I think jackinnj's point is more that a super regulator can give you a lower output impedance and just as low noise than a battery. Therefore if properly configured and implemented the super reg will give you comparable performance, but will not require charging. Charging to me would be a major pain in the behind, I tried battery powered a few years ago and it frustrated me no end. The only way to get around that would be to include some charging system that kept the batteries topped off so I'd never end up with them going flat on me during a listening session - this obviously adds in cost. From my point of view the super reg just seems more convenient.
I don't know strictly how much current a super reg can provide, but it can be quite a lot. I easily pull 500mA+ out of mine, but if you use an LM317 as a pre reg, then it acts as protection vs over current and short circuit, so providing you keep the pass transistor suitably cool you can probably get ~1.5A out of one. I don't know why this is important though when the clock will only draw a few tens of mAs.
With an LME49710 as the error amp, and a D44H11 pass transistor you're probably talking less than $10 for parts.
I've used the Jung regulator to about 600mA.
I'm not anti-battery (did I say I was) -- they have a smooth, non quirky aspect to the sound as a product of their use. They have great rejection of low frequency (under 50Hz) energy.
Guess I should look at a jitter histogram...
Guys...
It appears that you are not reading the details of the posts. No one has ever established, to my knowledge, that any regulator will have a lower output z than a LiFePO4 battery. So far we have seen data only for older battery tech, which is not what qusp proposed.
Additionally, cost is not really a factor, but even if it was, one needs to consider the following costs for an AC based supply: transformer+diodes+smoothing caps+pre-regulator+local regulators of your choice. I am already using this set up, with a dedicated transformer for the oscillators power supply.
A battery requires: a single LiFePO4 cell for each oscillator, battery management, a charger, and a switch and jack for charging. I think the costs are going to be close enough not to worry about anyway, if anything the AC supply will be more.
For powering oscillators, high current output is not required, so ultimate current capability is not an issue. Low noise is crucial, and isolation from the rest of the DAC is crucial as well. Some seem to believe that very low output z is needed, (and I suspect that a LiFePO4 cell meets this need) but, Demian Martin does not seem to think so: see his dedicated oscillator regulator design in the Ian's FIFO thread. I have a lot of respect for Demian, such that I think he probably knows what he is talking about. Additionally, no regulator is going to have low impedance at MHz range clock frequencies anyway, right?
This thread is about powering oscillators, the merits of powering other circuits with batteries is OT here. I feel that qusp has presented an interesting alternative worth considering for powering oscillators.
It appears that you are not reading the details of the posts. No one has ever established, to my knowledge, that any regulator will have a lower output z than a LiFePO4 battery. So far we have seen data only for older battery tech, which is not what qusp proposed.
Additionally, cost is not really a factor, but even if it was, one needs to consider the following costs for an AC based supply: transformer+diodes+smoothing caps+pre-regulator+local regulators of your choice. I am already using this set up, with a dedicated transformer for the oscillators power supply.
A battery requires: a single LiFePO4 cell for each oscillator, battery management, a charger, and a switch and jack for charging. I think the costs are going to be close enough not to worry about anyway, if anything the AC supply will be more.
For powering oscillators, high current output is not required, so ultimate current capability is not an issue. Low noise is crucial, and isolation from the rest of the DAC is crucial as well. Some seem to believe that very low output z is needed, (and I suspect that a LiFePO4 cell meets this need) but, Demian Martin does not seem to think so: see his dedicated oscillator regulator design in the Ian's FIFO thread. I have a lot of respect for Demian, such that I think he probably knows what he is talking about. Additionally, no regulator is going to have low impedance at MHz range clock frequencies anyway, right?
This thread is about powering oscillators, the merits of powering other circuits with batteries is OT here. I feel that qusp has presented an interesting alternative worth considering for powering oscillators.
Last edited:
Hi Barrows
FWIW, I use a +-15V LM317/LM337 supply with my highly modified MF X-DAC V3. It is further smoothed by a JLH PSU addon , which is as you are aware, is a variety of Super Regulator, but without a voltage drop.I then follow this by a +5v version of a similar topology to the PFM "Flea",which uses a highly filtered reference voltage derived from the higher voltage supply, then via an opamp. This supply then goes via the existing ferrite beads and ceramic bypass circuit to the oscillator module. I believe that it's the extremely low noise stable reference, that matters more than just a very low output impedance supply.
Just as changing from a Vanguard 1PPM oscillator module to a .3PPM Vanguard oscillator made a worthwhile audible improvement, so did powering the .3PPM from this opamp circuit give a similar improvement over the same highly filtered supply followed by a 7805. The further audible improvements were especially evident with high resolution material.
