AM audio muffled with BW >= 10972 Hz [fixed in v1.222]

G4DYAG4DYA
edited September 2018 in Problems Now Fixed
Hi all - with an AM broadcast station, if you set the passband width in the URL to 10971 Hz or lower it sounds right, but if you choose a bandwidth between 10972 and 12000 Hz it sounds quite a bit narrower. I think this is a recent bug. Currently running 1.218 with four WF channels.

Comments

  • jksjks
    No, it's been that way forever. The reasons are not understood.
  • jksjks
    I've been working on an audio FFT display (as mentioned in the new admin "mode" tab). The problem is clearly visible and the FFT will be a good diagnostic tool. Hopefully this issue will finally get fixed.
    WA2ZKDPowernumpty
  • jksjks
    Okay, v1.220 fixes this.

    The problem was the AM post-detector "clean up" FIR filter. This filter was inherited from the CuteSDR code. The problem is that when the filter bandwidth approaches the sample rate (12 kHz) bad things happen during the filter tap construction due to the Nyquist limit and the filter not having a small transition band (which would cause it to ring).

    I couldn't hear that the filter was doing much anyway so I just removed it. There is other LPF filtering code in the browser (written in Javascript) to handle artifacts from the interpolation from 12 kHz to the PCs audio sample rate of (usually) 44.1 kHz. And also artifacts when decompression is required. Because the sample rate is larger than 12 kHz the LPF doesn't get into Nyquist trouble.
  • G4DYAG4DYA
    @jks: Excellent stuff - thank you.
  • VK3KHZVK3KHZ
    The AM audio now sounds completely bandwidth restricted with no top-end at all
  • VK3KHZVK3KHZ
    I have done some audio frequency response measurements. The response rolls off quite dramatically, with a slope of ~20dB down across the passband, see attached graph. can something be done to improve the EQ ?

    Attachments:
    https://forum.kiwisdr.com/uploads/Uploader/b8/4030d38aad332a919102a673f464ae.png
  • jksjks
    edited September 2018
    Is that spectrum plot in dBV or dB(power)? What you're seeing seems to be the known FPGA CIC filter droop problem. Although if your plot is in dB(power) the slope is much worse than what I measure.

    For some reason when the AM post-detector FIR filter was enabled it "normalized" the response inside the filter passband even for large passband values. I don't understand why that happens. Maybe some DSP expert can explain it to us. Anyway, I looked a little closer at the original problem that caused me to remove the FIR filter in the last update (passbands over 10971 Hz suddenly get much narrower filter responses). I found a workaround, so now I'll just re-enable the FIR filter and we should be good.

    Eventually the CIC problem needs to be fixed. You can really see it in the spectrum display with IQ mode and >= 10 kHz passband. This is probably negatively impacting things like DRM reception.
  • jksjks
    Works much better in v1.222..
  • VK3KHZVK3KHZ
    Thanks John, I will test later today
    Cheers
    PK ;-)
Sign In or Register to comment.