measuring rail currents on a board

You should be able to do something with a Hall effect current sensor.

The brutal solution would be to cut a rail, and bridge the gap with the Hall sensor.

offers a 1.2 mOhm resistance which shouldn't be a problem.

Just glueing the Hall sensor onto the track ought to work almost as well. I haven't done it so I can't make any promises.

all sensor.

I haven't done it so I can't make any promises.

did you not see that it was ~50mA ?

that hall sensor has twice that just in noise

Yeah, when you forget the zero-ohm jumper on the output of each power supply, life gets more difficult.

When bringing up the first article, I always disconnect the supplies to make sure they're right before possibly blowing up everything downstream.

For high current stuff, the Keystone 5110 is a solid metal 0603 jumper that will take way more amps than the PCB traces. I use it mostly in thermoelectric cooler supplies.

Cheers

Phil Hobbs

Sometimes I put a 1 ohm, or 0.1 ohm, resistor in series with linear reg inputs. Maybe add a cap to ground and get some added noise filtering. I didn't do it on this board for some reason.

I did include an interesting TDR test trace, SMAs on the right. It crosses some power-pour boundaries with no visible effect, and has vias of various sizes, which has effects.

Where should I put the Hall sensor on a big layer 5 power pour?

I could have used a linear reg for the FPGA core power. I had no expectation it would be so low.

I used a 1206 and 1210 jumper on my latest board. I was intending to use thick film resistors but didn't realize how badly they suck. I ended up ordering the Keystone parts. Expensive but they'll take something like 50A. That ought to do it (there are two in parallel).

Hall sensor.

l. I haven't done it so I can't make any promises.

It didn't register (and it should have).

There are others. I'm not going to do an exhaustive search to help John Lar kin solve his problems - it's not as if he'd read anything that I posted.

The situation is that the customer reserved the right to conduct their own design review, with a schedule milestone date. The lead engineer claimed that he did, and OKd us to go ahead with the board layout, which we did. Everything worked first time. But the guy quit and the new team did another review, after working units were shipped, and gave us a giant (Excel, of course... these people think in Excel) list of items, caregorized as MUST and SHOULD and NICE TO HAVE. Among other things, they want a full power analysis, as an Excel spred sheet of course. I figured it would be easier to measure the supply currents. I guess it would be even easier to tell them that the required design review happened months ago.

One of their guys apparently read the HoJo Black Magic books, with predictably absurd results, like whining about traces crossing power plane transitions and wanting us to add more ground planes. Or complaining about silly bypassing things. They want to see a big cap at the input of every linear regulator *on the schematic*, where the reg inputs are all on one big, well bypassed, power pour.

I did include a TDR test trace that zigs through all the signal layers and crosses some layer 5 power plane patches, with small gaps. The pour crossings are invisible on 20 GHz TDR, of course. But I did include a variety of vias, which was interesting. The smallest via was the best to maintain a smooth 50 ohm path between layers.

The Saturn PCB software calculates via capacitance, which is useful on another project. This is a special 0603 inductor footprint:

where we'll disappear one normal thru-hole pad for a photodiode and solder the lead directly on the inductor. Saturn says that the thru-hole pad would be 8 pF.

I remember a miniature field sensor ages ago, and expensive. It was so simple that anyone could build one. I think it used a small (RF?) inductor inside a copper tube that had a slit on the side to prevent shorting of the secondary EMF.

Obviously, that would be OK for spikes and RFI detection. So, maybe use one of those 2-hole balun cores, 2 separate windings physically balanced as possible for starters. Set up a push-pull oscillator based on this. Mark-space ratio at zero external field should be 1:1.000 and should be sensitive to external DC magnetic field.

A mag field sensor, like a flux gate, isn't hard. But how could you calibrate power regulator current from mag field?

I guess you could set up the sensor and add extra load to the rail briefly, and see how much the field increases, to get a gain factor. Incremental loading could calibrate other methods too, like measuring voltage drops in traces or maybe planes.

Other nearby currents might wreck a mag field measurement. I have inductors carrying current, too, and they will leak big fields.

One way to measure current is to look at ripple voltage. In a classic rectified AC supply, the discharge slope between line peaks lets you compute current.

My switchers are both running in power-save burp mode, so there are intervals when the load is just being run from the caps on the rail. So I can measure the slope and compute current. One could wiggle the enable pin on a switcher too.

If you use the more advanced DC/DC controllers (such as (e.g.) a PMIC with multiple rails), you can just read a (rough) estimate of the current out over I2C/SMB.

A lot of them also allow the output voltages to be trimmed by writing to registers over I2C. This can be handy for manufacturing tests or investigating the margin in your design.

Regards Allan

You can calculate the magnetic field around a conductor as a function of the current running through it.

Working out where the magnetic field sensor is actually sitting with respect to the conductor might be a bit trickier.

They might. They are mostly designed not to, because if you provide a low reluctance path around the winding, you get more inductance per ounce of wire.

If you know the capacitance accurately. Reservoir capacitors tend to be loosely toleranced, and can have other vices (like voltage dependent capacitance).

Lots of desperate improvisation there.

Chicken-egg issue: I want to check the hardware, especially the power supplies, before I turn the board over to my programmer to test his software.

But it's too late for that. The customer is demanding a power analysis just to be fussy. It's hard to spot anything warm on a thermal imager.

Has anyone here done the full digital I2C power supply control thing?

Not with one of the fancy chips, but we often dork the feedback node of a SMPS with a DAC or PWM to make class-H TEC drivers. Our fave M0+ chip is the LPC845, which has 12 ADC channels, which makes power supply monitoring easy.

Cheers

Phil Hobbs

In the board in question, four of the analog-section power supplies are DAC trimmed, but none of the things fundamantal to bringing it up.

When we run out of DACs, we PWM.

The LPCs that we have been using for a while are going EOL. The next gen products will be STs. There are some nicer things about the STs, like a megabyte of flash and some nice pin generality, and uniformity across chip families.

The LPC17xx ones? Bummer.

Cheers

Phil Hobbs

The 3250 is officially going. The 17xx parts are estimated good for 5 more years or so, but we're going with ST for new designs.

One of our boards has 13 of the LPC1754s.

The ST library code (especially the headers) is _horrible_, which is why we've never used their MCUs. I gather you folks have some sensible alternative, but what?

Cheers

Phil Hobbs

Paul, my embedded programmer, likes to program things himself, his way. He has his own tcp/ip stack and drivers and web page things.

We have one ST thing running now, the laser controller. No apparent problems.

The ST JTAG pod is slick. It enumerates as a memory stick, so to program the chip you just drag the file into it.