LM317-based PSU design questions

[jaygunn]

I am looking at creating a fairly standard bipolar IC-regulated PSU with very good ripple rejection. Of course schematics for these are readily available, e.g.:

http://www.jlmaudio.com/ACDCVer2sch.pdf
http://www.tnt-audio.com/clinica/regulators_noise2_e.html
http://www.beatbazar.com/guests/ssltech/kps-1/index.htm
http://web.telia.com/~u31617586/#48 volt phantom power supply

This one is edited from the one on JLM Audio's site. I have a couple of specific questions I'd appreciate help with.

Q1: Do we really need major ripple reduction on the INPUT to the LM317? I have seen as much as 6,600uF (where C5 and C6 sit in this schematic), and as little as 0.1uF. When I read TI's datasheet for the LM317 it specs ripple as a function of temp, output current and output voltage, but NOT as a function of input ripple.
The reason I care is that very big values of C5/C6 puts a big demand on the bridge diodes and I am thinking of using FRED diodes there instead of the usual IN4007.

Q2: One of the stated benefits of regulation is low PSU impedance. I am tempted to add one or more L-C filters after the PSU, like the following, for extra cleanliness. The question is whether the added series resistances of these filters greatly increases PSU impedance and do I care?

Suppose I put two L-C stages with 10R resistors and 1000uF caps after the
regulator output, maybe bypassed with 1uF tants. Any problems? Maybe slow ramp-up (which seems like a benefit to me) or huge current through the regulator (which might be fatal?),

This PSU would be used in a separate chassis to supply DC to 4 channels of mic pre and 4 channels of compression (2 pico, 2 Forssell opto). I'll provide +-24VDC and +-18VDC from one PSU and +-15VDC and +-12VDC from another in the same box, using 2 perf-boarded PSUs of the same general design. So load will vary according to which devices are in use.

Q3: Is it important whether the optional extra L-C filtering is physically placed on the PSU board in its chassis or on the preamp or compressor board in its separate box? I.e. close to the PSU or close to the ICs that are using the power? Of course all ICs in the preamps and compressors have local bypass caps that can help supply instantaneous demands for extra current.

John

 

[owel]

I'd take a stab at this.

1. Do we really need major ripple reduction on the INPUT to the LM317? I have seen as much as 6,600uF (where C5 and C6 sit in this schematic), and as little as 0.1uF.

Are you looking at audiophile PSU designs?

I only use a 1000u/50V and a 1000u/80V (for the 48V section) paralleled with a 0.1u/100V NPO COG. This is enough. You don't need 6800uf.

I am tempted to add one or more L-C filters after the PSU, like the following, for extra cleanliness.

I don't think LC filters are necessary. The LM317 is good enough.

On my PSU, I only have a 1uf/50V (paralleled with a 0.1uf) on the outputs of the regulator.

For the 48V section, I did add a single 1000uf/80V at the regulator output.

From my testing, some phantom powered mics exhibit hum (while some don't). Adding an extra 1000uf at the 48V output fixed that hum problem that some condenser mics exhibit.


What is more important though is that C12 and C13 in the schematic you posted. Don't leave those out.

Of course all ICs in the preamps and compressors have local bypass caps that can help supply instantaneous demands for extra current.

I don't think bypass caps help with current demands. But they do help shunting high-freq. PSU noise in your circuit. Stick them as close as possible to the power pins of your opamp.

 

[Ptownkid]

Yeah, i've built versions of the green pre psu stripped down to a total of 6 caps total and had next to no noise at all on the output.

 

[burdij]

I don't think you want to hang all of that extra capacitance off the output of the regulator chips. These regulators are servos that require a fairly quick response to changes on the output pins. If you want to suppress ripple more, put more capacitance before the regulators. 1000uf or 2000uf really isn't enough. Typically I use at least 6800uf on each polarity.

The secret to good regulator performance is to make sure you have enough margin between the input voltage from the reservoir caps and the voltage output you want. The performance and rejection of AC ripple degrades rapidly if the margin is too small. Consult the datasheet of the part for the minimum values. The LM337 negative part has different (and less desirable) characteristics than the LM317 positive part, too.

 

[pstamler]

Making the input filter a 2-stage (C-R-C) will let you whack down the ripple to negligible proportions without stressing out the diodes; make the second cap the bigger one. It will, however, require a slightly higher transformer voltage, and will generate some heat in the resistor.

Peace,
Paul

 

[jaygunn]

So Paul and Burji, you're going for ripple reduction before the LM3x7 with the C-R-C filter...Paul is managing the in-rush problem by making the 2nd cap the bigger, hungrier one...got it

Then on the output of the reg what size cap do you suggest? (Corresponding to C11 on the upper schematic)

And in the preamp or compressor (in a separate box), would you suggest an R-C on the voltage rails? or just a biggish electrloytic?

