Failed simple smps design: A case study

[Born2BeWired]

I constructed an adjustable power switching regulator circuit using Allegro’s simple switcher A8447ST, hoping that I could share this circuit with anyone in this forum who would like to explore switching power supplies by building their own. The circuit I came up with is not so different from the datasheet circuit, modifying it just a bit to make it work respectably within desired output voltage range.

Using a 24VDC test source and ProDigit 3311 Electronic Load in 1Amp constant current mode as test load, I got a load and line regulation performance figure better than 1%; pretty good for this kind of circuit. Output voltage range measured 0.8VDC to 15VDC, and the IC worked with 80% efficiency at 5VDC output, keeping the IC running cool even at these low output voltage settings. But when output noise was measured, the results were so horrible, better to keep them hidden from view. This is basically a failed project. Why then am I posting it? Believe it or not, It could be more fun when circuits do not work as expected, because failures brings more opportunities to learn new things than some boring outright success. A failure is not a failure at all if you learn something from it.



The circuit, being built on a prototyping PCB without the benefit of ground planes and restricting layout, I actually expected it to have a noisy output. It is just that the output noise was far worse than I had hoped for. The problematic noise is the ringing that occurs during the switching transitions that oscillates with about a 100MHz frequency. The main switching inductor is the natural suspect that causes this problem, and it is actually easy to verify this.

If I clip the ground wire of my oscilloscope probe on its tip, I actually form a loop antenna capable of picking magnetic fields that happens to pass through the loop. The picture below shows what the probe loop picked up when it was put over the power supply PCB.





The toroidal inductor is leaking too much stray fields. Most of this stray energy are coupled with the components around it. There is very little we can do about it; this problem can be solved only by replacing the coil with one that has more adequate shielding.

We have to accept the fact the switching noise cannot be fully eliminated with SMPS circuit. The question is more on how much noise can be considered as acceptable? To answer this, I picked one similar simple switcher made by a reputable US company, and measured its output noise under similar test conditions, so that we can have something we can use as a baseline for comparison.

This simple switcher is based on LM2576S. The power inductor is the black cube with 101 (100uH)  markings. The PCB copper layout can be clearly seen in this photo.



During the tests, a 470uF low ESR capacitor was connected through J10 pads, where the regulator output is also drawn out.

This is the actual oscilloscope waveforms taken from the J10 output.



TDS784D Oscilloscope is band limited to 250MHz, with the vertical channel 1 set to 50mV/div ac coupled. Tek P6139A 1:10 probes are used. Trace shows ripple voltage of about 50mVp-p, with ringing appearing as spikes with 300mVp-p amplitude.

Is this noise low enough? I don't know, really .  Especially after I watched the video of Jim Williams measuring his SMPS circuit churning out noise at the microvolt level!
http://www.linear.com/designtools/video/ltchannel_switchregnoise.jsp

I seldom build SMPS circuits and lack the experience to say anything with confidence in this field.
Master Glutnix is the authority on this subject. I hope mabasa nya ito and throw in his expertise to further educate us. 

[rdpzycho]

ang galing niyo talaga sir Born. previous work, meron na akong coax cable na naka-loop na kaagad for the same type of measurement. then common mode noise measurement naman using closely grounded oscilloscope probe.

tama kayo sir, mas maraming natututunan sa failed circuit than outright working circuit. at 'yung matututunan dito madali na i-apply sa next project.

[maldihtah13]

Subscribing. 

Wala bang magshashare ng LM2577 to LM2579 na home made dito?

Dami ko kasi stock pero di ko pa nagagawa dahil sa limited parts.

[Born2BeWired]

The ringing 'spikes', being high frequency (100MHz) in nature, may not be a problem at all in some applications. Examples of this will be with CMOS and LSTTL circuits, and low frequency circuits in general. These circuits generally do not respond to signals in the 100MHz region (Note: Some LSTTL gates do). To illustrate this further, if we check the noise output with a 20MHz oscilloscope, the spike can hardly be seen.

The truth is, I already used A8447ST several times, largely ignoring the presence of this noise, since it does not interfere with the operation of the rest of my circuit (microcontroller circuit).

For low noise and high speed circuits, this noise could be a major source of trouble.

