Ooops, sure ! You're right !
Just to stay in dielectric types, I didn't find any info on the Panasonic mysterious "F" type... The ECJ5 (5 refering to the package size) is not even referenced on Panasonic's site afaik. But ECJ datasheet seem to make a curious parallel between Y5V and 'F'...
It's late and I'm not sure to get it
What I understand is that you say that the ceramic has to be located closer to the supply pin than the 'lytic, which I totally agree. But the way you say it may induce a little confusion. You seem to say that propagative phenomena occur, which is not the case at these frequencies and with the traces lengthes we're talking about. Elements are not distributed, and the classical Kirchoff (not sure of the spelling) law still applies. The ceramic caps (small and high value) have to be located as close as possible to the supply pin to avoid degradation of their low inductance by additional traces lengthes. To the contrary, some trace to the 'lytic won't degrade its poor ESL. But may be I'm just tired. I'd rather go to sleep now, hoping not to dream of caps Well, I guess I didn't express myself very clearly indeed. For 1-2 cm lengths, the voltage will be roughly the same all over the trace (except for minute variations due to HF current times the trace resistance per unit length) and neither the trace nor the cap will be a very efficient antenna.
What I was trying to say was that the current should be confined to as small a loop area as possible. Long leads carrying a HF current may inductively couple into other leads. But this is a near field effect which should not be confused with an antenna that radiates an electromagnetic wave.
Thanks for pointing out the sloppy terminology to me, Dr. T!
Eric
What I was trying to say was that the current should be confined to as small a loop area as possible. Long leads carrying a HF current may inductively couple into other leads. But this is a near field effect which should not be confused with an antenna that radiates an electromagnetic wave.
Thanks for pointing out the sloppy terminology to me, Dr. T!
Eric
Back to the workbench - Part 1
In order to get rid of this stuff, I've made some additional measurements that might interest you.
I started from the last configuration I studied : 100uF/ZA/Rubycon // 33nF/X7R/0612 // 220nF/X7R/0805, this latter still being soldered on solderside. Yes, no Murata here, since it damps everything and particular effect are not easily seen.Then I stacked a 1nF/NPO/0603 on the top of the 0612 cap, just to see...
Curves below show the results without (black) and with (red) this little cap. Spectacular, uh ? As awaited, it doesn't make even a little difference. I double checked, double measured, just to find the same results. May be the configuration I tested is not optimal to reveal the effects of this cap, and I should have tried leaving the 'lytic and the NPO cap alone. But I didn't find it worth the time.
In order to get rid of this stuff, I've made some additional measurements that might interest you
.I started from the last configuration I studied : 100uF/ZA/Rubycon // 33nF/X7R/0612 // 220nF/X7R/0805, this latter still being soldered on solderside. Yes, no Murata here, since it damps everything and particular effect are not easily seen.Then I stacked a 1nF/NPO/0603 on the top of the 0612 cap, just to see...
Curves below show the results without (black) and with (red) this little cap. Spectacular, uh ? As awaited, it doesn't make even a little difference
. I double checked, double measured, just to find the same results. May be the configuration I tested is not optimal to reveal the effects of this cap, and I should have tried leaving the 'lytic and the NPO cap alone. But I didn't find it worth the time.Attachments
Back to the workbench - Part 2
Well, what about the electrolytic itself ?
I restarted from the following config. : Rubycon 100uF/ZA // 100nF/X7R/0612 // 220nF/X7R/0805 // Murata 22uF/1210. Then I removed the Rubycon cap. The top graph of attached pic (black with ZA, red without) shows that the 'lytic doesn't seem to be necessary, a least in this application. Tiroth, you were right asking. Apparently, the Murata cap does all the necessary job. Too good to be true
Just for completeness, I soldered a 120uF/25V/FC from Panasonic, and compared it with the ZA, all the other caps being the same. Little surprise here : the Panasonic, with all his surrounding friends, seems to do a better job than the Rubycon Bottom graph shows the ZA in black compared with the FC in red. I zoomed the curves between 0 and 500MHz, since nothing happens above.
