38度发烧友--38Hot Volt-Nuts

 找回密码
 立即注册

QQ登录

只需一步,快速开始

查看: 167|回复: 0

小心ADC的ENOB

[复制链接]
发表于 2013-8-2 23:58:24 | 显示全部楼层 |阅读模式
本帖最后由 shichen717 于 2013-8-2 23:59 编辑

厂商通常给出高位低速ADC的有效分辨率(effective resolution)和ENOB(effective number of bit),并将其并列列出,例如20(18.2),用以表达ADC具有至少18.2位无闪烁(No flicker)分辨率,相当于5 1/2位分辨率。

这是一个严重的误导。

测量直流信号时,真正需要的是无噪声分辨率(Noise free resolution)=有效分辨率-2.7bit。因此手册应给出的指标应该为20(17.3)。17.3bit才是此ADC的有用分辨率。比ENOB低0.9bit,差不多1bit。

ENOB实际是ADC的动态指标,与直流测量关系不大,使用ENOB估算直流性能会造成很大的困扰,实际测量的分辨率永远达不到ENOB。原因在于ENOB使用的峰峰值系数为12^0.5=3.46,无噪声分辨率使用的峰峰值系数为6.6,比ENOB大1.9倍。

例如AD7705,双极性PGA=32,SR=50Hz时,AD手册给出的数据ENOB=16bit,enrms=0.66uV,enp-p=6.6enrms=4.36uV,但AD7705此时的1LSB=2.38uV,无噪声分辨率仅为16-lg2(4.36/2.38)=15.1bit,实际测量也确实如此,几秒内2LSB,长时间3LSB的峰峰值噪声(无内部缓冲器)。AD宣称其缓冲器不影响噪声,但在使用内部缓冲器时的长时间测量结果居然达到5LSB峰峰值噪声。

因此需要谨慎对待ENOB,至少再减去0.9bit才是有用的无噪声分辨率。实际上6.6只保证99.9%的峰值,对于50Hz采样率而言,最多只能坚持20秒,对于长期测量,7可能是更好的系数,此时AD7705的分辨率只有15bit。

回过头看看20(18.2)bit的实际分辨率其实只有17.3bit,只有4 7/8位,做5 1/2位电压表末位会始终跳动。

(不要提后续的数字滤波处理,不到万不得已最好不用,数字滤波会严重降低响应速率,那样的滞后会让使用者非常痛苦,例如34401A。尤其对于不能使用数字滤波的实时触发测量,ADC的无噪声分辨率才是最重要的,在工业应用中,触发测量才是数字万用表最重要的功能。)


发表于 2013-8-3 00:06:43 | 显示全部楼层
顶一个
回复

使用道具 举报

发表于 2013-8-3 09:19:47 | 显示全部楼层
好好体会一下.
回复

使用道具 举报

发表于 2013-8-3 10:39:21 | 显示全部楼层
我也顶一个,LZ有什么好的高速高位ADC介绍?
回复

使用道具 举报

发表于 2013-8-3 12:48:48 | 显示全部楼层
嗯,不错的文章 ,我也一直把ENOB当直流有效位。
回复

使用道具 举报

发表于 2013-8-3 13:56:29 | 显示全部楼层
本帖最后由 analex 于 2013-8-3 14:04 编辑

AD7705,双极性PGA=32,SR=50Hz时,AD手册给出的数据 ENOB=16bit, 无噪声分辨率=ENOB - 2.7bit = 13.3bit. 楼主看到2~3LSB跳动属正常。
DS中Table 8无噪声分辨率仅13bit,而对应的噪声仅比Table 5中大了1.51倍, 很不正常。由此推断Table 6大概是写错了。
做直流设计,还是看噪声电压有效值比较方便。 ENOB, EB,Noise Free...等概念表达的是动态范围的概念,不直接反映噪声的绝对值,在直流设计中容易把人绕进去了。
回复

使用道具 举报

发表于 2013-8-3 16:22:58 | 显示全部楼层
刚好在搞这个器件,也刚好看到这方面的资料
When I look at the output of the ADC, I don’t get the full resolution; a few of the LSBs are flickering. Why is this happening?

