I hope that this is not too far OT although this is not an active filter.
For some time I have thought of trying out a balanced filter, so that I can measure the distortion of my latest project, which is a DAC/ADC for an audio analyzer.
I started by simulating it in Tina TI. The schematic and the simulation result are shown below.
I then built it using 20 nF polystyrene capacitors and a combination of resistors to get to approximately the right values (6k8 + 1k2). The measured resistor values were a bit too low and the initial measurement shoved a notch frequency of 1004.8 Hz. I then added small resistors in series with the the ones already in place to correct the notch frequency. After the correction I measure a notch frequency of 999.95 Hz, so almost exactly right.
I didn't mount the 1 T resistor (R7) since I was out of stock
Well actually it is only there to keep the simulator happy, because it doesn't like nodes without a DC path to ground.
For some reason the notch depth of the initial version was deeper that the notch depth of the corrected one. The depths were measured at around -90 dB and -80 dB respectively. Probably a question of the matching in the circuit.
The slight peak and drop off at the high frequencies are caused by the DAC/ADC system.
At the low frequencies there is some attenuation caused by the resistors in the filter and the 50 kohm input impedance of the ADC system. But for measurement of harmonics this is of course no problem. I measured the attenuation at the harmonics and put that into a correction file, so that I don't need to correct the harmonic levels manually.
I made some measurements of the DAC distortion, see the attached figure for one example. The results were in line with the DAC specification (AK4399). So the filter does what it was designed for.
There is some mains hum. Nex step is to build it into a metal box. This was measured with an unshielded filter.
The maximum level without the filter is -1.3 dBFS.
Only the left channel is relevant. The right channel was just floating.
For some time I have thought of trying out a balanced filter, so that I can measure the distortion of my latest project, which is a DAC/ADC for an audio analyzer.
I started by simulating it in Tina TI. The schematic and the simulation result are shown below.
I then built it using 20 nF polystyrene capacitors and a combination of resistors to get to approximately the right values (6k8 + 1k2). The measured resistor values were a bit too low and the initial measurement shoved a notch frequency of 1004.8 Hz. I then added small resistors in series with the the ones already in place to correct the notch frequency. After the correction I measure a notch frequency of 999.95 Hz, so almost exactly right.
I didn't mount the 1 T resistor (R7) since I was out of stock
Well actually it is only there to keep the simulator happy, because it doesn't like nodes without a DC path to ground.For some reason the notch depth of the initial version was deeper that the notch depth of the corrected one. The depths were measured at around -90 dB and -80 dB respectively. Probably a question of the matching in the circuit.
The slight peak and drop off at the high frequencies are caused by the DAC/ADC system.
At the low frequencies there is some attenuation caused by the resistors in the filter and the 50 kohm input impedance of the ADC system. But for measurement of harmonics this is of course no problem. I measured the attenuation at the harmonics and put that into a correction file, so that I don't need to correct the harmonic levels manually.
I made some measurements of the DAC distortion, see the attached figure for one example. The results were in line with the DAC specification (AK4399). So the filter does what it was designed for.
There is some mains hum. Nex step is to build it into a metal box. This was measured with an unshielded filter.
The maximum level without the filter is -1.3 dBFS.
Only the left channel is relevant. The right channel was just floating.
Attachments
Richard,
You know if Bas as in Lim is still around?
He used to be one of my go to guys.
Demian,
Is the B&K a non-worthy piece to be sold or to hang on too?
Or still worthwhile to make the active circuit into it?
Last edited:
I dont know for sure.... I havent heard he died yet. Several years ago I asked to buy his business and move it to the Phillipines (were he is from). At that time he had more than one heart attack. But his office employee (Shelley) would take your order or answer questions.
THx-RNMarsh
I didn't know what happened Bas. I hope he's doing well, or better or stabilized
at least. Did anyone buy the business? Like that capacitor guy from AZ
who makes a lot of caps for a lot of different companies?
Or
Did it float off to never never land?
That would have been a natural good choice of buying the biz,
maybe he didn't think he was even healthy enough to make a go
of it back home.
Were you able to second source your caps?
I recall he was getting worried about supply of films as
environmentalist were trying to stop production of some
materials...I guess Bayer gave in.
at least. Did anyone buy the business? Like that capacitor guy from AZ
who makes a lot of caps for a lot of different companies?
Or
Did it float off to never never land?
That would have been a natural good choice of buying the biz,
maybe he didn't think he was even healthy enough to make a go
of it back home.
Were you able to second source your caps?
I recall he was getting worried about supply of films as
environmentalist were trying to stop production of some
materials...I guess Bayer gave in.
Twin t vs. bridged t distortion in simulation
I had both simulations running and thought I would try distortion simulations. I don't put a lot of faith in these, the models are not really 180 db accurate, but they should show any difference. Remarkably they do show a significant difference, one that mirrors what I have been measuring.
The twin T is the variant I have been tweaking that started with the PCB from JensH. I adjusted it for 40 dB notch depth and 40 dB of gain. In the actual notch I have been running into a -135 to -140 dB 2nd harmonic that I'm not sure is real. In the simulation it shows up scaled at approx -160 or so when scaled. its presence in the simulation suggests any real nonlinearity on the opamps will kick it up further.
In the bridged T the second harmonic is completely missing which is a good sign. Now I will need to build the circuit and see if its an improvement.
Attached are the LTspice files and the model pulled from the TI-tina file for the LME49990.
I had both simulations running and thought I would try distortion simulations. I don't put a lot of faith in these, the models are not really 180 db accurate, but they should show any difference. Remarkably they do show a significant difference, one that mirrors what I have been measuring.
