Proper procedure for reducing ripple on DC power rail? (e.g. oscilloscope reading -> filter design)

Thread Starter

Mahonroy

Joined Oct 21, 2014
406
Hey guys,
I have a 3.3 volt DC power rail. I have an integrated circuit that can only operate correctly if its power input has less than 30mV ripple.

What is the procedure for building the corresponding filter circuitry to accommodate those requirements?
Is there a reading I can take with the oscilloscope, then build a circuit accordingly, then verify again with the oscilloscope? Is there a way to go about designing this where I don't even need to use an oscilloscope - e.g. do some math and it should guarantee the 30mV max ripple requirements?

What about additional noise that is on the line, will it take care of that as well?

How do you guys go about this, whats the process?

Here are some more specifics if they are needed for this example:
  • Integrated circuit has a max current draw spikes of 40mA, pretty much anywhere from 0.1mA to 40mA. Averaging 20mA.
  • Lets say the voltage regulator can handle a max output of 250mA.
  • 3.3V, 30mV ripple max
 

ronv

Joined Nov 12, 2008
3,770
Hey guys,
I have a 3.3 volt DC power rail. I have an integrated circuit that can only operate correctly if its power input has less than 30mV ripple.

What is the procedure for building the corresponding filter circuitry to accommodate those requirements?
Is there a reading I can take with the oscilloscope, then build a circuit accordingly, then verify again with the oscilloscope? Is there a way to go about designing this where I don't even need to use an oscilloscope - e.g. do some math and it should guarantee the 30mV max ripple requirements?

What about additional noise that is on the line, will it take care of that as well?

How do you guys go about this, whats the process?

Here are some more specifics if they are needed for this example:
  • Integrated circuit has a max current draw spikes of 40mA, pretty much anywhere from 0.1mA to 40mA. Averaging 20mA.
  • Lets say the voltage regulator can handle a max output of 250mA.
  • 3.3V, 30mV ripple max
A lot depends on the power supply. Is it a switching supply? They are normally a little noisier. If it is a switcher, probably the easiest way is to scope it to see the frequency of the noise and the amplitude then design around that.
 

#12

Joined Nov 30, 2010
18,224
1.414 x capacitance x p-p voltage ripple x frequency = current.
The only things you gave us were the ripple voltage and the current.
Plug in the frequency and you can find the capacitor.
 

hp1729

Joined Nov 23, 2015
2,304
Hey guys,
I have a 3.3 volt DC power rail. I have an integrated circuit that can only operate correctly if its power input has less than 30mV ripple.

What is the procedure for building the corresponding filter circuitry to accommodate those requirements?
Is there a reading I can take with the oscilloscope, then build a circuit accordingly, then verify again with the oscilloscope? Is there a way to go about designing this where I don't even need to use an oscilloscope - e.g. do some math and it should guarantee the 30mV max ripple requirements?

What about additional noise that is on the line, will it take care of that as well?

How do you guys go about this, whats the process?

Here are some more specifics if they are needed for this example:
  • Integrated circuit has a max current draw spikes of 40mA, pretty much anywhere from 0.1mA to 40mA. Averaging 20mA.
  • Lets say the voltage regulator can handle a max output of 250mA.
  • 3.3V, 30mV ripple max
What am I missing here? How can you apply a remedy without asking where the problem originates? What chip or stage is causing the problem? Is it the power supply?
 

Thread Starter

Mahonroy

Joined Oct 21, 2014
406
A lot depends on the power supply. Is it a switching supply? They are normally a little noisier. If it is a switcher, probably the easiest way is to scope it to see the frequency of the noise and the amplitude then design around that.
I have 3 separate designs I'm working with. I am wondering if the process would be the same for each design?
  • 12VDC power adapter into a 5V switcher (RECOM r-78e5.0-1.0). There is then a 3.3V LDO voltage regulator (MCP1700) connected to the 5V rail.
  • 12VDC power adapter into a 3.3V switcher (RECOM r-78e3.3-0.5).
  • LNK305 switcher generating a 4.5V rail. Then the 3.3V LDO voltage regulator (MCP1700) connected to this 4.5V rail.