Well, good news indeed. We can use a cheap FC, or even remove it ! However, I'd rather keep it : I've no proof that the configuration is not sensitive to the application. The measurements are made for a particular app, with somewhat high frequencies (nothing below 195 kHz), and as Dr. Capslock stated a little above, there may be some cases where the 'lytic is needed.
I still have one measurement to do : replacing the costly and rapidly derating 22uF/10V/1210 ceramic cap with a 4.7uF/16V/1206, which seem to be less derating at 5V, and, most of all, cheaper. But I keep it for the next week
Hope this helps a little.
Well, what about the electrolytic itself ?
I restarted from the following config. : Rubycon 100uF/ZA // 100nF/X7R/0612 // 220nF/X7R/0805 // Murata 22uF/1210. Then I removed the Rubycon cap. The top graph of attached pic (black with ZA, red without) shows that the 'lytic doesn't seem to be necessary, a least in this application. Tiroth, you were right asking
. Apparently, the Murata cap does all the necessary job. Too good to be true Just for completeness, I soldered a 120uF/25V/FC from Panasonic, and compared it with the ZA, all the other caps being the same. Little surprise here : the Panasonic, with all his surrounding friends, seems to do a better job than the Rubycon
Bottom graph shows the ZA in black compared with the FC in red. I zoomed the curves between 0 and 500MHz, since nothing happens above.Well, good news indeed. We can use a cheap FC, or even remove it ! However, I'd rather keep it : I've no proof that the configuration is not sensitive to the application. The measurements are made for a particular app, with somewhat high frequencies (nothing below 195 kHz), and as Dr. Capslock stated a little above, there may be some cases where the 'lytic is needed.
I still have one measurement to do : replacing the costly and rapidly derating 22uF/10V/1210 ceramic cap with a 4.7uF/16V/1206, which seem to be less derating at 5V, and, most of all, cheaper. But I keep it for the next week
Hope this helps a little.
Attachments
Well, beats me! The FC has only 20% higher nominal capacitance. On the other hand, ESR is 0.25R at 100 kHz compared to 0.078R for the ZA, also at 100 kHz.
So what is this? ESL? Geometry? Or does the poorer electrode material and higher resistance keep the HF current from entering the cap, hence keeping it from becoming an "antenna"?
Eric
So what is this? ESL? Geometry? Or does the poorer electrode material and higher resistance keep the HF current from entering the cap, hence keeping it from becoming an "antenna"?
Eric
Don't really know But I do not measure anything at 100kHz. My measurements start around 1MHz, and the spectrum analyzer I used has only 700 measurement points. So who knows how these caps behave at frequencies > 50MHz, and what the given ESL data would be here... May be I'm missing something in the LF band, but I'm too lazy to restart my tests focusing in the LF region. Does someone want to try with a LF analyzer ?
But I do not measure anything at 100kHz. My measurements start around 1MHz, and the spectrum analyzer I used has only 700 measurement points. So who knows how these caps behave at frequencies > 50MHz, and what the given ESL data would be here... May be I'm missing something in the LF band, but I'm too lazy to restart my tests focusing in the LF region. Does someone want to try with a LF analyzer ?Vindication!...maybe. Like you said, the bulk cap can't hurt, and might have some effect <1MHz.
I'll be interested to see what effect the 4.7uF cap has compared to the 22uF. If it is comparable, I might consider going with a tantalum for the 1210 cap rather than a $$ ceramic, since I'm already stacking 3.3uF on the 0805 pad.
I'll be interested to see what effect the 4.7uF cap has compared to the 22uF. If it is comparable, I might consider going with a tantalum for the 1210 cap rather than a $$ ceramic, since I'm already stacking 3.3uF on the 0805 pad.
LF properties...
