Forum Discussion

Altera_Forum's avatar
Altera_Forum
Icon for Honored Contributor rankHonored Contributor
15 years ago

Making an FIR filter work correctly ...

i'm using the following components from altera:

- altera DSP Development Kit Stratix III Edition with a EP3SL150N board

- Stratix III EP3SL150F1152N FPGA

- HSMC with 2x 14-Bit DAC and 2x 14-Bit ADC

- Quartus II 8.0 wich SP1 (subscription edition / academic)

i'm using the following measurement equipment:

- Tektronix AFG3101 signal generator (one channel)

- Agilent InfiniiVision DSO7012A Oscilloscope (two channels)

- mRS mini VNA vector network analyzer

- mini circuits SLP-50+ alias filter (at DAC output)

The attached .zip-file contains all information about my project, including measurement results and an archive file of the complete project.

I have the following problem: my goal is to have the filter working correctly, allowing input / output voltages up to approx. 2V. Currently, the signal is deviated inside the filter and becomes useless. I've tried many things regarding "signed" and "unsigned" representations, but still hve not found a combination that works - or is the problem located somewhere else?

I desperately need help - I'm walking trough a horrible hell of try and error for about 4 weeks now ... :confused::confused::confused:

remark: The filter curve is a measured attenuation of a house installation in the broadband powerline frequency band.

26 Replies

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    Hi,

    In my short time, I can only suggest using signaltap. I use it regulary for debugging and is wonderful if done correctly and in conjunction with a math tool. It will debug the most difficult issues up to fpga io.

    You can capture filtered output, see in time domain or frequency domain(fft).

    If you intend to go that way then I can post few Matlab statements to read signaltap text files and display results.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    I've you would have opened the .zip-file, you would have seen, that I already used signaltap ... :mad: ... but it just didn't help me ...

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    May I ask what are these coefficients meant to be (Low pass?). They look very strange to me.

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    I'll quote myself ... :D

    --- Quote Start ---

    remark: The filter curve is a measured attenuation of a house installation in the broadband powerline frequency band.

    --- Quote End ---

  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    I am afraid you need to clear up the issue.

    Are you talking about an equaliser response derived from measured input. However you are using low pass filter in your setup yet the filter coefficients

    are bizarre.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    As with your previous thread, I see two different problems, that should be clearly distinguished

    - about signed/unsigned data representation

    - about intended versus achieved filter characteristics and attenuation

    The first point is basically simple, although I don't understand your results in every detail. The shown SignalTap waveform (also included in your project *.qar) can be simply turned into plausible data by changing the output data representation to signed. I didn't check at which point of the design the data representation is changed, but it obviously is. As previously discussed, you have to care to use the correct data representation to interface ADC and DAC. It's onbviously O.K. for the ADC.

    For the filter characteristic/scaling topic, you have to know, what your filter is intended to do and how the FIR compiler works. As I already mentioned, I'm not familiar with this tool. But I'm sure, that it's operation can be understood. This has been at least the case with the previous scaling issue, in my opinion.

    P.S.: You can also try the free ModelSim Altera edition, but I fear, there is a considerable initial learning amount from where you start.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    @kaz:

    A house installation has a kind of low pass characteristic, you are right. The coefficients come directly from a transmission measurement with a network analyzer (a really good one from Rhode&Schwarz), and are converted by coeff.exe from altera into compatible data.

    @FvM:

    Turning the output from "unsigned" to "signed" does not improve the signal tap signal - if you have a closer look, you will see that the sinus is somehow "plattend" to an unpure sinus.

    The FIR compiler manual is not very helpful in my case - I've gone through it after your last remark, and didn't find anytghing in detail about coefficient scaling or the signed/unsigned problems - the only sentence about it says that "Auto" provides the best SNR.

    As said before - everthing around the filter (seems to work) works correctly. I mean, this should be no rocket science - I have just a digital filter on the board and want it to work - it can just not be that I'm the first one attempting this feat ...
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    Fair enough.

    Try use one number system e.g. signed for all signal path i.e. input/coeffs/output.

    Also try multipliers instead of DA if you can afford it or use lower taps
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    to your first hint:

    If I want to make everything signed - how do I do that?

    - The coefficients are already signed

    - I switch the input data of the FIR in the drop down menu to "signed"

    - I switch the reprasentations to "signed line chart"

    So far, I already tried it out - now to the points that are unclear to me ...

    - How can I access / influence the input data?

    - Is there another option that needs to be changed?

    to your second hint

    --- Quote Start ---

    Also try multipliers instead of DA if you can afford it or use lower taps

    --- Quote End ---

    You need to describe this a little bit more in detail, if you want me to understand this. How can I replace DACs with multipliers? Whow should I access such multipliers from outside the board? What are "lower taps"? LEO tells me, that "taps" means "Ablassventil" in german, and I do not know what is exactly meant by that in this case.
  • Altera_Forum's avatar
    Altera_Forum
    Icon for Honored Contributor rankHonored Contributor
    15 years ago

    For the input to be signed you need an ADC that gives you signed. If it is other systems then convert.

    With regard to multipliers I meant by DA = Distributed Arithmetic not DAC, sorry.

    100 multipliers can be too much but a lower coefficient set can help testing, you may for example decimate the coeffs to 50 by dicarding very other...