1.414 x capacitance x p-p voltage ripple x frequency = current.
The only things you gave us were the ripple voltage and the current.
Plug in the frequency and you can find the capacitor.
So you must know the exact ripple frequency in order to filter it out? What if there is some randomness to it? What about external interference and noise?

What am I missing here? How can you apply a remedy without asking where the problem originates? What chip or stage is causing the problem? Is it the power supply?
Is there a different process to this if you are trying to filter at the source vs at the destination?

Thanks again for the responses!
 

OBW0549

Joined Mar 2, 2015
3,566
I have a 3.3 volt DC power rail. I have an integrated circuit that can only operate correctly if its power input has less than 30mV ripple.
What IC is this (manufacturer and part number)? Analog or digital? What is the application circuit of this chip-- can you show us a schematic diagram? How much ripple are you experiencing right now? This info might help in arriving at a solution.

I have 3 separate designs I'm working with. I am wondering if the process would be the same for each design?
  • 12VDC power adapter into a 5V switcher (RECOM r-78e5.0-1.0). There is then a 3.3V LDO voltage regulator (MCP1700) connected to the 5V rail.
  • 12VDC power adapter into a 3.3V switcher (RECOM r-78e3.3-0.5).
  • LNK305 switcher generating a 4.5V rail. Then the 3.3V LDO voltage regulator (MCP1700) connected to this 4.5V rail.
I'm assuming the problem here is high-frequency ripple emanating from the switching supplies you're using to power the LDO, not excessive mains-frequency ripple caused by an inadequate post-rectifier filter capacitor.

Because of their topology (i.e., incoming voltage connects to the emitter of a series-pass BJT, or the source of a series-pass MOSFET), LDO regulators typically have very poor PSRR (power supply rejection ratio, a.k.a. ripple rejection) at high frequencies; and micropower LDOs such as the MCP1700 are extremely poor. What this means is that any high-frequency ripple on the LDO's input tends to pass right through the regulator with little or no attenuation (for example, see Figure 2-15 on page 7 of the MCP1700 data sheet).

The RECOM switchers operate at a nominal frequency of 330 kHz and produce up to 120 mV peak-to-peak on their outputs; the LNK305 runs at 66 kHz but doesn't specify ripple voltage, so you can assume it's pretty bad. At these frequencies, the MCP1700 provides essentially no ripple rejection, and this is most likely the source of your problem.

As a general rule, if you're looking for a low-noise regulated supply, the combination if a switcher followed by an LDO is NOT a very good choice.

The way I see it, you have two alternatives: a) since you have plenty of input voltage available, you can use a non-LDO regulator such as an LM317 which has much better PSRR, or b) if you want to continue using the MCP1700 you can put a simple RC lowpass filter between the switcher and your LDO to reduce the noise at the LDO's input. A 10Ω series resistor followed by a 100 μF shunt capacitor, for example, will give you roughly 55 db of attenuation at 100 kHz.

Either one would probably take care of your problem.
 

#12

Joined Nov 30, 2010
18,224
So you must know the exact ripple frequency in order to filter it out?
You design the capacitor for the lowest frequency of ripple voltage and the higher frequencies magically pass through the filter capacitor to ground as if the impedance of a capacitor is less at higher frequencies.
 

Thread Starter

Mahonroy

Joined Oct 21, 2014
406
What IC is this (manufacturer and part number)? Analog or digital? What is the application circuit of this chip-- can you show us a schematic diagram? How much ripple are you experiencing right now? This info might help in arriving at a solution.


I'm assuming the problem here is high-frequency ripple emanating from the switching supplies you're using to power the LDO, not excessive mains-frequency ripple caused by an inadequate post-rectifier filter capacitor.

