Newbie Q: Why not just low-pass filter SMPS vs. using a linear lab PSU?

Thread Starter

ballsystemlord

Joined Nov 19, 2018
167
Hello,
Among the items I don't have, is a lab power supply. Currently, I'm using a buck convertor attached to a small 24v/10A PSU from China. This works well enough when I need power, but IDK how good it would be when I need low noise performance.

Now I must confess that I've not read far enough into my books to know the answer to this. I'll get there. In the interim, please be gentle.

My current plan is to utilize an old PSU and make myself a SMPS from it. But after reading up on such matters I'm uncertain if I'd be possible to get low noise performance from it. Hence my question.

If the answer is "You can't get low noise performance from an SMPS PSU," I won't abandon my PSU idea. It'd be a good learning experience and I can always use it where a SMPS's higher noise designs are practical. This is more for setting realistic goals and expectations.

Can you use a low-pass filter to get similar noise performance to a linear PSU from an SMPS PSU? If "no," why not?

Thanks!
 
Last edited:

Jolly13

Joined Mar 25, 2018
27
All about the definition of low noise. OF course your power supply will work for a lot of thing. Not sure who told you other wise.
 

Papabravo

Joined Feb 24, 2006
21,258
You can build an SMPS that will power a 100 Watt transceiver doing weak signal work and never notice the power supply noise in the HF bands from 2-54 MHz. I will say that the power supply was a commercial product, and it was comparable in price to a typical linear power supply for the same application. The cost was about 20% of what the radio cost.

ETA. When you talk about noise you need to specify the frequency range of interest.
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
167
ETA. When you talk about noise you need to specify the frequency range of interest.
I can't say what my frequency range of interest will be in the future.
If I was to give a current goal, I'd say my current range of interest is at DC to 20MHz at -60dbV for noise. But if I can get lower noise, say -80dbV, I'd go for it.
 

Papabravo

Joined Feb 24, 2006
21,258
I can't say what my frequency range of interest will be in the future.
If I was to give a current goal, I'd say my current range of interest is at DC to 20MHz at -60dbV for noise. But if I can get lower noise, say -80dbV, I'd go for it.
I'm not sure what you mean. Do you plan to buy one and test it to see if it meets your requirements or are going to try to build one?
 

WBahn

Joined Mar 31, 2012
30,236
Hello,
Among the items I don't have, is a lab power supply. Currently, I'm using a buck convertor attached to a small 24v/10A PSU from China. This works well enough when I need power, but IDK how good it would be when I need low noise performance.

Now I must confess that I've not read far enough into my books to know the answer to this. I'll get there. In the interim, please be gentle.

My current plan is to utilize an old PSU and make myself a SMPS from it. But after reading up on such matters I'm uncertain if I'd be possible to get low noise performance from it. Hence my question.

If the answer is "You can't get low noise performance from an SMPS PSU," I won't abandon my PSU idea. It'd be a good learning experience and I can always use it where a SMPS's higher noise designs are practical. This is more for setting realistic goals and expectations.

Can you use a low-pass filter to get similar noise performance to a linear PSU from an SMPS PSU? If "no," why not?

Thanks!
It all comes down to what qualifies as "low noise" for your application and how much effort and expense you are willing to throw at the problem.

There's also noise and then there's noise.

We I was designing detector and imaging readout ICs, for a long time we used only linear supplies because any time we used a switcher we could see unacceptable artifacts in the data on our more sensitive chips. But then we had a chip that, for a variety of reasons, we had to implement a switcher on the chip, which meant that the noise was going to be considerably worse that using an external switcher. But since we had control of the switcher, we could force it to switch when we wanted it to and synchronize it with the readout process, which turned the random switching noise into fixed pattern noise that could be calibrated out very easily.
 

Irving

Joined Jan 30, 2016
3,973
My lab supply has a linear output stage but a tracking switching pre-regulator that runs about 2 - 3v above the desired output. This way you get the efficiency of a switcher but the low noise of a linear supply without the heat generation...
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
167
I'm not sure what you mean. Do you plan to buy one and test it to see if it meets your requirements or are going to try to build one?
As I mentioned in my original post, I had planned on modding a PC PSU. Should that fail, I'd assume that I could read the specs of what's available and purchase one.
 

BobTPH

Joined Jun 5, 2013
9,124
Have you seen any commercial consumer products that use a linear supply lately? I suppose there are auduiophool products that do, but they also use oxygen free power cables.
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
167
Have you seen any commercial consumer products that use a linear supply lately? I suppose there are auduiophool products that do, but they also use oxygen free power cables.
You forgot that such audio equipment also uses vacuum tubes. ;)

Electrical test equipment and regular "consumer" products are not really comparable IMHO. Electrical test equipment is, or should be, striving for excellence, not stooping to the status quo.

