The achievable accuracy for systems with multiple ADCs depends directly on the reference voltages applied to the ADCs. Medical-ultrasound-imaging systems, for example, commonly include a large number of ADCs in the system's beam-former electronics, with the ADCs usually organized in groups of 16, 24, 32, and so on. To obtain maximum beam accuracy, you must minimize errors in the ADC path. Poor accuracy of the reference voltages of the individual ADCs degrades the overall system accuracy. Another source of degradation is the distributed load, which comprises many individual resistive and capacitive loads. Several approaches are available to provide the reference voltage for such ADC arrays:

  • Individual on-chip references. Though this approach offers a convenient local connection to each ADC, it can result in relatively poor matching among the converters.

  • A single external reference voltage applied to all ADC reference inputs. Such a configuration allows you to engineer an external reference voltage of arbitrary accuracy but incurs errors from the small variations among the internal ladders of the ADCs.

  • An external reference directly driving the ADCs'reference-ladder taps. This option delivers maximum gain accuracy by directly controlling the reference voltage applied to each ADC ladder. However, it requires driving the relatively low resistance of the ladders. Moreover, some ADCs do not allow access to that internal bias point.

ADC accuracy

In many applications, gain and noise level have a major effect on ADC accuracy. The gain of an ADC is in effect the slope of its transfer function, which relates analog inputs to the allowable range of digital-output codes. One way to quantify gain is to measure the full-scale input range, which is a direct function of the reference-voltage level. For medical- ultrasound-imaging systems, variations in the full-scale ranges of the ADCs can cause errors in beam formation. The variations also affect the ADCs' clipping point—an effect that may be important is certain signal-demodulation schemes. An ADC's noise level determines its usable dynamic range. This dynamic range should be as great as possible. The reference-noise component of ADC noise can be additive or multiplicative. Local bypass capacitors on the individual ADCs can easily filter additive noise. Multiplicative noise, on the other hand, is more insidious. For ultrasound applications, reference noise in the audio-frequency spectrum can modulate large "stationary" signals in the RF spectrum. Such signals arise from stationary tissue in the ultrasound target.

Audio modulation produces sidebands in the RF signal that a Doppler detector can demodulate, producing audio tones. To estimate the amount of audio noise tolerable in an ultrasound application, assume a nearly full-scale RF signal applied to a 10-bit ADC such as the MAX1448. The device's dynamic range of almost 60 dB equates to a noise floor of –60 dBFS (relative to full-scale). You can normalize that noise level to a 1-Hz bandwidth. The Nyquist bandwidth for an 80-MHz sampling rate is 40 MHz. The correction factor is =76 dB, which places the ADC's noise floor at –60 dBFS–76 dBFS=–136 dBFS. Because a conservative design requires the reference-voltage noise to be at least 20 dB lower (–156 dBFS), a 2V reference requires an extremely low noise level of 33 nV p-p (approximately 8 nV/).

A multiple-ADC array may require a more accurate reference voltage than the one internal to each converter. The reference voltage internal to MAX144x converters, for example, has an accuracy of ±1%. The following two circuits are reference designs for such arrays. They feature a single, common low-frequency noise filter, and they offer high-frequency noise suppression via local decoupling capacitors connected to individual ADCs.

Single external reference

Multiple-converter systems based on the MAX144x family are well-suited for use with a common reference voltage. You can the REFIN pin of these converters to an external reference source and thus eliminate the need for any circuit modification. Moreover, the high input impedance of REFIN (even of multiple REFIN terminals connected in parallel) results in only a small load-current drain. Figure 1 shows a precision source, such as the MAX6062, that generates an external dc level of 2.048V and exhibits a noise-voltage density of 150 nV/ . The output of the IC passes through a one-pole lowpass filter with 10-Hz cutoff frequency to op amp IC2, which buffers the reference. The buffered reference voltage then passes through a second 10-Hz lowpass filter. IC2exhibits a low offset voltage for high gain accuracy and a low noise level. The passive 10-Hz filter following the buffer attenuates noise produced in the voltage-reference IC and buffer stage. The filtered noise density, which decreases with frequency, meets the noise levels required for precision-ADC operation.

Converters of the MAX144x family specify a typical gain error of ±4.4% (better than ±0.5 dB). This performance is better than the gain tolerance of all other building blocks in the signal path of an ultrasound receiver. Note that the circuit in Figure 1ensures proper power-up/power-down sequencing, because all active parts receive their power from the same supply-voltage rail. This approach yields excellent gain matching and an extremely low noise level with minimal circuitry. The circuit should prove adequate in many applications that require multiple gain-matched ADCs.

Precision external reference

For applications requiring more stringent gain matching, the MAX144x family fills the bill. In Figure 2, connecting each REFIN to analog ground disables the internal reference of each device. You can thus directly drive the internal reference ladders from a set of external reference sources. These voltages can have an arbitrarily tight tolerance; the ADCs typically track them within 0.1%. ADCs of this family have 4-kΩ resistance across the ladder's reference connection, so it's easy for the reference source to drive the load, even with many ADCs connected in parallel. IC1 generates a dc level of 2.500V, followed by a 10-Hz lowpass filter and a precision voltage divider. The buffered outputs of this divider provide 2, 1.5, and 1V, with an accuracy that depends on the tolerances of the divider resistors. The quad op amp IC2, selected for its low noise and dc offset, buffers the three voltages.