Because of their topology (i.e., incoming voltage connects to the emitter of a series-pass BJT, or the source of a series-pass MOSFET), LDO regulators typically have very poor PSRR (power supply rejection ratio, a.k.a. ripple rejection) at high frequencies; and micropower LDOs such as the MCP1700 are extremely poor. What this means is that any high-frequency ripple on the LDO's input tends to pass right through the regulator with little or no attenuation (for example, see Figure 2-15 on page 7 of the MCP1700 data sheet).

The RECOM switchers operate at a nominal frequency of 330 kHz and produce up to 120 mV peak-to-peak on their outputs; the LNK305 runs at 66 kHz but doesn't specify ripple voltage, so you can assume it's pretty bad. At these frequencies, the MCP1700 provides essentially no ripple rejection, and this is most likely the source of your problem.

As a general rule, if you're looking for a low-noise regulated supply, the combination if a switcher followed by an LDO is NOT a very good choice.

The way I see it, you have two alternatives: a) since you have plenty of input voltage available, you can use a non-LDO regulator such as an LM317 which has much better PSRR, or b) if you want to continue using the MCP1700 you can put a simple RC lowpass filter between the switcher and your LDO to reduce the noise at the LDO's input. A 10Ω series resistor followed by a 100 μF shunt capacitor, for example, will give you roughly 55 db of attenuation at 100 kHz.

Either one would probably take care of your problem.
Thanks a lot for this good information. This really does sound like what is going on.

How did you come up with the "10Ω series resistor followed by a 100 μF shunt capacitor" values? Is there a formula you used?

Would it be best to do the RC filter, as well as the non-LDO voltage regulator? Or would one of the 2 methods suffice?

Is there anything else I should look for as far as the linear voltage regulators go? It looks like I am out of luck trying to find a linear regular that matches the same footprint as the MCP1700 to try out the theory. In that case, if I can find a non-adjustable one that would cut down on components. So you are looking at the rejection rate, how can you tell one is better than the other, or what I should be looking for?

I wonder if there is a patch I could try on this circuit to test it out before I incorporate that? The red arrow is where the voltage ripple needs to be small. Think I could replace L3 with a 10Ω, C1 with the 100uF, L4 with a 0Ω?
nrf_troubles7.jpg
 

OBW0549

Joined Mar 2, 2015
3,566
How did you come up with the "10Ω series resistor followed by a 100 μF shunt capacitor" values?
I initially chose them by pure guesstimation, based on >50 years experience. I then confirmed my guesstimate by calculation.

Is there a formula you used?
The cutoff frequency of a low-pass filter (i.e., the frequency at which it begins to attenuate) is given by Fc = 1 / (2 * pi * R * C). And for a single-pole low-pass filter, the attenuation in db at a particular frequency is given by Atten(db) = 20 * log10(F / Fc).

Would it be best to do the RC filter, as well as the non-LDO voltage regulator? Or would one of the 2 methods suffice?
I think either one of the two methods would suffice. If you want to be really paranoid, do both.

So you are looking at the rejection rate, how can you tell one is better than the other, or what I should be looking for?
You look at the power supply rejection ratio (PSRR) spec, also known as "ripple rejection." It will be expressed in db. The larger the number, the better.

I wonder if there is a patch I could try on this circuit to test it out before I incorporate that? The red arrow is where the voltage ripple needs to be small.
I would put it in the +5V line. Snip the connection between R4 (the 1 kΩ resistor) and C9, and insert the filter there.

Think I could replace L3 with a 10Ω, C1 with the 100uF, L4 with a 0Ω?
Absolutely NOT! The 10Ω resistor and 100uF capacitor must go in front of the MCP1700's input, NOT after its output; if you put the resistor in the 3.3V output line, all you'd accomplish would be to destroy your voltage regulation because of the voltage drop across the resistor.

Do replace L4 with a 0Ω jumper, though; L4 has no business being there.
 

