Advertisement

Blog

DAC BASICS, Part 4: The Pesky DAC Output Glitch-Impulse

In DAC BASICS, Part 3: Static specifications, we talked about the static specifications and their impact DC characteristics, such as offset, gain, and linearity. These characteristics are generally consistent across the various topologies of the R-2R and string digital-to-analog converters (DACs). In contrast, the glitch-impulse behavior of the R-2R and string DACs are dramatically different.

One can observe a DACs dynamic non-linearity while running at its operating sample rate. There are many reasons, but the most significant ones are glitch-impulse, slew rate / settling time and sample jitter.

One can observe glitch-impulses while running across the DAC’s output range at a consistent sampling rate. Figure 1 shows this phenomenon with the DAC8881, a 16-bit R-2R DAC.

Figure 1

This 16-bit DAC (R-2R) output shows glitch-impulse characteristics located at 7FFFh - 8000h code changes.

This 16-bit DAC (R-2R) output shows glitch-impulse characteristics located at 7FFFh – 8000h code changes.

What is happening?

Ideally, the DAC’s output moves from one voltage value to the next in its expected direction. Contrary to this ideal example, real DAC circuits have undershoot or overshoot characteristics during some code-to-code transitions.

This characteristic is not consistently the same from every code-to-code transition. Some transitions are more dramatic than others. The glitch-impulse specification quantifies this characteristic. The DAC glitch-impulses can disrupt a closed-loop system by momentarily outputting erroneous voltages.

Figure 2 shows an example of a DAC with a single lobe glitch-impulse. A string DAC usually produces this type of glitch-impulse.

Figure 2

Single lobe DAC output glitch-impulse characteristic.

Single lobe DAC output glitch-impulse characteristic.

In Figure 2 , the code transition is from 7FFFh to 8000h. If you convert these numbers to a binary form, notice that every bit in these two hex codes are either switching from 1 to 0 or 0 to 1.

The glitch-impulse specification quantifies the amount of energy in this glitching phenomenon, carrying the units of nano-volts-seconds or nV-sec (GI). This glitch-impulse quantity is equal to the area under the curve.

Single-lobe glitch-impulses are a consequence of the DAC internal switches being out of sync. What causes this DAC phenomenon? The synchronization of the internal DAC switches is not always perfect. As the parasitic switch capacitances charge or discharge, one can see these charge exchanges at the DAC’s output.

R-2R DACs produce two regions of the glitch-impulse error (Figure 3 ). With the dual impulse error, subtract the positive glitch-impulse (G2) from the negative glitch-impulse (G1) to produce the final glitch-impulse specification.

Figure 3

DACs with an R-2R internal structure exhibit a double lobe glitch-impulse.

DACs with an R-2R internal structure exhibit a double lobe glitch-impulse.

Again, in Figure 3 the code transition is from 7FFFh to 8000h.

In order to understand the sources of a DAC glitch-impulse, we must first define the major-carry transition. A major-carry transition is the point where a MSB is changing from low to high while the lower bits are changing from high to low (or vice versa). An example of this code change is from 0111b to 1000b, or even more dramatic a change from 1000 0000b to 0111 1111b.

One may think that this phenomenon occurs where the output of the DAC exhibits a large change in voltage. This is not the case with virtually every DAC coding scheme. Refer to reference 1 for more details.

Figure 4 and Figure 5 illustrate the impact of this type of glitch with an 8-bit DAC. To the DAC user this phenomena occurs with a single LSB step, or in a 5 V, 8-bit system a 19.5 mV step.

Figure 4

In this 8-bit DAC configuration, the internal switches have seven R-2R legs tied to VREF while there is one R-2R leg tied to ground.

In this 8-bit DAC configuration, the internal switches have seven R-2R legs tied to VREF while there is one R-2R leg tied to ground.

Figure 5

In this DAC configuration, the internal switches have one R-2R legs tied to VREF while there are seven R-2R legs tied to ground.

In this DAC configuration, the internal switches have one R-2R legs tied to VREF while there are seven R-2R legs tied to ground.

As the DAC loads a code, there are two areas that produce output glitches: switching synchronization, and switch charge transfer of simultaneously triggered multiple switches.

The string DAC has a single switch topology. A string DAC taps into different points on a giant resistor string. The switching network does not require multiple transitions across a major carry and, therefore, is less prone to glitch. There will still be a minor glitch from the switch charge, but it is minimal in comparison to R-2R structure DACs.

