Q: ultra clean power supply [GGG]

[km-r]

hey!

im planning to build the ultra clean psu. got some questions though...

why is the transformer rated at 25vac [35v dc rectified  ]
isnt this overkill for a 9v supply?
aint that gonna make a ic regulator dissipate lots of juice?
im going to use 5% carbon comp resistors... is that fine?

thanks!

[tackleberry]

The ultra cleans I have seen use cascaded regulators from 24v down to 12v or 9v. First regulator is 15v and reduces hum by 50db then next even more etc. You could get a transformer for 12v or 9v but doesnt leave much room if any for the regulators to clean the signal.

[R.G.]

im planning to build the ultra clean psu.

I've always internally objected to the name of that thing. "Ultra clean" it's not. It's just a regulated power supply, no special tricks to get cleaner than it would otherwise be. That's not to say it's bad, just misnamed IMHO. It's like a company naming itself "Guaranteed Good" and then not having any guarantees on its products. You assume things from the name that are not necessarily there.

why is the transformer rated at 25vac [35v dc rectified  ]

Unknown, Captain. It's been that way since that schematic was posted to the net. Presumably it's to get a lot of headroom or use a standard transformer, or because that's what the first builder had handy. It's not necessary. A minimum of about 12-13Vdc is needed over all power supply changes: Low line, ripple sag, droop from loading, etc. Which leads us to the discussion of tolerance, which we'll have in a minute.

isnt this overkill for a 9v supply?

IMHO, yes.

aint that gonna make a ic regulator dissipate lots of juice?

The regulator will dissipate much more this way than otherwise, yes. Maybe the original designer thought it was cleaner somehow.

im going to use 5% carbon comp resistors... is that fine?

Maybe. Let's talk about tolerances.

Outside the quantum physics world, there are no exact quantities. Even inside the quantum mechanics world, Heisenberg's uncertainty principle applies and you can never know both a particle's position or energy at the same time. Measuring one disturbs the other. In the macroscopic world, we just can't measure that finely. A lot of engineering is to know how to measure as closely as possible, then to specify how to put things together so the tolerances are acceptable.

The output voltage of an LM317 is specified as Vout = Vref*(1+R2/R1)+Iadj*R2.

Hmmm... If you're using +/-5% resistors, then both R2/R1 and Iadj*R2 will vary with the resistor tolerances. According to the datasheet, so will Vref and Iadj. The output voltage will be biggest when Vref, R2 and Iadj are their max values and R1 is minimum. Likewise, it will be smallest for Vref, R2 and Iadj being at the small end of their range and R1 being biggest.

Let's play numbers. For the biggest output voltage, Vout = 1.30*(1+(1528.8/228.95) +(100uA*1528. = 9.9806 + .15288 =10.133V
For the smallest output voltage, Vout = 1.20*(1+1383.2/253.05)+(50uA*1383.2) = 7.759 + 0.06916 = 7.828V

Huh? I thought it was supposed to be 9V. Well, it is. It's 9V +12.6%, - 13%. What happened? I deliberately selected the worst case values for the resistors and the chip parameters. The biggest determiner of the output voltage is the ratio of R2 and R1, and those could be one high and one low. I also stuck in the min and max values from the datasheet for Vref and Iadj. Ugly, huh? 

That's not even all. The LM317 also lists a tolerance for the line regulation (i.e. how much the output varies as the input voltage changes; should be 0 but isn't) and load regulation (how much the output drops as you load it)  that would make the real numbers even worse.

The whole area of worst case design is what you have to do for things that simply have to work, no matter what parts you put in. At least it tells you how bad it can be.

While we're on tolerances and datasheets; the datasheet tells us that the ripple rejection will be a minimum of 66db.  It would "typically" be 80db, they say. What that means is "It COULD be as low as this much on a good day." Don't ever stake your job on a "typical" number. Notice that the datasheet does not tell us the input-output differential voltage, only a graph of typical values. It could be as high as 2.25V at 1.5A output, or higher. So you're going to need at least your output voltage plus 2.25V at the bottom of the raw power supply ripple or the ripple voltage will come though. That's 10.133 + 2.25 = 12.383V if your resistors add up wrong. If you used a transformer that actually put out 11Vac under load, the DC output would be 11*1.414 = 15.55V, minus two diode drops of about 0.7V each, and that means that you could stand 14.15V-12.383 = 1.767V of ripple on the DC power supply. That's about 4.527mF (4527uF) per amp of DC output to keep the ripple down.

But there are tolerances on the transformer voltage, the diodes forward voltage and +12/-10% on the AC power line itself.

By the time you get through with the evaluations, you probably need a 14-15V transformer. But 25Vac is excessive.

Remember tolerances. You'll save yourself some surprises. 

[km-r]

thanks for the reply guys, especially mr RG!

i have to believe that this regulator is quieter than just one 78xx ic regulator because IIRC the LM317 is a shunt regulator blah blah...
and 78xx regulators are zener based? ive also seen designs of "Zero-ripple power supplies" mehhh...

so with the math done by mr RG, i can say that the 1% component tolerances are recommended so that the output voltage will be very close to the original build...

still i will build this PSU with a lower rating transformer.

with mr RG's explanation...
i will be led to the option of using trimmers to adjust the output voltage and seal it with epoxy.
and now is there a threat of voltage output drift due to temperature?
would this significantly alter the voltage in the practical world?

[R.G.]

i have to believe that this regulator is quieter than just one 78xx ic regulator because IIRC the LM317 is a shunt regulator blah blah...
and 78xx regulators are zener based? ive also seen designs of "Zero-ripple power supplies"

Actually, no, the LM317 is a series regulator, not a shunt regulator. I has an internal band-gap reference of nominally 1.25V. Looked at another way, it forces the voltage across R1 to be 1.25V by cutting back on current through it until this is true. Since all of the current in R1 must go through R2 except the Iadj current, which is very small, then the voltage across R2 is the regulated current through R1 times R2. And the output voltage must be the sum of those two.

Zener diodes are shunt regulators.

Zeners are noisy when starved for current. They can be very quiet when run at higher currents and shunted by capacitors.

The 78xx family may be zener based, but both they and the LM317 are specified for maximum noise in their datasheets and the numbers are very similar.

Zero ripple power supplies use some kind of circuit to feed some ripple forward to cancel the remaining output ripple. It's touchy and not worth the effort IMHO, when you're reducing ripple by 6 db (factor of 2000).

so with the math done by mr RG, i can say that the 1% component tolerances are recommended so that the output voltage will be very close to the original build...

Good idea.

still i will build this PSU with a lower rating transformer.

Good idea.

i will be led to the option of using trimmers to adjust the output voltage and seal it with epoxy.

With 1% resistors, probably not necessary.

and now is there a threat of voltage output drift due to temperature?

It's specified in the data sheet if you want to calculate it. But temp drift of the regulator reference is a fraction of 1% as I recall.

would this significantly alter the voltage in the practical world?

No.