A microwave chip encodes data into radio signals, reducing transmission needs and helping satellites and drones communicate with less bandwidth and power.
Researchers at Cornell University have developed a microwave-based computing chip that can encode information directly into radio signals, reducing the amount of data that needs to be transmitted while adding a layer of hardware-based security. The team says the technology could improve wireless communication in satellites, drones and other systems with limited bandwidth and power.
The researchers showed that the chip creates microwave token embeddings, which work in a similar way to the tokens used in large language models. These tokens allow information to be represented as microwave pulse patterns instead of conventional digital data, enabling more efficient transmission.
The chip is based on a microwave neural network introduced by the team last year. Rather than converting analog radio signals into digital data before processing them, it performs computation directly using the properties of microwave signals. This reduces the need for conventional signal processing and lowers bandwidth and energy requirements.
According to the researchers, commands such as navigation instructions for a satellite or drone can be converted into a small number of microwave pulses. Another microwave neural network with the same configuration can interpret these pulses directly, eliminating the need to transmit longer streams of digital data.
The team also found that every microwave neural network has unique physical characteristics, meaning only a similarly configured chip can correctly decode the transmitted information. They say this could provide an additional level of security for wireless communication.
The researchers also demonstrated that the chip can compress data using probabilistic bits, or p-bits, whose values change based on incoming data. In one test, they compressed a satellite image of a tropical storm and reconstructed it while reducing the transmitted data by about eight times. They believe this could help small satellites send more useful image data to Earth despite limitations in power and communication bandwidth.