Regards
Alex
FWIW, I use a +-15V LM317/LM337 supply with my highly modified MF X-DAC V3. It is further smoothed by a JLH PSU addon , which is as you are aware, is a variety of Super Regulator, but without a voltage drop.I then follow this by a +5v version of a similar topology to the PFM "Flea",which uses a highly filtered reference voltage derived from the higher voltage supply, then via an opamp. This supply then goes via the existing ferrite beads and ceramic bypass circuit to the oscillator module. I believe that it's the extremely low noise stable reference, that matters more than just a very low output impedance supply.
Just as changing from a Vanguard 1PPM oscillator module to a .3PPM Vanguard oscillator made a worthwhile audible improvement, so did powering the .3PPM from this opamp circuit give a similar improvement over the same highly filtered supply followed by a 7805. The further audible improvements were especially evident with high resolution material.
Regards
Alex
I for one read the entire thread.
This is irrelevant because even on a theoretical level, both potentially will have an output impedance that is extremely low.
Neither is a battery, track/trace/wire inductance takes over, it's one of the reasons we use decoupling capacitors.
One advantage of the super regulator vs the battery is that you can use the regulators remote sensing ability to lower the seen impedance of the powered device beyond that of the traces/wires. This is one of the areas where the super regulator actually excels - powering a single device only.
Exactly...
"Neither is a battery, track/trace/wire inductance takes over, it's one of the reasons we use decoupling capacitors."
Exactly my thinking, and why I do not think that super low z out should be the determining factor for an oscillator power supply. My suspicion is that low noise, and complete isolation from the rest of the DAC, are the desirable factors. Seems like a LiFePO4, decoupled by .1uF right at the point of load, could be a very good approach. Minimizing the lead lengths from the battery may be worthwhile as well.
Anyone know if there are any known resonance factors with batteries-leads-caps?
"Neither is a battery, track/trace/wire inductance takes over, it's one of the reasons we use decoupling capacitors."
Exactly my thinking, and why I do not think that super low z out should be the determining factor for an oscillator power supply. My suspicion is that low noise, and complete isolation from the rest of the DAC, are the desirable factors. Seems like a LiFePO4, decoupled by .1uF right at the point of load, could be a very good approach. Minimizing the lead lengths from the battery may be worthwhile as well.
Anyone know if there are any known resonance factors with batteries-leads-caps?
The "6 Moons" review of the Red Wine battery supply said that Zo was 20milliOhms -- one has no idea how they accounted for the inductive portion.
I've read several papers by Chinese cell manufacturers and they specify Zo of 20 to 100milliOhms.
The Jung regulator measures Zo under a micro-Ohm -- this is because the bandwidth of the error amplifier is very high, and Zo is inversely related to bandwidth.
Send me an LiFePO4 and I will measure the Zo from 1Hz to 40MHz free of charge. I'll even send the battery back.
Barrows has hit the nial on the head, PSU choice is secondary to porper decoupling and preferably power planes with interplane capacitance. Though 1uf is not a decoupler but a reservoir cap, ebven a 0.1uf ceramic in a 0402 package is only good for 12MHz. The initial charge come from capacitance within the device, then the board interplane capacitance, then the small decouplers, these are then filled up by the reservoir caps and finaly the PSU fills themup. This is the MAIN function of decoupling capacitors to provide the charge required for a circuit to switch. With todays fast rise time devices, power planes are a must, and planear capacitance is critical and becomes more so the higher the frequency. For clocks I would recomend some very small COG caps(pFs) with 0.1uF X7R's as intermediate reservoir caps, then some 10uF X7R then a few larger elecs or tants.
Using some software similar to this:
http://www.algozen.com/DS_CADSTAR_LT_PowerIntegrityAdvanced_ENG_2011_10_05.pdf is an eye opener, and generaly shows how bad rules of thumb we have all worked to for years are.
Using some software similar to this:
http://www.algozen.com/DS_CADSTAR_LT_PowerIntegrityAdvanced_ENG_2011_10_05.pdf is an eye opener, and generaly shows how bad rules of thumb we have all worked to for years are.
Thanks...
marce. I am using Crystek CCHD oscillators. If one happens to look inside one of these, it is clear that Crystek has supplied final decoupling internally. Must be time to look at the app notes from Crystek on using these parts. I suspect that .1 uF at the power pin may be totally adequate considering the additional onboard decoupling/filtering.
marce. I am using Crystek CCHD oscillators. If one happens to look inside one of these, it is clear that Crystek has supplied final decoupling internally. Must be time to look at the app notes from Crystek on using these parts. I suspect that .1 uF at the power pin may be totally adequate considering the additional onboard decoupling/filtering.