Note to all: Using the forum Search function I noted a spotted by PRR that a good preamp design tolerates a certain amount of junk on the rails and still sounds clean and pure. So maybe the search for really clean power is a waste of time. OTOH I am building this equipment for live and studio recording in India where the power is really unstable, so I have been thinking the more filtering the better...

 

[PRR]

> Do we really need major ripple reduction on the INPUT to the LM317?

On any non-trivial DC design, you want low ripple.

The 317 can't raise voltage, not even for an instant, only lower it. So the valley of the ripple must be well above the output.

The heat in the 317 is more about the peak of the ripple. Unless load is small, you don't want high peaks.

This suggests input ripple in the 10%-5% range. 30% won't do.

> very good ripple rejection.

When you add ripple valley, wall-voltage variation, the poorer performance as you approach drop-out, and your ripple phobia... big cap is your friend.

Since you have not specified your input AC voltage or your load DC current, design can't start yet.

But as a wild-man guess: eight chip-boxes isn't going to pull huge power. 1,000uFd per Ampere of load is a good first-pencil number. Your load is probably way under 1 Amp. A 1,000uFd cap is too cheap, in DIY terms, to sharpen a pencil and see if 470uFd would do.... $2 brain-strain to save $0.20 of cap, just use 1,000uFd. 'scope the raw DC at full load and low wall voltage, see if you get at all close to dropout voltage.

> very big values of C5/C6 puts a big demand on the bridge diodes

Once you get your ripple down to not-huge, diode stresses do NOT rise a lot. The peak current is then more about your transformer (winding resistance), not your cap.

I just modeled 20VAC 100 ohm load, nominal 240mA with enough voltage to regulate to 15V or 18V. 100uFd gives 0.75A peak 0.23A RMS in the rectifier but just 13V at ripple valleys, which won't work. 1,000uFd gives 21.6Vmin 1.7A peak 0.4A RMS in rectifier. 6,000u gives 22.7Vmin 1.8A peak 0.4A RMS. Note that a 6:1 increase of cap gives only 6% increase of rectifier strain.

> using FRED diodes there instead of the usual IN4007

If you think you can't hear your ripple, can you really hear your diodes?

> a good preamp design tolerates a certain amount of junk

To me, that is just "obvious". If the core circuit won't eat dirt, it should have local filtering OR this over-fussy fault should be loudly noted to the power supply designer. A triode tube has ~~6dB ripple rejection. Its rail must be 60dB-100dB cleaner than its nominal output signal. Rail filtering is part of designing a tube amp. Either amp and PSU are designed together, or the amp-chassis has ample local filtering so it will eat whatever crap the PSU designer allows, AND whatever crap cross-talks from other circuits on the same PSU. Heavily loaded AB chips throw half-wave audio at the rails. Your opto-limiter is liable to thump the PSU unless this was considered in its design.

> perf-boarded PSUs

Then stop thinking too much and just do it. This is DIY. If it doesn't work, fix it. If you think it could be sexier, fix it. If it blows up, fix it. If you just aren't sure, tack it together all spread out and try it. You don't have to nail it down before you begin production.

To think outside the box: if your wall voltage is steady, and your preamps are not fussy about exact voltage, find a sale on 24VCT 0.3A iron, use 1A bridge, four 1,000uFd caps, and a couple 220R resistors, for EACH preamp.

 

[PRR]

http://tangentsoft.net/audio/
http://tangentsoft.net/elec/
http://tangentsoft.net/elec/ps-est.html

 

[bcarso]

EDIT (hit send before seeing PRR's contribution)


[quote author="jaygunn"]Q1: Do we really need major ripple reduction on the INPUT to the LM317? I have seen as much as 6,600uF (where C5 and C6 sit in this schematic), and as little as 0.1uF. When I read TI's datasheet for the LM317 it specs ripple as a function of temp, output current and output voltage, but NOT as a function of input ripple.
John[/quote]

(my italics)

Beg to differ: on page 3 of the TI spec* I just pulled up, it specifies ripple rejection for 10V out and 120Hz, for two different values of the Cadj cap (0 and 10uF). It is not terribly good but often adequate for equipment with decent power supply rejection. Note that they say 120Hz, and they may actually mean some typical sawtoothish charging ripple, which has energy at higher harmonics. Since the regulator is a typical feedback system its rejection of the higher harmonics may be falling off a bit, hence passing more crud to the output compared to how it would do with a 120Hz sinusoid.

