DSP Network Research Articles

Search and hunt based-driving scene perception network for distance estimation and object detection

Visual tracking is generally utilized in numerous applications including surveillance and autonomous driving to provide a wide angle of vision. However, different types of alteration, including occlusion, pose, and illumination present in the video sequence act as one of the crucial troubles considered in visual tracking. Hence, this research developed a modified Deep learning (DL) approach useful for distance estimation, object detection, and pixel-wise semantic segmentation by shared convolutional architecture. The proposed object detection model employs a modified Driving Scene Perception network (DSP), which makes use of multi-task learning to enhance the efficacy of distance evaluation, semantic segmentation, and object detection. The prime significance of the research relies on the Search and Hunt optimization (SaHO) algorithm, which optimally selects the hyperparameter value of the DSP network. The experimental analysis substantiates that the proposed modified DSP network effectively estimates the distance between the moving object and the camera. The experimental analysis depicts the effectiveness of the DSP with SaHO model for both segmentation and object detection. For object detection, the DSP with SaHO model attained high rates of 0.98, 0.94, and 0.96 for accuracy, specificity, and sensitivity respectively utilizing the cityscape dataset. Additionally, the DSP with SaHO approach attained the lowest errors for segmentation and surpassed other existing techniques that revealed the efficiency of the DSP with the SaHO model.

An efficient DSP network for the real-time auralization of complex urban scenarios

The most commonly used technique for auralization is a simulation step to determine sound propagation based on Geometrical Acoustics and a subsequent assembly of the simulation results into a filter or (binaural) impulse response. Convolution is used to apply the propagation effects to an anechoic recording per source-receiver pair, which is efficient for a scenario including a few sound sources and many propagation paths. The procedure offers a good balance of speed and accuracy, and has seen much development. However, it reveals an intrinsic difficulty upon scene adaption (filter exchange) for fast-moving sound sources, because in principle it requires a steady-state environment. In particular, this constellation is found in urban scenarios, where many fast-moving sound sources can be expected while essentially only a manageable number of propagation paths contribute to the perceived overall sound field. An efficient DSP network is put to discussion that handles propagation paths individually using a single-input multiple-output variable delay line per sound source and applies spectral influences like diffraction effects using short IIR filters. Binaural cues are maintained by a clustering approach that combines incident waves and offers a constant convolution cost.The most commonly used technique for auralization is a simulation step to determine sound propagation based on Geometrical Acoustics and a subsequent assembly of the simulation results into a filter or (binaural) impulse response. Convolution is used to apply the propagation effects to an anechoic recording per source-receiver pair, which is efficient for a scenario including a few sound sources and many propagation paths. The procedure offers a good balance of speed and accuracy, and has seen much development. However, it reveals an intrinsic difficulty upon scene adaption (filter exchange) for fast-moving sound sources, because in principle it requires a steady-state environment. In particular, this constellation is found in urban scenarios, where many fast-moving sound sources can be expected while essentially only a manageable number of propagation paths contribute to the perceived overall sound field. An efficient DSP network is put to discussion that handles propagation paths individually using a si...

Asic Implementation of 12-Bit Radix-8 Booth Multiplier

Multipliers are playing a vital role in DSP and Neural Networks applications. Many methods have been introduced to work on multipliers that offer high speed, less power consumption and reduced area. Booth Algorithm demonstrates an efficient way of signed binary multiplication. In this paper, physical design of 12-bit radix-8 booth multiplier for signed multiplication is presented with an aim to improve the performance metrics such as power, area and delay. The performance of 12-bit radix-8 booth multiplier is compared with the 64-bit radix-16 booth multiplier.

Research of Acoustic Emission Monitoring Wireless Sensor Network System Based on DSP

Taking crop Acoustic emission (AE) signals as the main monitoring index for plant diseases or dehydration, the crop AE wireless sensor network monitoring system was established based on DSP and ZigBee network. Crop AE signals were conducted on A/D transformation and data processing and then sent to the monitoring center based on wireless communication module SZ05. AE signals received at the monitoring center were demonstrated in waveform display. Based on the variation of AE signal’s frequency the system can diagnose disease and prevent dehydration for crops.

Design of Temperature Control System Based on DSP and Fuzzy Neural Network

Using fuzzy neural network to tune PID parameters, and DSP as processor, it was designed that a set of electric boiler temperature control system based on PID parameters self-tuning, including the design of each hardware module and each software subroutine of the system. Experimental results show that compared with the traditional PID temperature control system, this temperature control system has the advantages such as good control effect, easy parameter adjustment, strong anti-jamming capability, better adaptability and robustness, has the feasibility and practical value.