Thread Starter

Mahonroy

Joined Oct 21, 2014
406
I would put it in the +5V line. Snip the connection between R4 (the 1 kΩ resistor) and C9, and insert the filter there.
I gave this a shot. For some reason it didn't seem to make any difference? Unless I am not using the scope correctly? I took video of before and after, what do you think?
Here is the change I made (oh, since the ferrite bead was weird, I used the version without them), everything else identical. I measured across the power and ground of the nRF24L01 :

Before:
nrf_problem1.jpg
Video of oscilloscope:
https://goo.gl/photos/TxpybHEWfKQrqGRp7
https://goo.gl/photos/h5W5WoXW97UWjsgS8


After:
nrf_problem2.jpg
Video of oscilloscope:
https://goo.gl/photos/A2G4f1x3rvNkj4dt8
https://goo.gl/photos/BZsGnWFYK9TVxs418

So it looks like the sharp edges were smoothed out a little, but didn't change much. What do you guys think?
Thanks again for all of the help!
 

OBW0549

Joined Mar 2, 2015
3,566
Wow. Something is terribly wrong, there.

Whatever the problem, it doesn't appear to be what I thought it was (feed-through of switching regulator noise via the LDO). If it had been, that filter would have made a huge difference. This is starting to look like the kind of noise you get from a switching regulator when you don't lay out the ground topology correctly: a huge amount of high-frequency, high-amplitude spikey "hash."

Maybe someone else can step in here and offer better advice, but I've about reached the limit of what I can do by "remote diagnosis" other than to say something is clearly very, very wrong. Given the number of things that might be causing this, it would probably be appropriate to find yourself a local EE who can get hands-on with this thing and do some in-depth digging; that's probably what it will take.
 

Thread Starter

Mahonroy

Joined Oct 21, 2014
406
Do you think those high frequency thin spikes might be from the oscilloscope picking up interference, and in reality its not actually on my PCB? I plugged in just a standard 12VDC power adapter and I can see thin spikes on that too.

Another thing I had tried was gutting the switching regulator portion and rigged up an external 5V power source into the 5V rail, and I see similar things.

Yeah I know what you mean about being hands on, diagnosing this stuff is tricky.
 

OBW0549

Joined Mar 2, 2015
3,566
Do you think those high frequency thin spikes might be from the oscilloscope picking up interference, and in reality its not actually on my PCB?
That is entirely possible. Make sure your scope probe ground is connected to your circuit ground through a VERY short lead, otherwise you can pick up all kinds of crap.

I plugged in just a standard 12VDC power adapter and I can see thin spikes on that too.
If it's a switching-type power adapter, that wouldn't be very surprising; but the noise pattern should be different, at least.

Another thing I had tried was gutting the switching regulator portion and rigged up an external 5V power source into the 5V rail, and I see similar things.
Again, if the external 5V supply is a switching type, you could be getting noise from that, too.

Yeah I know what you mean about being hands on, diagnosing this stuff is tricky.
One thing that might help narrow down the list of possible culprits would be to ditch the power adapters and switching supply entirely, and run this thing from a linear regulator (like a 7805 with appropriate input & output capacitors) powered off a 9V battery. If your circuit quiets down, that will tell you that the switcher or adapter was the problem; if the linear regulator doesn't help any, that's a sign that the problem is some other source of interference.
 

Thread Starter

Mahonroy

Joined Oct 21, 2014
406
That is entirely possible. Make sure your scope probe ground is connected to your circuit ground through a VERY short lead, otherwise you can pick up all kinds of crap.


If it's a switching-type power adapter, that wouldn't be very surprising; but the noise pattern should be different, at least.


Again, if the external 5V supply is a switching type, you could be getting noise from that, too.


One thing that might help narrow down the list of possible culprits would be to ditch the power adapters and switching supply entirely, and run this thing from a linear regulator (like a 7805 with appropriate input & output capacitors) powered off a 9V battery. If your circuit quiets down, that will tell you that the switcher or adapter was the problem; if the linear regulator doesn't help any, that's a sign that the problem is some other source of interference.
I was actually rigging that exact thing up... just got done testing it. I took a new board without the RC modification, completely severed the circuit between C8 and R4, rigged up a 7805 and a power adapter. Here are the results:
https://goo.gl/photos/U3rpHXU7YF95NMjt5

I will now try with a 9V battery instead of that adapter.
 
Top