Semi-supervised Learning (SSL) has received increasing attention in autonomous driving to reduce the enormous burden of 3D annotation. In this paper, we propose UpCycling, a novel SSL framework for 3D object detection with zero additional raw-level point cloud: learning from unlabeled de-identified intermediate features (i.e., "smashed" data) to preserve privacy. Since these intermediate features are naturally produced by the inference pipeline, no additional computation is required on autonomous vehicles. However, generating effective consistency loss for unlabeled feature-level scene turns out to be a critical challenge. The latest SSL frameworks for 3D object detection that enforce consistency regularization between different augmentations of an unlabeled raw-point scene become detrimental when applied to intermediate features. To solve the problem, we introduce a novel combination of hybrid pseudo labels and feature-level Ground Truth sampling (F-GT), which safely augments unlabeled multi-type 3D scene features and provides high-quality supervision. We implement UpCycling on two representative 3D object detection models: SECOND-IoU and PV-RCNN. Experiments on widely-used datasets (Waymo, KITTI, and Lyft) verify that UpCycling outperforms other augmentation methods applied at the feature level. In addition, while preserving privacy, UpCycling performs better or comparably to the state-of-the-art methods that utilize raw-level unlabeled data in both domain adaptation and partial-label scenarios.

We present${\sf Supremo}$, a cloud-assisted system for low-latency image super-resolution (SR) in mobile devices. As SR is extremely compute-intensive, we first further optimize state-of-the-art DNN to reduce the inference latency. Furthermore, we design a mobile-cloud cooperative execution pipeline composed of specialized data compression algorithms to minimize end-to-end latency with minimal image quality degradation. Finally, we extend${\sf Supremo}$to video applications by formulating a dynamic optimal control algorithm to design${\sf Supremo-Opt}$, which aims to maximize the impact of SR while satisfying latency and resource constraints under practical network conditions.${\sf Supremo}$upscales 360p image to 1080p in 122 ms, which is 43.68× faster than on-device GPU execution. Compared to cloud offloading-based solutions,${\sf Supremo}$reduces wireless network bandwidth consumption and end-to-end latency by 15.23× and 4.85× compared to baseline approach of sending and receiving whole images, and achieves 2.39 dB higher PSNR compared to using conventional JPEG to achieve similar data size compression. Furthermore,${\sf Supremo-Opt}$guarantees robust performance in practical scenarios.

For emerging high-density Wi-Fi networks, there have been plenty of studies that claim the need for more aggressive channel access to enhance spatial reuse. Against those previous ideas, this paper presents a different perspective that existing Wi-Fi devices already have excessive transmission opportunities, even without protecting each other in certain scenarios. We shed light on an anomaly within actual carrier sensing (CS) behaviors, which makes some neighboring devices become blind to each other and transmit simultaneously, due to undetected preambles. Through experimental study and analysis, we reveal both sides of the anomaly heavily affecting the overall network performance. Based on the observations, we design REFRAIN, a standard-compliant PHY/MAC framework, which copes with and further exploits the anomaly for better spatial reuse. Our prototype using NI USRP and commercial Wi-Fi devices shows the feasibility and effectiveness of our approach, while extensive simulation results demonstrate that REFRAIN achieves up to 1.57× higher average throughput by promoting valid transmissions, without modifying the 802.11 CS specification at all.

AI assistants have found their place in households but most of the existing assistants use single modal interaction. We present a language assistant for kids called Hola (Hang out with the Language Assistant) which is a true multimodal assistant. Hola is a small mobile robot based assistant capable of understanding the objects around it and responding to questions about objects that it can see. Hola is also able to adjust the camera position and its own position to make an extra attempt to understand the object using robot control mechanism. The technology behind it uses a combination of natural language understanding, object detection, and hand pose detection. In addition, Hola also supports reading book in the form of storytelling for kids using OCR. Children can ask a question about any word that they do not understand and Hola can retrieve the information from the internet and tells the meaning, other details of the word. After reading the book or a page, the robot asks the child based on the words used in the book to confirm the child’s understanding of the book.

LTE-Unlicensed (LTE-U) supports LTE downlink operation in 5GHz unlicensed bands where Wi-Fi has been a traditional user for a long time. To achieve fair coexistence with Wi-Fi, LTE-U employs carrier sense adaptive transmission (CSAT), but it does not guarantee perfectly fair coexistence. Therefore, many studies have dealt with unfair coexistence problems of Wi-Fi and LTE-U. However, in this paper, our experiment results show that a Wi-Fi station not only suffers from unfair coexistence but also wastes energy and air time when it coexists with LTE-U. To cope with this problem, we propose AWARE, a station-driven adaptive Wi-Fi power save operation coexisting with LTE-U, which detects LTE-U and adjusts Wi-Fi power state adaptively. We implement AWARE on a commercial 802.11n device, and our evaluation shows that a Wi-Fi station achieves almost the same throughput while reducing its power consumption by up to 33% and enhancing the throughput of neighbor Wi-Fi stations by up to 43%.

