More studies are being done. I am learning more about SMPS. I am not rushing with this as I know this is a learning process and it will take some time.
Although I did not see a snubber on the primary side, it actaully exists already in the original schematic and board design, except that it was not implemented. It is the standard RCD snubber for most of the flyback inverters but in Marantz implementation the R is not installed. Why is it so?
I am also reviewing the idea of snubbing the secondary sides. There have been such recommendations in audio forums, but I have not seen any of such recommendations in manufacturers' datasheets or implementations. I think they will cause no harm in audio (except with a slight degrade of efficiency perhaps, against which SMPS are designed for) and will possibly do good. But I can now see that the ringings I saw from my measurements mainly come from the primary switching interacting with the transformer leakage inductance and the MOSFET capacitance.
Indeed, the wave forms I saw are probably not much different from those in most SMPS papers I have read. The Marantz SMPS has one more resonance which is not in any others.
I now also believe that there is no way I can ultimately get rid of most of the SMPS switching junks, given the contraints that I am working on an existing player. I can only reduce the noise to some extent.
Although I did not see a snubber on the primary side, it actaully exists already in the original schematic and board design, except that it was not implemented. It is the standard RCD snubber for most of the flyback inverters but in Marantz implementation the R is not installed. Why is it so?
I am also reviewing the idea of snubbing the secondary sides. There have been such recommendations in audio forums, but I have not seen any of such recommendations in manufacturers' datasheets or implementations. I think they will cause no harm in audio (except with a slight degrade of efficiency perhaps, against which SMPS are designed for) and will possibly do good. But I can now see that the ringings I saw from my measurements mainly come from the primary switching interacting with the transformer leakage inductance and the MOSFET capacitance.
Indeed, the wave forms I saw are probably not much different from those in most SMPS papers I have read. The Marantz SMPS has one more resonance which is not in any others.
I now also believe that there is no way I can ultimately get rid of most of the SMPS switching junks, given the contraints that I am working on an existing player. I can only reduce the noise to some extent.
Just for a short update.
It seems that a cheap LCR metre can give quite accurate results. I have been able to match the inductance (measured with LCR metre), capacitance (based on datasheet of MOSFETs, diodes and capacitors) and resonance frequencies (observed from 100MHz digital scope using probe-in-the-air) and they all match with known formulas. This surprises me because inductance values may vary wildly with frequencies.
So now I have no problem with fitting in suitable secondary snubbers.
It is still not easy to calculate the primary side snubbers, since I have not measured the high voltage points and the currents.
Currents are difficult to prodict because it may draw little current when the machine is idle and a lot of current when the disc is spinning. There are no test points on the board for me to clip my probes onto, and it is not easy to install my own. The trouble is, the board must be mounted on the chassis with all rails connected or the controller would shut the thing down. I could not turn the board up-side-down to do measurements.
Since the primary side voltages can swing form wildly within hundreds of volts the power rating of the damping resistor can be very high, to an extent that it may be impossible to fit them in (5W plus is possible).
More analysis is needed.
It seems that a cheap LCR metre can give quite accurate results. I have been able to match the inductance (measured with LCR metre), capacitance (based on datasheet of MOSFETs, diodes and capacitors) and resonance frequencies (observed from 100MHz digital scope using probe-in-the-air) and they all match with known formulas. This surprises me because inductance values may vary wildly with frequencies.
So now I have no problem with fitting in suitable secondary snubbers.
It is still not easy to calculate the primary side snubbers, since I have not measured the high voltage points and the currents.
Currents are difficult to prodict because it may draw little current when the machine is idle and a lot of current when the disc is spinning. There are no test points on the board for me to clip my probes onto, and it is not easy to install my own. The trouble is, the board must be mounted on the chassis with all rails connected or the controller would shut the thing down. I could not turn the board up-side-down to do measurements.
Since the primary side voltages can swing form wildly within hundreds of volts the power rating of the damping resistor can be very high, to an extent that it may be impossible to fit them in (5W plus is possible).
More analysis is needed.
