What causes low frequency ripple in a PSU?

[WatchfulEye]

I bought a super-cheap bench PSU mainly for LEDs, car battery charging and other non-critical purposes (sold on amazon under the name Nankandf, but the OEM appears to be Wanptek). Model number is TPS3010.

I thought I would just test the output with a scope for interest purposes, and becuase I had an older, but similar supply which suffered catastrophic noise problems and completely messed up my analogue circuits. The basic functions, accuracy and user interface on this newer model are all fairly satisfactory.

However, it has a fair bit of ripple - but what I don't quite understand is why the ripple is not at the switching frequency (100 kHz), but instead it is at 1 kHz. For example, at 25 V 1.5 A, the PSU produces nearly 20 mV p-p of 1 kHz noise (see waveform), which I found quite surprising.

For my education, I'd be interested to understand where this low frequency noise is coming from? Could this be a loop compensation issue (the ripple is only present in CV mode, in CC mode, the ripple is 100 kHz and much smaller)?

Examination of the PSU circuit shows it to use a standard TL494 controller with conventional voltage and current limit feedback for CV/CC operation.

[bdunham7]

How does it vary, if at all, with loads going from no load to the most you can get while staying in CV mode?

[CaptDon]

The regulation control loop may have a tiny bit of hysteresis for overall stability and the 1KHz is the response time of the control loop. It ends up chasing its tail.

[WatchfulEye]

How does it vary, if at all, with loads going from no load to the most you can get while staying in CV mode?

it is dependent on both voltage and current (mainly voltage where it is roughly proportional), but doesn't disappear until constant current mode takes effect, or the current is low (<100 mA). The frequency is fixed.

I guess the hysteresis idea might be it.

[Konkedout]

If you see 1 KHz "ripple" on the output, it is probably instability or oscillation.  This would come from a badly designed internal feedback loop.  You MIGHT be able to quash it by putting a large value electrolytic capacitor on the output; maybe with a low value resistor (100 milliohms for example) in series with the capacitor.  Do you have any large value electrolytics laying around that can handle the voltage?  If so, the series resistor should be chosen approximately by:

R=1/(2*pi*1KHz*C) where C is in Farads.

If you figure this on Faraday.   

Just to say: It is very likely that you have a poorly designed power supply with an unstable feedback loop.  The dang thing is oscillating.  Without digging into the electronics, the easiest way to fix it is by external damping as I describe above.  This can improve the feedback loop gain and phase margin.

You can buy "low ESR" or "low impedance" electrolytics such as 2200 uF 35V for < $2 US.    You want a low impedance capacitor in order to know what you have.
 My formula above will provide the best feedback loop damping (for a given capacitance) at 1 KHz.  If you use a few of them in parallel then the series resistor should scale down in inverse proportion, or each capacitor can have its own series resistor; that may be easier to implement well.    If you can do this with (educated guess) 3-10 millifarads of total capacitance and the right calculated resistor value (+/-10%) there is a very good chance that you can kill this oscillation.  Your series resistance calculation should ideally include the data sheet ESR of the capacitor.  But that actual ESR in the capacitor should typically be slightly lower than the specified number in the datasheet.

I have been designing power supplies since 1980.

[jwet]

You've gotten good responses but I'd like to add a bit.  Its not that the supply is a poor design most likely.

Switch mode supplies generate their lowest and fundamental only (switching frequency) ripple at full load.  This is generally when they are in what is called "continuous conduction mode" where the current in the inductor ramps up and down during on cycles of the switch.  In this mode, the current doesn't decay to zero before the next cycle starts.  The average current will be the output current and the ripple current on the inductor might only be 1/3 of the output current.  This current ripple gets stored by the output C of the supply and gives a ripple voltage of IR where the R is the high frequency ESR of the output cap.  Adding lots of C will reduce ripple but its not the C as much as the reduction of ESR with the added C.  Tantalums and Organic Semiconductor Electrolytics (Panasonic was Sanyo).  A bit more on this later.  Adding lot of high ESR C can make low frequency skipping worse.

At lighter loads, the controller reduces the PWM of the on current in the inductor until the current in the inductor reaches zero between cycles.  This is called "discontinuous conduction mode".  Ripple increases since the P-P current is now increased, it goes from zero up to about twice the output current (vs. 1/3).  This higher I times cap ESR will create significantly more ripple voltage.

