Power supply poor performance.

Yes.

Scope shot?

More importantly, what's pin 2 look like? And the current through R1?

Aha! Ferrite is shit for DC filtering. You're probably saturating them. Try a ferrite rod or powdered iron toroid.

How many turns? What's "300uH" based on, is it measured? At what bias?

Layout looks pretty good. D5 and C2 seem to be on the wrong sides, C2 should be closer to the chip I guess, but with L1 where it is, that might be tricky. Maybe it can get just close enough to that mounting hole, or see what it does rotated. Anyway, that ground plane everything's connected to is pretty wide and this should make very little difference. I'd say it's a good tight layout.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

You might want to experiment with Tantalums for C1, C3. And looking at your layout there is potential of coupling the input switching noise to the feed back pin. A small capacitance might help, with out affecting loop response. Also this type of switcher creates a lot of hash, so a input filter of some type (LC) is usefull.

Cheers

Thanks for the reply. I'll try out your suggestions and see what happens.

OK. I'll upload the shots but it'll have to wait a bit.

Ah. I'm not completely ignorant about the saturation thing, but I have limited experience with unbalanced filtering and it just didn't cross my mind. Thanks for the heads-up.

12 turns. Measured without dc bias. The core is something I salvaged from junk.

I could move C2 to the copper side and solder it directly to the IC pin but, as you said, I wouldn't expect that to make a lot of difference. Thanks for the reply.

Couple of observations and tests to try.

1) L1 has high dv/dt and is unshielded so it will radiate noise readily.

2) Have you designed L1 and L2 with DC load in mind? you have to gap a ferrite core or use a powered core with a distributed air gap so the cores doesn't saturate. Magnetics inc has a good little DC inductor design program for this purpose. You might consider shielding that core as well.

3) use X5R/X7R ceramic caps in parallel with the large electrolytics. lower quality ceramic disks (Z5U) are not as good at filtering high frequencies. Even the input capacitor. 0.1uF shoudl be fine.

4) If your second stage is a thin spike filter the indictor is way too large and you're using the wrong type of capacitor. Large inductors have high parasitic capacitance so high frequencies shoot right through them and the electrolytic caps have high ESL/ESR so high frequencies skip right across them. (recommend 30uH with high self resonant frequency, DC current rated, ~1uF X7R)

5) C2 is in a high noise spot next to D5 (ringing) and L1 with the high dV/dt. That trace sets the frequency of the device and is noise sensitive. the current path from the anode of the diode is between C2 and pin 4 as well opening hte potential for noise. I would connect C2 directly between pins 3 and 4.

6) With a light load the inductor goes "discontinuous" and is generally more noisy than continuous mode. With 24mA I'm going to assume you're in discontiuous mode. discontinuous mode isn't bad in itself but something to be aware of.

7) (this is a biggie) When probing for noise connect the tip of the probe to the ground clip and touch this to board ground. If you're seeing a lot of noise there is a lot of common mode noise between the board ground and earth ground. you have to make sure the board ground and scope grounds are at very close potentials so there is no noise so you can believe what you see whne you measure. Also a long ground probe lead can be a problem with switching power supplies. The loop of the wire can pickup noise that is otherwise not there. I take the probe clip off and wrap a 24-26AWG bus wire around the top of the probe and leave a tail just long enough the reach the nearest ground form where I'm probing. I've seen measures noise reductions by a factor of 10 or better using this method vs the wired ground probe.

Give those a try and let us know what you find.

The ol 'shorted' probe test?

This is more about RFI in general than common mode specifically. When it switches, current loops and voltage loops throw off electromagnetic radiation. With the probe anywhere in the near field, you'll pick up a delicious burst of switching noise. The solution is a coax probe, where possible. For instance,

The red/black twisted pair running up left of center goes to a 0.47 ohm current sense resistor. It sees this waveform: regardless of how the 10x probe is connected (notice the clip with wire above the heatsink, and its ground clip coming across above the transformer).

If I tried watching the same current with the 10x probe, it would be unrecognizable from all the trash. Notice the complete absence of ringing and hash when the transistor switches, or when the secondary diode turns off (discontinuous mode).

