Not sure what science your referring too. However here is a DAC I built last year, as well as others. And it's well documented from diy members including an ee. That in some cases by replacing the low uF ceramic decoupling cap on the op amps rails with a low esr and higher capacitance lytic, resulted in improves Sq.
Here is one of many instances that validate that such a mod made a nice improvement.
http://www.diyaudio.com/forums/digi...gn-mod-not-play-music-not-12.html#post4337278
Last edited:
Well, here's your science, since the pin is a Reference Voltage (not a power supply) :
1) Higher capacitance value provides better reference stability and lower noise, which is nice to have (at the price of a slower startup)
2) High-K ceramic dielectrics (X7R and friends) exhibit HUGE microphony (it is, after all, piezoelectric). Stick a X7R cap on the REF pin of your LDO, shake the board a little, and watch the output voltage wiggle... replacing the cap with a non-microphonic type (Alu, tant, C0G, whatever) fixes the issue.
No mystery here...
C0G is best if you can find the value you need
If you need 100µF, well...
What small signal opamp needs that kind of bypass?!
We're not talking 60dB low noise preamps that need ac coupling here, haha.
C0G is best if you can find the value you need
If you need 100µF, well...
I suppose you can parallel stack a bunch of GOGs until you reach 100UF. =)
C0G/NP0 is great for filters, like post-DAC antialias, EMI filter or active crossover, that kind of stuff... if you manage to measure any kind of THD on one of these caps, clean your solder flux and try again!
For rail decoupling, well, the values are simply too small. A 10nF (or even 100nF) ceramic cap opens a can of worms, as it will resonate with the nearby 1uF MLCC and/or the low ESR alu cap.
For a reference bypass I'd tend to look at the noise behavior, if increasing capacitance lowers noise, as is usually the case, then an aluminium cap would be the way to go.
For rail decoupling, well, the values are simply too small. A 10nF (or even 100nF) ceramic cap opens a can of worms, as it will resonate with the nearby 1uF MLCC and/or the low ESR alu cap.
For a reference bypass I'd tend to look at the noise behavior, if increasing capacitance lowers noise, as is usually the case, then an aluminium cap would be the way to go.
That's exactly what many LDO manufacturer datasheet recommend !!!!
are they all wrong ?
.
If you use your LDO to power a CPU which doesn't care about 50mV noise, no problem.
If you use a 10µV noise LDO and expect to meet spec with a piezoelectric X7R cap on the REF pin, then it will only work if the board does not vibrate. That would depend on the way it's assembled, etc. ADP151, for example, sidesteps the problem by not requiring an external capacitor. Plus, it's a nice LDO, fast transient response, reasonably low Z...
When in doubt, just measure it! It's not that difficult... just stick a scope on the output and gently tap on the board with your finger... it's a non-quantitative test (unless your finger is calibrated) but it does tell the story.
Is a 5 cents aluminium capacitor too expensive so you need to put a 2 cents ceramic?
Hell, I built Groner's excellent low noise preamp : the input cap (a big MKP) has microphony too... a lot less than piezo ceramics, but still not negligible.
If you use a 10µV noise LDO and expect to meet spec with a piezoelectric X7R cap on the REF pin, then it will only work if the board does not vibrate. That would depend on the way it's assembled, etc. ADP151, for example, sidesteps the problem by not requiring an external capacitor. Plus, it's a nice LDO, fast transient response, reasonably low Z...
When in doubt, just measure it! It's not that difficult... just stick a scope on the output and gently tap on the board with your finger... it's a non-quantitative test (unless your finger is calibrated) but it does tell the story.
Is a 5 cents aluminium capacitor too expensive so you need to put a 2 cents ceramic?
Hell, I built Groner's excellent low noise preamp : the input cap (a big MKP) has microphony too... a lot less than piezo ceramics, but still not negligible.
I started experimenting with polymer capacitor types a couple months ago when I got into a situation where I figured out that the Class A op amp circuits in one of my DAC modifications were being starved. I had 5 of the discrete Sparkos Labs Class A bipolar op amps in the DAC and it is very easy to starve these of power. Starving can happen in a couple of different ways. The first is if the capacitor/power supply runs out of gas (i.e. voltage drops) when responding to very fast high frequency spikes (like hitting of cymbols, snare ring taps, etc.). When this occurs, the waveform in the op amp essentially flatlines when it runs out of voltage for slew and you get a clipping type sound. This comes across as a bright/harsh clicking sound instead of the natural ring of the high frequency waveforms. The second way to starve Class A is not to have enough overall constant voltage for the Class A circuit to work with. In this situation, the sound becomes very warm and lush. The fast high frequency excitement is lost – however it is a very interesting sound that some may like. For me, it becomes boring when listening for any length of time.
