Tony ,I was reading about your FDNR experiments and wondering if I can use it as a low pass filter for POST DAC filtering?
split from here --> http://www.diyaudio.com/forums/analog-line-level/164886-synergy-active-crossover-6.html#post3953096
Note that some early posts may have context from the original thread that seems disjoint.
split from here --> http://www.diyaudio.com/forums/analog-line-level/164886-synergy-active-crossover-6.html#post3953096Note that some early posts may have context from the original thread that seems disjoint.
Last edited by a moderator:
Hi kinku, I believe that there are commercial units which use fdnr's for their anti-aliasing filters. I'm still uncertain about the performance in my circuit, it may be the very low corner frequency that is giving me trouble. If you are familiar with spice go back a few pages and download the spice model that I put up for frank. My blog page in the comments section has some info on how to (cheat) calculate the values using an online calculator.
Walt Jung did an excelent piece on FDNR's in an AD ap note where he presents a design for this exact purpose. http://www.analog.com/library/analogDialogue/archives/39-05/Web_Ch5_final_PtB_F.pdf
quote of section 5.8:
attached is a screen grab of their circuit.
One of the things I have wondered is whether the capacitor size I am using is part of the problem. They are using 10nF caps whereas to get the low frequencies I am using I needed to go with 100nF caps (or use very large resistors).
You can change the relative positions of the caps and resistors, and the filter remains the same slope, but the performance can differ greatly. My current topology is the one that seemed to work best for me.
Also the choice of opamp seems to be quite important. I'm going to try an LM4562 as in simulation it performs much better (going out to 100KHz before the knee in the slope.
Hope that helps.
Tony.
Walt Jung did an excelent piece on FDNR's in an AD ap note where he presents a design for this exact purpose. http://www.analog.com/library/analogDialogue/archives/39-05/Web_Ch5_final_PtB_F.pdf
quote of section 5.8:
attached is a screen grab of their circuit.
One of the things I have wondered is whether the capacitor size I am using is part of the problem. They are using 10nF caps whereas to get the low frequencies I am using I needed to go with 100nF caps (or use very large resistors).
You can change the relative positions of the caps and resistors, and the filter remains the same slope, but the performance can differ greatly. My current topology is the one that seemed to work best for me.
Also the choice of opamp seems to be quite important. I'm going to try an LM4562 as in simulation it performs much better (going out to 100KHz before the knee in the slope.
Hope that helps.
Tony.
Attachments
Last edited:
This is a silly question,but I guess there are few FDNR filters out there using different OPAMPs. When I change OPAMP which are the components that need to be changed in those FDNR circuits. Or those passive components are only deciding the frequencies and OPAMPS can be changed in FDNR without any additional components?
Interesting...im a novice so ive not seen that circuit.
If i take a wild guess (please correct me if I'm a dope) :
1/ the FDNR is not unity gain?
2/ the gain required is actually quite high?
3/ does that inherently result in some instability? I.e. OLG & GBW?
4/ If you posted a link about FDNRs then disregard the above, and could you kindly remind me? Interesting stuff.
5/ if 1-3 are even partially correct (I'm guessing) then the choice of op amp would probably be quite critical as you've noted.
If i take a wild guess (please correct me if I'm a dope) :
1/ the FDNR is not unity gain?
2/ the gain required is actually quite high?
3/ does that inherently result in some instability? I.e. OLG & GBW?
4/ If you posted a link about FDNRs then disregard the above, and could you kindly remind me? Interesting stuff.
5/ if 1-3 are even partially correct (I'm guessing) then the choice of op amp would probably be quite critical as you've noted.
Yes I think you are on the money Mondo.
Kinku, the opamp itself does not change the filter slope, but the characteristics as mondo says, do affect it's performance. There are actually only two components that affect the filter slope for a 2nd order slope. If you look at only the first FDNR above in the diagram they are the 3.09K resistor and the 3.4K resistor.
Think of the 3.09K as being the inductor, and the 3.4K as being the capacitor in a normal passive circuit. The rest are just components that make that work. If you change the values of the caps then the values of the resistors must change as well. If you changed to 100nF caps you would need to drop the resistors to 309 ohms and 340 ohms (too low). The 2K resistors you can make whatever you want. you could make them 5K 10K 20K (as long as they are the same) with no impact on the slope of the filter.
