4
\$\begingroup\$

An ADC will need an anti-aliasing filter to remove signal spectrum above the Nyquist frequency and a buffer so that the signal source is not loaded by the ADC input.

For a DAC, what does one need to be able to use it in a circuit? I assume that a low filter could be required since the output of DAC will have high frequency noise component due to fact that the digital input changes in discrete points in time.

Note: This question is from perspective of a design engineer and not related to theory of how DAC works as described in university textbooks.

\$\endgroup\$
6
  • 1
    \$\begingroup\$ There are also applications that don't care about the output signal bandwidth, for example if you build a power supply with adjustable overcurrent protection, you'd use a DAC and feed its output into a comparator to find out if you are above the limit. \$\endgroup\$ Aug 22, 2023 at 4:59
  • 10
    \$\begingroup\$ "Not related to … textbooks": What exactly do you think the textbooks are trying to teach you? Any textbook on digital-to-analog conversion will tell you you need a reconstruction filter, why, and what the design requirements are. "From a design engineer perspective" cannot be an excuse for saying "I don't intend to understand the basics"! \$\endgroup\$ Aug 22, 2023 at 7:57
  • 1
    \$\begingroup\$ Textbooks talk about theory since it is about building foundation. Most school textbooks only teach the most basic stuff. \$\endgroup\$
    – gyuunyuu
    Aug 22, 2023 at 11:09
  • 2
    \$\begingroup\$ The DAC's complement to an ADC's anti-aliasing filter is an "analog reconstruction filter" \$\endgroup\$
    – Ben Voigt
    Aug 22, 2023 at 16:20
  • 1
    \$\begingroup\$ "need an anti-aliasing filter to remove signal spectrum above the nyquist frequency" -- these are strong assumptions. There are applications where they aren't needed, or at least not in the simple way I think this statement was meant. Conversely, DACs needn't use any, and there are real applications for aliasing e.g. hackaday.com/tag/osmo-fl2k . It is specific to the conditions of the Nyquist sampling theorem where filtering on both ends is needed. (Which is many applications, but not all.) \$\endgroup\$ Aug 22, 2023 at 23:13

4 Answers 4

9
\$\begingroup\$

Well, it depends what compromise you are ready to take. For example, an ADC will often use a follower to deal with it's impedance. and a low pass filter to prevent nyquist harmonics.

A DAC could also benefit of a follower to improve it's output impedance. Other then that, a low pass filter and you are good to go! The frequency of your filter should be at most half of your sampling rate or the highest frequency component you will be generating. This will improve noise performance.

\$\endgroup\$
4
  • 6
    \$\begingroup\$ Good answer, except that the cutoff frequency of the filter should match the application signal bandwidth. This is half of the sampling frequency only if you're trying to use the minimum possible sample rate for the application -- in which case, it needs to be a fairly high-order filter. Using oversampling to relax the filter requirements is actually fairly common in embedded applications. \$\endgroup\$
    – Dave Tweed
    Aug 22, 2023 at 1:04
  • 1
    \$\begingroup\$ Excellent point! I added it to the answer! \$\endgroup\$
    – Julien
    Aug 22, 2023 at 1:07
  • \$\begingroup\$ @Dave, when have you had to use DACs? \$\endgroup\$
    – gyuunyuu
    Aug 22, 2023 at 12:55
  • 1
    \$\begingroup\$ @gyuunyuu: I guess mostly in audio and video signal processing, but also in control applications, such as GPS-disciplined oscillators and similar areas. These days, I'm mostly working with high-speed ADCs, capturing the raw signals from image sensors for subsequent processing in the digital domain. \$\endgroup\$
    – Dave Tweed
    Aug 22, 2023 at 13:49
2
\$\begingroup\$

ADCs need an anti-aliasing filter if aliasing would otherwise be a problem. Similarly DACs need a reconstruction filter if step changes in the output would otherwise be a problem.

For DACs Perhaps the steps are small enough that they are below the noise floor. Perhaps a step change that happens as fast as possible is what I actually wanted in the first place. For ADCs perhaps my input signal simply doesn't contain enough high-frequency content to be worth filtering it out.

And filters, particularly those with sharp cutoffs come with problems of their own. They can cause phase distortions. They can turn a rapid step change into a long period of ringing.

The need for analogue anti-aliasing and reconstruction filters can often be reduced or even eliminated by increasing the sample-rate of the ADC/DAC. Depending on what the rest of your system is doing you may either use digital filters to bring down the sample rate, or just run your whole system at the higher rate.

Filters on the output of DACs can also be used to compensate for low bit depth of the DAC. You can deliberately introduce high-frequency noise into the digital bitsteam and then filter it out with an analog filter to increase the effective bit depth of your DAC. In an extreme case, one can use a DAC with only a single bit.

Ultimately design is about achieving a goal. Sometimes you need analogue filters to achieve that goal. Sometimes you don't.

\$\endgroup\$
2
  • \$\begingroup\$ Any signal from real world source (sensor) will have a lot of frequency components inside it. Or so I believe. So, quite possibly we will always benefit from having anti-aliasing filter. Better safe than sorry. \$\endgroup\$
    – quantum231
    Aug 22, 2023 at 22:51
  • \$\begingroup\$ But how strong are those frequency components? are they strong enough to pose a real issue? is software filtering enough to reduce them to an acceptable level? These are all questions the designer needs to ask themselves. \$\endgroup\$ Aug 23, 2023 at 3:42
2
\$\begingroup\$

DACs may also have an scaling in the form of (DAC output * Gain) + Offset.

For example, for a DAC with a 0 -- 5V output, if you want the output range to be -10 -- 10V, you must apply a gain 4V/V of and offset of -10V. A simple realization of this gain plus offset is from a op-amp.

Can see some examples here: Offset and scale a DAC output with opamps

\$\endgroup\$
1
\$\begingroup\$

DACs require some kind of output filtering to compensate for their step-wise output (referred to as zero order hold.) This is often in the form of a low-pass 'anti-aliasing' filter, which is really a sinx/x (sinc) reconstruction filter.

More about that here: https://www.analog.com/en/technical-articles/equalizing-techniques-flatten-dac-frequency-response.html

\$\endgroup\$
4
  • \$\begingroup\$ Assuming the DAC even has a step-wise output. Many audio DACs don't and even if they do, it will be filtered out by e.g. inductance of headphones. \$\endgroup\$
    – Justme
    Aug 22, 2023 at 21:03
  • \$\begingroup\$ Audio DACs these days use a variety of modulation strategies besides zero-order and are a whole other discussion, probably more than OP was asking for. \$\endgroup\$ Aug 22, 2023 at 21:06
  • \$\begingroup\$ Most audio DACs are ΣΔ, so the sample rate is very high, about 3 MHz. So they usually don't need anti-imaging filtering. No more than a passive RC circuit and maybe a voltage follower. \$\endgroup\$ Aug 23, 2023 at 3:33
  • \$\begingroup\$ Again, while interesting, beyond the scope of OP’s question… \$\endgroup\$ Aug 23, 2023 at 5:32

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.