Power Supply 101 (Linear Supply)

[Plcamp]

I remember years ago on DIYaudio site, a gent studied power amp bridge rectifier snubber circuits, and concluded the optimum values (for minimal noise) were dependent on transformer parameters. As I recall several common power amp transformers were optimized, including old Haflers.

 

[solderdude]

1A must be sufficient for stuff like pre's,active X-overs.etc.
Yes.
Yes.
The most common around,about 30VA,double windings.
Yes.

(the simpler,the better)

Just build a low-noise dual voltage power supply (according to application notes)
Use an R-core transformer.
Add a common mode filter mounted in an IEC socket.

 

[Sokel]

Pre-made linear supplies are commonly available, unless you just want to build your own.

I do built my own,I have built Salas,generic 15-0-15 (even with double 7815 and not 7915,whatever is in the drawer,etc.

Ok,Salas is really nice but when it comes to 7815-7915 15-0-15 it's nice too the generic way but more experienced people,educated,like here must have a few tips.

 

[Mnyb]

Dangerous if there is not a constant current drawn from them. Best avoided.

Thankyou ! I can understand this , hmm . I've learned something

 

[fpitas]

If you SIM a step driving or loading an inductor followed by a capacitor, very often you get a large ringing waveform. That can be damped, but you have to be aware, and careful that the peak voltage doesn't overvoltage something.

 

[DonH56]

"Standard" linear power supply, with EI transformer, +/- 15V output, LM7815/LM7915 stabilizers in a conventional scheme, has quite excellent S/N parameters and spectrum

View attachment 294792
Red line corresponds to +15V output voltage. 100Hz line is 106dB below. Mains residuals are quickly attenuated. S/N (45kHz) is 101.36dB.


With wider BW
View attachment 294791
Spectrum is clean above 1kHz up to 80kHz.

Note this is representative of a regulated supply typically used for low-level circuits; the original post applies to an unregulated supply typical for (class A/AB) ampliifer output stages. Layout, particularly paying attention to grounding and output decoupling, is important to achieving high rejection of 50/100 and 60/120 Hz noise (but below -80 dB is excellent).

 

[DonH56]

It would be valuable to have an exact (most around the net is almost generic) schematic for a nice 15-0-15 to follow.
I mean the exact value components,layout,etc.

In a different thread, please... There are many ways to do that but it depends on the application as well.

 

[sarumbear]

It's a "debate" between audiophools (or ignorant people) and rational people.


Not a good "permanent" design. You can't guarantee that the current will divide equally. And if one diode fails & shorts out all kinds of bad things happen.


You won't find 100uF filter caps in a power amplifier.

In a "good design" there will be a voltage regulator and besides regulating the voltage it will filter out (most of) the remaining ripple (and it will also help with any other noise that gets-through the transformer).

Have you replied for the sake of replying?

 

[DonH56]

Thanks nice writeup .

The obvious follow up question what about LC or LCL or LCLC type of linear supplies

I reckon they are(where) common in Tube gear of old ( And the LC type is in industrial VSD drives with sometimes some LCL filter at the input too )

Nowadays when building power amps a normal linear supply as you described is deemed "good enough" even in expensive amps and any upgrade is usually a switchmode supply like in the AHB2 ?

Not seen anyone going to the extra trouble off adding heavy and large filter inductors in linear power supplies in a long time.
The cynic in me tells that its not an understood and marketable feature , moare microfarads is

Again, this started addressing power amplifiers and their output stages, but fundamentally was just to explain the operation of a basic supply, not provide design details nor address regulation which I considered beyond a "101" level for the average audiophile.

I have rarely seen LC filters in linear supplies outside tube circuits these days. They are pretty much mandatory for SMPS' but can be smaller since L and C values can be much smaller due to the high switching frequency. Not being a power supply expert in my (former) day job I have not put together anything on SMPS designs, though I spent a lot of time developing output filter networks for (after) them. The principles of controlling resonances in an LC circuit are the same and something that must be carefully considered when implementing an LC filter. For example, here is a plot of an LC filter showing a supply output with no snubber (RC damping circuit) and two different schemes for suppressing the low-frequency peak that can increase noise and potentially lead to an unstable regulator. This shows the danger in just throwing together components without considering the result.