This paper sheds light on the necessity for transmit power control of 802.11 clear-to-send (CTS) frame for better spatial reuse in densely deployed wireless local area networks (WLANs). We first study the effect of the interference caused by CTS frames through both experiments and numerical analysis. Based on the analysis, we raise a limited spatial reuse problem caused by RTS/CTS mechanism and establish a guideline for controlling CTS transmit power to solve the problem. Next, a standard-compliant CTS transmit power control method, namely Quiet CTS (QCTS), is proposed. Through extensive simulations and prototyping using both software-defined radio and commercial devices, we demonstrate that QCTS enhances the average network throughput by up to 50% by enabling more simultaneous data transmissions.

Acoustic communication using microphones and speakers of smart devices is one of the most spotlighted wireless technologies in recent years. In particular, chirp-based acoustic communication is widely adopted for smart device applications because of its robustness to frequency selectivity. Since many chirp-based acoustic applications run in the background on mobile devices, the power consumption of chirp-based acoustic communication is a critically important issue. Using energy detectors (EDs), which determine the existence of a valid acoustic signal based on energy level, applications can reduce power consumption by working only when a valid signal exists. However, conventional ED fails to distinguish between valid signals and high-energy noise in everyday life. In this paper, we propose No Entry, a novel ED for chirp-based acoustic communication systems. No Entry avoids not only high-energy noises but also a different modulation-based acoustic signal by utilizing the frequency sweeping characteristic of chirp signals. We implement prototype Android applications to evaluate the detection accuracy and power consumption. Compared with the state-of-the-art schemes, No Entry reduces energy consumption by 30% while achieving a greater detection performance.

With increasing demand for wireless connectivity, IEEE 802.11 WLANs have become ubiquitous and continue to grow in number. This leads to high density of basic service sets with the significant co-channel interference (CCI) among them. This paper sheds light on the CCI caused by 802.11 MAC ACK frames, which has been less studied than the CCI caused by data frames. Based on stochastic geometry analysis, we propose Quiet ACK (QACK), a dynamic transmit power control algorithm for ACK frames. Fine-grained transmit power adjustment is enabled by CCI detection and CCI power estimation in the middle of a data frame reception. Our prototype using software-defined radio shows the feasibility and performance gain of QACK, i.e., 1.5x higher throughput than the legacy 802.11 WLANs. The performance of QACK is further evaluated in more general WLAN environments via extensive simulations using ns-3.

Antenna selection, a cost-effective way to enhance network performance, has been employed in a limited manner in wireless local area networks (WLANs) due to the lack of channel information at the transmitter. In this paper, a practical antenna selection system without using channel information is proposed. We first describe the practical issues of antenna selection system for infrastructure-based WLANs, and then, analyze its characteristics through an extensive measurement study. Based on that, we propose antenna selection algorithms of access point (AP) for both (1) unicast transmission and (2) multicast transmission and reception. The proposed algorithms are comparatively evaluated using prototype implementation in a commercial AP. It is demonstrated that the proposed algorithms achieve up to 34% throughput gain and 54% frame error rate reduction for unicast and multicast transmissions, respectively.

Femtocell base station (BS) is a low-power, low-price BS based on the cellular communication technology. It is expected to become a cost-effective solution for improving the communication performance of indoor users, whose traffic demands are large in general. We propose long-term parameter optimization schemes for open and hybrid femtocells, which maximize the average throughput of macrocell users by offloading the macro users' downlink traffic to femtocells. To achieve this goal, load balancing between femtocells and macrocells is needed, and hence, we jointly optimize the ratio of dedicated resources for femtocells as well as the femtocell service area in open access femtocell networks by numerical analysis. Then, we extend our algorithm to hybrid access femtocells, where some intra-femtocell resources are dedicated only for femtocell owners, while remaining resources are shared with foreign macrocell users. Our evaluation results show that the proposed parameter optimization schemes significantly enhance the performance of macrocell users because of the large offloading gain. The benefits provided to femtocell users are also adaptively maintained according to the femtocell users' requirements. The results in this paper also provide insights about the situations where femtocell deployment on dedicated channels is preferred to the cochannel deployment.

The demand for wireless communication in enterprise network has been increasing with the advent of "smart working environment." In enterprise wireless network, high density of access points (APs) and limited frequency resources raise the significance of radio resource management. In this paper, we propose a centralized algorithm, named JACPoT, for jointly executable transmission power control (TPC) and dynamic channel assignment (DCA). The proposed algorithm first guarantees that the links between AP and client stations are robust to the inter-cell interference via TPC. Next, it minimizes contention between neighboring APs via DCA utilizing a novel metric and annealed Gibbs sampler. Through ns-3 simulations based on measurement in real office environments, we demonstrate that JACPoT reduces uplink frame collisions by 36% and significantly enhances both data and voice traffic delivery.