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

I'm especially interested in the inductor core situation. If you do change it, and it helps, let us know. There are some other good points made but I thought I'd focus on the core, because I'm still learning about these details and wanted to consider it a little.

You mentioned 300uH measured and 24mA load and 12 turns on the inductor. With ferrite, let's use a Bsat = 0.1T to be safer. We assume not to know mu_r, for now. That still tells a lot.

L = mu_0 * mu_r * N^2 * A_e / l_m

but also,

Bsat = mu_0 * mu_r * N * I / l_m

This second equation can be solved to place the unknowns on one side and the knowns on the other:

(mu_r / l_m) = Bsat / (mu_0 * N * I)

That can be stuffed into L, as:

L = N * A_e * Bsat / I

But we know L, so solve for the unknown A_e as:

A_e = L * I / (N * Bsat)

Stuffing in L=300e-6, I=24e-3, N=12, and Bsat=0.1 yields a value for A_e of 6e-6 m^2. Assuming a circular profile, this is about a radius of 1.4mm or a diameter of 2.8mm. However, I don't imagine that the average load current is the peak inductor current. So this is actually too small to avoid saturation -- the diameter will need to be greater, I think.

L goes proportional to N^2 while B goes proportional to N, so if you reduce the mu_r of your core you will only have to wind more turns by the ratio of change in mu_r and reduces B allowing more Ipeak. So you might consider finding a significantly lower mu_r core and winding a few more turns to get back to your L value. Here, I'm talking about L1, I think. L2 looks like part of a filter to me so I have little comment about that, now.

Jon

accurately measuring the spikes at the output of a switchers is an art and a science.

The loop formed by the scope probe itself and the ground clip can even pick up spikes that are not really on the output.

Connect the scope probe to the ground where the ground clip is connected and you should see nothing but you will probably still see the spikes. This indicates that you need to improve the measurement.

I'm too lazy to search the web for you but you can look around the web yourself for techniques to measure ripple of switches and see the various techniques of filtering or common mode rejection that are suggested to get a more accurate measurement.

To be clear, I am saying that the spikes you are seeing on the scope are worse then are really there due to measurment technique issues.

Mark

You did attach a load to the output? Compare the input to the regulator to the output at no load and then try adding a load of maybe 10% of max and see if it does any better. I don't see why you need a second L-C stage. The led may not be a large enough load.

Also, your circuit seems a bit different than the buck in the datasheet. Some of your resistor values are different(not sure if thats intentional or not). Other than the the circuit looks correct.

On page 7 they give the characteristics of that circuit so you should be seeing approximately the same(again, some of your component values are different(R2 and R3).

Try a larger load of at least 100mA or even shorted and see what you get. As has been mentioned, in an smps design, the output depends on the duty cycle. For small loads it is non-linear and the duty is significant. For large loads the regulation is almost independent of the duty. Hence the first thing is try a larger load and see if that improves anything.

I have a small 0.5W NCP3063 pcb inverting. Heres a shot of the input and output ripple below.

Is this what your output looks like?

That's just using a bead and a cheap 100uf AL cap for my post LC filter. Yellow is input ripple; blue is output ripple 38mVpp full load. Are you talking about those skinny spikes riding the ripple on the blue trace? I'm measuring this with a short ground pin on my probe and right over a 0.1uf X7R 0603 cap right at the output.

You're always going to have weird ripple on Hysteric controllers but yours does sound really high. The inductance on your post LC filter seems unnecessarily large as well.

The advice mook gave is good particularly when measuring with your scope.Your probe acts like an antenna and picks up switching noise.

Try holding your probe in the air and keep moving it closer and you will see waveforms on your scope without even touching the circuit.

That implies A_L = 2.08 uH/T^2 (unbiased), which is pretty high, typical for an ungapped high-mu ferrite toroid. They're typically around 1-10 amp-turns (AT) saturation, pretty easy to saturate. If the peaks are four times higher than average current, that's easily 0.024A * 12T * 4 = 1.15AT, which might be enough to saturate it. If it's highly discontinuous, the peaks could be much higher, bringing it into saturation. And if it is discontinuous, that could explain the unusually low frequency.

