An old article by Jim Williams...
I was looking over some old Analog Dialogues that are available on the web and I ran across this very old article by Jim Williams [1] (one of his first),
It is from 1976 and it covers the classic "Weighing Scale" application [2]. It is a reprint of an article that first appeared in EDN Magazine on October 5th, 1976. The title of the original article was,
"This 30ppm scale proves that analog designs aren't dead yet" [3].
It is interesting that way back in 1976 the main thing on some Analog Designers minds was: "Analog isn't dead yet!", still sounds familiar today. A recent issue of a popular magazine recently had a cover that boldly stated that “Analog is not dead”, yet sadly the issue contained really no Analog articles at all. So Analog may still not be dead, it just isn't written about much anymore.
Back to Jim's Article
I called this a classic weighing scale design because numerous clever circuits have been presented over the years to make a high resolution scale settle out to very high precision very fast. A difficult problem that usually involves the use of a Nonlinear Filter. This article (Last Figure) shows that function as a Black Box and provides very little detail about it other than to describe its function as,
“...Carefully designed nonlinear filter permits large changes of weight to be stably registered within 5 seconds, small changes within one second and fast disturbances … To be rejected”
This is the age old problem in electronics, that is,
“How do I make a very high resolution measurement in the presence of noise quickly?”
Numerous articles over the years have presented solutions to this problem one way or another. The simplest way is to put back to back diodes across the resistor in a RC Low Pass filter as shown in Figure 1. While this works, it is pretty limited in it's adjustability. You have the choice of a 0.6 volt or 0.3 volt forward voltage drop for the diodes depending on weather you use Silicon or Schottky barrier diodes. Over the years numerous other circuits have been presented to increase the utility of this simple circuit even more. Perhaps the ultimate form of the purely Analog non-linear filter was presented by Burr-Brown Engineers: Messrs. Stitt and Burt in 1991 (See Figure 2 and Reference 4).
Figure 1 – The simplest form of a nonlinear filter is to put bypass diodes around a RC filter. It is simple, but it isn't very adaptable or stable over temperature.
Hewlett-Packard also faced a similar problem with their Microwave Power Meters starting as I recall with the Digital Model 436 Series and continuing to this day. The first digital power meter, The Model 436 [5] used a hybrid approach, incorporating a analog filter, and also including a digital box car or moving average filter. The combination of the two allow for fast settling to a low noise value. Later models of this power meter series also incorporate a non-linear digital filter by resetting the moving average if the input exceeds around 12% of the average value. Thus improving on purely analog filtering even more.
Figure 2 – Reference 4 shows how to make a much improved nonlinear filter. By the addition of an OPAMP the diode threshold can be made adjustable over a very large range. Stability over temperature issues remain however.
Analog Design is now Hybridized
This hybrid trend continues unabated. While there may be relatively few purely analog designs anymore, there are many examples of hybrid designs that take the best of both the Analog and Digital worlds.
If your latest design digitizes the signal of interest you can apply an infinite variety of digital filtering both linear and non-linear. You can even change the filtering characteristics on the fly when the system is running.
One of the hybrid approaches I have used in noisy industrial settings is to use an purely analog filter set to the maximum bandwidth needed. This acts like a roofing filter and if properly designed will even tame Radio Frequency interference and add ESD suppression. The signal is then digitized by a microprocessor and the samples are put in a buffer. To non-linearly remove impulse noise a median filter is applied. This discards the largest and lowest recorded value, then the rest of the buffer is averaged yielding an average with large peaks removed from the measurement. Even here a fair amount of Analog knowledge is needed to make the filter work as a low pass filter for the Analog and reject RF and ESD as required by the system usage. Careful experimentation and testing to optimize the digital filter is also required for best performance. The degree of filter memory is also adjustable from remembering all samples to remembering no samples between averaging cycles. The RC filter always has some memory and this leads to slow settling times. Best of all however, if the Digitizer is stable over temperature, all the digital filtering will be drift free also.
So I wouldn’t say Analog Design is Dead, or even sick! Analog Design has just been hybridized by Digital Processing to make our systems even more bulletproof and lower cost to boot, we should take advantage of this whenever possible.
Original Article Snippet from Analog Dialog [2]
References
[1] Analog Dialogue was distributed in printed form starting in 1967, today it continues as an online electronic publication.
http://www.analog.com/library/analogdialogue/archives.html
[2] Analog Dialogue, Vol 10, No 2, 1976
http://www.analog.com/library/analogDialogue/cd/vol10n2.pdf
The screen captures here are reproduced from this source and are copyright and owned by Analog Devices, Inc.
[3] Thanks to Kent Lundber of MIT for cataloging all of Jim's articles, where I found that this was indeed one of Jim's first articles.
http://web.mit.edu/klund/www/jw/jwbib.pdf
[4] Stitt, Mark and Burt, Rod, “Fast Settling Low-Pass Filter”, Burr-Brown Corporation, AB-022, January1991
http://www.ti.com/lit/an/sboa011/sboa011.pdf
[5] HP Journal, October, 1975.
http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1975-10.pdf
By: Steve Hageman
www.AnalogHome.com
No comments:
Post a Comment