Adding larger capacitors on the power rails really do not help here as the larger caps cannot discharge fast enough to respond to the immediate transients that occur in the upper midrange and highs. I do not have enough capacitor points to put in a larger bank of smaller size capacitors (such as a number of 47uf caps – as I found that 100uf is even too slow to give the Class A circuits fast enough voltage spikes). With a 100uf, the midrange becomes recessed and you do hear some of the high-frequency clipping when it cannot provide the full voltage spike. At this point, I started trying polymer types as I know that they excel as caps for digital power supply lines (such as power supply and VREF for DAC chips). I do not have the ability to surround each Class A opamp with 4-6 capacitors, as is done in this Krell Class A preamp (they do know what they are doing with Class A circuits!):
http://www.gzhifi.com/fa/images/201408/1406922675180193346.JPG
Well, the first few hours were actually pretty good. Then the burn-in process started. Burn in was truly a very long and painful experience as within 10 hours, all the high frequencies went out. It was like there was a big rolloff on anything after 6khz. As burn-in proceeded, it slowly expanded it’s higher frequency range, but it seemed like there was a bump in whatever frequency point it was burning in (to my ears, this is what it sounded like). For example, if it was burning in at the 7khz point, there would be a big bump in frequencies at 7khz.
After 50 hours, the bass was lost.
After 100 hours, the sound was very forcefully bright and forward (however, ultra high frequencies were not there yet).
The bass slowly came back between 100-150 hours.
Somewhere between 150 and 200 hours it finally burned it all the way and I had a very smooth natural frequency response.
As I listened to this, I could really describe it as providing a very constant and smooth sound. The key word here is “smooth”. All the frequencies were there and I did not lose any midrange body or any bass/midbass frequencies. However, upon some critical listening, I found that it was just not responding fast enough for those sharp/spiked transients. This gave the music a very smooth laid-back effect. It sounded very pleasant and there was absolutely no listening fatigue, but it just did not provide that last bit of excitement/slam that was needed. I think what the polymer cap is doing here is slowly turning on the voltage output when it senses a decrease in voltage. It’s a gradual delay in providing voltage. However, once it starts discharging voltage, it continues to do so in a very authoritative manner. The end result here is that it is making the slow slew curve of the Class A circuit even more slower, therefore removing any of the immediacy and slam that is naturally in the music. This may actually help conventional op amps that have a very immediate slew (make them act more like Class A devices), but I have yet to experiment with them in this scenario. This is very different from normal eletrolytics where they will discharge fast and immediate – however the voltage from the electrolytic will start to drop immediately and fast, where the polymer will attempt to keep the voltage as constant as possible as it discharges until it gets low enough where the voltage does start to drop.
I’m referencing my other post here to explain what I mean by “slew curve”:
http://www.diyaudio.com/forums/digi...te-opamp-output-my-cd-player.html#post4741445
What I had to do was solder an Elna Cerafine directly to the pins of the op amps (and I eventually settled on just the I/V op amps as the Elna signature is just too much for me). Using polymers as all pre/post regulator points and having a conventional electrolytic at the last point before the op amp really solved my problems. The polymers gave me the solid/constant voltage I needed to keep the Class A circuits happy and the electrolytic on the op-amp pins gave me the immediacy/slam I wanted.
I will be moving on to experiment with Nichicon FG/KZ directly on the op amps as I don’t like the Elna sonic signature (as a power supply cap). Both the Elna caps (Cerafine and Silmic) have this warm glare to the sound (not neutral at all). The difference being that the high frequencies are rolled off with the Silmic. The Cerafine also has this etched/metallic sound in the high frequencies that just doesn’t sound good to me. That being said, as a DC blocking signal capacitor, the Silmic just awesome!
