Hongda Wang

Bio: Hongda Wang is an academic researcher from The Chinese University of Hong Kong. The author has contributed to research in topics: Time domain & Field-programmable gate array. The author has an hindex of 2, co-authored 10 publications receiving 38 citations.

Hardware Design of Real Time Epileptic Seizure Detection Based on STFT and SVM

Abstract: Closed-loop stimulation of many neurological disorders, such as epilepsy, is an emerging technology and regarded as a promising alternative for surgical and drug treatment. In this paper, a real-time seizure detection algorithm based on STFT and support vector machine (SVM) and its field-programmable gate array (FPGA) implementation are proposed. With a two-stage patient-specific channel selection and feature selection mechanism, those redundant and uncorrelated spectral features are removed from the entire feature set. The evaluation results on CHB-MIT epilepsy database show that the mean detection latency of the proposed algorithm is 6 s, the sensitivity is 98.4%, and the false detection rate is 0.356/h. The performance of our proposed algorithm is comparable to other existing seizure detection algorithms. Moreover, we implement the proposed seizure detection algorithm on Xilinx Zynq-7000 XC7Z020 with high level synthesis. Each classification of the input electroencephalography signal can be finished within 313 $\mu \text{s}$ , and the power consumption of the programmable logic is only 380 mW at 100 MHz. In hardware implementation, an optimization strategy for the nested-loop structure within nonlinear SVM is proposed to improve pipeline efficiency. Compared with existing method, the experimental result shows that our method can speed up the nonlinear SVM by $1.70\times $ , $1.53\times $ , $1.37\times $ , and $1.26\times $ with the unroll factor equal to 1–4 at the same DSP utilization rate. The evaluation results affirm the possibility of integrating the proposed algorithm and FPGA implementation into a wearable seizure control device.

Integrating channel selection and feature selection in a real time epileptic seizure detection system

Abstract: Automated real time seizure detection is difficult since detection sensitivity, false detection rate and seizure onset detection latency need to be considered simultaneously. Traditional pattern recognition and classification system usually suffers huge performance variation due to patient specificity and algorithm inadaptability. To address this problem, we propose a two stage seizure detection system which integrates off-line channel selection and feature selection before the construction of the final model. This system allows patient specific channel selection and flexible feature set extraction for individual patient, so that a more compact and reliable model could be developed. Employing the two stage scheme not only decreases hardware cost in signal readout and feature extraction, but also remarkably improves detection sensitivity and reduces false detections. Mutual information based method is used for channel selection, while Random Forests and nonlinear SVM-RFE are evaluated for feature selection. The whole system achieves a mean detection latency of 6 seconds and a false detection rate of 0.356 per hour. Based on the test dataset, the sensitivity is found to 74.2% by sample or 98.4% by record with only two detection misses. Our design is also hardware-friendly, which could be implemented as a single chip closed loop neural modulation system.

BOTDA Fiber Sensor System Based on FPGA Accelerated Support Vector Regression

Abstract: Brillouin optical time domain analyzer (BOTDA) fiber sensors have shown strong capability in static long haul distributed temperature/strain sensing. However, in applications such as structural health monitoring and leakage detection, real-time measurement is quite necessary. The measurement time of temperature/strain in a BOTDA system includes data acquisition time and post-processing time. In this article, we propose to use hardware accelerated support vector regression (SVR) for the post-processing of the collected BOTDA data. Ideal Lorentzian curves under different temperatures with different linewidths are used to train the SVR model to determine the linear SVR decision function. The performances of SVR are evaluated under different signal-to-noise ratios (SNRs) experimentally. After the model coefficients are determined, algorithm-specific hardware accelerators based on field-programmable gate arrays (FPGAs) are used to realize SVR decision function. During the implementation, hardware optimization techniques based on loop dependence analysis and batch processing are proposed to reduce the execution latency. Our FPGA implementations can achieve up to $42\times $ speedup compared with software implementation on an i7- $5960x$ computer. The post-processing time for 96 100 Brillouin gain spectrums (BGSs) along with 38.44-km fiber under test (FUT) is only 0.46 s with FPGA board ZCU104, making the post-processing time no longer a limiting factor for dynamic sensing. Moreover, the energy efficiency of our FPGA implementation can reach up to $226.1\times $ higher than the software implementation based on CPU.

Systolic array based VLSI architecture for high throughput 2-D discrete wavelet transform

Abstract: A new data scan method is proposed for 2-D discrete wavelet transform to access more pixels in one clock cycle. Unlike existing stripe based method, in our design adjacent even and odd rows are read and processed at the same time. The concurrent output from even and odd row transform units inherently eliminate the data sequencing between row transform and column transform. Thus the transposition memory is not needed any more. For the row transform unit, a novel systolic array structure is constructed with pipeline technique employed to reduce the critical path delay. Without too many additional registers, the improved critical path delay of Tm is superior to most of the stripe based designs. For the column transform unit, a conventional lifting based two-input/two-output structure is adopted. Theoretical analysis shows that this design is suitable for applications which have demanding throughput rate and high operation frequency requirements. Synthesis results in UMC 130nm process show that the Area Delay Product is 23%, 27.3% and 29.6% better than the best existing stripe based structure for S=2, 4 and 8.