Last edited:
I am pretty sure I have now worked out one snubber R and C on the primary side. I used multiple formulas given by various manufacturers' datasheets / application notes and they all ended up with some fairly close numbers.
I worked out R = 820R and C=100pF. This snubber is to shunt the MOSFET switch from the MOSFET gate to source. The MOSFET capacitance and the leakage inductance of the transformer primary form a LC circuit and causes severe ringing at 3.5MHz with an amplitude as high as 150V maximum at the inital switch-off of the switching MOSFET which gradually dies down after 1/4 of a duty cycle at the switching frequency of 100kHz.
However, I am having problem calculating the power dissipation of R.
When the switch turns open, the MOSFET gate voltage is raised to about 350VDC plus 150VAC at 3.5MHz (ringing frequency) on the top (worst case). The MOSFET source is at ground potential.
In other words, 350VDC + 150VAC at 3.5MHz (worst case) is across 820R and 220pF.
So how much power the resistor R dissipates?
I am not sure if I could calculate it this way. The snubber does not pass DC so the 350VDC can be ignored. The RC forms a corner frequency just below 1MHz so the impedance at 3.5MHz is 820R roughly. The power dissipation is only for the AC component which is 106 (RMS) * 106 (RMS) / 820 = 13.7W.
If the above is right I would be in trouble, as I could never find the space to fit in a 13.7W resistor.
But since the ripple frequency amplitude gradually dies down after about 1/4 of a duty cycle, the actual power dissipation would be less than that. But I don't know how to calculate it accurately. I also assumed that after the snubber is in place the ringing may disappear!
The best I could do would be to mount two 2W resistor in series with the C.
I don't know if I would risk catching fire...
Another problem is that 2 x 2W resistors and one capacitor would have about 25nH inductance. Would it still be able to work at 3.5MHz?
I worked out R = 820R and C=100pF. This snubber is to shunt the MOSFET switch from the MOSFET gate to source. The MOSFET capacitance and the leakage inductance of the transformer primary form a LC circuit and causes severe ringing at 3.5MHz with an amplitude as high as 150V maximum at the inital switch-off of the switching MOSFET which gradually dies down after 1/4 of a duty cycle at the switching frequency of 100kHz.
However, I am having problem calculating the power dissipation of R.
When the switch turns open, the MOSFET gate voltage is raised to about 350VDC plus 150VAC at 3.5MHz (ringing frequency) on the top (worst case). The MOSFET source is at ground potential.
In other words, 350VDC + 150VAC at 3.5MHz (worst case) is across 820R and 220pF.
So how much power the resistor R dissipates?
I am not sure if I could calculate it this way. The snubber does not pass DC so the 350VDC can be ignored. The RC forms a corner frequency just below 1MHz so the impedance at 3.5MHz is 820R roughly. The power dissipation is only for the AC component which is 106 (RMS) * 106 (RMS) / 820 = 13.7W.
If the above is right I would be in trouble, as I could never find the space to fit in a 13.7W resistor.
But since the ripple frequency amplitude gradually dies down after about 1/4 of a duty cycle, the actual power dissipation would be less than that. But I don't know how to calculate it accurately. I also assumed that after the snubber is in place the ringing may disappear!
The best I could do would be to mount two 2W resistor in series with the C.
I don't know if I would risk catching fire...
Another problem is that 2 x 2W resistors and one capacitor would have about 25nH inductance. Would it still be able to work at 3.5MHz?
Last edited:
Cool!
So divide the 13.7 Watts by a little less than 4, since it dies out gradually after 1/4 of a cycle. You still want to leave about a 100% margin, i.e. use only about half of the resistor's rated dissipation, at least for the "production" version.
Also, you could lower the series capacitance, so less energy gets through to the resistor. It would work less and less well, as you lowered the C value. But if it works "well-enough", then it works well-enough!
Also, you can test your calculations, ahead of time, if it would be possible to touch the leads of a cap that is 3X the parasitic C that you believe is present, across the two points where the snubber will be connected, while watching a scope or frequency counter. If the frequency of the ringing is cut in half, then you had the correct cap value figured out.
