battery chargers, again
in [Design]
|
From: John Larkin on 1 Oct 2009 16:44 On Thu, 01 Oct 2009 13:13:16 -0700, Joerg <invalid(a)invalid.invalid> wrote: >John Larkin wrote: >> On Wed, 30 Sep 2009 14:23:38 -0700, Joerg <invalid(a)invalid.invalid> >> wrote: >> >>> John Larkin wrote: >>>> On Wed, 30 Sep 2009 13:01:31 -0700, Joerg <invalid(a)invalid.invalid> >>>> wrote: >>>> >>>>> Jan Panteltje wrote: >>>>>> On a sunny day (Wed, 30 Sep 2009 11:17:34 -0700) it happened Joerg >>>>>> <invalid(a)invalid.invalid> wrote in <7ihlm2F3179urU2(a)mid.individual.net>: >>>>>> >>>>>>>> Are you referring to the loop filter? >>>>>>>> There is no loop filter. >>>>>>>> >>>>>>> No, to the output filter. You can't possibly let the ripple just saunter >>>>>>> out the banana jacks of a bench supply and pollute whatever it find >>>>>>> downstream. >>>>>> All depends, 10mV is fine with me:-) >>>>>> >>>>>> You really got to come up with some numbers, else it all makes no sense, >>>>>> >>>>> Ok, this is the number I typically need: Not being able to show the >>>>> ripple on a scope when set to 2mV/div. Good enough? >>>>> >>>>> Seriously, 10mV piping out would be disastrous when doing things like >>>>> ultrasound experiments. >>>> I'm working on a smallish board that has four switchers in one corner >>>> and an ADC that's looking for a 1 nV spectral line in the other, range >>>> of interest straddling all the switcher frequencies. I'm planning on >>>> spread-spectrum wobulating all the switchers, just in case. >>>> >>> Spread spectrum just smears the noise so it looks lower (and is to some >>> extent but not a lot). Like throwing excess junk under the sofa and >>> pretending it ain't there no more. >> >> It will take a big spectral spike and make it into wide and low hump. >> If the psd gets below my 1 nv/rtHz noise floor, it's as good as gone. >> >>> Since you wrote spectral line, how about rotating the switchers through >>> three different frequencies and have the ADC always look in the quiet bands? >> >> No, we've got to digitize in one burst and see what's there. The next >> shot may be different chemicals, and we need all the data. >> > >Nasty situation ... > >If you can confine them all to a corner and the loads don't change too >rapidly you could LC-filter the dickens out of every rail. Yes. And maybe slit the ground plane in artistic patterns to confine most of the circulating currents to the "power" corner of the board. This is it so far: ftp://jjlarkin.lmi.net/Rev30.jpg The switchers are upper-right, and the differential-input RF stuff is the two SMBs south-east of center. The layout is tedious, because we're optimizing the BGA FPGA routing as we go along... can't just draw the schematic and go forward to the board. This *is* rev 30. John
From: Joerg on 1 Oct 2009 16:48 John Larkin wrote: > On Thu, 01 Oct 2009 13:10:44 -0700, Joerg <invalid(a)invalid.invalid> > wrote: > >> John Larkin wrote: >>> On Wed, 30 Sep 2009 14:40:06 -0700, Joerg <invalid(a)invalid.invalid> >>> wrote: >>> >>>> John Larkin wrote: >>>>> On Wed, 30 Sep 2009 16:04:00 -0500, Vladimir Vassilevsky >>>>> <nospam(a)nowhere.com> wrote: >>>>> >>>>>> John Larkin wrote: >>>>>> >>>>>> >>>>>>> I'm working on a smallish board that has four switchers in one corner >>>>>>> and an ADC that's looking for a 1 nV spectral line in the other, >>>>>> The 1nV over the background of how many nV/root(Hz) ? >>>>> About 1. That's our target noise floor. The existing system is >>>>> ballpark 100 nv/rtHz with huge birdies all over the place. >>>>> >>>>> Never let scientists design electronics. >>>>> >>>>>>> range >>>>>>> of interest straddling all the switcher frequencies. >>>>>> Synchronize the switchers away from the particular frequency of interest? >>>>> The signals can be all over the place. >>>>> >>>>>>> I'm planning on >>>>>>> spread-spectrum wobulating all the switchers, just in case. >>>>>> So the dirt will be in band for sure? >>>>> Yup. Operating range is audio to many MHz, all at once. We digitize at >>>>> 64 Ms/s and FFT and see what's there. >>>>> >>>> How many MHz is "many"? And how much power does the biggest of the >>>> switchers have to deliver? >>> 20-ish. And just a couple of watts. I need 5, 3.3, 2.5, 1.8, 1.2, and >>> -5. All from +12. >>> >> Theoretically you could run it at 27.12MHz but that's nasty. If you >> can't stomach the dissipation of linears then you may be stuck with some >> spectral dodging scheme like the one I described. We do that in >> ultrasound sometimes but there I could always get a few clock control >> lines from some a DSP or FPGA. If I bought the digital and SW guys some >> beers that night ... > > I've got a couple of switcher eval boards around. Maybe I should fire > some up and experiment with FMing the switcher freqs and see what that > does to the spectrum and to the output voltage. > But first check the limit values. Or at least try them out with a potmeter for the frequency set resistor. Easiest would be to have PWM chips with a sync input. > Or better yet have somebody do it for me. > Can't you get a student who is eager to learn and make a buck from the university? Maybe throw in free parking :-) -- Regards, Joerg http://www.analogconsultants.com/ "gmail" domain blocked because of excessive spam. Use another domain or send PM.
