More bigger MLCCs will vibrate your board even better according to current drawn by load. These things are piezo transducers. Also the ESR is too low.
> those chips have a big steady-state, but on top of that, like any chip, there is a transient state.
If you want to keep your ground clean from choppy current drawn by the big digital chip, put shunt regulator next to it, return current near chip's ground pins, plus careful layout of decoupling caps.
If you don't, use fast LDO like NCP566. There are others in the same style from OnSemi. Performance is just ridiculous. 1 mOhm output impedance, DC to about 20kHz, microsecond transient response, low noise, high PSRR, cheap, but it is picky about caps. I mean, you need caps that match its performance. So you will end up with polymers that are impossible to use properly. Also, obviously, it's not a shunt, so all the noise current drawn by the load ends up in the main supply. So you need more caps and RCs and ferrite beads.
> ESR of electrolytics (os-con is roughly equal) is so high that they can't match the transient reponses.
ESR is good. ESR is nice. Everyone loves ESR.
ESR MAKES YOUR TRANSIENT RESPONSE BETTER, not worse.
The best electrolytic cap for low voltage decoupling, which is Panasonic FR 470µF 6.3V, has 2.5mm pin spacing and 0.1 ohm ESR, ideal combo with its own inductance of about 2-3 nH for a smooth transition to 1µF MLCC without antiresonance peak. If you want to use 100nF MLCC instead of 1µF, then you need more ESR to avoid resonance, solution is a cheaper aluminium cap. But of course, the result is worse impedance.
The famous purple polymer caps are mostly useless for audio, too little ESR. OSCONs and the like, properly used, belong on your GPU.
Why? Their ESR is too low for their inductance, so it will ring with the next cap which is MLCC. The way to use them properly is with a few dozen MLCCs per OSCON, on 4-6 layer board, with close coupled power and ground planes, then it won't ring. Just like on a GPU. Too expensive.
Low inductance.
Note the 8 cents price of the FR cap I posted above means "low inductance" because you just put 10 in parallel on power and ground planes. Costs the same as a fancy polymer cap, but 10x less inductance. That works wonders, but it takes some space.
Note it's the same for all the fancy polymers, the super low ESL tantalum polymers, etc. If you absolutely need it to be tiny, then yes. Otherwise, if you have space, a bunch of cheap thru hole aluminium caps in parallel will cost less and give lower impedance.
ESR of MLCC is always too low. Very annoying, screws up the transient, and causes ringing with the next cap. So if you use low-ESR MLCC then you have to use high ESR bulk cap to avoid ringing. You get less impedance at the frequency the MLCC is decoupling, but more impedance where the other caps are working.
It's like whackamole. If you drive down the impedance too much somewhere, a peak pops up at another frequency. It gets expensive quickly.
Audiophile dislike ceramics. But C0G has best distortion performance and is very cheap. It beats everything else.
What matters is impedance.
At HF it's inductance, which is package size.
The low value cap works better because it is physically smaller. But 0402 caps are so annoying to hand solder.
Basically, get some NFM21PS106 from Murata.
Once I put a X7R MLCC as "VREF decoupling" cap of a low noise LDO. Big mistake, the thing turned into a microphone. Switched to C0G, problem solved.
While I agree with most of what you've said here, I don't think there is a practical issue using X7R as reference decoupling for a regulator. The LT3042/5 is characterized by Linear themselves using X7R. Similarly, LTC6655 and other references are the same. Everyone is aware they are piezoelectric, so if this is critical in a design then you could avoid it. If it were such a disaster they would not use it on the reference design which generated all of their pristine measurements. Obviously if the cap value is low enough for C0G then you should use that... but it's not for LT304x.
When X7R caps are squeezed they generate charge, so the effect is proportional to the impedance across them that can get rid of the charge. If they're used at the output of a regulator, that's a very low impedance node, so you get zero effect.
The LDO I had problems with (forgot which one) had a very high impedance reference node, probably to get a good filtering with a low value cap, so it was very sensitive.
Out of curiosity I checked LT3042 datasheet, and they mention it, look at the bottom left of page 16.
"given the high SET pin impedance, any piezoelectric response from a ceramic SET pin capacitor generates significant output noise – peak-to-peak excursions of hundreds of µVs."
Also I was quite surprised that the output voltage of some other LDOs that do not have reference bypass caps does wiggle quite a bit when flexing the board.
The LDO I had problems with (forgot which one) had a very high impedance reference node, probably to get a good filtering with a low value cap, so it was very sensitive.
Out of curiosity I checked LT3042 datasheet, and they mention it, look at the bottom left of page 16.
"given the high SET pin impedance, any piezoelectric response from a ceramic SET pin capacitor generates significant output noise – peak-to-peak excursions of hundreds of µVs."
Also I was quite surprised that the output voltage of some other LDOs that do not have reference bypass caps does wiggle quite a bit when flexing the board.