So divide the 13.7 Watts by a little less than 4, since it dies out gradually after 1/4 of a cycle. You still want to leave about a 100% margin, i.e. use only about half of the resistor's rated dissipation, at least for the "production" version.
Also, you could lower the series capacitance, so less energy gets through to the resistor. It would work less and less well, as you lowered the C value. But if it works "well-enough", then it works well-enough!
Also, you can test your calculations, ahead of time, if it would be possible to touch the leads of a cap that is 3X the parasitic C that you believe is present, across the two points where the snubber will be connected, while watching a scope or frequency counter. If the frequency of the ringing is cut in half, then you had the correct cap value figured out.
It looks like it will get close to the margin with the resistor rating.
I found some 3W metal film capacitors from Farnell's catelogue that may fit and I can put 2 in series. But 6W may still be low.
The question is, if these resistors are destroyed by over current / heat, will they be open or short? If open, there is no issue. If short, it may cause other components to fail, and the result can be unpredictable.
I found some 3W metal film capacitors from Farnell's catelogue that may fit and I can put 2 in series. But 6W may still be low.
The question is, if these resistors are destroyed by over current / heat, will they be open or short? If open, there is no issue. If short, it may cause other components to fail, and the result can be unpredictable.
I spent 10 minutes searching. One person suggested that when burnt out metal film resistors will short while metal oxide resistors will open. So I will get 3W metal oxide resistors instead. Again the parts will come from Farnell UK so it is going to take a week.
There are so many things involved in improving this SMPS which make the process tedious and fun at the same time.
I am sure that the 5V secondary winding is the one that creates the big 8MHz resonance. The big TO220 Schottky diode MBRF1060 in the 5V is the only diode I have not upgraded. I am really tempted to upgrade it with MBR40250 ultrafast soft recovery diode (trr=35nS) so that the resonances can be reduced. Although the 5V is not for analogue audio, at such high frequencies the resonant noise will pass through the transformer into the 12V analogue output with little resistance.
But the issue is that MBR40250 will incur an extra 0.15V voltage drop. I have measured the 5V output and the voltage is 5.13V. If I changed the diode to MBR40250 it will possibly be 4.98V. Would you do it?
There are so many things involved in improving this SMPS which make the process tedious and fun at the same time.
I am sure that the 5V secondary winding is the one that creates the big 8MHz resonance. The big TO220 Schottky diode MBRF1060 in the 5V is the only diode I have not upgraded. I am really tempted to upgrade it with MBR40250 ultrafast soft recovery diode (trr=35nS) so that the resonances can be reduced. Although the 5V is not for analogue audio, at such high frequencies the resonant noise will pass through the transformer into the 12V analogue output with little resistance.
But the issue is that MBR40250 will incur an extra 0.15V voltage drop. I have measured the 5V output and the voltage is 5.13V. If I changed the diode to MBR40250 it will possibly be 4.98V. Would you do it?
I am just wondering if it makes any difference in the snubber performance when the two resistors and a single capacitor are connected as CRR, RCR or RRC. RCR saves space, has the shortest path and looks the best.
??????
??????
The MOSFET capacitance looks like to have around 50pF based on the calculation from the measured resonance frequency as well as the leakage inductance, and that is fairly close to the MOSFET datasheet.
Using the formula of C = 2 * pi * sqrt(L * C) / sqrt(L / C) for a Q=0.5 the calculated C is 330pF.
I chose 100pF, so we are in the same pace.
Now pinging abraxalito...
I have read a lot of your recommendations of using LM6172 for I/V. I am happy to take your words and give it a try. The Marantz UD7007 uses op275 for I/V at the output of the DAC PCM1795.
How would you bypass the LM6172? For the op275, Marantz had 100uF electrolytic, as well as 18nF polyester, which should work fine. I upgraded them into 470uF/16V and 0.1uF MKP + 0.5R. Now for the 100MHz device of LM6172, the MKP would have too high inductance and it needs to be changed to a ceramic X7R, and perhaps 0.01uF would do. But this would create some noise within the LM6172 bandwidth unless 1R is in series, but then the 1R would probably reduce the effectiveness of the bypass due to the resistance as well as the increased inductance. So the best compromise seems to be 0.01uF X7R directly from the supply pins to ground. I am not sure if a 0.1uF X7R across the supply pins would work better.