Bsat ~ 0.4T is more typical, though you might want to drop it to 0.2 or 0.1 for better linearity, or for high frequency use (transformer duty only, DC choke is different). I usually go with ~0.2T, which is a practical factor-of-2 headroom, just so I can say I've made the allowance.

Notice that assuming peak B is equivalent to a certain amount of applied voltseconds. This is simply because EMF = -dB/dt * A_e, or Vs = -B * A_e. In terms of easy-to-measure parameters, you can get AT(sat) and A_L from the inductor's V-I curve, which gets you V*s = AT(sat) * A_L / N.

I like to work with Vs because it's more useful to circuit analysis. How many turns do I need? Integrate voltage over a quarter wave period to get V.s (it's usually a square wave, so that's just Vpk * 4/f), and I've already measured the core's A_L and AT(sat), so just divide and you get turns.

In this case, if AT(sat) = 1AT and A_L = 2uH/T^2, then Vs = 2*1 * 12 =

24uVs. At 9V, that's only 2.7us on-time. Pawihte, are you seeing ~3us wide pulses?

Notice that mu_r disappears -- that means it doesn't matter how much magnetization you apply, it's the volts and time that gets it up to saturation. (You could do the same to a powdered iron core (let's say

0.1uH/T^2 and 300AT(sat)), although magnetization current gets absurdly large for transformer duty!)

L2 has to carry the same current, so it'll need the same characteristics, although it has less delta B, which basically means it can be lossier = cheaper (one of those ugly yellow/white toroids?).

Powdered iron cores that size are in the 80nH/T^2 range, so you need about sqrt(300 / 0.08) = 61 turns to get there. Saturation is in the >200AT range, so you can safely dump over 3A though it -- more likely you won't even be able to get enough turns inside it to see it saturate before I^2*R losses overwhelm it.

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Oh, I remembered something else about probes and grounding:

If you put the probe itself through a ferrite bead, you can observe its effect, if any. Ideally, this won't change anything. If it changes, then you have current going somewhere it shouldn't, either up through the mains, or between probes (if you're using more than one).

Last time I did this and saw the effect, I got:

- No bead: fast ringing, medium amplitude, medium decay

- With bead: slow ringing, same amplitude, medium decay

- Damped bead: slow ringing, low amplitude, fast decay

Bead setup: three turns through large ferrite bead. Last test: 10 ohm resistor soldered in place through the ferrite bead (so it's acting like a leaky shorted turn).

If you are getting probe current, adding a ferrite bead doesn't really help your measurement, but it does change it, so you can at least guess what hides behind the "probe cable resonance" jigglies.

BTW, I happen to have some 3" high-mu ferrite toroids, which are *beautiful* for putting a couple turns around a big fat probe. (They're supposedly

12uH/T^2, so 3T will get over 100uH easily!)

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

You're right. I touched the probe tip to the ground-clip and the spikes remained almost the same. In fact, I should have thought of this myself. Although it must be obvious that I have large gaps in my knowledge, I've certainly been aware of such induced pickups for a long time - decades actually. My only excuse is that it was already the small hours of the morning when I ran the trial. Thanks for pointing it out.

As a quick test, I placed the whole thing inside a tincan. I wrapped the whole thing, including the mains transformer, in bubble plastic wrapping without even grounding the circuit to the can. Only the 9V output lines (7" of flex) and the mains wire were outside. The spikes dropped from about +/- 2V to +/- 20mV !

I'll check out the suggestions made by others and report back.

Correction: I was too quick to draw a conclusion. I took out the whole thing from the tincan again, but the spikes are still only

+/-20mV. (This time I left the scope probe and ground clip attached to the ends of the output wires). I further observed that even slightly moving the setup (millimeters) caused the spikes to shoot up again, and they remain high until I turn it off and then on again. Even just touching the wires have an effect. It seems something's unstable.

I increased the load to ~0.1A, then to 0.6A. The spikes remained fairly stable at approx +/-10-20mV until I momentarily touch the output +, at which point the spikes shoot up again until I turn it off and on. Normal ripple, which was effectively nil at light load, was about 5mVp-p at 0.6A.

Frequency is now about 45kHz. The ~15kHz I reported earlier was probably an observation error on my part. There was a lot of jitter. I think it's time to work on the inductors.