Calculating adjust pin capacitance on lm1086 - help!

[hacker]

I'm looking at doing this new PSU of mine properly, and instead of picking random values for components, I'm going to try and do it properly. So, I've got the LM1086 datasheet out (I'm using those for the built-in pre-regulation) and am looking at the calculations necessary to get the best performance from the regs.

Basic facts:

- 25-0-25 c/t trafo -> dual schottky -> 10000uF kendeil (referred to as Cin)
- Input to lm1086 is therefore 36-37v
- R1 = 100R
- R2 = 2k49
- Adj pin tant = 10uf (referred to as Cadj)
- Output decoupling = 200uF (2x 100uF) Rubycon ZA (referred to as Cout)

The datasheet gives this formula for determining if the value of Cadj is suitable:

1/(2pi * ripple_frequency * Cadj) <= R1

For me, that gives:

1/(6.28 * 100 * 0.00001) = 159

Therefore, Cadj is not big enough. If I was to use a 22uF tant for Cadj, we get:

1/(6.28 * 100 * 0.000022) = 72.3

That's better...I think. Is my math ok, or have I done a brain fart somewhere? The other thing that worries me now is the low-Z of the Rubycon ZA caps on the output of the LM1086... From the datasheet:

Stability consideration primarily concern the phase response of the feedback loop. In order for stable operation, the loop must maintain negative feedback. The LM1086 requires a certain amount series resistance with capacitive loads. This series resistance introduces a zero within the loop to increase phase margin and thus increase stability. The equivalent series resistance (ESR) of solid tantalum or aluminum electrolytic capacitors is used to provide the appropriate zero (approximately 500 kHz).

It sounds from that paragraph like low-Z isn't wanted, but from here:

It is desirable to have large output capacitance for applications that entail large changes in load current (microprocessors for example). The higher the capacitance, the larger the available charge per demand. It is also desirable to provide low ESR to reduce the change in output voltage:

ΔV = ΔI x ESR

It is common practice to use several tantalum and ceramic capacitors in parallel to reduce this change in the output voltage by reducing the overall ESR. Output capacitance can be increased indefinitely to improve
transient response and stability.

So, what's the right answer???

Carl

 

[martin clark]

Quick answer - there is usually a range of ESR which guarantees stability. Have a read through the PDF linked here:
http://focus.ti.com/analog/docs/techdocsabstract.tsp?familyId=400&abstractName=slva115

Paralleled caps are interesting. Although the composite will behave like it has vanishingly-small ESR overall, the individual caps - esp. if there's a tantalum or regular Electrolytic in there - may have significant ESR. These will then tend to 'snub' or damp the rest. Plenty on this in appendices to Jim Williams' excellent Linear Tech App Note AN47 ('high speed amplifier techniques' or somesuch)

As a ballpark though, it's hard to go wrong with something like the 10-22uF you've calculated!

 

[hacker]

Martin, that link you provided is perfect. 200uF of ultra-low ESR capacitance such as that provided by the ZAs is perfect and gives the best of both worlds for the reg: high capacitance/low ESR for excellent transient delivery.

All I need now is a set of 22u tants for Cadj (I assume low-voltage tants such as 6.3v will be fine if the 1086 develops 1.25v between adjust/output pins) and Bob's my uncle.

Thanks,
Carl

 

[martin clark]

assume low-voltage tants such as 6.3v will be fine

Maybe not! The tant sees the voltage across R2, which is full output voltage less 1.25V. Watch it, overvolted tants explode & burn quite nicely

 

[hacker]

Maybe not! The tant sees the voltage across R2, which is full output voltage less 1.25V. Watch it, overvolted tants explode & burn quite nicely

Yikes! My bad... that could've been interesting! I'll get some that are rated at 35v.

Cheers,
Carl

 

[teddy_pardo]

Carl,

Cadj is between the adj pin and the ground, not between the out and adj !!!

You'll need more than 6.3v. BTW, AFAIK the type of Cadj capacitor is not critical. For the output, the data-sheet recommends 22uF tant. I bought a pack of 50 on eBay.

Don't forget that tants have polarity too!!!