Large capacitors cost money so I suspect part of the issue is simply cost-cutting, particularly with the reduced FTC testing requirements reducing the need for continuous power delivery. It is also important to note that large current spikes cause large voltage spikes across inductors. Handwaving, inductors resist current changes; capacitors resist voltage changes. Since most amplifiers are not pure class A designs with steady (constant) current flow from the supply, capacitors provide better decoupling and filtering since the goal is to provide constant voltage to the circuit no matter the current demands.

 

[Multicore]

Funny how we call that linear. At the heart of it is the diode bridge rectifier!

 

[fpitas]

Funny how we call that linear. At the heart of it is the diode bridge rectifier!

Right? Which us RF guys use as a mixer sometimes.

 

[DonH56]

Funny how we call that linear. At the heart of it is the diode bridge rectifier!

I am trying to remember where the "linear power supply" term came from. It is not something I or my peers used "growing up". I do not know, but suspect it was meant to distinguish from SMPS designs which have been labeled suspect by all "true" audiophiles. Since I am not a power supply guru (not ignorant since it was part of any design I ever did) perhaps somebody who is can answer.

I'm still debating if the first g in "giga" is hard or soft...

 

[fpitas]

I am trying to remember where the "linear power supply" term came from. It is not something I or my peers used "growing up". I do not know, but suspect it was meant to distinguish from SMPS designs which have been labeled suspect by all "true" audiophiles. Since I am not a power supply guru (not ignorant since it was part of any design I ever did) perhaps somebody who is can answer.

I'm still debating if the first g in "giga" is hard or soft...

I've always heard it pronounced like the g in good. No idea why a "linear" supply is so-called, though.

 

[DonH56]

I've always heard it pronounced like the g in good.

That is the way I pronounce it, but I had coworkers who were adamant it was the other way, like gee.

No idea why a "linear" supply is so-called, though.

I have speculated over the years but don't have a good answer. I read somewhere (don't ask me where) that it was used to distinguish passive (diode-based) from active (switching transistor-based) rectifier supplies, but either involves a nonlinear circuit element. Is a tube rectifier an active or passive device? Too old-school for me, I guess.

 

[DonH56]

Promoted to home page. Thanks for porting over this article.

Thanks Amir. I cleaned it up a little but otherwise just plopped it down. I had started a more in-depth version but got busy doing other stuff (couple of performances over the weekend, getting ready for an off-road trip next weekend, power supplies dropped low in priority ).

 

[pma]

Note this is representative of a regulated supply typically used for low-level circuits; the original post applies to an unregulated supply typical for (class A/AB) ampliifer output stages. Layout, particularly paying attention to grounding and output decoupling, is important to achieving high rejection of 50/100 and 60/120 Hz noise (but below -80 dB is excellent).

Understood, just tell me .

Below is the spectrum from 2 x 27Vdc/10A "linear" power supply, center tapped diode bridge, 2 x 14100uF capacitor bank, 5.75A rail load current, +27V rail spectrum. In my opinion, we can find many textbook examples on the Internet, but really a few real-world examples and measurements.

 

[DonH56]

Understood, just tell me .

Below is the spectrum from 2 x 27Vdc/10A "linear" power supply, center tapped diode bridge, 2 x 14100uF capacitor bank, 5.75A rail load current, +27V rail spectrum. In my opinion, we can find many textbook examples on the Internet, but really a few real-world examples and measurements.

View attachment 295079

Thank you! And now I realize my original plots were linear in magnitude, as I was focused on the actual ripple current and voltage, whereas dB would make it easier to see. Blah. I don't have the simulator I used installed any more so I'll defer to your plots. I have plots from SMPS supplies from work, but they won't let me publish them, and nothing from "linear" supplies (maybe in a box in the black hole of a basement).

 

[Multicore]

I've always heard it pronounced like the g in good. No idea why a "linear" supply is so-called, though.

I have heard it pronouced like the drink mixer's jigger. Not for a long time, though. And it might have been with respect to Watts rather than bytes.

 

[theREALdotnet]

I'm still debating if the first g in "giga" is hard or soft...

It’s hard, unless we’re talking Watts and the value is exactly 1.21.

 

[pma]

Thank you! And now I realize my original plots were linear in magnitude, as I was focused on the actual ripple current and voltage, whereas dB would make it easier to see. Blah. I don't have the simulator I used installed any more so I'll defer to your plots. I have plots from SMPS supplies from work, but they won't let me publish them, and nothing from "linear" supplies (maybe in a box in the black hole of a basement).

I have repeated the measurement for you in linear Y-axis scale, directly in Vpeak, to read the ripple component. We lose all the resolution then, however.

Ripple in time domain