In this paper, the realizable short-code direct sequence code-division multiple-access (DS-CDMA) system occupying Nyquist chip bandwidth and exhibiting zero direct-current (DC) is developed. By limiting the spreading waveform to exhibit zero DC and continuous bandlimited spectrum, the realizability and DC nulling constraints on spreading codewords are derived to enable such system realization. Based on both constraints, a low-complexity implementation structure in spreading and despreading is also addressed. Several frequently-adopted spreading codes are examined for realization feasibility. It is shown that Nyquist chip bandwidth and zero-DC transmission can be practically achieved for the bandlimited DS-CDMA system using two conventional even-length orthogonal spreading codes at the sacrifice of very little user capacity. A new odd-length orthogonal spreading code is also proposed to enable such system realization.
Under the boom of low-orbit satellite Internet constellation construction, an LDPC decoder design method is proposed for the outgoing physical layer of satellite Internet in order to carry out effective monitoring for typical satellite signals with high bandwidth and low time delay. The scheme uses a partially parallel architecture, where the node update step based on the normalized min-sum algorithm omits the resource consumption and compresses the number of clocks consumed by operations such as multi-level comparison of message values, which helps to improve the decoder throughput rate. Based on Virtex-7 series hardware platform, we realize the high rate decoding of long and short frame LDPC codes with different modulation methods in DVB-S2/S2X standard. At 204.8 MHz operating frequency, the decoder can reach 768 Mbps throughput. The decoder implemented in this paper is capable of meeting the rate and performance requirements of typical satellite systems for targeted signal monitoring.
A safety helmet detection model has been developed based on the original YOLOv5 network model, which addresses the problems of low accuracy and poor robustness for small targets in complex natural environments. Firstly, we add the layer to detect small targets, resulting in the shallow detection scale, which makes the model better able to detect small objects. Secondly, introducing the transformer self-attention module in the backbone network, which can extract global features with rich semantic information, improves the detection performance. Thirdly, The Ghost module demands fewer parameters and less computational complexity for features to be generated, reducing the pressure to learn model parameters. Finally, the loss function EIoU Loss is applied to improve the detection accuracy and accelerate the convergence of the model. Experimental results show that the average accuracy means (MAP), accuracy rates (P), and recall rates (R) of the improved algorithm are increased by 1.2%, 0.4%, and 0.2%, respectively while reducing the model size, which proves that the algorithm is more effective than the original algorithm at detecting safety helmets.
The rise and development of software-defined networking (SDN) has brought a new development direction to the Internet. As a new network architecture, SDN has better flexibility and programmability than traditional network architectures, which brings new opportunities for the development of traffic management. In this paper, the development and trend of SDN are firstly introduced, and then some representative works of SDN in data center traffic management are analyzed. Finally, some research results related to traffic management based on the combination of SDN and artificial intelligence (AI) technology are expounded, and the challenges and development prospects of SDN traffic management are pointed out.
The development of 5G networks has brought higher requirements for services such as switching, routing, and QoS (Quality of Service) of the carrying network and the core network. Packet classification is an essential technology which plays an important role in guaranteeing the high bandwidth and low latency of the network. Many hardware-based algorithms have been presented to meet the high requirements. However, the resource overhead on chip become greater and greater as the network scale expands so fast that it is urgent to improve the storage efficiency. Thus, an effective classification method based on extended bloom filter and cuckoo hash is proposed in this paper. It preserves the feature of the extended bloom filter by locating entries according to the comparison result of multiple hash counter groups, and further introduces the hash index comparison which slows down the increase of the counter groups greatly. In addition, a cuckoo hash with limited number of collisions and a reasonable deletion mechanism are also presented in this paper to ensure higher stability. The experimental results show that the method has the advantages of simple structure, wire-speed lookup capability, high stability, and high storage utilization.