Tomorrow is another day, says the poet. I wanted to take a little rest without caps, but the brand new day found me full of energy, and as I'm almost as curious as you are, and as I had some time to lose, I switched on my favourite spectrum analyzer again.
I narrowed the bandwidth down to 60 MHz, and looked at what was going on. First, having a 120uF/FC // 100nF/0612 // 220nF/0805 // 22uF Murata soldered on the PCB, I measured them and removed the FC cap. Results are on the top graph of attached pic : with the FC (red) and without it (black). Given the amplitude levels (all in dBV), shall we say that the 'lytic has to be there ? I'll leave it to you
Middle graph compares once more the Panasonic/120uF/FC (red) with the Rubycon/100uF/ZA (black), the other caps still being the same. As I have previously noticed, the Panasonic does a nice job at minimum cost
Lastly, keeping the Rubycon/ZA, the 100nF/0612 and the 220nF/0805, I removed the 22uF Murata. Bottom graph shows the results, with (black) and without (red). No comments This little cap won't end amazing me...
Tomorrow is another day, says the poet. I wanted to take a little rest without caps, but the brand new day found me full of energy, and as I'm almost as curious as you are, and as I had some time to lose, I switched on my favourite spectrum analyzer again.
I narrowed the bandwidth down to 60 MHz, and looked at what was going on. First, having a 120uF/FC // 100nF/0612 // 220nF/0805 // 22uF Murata soldered on the PCB, I measured them and removed the FC cap. Results are on the top graph of attached pic : with the FC (red) and without it (black). Given the amplitude levels (all in dBV), shall we say that the 'lytic has to be there ? I'll leave it to you
Middle graph compares once more the Panasonic/120uF/FC (red) with the Rubycon/100uF/ZA (black), the other caps still being the same. As I have previously noticed, the Panasonic does a nice job at minimum cost
Lastly, keeping the Rubycon/ZA, the 100nF/0612 and the 220nF/0805, I removed the 22uF Murata. Bottom graph shows the results, with (black) and without (red). No comments
This little cap won't end amazing me...
What about saving a few bucks ?
In the meantime (mealtime more exactly), I had received some Murata 4.7uF/16V/1206/Y5V. I couldn't resist to test'em Results as usual on the attached pic. Top graph is the LF comparison of the 22uF/10V/1210 (black) with the newcomer (red). Logically, the 22uF is better than the 4.7uF, but only by a small amount. Widening the frequency range (< 500MHz, but absolutely no difference above), bottom graph shows the same behavior.
Well, I might say that if cost is not an issue, you could go with the 22uF But is the global result worth the price ? Personally, I will use the cheaper 4.7uF without any mental inhibition.
Just to sum up, the absolute cost/quality winner of my contest is:
Panasonic 120uF/25V/FC // 100nF/X7R/0612 // 220nF/X7R/0805 // Murata 4.7uF/16V/Y5V/1206
And thanks to all the people who have made the clever suggestions to improve quality and reduce costs
What I'm looking for now is some little feedback from you, Gentlemen. I'm not (by far) finished with my future DAC design, and if someone could test and report audio results before I can actually post mine, that would be great...
Last thing, just to relaunch the debate : What about the piezo-electric properties of Y5V compared to X7R ? Ceramic layers are so thin in these caps that vibrations may induce some unwanted noise. So we will have to damp them. Any suggestion for a damping material have nice HF dielectric properties ?
In the meantime (mealtime more exactly), I had received some Murata 4.7uF/16V/1206/Y5V. I couldn't resist to test'em
Results as usual on the attached pic. Top graph is the LF comparison of the 22uF/10V/1210 (black) with the newcomer (red). Logically, the 22uF is better than the 4.7uF, but only by a small amount. Widening the frequency range (< 500MHz, but absolutely no difference above), bottom graph shows the same behavior. Well, I might say that if cost is not an issue, you could go with the 22uF
But is the global result worth the price ? Personally, I will use the cheaper 4.7uF without any mental inhibition.Just to sum up, the absolute cost/quality winner of my contest is:
Panasonic 120uF/25V/FC // 100nF/X7R/0612 // 220nF/X7R/0805 // Murata 4.7uF/16V/Y5V/1206
And thanks to all the people who have made the clever suggestions to improve quality and reduce costs
What I'm looking for now is some little feedback from you, Gentlemen. I'm not (by far) finished with my future DAC design, and if someone could test and report audio results before I can actually post mine, that would be great...