This case will be an interesting one to solve. Not for the sake of the circuit itself, because I thought, it will give everyone a glimpse of some of the tools of the trade, techniques, in solving a problematic circuit.

I already revealed some technique, which, kung nasa dating trabaho pa ako, will give an excuse for my boss to kill me, dahil considered na trade secret ang mga ito. Master rdp mentioned the coax loop, another tool that cost next to nothing, yet so effective.

I hope yung ibang masters natin, especially master Glutnix, will join us. Of course, do not reveal information that will get you in trouble 

[subzer0]

buck regulator ba to sir born?anu po yung switching frequency nyang ginawa mo po?

[TinTopHack]

Born, can you publish the schematics here? I do not have much experience with SMPS but in my view radiated switching noise from the inductor is just the effect. It will not be that big if it is small at the source. The real cause is the device that is switching on and off, the circuit around it and the layout. In my short stint with a company that makes SMPS, meeting output noise specs and EMI specs was a real challenge. A lot of circuit modifications have to be made after the first few boards are made and tested to reduce output noise and EMI. One thing nice about switching spikes is that the source can usually be easily identified. Once identified, you can apply the appropriate method of suppression or you can search for a new method.

[motion55]

@Born2BeWired

It would be interesting to take a scope reading by running a twisted pair of wires from the output to the scope. One end of the twisted pair soldered to one of the output capacitors and the scope probe and ground connects on the far end.

[e.novacek]

Inductor usually contribute some "noise" magnetic field generator and hence, it is a good practice to shield the inductor.. take note also the brand of inductor.. the higher the current rating, the lower the noise generated ..

I think the "ringing spikes" that occurs in the circuit .. i will coin the term "frequency noise at low-level" for clarity.. has something to do with the "magnetostriction" , "gibbs phenomenon" and "air gaps"..

There are some IC's (or regulators specifically) available that emits "Low-EMI" noise

In communication systems, noise voltage can be calculated and should be in the micro-volts level under normal circumstances..

Therefore, if the noise voltage under SMPS is "under the micro-volts umbrella" then the noise level is acceptable ..

[TinTopHack]

@Born2BeWired

It would be interesting to take a scope reading by running a twisted pair of wires from the output to the scope. One end of the twisted pair soldered to one of the output capacitors and the scope probe and ground connects on the far end.

Bro baka maktulong din to. Kasi minsan ang twisted pair is not good enough. Please make sure the wires are as short as possible. We used this method in my old job.

[subzer0]

pero sa tigin ko po mas ok po pag X1 lang po ng probe at AC coupled at 20Mhz BW is farily enough tapos lagyan din ng kahit low esr na  cap across output(typical 0.1uF na ceramic ay ok but mas ok pag 1uF na tantalum type) pag nagmemeasure ng output ripple.saka masyadong atang malaki yung toroid kasi base dun sa mataas na switching frequency kailangang maliit lang yung toroid inductor kasi yun yung purpose ng high switching freq to maximize the power density at dapat maliit din yung area ng loop traces ng PCB kumbaga madami ng parasitics considerations pag masyadong malaki yung loop ng PCB traces.maingay po talaga smps nakikita lalo ng approaching MHz sa switching freq kaya minsan ginagawa nilang ishield yung switching devices using heatsink pero sa nakikita ko naman dun sa board na isa ok naman layout nya pero malaki ripple parin kulang pa ng cap sa output.liliit po yung ripple using low esr output caps,small loop area for pcb layout traces at syempre compensation para kahit anong noise na papasok,stable parin yung output  but pag high freq switching freq better to use resonant type kasi pag PWM lang na hard switching malaki switching losses but disadvantages naman ng resonant,efficient lang po pag malapit na magfull load.

[motion55]

I don't think it's the main inductor that is at fault here. The pulse is too fast (@100Mhz) for the 100uH sized coil. It is probably in the freewheeling diode and the lead inductance and the IC lead inductances. Remember the diode has to suddenly conduct 1-3Amps of current and while conducting current suddenly a reverse voltage is applied to it. You can see those pulses in a circuit simulator. For example the diode reverse recovery alone manifests as a huge pulse of current through the diode.

[marcelino]

ayos, ganda ng discussion dito!

[mvs sarma]

with a toroid core shielding is not an issue, as feared by e.novacek, I feel..

