The Impact of Artificial Intelligence in Massive MIMO Technology

Massive MIMO (multiple-input and multiple-output) is a well-known technology that has been developed and deployed to cope with the expected high data traffic demand, leveraging existing cell sites (where possible) and the large bandwidths allocated to 5G in c-band and mmWave. In spite of all the advantages that the technology brings for Mobile Network Operators (MNOs), such as enhanced spectral efficiency and better end user data throughputs, there are technical challenges that have driven further investment in Research and Development (R&D) in order to improve the performance of the solution along with innovation in physical features that lead to ease the cell site installation. ABI Research summarizes some of the technical challenges as follows:

• Pilot Contamination occurs if the same set of orthogonal pilots are used by different User Equipment (UE) in both the neighbor and home cells. The base station uses the uplink pilot signals from the UE to estimate the downlink channel. Thus, if there are orthogonal pilots from different cells, the base station will beamform towards the user in its home cell and the undesired user in the neighbor cell. This interference scenario can limit the achievable throughout.
• Signal Detection is a computationally complex mechanism in the uplink channel due to the large number of antennas of the massive MIMO radios, as well as all UEs (within the cell), transmitting signals through different wireless paths. These signals are superimposed at the base station creating interference.
• Channel Estimation is a mechanism utilized by the massive MIMO solutions to beamform downlink data towards the end user, which relies on the Channel State Information (CSI). The CSI is the information of the communication link from the UE to the base station and contains the effects of scattering and fading, among others, which is estimated by the base station based on the pilot signals sent by the UE in the uplink channel. Channel estimation becomes challenging due to the effects of pilot contamination as well as systems using Frequency Division Duplex (FDD). In FDD systems, the CSI is estimated in both uplink and downlink, and the UE has to acknowledge the base station with the estimated channel information for the downlink transmission; thus, increasing the complexity of channel estimation mechanism.
• Hardware Impairment occurs due to the use of low-cost components that increase the effects of hardware imperfection such as phase noise, magnetization, and amplifier distortion. In addition, due to the large number of antennas in massive MIMO, there is mutual coupling between the antenna elements that change the load impedance and cause distortion.

Many technical challenges of the massive MIMO platforms have been addressed by the industry via the implementation of different techniques and mechanisms for optimal performance. However, advanced technologies, such as Artificial Intelligence (AI), for massive MIMO have the potential to further address the aforementioned complex challenges, saving a considerable amount of computational power while enhancing the overall performance of the massive MIMO solutions.

In fact, different vendors have started to implement AI algorithms in the massive MIMO platforms. ABI Research provides some examples below:

• Huawei has introduced an algorithm called Adaptive High Resolution (AHR) that relays in three fundamental pillars: High Accuracy Channel Estimation (HACE), UCS (User & Channel & Service) adaptive algorithm, and High-Resolution Beams (HRB). During the HAS 2021 Global Analyst Summit, Huawei confirmed that the solution implements AI algorithms to predict the channel status of the specific UE and increase the beam accuracy. According to Huawei, this technology is already implemented in different carriers’ networks and can increase 1.6x user experience and approximately 2x cell capacity.
• ZTE has introduced the Automatic Antenna Pattern Control (AAPC) self-optimization solution that utilizes AI to simplify the optimization and Operation and Management (O&M) of the 5G Massive MIMO solutions searching for the optimal antenna parameters. The solution has been trialed with China Mobile achieving nearly 12% increase network coverage and higher download rates by approximately 10%.
• Samsung has introduced the Mobility Enhancer AI feature along with advanced signal processing technology to enhance the accuracy of the beamforming. The feature – which according to Samsung is a software upgrade only – is expected to be launched in 2021 and it is likely to improve the network throughput by 30%.

In addition, the implementation of AI algorithms has gone beyond the enhancement of system throughput and spectral efficiency. AI algorithms are also being implemented to further enhance the energy consumption of the 5G networks. For example, AI algorithms are being implemented to recognize and separate the user type of traffic diverting short bandwidth-based traffic (i.e., web browser, WhatsApp) to the 4G network and high bandwidth data applications (i.e., video) to the 5G network. Thus, the 5G network can be set to idle and sleep mode if there is no 5G traffic demand, further reducing power consumption of the overall 5G network.

Massive MIMO solutions are transitioning from being recognized as an enabling technology to a mature technical solution with mass market adoption due to the acceleration in the roll out of 5G across different regions. Vendors are investing very heavily in R&D, dedicating huge effort in addressing pressing performance drivers for MNOs, including spectral efficiency, enhanced user throughput, energy saving mechanisms, as well as simplifying the cell site installation.

ABI Research expects that the performance and energy saving mechanism of the massive MIMO platforms will continue to be enhanced due to the implementation of AI algorithms in future massive MIMO generations. Mechanisms such as channel estimation, along with hardware distortion, uplink signal detection, and beam alignment have the potential to be further improved by using AI while reducing system complexity of known algorithms. In addition, AI has a huge potential to also become one of the mainstream pillars in 5G and in the development of future wireless communications that will require the imminent transition from a mathematical model-dependent optimization to data-dependent deep learning.