What has become of this? (Low noise DC/DC cnv)

Wow, that's a long link! It's message-ID and here's my attempt at a shorter Google link:

thread/bb3edbc02ec23260

Click on view as tree for better navigating.

My project is on hold right now, since I lack the time to work on this problem. An efficient simple 1MHz level-controlled sine-wave power oscillator could be a critical part of the solution.

Here's a copy of my Aug 4, 2004 post:

My dc-dc converter specs: 5Vdc in, 1/2 to 1W max capability, three simultaneous "unregulated" output voltages, small, very low noise, high switching frequency (600kHz min to 5MHz better, etc.) -- by "unregulated," I mean fixed input-to-output voltage ratios, 10-20% tolerance allowed (if necessary I can adjust transformer windings, and the loads will be constant).

I'm looking at Linear Technology's LT3439, almost exactly what I need, push-pull transformer drive, etc., but yet misses by a mile.

The LT3439 is one of LTC's line of slew-rate controlled low-noise switching converters, but like the others it's limited to 250 or 300kHz, far too low for my application, in which I need to put the switching noise well above my sensitive amplifier circuit's range.

Does anyone know a switching IC that's similar to the LT3439 but runs at MHz frequencies? Maybe with sine-wave transformer drive?

--
 Thanks,
    - Win

First, I should say a very interesting discussion and a number of excellent responses are archived in the above thread; as a result the ball is solidly in my court (and now also in Jeroen's court).

I don't recall among the excellent suggestions anyone mentioning a high-frequency resonant-mode controller IC. As it happens, I have worked with an excellent TI phase-shift resonant mode controller at 1MHz, so I can tell you they are very noisy beasts! In addition to the high switching noise of a fast H-bridge, there's real potential for serious sub-harmonic noise. That's because these circuits must absolutely be operated in a feedback-loop mode to control the output voltage, and the loop always has minor (or major) instabilities, perhaps in part because of the noisy environment. No, I wouldn't want to try going down that route. But thanks for the suggestion.

--
 Thanks,
    - Win

Good luck on finding a chip. To get the most market, chips are designed to be application catch-alls. Every design I did for low noise required custom circuits. You can use some of the pieces in the chips as part of the control loop. They always have references, gain blocks and switches.

Again, good luck.

Steve.

--
Steven D. Swift, novatech@eskimo.com, http://www.novatech-instr.com
NOVATECH INSTRUMENTS, INC.      P.O. Box 55997
206.301.8986, fax 206.363.4367  Seattle, Washington 98155 USA

How about a resonant mode (zero voltage switch / zero current switch) controller chip? These go well beyond 1MHz at the power levels you require.

--
Reply to nico@nctdevpuntnl (punt=.)
Bedrijven en winkels vindt U op www.adresboekje.nl

I read in sci.electronics.design that Jeroen wrote (in ) about 'Low noise DC/DC converter', on Sun, 26 Dec 2004:

In days of yore, this sort of thing was done with an r.f. oscillator. Class C push-pull, or, for very low noise at the expense of efficiency, Class A.

Unless high efficiency is a major requirement, I don't see why anyone should think 'low noise' and 'switching' simultaneously in the same brain.

--
Regards, John Woodgate, OOO - Own Opinions Only. 
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

I'm currently exploring the interesting details of building and designing transformers. I need an isolated 5V to -15V/+15V/3V3 convertor, that should be very low noise, as it supplies a 12 bits ADC with additional signal conditioning circuits. Currently I'm using two C&D Tech. 1W DC/DC bricks followed by low noise LDOs. The noise (or better, the switching residue) can generate errors up to 48 ADC counts. Actually these little bricks seem to generate more noise then three switchers in the digital part of the board (a LT1940 and a LT3431 to create a 1V5/1.4A, 3V3/1.4A and 5V/3A supply).

I'm going to try the LT3439 (if those samples ever arrive...) or otherwise I give up my pride and use a off-the-shelf 1"x2" Beta-Dyne brick. Currently I'm still having fun with magnetics and a very simple push pull drive circuit.

Jeroen

Indeed, it's not too hard to get the switching-converter output noise voltage basically down to zero, when measured across that last filter capacitor with a good probing system. But a short distance away, where used, the noise may well be high again, due to several causes. One is noisy ground currents. Another is noisy magnetic fields. Either can be a killer and can be VERY hard to completely eliminate.

A nice reference.

--
 Thanks,
    - Win

should

can

(a

Hi there.

Have you tried hanging one or more extra LC low pass filter(s) on the output of the DC/DC converter modules? In my experience playing with DC/DC converters this extra LC filter can sometimes greatly improve at least the aesthetic appearance of the output voltage as viewed on an oscilloscope. On the other hand, I have been significantly disappointed by the noise reducing capability of hanging a linear regulator on the output of the switching supply. A typical DC/DC converter runs at 100kHz+, so presumably most of the noise produced by it will be at that frequency and higher. Unfortunately the supply voltage rejection of normal linear regulators seems to be fairly low at 100kHz and even lower in the MHz range which (at least for me) is the most aesthetically unpleasing noise to look at on the oscilloscope.

What are the relevant frequencies in your circuit?

