x-axis. As a user is walking around the room with their Bluetooth headset on and enjoying music streaming to their headset, packets are being continuously transmitted from their phone, Alexa or similar, to their headset. If there is a large amount of sporadic packet loss, the music experience is not very enjoyable. This is what happens in many instances with standard Bluetooth as shown by the area under the pink on the graph on the left, which shows that the rate at which packets are received varies widely as the user walks around. With ML enabled—the area under the green line—the packet reception rate is constant, thus maintaining music quality. 

In addition, as the right part of the graph in Figure 5 (see page 2) shows, the total number of rounds that are consumed by the link to deliver packets across is much higher for Bluetooth without ML (pink) as opposed to Bluetooth with ML (green).

When the left and right side are considered together, this shows that without ML, not only are fewer packets received, but also that there are more total retransmissions needed. This is a classic “double whammy” that ML helps alleviate.

Not only can efficiency be improved, but the ML algorithms also learn on their own to arrive at the best performance under different channel conditions and evolving interference environments. For example, when WiFi and Bluetooth are active in the same room, they interfere with each other because they both operate in the same 2.4GHz band in many cases. ML has been demonstrated to reduce the impact of such interference by a factor of two to four times so that both WiFi and Bluetooth can continue to maintain good performance.

Energy per bit of information sent is another important consideration for a wireless system, as most wireless devices are battery-powered. Some Bluetooth devices like hearing aids are even more power-constrained compared to phones or headsets. For such devices, energy efficiency is critical. As a hearing aid or earbud gets closer to its range limit, as when a user walks a few feet away from their phone, the energy required to transmit one bit of information on Bluetooth may increase from one to two nano Joules to 100nJ or more. ML-based systems have been shown to extend the range where Bluetooth continues to operate power efficiently by two to four times.

Machine learning improves cybersecurity

Machine learning also helps in keeping wireless links more secure. Because today’s wireless protocols re-transmit packets that are not received, these retransmissions increase the probability that an intruder can sniff the packet. Once the packet is sniffed correctly, the intruder has access to ciphertext, which can then be used for sophisticated crypto-analysis. ML-based feedback systems provide better ciphertext security when compared to current systems. Unlike current systems, ML-based multi-bit feedback never re-transmits packets, which reduces the probability of sniffing a ciphertext. This reduces the probability of crypto-analysis.

Exploration continues on the function of multi-bit feedback in multiple applications. Its technology works especially when there is power asymmetry between the transmit and receive directions, for example when the transmitter is more power-constrained than the receiver. This power asymmetry often arises in cellular, WiFi and satellite communications. While these transports are more sophisticated than Bluetooth and have advanced methodologies and protocols to combat packet loss (such as modern codes, HARQ, block ACKs, and more), they will still benefit from the efficiency improvement provided by the AI/ML-based technology.

Bluetooth is merely the most obvious entry point for this technology. New applications will also benefit from the enhanced security benefits. These innovative developments are indispensable in an increasingly machine-to-machine world.