Notch filter compensation
The problem is how to calculate or estimate the insertion loss at the fundamental by means of interpolation. Perhaps B-splines again?
Cheers, E.
But I'm still busy with this: DiAna, a software Distortion Analyzer
The problem is how to calculate or estimate the insertion loss at the fundamental by means of interpolation. Perhaps B-splines again?
Cheers, E.
Last edited:
Notch filter characterization
See: DiAna, a software Distortion Analyzer and further. However, be aware of possible pitfalls like overloading etc.
Cheers, E.
Hi Jan,
See: DiAna, a software Distortion Analyzer and further. However, be aware of possible pitfalls like overloading etc.
Cheers, E.
Edmond: in my distortion compensation project I tried estimating the RC filter component values by nonlinear regression (curve-fitting) of the respective circuit equations to the measured transfer values (amplitude and phase shift), initialized with imprecise values of the circuit components. Worked very precisely. Once the circuit model has the exact component values determined by curve-fitting, you can probe it for any parameter at any condition.
In your case the equations would be more complicated since the filter is a bit more complex, but you could offer a few typical circuits + user would enter the nominal component values. Good libraries can curve-fit complex values which makes the equations much simplier.
The equation for RC LPF was trivial with complex numbers nonlinear-compensation/lowPass.m at 078fb9f1c572568f343bb0bd652b2ea55874896d * pavhofman/nonlinear-compensation * GitHub
Finding exact values of R1, C1, Rin to fit the measured transfer values (in my trial just randomized exact values) were only several lines nonlinear-compensation/testLP.m at 078fb9f1c572568f343bb0bd652b2ea55874896d * pavhofman/nonlinear-compensation * GitHub
The core of my algorithm is based on nonlinear curve-fitting and the fitting precision is extremely good (sum of squared errors at 1e-20), of course provided the equations are correct.
I am sure it would work, especially if you let some software derive the equations for you.
Maybe some interpolation will work too, but definitely not as accurately as the circuit model fitted to measured values. On the other hand you do not need any major precision in your interpolation, it does not matter if diana measures values with a few percent of relative error.
Just 2 cents.
In your case the equations would be more complicated since the filter is a bit more complex, but you could offer a few typical circuits + user would enter the nominal component values. Good libraries can curve-fit complex values which makes the equations much simplier.
The equation for RC LPF was trivial with complex numbers nonlinear-compensation/lowPass.m at 078fb9f1c572568f343bb0bd652b2ea55874896d * pavhofman/nonlinear-compensation * GitHub
Finding exact values of R1, C1, Rin to fit the measured transfer values (in my trial just randomized exact values) were only several lines nonlinear-compensation/testLP.m at 078fb9f1c572568f343bb0bd652b2ea55874896d * pavhofman/nonlinear-compensation * GitHub
The core of my algorithm is based on nonlinear curve-fitting and the fitting precision is extremely good (sum of squared errors at 1e-20), of course provided the equations are correct.
I am sure it would work, especially if you let some software derive the equations for you.
Maybe some interpolation will work too, but definitely not as accurately as the circuit model fitted to measured values. On the other hand you do not need any major precision in your interpolation, it does not matter if diana measures values with a few percent of relative error.
Just 2 cents.
Ralf, I'm trying to figure out the calculations.
For H2, the measurement (I think) is -94.6dB. The notch atten for H2 is -8.8dB. So the actual H2 is then -85.8dB, right?
The amp gain is 40.1dB. That gives me -125.9dB. But the calculated column says -124dB. Maybe I miss something, but I can't solve it.
How do you do it?
Edit: if I figure in the atten from 600R Zout and 2.5k Zin I get the missing 1.8dB. But that would be for both the (notched) fundamental as well as the harmonics, so should make no difference. I think ...
Jan
For H2, the measurement (I think) is -94.6dB. The notch atten for H2 is -8.8dB. So the actual H2 is then -85.8dB, right?
The amp gain is 40.1dB. That gives me -125.9dB. But the calculated column says -124dB. Maybe I miss something, but I can't solve it.
How do you do it?
Edit: if I figure in the atten from 600R Zout and 2.5k Zin I get the missing 1.8dB. But that would be for both the (notched) fundamental as well as the harmonics, so should make no difference. I think ...
Jan
Last edited:
Your conclusions are correct. The value of the attenuated fundamental doesn't matter in the calculation. I do the calulation with the unattenuated input fundamental. This has the advantage that I don't need to know the fundamental attenuation of the notch filter. In addition, if the oscillator frequency deviates from the notch frequency, then the attenuation is different again.
