By: Ravikiran Gopalan

We live increasingly in a “sensorized” world. Everything is connected to everything else. Connectivity is key to smart cars, smart cities, smart hospitals, smart armies, and more. As a result, connectivity reliability is crucial.

In addition, spectrum sharing by multiple technologies is making wireless connectivity more unreliable and more prone to dropouts. Traditional wireless receivers don't have the in-built intelligence to function in such crowded environments. Traditionally wireless communications have been built on a foundation of human-intuition-based behavior models and analytically tractable channel approximations. The real-life behaviors and channel conditions, however, can be quite different. While there is an abundance of data that is readily available and collectible about the real-life parameters, there is no existing mechanism by which the traditional wireless communications can be modified and optimized by using this real-life data

Machine learning improves wireless performance exponentially

Machine learning techniques, powered by real data, can be one such mechanism to provide radical improvements in wireless communications, like improvements made by ML in speech recognition and image detection. Not only can ML allow wireless receivers to improve reliability in the presence of wireless congestion, but there is also enormous potential to improve range, battery life, data rates and latency.

Bluetooth, WiFi, cellular, satellite communications and sensor networks are some of the types of wireless networks that can benefit from ML. Each of these are constantly expanding markets and there are already billions of connections that can be improved. In Bluetooth alone, there are billions of ear buds, headsets, car audio, hearing aids and many other types of endpoints already in use by consumers worldwide.

Improving wireless connectivity is of paramount importance, driving numerous startups to innovate and deliver solutions in this space. What has emerged is that by adding machine learning Bluetooth can be supercharged. An AI/ML-based feedback system quadruples Bluetooth range, doubles battery life, and multiplies reliability ten times over.

Below are a few examples of performance results from Aira Technologies. Consider music playing on a phone and streaming to your Bluetooth earbud or headset. Figure 1 below shows a representation of the Bluetooth audio across a few seconds when wireless link quality is good.

Figure 1: Bluetooth audio with good wireless link quality

Can you hear me now?

To understand a simple example of how ML has been harnessed to achieve such impressive results, it’s important to understand that wireless communication channels are inherently noisy and as a result, a large percentage of packets that are transmitted arrive with errors. Figure 2 shows a representation of the same Bluetooth audio stream when about half the packets are being lost.

Figure 2: Bluetooth audio with packet loss