From: John Larkin on 1 Oct 2009 17:00 On Thu, 01 Oct 2009 13:48:49 -0700, Joerg <invalid(a)invalid.invalid> wrote: >John Larkin wrote: >> On Thu, 01 Oct 2009 13:10:44 -0700, Joerg <invalid(a)invalid.invalid> >> wrote: >> >>> John Larkin wrote: >>>> On Wed, 30 Sep 2009 14:40:06 -0700, Joerg <invalid(a)invalid.invalid> >>>> wrote: >>>> >>>>> John Larkin wrote: >>>>>> On Wed, 30 Sep 2009 16:04:00 -0500, Vladimir Vassilevsky >>>>>> <nospam(a)nowhere.com> wrote: >>>>>> >>>>>>> John Larkin wrote: >>>>>>> >>>>>>> >>>>>>>> I'm working on a smallish board that has four switchers in one corner >>>>>>>> and an ADC that's looking for a 1 nV spectral line in the other, >>>>>>> The 1nV over the background of how many nV/root(Hz) ? >>>>>> About 1. That's our target noise floor. The existing system is >>>>>> ballpark 100 nv/rtHz with huge birdies all over the place. >>>>>> >>>>>> Never let scientists design electronics. >>>>>> >>>>>>>> range >>>>>>>> of interest straddling all the switcher frequencies. >>>>>>> Synchronize the switchers away from the particular frequency of interest? >>>>>> The signals can be all over the place. >>>>>> >>>>>>>> I'm planning on >>>>>>>> spread-spectrum wobulating all the switchers, just in case. >>>>>>> So the dirt will be in band for sure? >>>>>> Yup. Operating range is audio to many MHz, all at once. We digitize at >>>>>> 64 Ms/s and FFT and see what's there. >>>>>> >>>>> How many MHz is "many"? And how much power does the biggest of the >>>>> switchers have to deliver? >>>> 20-ish. And just a couple of watts. I need 5, 3.3, 2.5, 1.8, 1.2, and >>>> -5. All from +12. >>>> >>> Theoretically you could run it at 27.12MHz but that's nasty. If you >>> can't stomach the dissipation of linears then you may be stuck with some >>> spectral dodging scheme like the one I described. We do that in >>> ultrasound sometimes but there I could always get a few clock control >>> lines from some a DSP or FPGA. If I bought the digital and SW guys some >>> beers that night ... >> >> I've got a couple of switcher eval boards around. Maybe I should fire >> some up and experiment with FMing the switcher freqs and see what that >> does to the spectrum and to the output voltage. >> > >But first check the limit values. Or at least try them out with a >potmeter for the frequency set resistor. Easiest would be to have PWM >chips with a sync input. One of my guys wants to have the FPGA synthesize a bunch of FMd clocks and dump that into the reg sync inputs. I was planning to make an analog triangle and squirt a little of it into the pins where the freq set resistors connect. > > >> Or better yet have somebody do it for me. >> > >Can't you get a student who is eager to learn and make a buck from the >university? Maybe throw in free parking :-) We usually have an intern or two for the summer, but we seem to have run out. John
From: Joerg on 1 Oct 2009 17:06 John Larkin wrote: > On Thu, 01 Oct 2009 13:13:16 -0700, Joerg <invalid(a)invalid.invalid> > wrote: > >> John Larkin wrote: >>> On Wed, 30 Sep 2009 14:23:38 -0700, Joerg <invalid(a)invalid.invalid> >>> wrote: >>> >>>> John Larkin wrote: >>>>> On Wed, 30 Sep 2009 13:01:31 -0700, Joerg <invalid(a)invalid.invalid> >>>>> wrote: >>>>> >>>>>> Jan Panteltje wrote: >>>>>>> On a sunny day (Wed, 30 Sep 2009 11:17:34 -0700) it happened Joerg >>>>>>> <invalid(a)invalid.invalid> wrote in <7ihlm2F3179urU2(a)mid.individual.net>: >>>>>>> >>>>>>>>> Are you referring to the loop filter? >>>>>>>>> There is no loop filter. >>>>>>>>> >>>>>>>> No, to the output filter. You can't possibly let the ripple just saunter >>>>>>>> out the banana jacks of a bench supply and pollute whatever it find >>>>>>>> downstream. >>>>>>> All depends, 10mV is fine with me:-) >>>>>>> >>>>>>> You really got to come up with some numbers, else it all makes no sense, >>>>>>> >>>>>> Ok, this is the number I typically need: Not being able to show the >>>>>> ripple on a scope when set to 2mV/div. Good enough? >>>>>> >>>>>> Seriously, 10mV piping out would be disastrous when doing things