The key factor for training a good neural network lies in both model capacity and large-scale training data. As more datasets are available nowadays, one may wonder whether the success of deep learning descends from data augmentation only. In this paper, we propose a new dataset, namely, Extended ImageNet Classification (EIC) dataset based on the original ILSVRC CLS 2012 set to investigate if more training data is a crucial step. We address the problem of object localization where given an image, some boxes (also called anchors) are generated to localize multiple instances. Different from previous work to place all anchors at the last layer, we split boxes of different sizes at various resolutions in the network, since small anchors are more prone to be identified at larger spatial location in the shallow layers. Inspired by the hourglass work, we apply a conv-deconv network architecture to generate object proposals. The motivation is to fully leverage high-level summarized semantics and to utilize their up-sampling version to help guide local details in the low-level maps. Experimental results demonstrate the effectiveness of such a design. Based on the newly proposed dataset, we find more data could enhance the average recall, but a more balanced data distribution among categories could obtain better results at the cost of fewer training samples.

In spatial modulation (SM) systems, the distribution of amplitude-phase modulation (APM) symbol causes the tradeoff of detection performance between APM symbol and active antenna, which infers the overall capacity of SM system. In this paper, the APM constellation and symbol labeling of SM system is investigated. The optimized APM constellation is designed to maximize the system coded modulation average mutual information (CM-AMI). The symbol labeling of SM system is optimized with binary switching algorithm (BSA) using the symbol detection pairwise error probability (PEP) criteria. Moreover, the proposed APM constellation and symbol labeling are applied in a coded spatial modulation (SM) system with the bit-interleaved coded modulation and iterative demapping (BICM-ID). Simulation result shows the BICM-ID SM system with proposed constellation and symbol labeling has a signal-to-noise ratio (SNR) gain compared to conventional constellation system and independent demapping system.

Compared with Wireless Local Area Networks (WLANs) in the ISM band, smaller path loss of the spectrum in TV White Space (TVWS) provides larger coverage region with the same power budget. More nodes derived from the larger coverage region increase the collision probability, which damages the throughput efficiency of carrier sense multiple access#x002F;collision avoidance (CSMA#x002F;CA) protocol in 802.11af. The scheme of the Channel Access based on Frequency Contention (CAFC) exploits single sub-carrier of Filtered Bank Multi-Carrier (FBMC) to realize frequency contention. CAFC improves the system throughput efficiency by decreasing the collision probability and removing channel idle time. The set of possible frequency contention outcomes can be enlarged to further decrease the collision probability with little cost appended to the system complexity. Compared with conventional CSMA#x002F;CA in 802.11af, the simulation indicates that the CAFC scheme enhances the saturated throughput efficiency significantly. Besides, the CAFC scheme is designed to be compatible with 802.11af, which reduces the application cost in practice.

Generalized spatial modulation (GSM) is an extended architecture to the spatial modulation multiple antenna system. The GSM system uses multiple activated transmit antennas to further exploit the system transmission capacity. In this paper, we consider the GSM systems with channel diagonalization (CD-) transmitter pre-coder (TPC), channel inversion TPC, and without TPC. For each TPC scheme, the pairwise activated antenna detection discriminant and closed-form antenna detection pairwise error probability (PEP) of the GSM system are derived. We use the antenna detection PEP to calculate the analytical antenna detection symbol error rate union bound and the mutual information lower bound. We use asymptotic methods to analyze antenna detection diversity and coding gains of GSM system with or without TPC when the signal-to-noise ratio is high. Together with complexity analysis on the pre-coders and antenna detectors, we discover the CD-TPC trades the antenna detection performance for complexity reduction, while the CD-TPC scheme achieves a balance between the performance and complexity. This paper introduces methods to the performance evaluation of the GSM system, and provides a guideline for the GSM system design such as antenna configuration and TPC schemes selection.

Generalized spatial modulation (GSM) is an extended architecture to the spatial modulation (SM) multiple antenna system. Multiple activated transmit antennas (TAs) are applied in the GSM system to further exploit the system transmission capacity. In this paper, we consider the transmitter pre-coding (TPC) GSM systems with channel diagonalization (CD) and channel inversion (CI) pre-coding schemes. The pairwise activated TA detection discriminant and closed-form pairwise error probability (PEP) function are derived. We use the TA detection PEP to calculate the analytical TA detection symbol error rate (SER) union bound and the mutual information lower bound. The analytical results are verified by numerical simulation over Rayleigh fading channels. We investigate the system with different number of antennas and TPC schemes to evaluate the error probability and mutual information performance of the GSM system. The results show that the CD-TPC GSM system with equal power allocation has the most TA detection mutual information, and the CD-TPC system with water-filling power allocation has the most joint APM domain and space domain mutual information.

Spatial modulation (SM) is a recently proposed multiple-input multiple-output (MIMO) architecture that exploits the space domain for information transmission. In this paper, SM systems with continuous amplitude-phase modulation (APM) and the corresponding maximum likelihood (ML) active transmit antenna detector are considered. We introduced a method to compute the theoretical transition probability matrix and the mutual information of the ML antenna detector. For SM multiple-input single-output (MISO) system, we derive the error probability and the mutual information of the antenna detector in closed-form. For multiple receive antenna SM system, the union bound of error probability and the lower bound of antenna detection mutual information are proposed. The analytical results are verified by numerical simulation over Rayleigh fading channels. We investigate systems with different number of antennas and APM symbol input distributions to evaluate the performance of SM. The joint APM and antenna detection mutual information of SM systems is compared to the capacity of vertical Bell Labs layered space-time (V-BLAST) spatial multiplexing MIMO scheme. Our research gives achievable transmission rates of SM-MISO/MIMO systems and can be a guideline for suitable application scenarios.