[subzer0]

I don't think it's the main inductor that is at fault here. The pulse is too fast (@100Mhz) for the 100uH sized coil. It is probably in the freewheeling diode and the lead inductance and the IC lead inductances. Remember the diode has to suddenly conduct 1-3Amps of current and while conducting current suddenly a reverse voltage is applied to it. You can see those pulses in a circuit simulator. For example the diode reverse recovery alone manifests as a huge pulse of current through the diode.

magriring po talaga pag di sufficient yung reverse recovery ng diode,halimbawa 25ns ultrafast na ito,typical kaya na neto yung dynamic response na 100A/us.may naencounter akong problem regarding this,may nakasakay na low freq kaya nagsusuffer yung output ripple bagsak,tapos yun nakapagtataka,yun pala nagsubstitute ako,ng mas mataas na trr dun sa catch rectifier ng diode,nung binalik ko sa dati ok na.yung sa 50mVpp na ripple ni sir born masyadong malaki ripple,typical nasusukat ko mga less than 20 to 10 mVpp,flyback topology with output of 5V @ 10 A.

[subzer0]

sample ng ripple measurement  5V @ 10A na nasukat ko po,yung set-up gaya po kay sir TinTopHack,wala na akong nilagay na cap across,16mVpp ang result.

[Born2BeWired]

buck regulator ba to sir born?anu po yung switching frequency nyang ginawa mo po?

I don't even know kung anong tawag nila sa topology ng regulator na ito, but I believe it is a step down buck regulator. Around 380kHz and switching frequency nya sir.

Born, can you publish the schematics here? I do not have much experience with SMPS but in my view radiated switching noise from the inductor is just the effect. It will not be that big if it is small at the source. The real cause is the device that is switching on and off, the circuit around it and the layout. In my short stint with a company that makes SMPS, meeting output noise specs and EMI specs was a real challenge. A lot of circuit modifications have to be made after the first few boards are made and tested to reduce output noise and EMI. One thing nice about switching spikes is that the source can usually be easily identified. Once identified, you can apply the appropriate method of suppression or you can search for a new method.

How can I miss the schematic! Thanks Sir TipTop, I will post the schematic shortly.

It would be interesting to take a scope reading by running a twisted pair of wires from the output to the scope. One end of the twisted pair soldered to one of the output capacitors and the scope probe and ground connects on the far end.

Thanks sir motion, I will try that. Nasa workshop@home ko yung prototype unit, I will post the result later tonight: If my pldt dsl connection be kind enough to give me an internet access. (My pldt dsl do not work during peak hours. Parang yung dating mwss, you have to wait till the wee hours to get the service.)

[rdpzycho]

If my pldt dsl connection be kind enough to give me an internet access. (My pldt dsl do not work during peak hours. Parang yung dating mwss, you have to wait till the wee hours to get the service.)

OT: same here sir. nagpalit na rin sila ng mga lines dati pero ganun pa rin disconnected every 5 to 15 minutes from 10PM to 1AM.

[Born2BeWired]

Inductor usually contribute some "noise" magnetic field generator and hence, it is a good practice to shield the inductor.. take note also the brand of inductor.. the higher the current rating, the lower the noise generated ..

I think the "ringing spikes" that occurs in the circuit .. i will coin the term "frequency noise at low-level" for clarity.. has something to do with the "magnetostriction" , "gibbs phenomenon" and "air gaps"..

There are some IC's (or regulators specifically) available that emits "Low-EMI" noise

In communication systems, noise voltage can be calculated and should be in the micro-volts level under normal circumstances..

Therefore, if the noise voltage under SMPS is "under the micro-volts umbrella" then the noise level is acceptable ..

Yes sir, microvolts is what we want. Getting there is the problem. Sa ngayon, we are measuring 50mV ripples, accentuated by 100Mhz ringing noise. Right now, I am more bothered with the ringing noise, since it swings at several hundred millivolts level. I thought the ripple will be much easier to deal with, hence we can save it later for our anti climax.

Bro baka maktulong din to. Kasi minsan ang twisted pair is not good enough. Please make sure the wires are as short as possible. We used this method in my old job.