What is your experience?
I have read a lot of your recommendations of using LM6172 for I/V. I am happy to take your words and give it a try. The Marantz UD7007 uses op275 for I/V at the output of the DAC PCM1795.
How would you bypass the LM6172? For the op275, Marantz had 100uF electrolytic, as well as 18nF polyester, which should work fine. I upgraded them into 470uF/16V and 0.1uF MKP + 0.5R. Now for the 100MHz device of LM6172, the MKP would have too high inductance and it needs to be changed to a ceramic X7R, and perhaps 0.01uF would do. But this would create some noise within the LM6172 bandwidth unless 1R is in series, but then the 1R would probably reduce the effectiveness of the bypass due to the resistance as well as the increased inductance. So the best compromise seems to be 0.01uF X7R directly from the supply pins to ground. I am not sure if a 0.1uF X7R across the supply pins would work better.
What is your experience?
My bypassing changes according to my experience when I was using LM6172 I didn't understand what I do now about power supplies. Now I use plentiful small ceramics between the rails (I don't use dual supplies much these days) and always some series inductance before these 'see' the electrolytics. Normally the series L is provided by one or two ferrite beads (typically 1uH each).
I'd not bother with such low values of ceramic as 0.01uF - they might just be suitable for extremely fast logic but do sweet FA anywhere near audio. Get yourself some > 1uF ceramics. I reckon the best bang for the buck right now are 0603 4.7uF 10V X5R (does rather limit your rail voltage options, but I find no need to go beyond 10V these days). On full bias they do lose at least 75% of their value so treat them as 1uFs if you're using 10V. Then parallel them up to get as much capacitance as you'll need (normally I go for 10uF).
The next stage of supply will use low ESR lytics (currently I like Rubycon ZLHs) and preferably in the 1000s of uF. If you're feeding these from a reg, a series inductor between the reg and the lytics helps a lot in limiting the output noise.
I'd not bother with such low values of ceramic as 0.01uF - they might just be suitable for extremely fast logic but do sweet FA anywhere near audio. Get yourself some > 1uF ceramics. I reckon the best bang for the buck right now are 0603 4.7uF 10V X5R (does rather limit your rail voltage options, but I find no need to go beyond 10V these days). On full bias they do lose at least 75% of their value so treat them as 1uFs if you're using 10V. Then parallel them up to get as much capacitance as you'll need (normally I go for 10uF).
The next stage of supply will use low ESR lytics (currently I like Rubycon ZLHs) and preferably in the 1000s of uF. If you're feeding these from a reg, a series inductor between the reg and the lytics helps a lot in limiting the output noise.
Thanks. I have now formed my ideas on how to do PSU and I am pretty confident that they will work well. However, I am unable to use them in an existing player in which the PCB is fixed and there are only a limited things I can change.
The 470uF/16V I selected would probably work well with most regulators comparing to the normal 100uF, but going higher would put it into an unknown territory.
I simply don't have experience with a 100MHz part. It seems now that the best is to solder a large X7R or X5R cap directly across the supply bins, instead of from individual supply pins to ground, because such low impedance high Q caps can cause execessive resonance with the power supply regulator. I am not too sure if a cap across the supply pins would also cause resonance. I tried that a while ago on a discrete shunt reg and it resonated like hell.
The 470uF/16V I selected would probably work well with most regulators comparing to the normal 100uF, but going higher would put it into an unknown territory.
I simply don't have experience with a 100MHz part. It seems now that the best is to solder a large X7R or X5R cap directly across the supply bins, instead of from individual supply pins to ground, because such low impedance high Q caps can cause execessive resonance with the power supply regulator. I am not too sure if a cap across the supply pins would also cause resonance. I tried that a while ago on a discrete shunt reg and it resonated like hell.
