Dynamic Radio Regulation with A.I.-Assisted Spectrum Etiquette

What is the Challenge

National regulatory bodies, for example the Federal Communications Commission (FCC), coordinate commercial radio spectrum usage and the regulation of radio emissions. The radio spectrum is divided into frequency bands, and licenses for the use of frequency bands are given to -among others- telecom operators. With licensed frequency bands, operators are provided with a long-term and exclusive right to use the radio resources of the assigned bands. Not always will a band be fully used, and with an increasing number of licenses and increasing number of dedicated frequency bands, this leads to a high percentage of radio spectrum being scarcely used - or not at all.

A common alternative to licensing is to provide non-exclusive access permissions to certain bands without such a license. Bands made available like this are known as license-exempt or unlicensed frequency bands. Because of their general availability, unlicensed frequency bands are used by many commercial and short-range wireless communication systems. Wi-Fi, Bluetooth, and ZigBee are popular examples.

However, such radio systems are often not designed to coordinate spectrum with dissimilar systems, and as result, they might interfere harmfully with each other when used in close proximity.

Spectrum Etiquette

In an ideal scenario, spectrum etiquette rules are followed by all radio systems that operate in an unlicensed band and share the spectrum. The rules help reaching fair access to the radio resources and a more efficient use of the radio spectrum.

Selecting the right parameters can be location-, device-, and scenario-dependent. Algorithms make decisions based on the observation of various factors in their environment. A decision making algorithm takes specific characteristics of the involved radio systems into account, including coverage ranges, the number of active devices, required data rates, mobility patterns. Default sharing rules are defined by regulation. Additional factors to consider are related to the behavior of other systems and the availability of certain frequencies. The algorithm uses all this information to adjust its operational parameters in real-time, such as frequency channel, transmission power level, and modulation and coding scheme.

The goal of a spectrum etiquette is always to maximize the efficiency of the radio spectrum usage, while ensuring fair access and minimizing mutually harmful interference.

Any sharing rule to address how unlicensed devices should behave should promote greater coexistence and spectrum efficiency in new unlicensed bands. The application of such rules must allow spectrum in new unlicensed bands to be used more intensively with a higher quality of service than is possible in the current unlicensed bands, without limiting innovation. The constraints to be kept in mind when developing sharing rules are:

• A sharing rule must not favour a specific radio standard, limit the innovation potential of new generations of technology, or give an advantage to certain services

• Minimalist sharing rules should have demonstrable benefits

• Sharing rules should not increase costs or complexity

Example Sharing Rule

The following example assumes a listen-before-talk medium access and a different channel bandwidth for the two different systems involved. The example rule is called “Frequency Channel Clustering “, and described as follows: “When selecting a frequency channel for operation, a radio device should select a frequency channel next to channels that are already used by similar radio systems.”

The objective of this sharing rule is to maximize the availability of spectrum opportunities for other, competing radio devices. This rule is illustrated in the figure below. The benefit of the rule is clearly visible. It will be helpful to the broadband device(s) if narrowband devices would select frequency channels next to each other, as indicated in the figure.

Use of Machine Learning

Devices that operate according to different radio standards with non-similar modulation- and coding schemes or protocols, are not directly able to negotiate or coordinate the access to the radio spectrum. Predictive analytics and pattern recognition are therefore often considered to be used for anticipating future spectrum allocations from other devices. This would give a device the means to react to the anticipated allocations, and select resources accordingly. The three main approaches to learn from past spectrum allocations are reinforcement learning, supervised learning, and unsupervised learning. They differ based on how data is presented to the device and how the device learns to make predictions or decisions.

In supervised learning, a decision-making algorithm is trained on a labeled dataset, which means that each training example is paired with an output label. The goal is to learn a mapping from inputs to outputs, allowing the model to make predictions on new, unseen data based on this learned mapping.

Unlike supervised learning, unsupervised learning involves training the algorithm on a dataset without any labels. The goal here is to discover underlying patterns, structures, or distributions in the data without any prior knowledge of outcomes.

Reinforcement Learning is fundamentally different from both supervised and unsupervised learning. In reinforcement learning, an agent learns to make decisions by performing actions in an environment to achieve some goal. The agent receives feedback through rewards or punishments and learns over time to maximize these rewards.

A.I. Assistance

In a new approach with the help of Artificial Intelligence (A.I.), rules can be reasoned about, they might evolve over time, and they can be proposed and tested by the interacting radio devices. Reasoning engines assisted by Artificial Intelligence can use a simple logic and might further improve the efficiency of the original etiquette rules published by the regulator. Iteration by iteration, the spectrum will then hopefully be used more efficiently.

Models derived from the theory of non-cooperative multi-stage games provide an underlying framework for the reasoning engines. They emulate the interaction and competition for spectrum, while each rational player (each interacting device) aims to maximize their own utility.

Back to: About Lovefield – Overview