Last thing, just to relaunch the debate : What about the piezo-electric properties of Y5V compared to X7R ? Ceramic layers are so thin in these caps that vibrations may induce some unwanted noise. So we will have to damp them. Any suggestion for a damping material have nice HF dielectric properties ?
Attachments
Well, the FC seems to be superior in spite of inferior spec values. By the way, www.schuro.de seem to carry even large values of FC (suitable for power amps) at comparatively reasonable prices.
Now for something that confuses me: BC components (formerly Philips Passive) type 136 low impedance cap specs give ESR at 120 Hz and impedance at 100 kHz. For 1000uF/25V, ESR is 0.22R and impedance is 0.034R.
To my understanding ESR (equivalent series resistance) is a real impedance. A HF complex impedance can be no smaller than the real series impedance. Strange, uh?
Now for something that confuses me: BC components (formerly Philips Passive) type 136 low impedance cap specs give ESR at 120 Hz and impedance at 100 kHz. For 1000uF/25V, ESR is 0.22R and impedance is 0.034R.
To my understanding ESR (equivalent series resistance) is a real impedance. A HF complex impedance can be no smaller than the real series impedance. Strange, uh?
Great, will keep me looking forward to those 4u7 caps even more so! I guess the top graph (22u vs. 4u7) with the <60 MHz setting was also with the Rubycon in parallel? Then it is an indication that the ESR/ESL of the Rubycon is still too high and it benefits from having a couple of uF in parallel with very low ESR/ESL. One could of course always parallel two 4u7 and they'd probably be both cheaper and more effective than a single 22uF.
If you could make your HC "noise generator" process also a signal divided down to <10 kHz, this would be a nice simulation of a sdata audio stream being handled. This might make the distinguishment between various electrolytics and various size ceramic caps clearer. But it's a lot of work...
Oh, considering the HF noise, it is beginning to surprise me why the 22uF gave such a big improvement compared to all the configurations of paralleled 10, 100 and 220 nF caps you tried. After all, the 4u7 already goes in the direction of the paralleled smaller caps. Maybe the Murata just has a better internal design? I can send you 1 and 2u2 caps from the same Murata series if you want.
Another wild guess is that the position for the 1210 cap in your layout is more effective than the 0805 position, but looking at your pictures and explanations, I couldn't say way.
I am not sure what you mean by contributing audio results, but I'll give it a try. I have designed a little DAC board that contains an ECL master oscillator, clock buffer to feed the clock to the CD player, and it will also hold an AD1854 or AD1852 DAC. I took exceptional care for good layout and decouling, and I used above mentioned 1 and 2u2 caps liberally. The first board had to use the almost humble AD1854J because I wanted to use the Sony format serial data from the cheapo player I hooked it up to (the 1852 needs a uC to set it to Sony). Still, the sound was exceptionally transparent and relaxed, much more so than I had ever observed in years of modyfing lots of different players with different DACs, some of which were highly respected.
A second board I built and hooked up to the same player seems to sound even slightly better. It contains the AD1854K (never understood why there would be selection grades of 1bit-DACs), the analog filter was dimensioned slightly differently, and I used Panasonic FC electrolytics and maybe 1-2 more of the 2u2 caps than on the first board. Now don't ask me what made the difference.
I have almost finished putting the component values into the silkscreen layer. If anybody is interested, I can mail out the eagle board file soon.
