Group Actions
Engagement
  • Author Author: michaelkellett
  • Date Created: Date Created
  • Views 1349 views
  • Likes 14 likes
  • Comments 16 comments
Related
Recommended

Silent Switcher - LT8640s meets KiCad

michaelkellett

The first part of this project kicked off with a blog “First Design with KiCad (8.0)” about 4 weeks ago. The idea was to build a replacement for obsolete TO3 can linear voltage regulators using the Linear Technologies (now owned by Analog Devices) LT8640s low noise switching regulator.

This was my first KiCad project and the pcb design went quite  well.

The boards came from PCBCart in China and I had no issues with them working with the KiCad Gerber and other outputs.

Because the board is an odd shape I had it made in a 1 x 1 panel.

Here are 4 with the solder paste screened on and some of the parts placed.

image

And here they are again after solder reflow.

image

There were two errors in the design. Both were my fault Frowning2

For some reason I connected BST to the output. I can’t think why because the data sheet says:

 image

To save anyone else the trouble of trying it – don’t  - it doesn’t work !

Easily fixed by cutting a track.

The second error was that I made the holes for the two press fit pins too large (Mill-Max 8979-0-00-15-00-00-03-0 from Mouser)

I specified a 1.9mm hole, which is bigger than the pin data sheet says – I should have gone smaller and opened it out if necessary. Not the end of the world, I squeezed the heads of the pins which gave them some grip and then soldered them.

 

Testing

I tested the board without any heat sinking.

I used a pair of Keysight 34465A DMMs to measure input voltage and current. Power was supplied by an AimTTi QL564TP power supply but its readout isn’t terribly accurate. I trusted the Rigol DL3031A active load to measure load voltage and current. It was connected with Kelvin leads using the remote voltage sense capability which would have worked better if I had turned it on !

I was in a hurry to get data so I didn’t automate the measuring but entered it by hand into Excel.

image

The first block of data from lines 3 to 21 is the measured values:

Vout, Iout, Vin and Iin (that’s I in )

The computed values are:

Power out, Power in, E (efficiency), D (dissipation), LinPin (the input power if it had been a linear regulator), LinD (dissipation if it had been a linear regulator) Dsaving (reduction in power input by using a switcher).

The second block of data (lines 24 to 42) is a copy of the first but rounded to three decimals.

image

And then that note !

Looking at Vout (2nd col, lines 4 – 12) the voltage regulation is a bit disappointing. This was because the remote voltage sense on the Rigol load was not turned on (it defaults to off on power up and can’t be made to save the previous state). So the voltage drop along the quite long wires to the Rigol is causing an error. Rather than repeat all the measurements I added a compensation factor to the output power, using this formula for the Pout col:

=A5*B5 + (0.032 * A5 * A5)

So I have added the power lost in the wires to the output power calculated by multiplying the voltage and current as measured by the Rigol load. The resistance of the wires is 0.032R which I measured by comparing the voltage measured by the Rigol with remote sensing turned off and then on.

The output voltage graph is wrong because of this error – it should show a drop of about 40mV at 4A rather than about 180mV (0.8% rather than 3.6%).

The other three graphs are correct and show that the efficiency remains above 90% for output currents up to 3A.

At 4A output the board is getting quite warm and the rise of the case temperature of the LT8640s above ambient is about 55K.  I’ll experiment with a heatsink later.

I’ve tried to show the benefit of using the switcher compared with a linear regulator. The data with a nominal 12V input represents a reasonable design if a simple transformer, rectifier and smoothing capacitor type power supply is used to provide the input power. The approx 1V drop caused by the wires in my test setup is not far from what might be expected from an unregulated supply.

So at 3A load current the switcher reduces the heat dumped into the rest of the system by 20W.

I did some quick noise and transient response measurements.

The Rigol load can do pulsing so I set it up to pulse the current from 1A to 3A for 50ms with a current slew rate of 0.5A/us.

