Graph-based Detector Research Articles

DenseSphere: Multimodal 3D object detection under a sparse point cloud based on spherical coordinate

Multimodal 3D object detection has gained significant attention due to the fusion of light detection and range (LiDAR) and RGB data. Existing 3D detection models in autonomous driving are typically trained on dense point cloud data from high-specification LiDAR sensors. However, budgetary constraints often lead to adopting low point-per-second (PPS) LiDAR sensors in real-world scenarios. The low PPS specification can generate a sparse point cloud. In this case, the existing models trained on dense data with a high PPS specification cannot achieve optimal performance under a sparse point cloud. To address this problem, we propose DenseSphere for robust multimodal 3D object detection under a sparse point cloud. Considering the data acquisition process of LiDAR sensors, DenseSphere involves the spherical coordinate-based point upsampler. Specifically, points are interpolated in the horizontal or vertical direction using bilateral interpolation. The interpolated points are refined using dilated pyramid blocks with various receptive fields. For efficient fusion with generated dense point cloud, we use a graph-based detector and hierarchical layers. Then, we demonstrate the performance of DenseSphere by comparing it with other multimodal 3D object detection models through experiments. The visual results and source code with the pretrained models are available at https://github.com/Jung-jongwon/DenseSphere.

Improved 2-D Graph-Based Detectors for 2-D Interference Channels

The 2-D interference channels, which consist of intersymbol interference as well as intertrack interference (ITI), appear in high-density recording channels. A graph-based detector has received much attention since it is a real 2-D detector, which can recover data from multiple tracks or each 2-D page simultaneously. In this paper, we propose two approaches to improve the performances of graph-based detectors. For the first method (M1), which processes data on a 2-D page, the system equalizes 2-D interference channel into two 1-D targets, horizontal and vertical directions followed by two corresponding graph detectors. The second method (M2) is similar to the first one, but the 2-D channel is equalized into two 2-D targets with and without cornered ITIs, respectively, in order to better handle the effects of cornered ITIs. Compared with the full-graph detector, the method M1 offers reduced-complexity due the fewer number of edges from the factor nodes. The simulation results on the 3 × 3 channel matrix with 2% cornered ITI show that the proposed methods achieve the gains of about 1.0 and 1.7 dB, respectively, at the BER of 10 -5 over the full graph detector. As the cornered ITI levels increase, the method M2 gives the larger gains over the others. When the media noise is accounted for, the method M2 still gives the best performances. Finally, the mutual information of the three graph detectors for the 2-D channels is computed and compared.

Investigation of Graph-Based Detection in Iterative Decoding for Bit-Patterned Media Recording

High-density bit-patterned media recording (BPMR) can be obtained by reducing the spacing between data bitislands in both the along-and across-track directions, thus leading to severe intersymbol interference (ISI) and intertrack interference (ITI) because of small bit and track pitches, respectively. Here, we propose to use the graph-based detector, instead of the trellis-based detector, in iterative decoding to combat the ISI and the ITI for a multi-head multi-track BPMR system. Specifically, the readback signal is sent to the graph-based detector before iteratively exchanging the soft information with a decoder. Experimental results indicate that at low to moderate complexity, the proposed scheme outperforms the existing schemes, especially at high recording density.

Application and Optimization of Factor Graph-Based Detector on 1D ISI Magnetic Recording Channel

This paper introduces the application of a factor graph (FG)-based detector on a one-dimensional (1D) intersymbol interference (ISI) magnetic recording channel. The conventional Bahl-Cocke-Jelinek-Raviv (BCJR) detection algorithm operates on a trellis, which defines an equivalence between a bit sequence and a path through the trellis, as well as all possible time evolutions of states of the channel. The BCJR algorithm outputs soft information which corresponds to the a-posteriori channel bit probabilities. Unlike the BCJR, the FG-based detector employs the use of the sum product algorithm (SPA) on a FG with the objective of calculating the same a-posteriori probabilities. Employing the binary SPA to this FG detector gives rise to a suboptimal performance due to the presence of short cycles within the channel factor graph. A nonbinary implementation of the same algorithm mitigates the effect of these cycles and thereby results in an improved performance as compared to the binary implementation. In this paper we also propose an alternative method that can be used to reduce the number of these cycles for performance enhancement. This method involves using an equalizer with a generalized partial response (GPR) target.

Modified Graph-Based Detection Methods for Two-Dimensional Interference Channels

Two-dimensional (2-D) interference channels with inter-symbol interference (ISI) and inter-track interference (ITI) exist in the magnetic recording systems at high areal density. A number of 2-D detection methods have recently been proposed for the multi-track processing of the 2-D channels. Graph-based detector with the belief propagation algorithm appears as an alternative method, but at a degraded performance and high complexity level. In this work, we propose two methods to modify the graph-based (GB) detector. One applies a serial scheduling to the GB detector, while the other modifies the GB detection by ignoring some connections during one direction of the reliability updates in the factor graph leading to the reduction of short cycles. The simulation results show that the proposed GB detectors give better bit error rate performances than the other GB detectors.