That's exactly the setup I am using when measuring the noise. The tek probe outer cover can be remove to expose the ground shield, and I take measurements by getting the output test points in direct contact with the probe tip and ground shell.

pero sa tigin ko po mas ok po pag X1 lang po ng probe at AC coupled at 20Mhz BW is farily enough tapos lagyan din ng kahit low esr na  cap across output(typical 0.1uF na ceramic ay ok but mas ok pag 1uF na tantalum type) pag nagmemeasure ng output ripple.saka masyadong atang malaki yung toroid kasi base dun sa mataas na switching frequency kailangang maliit lang yung toroid inductor kasi yun yung purpose ng high switching freq to maximize the power density at dapat maliit din yung area ng loop traces ng PCB kumbaga madami ng parasitics considerations pag masyadong malaki yung loop ng PCB traces.maingay po talaga smps nakikita lalo ng approaching MHz sa switching freq kaya minsan ginagawa nilang ishield yung switching devices using heatsink pero sa nakikita ko naman dun sa board na isa ok naman layout nya pero malaki ripple parin kulang pa ng cap sa output.liliit po yung ripple using low esr output caps,small loop area for pcb layout traces at syempre compensation para kahit anong noise na papasok,stable parin yung output  but pag high freq switching freq better to use resonant type kasi pag PWM lang na hard switching malaki switching losses but disadvantages naman ng resonant,efficient lang po pag malapit na magfull load.

sample ng ripple measurement  5V @ 10A na nasukat ko po,yung set-up gaya po kay sir TinTopHack,wala na akong nilagay na cap across,16mVpp ang result.

I can see you are indeed using a X1 probe, as your scope setting gives out a 1mV/div setting. You think noise signals above 20MHz is not that important, am I getting it right sir? It appears you are currently working with a SMPS manufacturing company. Can you give us an idea what level of noise, and and up to what bandwidth, is considered acceptable in the industry?

[Born2BeWired]

I don't think it's the main inductor that is at fault here. The pulse is too fast (@100Mhz) for the 100uH sized coil. It is probably in the freewheeling diode and the lead inductance and the IC lead inductances. Remember the diode has to suddenly conduct 1-3Amps of current and while conducting current suddenly a reverse voltage is applied to it. You can see those pulses in a circuit simulator. For example the diode reverse recovery alone manifests as a huge pulse of current through the diode.

The SMPS switching frequency is around 380KHz. I agree with you sir, the lead inductance could be the culprit. The A887ST is actually available in soic-8 package only. To work this on a prototyping board, I had to use an soic to dip conveter board, and then an IC socket. All these contributes to an increase in unwanted inductance.

There is that real possibility that in the end, we just arrive with a conclusion that it is not possible, or may be just incredibly difficult, to build a SMPS with acceptable output noise in a prototyping board. But I am not giving up yet, what I am learning right now is worth the effort

with a toroid core shielding is not an issue, as feared by e.novacek, I feel..

I also thought the toroid will behave well, but my initial test showed otherwise. But I do not discount the  possibility that I may have missed something..

[subzer0]

depende po talaga sa apps,sir born,sorry nalito ako dito sa discussion,iba naiisip ko regulation vs ripple,ripple po pala yung problem hehehe,btw una kasi basically we have to trouleshoot muna sa regulation kumabaga panghuli na po yung ripple,typical mga ripple consideration nasa 1% ng output(e.g. 5V dapat huwag lalampas sa 50mVpp).yung ripple kasi inherent na talaga yan sa switching elements(meron ata sa power supply magazine ripple theorem by vorperian,i think di ko na matandaan,sa CCM discussion paano nagkakaroon ng ripple)sa mga mosfets at catch diodes,dun naman sa sinabi ko about trr,basically sa budget loss malaki effect nya,sa ripple di naman masyado malaki effect,sa mga switch talaga kumbaga inherent na po yan talaga,pero dun sa nakikita ko sa sukat mo sir kung 5V ang output at 50mVpp marginal po yan dapat po talaga mababa,thanks nga po pala dun sa video ni jim williams sir born,about pala dun sa BW na 100Mhz na gusto mo,masyadong tight po yan,sa mga VRM modules typical apps po yan,kaya po nasabi ko na 20Mhz fairly enough as industry based.
sa nakikita kung solution,gagamit ka po talaga ng low esr caps to lower ripple noise.