Having ceramics directly in parallel with lytics definitely can cause rather large resonances - hence my use of series inductors which looks to tame these (in sim anyway). If you're shy to try the LM6172 then have a look at the LT1355 which is of a similar architecture but only 12MHz GBW.
I have found these 3.3uF/10V X5R that is rather small to fit. I could solder one across the power supply pins, I suppose.
SA10ZD335KAR - AVX - CAPACITOR CERAMIC 3.3UF, 10V, | element14 Australia
SA10ZD335KAR - AVX - CAPACITOR CERAMIC 3.3UF, 10V, | element14 Australia
I thought that you suggested that for I/V one needs high bandwidth and that was the reason LM6172 instead of LT1355 was selected. Is that view changed?
In simulations once the film / ceramic capacitors have values above 3.3uF then resonance does not seem to be a problem. I found some resistance is needed to lower the filter Q if inductors are used, which can be applied to new circuits but would get very clumpsy for a mod job.
Yes, it's combinations of values that can get you into trouble, correct simulation will demonstrate whether there's an issue or not - I've mentioned a few rules of thumb in some of my posts ...
For I/V I prefer to go passive nowadays - if you prefer an opamp then the faster the better because greater GBW minimizes the overloading of the input stage which occurs each and every sample from a DAC. I suggested the LT1355 only because you seemed to be running a little scared from a 100MHz part. The LT part should do better than any LTP-fronted opamp I reckon, irrespective of GBW. With the possible exception of ADI's multi-tanh designs....
<edit> I don't recommend any leaded part for decoupling duty.
<edit> I don't recommend any leaded part for decoupling duty.
I know it is very hard. Unless I can isolate the 117/317 regulators, putting a 3.3uF (or 6.8uF) X7R from the supply pins to ground can still cause big trouble.
In the original Marantz player, the 9MHz OP275 has heavy compensation, with a 1nF in parallel with 820R, which has a -3dB corner frequency at 200kHz.
If I change it to the 100MHz LM6172, I think I will reduce the 1nF to 470pF to reduce the phase shift at 20kHz. In that case, the corner frequency will be shifted to close to 400kHz instead. When it reaches 10MHz, the attenuation is 30dB already, and I think that is a plenty.
Given such a heavy compensation, I am now not sure if 0.1uF bypass ceramic caps are necessary. Perhaps my existing implementation of 0.1uF MKP + 1R could be sufficient? Let me know if you think it is not.
P.S. order was placed and parts will arrive in 5 days time from the U.K.
If I change it to the 100MHz LM6172, I think I will reduce the 1nF to 470pF to reduce the phase shift at 20kHz. In that case, the corner frequency will be shifted to close to 400kHz instead. When it reaches 10MHz, the attenuation is 30dB already, and I think that is a plenty.
Given such a heavy compensation, I am now not sure if 0.1uF bypass ceramic caps are necessary. Perhaps my existing implementation of 0.1uF MKP + 1R could be sufficient? Let me know if you think it is not.
P.S. order was placed and parts will arrive in 5 days time from the U.K.
There is something I have overlooked!
Only in the last minute, I found a tiny SMD 0.01uF/50V/Y5V cap from the positive supply pin of OP275 to ground.
However, there is no such bypass cap for the negative supply.
This is interesting. Since the OP275 has much higher PSSR on the positive rail (97dB at 20kHz) than the negative rail (60dB at 20kHz), one would have thought the additional 0.01uF bypass would be on the negative rail. It is the opposite.
The LM6172 is a complementary amplifier so I am wondering what is going to happen if it fits into the circuit designed for the OP275.
Any comments?
Only in the last minute, I found a tiny SMD 0.01uF/50V/Y5V cap from the positive supply pin of OP275 to ground.
However, there is no such bypass cap for the negative supply.
This is interesting. Since the OP275 has much higher PSSR on the positive rail (97dB at 20kHz) than the negative rail (60dB at 20kHz), one would have thought the additional 0.01uF bypass would be on the negative rail. It is the opposite.
The LM6172 is a complementary amplifier so I am wondering what is going to happen if it fits into the circuit designed for the OP275.
Any comments?
- Status
- Not open for further replies.