Eric
If you could make your HC "noise generator" process also a signal divided down to <10 kHz, this would be a nice simulation of a sdata audio stream being handled. This might make the distinguishment between various electrolytics and various size ceramic caps clearer. But it's a lot of work...
Oh, considering the HF noise, it is beginning to surprise me why the 22uF gave such a big improvement compared to all the configurations of paralleled 10, 100 and 220 nF caps you tried. After all, the 4u7 already goes in the direction of the paralleled smaller caps. Maybe the Murata just has a better internal design? I can send you 1 and 2u2 caps from the same Murata series if you want.
Another wild guess is that the position for the 1210 cap in your layout is more effective than the 0805 position, but looking at your pictures and explanations, I couldn't say way.
I am not sure what you mean by contributing audio results, but I'll give it a try. I have designed a little DAC board that contains an ECL master oscillator, clock buffer to feed the clock to the CD player, and it will also hold an AD1854 or AD1852 DAC. I took exceptional care for good layout and decouling, and I used above mentioned 1 and 2u2 caps liberally. The first board had to use the almost humble AD1854J because I wanted to use the Sony format serial data from the cheapo player I hooked it up to (the 1852 needs a uC to set it to Sony). Still, the sound was exceptionally transparent and relaxed, much more so than I had ever observed in years of modyfing lots of different players with different DACs, some of which were highly respected.
A second board I built and hooked up to the same player seems to sound even slightly better. It contains the AD1854K (never understood why there would be selection grades of 1bit-DACs), the analog filter was dimensioned slightly differently, and I used Panasonic FC electrolytics and maybe 1-2 more of the 2u2 caps than on the first board. Now don't ask me what made the difference.
I have almost finished putting the component values into the silkscreen layer. If anybody is interested, I can mail out the eagle board file soon.
Eric
microphonics and nonlinear properties
That question would almost merit a new thread. Obviously, other factors that ESR, ESL and DA may influence the audio quality of a cap.
The most obvious is microphonics, as pointed out by you. My guess is that a dielectric with a high dielectric constant would be more susceptible. But it would also depend on the mechanical properties of both the dielectric material and the overall design of the cap.
Most sensitive application in my eyes is not rail decoupling but reference decoupling, i.e. voltage reference in a DAC or zener diode in discrete current sources.
Then there is the question how a cap reacts to voltage drops and current draw. ESR, ESL, DA are, after all, just linear measurement methods, i.e. they are done with a huge DC offset and a small AC test signal.
Capacitance vs. voltage is specified but measurement is carried out again only by changing the DC offset from measurement to measurement. How is linearity with voltage drop spikes such as may occur when a DAC charges its internal capacitors? Again, this is not only a question of the dielectric properties alone but also of the mechanical properties as the electrodes may "move" when voltages change or currents flow.
Listenting tests might help but might only produce new myths. We should try to come up with tests for nonlinear capacitor properties....
Eric
ftorres said:Last thing, just to relaunch the debate : What about the piezo-electric properties of Y5V compared to X7R ? Ceramic layers are so thin in these caps that vibrations may induce some unwanted noise. So we will have to damp them. Any suggestion for a damping material have nice HF dielectric properties ?
That question would almost merit a new thread. Obviously, other factors that ESR, ESL and DA may influence the audio quality of a cap.
The most obvious is microphonics, as pointed out by you. My guess is that a dielectric with a high dielectric constant would be more susceptible. But it would also depend on the mechanical properties of both the dielectric material and the overall design of the cap.
Most sensitive application in my eyes is not rail decoupling but reference decoupling, i.e. voltage reference in a DAC or zener diode in discrete current sources.
Then there is the question how a cap reacts to voltage drops and current draw. ESR, ESL, DA are, after all, just linear measurement methods, i.e. they are done with a huge DC offset and a small AC test signal.