I don't actually fully understand the mechanism at work in the circuit, but my hunch is that the OLG of the opamp IS important (and I think why the LM4562/LME48960 is probably a good choice (it certainly sims much better than the opa2134). The other thing that I think is very important with this circuit is the phase margin of the opamp. The phase does not continue on down to 180 degrees phase shift but instead reverses and heads on up again. In the sims it only gets back to about 60 degrees before reversing again and heading down towards -180 but I think in practice it is behaving differently.
I originally planned to use the opa2604, but had more stability problems (on the breadboard) than I did with the opa2134, which is why that is what I have been using it. Sometime this weekend I want to put the 2604 in and measure and see what the result is. I think it is much more sensitive to layout that the 2134 so I'm interested to see how it performs in comparison on the board..
mondo I don't think I've posted this link in this thread before but have posted it elsewhere http://www.ti.com/lit/an/sbaa001/sbaa001.pdf probably time I re-read it and see if I understand more this time around
Tony.
Kinku, the opamp itself does not change the filter slope, but the characteristics as mondo says, do affect it's performance. There are actually only two components that affect the filter slope for a 2nd order slope. If you look at only the first FDNR above in the diagram they are the 3.09K resistor and the 3.4K resistor.
Think of the 3.09K as being the inductor, and the 3.4K as being the capacitor in a normal passive circuit. The rest are just components that make that work. If you change the values of the caps then the values of the resistors must change as well. If you changed to 100nF caps you would need to drop the resistors to 309 ohms and 340 ohms (too low). The 2K resistors you can make whatever you want. you could make them 5K 10K 20K (as long as they are the same) with no impact on the slope of the filter.
I don't actually fully understand the mechanism at work in the circuit, but my hunch is that the OLG of the opamp IS important (and I think why the LM4562/LME48960 is probably a good choice (it certainly sims much better than the opa2134). The other thing that I think is very important with this circuit is the phase margin of the opamp. The phase does not continue on down to 180 degrees phase shift but instead reverses and heads on up again. In the sims it only gets back to about 60 degrees before reversing again and heading down towards -180 but I think in practice it is behaving differently.
I originally planned to use the opa2604, but had more stability problems (on the breadboard) than I did with the opa2134, which is why that is what I have been using it. Sometime this weekend I want to put the 2604 in and measure and see what the result is. I think it is much more sensitive to layout that the 2134 so I'm interested to see how it performs in comparison on the board..
mondo I don't think I've posted this link in this thread before but have posted it elsewhere http://www.ti.com/lit/an/sbaa001/sbaa001.pdf probably time I re-read it and see if I understand more this time around
Tony.
Ah OK well in actual fact, in my particular implementation then I suspect the answer is yes. The two components that I will likely change are the damping resistors that I added. With the opa2134 they are 5 ohms (in series with the shut cap after the FDNR) and the extra 47 ohm resistor between the last cap and ground. For the LM4562 it looks (from the sims I have run) to be better with 2 ohms instead of 5, and 20 ohms instead of 47. The addition of these particular resistors is to damp a resonance that seems to occur, the LM4562 seems to be less susceptible to this resonance (again based on sims).
James, no I haven't done the high pass section yet (its half populated). BUT when I did it on breadboard it had none of the problems of the LP section. It was completely clean out to 48Khz from memory. I did have a listen when it was on breadboard (though I had a big noise and oscillation problem then), and despite the noise it sounded very good. Unfortunately I don't have any other active filter to compare against.
Basically what I was comparing was my MTM's running full range, to my MTM's running from 200Hz up and my Vifa M26WR-09-08's running from 200Hz down. Obviously bass improved, but what was most apparent was the improvement in midrange. However this is an apples and oranges comparison as I have no other active filter to compare against.
Tony.
James, no I haven't done the high pass section yet (its half populated). BUT when I did it on breadboard it had none of the problems of the LP section. It was completely clean out to 48Khz from memory. I did have a listen when it was on breadboard (though I had a big noise and oscillation problem then), and despite the noise it sounded very good. Unfortunately I don't have any other active filter to compare against.
Basically what I was comparing was my MTM's running full range, to my MTM's running from 200Hz up and my Vifa M26WR-09-08's running from 200Hz down. Obviously bass improved, but what was most apparent was the improvement in midrange. However this is an apples and oranges comparison as I have no other active filter to compare against.
Tony.
Kinku, I think I was getting the resonance in the real circuit on the breadboard. In the sim it shows up as a phase glitch and a peak in the FR. In the real world it seems similar. I've done some more tests tonight (with the opa2604) and it is performing very similarly to with the 2134. I haven't purchased the 4562's yet, so no tests.
