IISVU: Intelligent Imaging System via Ultrasound

Evolution spent a few million years teaching bats and dolphins to see with their ears. They click, listen to the bounce, and assemble a picture of a world that cameras would just call dark. Robots, somehow, mostly skipped that lesson: we give them cameras and lidars and then act surprised when fog, dust, or a wall in the way blinds them. In-air ultrasound already does the distance part really well. The catch is 3D imaging, which usually means packing dozens of transducers into a dense array that costs more than the optical systems it's supposed to beat, and throws away the power advantage that made ultrasound interesting in the first place. Nature didn't put a thousand ears on a bat. It got clever with a few. IISVU is my lab's attempt at that. An eight-channel piezoelectric micromachined ultrasonic transducer (PMUT) platform reconstructs 3D scenes from a sparse array using compressed sensing. A microcontroller generates eight independent pulse trains and drives lithium-niobate PMUTs independently at 15 V, with carriers from 41 to 58 kHz and delays up to 200 μs. Between transmit and receive, a programmable HV2707 switch matrix routes each element under firmware control. On the receive side, an eight-channel analog front end amplifies echoes by 54 dB into an 18-bit ADC that samples all channels simultaneously at 468.8 kSa/s. A full scene is 100 coded acquisitions; we change each element's frequency and firing delay between shots so every ping hits the scene with a different wavefront, and a sparsity-promoting solver stitches it back together. Everything from the pulsers to the ADC fits on a single board the size of a credit card. The same firmware-reconfigurable hardware does two imaging modes with no board changes. In the first, it localizes corner reflectors hidden behind an occluder, out of the array's line of sight. In the second, we rotate the array through a series of poses to form a synthetic aperture much wider than the array itself, which sharpens lateral resolution far beyond what the bare array can resolve. Dense arrays usually get that kind of resolution by adding many more channels. With IISVU, precision is limited by the electronics rather than by how many transducers we could afford to pack in. I'll share the numbers once the papers are published. I recently presented this work at the Gulf Coast Undergraduate Research Symposium and co-authored two accepted IEEE MEMS 2027 conference papers. Partners: Nikita L., Divij M., Ziv B., Tofic E. Org: Liwei Lin Lab

Tags