When the manufacturer says as little as 0.1uF, they are talking about providing a local low impedance at high frequencies at the regulator input pin relative to common. Often one sees "required if the part is x inches or more away from the main bulk capacitor". Obviously you need the bulk cap or your regulator will be putting out pulsed d.c.---unless you are using three-phase a.c. :razz:

The various suggestions made are mostly good ones. I'd add a couple: consider putting some R in series with the transformer outputs to limit peak charging currents a bit, unless there is high winding resistance already. Like Stamler's two-section filtering strategy and warning, you will need a slightly higher voltage transformer.

The second comment is that most of the extraneous ripple and less-than-spec regulation arise from poor layout, when charging currents and load currents are flowing through traces shared by the reference terminal components. It's amazing how many designs miss this---"Oh, we'll just use a big ground plane", etc.


*http://focus.ti.com/lit/ds/symlink/lm317.pdf

 

[jaygunn]

I really appreciate all these points and my eventual design will benefit a lot from this discussion.

Note that a 6:1 increase of cap gives only 6% increase of rectifier strain.

Very surprised at this and it's good news since it gives me permission to overbuild a bit.

If you think you can't hear your ripple, can you really hear your diodes?

Good point! Months ago I talked with a friend who's both an electrical engineer and an extreme "high end" enthusiast (tho' those seem contradictory). He made a claim that interested me and got me onto this FRED issue. He said that diode noise reflects back onto the line through the transformer primary, which then affects the power that every unit in the studio sees. He speculated that one of the reasons people like the sound of tube equipment is that if they use tube rectifiers, this noise factor is greatly reduced.

But I think I'll use 1n4007 diodes and just bypass them as a tip of this hat in his direction.

stop thinking too much and just do it. This is DIY.

Totally agree and will get started ASAP, I just wanted to be able to get into the ballpark with component placement and sizes.

(317 datasheet) specifies ripple rejection for 10V out and 120Hz, for two different values of the Cadj cap (0 and 10uF).

Agreed the use of the Cadj cap is essential. In this article its benefit is shown in 3rd graph (without) and 4th (with):

http://www.tnt-audio.com/clinica/regulators_noise2_e.html

Thanks again for your help.

 

[PRR]

> Very surprised at this

Think about it.

Ideal source, ideal diode, ideal cap, any load: the diode current must go to infinity for some short period.

In the real world there is always resistance everywhere. The peak current is approximately the amount of voltage sag divided by the series resistance. Same load, bigger cap, less voltage sag.

As you come up from 70% ripple to 20% ripple, yes the diode strain increases. For the same power, a welder (tolerates ripple) needs less diode than an audio amplifier.

But as you go 20% 5% 2% ripple, the diode strain hardly increases.

And in most cases, transformer winding resistance dominates the budget.

You base a common-sense blunt-pencil design on the transformer, not the cap. A "1 Amp" winding with 10% sag can deliver 10 Amps when sagged to zero voltage. That almost happens at start-up. But any sane value of cap will charge-up pretty quick. Often the first cycle brings it to 90% of final voltage. Yes, the diode must pass 10A for that cycle, but if you look close most power rectifiers have a One Cycle Surge rating around 10 times the nominal rating. It's like they know about transformers.

So base your diode rating comfortably above your transformer rating. You have a 100mA load, a 1A tranny is the smallest you have, you must use a 1 A rectifier. At the 1A level this is no big deal. But if a 10A transformer happens to be available, you better use a 10A rectifier (or add a bit of series resistance) even for minimal load. Once up to voltage a smaller diode may be fine, but that first 200A Chunnnggg.... may blow a 1N4007 before it comes up to voltage.

> diode noise reflects back onto the line

Could be. If you have ever done carrier current radio, you know how hard it can be to get a signal over power lines to where you want it, but like WiFi it is goofy and sometimes you do get great reception in odd places.

 

[pstamler]

[quote author="jaygunn"]So Paul and Burji, you're going for ripple reduction before the LM3x7 with the C-R-C filter...Paul is managing the in-rush problem by making the 2nd cap the bigger, hungrier one...got it

Then on the output of the reg what size cap do you suggest? (Corresponding to C11 on the upper schematic)[/quote]

I usually use 100uF.

And in the preamp or compressor (in a separate box), would you suggest an R-C on the voltage rails? or just a biggish electrloytic?

Biggish electrolytic. But I usually do it differently: since I use 2-stage filters, I put the diodes and the first filter stage in the external box, the second filter stage and the regulators in the main box with the audio circuits. That keeps the connections from the regulators to the audio boards nice and short. The biggish electrolytic is on the audio boards.

If your power is really unstable, design the boards assuming that the AC coming out of the wall is 20% lower than nominal. You'll need some extra heatsinking, though, for the days when the AC is 20% higher than nominal.

Peace,
Paul