Finally, as the relative load is reduced more, you get into pulse skipping mode where you're below the level where the smallest PWM fraction will keep you in regulation.  Basically, you start skipping cycles in some way (see DS for PS Controller IC) and become more like a PFM controller with frequency depending on load and generally a fraction of the switching frequency.  This tends to be very high ripple since the output cap isn't just a filter, its stores energy between cycles.  Ripple can be terrible.  Adding lots of C here will help since you need better storage.

You'd like to be in the first mode.  If your load varies a lot from heavy to light, you might be stuck but if its somewhat constant and you're in a skipping mode or DCM, you can increase the value of the inductor in the buck and push yourself more toward continous conduction.

I hope this helps in your understanding.  These little chinese modules are designed for some max current, don't oversize (or underload) them, they will misbehave.

[Konkedout]

Pulse skipping or "pulse frequency modulation" here is a possibility.  But the higher frequency that I see riding on top of the 1 KHz looks like the switcher is not pulse skipping.  And the sorta-sinusoidal 1KHz waveshape also looks more like classic feedback loop instability.  But I have been worng before....

Pulse skipping or PFM is usually a way of saving a bit of power when a synchronous rectified output is operating at light load.   If the output uses a (schottky) diode then a related operating mode might be difficult to avoid.

I would think that a bench supply ought not do pulse skipping because users are more often interested in low noise than in saving a couple of watts at light load.

[WatchfulEye]

Thanks guys, especially @konkedout.

I added 6mF of junkbox caps and some wire to add some resistance to the output and while it didn't kill the 1 kHz ripple, it cut it in half.

I dug out my older model PSU (which has terrible potentiometers and common mode noise) but this one doesn't have the 1 kHz noise. It looks near identical inside, so I reverse engineered the circuit. It supposedly uses "low esr" caps on the output, but as these are counterfeit, who knows. Nevertheless, it uses a type 2 compensator with quite a low bandwidth, buit nevertheless the design looks potentially at risk of instability.

A variety of simulations with the circuit suggest that gain peaking and loss of phase margin at 1 kHz are extremely easy to get if ESR is reduced or the bandwidth of the compensator is increased.  They also highlighted a potential solution - the conversion to a type 3 compensator. The addition of C1 in the attached screenshot.

When I get a chance, I'll take a look at the circuit in the new supply and see what has changed, but I think there may be a strategy to modify the compensation as well as just adding capacitance/ESR.

[jwet]

Konkedout is right and on the right track.  I didn't see the "Bench" part of your description.  I saw cheap and jumped to the conclusion that it was  board level module.  My comments stand for those but if its designed to be a lab supply, it has some kind of feedback instability- it might be a bad supply, its hard to believe that they all work that way.  Congrats.

[WatchfulEye]

Konkedout is right and on the right track.  I didn't see the "Bench" part of your description.  I saw cheap and jumped to the conclusion that it was  board level module.  My comments stand for those but if its designed to be a lab supply, it has some kind of feedback instability- it might be a bad supply, its hard to believe that they all work that way.  Congrats.

I looked up some of the youtube reviews. Some do show the output wave form - where there is enough detail shown, the noise is indeed at 1 kHz - so they probably do all work that way!

This is interesting, as my old supply from a different brand but is quite obviously, when disassembled, a near identical supply - same basic topology, same overall PCB layout style. The new model just switches junkbox potentiometers for an MCU and DAC, and adds a pair of USB-charge outputs on the front panel.

The new revision makes some changes to the output caps, so I wonder if that was enough to push things over the edge.

[bdunham7]

Keep in mind that there is a tradeoff between damping or adjusting any feedback loops and the load transient response.  Unless that 1kHz ripple is a problem, you might consider it just a feature of how the PSU works. 

[jwet]

A "good" lab supply should have limited output caps, big output caps make it difficult do current limiting, etc- easy to burn up sensitive loads with a big charged cap on the output.  Kind of another way of saying what bdunham7 said in a different way.  The transient response should be designed into the loop dynamics.

[WatchfulEye]

This rabbit hold went far deeper than I had originally intended.

What had originally started as an interesting educational session into loop stability, started not to make sense, because nothing would change the 1 kHz frequency.

So what is the root cause of 1 kHz noise on this PSU? It's the PWM carrier frequency used by the MCU to generate the set-point voltage.    Who would have thought?