Hardware acceleration of support vector machine based on high level synthesis

Abstract: Hardware accelerator design of machine learning algorithms is of great significance since huge amount of data are generated continuously and the computation load keeps increasing fast. Traditional solutions which rely on general purpose processors like X86 CPUs and ARM embedded processors cannot achieve high computation performance and energy efficiency due to lack of adaptability to specific algorithms. Domain-specific hardware accelerator is a promising alternative solution since the hardware is specifically designed to deal with one type of problems, the computation performance and energy efficiency can outperform traditional solutions by more than one order of magnitude. In this paper, we have proposed a hardware accelerator for support vector machine based on high level synthesis. Through the proposed loop dependence analysis and batch processing method, the speedup can reach about 153x on Xilinx ZC706 with unroll factor of 160, which is very close to its limit. This hardware implementation can be used in data centers as a coprocessor to accelerate the data processing speed in many classification applications.

Patient-Specific Seizure Detection Using Nonlinear Dynamics and Nullclines

Abstract: Nonlinear dynamics has recently been extensively used to study epilepsy due to the complex nature of the neuronal systems. This study presents a novel method that characterizes the dynamic behavior of pediatric seizure events and introduces a systematic approach to locate the nullclines on the phase space when the governing differential equations are unknown. Nullclines represent the locus of points in the solution space where the components of the velocity vectors are zero. A simulation study over 5 benchmark nonlinear systems with well-known differential equations in three-dimensional exhibits the characterization efficiency and accuracy of the proposed approach that is solely based on the reconstructed solution trajectory. Due to their unique characteristics in the nonlinear dynamics of epilepsy, discriminative features can be extracted based on the nullclines concept. Using a limited training data (only 25% of each EEG record) in order to mimic the real-world clinical practice, the proposed approach achieves 91.15% average sensitivity and 95.16% average specificity over the benchmark CHB-MIT dataset. Together with an elegant computational efficiency, the proposed approach can, therefore, be an automatic and reliable solution for patient-specific seizure detection in long EEG recordings.

Hardware Design of Real Time Epileptic Seizure Detection Based on STFT and SVM

Abstract: Closed-loop stimulation of many neurological disorders, such as epilepsy, is an emerging technology and regarded as a promising alternative for surgical and drug treatment. In this paper, a real-time seizure detection algorithm based on STFT and support vector machine (SVM) and its field-programmable gate array (FPGA) implementation are proposed. With a two-stage patient-specific channel selection and feature selection mechanism, those redundant and uncorrelated spectral features are removed from the entire feature set. The evaluation results on CHB-MIT epilepsy database show that the mean detection latency of the proposed algorithm is 6 s, the sensitivity is 98.4%, and the false detection rate is 0.356/h. The performance of our proposed algorithm is comparable to other existing seizure detection algorithms. Moreover, we implement the proposed seizure detection algorithm on Xilinx Zynq-7000 XC7Z020 with high level synthesis. Each classification of the input electroencephalography signal can be finished within 313 $\mu \text{s}$ , and the power consumption of the programmable logic is only 380 mW at 100 MHz. In hardware implementation, an optimization strategy for the nested-loop structure within nonlinear SVM is proposed to improve pipeline efficiency. Compared with existing method, the experimental result shows that our method can speed up the nonlinear SVM by $1.70\times $ , $1.53\times $ , $1.37\times $ , and $1.26\times $ with the unroll factor equal to 1–4 at the same DSP utilization rate. The evaluation results affirm the possibility of integrating the proposed algorithm and FPGA implementation into a wearable seizure control device.

Application of Raman and Brillouin Scattering Phenomena in Distributed Optical Fiber Sensing

Abstract: We present a review of the basic operating principles and measurement schemes of standalone and hybrid distributed optical fiber sensors based on Raman and Brillouin scattering phenomena. Such sensors have been attracting a great deal of attention due to the wide industrial applications they offer, ranging from energy to oil & gas, from transportation to structural health monitoring. In distributed sensors, the optical fiber itself acts as a sensing element providing unique measurement capabilities in terms of sensing distance, spatial resolution and number of sensing points. The most common configuration exploits optical time domain reflectometry in which optical pulses are sent along the sensing fiber and the backscattered light is detected and processed to extract physical parameters affecting its intensity, frequency, phase, polarization or spectral content. Raman and Brillouin scattering effects allow the distributed measurement of temperature and strain over tens of kilometers with meter-scale spatial resolution. The measurement is immune to electromagnetic interference, suitable for harsh environments and highly attractive whenever large industrial plants and infrastructures have to be continuously monitored to prevent critical events such as leakages in pipelines, fire in tunnels, structural problems in large infrastructures like bridges and rail tracks. We discuss the basic sensing mechanisms based on Raman and Brillouin scattering effects used in distributed measurements, followed by configurations commonly used in optical fiber sensors. Hybrid configurations which combine Raman and Brillouin-based sensing for simultaneous strain and temperature measurements over the same fiber using shared resources will also be addressed. We will also discuss advanced techniques based on pulse coding used to overcome the tradeoff between sensing distance and spatial resolution affecting both types of sensors, thereby allowing measurements over tens of kilometers with meter-scale spatial resolution.