But my point was that you have to (and I think you do) take into account the attenuation from DUT Zout and notch Zin, the 600R - 2.5k. Right? It is a fixed factor in your test setup.
Jan
I tried to characterize the AP notch filter but it doesn't seem to work. Probably some interference from what the AP thinks it should do.
I set the AP to fixed 1kHz notch, set DiAna to 1kHz too. Edmond, the waveform I see in characterisation mode, is that the ref channel or the test channel? It shows a rather distorted wave, 20% THD, looks like strong xover.
The output of the notch on the scope looks like the usual noise with little signal detectable, which is to be expected.
Jan
This would be a very nice and elegant solution. As a matter of fact, I was already thinking in that direction, but that means a project on its own right, because a lot of math is involved (and I don't have a library that takes most of the work off my hands). For the time being and out of curiosity, I will first try 3rd degree B-splines (simply because I'm familiar with it) followed up by a rational function interpolation. The latter is most likely more appropriate because poles are involved.
Yes it can. After an analysis you can select the option "Reference" (see pic), in which case the amplitude of the fundamental is substituted by the Output level as specified in the Distortion menu. Now the harmonic amplitudes are referred to this level. That's what you want, right?
Cheers, E.
Attachments
Another option is measuring the transfer of the filter at the required frequencies directly by diana, using a simple switchable adapter. Since the code already talks to the soundcard, no major complication.
Transfer parameters of the filter vary only very little, for this purpose the calibration would hold valid for months.
Transfer parameters of the filter vary only very little, for this purpose the calibration would hold valid for months.
I tried your calculator. The chain was:
Gen - Viktor1k, output 1.9Vrms, +5.5dBv
Notch - passive twin-t notch 1k
Preamp - OPA827 +40dB
Analyzer - XFi USB HD modded
Attachments
I remember when I first got the Shibasoku and wanted to confirm the harmonics. I had to create an essentially phase locked harmonic tone at a precise level and mix with the Victor oscillator to verify the harmonic accuracy. Complicated to do but essential to verify accuracy. It was possible to see the effect of the phase causing cancellation of harmonics. However this is an essential step when looking so deep. Verifying harmonic level and confirming they aren't cancelling is important. Its a key piece of Pavel's efforts.
I suspect a measurement error / misunderstanding here. Can you connect the oscillator directly to the ADC and adjust the level so that it has about 0dBFS and then insert the DUT + notch and not adjust the level. And set the level also to 0dBFS @ calculator.
Moreover I suspect you are using the notch circuit from Victor. Has it really an input impedance of about 5100 ohms @H2?
Last edited:
there is no DUT, DUT is a generator itself
analyzer 0dBFS = 0.5Vrms, -6dBV
notch H1 imp cc 5k
notch H2 imp cc 4k
notch H3 imp cc 3k5
The calculation is correct, let's see
From analyzer, [dBV]:
H1 -10
H2 -112
H3 -115
Amplified by 40dB, after correction [dBV]:
H1 -50
H2 -152
H3 -155
Notch attenaution [dB]
H1 -55
H2 -9
H3 -5
After correction, generator level [dBV]
H1 +5
H2 -143
H3 -150
Distortion in dBc
H2 -148
H3 -155
I actually think the calculation is more simple than that.....
Using the same numbers:
No notch:
H1 = -6dBV
Harmonics measured with notch:
H2 = -112 after 40 dB amplification
H3 = -115 after 40 dB amplification
Notch attenuation measured using H2 and H3 test signals:
H2att = -9 dBV
H3att = -5 dBV
Harmonics corrected for notch attenuation and amplification:
H2 = -112 + 9 - 40 = -143 dBV
H3 = -115 + 5 - 40 = -150 dBV
Relative to fundamental:
H2 = -143 + 6 = -137 dBc
H3 = -150 + 6 = -144 dBc
I can see that the result is different and that worries me, even though I can't see where I'm doing something wrong
As pointed out by diyralf the attenuation of the fundamental should be kept out of the equation as it is unreliable to quantify it accurately and irrelevant as long as it is big enough not to interfere with the quantification of the harmonics.
As long as the depth of the notch @ H2 and H3 are actually measured in the same circuit, and not table values, I also cannot see why we should correct for input impedance of the notch as this is already included.