But the debate about quality doesn't end with just linear PSUs. If people in forums such as these are to be believed, cheap caps, such as CapXon shouldn't be used. Yet countless times I see them in tear downs of +$1000 electrical test equipment. So either we say we want genuinly good products, or we say that we'll take just about anything that has pretty lights to see, a bunch of buttons to push, and plenty of knobs to turn.
 

WBahn

Joined Mar 31, 2012
30,236
Electrical test equipment and regular "consumer" products are not really comparable IMHO. Electrical test equipment is, or should be, striving for excellence, not stooping to the status quo.
But what does "striving for excellence" even mean?

Electrical test equipment should strive to be good enough for the types of measurements for which it is being marketed. If you make it significantly "better" than that, then you will almost certainly have to sell it at a sufficiently higher price point so that (1) you can't sell nearly as many, which means that you won't have as much capital to design further products, and that's assuming that you are even able to stay in business, and (2) a lot of people that could have bought your equipment had it been designed for a more rational level of performance now can't afford it, and so they not only don't have the super-duper superior test equipment that you are actually selling, but they don't even have the solidly performing test equipment that you should have been selling. No winners on either side when things are excessively overdesigned.

But the debate about quality doesn't end with just linear PSUs. If people in forums such as these are to be believed, cheap caps, such as CapXon shouldn't be used. Yet countless times I see them in tear downs of +$1000 electrical test equipment. So either we say we want genuinly good products, or we say that we'll take just about anything that has pretty lights to see, a bunch of buttons to push, and plenty of knobs to turn.
I know very little about the situation in the caps industry -- it's not something that I've really had to deal with for a very long time. So I can only talk in general terms.

Manufacturers at all levels are always under price pressure, so it is not surprising that they are naturally drawn to lower cost components. As long as they make the effort to establish that those components are truly good enough for the use at hand, there's no problem with that. But, of course, sometimes they don't establish that, either because they tried to do so and missed the mark, for whatever reason, or they didn't even try.
 

BobTPH

Joined Jun 5, 2013
9,124
So you are a newbie who is designing lab quality instruments? Okay.

In my 55 years as an electronics hobbyist, I have never owned a lab quality power supply, much less attempted to design one.
 

Ya’akov

Joined Jan 27, 2019
9,230
So, several times people mentioned that you need to know what the definition of “noise” is before you can answer questions about it. This is not something to gloss over, it is fundamental.

For example if you are measuring things in the audio range, 10 MHz ”noise” doesn’t even exist (it can’t be a noise if you can’t hear it) but the same amplitude of unwanted signal at 10KHz could be completely devastating to your application.

So, first, you have to decide what “noise” is for you, and you can’t do that without knowing the application of the power supply. Unfortunately, “a lab supply” isn’t an application that provides enough context to do it.

Engineering is (among other things) the art of balancing constraints. These can include available space, available power, available time, and—almost always—available money. Lab supplies with fantastic metrics cost fantastic prices for a reason.

It really doesn’t make sense to drive a stake in the ground somewhere arbitrary, lacking a driving application, and call that a requirement. Real requirements are based on specific and well characterized applications. I would suggest that unless your goal is to become a design engineer solely focused on cost-no-object bespoke lab gear you consider how valuable your open-ended ”requirements” for this project will be as a learning experience.

At the very least, set down a good definition of “noise” for your application, along with the requirements for output from the supply—particularly current—which will drive the difficulty and cost of it. Also consider the odd combination of repurposing a device engineered for low cost and narrow use as one for an application that typically demands high cost devices that are engineered with that budget in mind. You will not find that these purpose built devices are cheap ones with expensive back ends, they are, start to finish, engineered for their purpose.

None of this is intended to discourage, but rather to suggest that your goal of learning will be much more effective if you don’t remove all of the… “fiber” from it before you sit down to eat… so to speak.
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
167
So you are a newbie who is designing lab quality instruments? Okay.
We all have to start somewhere. Making leds go blink with an arduino just didn't interest me. I wanted something a bit more practical. And, among other things, repurposing a PC PSU to be a lab PSU seems like a good goal to have.

In my 55 years as an electronics hobbyist, I have never owned a lab quality power supply, much less attempted to design one.
Interesting. As lab PSUs come in both linear and SMPS design, I'm curious how did you manage in the field while not having any lab power supply? Did you just cobble together whatever was laying around when you needed something?

So, several times people mentioned that you need to know what the definition of “noise” is before you can answer questions about it. This is not something to gloss over, it is fundamental.

For example if you are measuring things in the audio range, 10 MHz ”noise” doesn’t even exist (it can’t be a noise if you can’t hear it) but the same amplitude of unwanted signal at 10KHz could be completely devastating to your application.