Yes, this is happening more often (Virtex 5s + have 0201 decouplers on the interposters) as people are realising that standard decoupling does not work like we think. Just spent an exciting week in Munich looking at this and SIV (signal integrity) software and discussing power delivery systems and decoupling. it was all digital and Rf based.
My recomendation for the COG comes from some high rel designs (mil) where we did extensive EMC and signal integrity testing. We were using oscillators, and the best results for both signal integrity and EMC were with a very small COG next to the pins, with X7Rs further out. Note that it was only on oscillators that the COG's were beneficial, elsehwere better results were achieved using X7R caps as we could get smaller packages (0201's have half the inductance of 0402's). Using 0402 devices you can halve the inductance by using via in pad, though this can be rather costly for DIY as you have to have them filled and capped.
The excellent RF operation of small value COGs is benefitial for clock decoupling, the smallest package you can find.
My recomendation for the COG comes from some high rel designs (mil) where we did extensive EMC and signal integrity testing. We were using oscillators, and the best results for both signal integrity and EMC were with a very small COG next to the pins, with X7Rs further out. Note that it was only on oscillators that the COG's were beneficial, elsehwere better results were achieved using X7R caps as we could get smaller packages (0201's have half the inductance of 0402's). Using 0402 devices you can halve the inductance by using via in pad, though this can be rather costly for DIY as you have to have them filled and capped.
The excellent RF operation of small value COGs is benefitial for clock decoupling, the smallest package you can find.
how do the latest high performance micas and thin film caps compare vs c0g marce? was there any talk of them? they would seem to be more linear again vs F/S, but as yet the thin film types are not available in large enough values, topping out around 20pf, where the same caps start being mica. have you started playing with embedded thin film types fabricated right into the PCB.
@barrows: I would actually be more likely to use a larger cap or none at all with the Crystek for this reason, or at the least not adding more than is on the PCB already, unless you have access to a scope it would be pretty easy to cause resonance with the internal decoupling. follow the app note I guess, but something to be aware of.
i'm very interested to play with the new DSPLL types, like the Si570 from silicon labs, hopefully Ian becomes motivated in that direction again soon. finding suitably fast 22.1x and 24x parts for sync mode with 8xOSF enabled on es901X is not easy, with NDK being oretty much the only suitable set frequency model you can buy in singles.
@barrows: I would actually be more likely to use a larger cap or none at all with the Crystek for this reason, or at the least not adding more than is on the PCB already, unless you have access to a scope it would be pretty easy to cause resonance with the internal decoupling. follow the app note I guess, but something to be aware of.
i'm very interested to play with the new DSPLL types, like the Si570 from silicon labs, hopefully Ian becomes motivated in that direction again soon. finding suitably fast 22.1x and 24x parts for sync mode with 8xOSF enabled on es901X is not easy, with NDK being oretty much the only suitable set frequency model you can buy in singles.
For decouplingit is the size of the cap that is critical, as soon as the peak in the frequency Vs ipedance curve is passed, the cap becomes inductive and stores energy, not what we want. It is not so much actual capacitor value, but reducing inductance to the smallest possible value, and that means small packages and short leads.
I have done embedded components, actual 0201s in the PCB and planar capacitance using very very thin dielectrics (the problem of doing this is the signal propagation goes down because of Er so everything has to be nearerto cater for this, no free ride with high speed). I have also done 0201s between the pins of BGA components on the same side of the board, so you place the caps first then the BGA's. This combined with your top two layers as power and ground gives the lowest loop area and thus the lowest inductance , with micro vias you have only 0.1-0.05mm of travel for the loop, whereas a standard 1.6mm board with central power and ground has 0.8mm X2 loop area approx to travel through the board.
At the moment this is only a minority of boards where we do these sort of things, and usually where pockets are deep and reliability is paramount, probably over the top for audio, but when I get some spare time!!! I would like to do the ultimate DAC layout using all the tools I have at my disposal, just as a fun exercise. At the moment though I seem to always on site, spent 7 of the last 10 weeks away from home (but I do get to design boards for some pretty impressive kit.
Due to their small size almost all decoupling on comercial designs is performed by X8R, X7R - X5S multi layer caps, the better kit uses X7R's and above, throw away commercial products will use the lower dialectrics for cost. MLCC's X7Rs are the best choice for local decoupling, and with uF values becoming available are starting to replace elecs for the locla bulk caps (1-10uF), plus SMD electrolitics dont survive ballistic shock tests or extreme acceleration tests (150G +).
