The future of the automobile industry depends on the innovations of imaging sensors used in autonomous vehicles. Soon, more cars on the road will be self-driven with greater accuracy and spatial awareness than ever before. One challenge of autonomous cars is developing imaging sensors to adapt to unfavorable driving conditions, such as fog, dust, or rain. Spatial resolution, or how clearly an image can be captured, is limited in the sensors and cameras currently on the road.
The prevailing imaging sensors used in autonomous vehicles rely on optical cameras for short-distance object detection and a remote sensing method known as light detection and ranging (LiDAR). These cars are advanced enough to detect nearby vehicles as well as determine the distance between the car and an object located further away. However, these basic features are not adequate in bad weather conditions, when the imaging reliability is drastically decreased. To address this issue, Zhi Hu and a team of researchers, all from professor Ruonan Han’s group at MIT, created a new receiver array that is capable of processing sub-terahertz signals, effectively optimizing the visual acuity of imaging sensors.
Sub-terahertz signals are electromagnetic waves with very short wavelengths, which allow for higher spatial resolution and lower transmission loss. “With lower transmission loss, the output signals of sub-terahertz receivers will still be strong…whereas signals output from LiDARs will be weak with a low signal-to-noise ratio,” Hu said. Sub-terahertz sensors have not been implemented until recently, when scientists discovered that the high costs of building sub-terahertz transmitters and receivers could be minimized with a widely-used low-cost process—the complementary metal-oxide-semiconductor (CMOS) process.
In order to interpret sub-terahertz signals, Han’s group built an array of heterodyne receivers. According to Hu, each heterodyne receiver requires several components, including an antenna to capture radio-frequency (RF) signals such as the sub-terahertz signals described above, a local oscillator (LO) to change the frequency of an incoming signal, and a frequency mixer that downshifts the frequency of the RF signal. These receivers are able to convert RF signals into directly processable, low-frequency electrical signals without losing encoded information.
The two current methods of detection are square-law detection and heterodyne detection. Square-law detectors only have RF signal input and extract the amplitude information, so the output signal is relatively weak; on the other hand, the output signal of a heterodyne detector is much stronger because both the RF and LO signals are read, and both the amplitude and phase information of RF signals can be extracted. “While researchers have previously been able to design scalable arrays of square-law detectors because of its simple implementation, [they] never built a scalable array of heterodyne detectors due to the challenges of making receivers compact and coupling LO signals for coherence,” Hu explained.
To tackle these challenges, the MIT team decided to change the entire design of the heterodyne array. Earlier models relied on a centralized architecture, in which each unit of the pixel array receives LO signals from a single source. However, this causes an inherent trade-off between scalability and sensitivity. By shifting the design toward decentralization, each unit now generates its own LO signal with higher signal power. Higher LO power per unit leads to higher signal-to-noise ratio (SNR) and also the desired higher sensitivity. However, because each unit now has its own LO signal, output signals are randomized and not coherent with each other. “The key enabler of decentralization is the LO signal coupler in each unit—it synchronizes the phase of the LO signal in the current unit with the phase of LO signals of adjacent units, ensuring coherence,” Hu said. With this breakthrough in the design, the researchers were able to successfully build the first large-scale heterodyne sub-terahertz receiver array.
Imaging sensors can vastly improve the safety of self-driving vehicles—instead of just being able to detect the presence of an object nearby, radars will now be able generate an image of the object with high spatial resolution. While these new sub-terahertz sensors are unlikely to fully replace LiDARs and optical cameras, they already provide greater reliability for object detection and recognition, especially in harsher weather conditions. Hu is optimistic about the prospects of their current projects, which include building arrays with high spatial resolution on a large scale as well as building more complex interdependent sub-terahertz circuits. With this research, the MIT team is on its way towards substantially promoting the role of sub-terahertz sensors in the automobile industry.