Power/Performance Bits: Aug. 3

Efficient ADC
Researchers at Brigham Young University, National Yang Ming Chiao Tung University, Texas Instruments, and University of California Los Angeles designed a new power-efficient high-speed analog-to-digital converter.

The ADC consumes only 21 milli-Watts of power at 10GHz for ultra-wideband wireless communications, much lower than other ADCs that consume hundreds of milli-Watts to Watts. The team said the device has the highest power efficiency yet recorded.

To make the ADC more efficient, the researchers focused on reducing loading from the DAC by scaling both the capacitor parallel plate area and spacing. They also grouped unit capacitors differently from the conventional way, grouping together unit capacitors that are part of the same bit in the DAC rather than having them be interleaved throughout. Doing so lowered the bottom-plate parasitic capacitance by three times, significantly lowering power consumption while increasing speed. They also used a bootstrapped switch with dual paths that could be independently optimized.

The power-efficient ADC. (Credit: Shiuh-hua Wood Chiang / BYU)

“We’ve proven the technology of the chip here at BYU and there is no question about the efficacy of this particular technique,” said Wood Chiang, a professor at BYU. “This work really pushes the envelope of what’s possible and will result in a lot of conveniences for consumers. Your Wi-Fi will continue to get better because of this technology, you’ll have faster upload and download speeds and you can watch 4K or even 8K with little to no lag while maintaining battery life.” Chiang said that other applications could include automotive, wearables, and implantable devices.

5G relay transceiver
Scientists at Tokyo Institute of Technology developed a wirelessly powered relay network for 5G systems. The batteryless 28-GHz phased-array relay transceiver aims to improve bandwidth, coverage, and reliability for non-line-of-sight 5G communications.

It uses a vector-summing backscattering technique with 24GHz local oscillator (LO) and 4GHz intermediate frequency (IF) signals.

Kenichi Okada, a professor at Tokyo Tech, said, “Backscatter communication makes it possible to harvest energy from incident signals and reflects back parts of the same signals while modulating the data. In this design, backscatter up-converts the 5G New Radio (5G NR) spectrum at 4GHz and transmits at 28GHz.”

In addition, the transmitter acts as a phase shifter, allowing it to alter the phase of an incoming signal. The backscattering and phase-shifting capabilities of the transmitter facilitate beamforming, where an array of antennas can be controlled to transmit signals in a specific direction.

Passive phase shifters and power combiners (which combine power fed at multiple ports) are used to boost the received signal power for wireless power transfer. The rectifier acts as a self-heterodyne mixer to split and recombine an incoming beam with a modulated version of itself. It also works as a full-wave rectifier with the 24 GHz wireless power transfer signal.

The entire phased-array relay transceiver is configured in an area as small as 1.8 mm². In the receive mode, the wirelessly powered 4×8 array module produces 3.1 mW of power. In the transmit mode, it produces -2.2 dBm of saturated Equivalent Isotropic Radiated Power, which is the output power radiated from an antenna in a single direction. The vector-summing backscatter covers a 360° phase range with 7-bit phase resolution while consuming 0.03 mW in both transmit and receive mode.

Okada added, “The proposed battery-free transceiver enhances 5G connectivity by serving as a repeater between indoor and outdoor environments. This, in turn, will improve user experiences and create new opportunities for operational efficiency in internet-of-things, industrial automation, and new communication services.”

Adding noise to improve image recognition
Researchers from the University of Texas at San Antonio, University of Central Florida, Air Force Research Laboratory, and SRI International propose adding noise as a way to improve explainability and resilience of neural networks for computer vision.

In most models that rely on neural ordinary differential equations (ODEs), a machine is trained with one input through one network, and then spreads through the hidden layers to create one response in the output layer.

Instead, the team’s used neural stochastic differential equations (SDEs), an approach that learns not just from one image but from a set of nearby images due to the injection of the noise in multiple layers of the neural network. When more noise is injected, it needs to find better ways of making explanations and attributions because the model created at the onset is based on evolving characteristics of the image.

“It’s about injecting noise into every layer,” said Sumit Jha, professor in the Department of Computer Science at UTSA. “The network is now forced to learn a more robust representation of the input in all of its internal layers. If every layer experiences more perturbations in every training, then the image representation will be more robust and you won’t see the AI fail just because you change a few pixels of the input image.”