We propose a novel progressive framework to optimize deep neural networks. The idea is to try to combine the stability of linear methods and the ability of learning complex and abstract internal representations of deep learning methods. We insert a linear loss layer between the input layer and the first hidden non-linear layer of a traditional deep model. The loss objective for optimization is a weighted sum of linear loss of the added new layer and non-linear loss of the last output layer. We modify the model structure of deep canonical correlation analysis (DCCA), i.e., adding a third semantic view to regularize text and image pairs and embedding the structure into our framework, for cross-modal retrieval tasks such as text-to-image search and image-to-text search. The experimental results show the performance of the modified model is better than similar state-of-art approaches on a dataset of National University of Singapore (NUS-WIDE). To validate the generalization ability of our framework, we apply our framework to RankNet, a ranking model optimized by stochastic gradient descent. Our method outperforms RankNet and converges more quickly, which indicates our progressive framework could provide a better and faster solution for deep neural networks.

Cloud computing emerges as a new computing pattern that can provide elastic services for any users around the world. It provides good chances to solve large scale scientific problems with fewer efforts. Application deployment remains an important issue in clouds. Appropriate scheduling mechanisms can shorten the total completion time of an application and therefore improve the quality of service (QoS) for cloud users. Unlike current scheduling algorithms which mostly focus on single task allocation, we propose a deadline based scheduling approach for data-intensive applications in clouds. It does not simply consider the total completion time of an application as the sum of all its subtasks’ completion time. Not only the computation capacity of virtual machine (VM) is considered, but also the communication delay and data access latencies are taken into account. Simulations show that our proposed approach has a decided advantage over the two other algorithms.

Rough set theory is an important tool to solve uncertain problems. Attribute reduction, as one of the core issues of rough set theory, has been proven to be an effective method for knowledge acquisition. Most of heuristic attribute reduction algorithms usually keep the positive region of a target set unchanged and ignore boundary region information. So, how to acquire knowledge from the boundary region of a target set in a multi-granulation space is an interesting issue. In this paper, a new concept, fuzziness of an approximation set of rough set is put forward firstly. Then the change rules of fuzziness in changing granularity spaces are analyzed. Finally, a new algorithm for attribute reduction based on the fuzziness of 0.5-approximation set is presented. Several experimental results show that the attribute reduction by the proposed method has relative better classification characteristics compared with various classification algorithms.

The rapid development of location-based social networks (LBSNs) has provided an unprecedented opportunity for better location-based services through point-of-interest (POI) recommendation. POI recommendation is personalized, location-aware, and context depended. However, extreme sparsity of user-POI matrix creates a severe challenge. In this paper we propose a textual-geographical-social aware probabilistic matrix factorization method for POI recommendation. Our model is textual-geographical-social aware probabilistic matrix factorization called TGS-PMF, it exploits textual information, geographical information, social information, and incorporates these factors effectively. First, we exploit an aggregated latent Dirichlet allocation (LDA) model to learn the interest topics of users and infer the interest POIs by mining textual information associated with POIs and generate interest relevance score. Second, we propose a kernel estimation method with an adaptive bandwidth to model the geographical correlations and generate geographical relevance score. Third, we build social relevance through the power-law distribution of user social relations to generate social relevance score. Then, our exploit probabilistic matrix factorization model (PMF) to integrate the interest, geographical, social relevance scores for POI recommendation. Finally, we implement experiments on a real LBSN check-in dataset. Experimental results show that TGS-PMF achieves significantly superior recommendation quality compared to other state-of-the-art POI recommendation techniques.

In heterogeneous networks (HetNets), it is desirable to offload users from macro cells to small cells to achieve load balancing. However, the offloaded users suffer a strong inter-tier interference. To guarantee the performance of the offloaded users, the interference from macro cells should be carefully managed. In this paper, we jointly optimize load balancing and interference coordination in multi-antenna HetNets. Different from previous works, instead of almost blank subframes (ABS) on which the macro cells waste time resource, the macro cells suppress the interference to the offloaded users by zero-forcing beamforming (ZFBF) on interference nulling subframes (INS). Considering user association cannot be conduct frequently, we derive the long-term throughput of users over Rayleigh fading channels while previous works focused on instantaneous rate. From the perspective of the spectrum efficiency and user fairness, we formulate a long-term network-wide utility maximization problem. By decomposing the problem into two subproblems, we propose an efficient joint load balancing and interference coordination strategy. Simulation results show that our proposal can achieve good system performance gains over counterparts in term of the network utility, cell edge throughput and average throughput.

Caching popular content in the storage of small cells is deemed as an efficient way to decrease latency, offload backhaul and satisfy user’s demands. In order to investigate the performance of cache-enabled small cell networks, coverage probability is studied in both single-point transmission and cooperative multipoint (CoMP) transmission scenarios. Meanwhile, the caching distribution modeled as Zipf and uniform distribution are both considered. Assuming that small base stations (SBSs) are distributed as a homogeneous Poisson point process (HPPP), the closed-form expressions of coverage probability are derived in different transmission cases. Simulation results show that CoMP transmission achieves a higher coverage probability than that of single-point transmission. Furthermore, Zipf distribution-based caching is more preferable than uniform distribution-based caching in terms of coverage probability.