The R-2R DAC has multiple simultaneous switches switching during code transitions. Any lack in synchronization leads to a brief period where all switches are high or low, causing the DAC’s voltage output to migrate to the rail. The switches then recover creating a lobe in the opposite direction. The output then settles.

The voltage locations of these glitches are very predictable. With the R-2R DAC, the worst case glitch errors occur when all of the digital bits are toggling while still transitioning with a small voltage output change. This is the case when a DAC code changes with a major carry transition; from codes 1000… to 0111….

Examine real DAC behavior

Now that we have defined the candidate code transitions for glitch-impulse errors, we can take a closer look at an R-2R and string DAC glitch-impulses with the 16-bit DAC8881 (R-2R DAC) and the 16-bit DAC8562 (string DAC).

In Figure 6 , the glitch-impulse of the DAC8881 is 37.7 nV-sec and the glitch-impulse of the DAC8562 is 0.1 nV-sec. In both these pictures, the x-axis scale is 500 ns/div and the y-axis scale is 50 mV/div.

Figure 6

R-2R and string glitch-impulse performance

R-2R and string glitch-impulse performance

Glitch be gone

If a DAC glitch-impulse problem exists, one can use external components to either decrease the glitch amplitude (Figure 7a), or remove the glitch-impulse energy all together (Figure 7b ).

Figure 7

Reduce glitch-impulse error with a first-order lowpass filter (a) or sample/hold solution (b).

Reduce glitch-impulse error with a first-order lowpass filter (a) or sample/hold solution (b).

An RC filter after the DAC decreases the glitch amplitude (Figure 7a ). The glitch-impulse period determines the appropriate RC ratio. The 3 dB frequency of your RC filter is one decade prior to the glitch-impulse frequency. As you select the components, make sure the resistor is low, otherwise it will produce a voltage drop in combination with the resistive load. Since the glitch energy is never lost, the tradeoff for implementing a single pole low-pass filter is to spread out the error while increase the settling time.

A second approach is to use a sample / hold capacitor and amplifier (Figure 7b ). The external switch and amplifier removes the glitch from the DAC internal switching, leaving a small S/H switch transient. In this design, the switch is left open as the DAC crosses over a major carry transition. Once the transition is complete, the switch closes allowing the new output voltage to be set across the CH sampling capacitor. The capacitor continues to hold the new voltage as the external switch opens when the DAC is ready to update its output. This solution is more expensive and uses more board space, but you are able to reduce/remove the glitch without the cost of increased settling time.

Conclusion

Glitch-impulse is one of the important dynamic non-linearity DAC characteristics that you will run into as the device at its operating sampling rate. But, this is only one part of the picture. Other DAC characteristics that can impact high-speed circuits are the slew rate and settling time. Stay tuned for my next article which will be on this topic.

References

  1. Coding schemes used with data converters,” Jason Albanus, Application Bulletin (AN-175), Texas Instruments
  2. Use an RC filter to ‘deglitch’ a DAC

18 comments on “DAC BASICS, Part 4: The Pesky DAC Output Glitch-Impulse

  1. andrewmm
    April 29, 2015

    why do you say the glitch energy is not attenuated by the rc filter, just 'spread around'.

     

    A low pass filter, attenuates the high frequency energy. It dosn't reflect it down to lower frequencies.

        if it did not attenuate that would be great in the high power antena filters, they would not get hot. 

     

    If I put in two tones to a low pass filter, then the higher frequerncy one is attenuated, the noise floor does no change as the higher tone moves up in frequency, neither does the amplitude of thelower tone chanmge as I move the frequency of the higher tone.

     

    I've heard this said before abotu an rc network on th eoutput of a dac, but nether fomr some one that I could ask,

    what am I missing 

     

  2. Paul Bryson
    April 30, 2015

    You are right the signal IS attenuated by the LP filter. The LP filter does attenuate the higher frequency components of the glitch. But a short glitch pulse has a broad spectrum.  I would say the energy is “spread around” in the time domain – not the frequency domain.  The LP filter turns a high amplitude short pulse into a lower ampltude longer pulse.  

  3. andrewmm
    May 5, 2015

    Looking at glitch energy,

     

    the longer , low pulse I think that must have a less energy than the initial short pulse , if the filter is doing its job then it should be a lot less.

    The LPF does not convert the high frequenices into low frequencies, it absorbs the high frequencies, Thats why power filters get hot is it not.

     

     

     

  4. Paul Bryson
    May 5, 2015

    (Disclosure:  I do not consider myself an expert on this subject. And the following is me thinking out loud – so to speak.)

