dAJC: A 2.02-mW 50-Mb/s Direct Analog to MJPEG Converter for Video Sensor Nodes Using Switched Capacitor MAC-Quantizer With Process Calibration

With the advancement in the field of the Internet of Things (IoT) and Internet of Bodies (IoB), video camera applications utilizing video sensor nodes (VSNs) have become increasingly significant. These applications are pivotal in areas such as autonomous driving, health monitoring, machine vision, and security. Notable examples include their use in uncrewed aerial vehicles (UAVs), body-worn cameras, and surveillance systems. However, these applications often face challenges related to power consumption, primarily due to the large data volumes generated by analog-to-digital converters (ADCs) during the transmission of digitized high-resolution video signals. This data deluge introduces additional processing and storage overheads, exacerbating the problem of resource-constrained sensor nodes having limited battery capacity. To address this challenge, we propose a low-power solution that performs video compression directly at the sensor while shifting computation from the digital domain to the energy-efficient analog domain (suitable for peak signal-to-noise ratio (PSNR) values below 40–50 dB). Unlike conventional architectures, which rely on digital signal processing (DSP) blocks for computation and compression after the ADC stage, our approach eliminates the need for these power-intensive digital blocks at the sensor. Instead, we employ a switched capacitor (SC)-based computation unit operating in the analog domain, significantly reducing overall power consumption. Additionally, we utilize a dynamically activated ADC that activates only for significant samples, which constitute a small fraction ( ${\le }5$ %) of the total captured analog video samples. This approach enables us to achieve approximately ${\sim } 20{\times }$ lower analog-to-digital (A/D) conversion energy. By combining analog computation with the dynamically activated ADC, our proposed solution achieves at least ${\gt } 2{\times }$ reduction in power consumption compared to a digital implementation, with only a ~5-dB degradation in PSNR—an imperceptible difference to the human eye. We also propose a calibration scheme to mitigate the impact of variations introduced by analog computation, thereby enhancing the PSNR of the decoded video or image.