On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote:> I have seen designs where there is a cutout in the groundplane under the > power inductor in e.g. a buck converter. Inductor is nominally a > "shielded" one. > > I assume it is to prevent some kind of "shorted turn" effect? What do > you think? > > Might a continuous plane be better? It could help to shield any field > leakage and reduce emissions and circuit noise. > > I am asking generally, but say 1A, 500kHz.I inherited a SMPS a few years ago that was blanking out nearly the entire F.M. band. Part of the problem was the main inductor's gap, which faced down into a ground plane and impressed a very impressive signal across same. This inductor was "shielded," a drum core cemented inside a 5-sided ferrite box. I substituted a gap-up inductor, and (that part of) the problem was solved. Cheers, James Arthur
Groundplane under SMPS power inductor
Started by ●November 4, 2015
Reply by ●November 5, 20152015-11-05
Reply by ●November 6, 20152015-11-06
John Larkin <jjlarkin@highlandtechnology.com> writes:> On Thu, 05 Nov 2015 07:59:36 +0000, John Devereux > <john@devereux.me.uk> wrote: > >>John Larkin <jjlarkin@highlandtechnology.com> writes: >> >>> On Wed, 4 Nov 2015 14:48:32 -0800 (PST), makolber@yahoo.com wrote: >>> >>>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote: >>>>> John Larkin <jjlarkin@highlandtechnology.com> writes: >>>>> >>>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux >>>>> > <john@devereux.me.uk> wrote: >>>>> > >>>>> >> >>>>> >>I have seen designs where there is a cutout in the groundplane under the >>>>> >>power inductor in e.g. a buck converter. Inductor is nominally a >>>>> >>"shielded" one. >>>>> >> >>>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do >>>>> >>you think? >>>>> >> >>>>> >>>>if it is a magnetically shielded inductor, then it should not matter >>>> >>>>if it is an open magnetic circuit inductor, then it could matter. >>>> >>>>Mark >>> >>> People take great liberties with the word "shielded", but most such >>> surface-mount inductors leak more field out the top than the bottom. >> >>Yep, hence my wording "nominally sheilded". It is easy to probe the >>leakage field around the system, with a small wire loop on the end of a >>coax and a scope. (Someone suggested a SMT inductor instead of the loop, >>might have been you). > > A lot of the surfmount "shielded" inductors are drum (H-shaped) > ferrite cores, wound and then dropped into a ferrite tube. That makes > a ring-shaped air gap at both ends and allows lots of field leakage. > That's symmetric, except that the board end sometimes has a plastic > base which, with the solder joint, spaces the part a bit above the > board. I tried one, sitting on a PCB with a layer 2 ground plane. It > lost a couple per cent of its free-space inductance on the board in > its normal mounting position, maybe 5% if pushed onto the board upside > down.Hey you beat me to it, thanks. I found ~3% with a SRU1048 22uH. If I drill a 10mm hole under it the change is about 1.5%. So it makes a bit of a difference but does not eliminate the effect. This was all at 500kHz. The indicated Q went from 24 to 22 as it was brought up to the plane.> I think there are surface-mount power inductors with better shielding. > Pot cores and toroids are more expensive than shielded drum cores. > Unshielded drums leak huge fields. > > We don't usually worry about it. On the other hand, it's really > convenient to buy little potted dc/dc converters, cheap and all done. > They sit above the board and usually have toroidal inductors inside. > That can rip a lot of parts off your BOM. I sometimes use them even > when I don't need the isolation.-- John Devereux
Reply by ●November 6, 20152015-11-06
dagmargoodboat@yahoo.com writes:> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote: >> I have seen designs where there is a cutout in the groundplane under the >> power inductor in e.g. a buck converter. Inductor is nominally a >> "shielded" one. >> >> I assume it is to prevent some kind of "shorted turn" effect? What do >> you think? >> >> Might a continuous plane be better? It could help to shield any field >> leakage and reduce emissions and circuit noise. >> >> I am asking generally, but say 1A, 500kHz. > > I inherited a SMPS a few years ago that was blanking out nearly > the entire F.M. band. > > Part of the problem was the main inductor's gap, which faced down > into a ground plane and impressed a very impressive signal across > same. > > This inductor was "shielded," a drum core cemented inside a 5-sided > ferrite box. > > I substituted a gap-up inductor, and (that part of) the problem was > solved.Interesting, thanks. -- John Devereux
Reply by ●November 6, 20152015-11-06
On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmargoodboat@yahoo.com) said:> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote: > > I have seen designs where there is a cutout in the groundplane under the > > power inductor in e.g. a buck converter. Inductor is nominally a > > "shielded" one. > > > > I assume it is to prevent some kind of "shorted turn" effect? What do > > you think? > > > > Might a continuous plane be better? It could help to shield any field > > leakage and reduce emissions and circuit noise. > > > > I am asking generally, but say 1A, 500kHz. > > I inherited a SMPS a few years ago that was blanking out nearly > the entire F.M. band. > > Part of the problem was the main inductor's gap, which faced down > into a ground plane and impressed a very impressive signal across > same. > > This inductor was "shielded," a drum core cemented inside a 5-sided > ferrite box. > > I substituted a gap-up inductor, and (that part of) the problem was > solved. > > Cheers, > James Arthur >Interesting. I have a very similar problem at the moment, a board with two synchronised 90W boost converters, using the type of inductors you describe. They're made by W�rth, who are pretty good IME, but I've never tried this inductor construction before. The board is 4 layers with a solid ground plane under the inductor, and no traces or components on the opposite side to the inductor. Originally I thought any currents induced in the ground plane would have a small enough loop area, and the plane itself would be a low enough impedance, that this wouldn't be a problem. Clearly this analysis doesn't hold up at this sort of power level... I will try and find a gap-up inductor with the same specs which fits the PCB pads. I suspect the problem I will then have, is that there is an array of right-angle connectors on an adjacent board, whose PCB pins pass within a few mm of the top of one of the inductors. The boards are in a metal housing and the connectors and cabling are shielded, but still not ideal.