    I was more or less agreeing with you.  However, I don't believe that I have a good intuitive understanding of energy.  I mostly deal with voltages, currents and power.  That being said, your example of heat being dissipated as evidence that the energy of the signal is being reduced does not really track.  

    For example, in an RC LP filter the power dissipation is more a function of impedance than attenuation of the voltage signal. If I take any RC LP filter and increase the resistance by 10x and decrease the capacitance by the same factor; the voltage attenuation will remain the same but the power dissipation will decrease.  This can be extended to increase or decrease the power dissipation by any arbitrary amount.  

    So as a thought experiment, if I increase the resistance to a very large value and the capacitance to a tiny value then the shape of the DAC glitch is effected the same as with the original LP filter but no power is dissipated as heat.

    Here is another thought (My brain is hurting now.):  The way that the sum of an inifnite series of sinusoids can result in a single narrow pulse is that they must mostly all cancel out in the time domain.  So, if you filter out the high frequency components, could you conceivably increase the overall energy by removing some of the cancelling components?    

     

  5. andrewmm
    May 6, 2015

    In your example, I'd think of a RC low pass filter as a voltage divider.

       The impedance of the R and the C cause a frequency dependent attenuation.

    If I put 1 M watt at into a 1 nf capacitor 100 MHz, then the cpacitor will pop unless its big. 

     

    An interesting discussion I wish the orriginal auther of the article would get back to us,

        I have seen a similar article many times over the years, but this is the first time I have thought of asking why.

     

    may be I'm to old now that I dont mind stating that I dont know. 

     

     

  6. Paul Bryson
    May 6, 2015

    I don't understand your response.  It feels like we're not talking about the same thing. Yes, an RC low pass filter is a frequency dependent voltage divider. Yes, 1MW into any reasonably sized component would blow it up.  But the task at hand is filtering signals of a given voltage – not a given power.  For a given voltage waveform I can make the power dissipation as low I want by increasing the resistance.

    Maybe a point confusion is energy verus power.  They are not the same thing.  The energy of the signal is the integral over time of the voltage squared – power doesn't enter into it.

     

  7. andrewmm
    May 7, 2015

    Are your saying the area under the curve of the glitch after a low pass filter is the same as the area under the curve of the glitch before the low pass filter ?

     

  8. Paul Bryson
    May 7, 2015

    I didn't say anything like that.  I have just been trying to express that the heat generated when filtering a voltage signal does not necessariiy have anything to do with the energy of the voltage signal either before or after filtering.  i.e. I can filter the voltage with near zero heat produced (as near as I want).

    The ultimate answer to the question of whether the energy of the pulse is the same before and after filtering is a calculus problem which I have been too lazy to do.  My calculus is so rusty that it would take me hours to work it out.  I just don't have any experience working with the energy of a signal. 

    I did find something that leads to an intuitive analysis of this problem.  Parseval's theorem states:

    the energy of signal x(t),  E = ∫|x(t)|²dt = ∫|X(f)|²df

    So if I properly understand this, the energy of a signal is equal to the area under the square of the signal which is also equal to the area under the square of the spectral density (spectrum amplitude).  So, low pass filtering a broadband signal should reduce its energy.

    But does it really matter?  It is the voltage waveform and spectrum that matters in any application that I can think of.  The summary: The pulse amplitude is reduced but its duration is increased. Its bandwidth is also reduced.

     

     

  9. andrewmm
    May 8, 2015

    Thanks

     

    like you say,  my calculus is well rusty, in my case almost 50 year rusty..

     

    its an interesting problem that I wish the auther of the article would answer,

       

    at the end of the day, as you say its the amount the glitch disturbs the output that matters,

       

    My hunch is oposite to yours , in that a low pass filter does attenuate the glitch, and not just smear the glitch out over time.

    But i'd like to hear an experts mathamatics / experimental views. 

    Thanks for the thoughts, an interesting one ah.

     

    ( its also interesting to see how the picture of bonnie has changed over the years ) 

     

     

  10. Paul Bryson
    May 10, 2015

    Wow.  I said exactly the opposite of what you think I said. 

    I said the following things:

    1. The voltage waveform IS attenuated by the low pass filter.

    2. Production of heat by a low pass filter is not convincing evidence that the resulting waveform has reduced energy.  Because I can change the energy of the signal without creating significant heat. … or I can create a lot of heat. 

    3. Parseval's theorem seems to indicate that reducing part of the frequency spectrum, i.e low pass filtering WILL reduce the energy of the glitch.

