基于过采样的混合滤波器组性能优化

模拟误差是制约混合滤波器组信号重构精度的主要原因,如何降低由模拟误差导致的分解滤波器组系数误差成为了首要问题。引入过采样技术,研究在不同过采样率下8通道混合滤波器组的性能,寻找到过采样率最优值约等于7%。在不同模拟误差下对基于过采样的混合滤波器组的性能进行仿真,结果表明,原型结构和双阶型结构的混叠值相近,但后者对模拟误差的敏感度比前者大。信号重构阶段,在最小二乘法的基础上采用频带加权法进行误差校准。在1%模拟误差内,采用7%过采样率的原型混合滤波器组相比无过采样,平均混叠值下降了约50 dB,最大混叠值下降了约94 dB。仿真验证了引入过采样的有效性。     

基于加权表决的POF信号时钟数据恢复方法

针对POF数据传输特点,提出了一种基于加权表决的全数字时钟数据恢复(CDR)方法.引入数据窗口,消除了传统数字CDR电路中存在的恢复时钟1/N UI峰峰抖动,运用加权表决恢复数据,增强了电路抗干扰能力.该方法基于FPGA通过单PLL实现了100Mb/s的POF数据传输.仿真和实验测试结果表明,该方法可快速同步相位变化,电路输入抖动容限可达0.28UI.     

基于过采样的多种生物信息同步数据采集电路

提出一种新型的生物电放大器,巧妙地利用成形信号,不仅提高了过采样性能,还实现了多种生物医学信息的检测;文章以心电信号为例,同步检测了呼吸和导联脱落信号。实验表明该电路不仅能够满足人体生物医学信号动态监测的要求,还具有工艺性好、免调试、对器件性能要求低等一系列优点。该电路不仅可以用于生物电信号的检测,也可用于各种微弱信号的检测。     

基于3CNN-BiGRU的睡眠自动分期研究

针对单通道脑电信号睡眠自动分期效率和准确率问题,提出采用三尺度并行卷积神经网络提取睡眠信号特征和双向门控循环单元学习睡眠阶段之间内部时间关系的3CNN-BiGRU睡眠自动分期模型。首先对原始单通道脑电信号进行带通滤波处理,并采用合成少数类过采样技术进行类平衡,然后送入搭建的模型中进行训练和验证实验,其中采用预训练和微调训练对模型进行优化,采用10次和20次交叉验证提高训练可靠性。不同数据集下的不同模型对比实验结果表明,3CNN-BiGRU模型取得了更高的训练效率和更好的分期准确率。     

一种6倍无源增益低OSR低功耗的二阶NS SAR ADC

针对一阶噪声整形(NS)往往需要增加功耗而以较高的过采样比(OSR)来实现较高的有效位数(ENOB),提出了一种低OSR、低功耗的二阶无源NS SAR ADC。该无源NS模块较高的无源增益可以更好地抑制比较器的噪声;其残差电压是通过开关MOS阵列复用积分电容实现采样,从而无需额外的残差采样电容,避免了残差采样电容清零和残差采样时kT/C噪声的产生,因此减小了总的kT/C噪声。180 nm CMOS工艺仿真结果表明,在不使用数字校准的情况下,所设计的10位二阶无源NS SAR ADC电路以100 kS/s的采样率和5的OSR,实现了13.5位ENOB,电路功耗仅为6.98 μW。     

新型空间红外扫描传感器

针对天基红外扫描传感器,实例分析了SBIRS-HIGH和SPIRITⅢ传感器的特性.总结了新型天基红外扫描传感器系统设计的技术特点,包括多工作波段,时间延迟积分,过采样,大动态测量范围以及与凝视传感器相互切换等技术方面.     

一种基于多信道模型的IIR信道盲辩识算法

本文的主要贡献在于:1)证明了对单输入单输出(Single-Input-Single-Output SISO)IIR信道的输出进行过采样可以将其转化成单输入多输出(Single-Input-Multiple-Output SIMO)的多信道模型,不同的子信道具有相同的AR系数。2)由SISO向SIMO模型的转化时,本文指出子信道数是可变的,并且当子信道数等于过采样的倍数时,辨识效果最好。实验证明本文算法估计效果好,计算量小。特别是当用于估计的点数较少时仍能保持良好的性能,因而可适用于快变信道。     

Clock-Feedthrough Compensated First-Generation SI Circuits and Systems

In this paper we discuss the clock-feedthrough problemin switched-current circuits. We present a clock-feedthroughcompensated first-generation SI memory cell that ideally cancelsboth constant and signal-dependent clock-feedthrough. It is shownhow to optimize the memory cell performance according to a generalcost function. Measured total harmonic distortion of the memorycell is less than -65 dB when optimized for low-power. The implementationof a second-order delta-sigma modulator using the presented memorycell is also described. Measurements confirmed a dynamic rangeof 11 bits. All circuits were implemented in a single-poly CMOSprocess.     

Nonlinear Quantization in Low Oversampling Ratio Sigma-Delta Noise Shapers for RF Applications

Baseband signal processing for current base stations or 3rd generation mobile systems will impose high bandwidth and high VLSI integration demand. Many of the desired integration aspects can be satisfied with sigma-delta converter front-ends. However, under the technology constraints there are simultaneous requirements for high sample rate and low oversampling ratio in order to achieve the desired baseband width. In this paper, we present system architecture results for the 4th-order cascaded noise shaper architectures to be used in baseband front-ends. We show that the cascaded structures with proper scaling will satisfy simultaneous demand on linearity (spurious free dynamic range) and high SQNR with low oversampling ratio based on usage of multibit quantizers outside the actual signal noise shaping path. We also present results for nonlinear quantization effects in low oversampling ratio cascaded noise shaper architectures. We analyse the effect of the non-linearity in both the A/D and D/A-block in quantization error quantizer path for the 4th-order cascaded topology and the design constraints associated to the performance of the used A/D and D/A structures. The performance requirement for the multi-bit quantizer for high SQNR is shown for the case of low oversampling ratios. The results show that non-uniform quantization around zero input are far more crucial to the SQNR than nonlinear quantization deviating from the ideal transfer function. As the key difference to standard multibit quantizers, no special error correction or error distribution schemes are required; the linearity requirements are satisfied with 0.2 LSB accuracy of the few bit quantizer. Finally, the performance of non-linear quantization using multitone test signals are also shown.     

基于门控循环单元的非均衡数据驱动异常用电检测方法

异常用电检测能够及时发现异常用电行为,在减少能源浪费和经济损失的同时能够维持安全、稳定的电网运行环境。智能电表的普及使得用电数据获取十分容易,为数据驱动的异常用电检测方法提供了充足的数据支持。然而,在实际应用过程中,异常数据较少导致的数据非均衡问题严重影响了模型的训练效果。因此,针对上述问题提出了一种针对非均衡数据的门控循环单元异常用电检测方法。该方法利用边界合成少数类过采样技术实现了对少数类数据的有效扩充。为了更好的捕捉用电数据的时序特征,采用了门控循环单元实现对用电数据的分类。为了验证该方法的有效性,基于非均衡数据集进行了对比实验。实验结果表明,该方法能够更好的数据扩充效果以及更准确的异常用电检测效果。