The response transients for load current rising or falling a very similar. They should be treated with caution because active loads have their own transient difficulties and I have not yet checked the Rigol DL3031A  transient behaviour.

image

image

 

The output noise at 1A load current looks OK at 10mV pk-pk

image

 

I’m not totally convinced by the noise (it was measured hastily). Looking with a slower time base shows the nose envelope with some very unexplained modulation at low 10s of kHz frequencies – looks a bit like AM radio !

image

The noise with the scope probe connected to the ground terminal shows unexplained little spikes at about 20kHz repetition rate – I don’t think the LT8640 is making these, but more work is required to be sure.

 image

Summary

This first quick look at the performance of my TO3 shaped LT8840s Silent Switcher board shows that it works fairly well.

It can deliver 3A safely (and 4 at a pinch) with no heat sinking and has acceptable regulation, noise and transient performance.

As you would expect from a switcher it saves a lot of power but this comes at a very high financial cost – the BOM cost of this design is about £21 (allowing £5 each for pcbs). To compare the sublime with the ridiculous, I just bought 20 L7805CV TO220 5V regulators from Amazon – sourced from “BOJACK” for £7.88 (£0.39 each) – they won’t manage 3A and I have no idea who made them. A good ST 7805 in an insulated TO220 package costs about £1.

I’m thinking of building  a much better test rig  for small PSUs. When measuring efficiency we need to compare power in to power out and with good regulators we may be interested in measuring efficiency above 95% and would like better than 0.1% accuracy of the efficiency measurement.

This implies that the voltages and currents should be measured to (quite a lot) better than 0.01% accuracy.

There are very few (if any) bench top DMMs capable measuring 10A DC that accurately.  I have a cunning plan but that’ll need to wait for another blog.

MK

Parents

  • Interesting blog. I really must have a go with KiCAD and see how I get on with it.

    "Looking with a slower time base shows the nose envelope with some very unexplained modulation at low 10s of kHz frequencies – looks a bit like AM radio !"

    Would that be a side effect of the spread-spectrum clock?

    Although the spread spectrum clock will be simply frequency modulated, the cycle time will then be varying and that will interact with the ramping of the inductor current and translate back to a voltage with the ESR of the output caps. So you might then see a small amount of AM modulation on the output. Does that sound plausible?

  • in reply to jc2048

    Are you thinking you could recover the spread spectrum clock? Relaxed Would a spectrum analyzer show what is going on?

Comment Children
  • in reply to dougw

    The sync/mode pin is floating

    image

    So the LT8640s should be in FCM, mode 2, not spread spectrum.

    MK

  • in reply to dougw

    I was just thinking of some of the possible reasons why the AM might be there. But since Michael has the spread spectrum on the clock disabled, it can't be that.

    In reality he's much better situated than us to find out, given that he's got the boards, some nice test equipment, and plenty of experience.

    I don't know much about using spectrum analysers for looking at these kind of issues. To be honest, I don't know much about spectrum analysers at all, never having used one. I imagine protecting the input might be a concern - even with capacitor coupling of the input, there could be a large voltage step as things get connected, before the capacitor charges to remove the dc. Also, RF measuring instruments work with power and assume 50 ohms matching, so that would affect absolute readings, wouldn't it, though for simply understanding what's going on that doesn't necessarily matter. Presumably, a digital 'real-time' analyser might be more useful than a scanned type - I assume on a waterfall display you'd see the modulation of the clock mapped out - but you're pushing the extent of my knowledge here.

    Isn't it simpler to start with the FFT in the 'scope, though?

  • in reply to jc2048

    I just now peeked at the eval board documentation, they too mention using a 'scope for examining the noise, rather than a spectrum analyzer. The main downside of the traditional spectrum analyzer instruments is that most won't measure below approx. 9 kHz.

    As you say, even if a spectrum analyzer was used, the impedance wouldn't match and so the readings would be hard to interpret. So all those are reasons why the 'scope is preferred.

    There is a 'line impedance stabilization network [LISN] that is often used in the reverse scenario, where a load is being tested for emissions, but it can also be used for testing the power supply itself, i.e. the DC-DC converter, connected to a load. This doc discusses it at section 2.5:https://www.tekbox.com/product/TBOH01_Manual.pdf - but since it's usually for EMC compliance testing, that setup still would have the issue that at low frequencies it wouldn't be of use.