So, first, you have to decide what “noise” is for you, and you can’t do that without knowing the application of the power supply. Unfortunately, “a lab supply” isn’t an application that provides enough context to do it.
I wasn't really trying to gloss over what noise is. 20Mhz wasn't arbitrary. That's the frequency which PSUs are spec'd out to on noise. Likewise, -60dbv places the noise in the single digit millivolt range. That's the least significant digit of a 5 digit DMM when measuring greater than 1v. Granted, as you pointed out, different applications can tollerate different levels of noise at different frequencies.

Engineering is (among other things) the art of balancing constraints. These can include available space, available power, available time, and—almost always—available money. Lab supplies with fantastic metrics cost fantastic prices for a reason.

It really doesn’t make sense to drive a stake in the ground somewhere arbitrary, lacking a driving application, and call that a requirement. Real requirements are based on specific and well characterized applications. I would suggest that unless your goal is to become a design engineer solely focused on cost-no-object bespoke lab gear you consider how valuable your open-ended ”requirements” for this project will be as a learning experience.
As for an application, my latest is trying to figure out what the PPM/C of a shunt resistor is. The manufacturer provided everything but that. I don't expect the PSU I'm working to modify to do the testing. This is an example of what I'd do with a lab PSU.

That being said, that's only one application. It's not everything.

At the very least, set down a good definition of “noise” for your application, along with the requirements for output from the supply—particularly current—which will drive the difficulty and cost of it. Also consider the odd combination of repurposing a device engineered for low cost and narrow use as one for an application that typically demands high cost devices that are engineered with that budget in mind. You will not find that these purpose built devices are cheap ones with expensive back ends, they are, start to finish, engineered for their purpose.

None of this is intended to discourage, but rather to suggest that your goal of learning will be much more effective if you don’t remove all of the… “fiber” from it before you sit down to eat… so to speak.
I think I understand what you mean. The reason I thought I could modify a PC PSU to be a bench PSU is because I've read about others doing it. I don't know of anyone who's caracterized the noise or other performance metrics of the complete system. Hence why I'm asking this question.
 

Pyrex

Joined Feb 16, 2022
291
Hi,
it's a good idea to repurposing a PC PSU to be a lab PSU . You can use the PSU for a lot of purposes.
On the other hand, if you want to reduce the noise to a very low level, it will be a lot of trouble. In practice, is it worth it to conduct a series of experiments and get a dubious result?
Most of the circuits you experiment with are insensitive to noise.

For those circuits that are sensitive to noise, have a linear power supply. In most cases, those circuits are of low power , and a small and light linear power supply will be enough. I'm writing this from my practice
 

BobTPH

Joined Jun 5, 2013
9,124
Interesting. As lab PSUs come in both linear and SMPS design, I'm curious how did you manage in the field while not having any lab power supply? Did you just cobble together whatever was laying around when you needed something?
No, I have a couple of adjustable power supplies, but they are not lab quality instruments. Perhaps we are using the term differently.

Most of my projects are powered by phone chargers or 12-24V bricks.
 

ElectricSpidey

Joined Dec 2, 2017
2,828
Can you use a low-pass filter to get similar noise performance to a linear PSU from an SMPS PSU? If "no," why not?
If it were that easy everyone would be doing it.

If you want to build a high quality "bench" supply, the last thing you want to start with is some "old" PC power supply.

My bench supplies are all home built linear, but I have yet to build a project that doesn't run from a switcher in the field.
 

Ya’akov

Joined Jan 27, 2019
9,230
think I understand what you mean.
You certainly don’t seem clueless, so I hope my post didn‘t give you the impression I thought you were. I was trying to emphasize the idea that the best learning exercises are practical and practical engineering projects are always solving a specific problem, with the requirements gathering and specification being a critical and far from easy part.

Every constraint you specify becomes a fixed point around which your solution must pivot. This can have surprisingly far reaching consequences for what seem to be trivial constraints as the pivoting multiplies. For this reason, it is important to learn the difference between organic and artificial constraints*
*the terminology I am using here is not without drawbacks, after all every built thing is an “artifact” but I think the labels work

Organic constraints include things like the laws of physics, financial limits, and (real) deadlines while artificial constraints include things like “it can’t use more than X components” (for some reason that can’t be articulated), “no MCUs” (because MCUs are harder than discrete logic, it seems), or others I suspect you can imagine.

I feel that you’ve driven stakes into incompatible locations in your solution space and things are going to be more complicated and less satisfactory because of it. Those things being “I want to build my own lab supply“, and “I want to do it but converting a PC SMPS”.

If the constraint is building your own lab supply equivalent, the PC PS is not a good constraint, on the other hand if you want to use a PC PS, the lab supply target is not a good one. Instead, you might choose to drop the PC PS as a constraint while not ignoring it as a possibility—or you might instead want to make the PC PS into the best benchtop power supply it could be with a lab supply as an aspiration.