I have done embedded components, actual 0201s in the PCB and planar capacitance using very very thin dielectrics (the problem of doing this is the signal propagation goes down because of Er so everything has to be nearerto cater for this, no free ride with high speed). I have also done 0201s between the pins of BGA components on the same side of the board, so you place the caps first then the BGA's. This combined with your top two layers as power and ground gives the lowest loop area and thus the lowest inductance , with micro vias you have only 0.1-0.05mm of travel for the loop, whereas a standard 1.6mm board with central power and ground has 0.8mm X2 loop area approx to travel through the board.
At the moment this is only a minority of boards where we do these sort of things, and usually where pockets are deep and reliability is paramount, probably over the top for audio, but when I get some spare time!!! I would like to do the ultimate DAC layout using all the tools I have at my disposal, just as a fun exercise. At the moment though I seem to always on site, spent 7 of the last 10 weeks away from home (but I do get to design boards for some pretty impressive kit.
Due to their small size almost all decoupling on comercial designs is performed by X8R, X7R - X5S multi layer caps, the better kit uses X7R's and above, throw away commercial products will use the lower dialectrics for cost. MLCC's X7Rs are the best choice for local decoupling, and with uF values becoming available are starting to replace elecs for the locla bulk caps (1-10uF), plus SMD electrolitics dont survive ballistic shock tests or extreme acceleration tests (150G +).
Last edited:
Yup...
"@barrows: I would actually be more likely to use a larger cap or none at all with the Crystek for this reason, or at the least not adding more than is on the PCB already, unless you have access to a scope it would be pretty easy to cause resonance with the internal decoupling. follow the app note I guess, but something to be aware of."
I am going to try and have a look at Crystek app notes if they have any. I am dealing with a commercial product, so there is only so much I can do, but adding/changing a single decoupling ceramic at the pin of the oscillator is pretty easy.
marce, hope that you have time to produce a DAC board design sometime, it sounds like you have the skills to really get it right!
"@barrows: I would actually be more likely to use a larger cap or none at all with the Crystek for this reason, or at the least not adding more than is on the PCB already, unless you have access to a scope it would be pretty easy to cause resonance with the internal decoupling. follow the app note I guess, but something to be aware of."
I am going to try and have a look at Crystek app notes if they have any. I am dealing with a commercial product, so there is only so much I can do, but adding/changing a single decoupling ceramic at the pin of the oscillator is pretty easy.
marce, hope that you have time to produce a DAC board design sometime, it sounds like you have the skills to really get it right!
The most important parameter for clock supply is an as low as possible 1/f noise.
You cannot achieve this with a noisy supply and throw (small) decoupling caps at it. This helps for noise higher up in frequency, but at low f not, obviously.
The regulator or the battery has to have very low 1/f noise itself.
You cannot achieve this with a noisy supply and throw (small) decoupling caps at it. This helps for noise higher up in frequency, but at low f not, obviously.
The regulator or the battery has to have very low 1/f noise itself.
If you have a low noise, high output impedance supply, the clock is going to stutter, you're going to get logic indecision. That's just the way the physics works.
Since RS has some LiFePO4 in stock, I will test the Zout with the same apparatus used for the regulator bake-off featured in the Linear Audio V4. From what I've read in manufacturers' data sheets I'm going to guess it will be in the tens of milli-Ohms.
no, you cant just pick any lifepo4, some are pretty ordinary. the one most people are usually talking about is a specific variant made by A123 and the model # is ANR26650M1A or M1B. They have significantly lower impedance, better current delivery and lifetime than most. but watch out, its often faked and sold on the bay, you cant buy direct from the manufacturer and there are only a coupe of distributors that will sell legit new stock direct. if it wasnt so expensive to send it via Auspost as they havent upgraded their policy on polymer cells, I would send you one. perhaps someone over there will?
Last edited:
Well, it depends how you define high output impedance. 10, 100mOhms? And at what frequency?
Btw, as far as I have seen, clocks usually do have quite a constant demand at LF, there will not be much sag even with highish output impedance.
As soon as you go into the MHz, output impedance is mainly depending on the decoupling network. Feedback based regulators behave inductively, because the loop gain will reduce with frequency.
Aha...
Sounds like a case for maybe a Belleson with Bybees on its output! I know, some of you are laughing... but I have a couple of small Bybees here already...
Sounds like a case for maybe a Belleson with Bybees on its output! I know, some of you are laughing... but I have a couple of small Bybees here already...
Bad choice.
Bybees ranked last in our survey of 13 regulators. Belleson is prone to oscillation.
Edit: A123 went bankrupt 3 weeks ago -- with guess who as their largest creditor.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.