It looks like the Polymers physical size max out at about 10mm x 14mm. The actual voltage becomes a sliding scale. This means you can put an 1800uf / 6.3V cap in for your DAC circuit (which is awesome) – and that means a much cleaner bass with less distortion. For higher voltages, you can put in a 390uf 25V or 1000uf 16V. For 35V, I use the 120uf Panasonic SEPF. I could do the Nichicon, which provided 150uf, but I liked the slighter better ESR of the Panasonic.
Adding larger capacitors on the power rails really do not help here as the larger caps cannot discharge fast enough to respond to the immediate transients that occur in the upper midrange and highs. I do not have enough capacitor points to put in a larger bank of smaller size capacitors (such as a number of 47uf caps – as I found that 100uf is even too slow to give the Class A circuits fast enough voltage spikes). With a 100uf, the midrange becomes recessed and you do hear some of the high-frequency clipping when it cannot provide the full voltage spike. At this point, I started trying polymer types as I know that they excel as caps for digital power supply lines (such as power supply and VREF for DAC chips). I do not have the ability to surround each Class A opamp with 4-6 capacitors, as is done in this Krell Class A preamp (they do know what they are doing with Class A circuits!):
http://www.gzhifi.com/fa/images/201408/1406922675180193346.JPG
Well, the first few hours were actually pretty good. Then the burn-in process started. Burn in was truly a very long and painful experience as within 10 hours, all the high frequencies went out. It was like there was a big rolloff on anything after 6khz. As burn-in proceeded, it slowly expanded it’s higher frequency range, but it seemed like there was a bump in whatever frequency point it was burning in (to my ears, this is what it sounded like). For example, if it was burning in at the 7khz point, there would be a big bump in frequencies at 7khz.
After 50 hours, the bass was lost.
After 100 hours, the sound was very forcefully bright and forward (however, ultra high frequencies were not there yet).
The bass slowly came back between 100-150 hours.
Somewhere between 150 and 200 hours it finally burned it all the way and I had a very smooth natural frequency response.
As I listened to this, I could really describe it as providing a very constant and smooth sound. The key word here is “smooth”. All the frequencies were there and I did not lose any midrange body or any bass/midbass frequencies. However, upon some critical listening, I found that it was just not responding fast enough for those sharp/spiked transients. This gave the music a very smooth laid-back effect. It sounded very pleasant and there was absolutely no listening fatigue, but it just did not provide that last bit of excitement/slam that was needed. I think what the polymer cap is doing here is slowly turning on the voltage output when it senses a decrease in voltage. It’s a gradual delay in providing voltage. However, once it starts discharging voltage, it continues to do so in a very authoritative manner. The end result here is that it is making the slow slew curve of the Class A circuit even more slower, therefore removing any of the immediacy and slam that is naturally in the music. This may actually help conventional op amps that have a very immediate slew (make them act more like Class A devices), but I have yet to experiment with them in this scenario. This is very different from normal eletrolytics where they will discharge fast and immediate – however the voltage from the electrolytic will start to drop immediately and fast, where the polymer will attempt to keep the voltage as constant as possible as it discharges until it gets low enough where the voltage does start to drop.
I’m referencing my other post here to explain what I mean by “slew curve”:
http://www.diyaudio.com/forums/digi...te-opamp-output-my-cd-player.html#post4741445
What I had to do was solder an Elna Cerafine directly to the pins of the op amps (and I eventually settled on just the I/V op amps as the Elna signature is just too much for me). Using polymers as all pre/post regulator points and having a conventional electrolytic at the last point before the op amp really solved my problems. The polymers gave me the solid/constant voltage I needed to keep the Class A circuits happy and the electrolytic on the op-amp pins gave me the immediacy/slam I wanted.
I will be moving on to experiment with Nichicon FG/KZ directly on the op amps as I don’t like the Elna sonic signature (as a power supply cap). Both the Elna caps (Cerafine and Silmic) have this warm glare to the sound (not neutral at all). The difference being that the high frequencies are rolled off with the Silmic. The Cerafine also has this etched/metallic sound in the high frequencies that just doesn’t sound good to me. That being said, as a DC blocking signal capacitor, the Silmic just awesome!
It looks like the Polymers physical size max out at about 10mm x 14mm. The actual voltage becomes a sliding scale. This means you can put an 1800uf / 6.3V cap in for your DAC circuit (which is awesome) – and that means a much cleaner bass with less distortion. For higher voltages, you can put in a 390uf 25V or 1000uf 16V. For 35V, I use the 120uf Panasonic SEPF. I could do the Nichicon, which provided 150uf, but I liked the slighter better ESR of the Panasonic.