The individual voltage followers connect to 10-Hz lowpass filters, which filter both the reference-voltage and buffer-amplifier noise to a level of 3 nV/. The 2 and 1V reference voltages set the differential full-scale range of the associated ADCs at 2V p-p. The 2 and 1V buffers drive the ADCs' internal ladder resistances between them. The load is 4 kΩ divided by the number of ADCs in the circuit. As an example, 32 ADCs draw 8 mA from the supplies, a load current that is well within the capability of IC2. The gain accuracy of the configuration in Figure 2 can be almost arbitrarily tight, depending on the accuracy grade of IC1 and the tolerances of the resistors in the voltage divider. The gain matching of the ADCs in such a configuration is typically 0.1%. With a noise level below 3 nV/ at 100 Hz, this circuit provides exemplary performance. As in Figure 1, the common power supply for all active components removes any concern about power-supply sequencing.

Circuit provides reference for multiple ADCs的更多相关文章

  1. Kong(V1.0.2) Health Checks and Circuit Breakers Reference

    介绍 您可以让Kong代理的API使用ring-balancer,通过添加包含一个或多个目标实体的 upstream 实体进行配置,每个 target指向不同的IP地址(或主机名)和端口.ring-b ...

  2. Satisfying memory ordering requirements between partial reads and non-snoop accesses

    A method and apparatus for preserving memory ordering in a cache coherent link based interconnect in ...

  3. IOS CALayer的阴影属性

    @property(nullable) CGColorRef shadowColor; /* The opacity of the shadow. Defaults to 0. Specifying ...

  4. Articles Every Programmer Must Read

    http://javarevisited.blogspot.sg/2014/05/10-articles-every-programmer-must-read.html Being a Java pr ...

  5. iOS的阴影绘制及性能优化

    今天来讲讲iOS开发过程中的阴影绘制及其潜在的绘图性能问题.虽然在开发过程中,我们使用阴影功能的机会不是很多,但是如果用了,有可能引起如卡顿等性能问题,所以,还是有必要来探究一下阴影的绘制过程,及如何 ...

  6. linux驱动编写(电源管理驱动)

    对于嵌入式设备来说,合适的电源管理,不仅可以延长电池的寿命,而且可以省电,延长设备运行时间,在提高用户体验方面有很大的好处.所以,各个soc厂家在这方面花了很多的功夫.下面,我们可以看看linux是如 ...

  7. handle exceptions, opening and closing database connections

    https://www.tutorialspoint.com/spring/spring_jdbc_framework.htm Spring - JDBC Framework Overview Whi ...

  8. Kalibr installation tutorial

    How to install Kalibr I was confused about installing Kalibr, but there is no even one hint in READM ...

  9. PatentTips - Reducing Write Amplification in a Flash Memory

    BACKGROUND OF THE INVENTION Conventional NAND Flash memories move data in the background to write ov ...

随机推荐

  1. [node.js] async/await如何优雅处理异常?

    node.js的世界,从callback开始,不会止于async. 所有人都在骂为什么不能完全进化,其实我感觉这就是老外的细心,为了承上.这也就是为什么async其实就是promise一样,假如不是一 ...

  2. BZOJ 2049: [Sdoi2008]Cave 洞穴勘测——LCT

    传送门:http://www.lydsy.com/JudgeOnline/problem.php?id=2049 省选之前来切一道数据结构模板题. 题意 这是一道模板题. N个点,M次操作,每次加边/ ...

  3. AD服务器安装文档

    Windows Server 2008 R2 AD服务器搭建 1. AD服务器简介 应用到: Windows Server 2003, Windows Server 2003 R2, Windows ...

  4. java你应该学会什么

    给初学者之一:浅谈java及应用学java 先说什么是Javajava是一种面向对象语言,真正的面向对象,任何函数和变量都以类(class)封装起来至于什么是对象什么是类,我就不废话了关于这两个概念的 ...

  5. java基础10 单例模式之饿汉式和懒汉式单例

    前言: 软件行业中有23中设计模式 单例模式    模版模式    装饰者模式    观察者模式    工厂模式    ........... 单例模式 1. 单例模式包括 1.1 饿汉式单例 1.2 ...

  6. SGU 217. Two Cylinders

    题意:给空间内两根圆柱,求轴线垂直相交时公共部分的体积. 暴力积分即可. ID: Date'n'Time: Name: Task: .Ext: Status: Time: Memory: 158937 ...

  7. linux的文件隐藏属性 chattr lsattr

    <鸟哥的Linux私房菜 基础学习篇(第三版)> 7.4.2 读书笔记 显然,这里要说的并不是rwx那9个权限,而是要聊chattr和lsattr这两个命令.这两个命令在只能在Ext2/E ...

  8. HTML 如何显示英文单、双引号

    // 过滤英文引号替换成中文引号 function pregstring($str){ return preg_replace('/"([^"]*)/','&quot${1 ...

  9. python 2 如何安装 MySQL 数据库操作库

    我试了好几种网上的办法,在 windows 10 VS2017 环境下不是缺了头文件,就是缺 .lib,反正十分繁琐,以后我也懒得搞了,都用 annaconda 来搞定就好了,时间宝贵. 在 控制台中 ...

  10. ZOJ 3954 Seven-Segment Display

    二分图匹配. 先检查每个数字$1$的个数是否满足条件,不满足直接就是无解.剩下的情况可以建立二分图,如果现在的某一列可以对应于原图的某一列,那么建边.如果二分图的最大匹配是$7$,则有解,否则误解. ...