Yes, but if you don't vibrate them there isn't a problem. I have used LT3045 with X7R MLCC on the SET pin as the reference design does and it performs incredibly well. They mention it, but all of their measurements and their eval board uses X7R in that position. Obviously, they are not suitable for high vibration environments.
Last edited:
hello peufue,
Right or wrong ?
Thus, how a high ESR cap can be good for the chip, when this cap is slow to match the demand in current of this poor chip ?
By the way, when you state "dead low ESR of MLCC is bad", it's based to knowledge or experiments ?
Please note that I'm talking only about the caps linked the power supply pins of CPLD/SoC/FPGA etc...
Rgds
The lower the ESR, faster the cap reacts to match the transient response of a chip (case where the chip needs current).
Right or wrong ?
Thus, how a high ESR cap can be good for the chip, when this cap is slow to match the demand in current of this poor chip ?
By the way, when you state "dead low ESR of MLCC is bad", it's based to knowledge or experiments ?
Please note that I'm talking only about the caps linked the power supply pins of CPLD/SoC/FPGA etc...
Rgds
> The lower the ESR, faster the cap reacts to match the transient response of a chip (case where the chip needs current). Right or wrong ?
It's the whole impedance of the entire power rail that matters. At HF, that means mostly inductance. Below that, it's about getting rid of the antiresonance peaks that increase impedance and cause ringing. It's the peaks that matter. The point is you get lower impedance with the right amount of ESR in your caps, because that allows you to get rid of the resonance peaks. Too much ESR, it'll be wimpy. Too little, it'll ring.
> Please note that I'm talking only about the caps linked the power supply pins of CPLD/SoC/FPGA etc...
That's a different scenario than audio design. I'm talking about stuff like DACs, not the BGA chips with 200 power and ground pins.
If your board has close coupled planes then you can parallel caps with low inductance between them, so you get less antiresonance peaking, and you can use the fashionable OSCONs and the like. On a 4 layer board, it's tricky, since the only close coupled plane you got is L3/L4 and it's on the other side of the board, unless you use power pours under the IC, but then the caps have to be on the other side, which increases their inductance, etc. On a 2-layer board, polymer caps are pretty much not usable, and MLCCs wired in parallel with traces (=inductance) will ring and make a lot of noise.
It's the whole impedance of the entire power rail that matters. At HF, that means mostly inductance. Below that, it's about getting rid of the antiresonance peaks that increase impedance and cause ringing. It's the peaks that matter. The point is you get lower impedance with the right amount of ESR in your caps, because that allows you to get rid of the resonance peaks. Too much ESR, it'll be wimpy. Too little, it'll ring.
> Please note that I'm talking only about the caps linked the power supply pins of CPLD/SoC/FPGA etc...
That's a different scenario than audio design. I'm talking about stuff like DACs, not the BGA chips with 200 power and ground pins.
If your board has close coupled planes then you can parallel caps with low inductance between them, so you get less antiresonance peaking, and you can use the fashionable OSCONs and the like. On a 4 layer board, it's tricky, since the only close coupled plane you got is L3/L4 and it's on the other side of the board, unless you use power pours under the IC, but then the caps have to be on the other side, which increases their inductance, etc. On a 2-layer board, polymer caps are pretty much not usable, and MLCCs wired in parallel with traces (=inductance) will ring and make a lot of noise.
Then we are not talking about "simple" decoupling anymore, but PDN (Power Delivery Network) design with PCB stackups usually using upwards of 8 Layers. For dealing with that kind of stuff you want as precisely specified passives as possible in combination with simulation tools like Keysight ADS. And you might also want a multi GHz vector network analyser and some other goodies on your bench
Don't know how this discussion veered off into power supply decouplers. Discussion started as question about audio couplers & filter circuits. I have lots of audio couplers & filter circuits in organs and amps that originally had shorted tantalum or dried up non-polar electrolytics. Film caps are a great idea but the ones I can buy in stock at distributors with leads are 63 volt rated, 2" long by 1/2" diameter. Much too big. You Europeans may be able to buy exquisite special parts, but if I try to get anything through customs, it often gets lost. I knew the son of a US customs agent once; sophomore in college drove a nice one year old Oldsmobile while I drove a 15 year old Ford.
Nice to read about advantages of mythical COG ceramics. I haven't seen any 1 to 10 uf in stock at newark or digikey for 10 years. Yes, if you can afford a 10000 piece purchase, AVX will make you some in a year or two.
Rather than install new 2.2 50 vNP coupler caps on the input of my mixers & amps, I've been installing 2.2 uf 50 v x7r. NP electrolytic caps come with no life rating, may be sealed with red gum rubber for all I know. Tantalum caps come with no life rating and I've encountered dozens of shorted ones in my repairs. I don't trust them farther than I can throw them. I don't like doing a job over & over. After conversion to MLCC ceramic, the amps & organs don't start sounding like a cheap transistor radio. A new input cap builds up anemic volume sometimes. And since the amps are not mounted inside a speaker, microphonic is not a problem. Microphonic hasn't been a problem in the H100 organs which do have 3 speakers in the case.