A bit off topic but hey this my thread. I am not convinced that the electrical characteristics of the polymer caps change significantly enough with time to make bass or treble disappear but hey this what makes it all interesting.
Sent from my iPad using Tapatalk
I see the need for three grades of local supply rail decoupling.
High Frequency decoupling that handles the very highest frequencies all of which are ultrasonic.
Medium Frequency decoupling that handles the extreme treble and the lower end of the ultrasonic.
Low Frequency decoupling, which handles all the bass and all the mid and the lower end of the treble.
HF requires capacitance that can discharge, or charge very quickly. The demands ultra low inductance and that in turn demands low loss ceramic capacitors on the supply pins, with ultra short connections. X7R smd fit this best. Dual polarity requires at least two of these and these must be connected to each other at the chip, again with the shortest possible leads. Large values of capacitance are not required because the transients by definition must be very short lived. 100nF to 1uF is generally sufficient.
MF can be located a bit further away from the supply pins, These will recharge the HF decoupling and supply current in the longer term to the chip. Typically electrolytic capacitors of small size and low inductance located a few mm away from the supply pins can meet this medium speed duty. The junction between the two MF decoupling caps are included in the route length assessments. This MF centre tap is connected to the HF centre tap with a very short trace. These typically can be from 1uF to 10uF
LF can be located at the edge of the PCB. These only handle lower frequency glitches and provide current to recharge the HF &MF decoupling. Higher trace inductance is acceptable and higher capacitor inductance is also acceptable. Electrolytics are always used for this duty. Again the junction between the two LF decoupling caps are included in the route length assessments. This LF centre tap is connected to the HF/MF centre tap with a trace that is close coupled to the supply lines. Typically 47uF to 2200uF.
In all of the situations above, once a package style has been adopted to fit the layout and inductance parameter, one should select the highest capacitance that fits inside that package.
All electrolytics should be reformed before fitting.
It is my view that once the correct type and size and value have been selected and located that there will be no audible effects by swapping in different manufacturers or different styles of manufacture. "Burning in" will not happen, since leakage currents will not change during the early life of the electrolytics.
High Frequency decoupling that handles the very highest frequencies all of which are ultrasonic.
Medium Frequency decoupling that handles the extreme treble and the lower end of the ultrasonic.
Low Frequency decoupling, which handles all the bass and all the mid and the lower end of the treble.
HF requires capacitance that can discharge, or charge very quickly. The demands ultra low inductance and that in turn demands low loss ceramic capacitors on the supply pins, with ultra short connections. X7R smd fit this best. Dual polarity requires at least two of these and these must be connected to each other at the chip, again with the shortest possible leads. Large values of capacitance are not required because the transients by definition must be very short lived. 100nF to 1uF is generally sufficient.
MF can be located a bit further away from the supply pins, These will recharge the HF decoupling and supply current in the longer term to the chip. Typically electrolytic capacitors of small size and low inductance located a few mm away from the supply pins can meet this medium speed duty. The junction between the two MF decoupling caps are included in the route length assessments. This MF centre tap is connected to the HF centre tap with a very short trace. These typically can be from 1uF to 10uF
LF can be located at the edge of the PCB. These only handle lower frequency glitches and provide current to recharge the HF &MF decoupling. Higher trace inductance is acceptable and higher capacitor inductance is also acceptable. Electrolytics are always used for this duty. Again the junction between the two LF decoupling caps are included in the route length assessments. This LF centre tap is connected to the HF/MF centre tap with a trace that is close coupled to the supply lines. Typically 47uF to 2200uF.
In all of the situations above, once a package style has been adopted to fit the layout and inductance parameter, one should select the highest capacitance that fits inside that package.
All electrolytics should be reformed before fitting.
It is my view that once the correct type and size and value have been selected and located that there will be no audible effects by swapping in different manufacturers or different styles of manufacture. "Burning in" will not happen, since leakage currents will not change during the early life of the electrolytics.
Last edited:
Where do people get the quaint idea that 20kHz is 'very fast high frequency'? In modern electronics terms it is almost DC!ainami said:
Electrolytics only need to be reformed before fitting if they have been unused for a few years since manufacture. AndrewT seems to have a bee in his bonnet about this!
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.