No abstract

No abstract

In this paper, a new methodology for choosing design parameters of level-crossing analog-to-digital converters (LC-ADCs) is presented that improves sampling accuracy and reduces the data stream rate. Using the MIT-BIH Arrhythmia dataset, several LC-ADC models are designed, simulated and then evaluated in terms of compression and signal-to-distortion ratio. A new one-dimensional convolutional neural network (1D-CNN) based classifier is presented. The 1D-CNN is used to evaluate the event-driven data from several LC-ADC models. With uniformly sampled data, the 1D-CNN has 99.49%, 92.4% and 94.78% overall accuracy, sensitivity and specificity, respectively. In comparison, a 7-bit LC-ADC with 2385Hz clock frequency and 6-bit clock resolution offers 99.2%, 89.98% and 91.64% overall accuracy, sensitivity and specificity, respectively. It also offers 3x data compression while maintaining a signal-to-distortion ratio of 21.19dB. Furthermore, it only requires 49% floatingpoint operations per second (FLOPS) for cardiac arrhythmia classification in comparison with the uniformly sampled ADC.Finally, an open-source event-driven arrhythmia database is presented.

In neural spike sorting systems, the performance of the spike detector has to be maximized because it affects the performance of all subsequent blocks. Non-linear energy operator (NEO), is a popular spike detector due to its detection accuracy and its hardware friendly architecture. However, it involves a thresholding stage, whose value is usually approximated and is thus not optimal. This approximation deteriorates the performance in real-time systems where signal to noise ratio (SNR) estimation is a challenge, especially at lower SNRs. In this paper, we propose an automatic and robust threshold calculation method using an empirical gradient technique. The method is tested on two different datasets. The results show that our optimized threshold improves the detection accuracy in both high SNR and low SNR signals. Boxplots are presented that provide a statistical analysis of improvements in accuracy, for instance, the 75th percentile was at 98.7% and 93.5% for the optimized NEO threshold and traditional NEO threshold, respectively.

No abstract

Real time on-chip spike detection is the first step in decoding neural spike trains in implantable brain machine interface systems. Nonlinear Energy Operator (NEO) is a transform widely used to distinguish neural spikes from background noise. In this paper we define a general form of energy operators, of which NEO is a specific example, which gives better spike-noise separation than NEO and its derivatives. This is because of a non-linear scaling applied to the general discrete energy operator. Using two well-known publically available datasets, the performance of several operators is compared. On data sets that contain multi-unit spikes with low Signal to Noise ratio, the detection accuracy was improved by approximately 15%.

High-density, intracranial recordings from micro-electrode arrays need to undergo Spike Sorting in order to associate the recorded neuronal spikes to particular neurons. This involves spike detection, feature extraction, and classification. To reduce the data transmission and power requirements, on-chip real-time processing is becoming very popular. However, high computational resources are required for classifiers in on-chip spike-sorters, making scalability a great challenge. In this review paper, we analyze several popular classifiers to propose five new hardware architectures using the off-chip training with on-chip classification approach. These include support vector classification, fuzzy C-means classification, self-organizing maps classification, moving-centroid K-means classification, and Cosine distance classification. The performance of these architectures is analyzed in terms of accuracy and resource requirement. We establish that the neural networks based Self-Organizing Maps classifier offers the most viable solution. A spike sorter based on the Self-Organizing Maps classifier, requires only 7.83% of computational resources of the best-reported spike sorter, hierarchical adaptive means, while offering a 3% better accuracy at 7 dB SNR.

No abstract