As far as ghz vector analyzers etc, I've just now gotten a 20 mhz scope working. One that was NOT made in ***** by serfs. **** thing had a hundred dried up electrolytics in it, and the PCB's were glued in to prevent repair.
Nice to read about advantages of mythical COG ceramics. I haven't seen any 1 to 10 uf in stock at newark or digikey for 10 years. Yes, if you can afford a 10000 piece purchase, AVX will make you some in a year or two.
Rather than install new 2.2 50 vNP coupler caps on the input of my mixers & amps, I've been installing 2.2 uf 50 v x7r. NP electrolytic caps come with no life rating, may be sealed with red gum rubber for all I know. Tantalum caps come with no life rating and I've encountered dozens of shorted ones in my repairs. I don't trust them farther than I can throw them. I don't like doing a job over & over. After conversion to MLCC ceramic, the amps & organs don't start sounding like a cheap transistor radio. A new input cap builds up anemic volume sometimes. And since the amps are not mounted inside a speaker, microphonic is not a problem. Microphonic hasn't been a problem in the H100 organs which do have 3 speakers in the case.
As far as ghz vector analyzers etc, I've just now gotten a 20 mhz scope working. One that was NOT made in ***** by serfs. **** thing had a hundred dried up electrolytics in it, and the PCB's were glued in to prevent repair.
Last edited:
hello indianajo,
This discussion "veered off" into PS (thanks to add to my Eng. voc. ;-) ) because I use MLCC X-type (X7R X5R) there. I wanted to create a new thread but the web interface showed there was already that thread in line, close to my goal, thus I woke up that old thread.
The main point of this thread is to show that :
- MLCC X-type are good for audio.
- Don't listen to the old EEs or young ones, that learnt at school "MLCC are bad for audio, use only electrolytics + film". They never tested MLCC thus they know nothing about it. If they had tested, they would realized instead that X-type improve SQ like hell (3D stage, precision, fluid etc, no harshness) thanks to their dead low ESR in combination with their rather high capacity (>1mF easy by staking them)
- I've tested & used X-type right next to the PS pins of chips (CPLD Xilinx, XMOS etc) => SQ improvement is night/day. By combining 2 C0G with the X-type, you have a X-type that do not lead to harsh sound (coz piezo of x-type). The trick is basic : plug the X-type on top of the decoupling SMD ceramic already on the PCB next to these PS pins, then you can unsoldier the electrolytic bulk upstream (useless)
- now, if you Indianajo, you say that they are suitable on signal path too : too cool ! => we can use them anywhere !
Rgds
This discussion "veered off" into PS (thanks to add to my Eng. voc. ;-) ) because I use MLCC X-type (X7R X5R) there. I wanted to create a new thread but the web interface showed there was already that thread in line, close to my goal, thus I woke up that old thread.
The main point of this thread is to show that :
- MLCC X-type are good for audio.
- Don't listen to the old EEs or young ones, that learnt at school "MLCC are bad for audio, use only electrolytics + film". They never tested MLCC thus they know nothing about it. If they had tested, they would realized instead that X-type improve SQ like hell (3D stage, precision, fluid etc, no harshness) thanks to their dead low ESR in combination with their rather high capacity (>1mF easy by staking them)
- I've tested & used X-type right next to the PS pins of chips (CPLD Xilinx, XMOS etc) => SQ improvement is night/day. By combining 2 C0G with the X-type, you have a X-type that do not lead to harsh sound (coz piezo of x-type). The trick is basic : plug the X-type on top of the decoupling SMD ceramic already on the PCB next to these PS pins, then you can unsoldier the electrolytic bulk upstream (useless)
- now, if you Indianajo, you say that they are suitable on signal path too : too cool ! => we can use them anywhere !
Rgds
General statements like "low ESR is bad for SMPS" are useless imho - the context has to be considered. The same applies to comparisions of impedance curves of different caps. Lab measurements of series resonant frequences are valid with extreme short wiring hard to be achieved in reality. For instance parallelling lo-capacitance cercos to the battery of 10000uF bulk caps does not make any sense, etc ppp....
Hi. I need some practical advice:
I need to replace the the SMD caps in a
battery driven 48V phantom power supply.
The corner frequency of the original caps, that decouple
the 48V DC voltage from the audio path, is way too high,
about 150hz.
Place is very small for a replacement, I can only use 1206 SMD parts.
Would the attached X7R caps (100v/1µF) be suitable?
AC is "only" microphone level signals...
https://datasheets.kyocera-avx.com/X7RDielectric.pdf
I need to replace the the SMD caps in a
battery driven 48V phantom power supply.
The corner frequency of the original caps, that decouple
the 48V DC voltage from the audio path, is way too high,
about 150hz.
Place is very small for a replacement, I can only use 1206 SMD parts.
Would the attached X7R caps (100v/1µF) be suitable?
AC is "only" microphone level signals...
https://datasheets.kyocera-avx.com/X7RDielectric.pdf