Capacitance vs. voltage is specified but measurement is carried out again only by changing the DC offset from measurement to measurement. How is linearity with voltage drop spikes such as may occur when a DAC charges its internal capacitors? Again, this is not only a question of the dielectric properties alone but also of the mechanical properties as the electrodes may "move" when voltages change or currents flow.
Listenting tests might help but might only produce new myths. We should try to come up with tests for nonlinear capacitor properties....
Eric
No, The 'lytic was quietly sleeping in its box during both measurements.
Eric, you don't want me to see next Christmas ? Yes in fact, I have a little 6 MHz canned XO, but not a lot of courage...
Well, you may be right. The 220nF/0805 cap is located on the solder side, and right between output and ground pins of the regulator. It should be interesting to try a 220nf/0603 stacked on the top of the 100nF/0612 instead, since they would have the same dimensions between solder pads, and stacking would be easier.
Well, it's almost what you've guessed. As it seems that only a few caps, and mainly the 4.7uF ceramic, improve decoupling efficiency in a spectacular way, I was wondering if some of us could add these little caps on their existing digital boards, and tell us if they hear an improvement or not. I plan to do it on my old DAC, but I have to find the time to do it... I'm afraid it won't be before september, and if someone is quicker, should be nice...
This is extremely interesting work. Thank you, ftorres.
I was just drawing some boards for my DAC, and it's a shame how much space all these capacitors take up
Anyway, I just wanted to note that this configuration seems extreme at first glance, but upon closer study is not too absurd. Even the datasheet for the TI PCM1704U DAC I am using specifies 100uF electrolytic, 4.7uF ceramic, and 100nF ceramic in parallel on each power pin.
I got pretty distracted building my preamp, but am turning back to the DAC project again. Perhaps I'll be able to report how this configuration sounds in real life, soon.
I was just drawing some boards for my DAC, and it's a shame how much space all these capacitors take up
Anyway, I just wanted to note that this configuration seems extreme at first glance, but upon closer study is not too absurd. Even the datasheet for the TI PCM1704U DAC I am using specifies 100uF electrolytic, 4.7uF ceramic, and 100nF ceramic in parallel on each power pin.
I got pretty distracted building my preamp, but am turning back to the DAC project again. Perhaps I'll be able to report how this configuration sounds in real life, soon.
Do we really need electrolytics?
The work done on this thread is totally killer! This is some of the most appropriate and interesting work I have seen for yonks.
It seems that a lot of effort has also gone into cost reduction which I applaud, but since I use small volumes and SMT components on both sides of the board, I am personally much more interested in space reduction, and removal if possible of through-hole caps.
My question is:
Would one be able to get by without the electrolytic caps altogether if using the 22u or even the 100u Murata (http://www.murata.com/cap/nproduct/capa32.html)?
Considering how good the large value Murata caps are, would it not be opportune to consider bypassing by a small value stacked Murata? How about a combination of X5R .33 http://www.murata.com/cap/nproduct/capa33.html and CoG 10pF 50V http://www.murata.com/cap/nproduct/capa29.html
How well can one expect these bypass methods to work on the analog section of a DAC chip such as 1704 or better still the newer TI PCM1730 family?
Petter
The work done on this thread is totally killer! This is some of the most appropriate and interesting work I have seen for yonks.
It seems that a lot of effort has also gone into cost reduction which I applaud, but since I use small volumes and SMT components on both sides of the board, I am personally much more interested in space reduction, and removal if possible of through-hole caps.
My question is:
Would one be able to get by without the electrolytic caps altogether if using the 22u or even the 100u Murata (http://www.murata.com/cap/nproduct/capa32.html)?
Considering how good the large value Murata caps are, would it not be opportune to consider bypassing by a small value stacked Murata? How about a combination of X5R .33 http://www.murata.com/cap/nproduct/capa33.html and CoG 10pF 50V http://www.murata.com/cap/nproduct/capa29.html
How well can one expect these bypass methods to work on the analog section of a DAC chip such as 1704 or better still the newer TI PCM1730 family?
Petter
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.