Are you familiar with and have addressed the other considerations when using high resolution AD converters?

output

On

reducing

seems

using

For some more good reading I might suggest this application note as well:

What are you using for the reference voltages on your ADC?

designing

to

board

output

On

reducing

seems

I did include LC filters on the inputs and outputs and I used low noise linear post regulators. It's a six layer board with two ground planes. With very close grounding measuring techniques I still have nice 100mVp-p spikes on my scope reading, nicely coincident with the switching of the regulators (there are a few test pins on them). Even with the 512x averaging mode on the scope these spikes stand out in an otherwise rock stable supply.

These spikes contains a lot of higher harmonics that simply shoot through every parasatic component of each part, including the filter inductors (the winding-to-window capacitance). I think the operating frequency of the DC/DC is of no point here; you either end up with 200.000 spikes per second or

2.000.000 spikes per second. The problem is even worsened by having two of these DC/DC that are not synced, they both run at approx. 100KHz depending on input voltage. The sampling rate of the ADC is around 37-40 ksamples/sec. Digital noise is of no concern in this design; all digital parts are non-operative while it's converting (it's a serial DAC).

I think no amount of mu shielding, bypassing, inductors, beads and ground planes will kill these spikes; it's simply easier not to generate these spikes to begin with.

"Jeroen" schreef in bericht news:41d08c45$0$6207$ snipped-for-privacy@news.xsall.nl...

[snip]

Would it be an idea to turn off the switcher while doing the ADC conversion, using a large enough storage capacitor to power the ADC?

--
Thanks, Frank.
(remove 'q' and 'invalid' when replying by email)

Are you using shielded inductors and star-grounding? It sounds like you have a serious magnetic induction problem. Then again oversized switching inductors can be used to kill the current ripple. Your scope reading spikes are almost certainly magnetically induced to get 100mVpp on close GND readings- this can't happen unless your switching inductors have some hellacious parasite shunt capacitance. It looks like shielded components, physical separation, differential signaling in critical paths, and optimal pickup loop minimization everywhere, are in order.

"Frank Bemelman" wrote

You may be able to take the measurements during the switcher's idle periods.

The noise from the switcher will be common mode (or should be common mode). You might be surprised at how well it is possible to measure in the presence of horrific common mode noise. Or you may be surprised at how badly you can measure ...

Accurately measuring switcher noise with a scope is not a simple task. Hooking a scope probe to the circuit will show gawd-awful noise that really isn't there. A very high speed differential probe is what's needed. A pretty good measurement can be had by _soldering_ a heavy duty coax (RG58 A/U works, if my memory serves) to the board with the exposed center conductor and braid as short as possible (V. important). By rights a terminator needs to be used at the scope, but if you keep the cable short it really isn't needed - cable prop time is a ~nanosecond/foot.

Can you quickly bastardize a test measurement circuit? I would give it a try before launching a massive switcher development effort.

--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer:  Electronics; Informatics; Photonics.
Remove spaces etc. to reply: n o lindan at net com dot com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/

ground

conversion,

I gave that a quick go, but did not work well. The operating frequency and the sampling frequency are quite close together, 37khz and 100khz. The problem is that I have two DC/DCs, which run async. If they ran in sync, I could synchronize the sampling with the DC/DCs.

Jeroen

conversion,

important).

I can set the impedance on the scope (a TDS3034) itself to 50ohms instead of

1M, would that work too?

"Jeroen" wrote

Yes. Check in the manual for the max DC voltage: 10 volts will dissipate

2 watts.

You won't need a terminator with a short cable, though.

If you keep the cable in the 1 foot range then reflections won't be seen with a 200MHz scope. FWIW signals travel down RG58 at 0.66c, c ~= 1 ns/foot, so the first fundamental of a 1' cable is .66 / 2 nS =

330 MHz.

--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer:  Electronics; Informatics; Photonics.
Remove spaces etc. to reply: n o lindan at net com dot com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/

You may want to 'float' the scope: disconnect the ground pin on the power cable. In the US there are adapters that do this (well actually used to add a grounded plug to an ungrounded household socket, almost never used for this purpose.)

Some scopes already come with isolated signal ground.

--
Nicholas O. Lindan, Cleveland, Ohio
Consulting Engineer:  Electronics; Informatics; Photonics.
Remove spaces etc. to reply: n o lindan at net com dot com
psst.. want to buy an f-stop timer? nolindan.com/da/fstop/

Have you ever tried a common mode transformer (as used in mains sections of power supplies)? Together with a capacitor these give a lot of attenuation of high frequency components and are way more effective than a choke.

--
Reply to nico@nctdevpuntnl (punt=.)
Bedrijven en winkels vindt U op www.adresboekje.nl

An oscilloscope can be deceiving, the spikes can also be induced in the probe (and the ground lead). I did some measurements on a 100Ms/s

10bit ADC recently and found the digitized signal was quite clean (given the circumstances; a prototype PCB with a ground plane, one layer for components and several wires). Connecting the oscilloscope somewhere in the input circuit made the signal a lot worse though.

I suggest you should test your circuit with a linear PSU and the switching PSU to see if it makes much difference t the digitized signal. BTW, if you are using an audio ADC, the input filter should already take care of high frequency components.

--
Reply to nico@nctdevpuntnl (punt=.)
Bedrijven en winkels vindt U op www.adresboekje.nl