Continuations and transducer composition

On-line transducers are an important class of computational agent; we construct and compose together many software systems using them, such as stream processors, layered network protocols, DSP networks and graphics pipelines. We show an interesting use of continuations, that, when taken in a CPS setting, exposes the control flow of these systems. This enables a CPS-based compiler to optimise systems composed of these transducers, using only standard, known analyses and optimisations. Critically, the analysis permits optimisation across the composition of these transducers, allowing efficient construction of systems in a hierarchical way.

A Hierarchical Control Architecture for High-Speed Visual Servoing

The efferent signal is the impulses from the brain to muscle or organ tissue. The afferent signal is the sensation that transmits the state of peripheral body parts to the brain. The motor control architectures of biological systems have hierarchical structures in which the efferent/afferent signals interact. Thanks to this architecture, flexible and reflective action is realized. In this paper, we propose a hierarchical control architecture for high-speed visual servoing on the basis of a biological signal interaction model. The proposed architecture has three modules: servo, motion planner and adaptation. The afferent signal corresponds to the feedback signal from the sensors; the efferent signal corresponds to the motion command. These signals interact in a hierarchical manner that realize a parameter adaptation mechanism. A series of dynamical tasks, tracking/grasping/handling of a moving object, is implemented as an example of high-speed visual servoing. The system contains a DSP network, high-speed active vision, dextrous hand and a seven-degrees-of-freedom manipulator. Real-time experiments are conducted and the results exhibit the responsiveness and flexibility of the proposed hierarchical architecture.

The COMPASS RICH-1 read-out system

This paper describes the reconfigurable read-out system for the 82944 RICH-1 channels of the COMPASS experiment (NA58) at CERN. The system is based on 192 identical large front-end boards (BORA board). BORA was designed for acquiring, digitizing, threshold subtracting and transmitting event data. The overall operation of the board is controlled and supervised by a DSP tightly interacting with an FPGA that acts as a parallel co-processor. The DSP allows characterizing each analog channel by locally calculating noise and pedestal. Each BORA communicates with the outside world through two optical fibers and through a dedicated DSP network. One optical fiber is used to receive event triggers, and the other one is used to transmit event data to subsequent processing stages of the acquisition system. The DSP network allows reconfiguring and reprogramming the DSPs and FPGAs as well as acquiring sample events to visualize the overall operation of the system. The whole RICH has eight DSP networks working in parallel. These networks are handled by DOLINA, a PC resident multiprocessor board containing eight DSPs. Each network is formed by 24 BORA DSPs and 1 DOLINA DSP. The read-out system can steadily work up to a trigger rate of 75 kHz with maximum pixel occupancy of 20%, reaching a transmission data rate of 5.13 Gbytes/s.

Adaptive Generalized Predictive Control Algorithm Implemented Over an Heterogeneous Parallel Architecture

In this paper a parallel implementation of an Adaptive Generalized Predictive Control (AGPC) algorithm is presented. Since the AGPC algorithm needs to be fed with knowledge of the plant transfer function, the parallelization of a standard Recursive Least Squares (RLS) estimator and a GPC predictor is discussed here. Also, since a matrix inversion operation is required in the GPC predictor algorithm, special attention is given to its parallelization. A small DSP network with up to 3 processors is used to investigate, the performance of the parallel implementation. To exploit an heterogeneous architecture the parallel algorithm is mapped over a network builded up of transputers as communication elements, and DSPs as computing elements. Further some heterogeneous topologies are compared. Execution times and efficiency results of the RLS and GPC steps are presented to show the performance of the parallel algorithm, over different topologies.

Implementation of an Adaptive Generalized Predictive Control Algorithm Over an Heterogeneous Parallel Architecture

In this paper a parallel implementation of an Adaptive Generalized Predictive Control (AGPC) algorithm is presented. Since the AGPC algorithm needs to be fed with knowledge of the plant transfer function, the parallelization of a standard Recursive Least Squares (RLS) estimator and a GPC predictor is discussed here. Also, since a matrix inversion operation is required in the GPC predictor algorithm, special attention is given to its parallelization. A small DSP network with up to 2 processors is used to investigate, the performance of the parallel implementation. To exploit an heterogeneous architecture the parallel algorithm is mapped over a network builded up of transputers as communication elements, and DSPs as computing elements. Further some heterogeneous topologies are compared. Execution times and efficiency results of the RLS and GPC steps are presented to show the performance of the parallel algorithm, over different topologies.

The Cocktail Party Listener

A complex computational model of the human ability to listen to certain signals in preference of others, also called the cocktail party phenomenon, is built on the basis of surveys into the relevant psychological, DSP, and neural network literature. This model is basically binaural and as such it makes use of both spectral data and spatial data in determining which speaker to listen to. The model uses two neural networks for filtering and speaker identification. Results from some experimentation with type and architecture of these networks are presented along with the results of the model. These results indicate that the model has a distinctive ability to focus on a particular speaker of choice.