While the ADC has a resolution of 16 bits, the accuracy of the part varies with update rate and gain. The datasheet lists the rms noise and achievable accuracy for different update rates and gains. For example, the AD7705 has a peak to peak resolution of 16 bits when the gain is 1 (analog input range of +VREF), an update rate of 50Hz and operated in bipolar mode. If the update rate is changed to 250 Hz, the peak-to-peak resolution is reduced to 13 bits. If the gain is now changed to 128 (+20 mV input range), the peak-to-peak resolution is reduced to 12 bits.
The accuracy of an ADC is specified as effective resolution or peak-to-peak resolution. The effective resolution is calculated using the rms noise that is given in the datasheet. The effective resolution equals log (input span / rms noise) / Log 2. The peak-to-peak resolution is the number of bits which do not flicker and is calculated using the peak-to-peak noise which equals 6.6 * rms noise. Therefore, the peak-to-peak noise equals log (input span / (6.6 * rms noise)) / Log 2.
The datasheet rms noise values are measured with the chosen analog input channel shorted to some voltage such as VREF(both terminals of the analog input channel are connected to VREF). Therefore, the user should short the analog input on their system board using a similar method to ensure that the best performance is being obtained from the part. After performing a calibration, commence conversions. Using several thousand samples, the rms noise can be calculated and, from these,.the peak-to-peak resolution can be determined. The obtained value should be compared with the values given in the datasheet. If the accuracy specified in the datasheet is not obtained, this is due to noise on the circuit board, for example, ensure that ground loops do not exist and ensure that the power supply is adequately decoupled using a 10uF tantalum capacitor in parallel with a 0.1uF ceramic capacitor from each power supply to its respective ground. The capacitors should be placed as close as possible to the ADC’s pins.
The datasheet rms noise values are measured with the chosen analog input channel shorted to some voltage such as VREF(both terminals of the analog input channel are connected to VREF). Therefore, the user should short the analog input on their system board using a similar method to ensure that the best performance is being obtained from the part. After performing a calibration, commence conversions. Using several thousand samples, the rms noise can be calculated and, from these,.the peak-to-peak resolution can be determined. The obtained value should be compared with the values given in the datasheet. If the accuracy specified in the datasheet is not obtained, this is due to noise on the circuit board, for example, ensure that ground loops do not exist and ensure that the power supply is adequately decoupled using a 10uF tantalum capacitor in parallel with a 0.1uF ceramic capacitor from each power supply to its respective ground. The capacitors should be placed as close as possible to the ADC’s pins.
回复

使用道具 举报

发表于 2013-8-3 16:28:56 | 显示全部楼层
上面摘自 How do I get the best performance from the AD7705/06/07 ADC?

昨天晚上看c8051f350的资料,看里面说晶振为2.4576M的晶振可以获得最佳性能
7705最是推荐2.4576M的晶振,一开始不知道为啥,后来想到是不是60hz的整2倍,一算还刚好就是

国内的50hz工频,就没有50hz整2倍的晶振,真扯淡、、、、、、、、、、、、、
回复

使用道具 举报

 楼主| 发表于 2014-5-10 18:02:57 | 显示全部楼层

何事?
回复

使用道具 举报

 楼主| 发表于 2014-5-10 18:20:46 | 显示全部楼层
blackspider_e 发表于 2013-8-3 16:28
上面摘自 How do I get the best performance from the AD7705/06/07 ADC?

昨天晚上看c8051f350的资料 ...

这仿佛是个思路,可以用2.4576/6*5=2.048MHz,但用7705的60Hz采样模式获得50Hz notch,不过过采样率会降低,不知会不会影响有效分辨率。

7705给出的notch只有60Hz一张图,不知50Hz notch如何实现。

回复

使用道具 举报

发表于 2014-8-26 12:41:10 | 显示全部楼层
用卡尔曼滤波处理怎么样?这种滤波器性能不是一般的好。
回复

使用道具 举报

 楼主| 发表于 2014-8-26 22:25:46 | 显示全部楼层
xjf20072608 发表于 2014-8-26 12:41
用卡尔曼滤波处理怎么样?这种滤波器性能不是一般的好。

何解?
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

快速回复 返回顶部 返回列表