I've done Distortion tests tonight. Though they are not particularly valid, I think they show promise. My focusrite 2i2 has very good distortion performance when running in balanced mode. However the performance drops markedly when outputting balanced into a single ended load.
1st pic, focusrite in balanced loopback 1Khz.
2nd pic focusrite in unbalanced loopback 50Hz (yes I forgot to do one at 1Khz :rolleyes)
3rd pic synergy at 50Hz (note only marginally higher distortion than the focusrite unbalanced loopback). around 0.003 % more.
4th pic synergy 100Hz distortion
5th pic synergy 200Hz distortion
6th pic synergy 1Khz distortion (note that it is getting worse as we go up, but I think this is basically because there is less difference between the fundamental and the 2nd harmonic (ie the 2nd harmonic stays relatively constant but the fundamental is attenuated a lot more).
7th pic is the fr curve as REW sees it. The crud between 10Khz and 50Khz is what concerns me, and there does appear to be an abhoration at 30Khz.... I used to get this in holm on the breadboard, but it is much cleaner on the PCB.
8th pic is holm impulse measurement. I clicked the reverberation button in the impulse response. I don't understand what it is showing me but I suspect it is a clue! 300ms seems a long way into the impulse though...
edit: added nith pic, which is rew measurement exported and imported into holm impulse. Interesting that it looks a whole lot better in holm impulse than it does in REW. I'm starting to wonder more and more about my measurements and whether it's simply time to build the proper ones and have a listen (I can here James cheering )
So the moral of this story is I need to get some of the LM opamps, and I also need to work on my little project for making a balanced to single ended converter so I can get decent performance out of my focusrite 2i2 when testing single ended equipment!
Tony.
I've done Distortion tests tonight. Though they are not particularly valid, I think they show promise. My focusrite 2i2 has very good distortion performance when running in balanced mode. However the performance drops markedly when outputting balanced into a single ended load.
1st pic, focusrite in balanced loopback 1Khz.
2nd pic focusrite in unbalanced loopback 50Hz (yes I forgot to do one at 1Khz :rolleyes)
3rd pic synergy at 50Hz (note only marginally higher distortion than the focusrite unbalanced loopback). around 0.003 % more.
4th pic synergy 100Hz distortion
5th pic synergy 200Hz distortion
6th pic synergy 1Khz distortion (note that it is getting worse as we go up, but I think this is basically because there is less difference between the fundamental and the 2nd harmonic (ie the 2nd harmonic stays relatively constant but the fundamental is attenuated a lot more).
7th pic is the fr curve as REW sees it. The crud between 10Khz and 50Khz is what concerns me, and there does appear to be an abhoration at 30Khz.... I used to get this in holm on the breadboard, but it is much cleaner on the PCB.
8th pic is holm impulse measurement. I clicked the reverberation button in the impulse response. I don't understand what it is showing me but I suspect it is a clue! 300ms seems a long way into the impulse though...
edit: added nith pic, which is rew measurement exported and imported into holm impulse. Interesting that it looks a whole lot better in holm impulse than it does in REW. I'm starting to wonder more and more about my measurements and whether it's simply time to build the proper ones and have a listen (I can here James cheering
) So the moral of this story is I need to get some of the LM opamps, and I also need to work on my little project for making a balanced to single ended converter so I can get decent performance out of my focusrite 2i2 when testing single ended equipment!
Tony.
Attachments
Tony, thank you do much. So it seems like Lm4562 is the way to go. Is there anyway you can make a schematic for the proposed filter with all the caps and resistors please,which will be easy for me to understand. I don't want to mess up the circuit since my measurement capabilities are limited. I am not sure about the shut caps last cap config in your post 115. Hope you can help.
Hi Kinku, here is just a snap of the low pass section, but I'll see what I can do as far as making a better one. I might even put in values that make sense for a post DAC filter (though I'm not 100% sure what the corner frequency should be... 20Khz or 40Khz).
You would almost certainly need to decrease the caps from 100nf to 10nf for the higher frequency
Also note that the pdf you attached earlier mentions about the cap positions, that is contrary to the way I have done it (and may be part of my problem). I arrived at this configuration after many tests on the breadboard, but again that may be something that is different for a high frequency circuit. It could also be that nasties that this helped with on the breadboard, may not be present with a proper layout.
Tony.
