I am defining the required antialiasing (analog) filter for a chirp generated by a DDS. The chirp is produced in a FPGA (NCO generation) and forwarded to the DAC for the samples generation (with interpolation from the DAC). Because of a dynamic range lower than 16 bits and the natural behavior of the DAC to produce high frequency nonlinearities, I need to include and antialiasing filter which I try to define.

I guess I need to fulfill these requirements:

  • I need to low pass my chirp bandwidth (let's say 10-50Mhz) and cut frequencies above the high frequency BW,
  • I have to ensure that the group delay is flat within my bandwidth,
  • I have to ensure a given stop band value before the image frequencies (so for instance with a 400MSPS DAC, I would have a 350 MHz image frequency),
  • I need to push the dB level of such frequencies under the quantization level for the DAC to avoid distorsions so for instance with a 10 bits DAC I need to have a dB margin from the Gain 0dB of roughly 60dB.

As I am not very used to filters I read articles on the web and could determine those few rule of thumbs to design the best suited filter for my application:

  • Butterworth are good for sharp (brick wall style) low pass filter but the group delay is bad (so my chirp will be distorded in phase domain)
  • Bessel filters are flat for group delay (means that my frequencies will not be phase delayed or at least will have the same delay), but are not as sharped.

For the considered case, with a stopband at 350MHz and a gain ~1 at 50MHz it seems possible to achieve a filtering, especially with Bessel filters (order 3 or 4 are OK).But if I would like to decrease my cost, hence the sampling frequency of my DAC, it turns quickly to be impossible because my stop band and my cut off frequency are getting closer and only a Butterworth filter could do the job? but because it is a group delayed filter it is not suitable for chirp generation am I right?

I tried to simulate this with ADIsimDDS (interesting online software) for instance AD9859 (400MSPS 10 bit) and my case study looks good but i cant see any of the group delay impact with respect to the different proposed external analog filters.

Thanks for your answers and advices.

  • $\begingroup$ What is your sampling rate, chirp frequency range AND chirp rate? Do you have blanking time from the end of the chirp before you "snap" back to the beginning or do you step back right away (right away requires much higher bandwidth and your filtering will end up transitioning this)? Or alternatively do you ramp down? Can you sketch your chirp and what the application is? Also you are describing analog filters (Butterworth, Bessel) which would not be the best choice to try an implement digitally versus a linear phase FIR filter (which will have constant group delay). $\endgroup$ – Dan Boschen Jun 20 '20 at 2:16
  • $\begingroup$ These two answers may help you further: dsp.stackexchange.com/questions/61391/… $\endgroup$ – Dan Boschen Jun 20 '20 at 12:24
  • $\begingroup$ @Dan Boschen The chirp frequency range is 40MHz BW with a central frequency at 30MHz. The total ramp time is in the range 500µs to 800 µs. It is only a upchirp (no ramping down). I am not sure but as the DAC is a source of distorsions, how a digitally implemented FIR filters can filter what will be produce in a next stage? $\endgroup$ – AtoM_84 Jun 22 '20 at 8:38

I recommend a digital FIR filter to both compensate for the DAC Sinc droop and added group delay distortion combined with a simpler analog filter focused on rejecting the higher images out of the DAC with the minimum number of components (this would eliminate using a Bessel filter, while a Butterworth or elliptic filter would be viable candidates). The analog filter cost and complexity can be minimized by utilizing digital pre-distortion to equalize the analog filters group delay variation with a reasonable tolerance.

For further details on the design details for the anti-alias filter, please see this post: Where should I set my anti-aliasing filter corner frequency for this signal?

This post may also be of interest as an implementation approach where I detail all the code for an optimized chirp for a flat FFT response which would also have minimum aliasing effects under the same sampling conditions. Instead of a time domain filter, the response is scaled with a Tukey window which provides a constant envelope chirp over most of its duration, while scaling the very start and end of the chip minimize aliasing effects. This could be done with a weighting factor of the amplitude of the chirp at the output of the DDS. Please see the very bottom of this post for more details showing a flat FFT response over most of the chirp duration: How can I plot the frequency response on a bode diagram with Fast Fourier Transform?

  • $\begingroup$ Thank you @Dan Boschen for your feedback. As my chirp is generated in a FPGA (NCO) and is converted by a DAC (12 to 14 bits dynamic range), it seems to me that I can only implement an analog filter at the output of the DAC to remove all DAC generated frequency images and artifacts generated in the analog signal... Am I wrong? So FIR filters cannot be used for such purpose? My idea was to use a Bessel filter if I am ok with a weak roll off or a Butterworth filter and compensating the group delay effects by a non linear FM at the NCO stage (if achievable). $\endgroup$ – AtoM_84 Jun 22 '20 at 8:23
  • $\begingroup$ You can use both- you will need a simpler filter after the DAC to eliminate its images but you can also use a digital filter to control the bandwidth of the fundamental signal itself—- with the right choice of sampling rate and pulse shape this can really simplify the analog filtering required and be much more effective in controlling the occupied bandwidth of the signal. $\endgroup$ – Dan Boschen Jun 22 '20 at 16:40
  • $\begingroup$ It depends... FIR have significantly higher latencies than their IIR equivalents. I disagree with using the term "out-perform" without context. $\endgroup$ – Ben Jul 20 '20 at 17:15
  • 1
    $\begingroup$ @Ben True, I updated it to "most cases". $\endgroup$ – Dan Boschen Jul 20 '20 at 17:24
  • $\begingroup$ @Ben actually I realize now his question didn't have anything to do with mapping of analog filter techniques so my entire first paragraph was out of context! $\endgroup$ – Dan Boschen Jul 20 '20 at 21:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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