4
\$\begingroup\$

While studying analog circuits and conducting experiments with op-amps, I couldn't help wondering why would I need an analog circuit to add/multiply/integrate electrical signals. In modern times, computers are quite cheap, and it seems a lot easier letting a computer calculate anything you want (it's fast, versatile, reliable...)

So, why use op amps and analog circuits?

\$\endgroup\$
3
  • 1
    \$\begingroup\$ Edited out the FET/BJT part of the question, that is worthy of a question of its own, and isn't really part of the original question. \$\endgroup\$
    – Matt Young
    Commented Apr 9, 2013 at 19:44
  • \$\begingroup\$ musical applications, especially guitar amps \$\endgroup\$
    – wim
    Commented Apr 10, 2013 at 2:35
  • \$\begingroup\$ Digital is made of analog stuff. Building NAND gate requires transistors/diode etc. \$\endgroup\$
    – niki_t1
    Commented Mar 7, 2017 at 7:15

5 Answers 5

10
\$\begingroup\$

The moral of the story is digital electronics need an interface to the outside world. Analog electronics are necessary to get the outside world's signals into a form that can be digitized. For example, how do you get that \$\frac{50\mu V}{degree }\$ from a thermocouple into a signal large enough to put into an ADC? Use an instrumentation amplifier.

Here's an example of something I built a while back: sine shaper

It implements the equation \$V_{out}=1.552V_{in}-0.000560V_{in}^3\$ That is an approximation for sine shaping, the and circuit itself shaped the triangle wave into a sine wave with less than .05% THD. It could have been done in the digital domain, but:

  1. the input signal was \$20V_{pp}\$, way to big for immediate conversion without some kind of zero and span, and

  2. the signal was already analog and it would have made no sense to convert to digital just to convert back to analog.

On the subject of taking signals digital, an analog anti-aliasing filter is essential before any ADC. This filter is just a low pass filter to ensure \$f_{sig}<f_{Nyquist}\$. That has to be done in the analog domain. Solid analog circuitry is essential to the operation of any embedded system.

\$\endgroup\$
1
  • 1
    \$\begingroup\$ Even with DAC overhead, analog arithmetic can be less expensive; e.g., "Low-Power, High-Performance Analog Neural Branch Prediction" (St. Amant et al., 2008) proposed using digital storage and analog accumulation for a perceptron-based branch predictor. \$\endgroup\$
    – user15426
    Commented Apr 9, 2013 at 22:36
13
\$\begingroup\$

There are lots of examples that people will use that point out the "exception" to the rule where you can't use digital design approaches for signals, like:

  • dealing with RF or very high frequency signals, most digital motherboards in computers have a very "analog" design cycle.

  • conversion from the analog domain to digital.

  • Real world effects like stray capacitance, inductance and protection against ESD etc.

  • and many others and they are right !

They are cases where the rules used for digital abstraction breakdown. In reality, there isn't such a thing as a digital circuit, it's just that it gets "packaged" to simplify the next level up of design. And in higher performance designs reality arises and this packaging breaks down.

But we need not even look at these exceptions to understand that "digital" is a handy abstraction. A very useful abstraction.

I'll take as an example the design of a simple digital circuit on the chip level. One that is not particularly fast or even particularly challenging. The designer goes in there and describes the design in say Verilog, sends the design off and gets the results back or loads it into a FPGA. So we are not dealing with high speed mother boards, or RF and WiFi etc. that might "look" analog like.

What this digital designer is NOT seeing is the care and attention taken to develop and publish the parameters under which certain sub cells he uses in his design. The cell designer, simulates the design of say a Dff under the PVT corners (Process, Voltage and Temperature) determines what level of error is necessary (3 sigma, 4 sigma etc.) and then comes up with the operating parameters under which the device operation can be considered to be "digital". Then once fabricated they are tested against these simulations for verifications and corrections are made. For a Dff this would be setup and hold times. As long as those timings are met under those conditions you can happily live with the assumptions that "hell analog isn't needed anymore". But the next abstraction comes into play, synchronous design. Now if we say that certain design regimes are followed, we can then design the individual cells in such a way you can cobble them together and not even violate those timing requirements above. Unless you're doing something particularly clever or stupid.

