Problem with PSU module design that shows severe noise at full load

Thread Starter

bloguetronica

Joined Apr 27, 2007
1,453
Hi all,

I'm having a problem with the attached circuit. It was meant to be a front-end for a power supply. Basically, the output voltage is controlled via SPI. It works, but the noise from the pre-regulator DC-DC circuit passes almost unharmed through the pass element on the output stage. Any suggestions on how to improve the PSRR of the output stage?

For now, I only have a beautiful piece of modern art. At least, I can control the output voltage down to the mV, but the output noise makes it useless.

Kind regards, Samuel Lourenço
 

Attachments

Last edited:

Papabravo

Joined Feb 24, 2006
16,498
I don't think that is noise. It is how a switch mode power supply output looks. What were you expecting?
Is the scope telling you that there is 76 mV P-P of ripple on the output?
So a bit less than 1% ripple on the 8 VDC output. That is actually not too bad.
You could try a pi-filter on the output if you are concerned. Maybe a 3rd order Butterworh with a corner at 30 kHz (10% of your switching frequency)
 
Last edited:

Thread Starter

bloguetronica

Joined Apr 27, 2007
1,453
I chose the wrong words, sorry. Indeed it is ripple coming from the DC-DC and passing practically unharmed through the output stage, which tells me that the linear stage has a poor PSRR and line regulation.

...
Is the scope telling you that there is 76 mV P-P of ripple on the output?
...
Yes, and I'm measuring this using the ground spring.

...
You could try a pi-filter on the output if you are concerned. Maybe a 3rd order Butterworh with a corner at 30 kHz (10% of your switching frequency)
I guess so. Which is better, right at the output, or between the DC-DC and the output stage? Also, what input and output parameters should I consider? Infinite input impedance and zero output impedance?

Thanks in advance!
 

Papabravo

Joined Feb 24, 2006
16,498
I chose the wrong words, sorry. Indeed it is ripple coming from the DC-DC and passing practically unharmed through the output stage, which tells me that the linear stage has a poor PSRR and line regulation.


Yes, and I'm measuring this using the ground spring.


I guess so. Which is better, right at the output, or between the DC-DC and the output stage? Also, what input and output parameters should I consider? Infinite input impedance and zero output impedance?

Thanks in advance!
PSRR is irrelevant here. The ripple does not come from the input to the DC-DC converter. The ripple comes from the rising and falling current in the inductor. It is an inherent part of the DC-DC conversion process. If you want to reduce the ripple to a lower value you put the pi-filter on the output of the DC-DC converter. If that is still not enough then you can increase the order of the pi-filter. The major benefit you have is the switching frequency of 300+ kHz. is many orders of magnitude higher than 50/60 Hz. which is what you have to deal with in a linear power supply.

Just curious. Are you completely unfamiliar with how a switch mode power supply works. I'm not trying to insult you I'm just trying to understand where you are coming from.
 

Thread Starter

bloguetronica

Joined Apr 27, 2007
1,453
I'm refering to the PSRR of the output stage after the DC-DC. Although it is a discrete design, the term applies, no? Anyway, the ripple generated by the DC-DC pre-regulator goes right through the output stage, and thus, the last stage brings little advantage. Sure it helps suppress higher frequency harmonics, and also allows regulation to the mV.

As for your question, I'm totally familiar with how a DC-DC buck converter works. However, mind that the last stage is linear.
 

Papabravo

Joined Feb 24, 2006
16,498
I'm refering to the PSRR of the output stage after the DC-DC. Although it is a discrete design, the term applies, no? Anyway, the ripple generated by the DC-DC pre-regulator goes right through the output stage, and thus, the last stage brings little advantage. Sure it helps suppress higher frequency harmonics, and also allows regulation to the mV.

As for your question, I'm totally familiar with how a DC-DC buck converter works. However, mind that the last stage is linear.
So, obviously the linear stage is doing nothing to attenuate the ripple. Is there some reason to expect that it would given it's design?
What is the loop gain in the linear regulator at 300 kHz?
If I was going to add a pi-filter it would be between VPRE and VOUT.
 

