A multichannel adaptive integrated MEMS/CMOS microphone

麦克风有许多不同类型：它们的主要功能是转换：将压力波（在一定频率范围内）转换为单个电信号，通常尽可能精确。在此之后，信号可以用于记录或解释（这是我们在这里感兴趣的）。解释中的一个主要问题是，信号在听觉频谱的某些部分可能具有大量的能量，但在其他部分则少得多，并且这种分布可能会迅速改变。通常，正是这些较低能量区域的能量对解释至关重要。目前的做法是对来自麦克风的单个电信号进行滤波（无论是使用FFT还是带通滤波器），然后检查这样产生的信号。我们提出了一种不同的方法，其中压力波被直接转换成多个电信号，对应于可听频谱的不同部分。通过激活换能器（即为它们提供快速调节的增益控制），我们将能够在额外灵敏度可用于解释任务的那些区域中增加滤波器的灵敏度，并在信号非常强的那些区域中降低灵敏度。动物承担的听觉解释任务（在混响环境中有多个声源时解决什么和在哪里的任务-通常情况下-是自主机器人的听觉系统需要承担的任务。动物听觉系统包括多个换能器，并为频谱的不同部分提供多种输出，同时动态调整其灵敏度和选择性。当前的麦克风提供单个电输出，然后将其处理成多个带通流（保持精确定时），或者处理成基于FFT的向量序列，例如倒谱系数（损失定时精度）。所提出的有源MEMS麦克风在转换时执行频谱击穿，提供固有的并行输出，同时保持精确的定时。此外，它是适应性的。这种自适应能力在当前麦克风中不存在，但在助听器中很重要。精确的定时信息对于使用耳间时间和电平差的源方向识别是重要的。在有多个有源声源的情况下，准确的前景声源解释需要一定程度的声音流，需要能够检查声音的特征，通常在能量相对较低的频谱区域。有源MEMS带通麦克风将由一个膜组成，该膜将由于外部压力波而振动。膜物理地连接到MEMS结构中的不同谐振元件（杆）-这些元件将具有一定范围的谐振频率。此外，这些条将充当MOS晶体管的栅极，导致它们的振动调制通过这些晶体管的电流。调制的电流将被编码为一组尖峰序列，并且这些尖峰被处理以提供信号，从而通过使用静电效应来改变晶体管对条的振动的响应来调节每个谐振器的灵敏度。调制将用于调整增益，使得频谱的安静区域被选择性地放大，而频谱的大声区域被选择性地衰减。通过这种方式，将有可能构建集成MEMS/CMOS麦克风，其可以在放大频谱的安静区域的同时衰减频谱的大声区域。尖峰编码输出将以与地址事件表示（AER）兼容的方式提供，使其与其他实验室现有和拟议的神经形态芯片兼容。

There are many different types of microphones: their primary function is transduction: converting pressure waves (within some range of frequencies) into a single electrical signal, usually as precisely as possible. After this, the signal may be used for recording or for interpretation (which is our interest here). A major problem in interpretation is that the signal may have a large amount of energy in some parts of the auditory spectrum, but much less in others, and that this distribution may alter rapidly. Often, it is the energy in these lower energy areas that is critical for interpretation. Current practice is to filter the single electrical signal from the microphone (whether using FFTs, or bandpass filters), then examine the signal so produced. We propose a different approach in which the pressure wave is directly transduced into multiple electrical signals, corresponding to different parts of the audible spectrum. By making the transducers active (i.e. providing them with a rapidly adjusting gain control), we will be able to increase the sensitivity of the filters in those areas where additional sensitivity can be useful in the interpretation task, and reduce the sensitivity in those areas where the signal is very strong. The auditory interpretation tasks undertaken by animals (solving the what and where tasks when there are - as is normally the case - multiple sound sources in a reverberant environment) is the same task that an autonomous robot's auditory system needs to undertake. Animal hearing systems include multiple transducers, and provide numerous outputs for different parts of the spectrum, whilst adjusting their sensitivity and selectivity dynamically. Current microphones provide a single electrical output, which is then either processed into a number of bandpass streams (maintaining precise timing), or into a sequence of FFT-based vectors, such as cepstral coefficients (losing timing precision). The proposed active MEMS microphone performs the spectral breakdown at transduction, providing an inherently parallel output whilst maintaining precise timing. Further, it is adaptive. This adaptive capability, non-existent in current microphones is important in hearing aids. Precise timing information is important for source direction identification using inter-aural time and level differences. Where there are multiple active sources, accurate foreground source interpretation requires some degree of sound streaming, requiring the ability to examine features of the sound, often in spectral areas which with relatively low energy.The active MEMS bandpassing microphone will consist of a membrane which will vibrate due to the external pressure wave. The membrane is physically linked to different resonant elements (bars) in the MEMS structure - these elements will have a range of resonant frequencies. Further, these bars will act as gates for MOS transistors, resulting in their vibration modulating the current passing through these transistors. The modulated current will be coded as a set of sequences of spikes, and these spikes processed to provide a signal to adjust the sensitivity of each of the resonators by using an electrostatic effect to change the response of the transistors to the vibration of the bars. The modulation will be used to adjust the gain so that quiet areas of the spectrum are selectively amplified and loud areas of the spectrum selectively attenuated. In this way, it will be possible to build an integrated MEMS/CMOS microphone which can attenuate loud areas of the spectrum concurrently with amplifying quiet areas of the spectrum. The spike coded output will be made available in a way compatible with the address-event representation (AER), making it compatible with existing and proposed neuromorphic chips form other laboratories.