You have allowed an overweening (and largely unjustified) concern for HF noise to blind you to the necessity for the PSU filter to reduce the 100Hz ripple which results from the rectification process. You MUST meet this requirement before you turn to any other.
In order to make a significant impact on ripple @ 100Hz, inductors in the range of Henries are required. Look at any tube supply. Have you any idea the dimensions and cost of a 10H choke to pass 2A? For this reason and others resistive series elements are generally employed in low voltage supplies.
You should look at and understand the design of some existing PSUs, before attempting to surpass the efforts of unknown numbers of engineers worldwide over more than a century. We never throw a good idea away.
The putative problem with the filter is parasitics. Stray RLC elements mean that, as frequency rises, capacitors look less like shorts and more like open circuits, and inductors look less like OCs and more like shorts. If, in practice, large Ls and Cs are being employed then smaller elements (with smaller strays, and consequently better frequency response) are employed side-by-side with the existing elements. Thus 10H+0H1, 100u||1u. You understand, 10 Henries in series with 100 millihenries, 100 microfarads in parallel with 1 microfarad? These additions are insignificant at low frequencies, but because of their physical dimensions they continue to work when the larger components stop working, and because the frequency is much higher, they don't need to be so big to achieve an equal reactance.
Series elements must obviously be dimensioned to pass equal currents. Resistors may be required to dissipate considerable heat.
High-speed PCBs will frequently employ decoupling capacitors of 100n||10n or 1n, placed absolutely as close to IC power pins as physically permitted, both to improve performance by reducing the inductive effects which limit frequency response, just as the leads do in the PSU filter, and for EMC reasons, (to reduce the loop area).
For such physical considerations it may ultimately be necessary to build separate filters and cascade them. Ultimate HF noise performance is more likely to depend on the regulator than the PSU filter, however. Radiated or conducted interference are best dealt with as a separate issue, dependent on environment.
In order to make a significant impact on ripple @ 100Hz, inductors in the range of Henries are required. Look at any tube supply. Have you any idea the dimensions and cost of a 10H choke to pass 2A? For this reason and others resistive series elements are generally employed in low voltage supplies.
You should look at and understand the design of some existing PSUs, before attempting to surpass the efforts of unknown numbers of engineers worldwide over more than a century. We never throw a good idea away.
The putative problem with the filter is parasitics. Stray RLC elements mean that, as frequency rises, capacitors look less like shorts and more like open circuits, and inductors look less like OCs and more like shorts. If, in practice, large Ls and Cs are being employed then smaller elements (with smaller strays, and consequently better frequency response) are employed side-by-side with the existing elements. Thus 10H+0H1, 100u||1u. You understand, 10 Henries in series with 100 millihenries, 100 microfarads in parallel with 1 microfarad? These additions are insignificant at low frequencies, but because of their physical dimensions they continue to work when the larger components stop working, and because the frequency is much higher, they don't need to be so big to achieve an equal reactance.
Series elements must obviously be dimensioned to pass equal currents. Resistors may be required to dissipate considerable heat.
High-speed PCBs will frequently employ decoupling capacitors of 100n||10n or 1n, placed absolutely as close to IC power pins as physically permitted, both to improve performance by reducing the inductive effects which limit frequency response, just as the leads do in the PSU filter, and for EMC reasons, (to reduce the loop area).
For such physical considerations it may ultimately be necessary to build separate filters and cascade them. Ultimate HF noise performance is more likely to depend on the regulator than the PSU filter, however. Radiated or conducted interference are best dealt with as a separate issue, dependent on environment.
Yes, the plan is to focus on the lower end of the frequency scale to begin with and once the baseline has been achieved, then look at the upper frequencies. I am going to do some more reading first before I alter the schematic again.
I have taken into consideration that HF decoupling has to be kept as close to the load as possible, which is why I have worked out a way to design the input power cable of the load device, as an integral part of the filter.
Last edited:
Thanks for this, I am going to look over a few previous designs to see what has been achieved in the past, and do some more reading on filter design before I continue.
Looks like faulty reasoning to me. Tube amp supplies are often choke-fed, this is a specific configuration suited to classA operation. However a choke-fed supply is not really a viable solution where the load current isn't known (or is variable). Simply because choke-fed is a common design choice does not mean its the only option for using chokes to reduce ripple.
If you want guidelines on doing high speed PCB design I can provide you with plenty of advice on doing them and how decoupling works etc and on how to design for EMC immunity up into the GHz but is would be better to discuss this on another thread otherwise this one will go all over the place.
Thanks marce, I am not quite at that stage yet, but I definitely will be interested in learning as soon as I have solid baseline schematic completed first.
Using a single PSU to power several digital/analogue circuits will add ground loops adding hum to the system.
Last edited:
It's got nothing to do with choke-fed. Your turn for a red face if you persist.