One more thing: potential solutions have a habit of displacling the problems they are meant to solve and taking on that rôle. For example (and this is mentioned only as a good example of something that could become pathological, not that it has) your suggestion that a low pass filter could be a way to work this has the potential to drive your projects off the rails and become a new and far less functional problem you will fight to solve.

Good luck on your project, please don’t take any of this as meant to discourage or criticize you. A really great engineer understands the critical rôle of the initial phase of the design process in creating truly successful solutions.
 

Irving

Joined Jan 30, 2016
3,973
I have a Tenma 72-6905 'lab' supply, spec'd at <=1mV p-p noise from 5Hz - 1MHz, over 0 - 60v, 0 - 6A output. This was considered acceptable 'lab' quality when new in 2001; it still manages to be better than that 24y later. This is a purely linear device. My Digimess SM3040, of a similar age, is spec'd similarly at <=3mV over 0 - 30v at 0 - 40A output. This is SMPS pre-reg + linear output. Both came from a radio comms engineering lab where I once worked and I purchased these at knockdown prices when it closed 15 years later, but both were around $400 new. I also have a Minibea ZX telecoms supply modded to be variable output 0 - 60v @ 0 - 50A that is a pure switcher. Its not 'lab' quality but intended originally for powering telecoms PABX. Its noisy - but still <= 20mV at full output. It was <$200 shipped from Hong Kong in 2019.

A current Tektronix 'lab' bench supply PWS2326 is spec'd at <3mV p-p noise 20Hz - 7MHz over 0 - 32v, 0 - 6A - this is a $1000 'low noise' device... The programmable equivalent, for 'lab' use and automated testing (PWS4602), is <5mV p-p noise 0 - 60v @ 0- 2.5A and $1500. Their bigger brother, the PWS2260B, 80v 27A is <80mV p-p noise(!) 20MHz and $3,000...

My point here is that 'lab' quality has no specific definition - its what is 'good enough' for your needs, but <3mV seems to be a good number. And price is not a definitive indicator of 'noise' quality.
 

Thread Starter

ballsystemlord

Joined Nov 19, 2018
167
You certainly don’t seem clueless, so I hope my post didn‘t give you the impression I thought you were. I was trying to emphasize the idea that the best learning exercises are practical and practical engineering projects are always solving a specific problem, with the requirements gathering and specification being a critical and far from easy part.

Every constraint you specify becomes a fixed point around which your solution must pivot. This can have surprisingly far reaching consequences for what seem to be trivial constraints as the pivoting multiplies. For this reason, it is important to learn the difference between organic and artificial constraints*
*the terminology I am using here is not without drawbacks, after all every built thing is an “artifact” but I think the labels work

Organic constraints include things like the laws of physics, financial limits, and (real) deadlines while artificial constraints include things like “it can’t use more than X components” (for some reason that can’t be articulated), “no MCUs” (because MCUs are harder than discrete logic, it seems), or others I suspect you can imagine.
Got it!

I feel that you’ve driven stakes into incompatible locations in your solution space and things are going to be more complicated and less satisfactory because of it. Those things being “I want to build my own lab supply“, and “I want to do it but converting a PC SMPS”.

If the constraint is building your own lab supply equivalent, the PC PS is not a good constraint, on the other hand if you want to use a PC PS, the lab supply target is not a good one. Instead, you might choose to drop the PC PS as a constraint while not ignoring it as a possibility—or you might instead want to make the PC PS into the best benchtop power supply it could be with a lab supply as an aspiration.
You read my mind. Hence my question. I was/am trying to set my expectations realistically regarding what can be done with an SMPS design. Of course, different designs will have different levels of noise and power output. Seeing as how some stuff utilizes what is "tried and true" and other things seek better performance and/or more features using the latest and greatest tech, I was curious what could be achieved.
Was a low-pass filter for an SMPS just hard to design well? Or maybe the tech is new* and so few if any are going with it? Or maybe it's impossible and linear PSUs are the way to go for good noise performance?

One more thing: potential solutions have a habit of displacing the problems they are meant to solve and taking on that role. For example (and this is mentioned only as a good example of something that could become pathological, not that it has) your suggestion that a low pass filter could be a way to work this has the potential to drive your projects off the rails and become a new and far less functional problem you will fight to solve.

Good luck on your project, please don’t take any of this as meant to discourage or criticize you. A really great engineer understands the critical role of the initial phase of the design process in creating truly successful solutions.
Thanks, I'll keep your good advice in mind.


* EDIT: When I say "new" I mean new chips or new strategies based on recent break throughs. Of course, low-pass filters are not a new concept.
 
Last edited:
Top