    4.  The voltage waveform matters more than the energy.

    5. When people say the pulse is “spread out” they are talking about the time domain – not the frequency domain.

    6. I implied but did not state: For a GIVEN IMPEDANCE, more energy = more power i.e. heat.  But heat is not likely to be of concern when talking about low level signals nor for very short pulses. (However, the voltage amplitude could exceed the dynamic range of the system.)

  11. andrewmm
    May 11, 2015

    Very many appologies,

    you are correct :

     

    I was mearly trying to shrink things into bite size portions to be served up and thought about some more over Tee this morning.

     

    Was not tryign to accuse yo of anything, or infer anything that was not there,

     

    just trying to cross reference and to try to see if I had things about right .

     

    Very many appologies for any offence taken, none was implied or satated, just trying to wind things up and wonder if anytone else has any other thoughts,

     

     

  12. Bonnie Baker
    May 25, 2015

    Paul and andrwemm,

    Excellent discussion!

    It is true, that the low pass filter does reduce the magnitude of the voltage spike, however there is not free lunch. The area under the glicth curve is averaged out over time. That is to say, the information is not lost, just reshaped. The trick is to use this filter to reduce the impact on the system (for instance you are intested in a certain level of accuracy: bits) however, it may take time to reach your final desired value.

    On another note, yes my picture has changed over time. I choose to look at it like a good red wine. I feel like I am improving with age.

     

  13. andrewmm
    May 29, 2015

    Thank you for replying to your article

     

    Re your picture, I must agree with the red wine and age analagy.

     

    Re the area under the curve,

        that would be proportional to the energy,

           and the thought is the R/C network is absorbing the energy, not spreading it over time

    If it was just spreading it over time, why do the R/C filers on the output of a power amp get hot ? are we saying its just the ESR / losses in the capacitor that causes the heat ?

     

     

     

  14. Bonnie Baker
    May 29, 2015

    andrewmm,

    The R/C network is not absorbing the energy. It is in fact spreading it over time. No energy is magically lost.

    The DAC output filter and amplifier do not get hot as you would expect in a power application scenario. At this juncture in the circuit, you are simply applying a low voltage analog filter. There are no large voltages (the DAC typically uses single supply, 0 to 5 volt supplies) and the resistances are reasonable (1 to 10k ohms). Also, the ESR is not becoming an issue in this application because the magnitude of dicrete resistor overshadows the maginitude of the capacitor ESR.

  15. andrewmm
    May 29, 2015

    cant see how an RC can smear the enmergy over time,

     

    If I put a defined pulse into a R/C circuit, assuming the filter is working,  the energy before the filter is more than after the filter, thats what filters do is it not ?

    Thats why power filters get hot,

     

    Yes in the dac case, they are smaller etc than power filters, but the energy does not know that, the energy before the filter is more than after it.

    I'm going to have to get a scope, a pulse gen and an rc and test this me thinks.

    Unless you have doen this that is , then I'd love to see the plots.

    Interesting conversation here in the lab.

    Were all 'old timmers' and no one can come up with the deifnit answer…

     

     

     

  16. Bonnie Baker
    May 29, 2015

    If you have a DAC with higher glitch characteristics, there are design considerations to help with the glitch. Using external components, you are able to either decrease the glitch amplitude. With this method your add a simple RC filter after the DAC. Doing so, will attenuate the amplitude of the glitch at the cost of increasing the settling time

    Choosing the appropriate RC ratio can be determined by looking at the period of the glitch and selecting a 3dB point a decade prior to the glitch frequency. When it comes to picking exact components, make sure that the R is kept fairly low or else it will produce a voltage drop what in combination with a resistive load.

    The attached scope plot shows the effects of an RC filter which extends settling time at the cost of a decreased glitch amplitude. I estimate that the glitch period is about 1us and chose a RC time constant value based off of 80KHz 3dB cutoff frequency.

  17. andrewmm
    May 29, 2015

    Thank you

     

    due to time differences, I'm abotu to go home for the weekend, 

       but will have play with scope on monday.

     

    could not see the attatchment you mention, but that could be my end of things.

     

    just talking the glitch,

      Are we saying the RC filter does attenuate the energy of the glitch, energy of the glitch before  the filter is more than after the filter

     

    Have good weekend,

     

     

  18. Bonnie Baker
    May 29, 2015

    I am sorry you did not see the attachment. When you get back from the week-end activities, send me your e-mail. I can get the digarams to you that way.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.