Now once you have your slowish, "digital" circuit running, yes you can operate under the assumptions that it's digital and you won't get bitten. But the reality is that everything is analog, the details are just hidden from you. So the next time you use a "digital" uProcessor like a PIC or Arduino know that someone, somewhere has actually made your life easier by taking care of the analog aspect of nature so you can be deluded into thinking your design is digital.

\$\endgroup\$
4
  • \$\begingroup\$ But it isn't really, because once you have the processing defined as an algorithm in the digital domain, the implementation is no longer that relevant. The digital domain abstracts not only that particular circuit, but any other workalike. \$\endgroup\$
    – Kaz
    Commented Apr 9, 2013 at 23:06
  • \$\begingroup\$ @Kaz Hypothetically, I have designed the world's most sophisticated synthesis engine, and furthermore I have adapted it to generate layouts and manufacturing instructions for all technologies in the history of EE. You send me verilog RTL code for a simple 4 bit shift register, I have all the library files here for BOTH a 28 nm CMOS metal gate process and a 1930's vintage Tube (Valve) based logic factory. By your argument the implementation is not important and therefore BOTH designs will run at the same speed regardless of which process they are targeted for CMOS or Tube. \$\endgroup\$ Commented Apr 9, 2013 at 23:47
  • \$\begingroup\$ +1, nice explanation and great reference read for future researchers. \$\endgroup\$ Commented Apr 10, 2013 at 5:54
  • \$\begingroup\$ another key area is electronics in space, or other high radiation environments, where random bit flips are all too common \$\endgroup\$
    – jk.
    Commented Apr 10, 2013 at 7:32
4
\$\begingroup\$

The real world is analog. So when you receive RF or read from a sensor, the signal starts analog. You may convert to digital immediately with an A/D. Also inital filtering needs to be analog to avoid sampling aliasing.

\$\endgroup\$
1
  • \$\begingroup\$ Thanks for the answer! your answer implies that the main use of analog circuits is to convert analog signals to digital, and calculating circuits is actually not used in modern systems. \$\endgroup\$ Commented Apr 9, 2013 at 20:50
1
\$\begingroup\$

There are also some things that are actually very hard to do in the digital domain, hard limiting in a sampled system for example requires heroic oversampling if aliasing is not to be a problem.... Doing the same thing in the analogue domain can be as simple as a resistor and two anti parallel diodes. Consider further that signal conditioning can be a monster problem irrespective of how the subsequent processing is handled, You could digitize directly, but often adding something like a log amp (Analogue) makes subsequent processing easier.

When things are slow enough (or bandwidth is low enough) going digital early and doing the work there often makes sense if the work is non trivial, but by the time you get to GHz edge rates driving long cables, you had better be thinking analog (Including thinking of the power and ground planes as being LCR networks), or it just will not work, and a scary amount of stuff fits into this space these days.

Digital is a lovely abstraction when you can get away with it however, but just like the simple model of the opamp it has very definite limits, and you have to know where the edges of the model are, same with ebers-moll for transistor models, usually it is a good enough simple abstraction, sometimes you need to get your actual physics on.

Regards, Dan.

\$\endgroup\$
0
\$\begingroup\$

It is more simple to do analog integrating or some other signal processing operation in analog fashion rather then using A/D converters. By using A/D converters you must solve digital synchronization problems, signal conditioning, etc. Only trouble that op-amp makes comes from voltage offset, but there are ways to bypass offset problems. However, most of present op-amps are in quad case, so you have four op-amps in only one chip. It means that you can easily do signal buffering etc.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.