CAREER: Taming Wireless Devices Cross-Layer Errors with Assistive Networked Edges
Project Synopsis
Wireless devices are inherently faulty due to imperfect electronics and noisy environments, which can result in multifaceted data errors in computing, caching, and communications (C3). These errors have been widely deemed harmful, so a great number of error-control mechanisms are built into wireless devices’ hardware and software layers to avoid errors. However, data errors can be benign or even beneficial to design objectives (e.g., perturbing gradients when training deep neural networks may help the model escape local optima) while existing error-control studies that either consider the “worst case” for absolute error removal or reactively adapt to the dynamics of errors are passive, non-functional, and often redundant.
Figure 1 shows a demonstrative example of different wireless applications, for which existing designs adjust error-correction margins according to the naturally evolved error modalities (X axis). Clearly, existing designs either waste error-correction margins (yellow-shaded area for error-free applications) or leave random and non-usable errors (dashed area for error-tolerant applications). Motivated by this research gap, this project’s ideology of “turning waste into wealth” by proactively harvesting, rendering, and controlling errors offers an innovation that will push the frontier of the state of the art. This project proposes three research thrusts to achieve this vision.
In our prior studies [TMC'19, ICDCS'17], we adopted this idea in an OFDM system as shown in Figure 2. Specifically, IEEE 802.11a adopts a stair-case link adaptation technique in accordance to the measured SNR. To cope with the worst-case channel conditions, there always exists unused channel coding capability. In light of it, we intentionally render some symbol errors by nullifying the transmit power of a selective set of OFDM sub-carriers - the one that inherently suffers from strong frequency selective fading. These "silence symbols'' are viewed by the receiver as erasure errors, which can be corrected by bit interleaving and residual coding capability. Moreover, the temporal-spectral distribution of “silence symbols” can be encoded to carry extra information for implicit communications, as shown in Figure 2(b). Based on this design, we showed significant performance gains in throughput and power saving simultaneously.
Personnel and Collaborators
|
Steven Puckett (Fall 2021 - )
MS., Electrical Engineering, University of Alabama at Birmingham Email: scp0019 at uah dot edu Research Interests: Low-power IoT and Edge Intelligence |
|
Na Gong
W. Nicholson Associate Professor Department of Electrical and Computer Engineering University of South Alabama Email: nagong at southalabama dot edu |
Research Progress
-
[1]
Two Birds with One Stone: Differential Privacy by Low-power SRAM Memory
J. Liu*, N. Gong*, and H. Das, IEEE Transactions on Information Forensics and Security (IEEE TIFS), Under Review. (*equal contribution)
J. Liu, C. Zhang, K. Xue, and Y. Fang, to appear at IEEE Transactions on Mobile Computing (IEEE TMC).
Education and Outreach Activities
under development
Disclaimer
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.