The twin T is the variant I have been tweaking that started with the PCB from JensH. I adjusted it for 40 dB notch depth and 40 dB of gain. In the actual notch I have been running into a -135 to -140 dB 2nd harmonic that I'm not sure is real. In the simulation it shows up scaled at approx -160 or so when scaled. its presence in the simulation suggests any real nonlinearity on the opamps will kick it up further.
In the bridged T the second harmonic is completely missing which is a good sign. Now I will need to build the circuit and see if its an improvement.
Attached are the LTspice files and the model pulled from the TI-tina file for the LME49990.
Attachments
In the sim it looks completely clean. I may dive deeper but I need a "peer review" to make sure I'm not setting something incorrectly or otherwise loading the tests incorrectly.
@Demian -- I ran a freq. sweep analysis on the 40dB Bridged-T circuit and found that the notch depth is crucially dependent on the value of the 1.01k feedback resistor -- an ohm of change can make a 20dB or more difference in depth, so temp stability might come to be an issue. But the circuit works well.
I didn't check the notch depth sensitivity to mismatch in the bridge tuning Rs -- a practical auto-tuned circuit will have a mismatch there -- maybe tomorrow night.
I would make no judgement about actual THD until building the real circuit.
Thanks again for the sim circuits.
I didn't check the notch depth sensitivity to mismatch in the bridge tuning Rs -- a practical auto-tuned circuit will have a mismatch there -- maybe tomorrow night.
I would make no judgement about actual THD until building the real circuit.
Thanks again for the sim circuits.
I noticed that it was pretty picky as well. Its mostly lifted from the Shibasoku with a few values changed and the 40 dB opamp on the output. In the simulation resetting it to 10KHz (504 Ohms for the 2 resistors) causes the notch depth to increase by 4 dB.
Looking at the differences in the 3 stages of the Shibasoku suggests there is more to the resistor ratios than is obvious. They have different ratios for each filter.
Looking at the differences in the 3 stages of the Shibasoku suggests there is more to the resistor ratios than is obvious. They have different ratios for each filter.
Weakly nonlinear behaviour is not modelled at all--the residual distortion seen is a side-effect of parasitic element included (e.g. clamp diodes), but not an attempt to model weak opamp distortion. If simulation results seem to resemble real-world measurements, that's mostly just pure accident.
Samuel
Then, the only way to really see/measure what is going on Samuel,
is to build the actual circuit and test it?
and
Any further Spice modelling is a waste of time.
Correct?
Basically yes, but you can estimate the distortion contribution of the passives and opamp(s) using first-order extrapolations from known measurement points (I think I've posted such a calculation earlier in this thread).
Generally speaking I'd advise against the use of "activated" or tunable notch filters for very low (< -130 dB) distortion measurements as they have more distortion mechanisms which need attention. Correcting the frequency response and using large FFTs to get the noise floor down are rather trivial means compared to the real challenge: ensuring sufficiently low distortion contribution of the notch filter.
As posted elsewhere I have designed a set of decadic passive notch filters, using C0G caps, series-parallel MiniMELF resistors and a variable attenuator (0 dB, 10 dB or 20 dB) ahead of the filter. This has proofed to be suitable for measurements in the -140 dB region (in particular it was used by AP in the development of the new APx555 which was just anounced officially today). I hope to implement a few further ideas in a second revision and write a brief text about it once time permits, but if interested I can forward the current Gerber files by e-mail.
Samuel
I remember the discussion of distortion in passives becoming significant at these low levels and finding passives that won't dominate the measurements. A proven layout and parts list would be a very useful tool. I can't justify an APx555 and have not figured out how to get someone to buy one for me so I and others watching will need to resort to other tricks like we are discussing.
A copy of the Gerbers would also be great. You have my e-mail.
The second harmonic in the twin t active filter is a bit of a puzzle. Assuming the opamp model doesn't have a distortion mechanism (as you said modeling tiny distortion is not likely part of the opamp model) its either a number problem from the large numbers and a round off or possibly something from a phase mismatch between the positive feedback and the forward path causing something? That would only happen if there is a nonlinearity to feed it.
A copy of the Gerbers would also be great. You have my e-mail.
The second harmonic in the twin t active filter is a bit of a puzzle. Assuming the opamp model doesn't have a distortion mechanism (as you said modeling tiny distortion is not likely part of the opamp model) its either a number problem from the large numbers and a round off or possibly something from a phase mismatch between the positive feedback and the forward path causing something? That would only happen if there is a nonlinearity to feed it.
RE passive parts problems, my active twin-t notch filter suffers from one obvious major problem -- it has relatively high Z, in the kohms, at 1kHz and below. I did use polypropylene caps, since I didn't have any C0Gs of the right values. I really should re-do it with C0Gs. But I think the metal-film resistors I used are working OK.
Although I hear what Samuel is saying about active filter distortions, I just can't see the distortion problems from the filter -- not saying they aren't there! -- but noise is still a big issue at the moment in my measurements, with ADC possible issues ranking second. I think distortions from the ADC may still represent the largest component, even at -140dB. I long for a better, quieter ADC -- I've reached the limit of what the EMU-0204 and QA-400 can do...
Although I hear what Samuel is saying about active filter distortions, I just can't see the distortion problems from the filter -- not saying they aren't there! -- but noise is still a big issue at the moment in my measurements, with ADC possible issues ranking second. I think distortions from the ADC may still represent the largest component, even at -140dB. I long for a better, quieter ADC -- I've reached the limit of what the EMU-0204 and QA-400 can do...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.