Unstable Phase Shift Issue

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

wordave

Member
Joined
Oct 5, 2010
Messages
23
Location
Newcastle/Huddersfield UK
I've been working on a set of multifeedback filters for a preamp I am designing and I've encountered a problem. The 100Hz filter I have designed is unstable. I've attached the frequency response and phase response tests as well as the schematic. I've also attached the phase shift vs frequency graph supplied in the TL074 data sheet (Op amp used in the tests).

Both the phase response graph and TL074 show different phase shifts in the <100Hz region. This is what I think is to blame for a unstable frequency response.

One question ... What is going on?! I've really exhausted my knowledge on this subject and would appreciate any insight anyone may have. I'm 99% sure its phase causing the problem, just not sure why.

Download the files here: http://www.inverseaudio.co.uk/Filterissues.zip
 
I would suspect the problem lies with C1, C2 and C5 which together place three LF poles in one of the feedback loops. Your phase plot also seems to indicate this has lead to a very poor phase margin around 100Hz.

Cheers

Ian
 
Samuel Groner said:
You should include R and C values in your schematic...

My apologies I thought I had. See new attachment.

ruffrecords said:
Your phase plot also seems to indicate this has lead to a very poor phase margin around 100Hz.

Could you possibly expand on your poor phase margin comment? I understand the concept of phase lag that can cause positive feedback/oscillations. I'm not sure how the phase margin becomes poor at 100Hz and then how that is effecting the frequency response.

Regards,
Dave
 

Attachments

  • Screen shot 2012-04-25 at 10.07.48.png
    Screen shot 2012-04-25 at 10.07.48.png
    22.6 KB · Views: 15
MFBP.jpg

Is this copied from something working or is it your own invention?

What is it supposed to do?  Is it a parametric?

What are the screen shots showing?

Does it work (no oscillation) when designed for frequencies other than 100Hz?
 
It is based on other ideas but I adapted it to what I need. It is a graphic equaliser (Multi feedback bandpass summed either in or out of phase to the boost and cut summer respectively).

It works fine above 100Hz. I built a 250Hz filter and that shows normal performance.

The phase graph shows the output compared to the input. The frequency response is a simple sine sweep (in boost, cut and flat positions).
 
Can you redo MFBP.jpg with both component values and numbers?  It's very hard to talk about these without both component values and numbers.

Your curves show what appear to be 4 phase responses and 3 frequency responses.  Which is boost / cut / flat?  If the 100Hz filter is oscillating, how did you do the responses?

Do you have similar curves for the working 250Hz filter?
 
Ok, here is separate frequencies response graphs in boost and cut positions for the 100Hz and 250Hz filters. There is also separate phase responses of the 100Hz and 250Hz bandpass filters. The entire schematic is also attached.

The 250Hz isn't perfect as you will see (same low end issues as 100Hz) but it's less extreme and much closer to a working filter.

Thank you for your patience and I hope these graphs can help diagnose the problem.

Dave

www.inverseaudio.co.uk/Filterproblems.zip
 
Dave, it looks like you are using an FFT type analyser for your frequency response.  What size FFT are you using?  The resolution looks like 30Hz.

When you get so close to the lower limit of the FFT, it gives wrong answers unless you do a bunch of careful windowing and other DSP guru stuff.  Even then you need to interpret the results carefully.

Try measuring with sine waves at spot frequencies and plot your curves by hand.

You haven't said if your 100Hz filter is oscillating.  If it is not oscillating, it is not unstable.  If it was unstable, you would not be able to measure a frequency response.
 
The only test I had available at the time was a sine sweep using spectrafoo software. It took a lot of trial and error with the signal generator just to get readings that are readable.

I don't think it is oscillating, although if it is, would that explain the issue?
 
wordave said:
The only test I had available at the time was a sine sweep using spectrafoo software.
  I don't think it is oscillating.  I think Spectrafoo is telling lies.

Try using Audacity, a free multi platform DAW to send & record some spot frequencies and plot the response by hand.  I think you will find this will give very different results at LF.  Trust the spot frequency hand plotted response.
 
The circuit is at Uni being marked and moderated, I'm asking these questions because I am discussing this issues in a technical report. As soon as I get it back I will plot the response by hand.

In the mean time, how can the FFT size in spectrafoo create such a error in a reading? I had the analyser set to 4096pt (10Hz), anything shorter than that started to slow the laptop down and made the whole program unstable.
 
The 30Hz steps in the response tell me that your FFT is quite small.  An FFT is only an approximation to a "frequency response".  The approximation is poorest at the highest & lowest frequencies.

As you approach the limits, the FFT becomes VERY sensitive to phase.  Extra phase shift eg cos you have extra boost, cut or even a higher slope, make the amplitude response wonky.

Some of this is explained in "Extending Quasi-Anechoic Measurements to Low Frequencies" - Eric Benjamin, AES oct04 San Francisco

If Spectrafoo allows it, try doing a 1M point FFT analysis.  That should push the problem out of the audio range but the spot frequency hand plot will still be more accurate.  I may be wrong but this is my guess.

I'm not a Mac person so I can't recommend software which does convenient response via alternative methods.  In Windoz, RMAA does both FFT and the equivalent of spot frequency measurements.

The bottom line is that an FFT response is 'accurate' only if the response flattens off before the lowest frequency point is reached and also before the Nyquist frequency.
 

Latest posts

Back
Top