AI techniques can overcome nonlinear problems by extracting patterns automatically and efficiently.
Wireless system design complexity keeps increasing, from mobile wireless technology moving from 3G and 4G to the expansive use cases of 5G, and the introduction of Industry 4.0. Driven by the need to optimally manage the sharing of valuable resources to an expanding set of users, a growing number of engineers are turning to artificial intelligence (AI) to solve the challenges introduced by modern systems.
From optimizing call performance through resource allocation to managing vehicle-to-vehicle (V2V) and vehicle-to-everything (V2X) communication between autonomous cars, AI has brought the sophistication necessary for today’s modern wireless applications. As the number and capabilities of those devices connected to networks expands, so too will the role of AI in wireless. For the future success of the technology, engineers should be aware of the key benefits and applications of AI in wireless systems today, as well as the best practices necessary for optimal implementation.
Drivers for AI in wireless
Three distinct use cases have defined mobile networks’ transition to 5G and have acted as the driving forces in engineers’ adoption of AI. These include:
- The optimization of speed and quality of mobile broadband networks,
- The need for ultra-reliable low rates, and
- Massive machine-type communication for time-sensitive connections between Industry 4.0 devices.
An expanding set of devices competing for the resources of the same network and an ever-increasing pool of users also leads to the increasing complexity of wireless systems. Formerly linear patterns of designs once understood by human-based rules and human processing of data are no longer sufficient. AI techniques, however, can overcome non-linear problems by extracting patterns automatically and efficiently. These techniques can do so beyond the ability of human-based approaches.
Integrating AI in a wireless environment enables machine learning and deep learning systems to recognize patterns within communications channels. These systems then optimize the resources given to that link to improve performance. As applications of a modern network compete for the same resources without the use of AI methodologies, managing these networks becomes a nearly impossible task.
The sophistication of AI also enables more efficient project management such as reduced order modeling. By incorporating simulated environments into an algorithmic model by estimating the behavior of source environments, engineers can quickly study a system’s dominant effect using minimal computational resources. Additional benefits to the use of AI in this context can include more time to explore design and carry out more iterations faster, cutting time in production cycles and associated costs.
Data quality is vital for the successful and effective deployment of AI. AI models need to be trained with a comprehensive range of data to adequately deal with real-world scenarios. Applications provide the data variability necessary for 5G network designers to train AI robustly by synthesizing new data based on primitives or by extracting them from over-the-air signals. Failure to explore a large training data set and iterate on different algorithms based on that data could result in a narrow local optimization instead of an overall global one, compromising the reliability of AI in real-world scenarios.
A robust approach to testing AI models in the field is similarly critical to success. If signals to test AI are captured only in narrow and localized geography, the lack of variability in that training data may negatively impact how an engineer may approach and optimize their system design. Without comprehensive field iterations, the parameters of individual cases cannot be used to optimize AI for specific locations, adversely impacting call performance.
Smart homes to autonomous vehicles
Digital transformation has been embraced in industries across the spectrum, from telecommunications to automotive. This increased adoption of transformation has necessitated the wide-scale adoption of AI and is the primary driver for its application.
Placing electronic communications in areas once mechanical orientated generates large amounts of data as applications that include smart homes, telecommunication networks and autonomous vehicles (AV) connect. The large quantities of data generated by these applications facilitate the development of future-looking AI techniques to accelerate the process of digital transformation, yet also stretch the resources of the joining network.
In telecommunications, AI is deployed at two levels — at the physical layer (PHY) and above PHY. The application of AI for improving call performance between two users is referred to as operating at PHY. Applications of AI techniques to physical layers include digital pre-distortion, channel estimation, and channel resource optimization. Additional applications include autoencoder design which spans automatic adjustments to transceiver parameters during a call. Figure 1 shows a roadmap for using data and AI to train models.
Channel optimization is the enhancement of the connection between two devices, principally network infrastructure and user equipment like handsets. Using AI helps to overcome signal variability in localized environments through processes such as fingerprinting and channel state information compression.
With fingerprinting, AI can optimize positioning and localization for wireless networks by mapping disruptions to propagation patterns in indoor environments, caused by individuals entering and disrupting the environment. AI then estimates, based on these individualized 5G signal variations, the position of the user. In so doing, AI overcomes traditional obstacles associated with localization methods using comparisons between received signal strength indication (RSSI) and the RSS in providers’ databases. Channel state information compression, on the other hand, is the use of AI to compress feedback data from user equipment to a base station. This ensures that the feedback loop informing the station’s attempt to improve call performance does not exceed the available bandwidth, leading to a dropped call.
Above-PHY uses are primarily in resource allocation and network management. As the number of users and use cases on the network exponentially increase, network designers are looking to AI techniques to respond to allocation demands in real-time. Applications such as beam management, spectrum allocation and scheduling function are used to optimize the management of a core system’s resources for the competing users and use cases of the network.
In the automotive industry, using AI for wireless connectivity makes safer autonomous driving possible. Autonomous vehicles and V2V/V2X vehicular communications rely on data from multiple sources, including LiDAR, radar, and wireless sensors, to interpret the environment. The hardware present in AVs must handle data from these competing sources to function effectively. AI enables sensor fusion (fusing competing signals to allow the vehicle’s software to make sense of its location and establish how it will interact with its environment by understanding omnidirectional messages).
This approach to communications allows the vehicle to establish a 360-degree field of “awareness” of other vehicles (Figure 2) and potential crash threats within its proximity. Whether through informing the driver for the vehicle or driving autonomously, the utilization of AI is leading to improved road traffic safety and reducing the number of crashes at intersections.
Expanding AI in wireless system design
As the use cases for wireless technology expand, so too does the need to implement AI within those systems. From
5G, to AV, to IoT, these applications would not have the sophistication necessary to function effectively without the use of AI. AI’s place in the engineering landscape, particularly wireless system design, has been growing exponentially in recent years and this pace of change can be expected to continue rising – and faster – as the use cases and the number of network users expand in the modern age.