You would almost certainly need to decrease the caps from 100nf to 10nf for the higher frequency
Also note that the pdf you attached earlier mentions about the cap positions, that is contrary to the way I have done it (and may be part of my problem). I arrived at this configuration after many tests on the breadboard, but again that may be something that is different for a high frequency circuit. It could also be that nasties that this helped with on the breadboard, may not be present with a proper layout.
Tony.
Attachments
More than likely James!!
Kinku, not quite ready yet (I did this with 24Khz corner freq to get something going) will revise tomorrow, but looks promising. Interestingly at this higher frequency I needed to increase the value of one of the damping resistors dramatically, but the other one I eliminated completely.
There is some peaking in the filter response. 2nd curve is showing what happens with a 25 ohm resistor at R42 instead of the 1st curves 300 ohm. Blue curve is how the passive filter would theoretically perform.
Interestingly pretty much the same phenomenon occurs with the phase reversal and resonance regardless of the frequency.
Will upload the spice model once it is sorted. (but it is after midnight so it won't be till tomorrow (assuming I get the time). This is only a third order, but the article I linked to earlier basically said a post dac filter only needed to be 3rd order if it was an oversampling dac...
you will note I had to reduce the size of the caps to 1nF (from the 100nF I'm using at 200Hz) the very interesting thing is that the resonance frequency has gone up to around 800Khz I'm thinking I may need to look at reducing the value of the caps from 100nF down a bit (and scale the resistor appropriately). I'd gone with 100nF for a couple of reasons, they make calculations easy (I can basically cheat and not use the formulas) and it means the resistors are not too big (important for noise), but maybe if I scale them down to say 47nF or 33nF the resistors won't become 10 times larger, and the resonance frequency may well get pushed up higher meaning that we don't get the knee till a much higher frequency well past anything we are interested in for audio... I'll have to do some experimenting!
Tony.
Kinku, not quite ready yet (I did this with 24Khz corner freq to get something going) will revise tomorrow, but looks promising. Interestingly at this higher frequency I needed to increase the value of one of the damping resistors dramatically, but the other one I eliminated completely.
There is some peaking in the filter response. 2nd curve is showing what happens with a 25 ohm resistor at R42 instead of the 1st curves 300 ohm. Blue curve is how the passive filter would theoretically perform.
Interestingly pretty much the same phenomenon occurs with the phase reversal and resonance regardless of the frequency.
Will upload the spice model once it is sorted. (but it is after midnight so it won't be till tomorrow (assuming I get the time). This is only a third order, but the article I linked to earlier basically said a post dac filter only needed to be 3rd order if it was an oversampling dac...
you will note I had to reduce the size of the caps to 1nF (from the 100nF I'm using at 200Hz) the very interesting thing is that the resonance frequency has gone up to around 800Khz I'm thinking I may need to look at reducing the value of the caps from 100nF down a bit (and scale the resistor appropriately). I'd gone with 100nF for a couple of reasons, they make calculations easy (I can basically cheat and not use the formulas) and it means the resistors are not too big (important for noise), but maybe if I scale them down to say 47nF or 33nF the resistors won't become 10 times larger, and the resonance frequency may well get pushed up higher meaning that we don't get the knee till a much higher frequency well past anything we are interested in for audio... I'll have to do some experimenting!
Tony.
Attachments
Last edited:
Great work Tony. Perhaps I can stack two 3rd orders back to back and achieve a better attenuation. I saw that with one of the BB papers about FDNR filters
http://www.ti.com/lit/an/sbaa001/sbaa001.pdf
In that paper authors say they had better phase response when they stacked two third order ones over making 6 th order filter.take your time, seems like I am getting a great help here.
http://www.ti.com/lit/an/sbaa001/sbaa001.pdf
In that paper authors say they had better phase response when they stacked two third order ones over making 6 th order filter.take your time, seems like I am getting a great help here.
Hi kinku, I tried cascading two (equal) 3rd order filters but this does not appear to work properly (rolloff starts to soon). Note that I have relied on online cacluators for doing my filter calculations, and these top out at 4th order so I'm not sure what is required to do higher order circuits.
However, I thought about your comment on what level of filtering you require and I do not believe it is feasible. You are effectively asking for a brickwall filter. The slope of which would need to be enourmous!
To be down 60db at 21.1Khz (but not starting until 18Khz you will need a phenominally steep filter. a 6th order filter is only 36db / octave so the corner frequency would need to be around 8Khz (note that is a guess) to be down 60db at 21Khz. I think you have miscalculated somewhere
I'm not at all familiar with DAC filters, what I might do is model the one from the TI app note. Or alternatively if you can provide a passive say 6th order (or even 9th order) filter I can translate that to an FDNR circuit and model it.
