MIT’s Photonic Processor Promises to Revolutionize 6G Wireless Signal Processing

In an era where the proliferation of connected devices is driving an insatiable demand for bandwidth, managing the wireless spectrum efficiently is becoming paramount. As tasks like teleworking and cloud computing become increasingly reliant on seamless connectivity, the finite nature of available wireless resources presents a significant challenge.

To address this challenge, engineers are turning to artificial intelligence (AI) to dynamically optimize the use of the wireless spectrum, aiming to reduce latency and enhance overall performance. However, many current AI-driven methods for classifying and processing wireless signals are energy-intensive and struggle to operate in real-time, hindering their practical application.

Now, researchers at MIT have unveiled a groundbreaking AI hardware accelerator specifically tailored for wireless signal processing. This innovative optical processor harnesses the speed of light to perform machine-learning computations, enabling the classification of wireless signals in mere nanoseconds.

The photonic chip achieves speeds approximately 100 times faster than the most advanced digital alternatives, while maintaining a classification accuracy of around 95 percent. Furthermore, this novel hardware accelerator boasts scalability and flexibility, making it suitable for a wide array of high-performance computing applications. In addition to its performance advantages, the chip is smaller, lighter, more cost-effective, and more energy-efficient than traditional digital AI hardware accelerators.

This device holds particular promise for future 6G wireless applications, such as cognitive radios that dynamically adjust data rates by adapting wireless modulation formats to the ever-changing wireless environment.

By empowering edge devices with real-time deep-learning capabilities, this hardware accelerator could unlock significant speed improvements in various applications beyond signal processing. Examples include enabling autonomous vehicles to react instantaneously to environmental changes and facilitating smart pacemakers that continuously monitor a patient’s heart health.

According to Dirk Englund, a professor at MIT’s Department of Electrical Engineering and Computer Science and the senior author of the paper published in Science Advances, this innovation could revolutionize real-time and reliable AI inference. The research team, which includes lead author Ronald Davis III PhD ’24, Zaijun Chen, and Ryan Hamerly, has demonstrated a light-speed processing capability that overcomes the limitations of conventional digital AI accelerators.

The researchers developed an optical neural network architecture specifically for signal processing, called a multiplicative analog frequency transform optical neural network (MAFT-ONN). The MAFT-ONN addresses the problem of scalability by encoding all signal data and performing all machine-learning operations within the frequency domain — before the wireless signals are digitized.

In simulations, the optical neural network achieved 85 percent accuracy in a single shot, which can quickly converge to more than 99 percent accuracy using multiple measurements. MAFT-ONN only required about 120 nanoseconds to perform entire process.