This paper proposes a combination technique of the frequency-domain random demodulation (FRD) and the broadband digital predistorter (DPD). This technique can linearize the power amplifiers (PAs) at a low sampling rate in the feedback loop. Based on the theory of compressed sensing (CS), the FRD method preprocesses the original signal using the frequency domain sampling signal with different stages through multiple parallel channels. Then the FRD method is applied to the broadband DPD system to restrict the sampling process in the feedback loop. The proposed technique is assessed using a 30 W Class-F wideband PA driven by a 20 MHz orthogonal frequency division multiplexing (OFDM) signal, and a 40 W GaN Doherty PA driven by a 40 MHz 4-carrier long-term evolution (LTE) signal. The simulation and experimental results show that good linearization performance can be achieved at a lower sampling rate with about 24 dBc adjacent channel power ratio (ACPR) improvement by applying the proposed combination technique FRD-DPD. Furthermore, the performance of normalized mean square error (NMSE) and error vector magnitude (EVM) also has been much improved compared with the conventional technique.

Due to the high cost and power consumption of the radio frequency (RF) chains, it is difficult to implement the full digital beamforming in millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems. Fortunately, the hybrid beamforming (HBF) is proposed to overcome these limitations by splitting the beamforming process between the analog and digital domains. In recent works, most HBF schemes improve the spectral efficiency based on greedy algorithms. However, the iterative process in greedy algorithms leads to high computational complexity. In this paper, a new method is proposed to achieve a reasonable compromise between complexity and performance. The novel algorithm utilizes the low-complexity Gram-Schmidt method to orthogonalize the candidate vectors. With the orthogonal candidate matrix, the slow greedy algorithm is avoided. Thus, the RF vectors are found simultaneously without any iteration. Additionally, the phase extraction is applied to satisfy the element-wise constant-magnitude constraint on the RF matrix. Simulation results demonstrate that the new HBF algorithm can make substantial improvements in complexity while maintaining good performance.

In IEEE 802.11 networks, many access points (APs) are required to cover a large area due to the limited coverage range of APs, and frequent handoffs may occur while a station (STA) is moving in an area covered by several APs. However, traditional handoff mechanisms employed at STAs introduce a few hundred milliseconds delay, which is far longer than what can be tolerated by some multimedia streams such as voice over Internet protocol (VoIP), it is a challenging issue for supporting seamless handoff service in IEEE 802.11 networks. In this paper, we propose a pre-scan based fast handoff scheme within an IEEE 802.11 enterprise wireless local area network (EWLAN) environment. The proposed scheme can help STA obtain the best alternative AP in advance after the pre-scan process, and when the handoff is actually triggered, STA can perform the authentication and reassociation process toward the alternative AP directly. Furthermore, we adopt Kalman filter to minimize the fluctuation of received signal strength (RSS), thus reducing the unnecessary pre-scan process and handoffs. We performed simulations to evaluate performance, and the simulation results show that the proposed scheme can effectively reduce the handoff delay.

This paper presents a modified circular-cut multiband fractal antenna with good radiation patterns designed for digital cellular system (DCS), personal communication system (PCS), 2.4/5.2/5.8 GHz wireless local area networks (WLAN) and 2.5/3.5/5.5 GHz worldwide interoperability for microwave access (WiMAX) applications simultaneously. Originally, the modified circular monopole antenna is designed to resonate at around 2.1 GHz and 3.6 GHz. After subtracting the circular iterative tree fractal structure, it can produce three other resonances at around 5.6 GHz, 6.47 GHz and 7.89 GHz. Besides, as the number of iterations increases, not only do the new frequency bands appear (it demonstrates the good self-similarity property of the proposed antenna), but also the operating bands shift from high frequency to low frequency (it shows the well space filling property). Furthermore, the proposed antenna owns a compact structure, which can achieve the 5.28 dBi of relative high gain. And the measured results are basically accordant to simulated results, which proves the effectiveness of the proposed antenna.

This paper focuses on the linear transceiver design for multiple input multiple output (MIMO) interference channel (IC), in which a bounded channel error model is assumed. Two optimization problems are formulated as minimizing maximum per-user mean square error (MSE) and sum MSE with the per-transmitter power constraint. Since these optimization problems are not jointly convex on their variable matrices, the transmitter and receiver can be optimized alternately respectively. For each matrix, an approximated approach is presented where the upper bound of constraint is derived so that it has less semidefinite, thus the problem can be viewed as second-order-cone programming (SOCP) and gets less computational complexity. Compared with the conventional S-procedure method, the proposed approach achieves similar performance, but reduces the complexity significantly, especially for the system with large scale number of antennas.

Compressed sensing (CS) provides a new approach to acquire data as a sampling technique and makes it sure that a sparse signal can be reconstructed from few measurements. The construction of compressed matrixes is a central problem in compressed sensing. This paper provides a construction of deterministic CS matrixes, which are also disjunct and inclusive matrixes, from singular pseudo-symplectic space over finite fields of characteristic 2. Our construction is superior to DeVore’s construction under some conditions and can be used to reconstruct sparse signals through an efficient algorithm.

As the widespread employment of firewalls on the Internet, user datagram protocol (UDP) based voice over Internet protocol (VoIP) system will be unable to transmit voice data. This paper proposed a novel method to transmit voice data based on transmission control protocol (TCP). The method adopts a disorder TCP transmission strategy, which allows discontinuous data packets in TCP queues read by application layer directly without waiting for the retransmission of lost data packets. A byte stream data boundary identification algorithm based on consistent overhead byte stuffing algorithm is designed to efficiently identify complete voice data packets from disordered TCP packets arrived so as to transmit the data to the audio processing module timely. Then, by implementing the prototype system and testing, we verified that the proposed algorithm can solve the high time delay, jitter and discontinuity problems in standard TCP protocol when transmitting voice data packets, which caused by its error control and retransmission mechanism. We proved that the method proposed in this paper is effective and practical.