Femtocell is a cost-effective solution to both soaring wireless data traffic and coverage hole problems. Along with technical issues, the economic benefit which Wireless Service Providers (WSPs) and mobile users get from the femtocell service is a key to successful adoption of femtocell technology. In this paper, we study pricing issues for femtocell service using game theoretic approach. We propose a realistic and detailed pricing framework, and model the interaction between a monopolist WSP and users by a leader-follower game. We also model the competition between two WSPs in a duopoly market at the adoption of femtocell service by a non-cooperative simultaneous game. The competition between femtocells and repeaters is also considered using the same pricing framework. Analysis results present important factors and pricing strategies for economic success of femtocell technology.

Various packet-based simulation tools (e.g., NS-3) have been employed for design, validation, and evaluation of new protocols for WiFi networks since they offer cost efficiency, scalability, and reproducibility. These benefits come, however, at the expense of lack of realism compared to live testbed experiments. This is attributed in a major part to the difficulty of capturing detailed characteristics of channel dynamics, bit-level protocol specification (PHY layer), and application/user behaviors in a high-fidelity manner. The performance gap predicted by simulation and live testbed becomes even more pronounced when one considers a wide diversity of device characteristics and the way each device is used by end users. For example, smartphones generally show worse WiFi performance than other WiFi devices (e.g., laptops and tablets) because smartphones suffer from additional signal loss due to hand-grips and the low antenna gains of their embedded antennas. The goal of this study is to significantly close the gap by incorporating survey- and measurement-based smartphone WiFi characteristics and realistic hand-grip models into traditional WiFi network simulators (NS-3 in this study). The enhanced WiFi simulation tool's performance prediction capability is validated through an comparative study between testbed experiments and simulations.

As design complexity increases and process technologies shrink, high resolution and quick turnaround diagnostics are always in high demand. This is especially important for the development of high performance microprocessors, where not only the gross defects are a big concern, but small "AC" defects (soft defects) are as well. To diagnose these "AC" defects, conventional electrical diagnostic techniques, such as software-based and tester-based approaches, might not be effective. In this paper, a dynamic laser stimulation technique is described. When combined with the Picosecond Imaging Circuit Analysis (PICA) technique, very high diagnostic resolution can be achieved. Several IBM Power6TM(P6) microprocessor [1] diagnostic examples are given to demonstrate the effectiveness of the proposed diagnostic strategy and methods.

Experimental results are presented for the defect-oriented testing of PLLs using the charge-based frequency measurement BIST technique. Using National Semiconductor Corp.'s 0.25 /spl mu/m CMOS 900 MHz PLL, we demonstrate a low-cost and practical BIST solution. The area overhead is minimized by using the existing PLL circuitry for the BIST structure. The test time is typically 12 /spl mu/sec, significantly faster than conventional testing.

We propose a new method of defect-oriented testing of PLL using charge-based frequency measurement BIST (CF-BIST) technique. As no test stimulus is required and the test output is pure digital, low-cost and practical implementation of on-chip BIST for a PLL is possible. Fault simulations using the 900 MHz PLL from National Semiconductor Corp. show higher fault coverage than previous test methods.

A quick literature survey revealed that many researchers have and continue to work on automatic test pattern generation for analog and mixed-signal circuits and systems, however, very few if any have addressed the problem of test set size. This paper presents a novel test set compaction algorithm which takes a generated test set and maximally reduces the number of test vectors required while maximizing the fault coverage. Results show that 58.33% reduction can be achieved. Smaller test set implies lower total test time and long test times have been identified as one of the bottlenecks in analog and mixed-signal test.

In this paper, we present an approach to construct a set of dynamic test signals for analog circuits and systems. Testability transfer factors are introduced and we use them as the basis to construct an efficient test set. Fault detectability and fault coverage are defined. Two circuits from the suite of analog and mixed-signal circuits are used to evaluate our approach. The fault coverage is 100% for both circuits studied. The approach presented may be used to construct inputs signals for the selection of an external stimulus applied through an arbitrary waveform generator.

Fast and efficient test generation techniques are key to reducing the current high cost of testing analog circuits. A new multi-frequency test generation technique for detecting catastrophic failures in this class of circuits is presented. Testability transfer factors are introduced and we use them to construct an efficient test set for analog circuits. Fault detectability and fault coverage are also defined. Two circuits from the suite [4] of analog and mixed-signal benchmark circuits are used to validate our approach.