Hardware Realization of Neural Oscillator with Learning Capability

在这项研究中，我提出了一种使用同步微扰优化方法的脉冲耦合振荡器学习方案及其硬件实现。由于脉冲耦合振荡器是一种递归神经网络，通常的优化方法很难找到合适的参数值。同时摄动法给出了一个简单的解。此外，该方法适合硬件实现。从这个角度出发，我提出并制作了硬件脉冲耦合振荡器和具有学习能力的递归神经网络。首先，通过C语言和MatLab仿真验证了所提出的具有学习能力的脉冲耦合振荡器的可行性。脉冲耦合振荡器通过学习过程产生具有期望间隔的脉冲序列。神经网络的硬件实现是一个有趣的问题。主要有两种方法……数字实现和模拟实现。作为前一种方法，现场可编程门阵列（FPGA）是一个有用的目标。利用VHDL语言设计了具有学习能力的脉冲耦合振荡器。然后将设计结果配置到FPGA上。验证了FPGA脉冲耦合振荡器系统的运行，具有学习能力。得到合适的脉冲序列。第二种方法是模拟实现。然后，采用现场可编程模拟阵列（FPAA）实现振荡器。对系统进行了模拟电路设计。电路运行通过电路模拟器进行验证。该系统通过同步摄动方法实现具有学习能力的脉冲耦合振荡器。接下来，将设计配置为FPAA。该电路实现了脉冲耦合振荡器。产生的脉冲序列的间隔取决于脉冲耦合振荡器中包含的参数。此外，采用同步摄动法在FPGA上实现了一些具有学习能力的递归神经网络。Hopfield网络和双向联想记忆是循环网络的典型例子。通常，这些具有学习能力的硬件递归神经网络系统很难实现。然而，我使用同时摄动法实现了Hopfield神经网络系统和具有学习能力的双向神经网络系统。我展示了这些系统的一些应用。因此，我可以通过同时摄动的方法来证实具有学习能力的脉冲耦合振荡器的有效性和可行性。制作了这些系统，并对其运行情况进行了测试。少

In this research, I propose a learning scheme for pulse coupled oscillators using the simultaneous perturbation optimization method and its hardware implementation. It was difficult and complicated for usual optimization method to find proper parameter values of the pulse coupled oscillator, since the oscillator is a kind of recurrent neural network. The simultaneous perturbation method gives a simple solution. Moreover, this approach is suitable for hardware realization. From this point of view, I proposed and fabricated the hardware pulse coupled oscillator and recurrent neural network with learning ability via the simultaneous perturbation method.First of all, I confirm feasibility of the proposed pulse coupled oscillator with learning capability through simulation by C language and MatLab. The pulse coupled oscillator can generate pulse train with desired interval through leaning process.Hardware realization of neural networks is an interesting issue. Mainly there are two approache … More s ; digital realization and analog one. As the former approach, field programmable gate array(FPGA) is useful target. I designed the pulse coupled oscillator with learning capability by VHDL. Then, design result is configured on FPGA. I verified the operation of the FPGA pulse coupled oscillator system with learning ability. Proper pulse train is obtained.The second approach is analog realization. Then, field programmable analog array(FPAA) is adopted to implement the oscillator. Analog circuit design of the system is carried out. The circuit operation is confirmed by a circuit simulator. The system could realize the pulse coupled oscillator with learning capability via the simultaneous perturbation method. Next, the design is configured to FPAA. The circuit realized the pulse coupled oscillator. Interval of generated pulse train varied depending on parameters contained in the pulse coupled oscillator.Moreover, some recurrent neural networks with learning capability were realized by FPGA using the simultaneous perturbation method. Hopfield network and bidirectional associative memory are typical examples of the recurrent networks. Usually, it was difficult to realize these hardware recurrent neural network systems with learning capability. However, I implemented the Hopfield neural network system and bidirectional neural network system with learning ability using the simultaneous perturbation method. I showed some application of these systems.As a result, I could confirm a validity and feasibility of the pulse coupled oscillator with learning capability via the simultaneous perturbation method. These systems were fabricated and tested the operation of these systems. Less