like >>>>>> ultrasound experiments. >>>>> I'm working on a smallish board that has four switchers in one corner >>>>> and an ADC that's looking for a 1 nV spectral line in the other, range >>>>> of interest straddling all the switcher frequencies. I'm planning on >>>>> spread-spectrum wobulating all the switchers, just in case. >>>>> >>>> Spread spectrum just smears the noise so it looks lower (and is to some >>>> extent but not a lot). Like throwing excess junk under the sofa and >>>> pretending it ain't there no more. >>> It will take a big spectral spike and make it into wide and low hump. >>> If the psd gets below my 1 nv/rtHz noise floor, it's as good as gone. >>> >>>> Since you wrote spectral line, how about rotating the switchers through >>>> three different frequencies and have the ADC always look in the quiet bands? >>> No, we've got to digitize in one burst and see what's there. The next >>> shot may be different chemicals, and we need all the data. >>> >> Nasty situation ... >> >> If you can confine them all to a corner and the loads don't change too >> rapidly you could LC-filter the dickens out of every rail. > > Yes. And maybe slit the ground plane in artistic patterns to confine > most of the circulating currents to the "power" corner of the board. > Slits can cause grief once it is connected to the outside world. Residual noise flowing on the power input side and the input cable shields (which will meet somewhere inside the city limits) eventually makes a loop. So far I've never used slits, only removed them. > This is it so far: > > ftp://jjlarkin.lmi.net/Rev30.jpg > > The switchers are upper-right, and the differential-input RF stuff is > the two SMBs south-east of center. > Hard to see, the jpeg isn't very hi-res. I think the inductors are crucial here. Even the ones that are called shielded by the marketeers have epoxy gaps of 20mils or more. If you can get toroids on there I'd really do that. Toroids don't talk. > The layout is tedious, because we're optimizing the BGA FPGA routing > as we go along... can't just draw the schematic and go forward to the > board. > Ouch. > This *is* rev 30. > Double-Ouch. I am firing up rev 1 of a really unorthodox cicuit this afternoon. Wish me luck, and if you hear a muffled boom north-east from you guys ... -- Regards, Joerg http://www.analogconsultants.com/ "gmail" domain blocked because of excessive spam. Use another domain or send PM.
From: Rich Grise on 1 Oct 2009 17:03
On Wed, 30 Sep 2009 19:17:08 -0700, John Larkin wrote: > On Wed, 30 Sep 2009 14:40:06 -0700, Joerg <invalid(a)invalid.invalid> >>John Larkin wrote: >>> On Wed, 30 Sep 2009 16:04:00 -0500, Vladimir Vassilevsky >>> <nospam(a)nowhere.com> wrote: >>> >>>> >>>> John Larkin wrote: >>>> >>>> >>>>> I'm working on a smallish board that has four switchers in one corner >>>>> and an ADC that's looking for a 1 nV spectral line in the other, >>>> The 1nV over the background of how many nV/root(Hz) ? >>> >>> About 1. That's our target noise floor. The existing system is >>> ballpark 100 nv/rtHz with huge birdies all over the place. >>> >>> Never let scientists design electronics. >>> >>>>> range >>>>> of interest straddling all the switcher frequencies. >>>> Synchronize the switchers away from the particular frequency of interest? >>> >>> The signals can be all over the place. >>> >>>>> I'm planning on >>>>> spread-spectrum wobulating all the switchers, just in case. >>>> So the dirt will be in band for sure? >>> >>> Yup. Operating range is audio to many MHz, all at once. We digitize at >>> 64 Ms/s and FFT and see what's there. >> >>How many MHz is "many"? And how much power does the biggest of the >>switchers have to deliver? > > 20-ish. And just a couple of watts. I need 5, 3.3, 2.5, 1.8, 1.2, and > -5. All from +12. > Reference everything to a virtual ground at +6 from the supply negative, use LDOs for the 5's, and ordinary regulators for the rest? If there's no COTS LDO at +5 or -5, it might be fun to build one with a chip and low Rdson MOSFET. ;-) Cheers! Rich |