Tony.
However, I thought about your comment on what level of filtering you require and I do not believe it is feasible. You are effectively asking for a brickwall filter. The slope of which would need to be enourmous!
To be down 60db at 21.1Khz (but not starting until 18Khz you will need a phenominally steep filter. a 6th order filter is only 36db / octave so the corner frequency would need to be around 8Khz (note that is a guess) to be down 60db at 21Khz. I think you have miscalculated somewhere
I'm not at all familiar with DAC filters, what I might do is model the one from the TI app note. Or alternatively if you can provide a passive say 6th order (or even 9th order) filter I can translate that to an FDNR circuit and model it.
Tony.
Tony,Thanks for your time. So ant imaging filter remove images of processing after DAC. The first image of DAC for CD audio which samples at 44.1KHz starts at approximately half of sampling frequency and then at 2x44.1 etc. So the easy way is oversample which can increase the first image appearing at distant frequency depending on the nth order of over sampling. Most CD players use 4x oversampling when they started digital age. Which will in turn shift the first image to 4x44.1-20Khz=above 150Khz ,assuming the highest audi frequency in CD media is 20Khz.
I am planning to go the NOS way since there is a theory that NOS sound better.
Analog Devices: Interactive Design Tools: Digital-to-Analog Converters :
To eliminate the first image I have to steeply cut at 20KHz and by 22KHz it should be down by 60-80dB. So the -60dB (being modest with what can be achieved) is the 1/2 of sampling frequency.
So what are the options to do this with minimum phase distortion?
I think the TI application note uses a filter between Bessel and Butterworth "
They at the end reports that a 6th order without cascading with the OPA2604 was difficult
with extremely high GBW would be required. ...........
Just curious did you use a buffer between the two 3rd order sections?
So in short I am into the NOS and unfortunately I am in the catch 22 the pioneers in design field was in. Either allow the first image to appear in my audio chain and allow my ear to filter out or make a steep filter to cut out that frequency spending extra bucks. I like to preserve the phase as much as possible.I saw a brickwall filter online "Brick-wall" lowpass audio filter needs no tuning | EDN
was thinking perhaps make one for 20KHz instead of the 15Khz for FM. It is using a FDNR topology I guess.
I was thinking why not use the LME49720 in that circuit.
I am planning to go the NOS way since there is a theory that NOS sound better.
Analog Devices: Interactive Design Tools: Digital-to-Analog Converters :
To eliminate the first image I have to steeply cut at 20KHz and by 22KHz it should be down by 60-80dB. So the -60dB (being modest with what can be achieved) is the 1/2 of sampling frequency.
So what are the options to do this with minimum phase distortion?
I think the TI application note uses a filter between Bessel and Butterworth "
Thus, we begin the design process by selecting a passive,
third order linear-phase filter design that will be realized
using this active approach. The passive design shown in
Figure 1 is neither a Butterworth nor a Bessel response; it is
something in between. The component values for this particular
response, optimized for phase linearity and stopband
attenuation, were found through exhaustive computer simulations
and empirical analysis."They at the end reports that a 6th order without cascading with the OPA2604 was difficult
"To make a sixth order filter, you can repeat the design
process above from a passive realization and directly implement
a filter. This implementation is very sensitive to the
gain-bandwidth product (GBW) match of all of the op amps
used, however; for a 40kHz cutoff frequency, an op amp with extremely high GBW would be required. ...........
A simpler solution is to cascade two of the third order
sections designed above. This cascaded design (Figure 10)
works equally well for most applications."Just curious did you use a buffer between the two 3rd order sections?
So in short I am into the NOS and unfortunately I am in the catch 22 the pioneers in design field was in. Either allow the first image to appear in my audio chain and allow my ear to filter out or make a steep filter to cut out that frequency spending extra bucks. I like to preserve the phase as much as possible.I saw a brickwall filter online "Brick-wall" lowpass audio filter needs no tuning | EDN
was thinking perhaps make one for 20KHz instead of the 15Khz for FM. It is using a FDNR topology I guess.
I was thinking why not use the LME49720 in that circuit.
Last edited:
Tony, do not get discouraged with the DAC part. Once the signal is out of DAC it is back in analog domain. The filtering in the post DAC circuit is completely analog. I am still hopeful that you can help me with your expertise in FDNR filter and OPAMP . That is why I want to avoid discussing this in digital line level.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.