Teddy

 

[hacker]

Carl,
AFAIK the type of Cadj capacitor is not critical. For the output, the data-sheet recommends 22uF tant. I bought a pack of 50 on eBay.

I read that, yes. It also says that capacitance can be increased indefinitely to improve transient response and that the higher the capacitance, the lower ESR can be without leading to instability. It would appear that this is a very good thing and will hopefully improve the performance of the preamp-based superregs downstream from the psu-based preregulators.

The last thing to do is get some low ESR 1000uf caps inside the pre itself, as I remember that had a tremendous affect on sound quality in a previous power setup.

 

[martin clark]

AFAIK the type of Cadj capacitor is not critical

No, it's not but (theoretically at least) bigger values give better LF regulation.

The reason the bypass cap works at all is that above DC, the AC error current through the divider effectively 'sees' the impedance to ground on this pin. Bypassing R2 with a cap means the AC is shunted to ground directly, and the adjust pin only 'sees' a very small AC voltage - this is where the improvement in ripple rejection comes from. This isalso why using a bypassed zener or LM339 'pedestal' can work very well too - low dynamic impedance.

For most purposes 10uF is plenty as Cbypass, but for really quiet things eg the 5v feed to the 74HC04 clock in the cd players (where low frequency noise matters) Naim go so far as to use 100uF tants. Now you know why

 

[hacker]

For most purposes 10uF is plenty as Cbypass, but for really quiet things eg the 5v feed to the 74HC04 clock in the cd players (where low frequency noise matters) Naim go so far as to use 100uF tants. Now you know why

That's interesting... Now that we know the advantage of using larger caps for Cadj, is there a disadvantage of using large Cadj values? If not, the immediate question is: why not always use a large Cadj cap if it's effective at reducing noise?

 

[martin clark]

Cost, space availability - things that matter from a manufacturing POV really. For homebrew - please yourself, though I think you'd be hard-pressed to hear or measure a difference beyond the normal 10-22uF. Really big caps are electrolytic, and that means leakage noise - so there's probably a point of diminishing returns. FWIW, I've never noticed anything worth having past 47-100uF, beyond which the benefit dissapears under the reg's inherent output noise.

If what you want is seriously low noise, there are much better solutions of course...

 

[Mr Tibbs]

That's interesting... Now that we know the advantage of using larger caps for Cadj, is there a disadvantage of using large Cadj values? If not, the immediate question is: why not always use a large Cadj cap if it's effective at reducing noise?

Somewhere on the datasheet you might find mention of Cadj value being related to how much C is across the output of the reg. A 22u Tant being needed (for better stability) once the output cap is above a certain value. Have another look and see if there is mention of that. I don't think you can go wrong with 22u, the only problem is they cost a bit more!

Mr Tibbs

 

[hacker]

I don't think you can go wrong with 22u, the only problem is they cost a bit more!

By jingo, you're not wrong! £7.68 for one 22uf/35v tant from Farnell!!! Looks like I'll be sticking with 10u... although I could always parallel 2 of them...

Edit: RS are slightly more sane at £1.48 per tant... you just gotta buy 10 at a time

 

[Mr Tibbs]

They're approx £1.30 from RS.

You can parallel two 10u.

Mr Tibbs

 

[teddy_pardo]

I bought these on eBay. Much cheaper and look OK.

 

[bivalve]

Bypassing R2 with a cap means the AC is shunted to ground directly, and the adjust pin only 'sees' a very small AC voltage - this is where the improvement in ripple rejection comes from.

Martin,

"I think that the bypass cap causes the adj pin to see ALL the AC error voltage, that is relative to the output pin. Otherwise it would be reduced by the R1 R2 divider and become less effective with higher output voltages."

The above quoted text is my original post, but I am a bit uneasy about it. My apologies Martin. Cbypass is also filtering the AC noise voltage of the internal reference. Perhaps this is its primary role and the improved ripple rejection comes about from a more stable and less noisy reference. It seems to me that although the reference DC voltage is developed across the top resistor in the divider, the AC error voltage must appear here as well.

Interestingly I think Cbypass is analogous to C7 in the superreg, but there the consensus is that the SR sounds better without it.

ALW originally recommended ZA caps for his SR in the prereg output position, but I gather they gave trouble and it was an oscillation issue.

David