This post is also available in:
עברית (Hebrew)
As wireless networks carry ever-growing amounts of data, communication systems face increasing pressure to transmit information faster while consuming less power and bandwidth. This challenge is particularly important for satellites, drones and other mobile platforms, where limited energy reserves and constrained communication links can restrict the amount of data that can be exchanged.
Researchers at Cornell University have developed a new method that could address both challenges using a microwave neural network. Building on a previously demonstrated chip capable of processing microwave signals directly, the team has now shown that the same hardware can encode information into a unique microwave “language”, enabling more efficient data compression and a new approach to hardware-based communication security.
Unlike conventional wireless systems, which typically convert analog radio signals into digital data before processing them, the microwave neural network performs computation directly in the microwave domain. According to TechXplore, the chip exploits the natural behavior of microwave signals to process information, reducing the amount of computation required by conventional digital electronics while operating with very low power consumption.
The researchers describe the new communication method as a form of microwave token embedding. Similar in concept to the tokens used by large language models, the chip transforms information into distinctive microwave pulse patterns that preserve relationships between different pieces of data. Instead of transmitting long sequences of digital instructions, a device can communicate using only a small number of carefully generated microwave pulses that another compatible microwave neural network can immediately interpret.
Because every microwave neural network has slightly different physical characteristics, the communication method also introduces a degree of built-in security. Successfully decoding a transmission requires another appropriately configured microwave neural network using the correct configuration sequence, making unauthorized interpretation significantly more difficult. The researchers compare this concept to a hardware-based public-private key mechanism.
The team also demonstrated the chip’s ability to compress data. In one experiment, they reconstructed a satellite image of a tropical storm after reducing the transmitted data by approximately eight times, while preserving many of the image’s most important features. This could help satellites transmit useful imagery even when communication bandwidth is severely limited.
Although developed for future communications networks, the technology also has important defense applications. Military drones, satellites and tactical communication systems frequently operate in contested environments where bandwidth is limited and secure communications are essential. Compressing sensor data before transmission while embedding it in hardware-specific microwave signatures could improve resilience against interception while allowing more information to be exchanged using the same communication link.
The researchers believe the technology could eventually support 6G wireless networks, direct-to-satellite communications and autonomous platforms that require rapid, low-power processing. By allowing the physics of microwave signals to perform part of the computation itself, the new approach offers a different path toward faster, more efficient and inherently more secure wireless communications.
The research was published here.