Well what's the need for a 10H choke then? I have no fear of my face turning red if it means I get to learn something 😀
Lots of electronic circuitry are powered of one main supply with no problems, including audio....system design and engineering can avoid these problems, to just say it will cause ground loops is a bit of an over reaction.
Every time someone suggests a possible problem, it is not addressed with an engineering solution or even some understanding but seen as a nuisance getting in the way.
Because a 150uH inductor has an impedance of less than 1/10 ohm at 100 Hz compared with >6000 ohms for 10H.
What's wrong with this structure?
At low frequency it looks like this (because the inductors don't do anything significant):-
At high frequency it looks like this (because the capacitors don't do anything significant):-
So you might as well replace it with one of these:
For virtually identical performance.
Which means that you've gone from 2 components to 5 components with absolutely no gain other than a warm self-satisfied glow.
A filter is a voltage divider, or cascade of voltage dividers. A single shunt capacitor or series inductor only acts as a voltage divider against existing impedances in the circuit. This means that it is less efficient than a divider with explicitly defined elements, particularly as the number of stages rises, and that its effect is probably unquantifiable.
What's wrong with this structure?
An externally hosted image should be here but it was not working when we last tested it.
At low frequency it looks like this (because the inductors don't do anything significant):-
An externally hosted image should be here but it was not working when we last tested it.
At high frequency it looks like this (because the capacitors don't do anything significant):-
An externally hosted image should be here but it was not working when we last tested it.
So you might as well replace it with one of these:
An externally hosted image should be here but it was not working when we last tested it.
For virtually identical performance.
Which means that you've gone from 2 components to 5 components with absolutely no gain other than a warm self-satisfied glow.
A filter is a voltage divider, or cascade of voltage dividers. A single shunt capacitor or series inductor only acts as a voltage divider against existing impedances in the circuit. This means that it is less efficient than a divider with explicitly defined elements, particularly as the number of stages rises, and that its effect is probably unquantifiable.
Obviously higher inductance presents higher impedance. However I don't see why 10H was introduced in your post - just plucked out of the air because it had >6000ohms impedance? What's the significance of 6000 ohms here?
None of your images show up for me so I won't be commenting on your questions. Perhaps they're all located on the far side of the 'Great Firewall' or something....
Yep, this is fairly basic stuff. No red face so far.
You lost me here though - what's 'explicitly defined elements' ? And how is a filter that's been simulated already 'inquantifiable' in its effect please?
A bit of reading matter, more can be found on Henry Otts site EMC UK etc.
Did you explain the possible mechanisms...no.
In the real world systems in a single box often have ONE main supply and wiring, if you get loops causing noise in this situation you need to do further research and testing, often within a case the noise is caused by some high impedance ground path that has been missed or by over zealous star grounding. As I have said in the real world this is a problem that can be engineered out (or better still not designed in).
Learning about EMC and how to combat it covers this, and in the past there has been many discussions and information put up regarding loops and how to avoid them...and how to distribute power within a box.
As to understanding, I face these problems in my job, which is primarily getting signals around PCBs and systems, often consisting of more that one box and often to much higher levels of EMC immunity than a simple system in a domestic environment such as home audio.
That could be fun then if not done properly...I would isolate the analogue side of the DAC and run the wires from the power supply as closely coupled for as far as possible to minimise loop areas, splitting to the separate boxes only when necessary. Its pretty common practice with some multibox systems such as communications systems.
You could mod a DAC until it resembles Quasimodo. Separate PSU's for analogue, digital and control. Each PSU designed for its own unique purpose. Ground disconnecting circuits to prevent ground loops with other connected equipment.
But that is not what is being "designed" here.
But that is not what is being "designed" here.
abraxalito
This process will require that you employ your imagination and intelligence to understand what I am attempting to explain to you, rather than avoid it.
The first picture shows a shunt C of 4500 microfarads, a series L of 150 microhenries, a shunt C of 4900 microfarads, a series L of 150 microhenries, a shunt C of 10000, as in the
OP's diagram, but single sided.
The second picture shows a shunt C of 19400 microfarads.
The third picture shows a series L of 300 microhenries.
I am sure you can picture the two possible different arrangements of a 19400uF C and a 300uH L depicted in the fourth picture.
Don't worry about the red face, it'll be along when the penny drops.
This process will require that you employ your imagination and intelligence to understand what I am attempting to explain to you, rather than avoid it.
The first picture shows a shunt C of 4500 microfarads, a series L of 150 microhenries, a shunt C of 4900 microfarads, a series L of 150 microhenries, a shunt C of 10000, as in the
OP's diagram, but single sided.
The second picture shows a shunt C of 19400 microfarads.
The third picture shows a series L of 300 microhenries.
I am sure you can picture the two possible different arrangements of a 19400uF C and a 300uH L depicted in the fourth picture.
Don't worry about the red face, it'll be along when the penny drops.
- Status
- Not open for further replies.