commodified. Virtualization and AI automation are, of course, features that a state-of-the-art closed RAN deployment also supports—they are not unique features for Open RAN but have been researched under the “Cloud RAN” name for a long period. However, AT&T has demonstrated that there is little incentive to build new networks in the traditional way when one can get openness and interoperability as well.

In conclusion, Open RAN is successful in the sense of being a natural feature in any new network deployment. However, the hyped interface openness and multi-vendor support are not the transformative aspects, and we cannot expect a major uptake until greenfield 6G deployment commences.

For the last five years, most of the research into wireless communications has been motivated by its potential role in 6G. As standardization efforts begin in 3GPP this year, we will discover which of the so-called “6G enablers” has attracted the industry’s attention. Probably, only a few new technology components will be introduced in the first 6G networks in 2029, and a few more will be gradually introduced over the next decade. In this post, I will provide some brief predictions on what to expect.

6G Frequency Band

In December 2023, the World Radiocommunication Conference identified three potential 6G frequency bands, which will be analyzed until the next conference in 2027 (see the image below). Hence, the first 6G networks will operate in one of these bands if nothing unforeseen happens. The most interesting new band is around 7.8 GHz, where 650-1275 MHz of spectrum might become available, depending on the country.

This band belongs to the upper mid-band, the previously overlooked range between the sub-7 GHz and the mmWave bands considered in 5G. It has recently been called the golden band since it offers more spectrum than current 5G networks in the 3.5 GHz band and much better propagation conditions than at mmWave frequencies. But honestly, the new spectrum availability is quite underwhelming: It is around twice the amount that 5G networks will already be using in 2029. Hence, we cannot expect any large capacity boost just from introducing these new bands.

Gigantic MIMO arrays

However, the new frequency bands will enable us to deploy many more antenna elements in the same form factor as current antenna arrays. Since the arrays are two-dimensional, the number grows quadratically with the carrier frequency. Hence, we can expect five times as many antennas in the 7.8 GHz band as at 3.5 GHz, which enables spatial multiplexing of five times more data. When combined with twice the amount of spectrum, we can reach an order of magnitude (10×) higher capacity in the first 6G networks than in 5G.

Since the 5G antenna technology is called “Massive MIMO” and 6G will utilize much larger antenna numbers, we need to find a new adjective. I think “Gigantic MIMO“, abbreviated as gMIMO, is a suitable term. The image below illustrates a 0.5 × 0.5 m array with 25 × 25 antenna elements in the 7.8 GHz band. Since practical base stations often have antenna numbers being a power of two, it is likely we will see at least 512 antenna elements in 6G.

During the last few months, my postdocs and I have looked into what the gMIMO technology could realistically achieve in 6G. We have written a magazine-style paper to discuss the upper mid-band in detail, describe how to reach the envisioned 6G performance targets, and explain what deployment practices are needed to utilize the near-field beamfocusing phenomenon for precise communication, localization, and sensing. We also identify five open research challenges, which we recommend you look into if you want to impact the actual 6G standardization and development.

You can download the paper here: Emil Björnson, Ferdi Kara, Nikolaos Kolomvakis, Alva Kosasih, Parisa Ramezani, and Murat Babek Salman, “Enabling 6G Performance in the Upper Mid-Band by Transitioning From Massive to Gigantic MIMO,” arXiv:2407.05630.

I started analyzing reconfigurable intelligent surfaces (RISs) in 2019, just when the hype began in the communication community. I was mostly interested in understanding the fundamentals of this technology because some claims that I found in papers were too good to be true. In 2020, we summarized our thoughts in the paper “Reconfigurable Intelligent Surfaces: Three Myths and Two Critical Questions,” which corrected three misconceptions that flourished then.

More importantly, we identified two open research problems that we found particularly important to solve if the RIS technology should become practically and commercially viable.

The first question was: What is a Convincing Use Case for RIS? The literature was already full of potential use cases; take any communication scenario and add an RIS, and you will improve the system performance by a little. However, the proposed use cases were either too niche-oriented to motivate technology development or aimed at “solving” problems that existing technology (e.g., relays or small cells) could already manage to some extent. A groundbreaking use case of clear commercial value was missing, and I think we have yet to find it.

The second question was: How can we estimate channels and control a RIS in real time? I was skeptical that it would be possible to solve this problem until I saw a demo video from the University of Surrey in late 2020 and participated in a measurement campaign at the Huazhong University of Science and Technology. I then decided to look for a solution together with my research group at KTH.

Real-time reconfigurability is challenging because a typical RIS has hundreds of elements; thus, there are hundreds of channel parameters to estimate every time the channel changes. A plausible solution is to narrow the scope to use cases where the channels can be parametrized efficiently. If the propagation geometry is such that the channel is determined by the same few parameters, regardless of how many RIS elements there are, we can focus on estimating those parameters instead. We realized that array signal processing contains the necessary tools for developing such geometry-based models and algorithms.

We now have a solid solution framework. In the following video, I explain our approach, from the first principles to the solution. The slides were originally prepared for a keynote speech at IEEE CAMSAP 2023 and then fine-tuned for my recent keynote at IEEE SITB 2024.

The technical details are found in the paper “Parametric Channel Estimation with Short Pilots in RIS-Assisted Near- and Far-Field Communications,” which will appear in the IEEE Transactions on Wireless Communications in 2024.