Thread Starter

bloguetronica

Joined Apr 27, 2007
1,453
So, obviously the linear stage is doing nothing to attenuate the ripple. Is there some reason to expect that it would given it's design?
What is the loop gain in the linear regulator at 300 kHz?
...
No particular reason, but I would expect it to do so, nonetheless. How should I determine the loop gain? Should it be increased? Reduced? This is a design I've slapped together, mainly the output stage. I've used it in other projects, under different circumstances, and it worked fine, but those designs hadn't a DC-DC pre-regulator on them.

...
If I was going to add a pi-filter it would be between VPRE and VOUT.
Indeed it would be the better solution.
 

michael8

Joined Jan 11, 2015
151
I'm a bit unsure, but don't r10 and c20 make a low pass filter in the linear regulator feedback path so that the 300Khz ripple can't get to the linear pass transistor to fix the ripple from the DC-DC?

Not obvious to me until I looked it up, the MJD122 is a darlington transistor with about 150pF input capacitance which along
with r12 is another low pass.
 

Thread Starter

bloguetronica

Joined Apr 27, 2007
1,453
Linear regulators work fine if the ripple is at the line frequency (50/60 Hz.), but their response is not fast enough at 300 kHz.
Here is an introductory article that explains "loop gain" to get you started.
https://www.analog.com/en/technical...nd-its-effect-on-analog-control-systems.html#
Making the calculation with discrete components is more challenging but follows the same principles.
Thanks!

I'm a bit unsure, but don't r10 and c20 make a low pass filter in the linear regulator feedback path so that the 300Khz ripple can't get to the linear pass transistor to fix the ripple from the DC-DC?

Not obvious to me until I looked it up, the MJD122 is a darlington transistor with about 150pF input capacitance which along
with r12 is another low pass.
Like Papabravo suggested, it depends on the frequencies. And, a slow response of the feedback loop is not desired, so anything low pass on that side is bad. You want the feedback loop to react as fast as possible, provided that the op-amp has sufficient phase margin to work in a stable manner.
 

Papabravo

Joined Feb 24, 2006
16,498
The early developers of SMPS used voltage feedback and discovered that the response was too slow to handle load transients -- like starting a DC motor. You would connect the supply to the motor and the voltage would drop dramatically until the motor begins to spin and the back EMF limits the current. Sometimes this wouldn't happen fast enough and the magic smoke would appear. Switching frequencies were much lower like 12-25 kHz if I remember correctly.

Just like in an opamp with a low slew rate, the low magnitude, high frequency ripple is essentially invisible to the regulator and as a result nothing much happens. Whenever you work with devices that have parasitic elements you can't see, and would normally ignore, you need to be aware of the additional pathways they provide for signal propagation.

There is one last thing you might be able to try. It might be possible to change the inductor/capacitor characteristics of L1 in combination with what looks like C5, C6 , and C7 to improve the magnitude of the ripple. How did you determine the values originally?
 

Thread Starter

bloguetronica

Joined Apr 27, 2007
1,453
Those values were calculated from the equations in the datasheet, and the inductor was picked up to fit the requirements. The DC-DC part works as it should. I don't remember the trade-offs I had to do, so I might worsen something if I meddle with the circuit. I remember simulating this, and the behavior I'm seeing on the DC-DC part is what it should be. I guess I'll have to implement a large Pi filter, with beefy capacitors on either side to compensate for the added inductance.
 

Papabravo

Joined Feb 24, 2006
16,498
Those values were calculated from the equations in the datasheet, and the inductor was picked up to fit the requirements. The DC-DC part works as it should. I don't remember the trade-offs I had to do, so I might worsen something if I meddle with the circuit. I remember simulating this, and the behavior I'm seeing on the DC-DC part is what it should be. I guess I'll have to implement a large Pi filter, with beefy capacitors on either side to compensate for the added inductance.
You still have some flexibility in choosing component values. It is not like datasheets provide a unique solution; what they provide is a typical solution.
You only need the 'beef' at low frequencies. Your switching frequency is 300+ KHz, filters with corner frequencies from 10-30 Khz are both easy to design and easy to implement.
 
Top