Reply by ●November 6, 20152015-11-06
RBlack <news@rblack01.plus.com> writes:> On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmargoodboat@yahoo.com) > said: >> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote: >> > I have seen designs where there is a cutout in the groundplane under the >> > power inductor in e.g. a buck converter. Inductor is nominally a >> > "shielded" one. >> > >> > I assume it is to prevent some kind of "shorted turn" effect? What do >> > you think? >> > >> > Might a continuous plane be better? It could help to shield any field >> > leakage and reduce emissions and circuit noise. >> > >> > I am asking generally, but say 1A, 500kHz. >> >> I inherited a SMPS a few years ago that was blanking out nearly >> the entire F.M. band. >> >> Part of the problem was the main inductor's gap, which faced down >> into a ground plane and impressed a very impressive signal across >> same. >> >> This inductor was "shielded," a drum core cemented inside a 5-sided >> ferrite box. >> >> I substituted a gap-up inductor, and (that part of) the problem was >> solved. >> >> Cheers, >> James Arthur >> > > Interesting. I have a very similar problem at the moment, a board with > two synchronised 90W boost converters, using the type of inductors you > describe. They're made by Würth, who are pretty good IME, but I've > never tried this inductor construction before. > > The board is 4 layers with a solid ground plane under the inductor, and > no traces or components on the opposite side to the inductor. > Originally I thought any currents induced in the ground plane would have > a small enough loop area, and the plane itself would be a low enough > impedance, that this wouldn't be a problem. Clearly this analysis > doesn't hold up at this sort of power level...Looking at it pessimistically, you could think of it as a voltage transformer. 12V on the "primary" and xxV around the "secondary" which is a circular region of ground plane under the inductor. If the inductor has 6 turns say that would be 2V induced with perfect coupling. If it only couples 5% that is still 100mV which is not nothing when it appears at RF on a cable or in a sensitive circuit. The rest of the ground plane tries to short this out but there could be a lot of current available, the same ratio that steps down the voltage steps up the available current. (Sorry for the above travesty of transformer theory and I know it's all a bit handwavey but I can see how it could be an issue!)> I will try and find a gap-up inductor with the same specs which fits the > PCB pads. I suspect the problem I will then have, is that there is an > array of right-angle connectors on an adjacent board, whose PCB pins > pass within a few mm of the top of one of the inductors. The boards are > in a metal housing and the connectors and cabling are shielded, but > still not ideal.Unfortunately I never see the leakage mentioned anywhere as an inductor specification. Except that it might be "low". The dirt-cheap multilayer inductors are quite good here, I guess the turns are fully embedded in ferrite. But they only seem to go to low power levels and are more lossy than other types. -- John Devereux
Reply by ●November 6, 20152015-11-06
On Friday, November 6, 2015 at 8:34:36 AM UTC-5, John Devereux wrote:> RBlack <news@rblack01.plus.com> writes: > > > On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmargoodboat@yahoo.com) > > said: > >> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote: > >> > I have seen designs where there is a cutout in the groundplane under the > >> > power inductor in e.g. a buck converter. Inductor is nominally a > >> > "shielded" one. > >> > > >> > I assume it is to prevent some kind of "shorted turn" effect? What do > >> > you think? > >> > > >> > Might a continuous plane be better? It could help to shield any field > >> > leakage and reduce emissions and circuit noise. > >> > > >> > I am asking generally, but say 1A, 500kHz. > >> > >> I inherited a SMPS a few years ago that was blanking out nearly > >> the entire F.M. band. > >> > >> Part of the problem was the main inductor's gap, which faced down > >> into a ground plane and impressed a very impressive signal across > >> same. > >> > >> This inductor was "shielded," a drum core cemented inside a 5-sided > >> ferrite box. > >> > >> I substituted a gap-up inductor, and (that part of) the problem was > >> solved. > >> > >> Cheers, > >> James Arthur > >> > > > > Interesting. I have a very similar problem at the moment, a board with > > two synchronised 90W boost converters, using the type of inductors you > > describe. They're made by W�rth, who are pretty good IME, but I've > > never tried this inductor construction before. > > > > The board is 4 layers with a solid ground plane under the inductor, and > > no traces or components on the opposite side to the inductor. > > Originally I thought any currents induced in the ground plane would have > > a small enough loop area, and the plane itself would be a low enough > > impedance, that this wouldn't be a problem. Clearly this analysis > > doesn't hold up at this sort of power level... > > Looking at it pessimistically, you could think of it as a voltage > transformer. 12V on the "primary" and xxV around the "secondary" which > is a circular region of ground plane under the inductor. > > If the inductor has 6 turns say that would be 2V induced with perfect > coupling. If it only couples 5% that is still 100mV which is not nothing > when it appears at RF on a cable or in a sensitive circuit. > > The rest of the ground plane tries to short this out but there could be > a lot of current available, the same ratio that steps down the voltage > steps up the available current. > > (Sorry for the above travesty of transformer theory and I know it's all > a bit handwavey but I can see how it could be an issue!) > > > I will try and find a gap-up inductor with the same specs which fits the > > PCB pads. I suspect the problem I will then have, is that there is an > > array of right-angle connectors on an adjacent board, whose PCB pins > > pass within a few mm of the top of one of the inductors. The boards are > > in a metal housing and the connectors and cabling are shielded, but > > still not ideal. > > Unfortunately I never see the leakage mentioned anywhere as an inductor > specification. Except that it might be "low". > > The dirt-cheap multilayer inductors are quite good here, I guess the > turns are fully embedded in ferrite. But they only seem to go to low > power levels and are more lossy than other types. > > > -- > > John DevereuxI pulled up the old artwork for a look and thought it worth sharing some more details of my experience. It was a buck SMPS, 12V input, 5V output @2.4A, ~460KHz. As received, Layer 1 was needed to connect the inductor, Layer 2 had non-critical signals, Layer 3 was solid ground plane. Layer 1 .---------------------------// .------/-----------. | |-----. | .-+++-. +5V output | | | | | C1 | | '-----'-----|-|- - -|-------// | .--|-|- - -|----------> | | .| '-xxx-'. . . . . | L1 |. | . . . . ..-------> | | .|. . . . . | | .-----. |. | . GND . .| | | | | .|. . . . . | | '-----' |. | . . . . .| '-------\----/--|--'. .TP1. . | | | /. . . (). . .| TP1 is a solid through-hole metal post, connected to the GND trace and to the ground plane on layer 3. You can see that the original ground run ran under L1, parallel to L1's winding progression, creating a transformer-coupling in the topside GND above TP1 (a solid metal grounded test post at the lower bound) and prior to the bulk filter cap C1's negative terminal. So, no doubt there was significant coupling into that trace despite a solid plane two layers deeper. That rendered C1 (and additional bypasses) ineffective at r.f. The top-right portion of the GND trace supplied a ribbon connector, providing a path for wicked radiation and conduction of the induced GND signal. The revised layout routed layer 1's GND well around L1 rather than under, and nailed it to layer 3's ground plane at both ends of the run, and in- between. The replacement inductor(*) was also a drum cemented into a ferrite box like the first unit, however, the new unit's down-facing gap was tighter, and filled with what appeared to be a ferrite-loaded cement. (The original unit had a larger downward-facing gap, air-filled, IIRC.) * Bourns SRR1240-150M The improved gap configuration wasn't obvious on the datasheet--I had to get a sample in hand. So, I can't say for sure that an inductor directly over a solid ground plane would be problematic from this experience. I can say there was significant transformer coupling due to fringe flux from the downward-facing gap. Cheers, James Arthur
Reply by ●November 6, 20152015-11-06
On Fri, 06 Nov 2015 07:59:52 +0000, John Devereux <john@devereux.me.uk> wrote:>John Larkin <jjlarkin@highlandtechnology.com> writes: > >> On Thu, 05 Nov 2015 07:59:36 +0000, John Devereux >> <john@devereux.me.uk> wrote: >> >>>John Larkin <jjlarkin@highlandtechnology.com> writes: >>> >>>> On Wed, 4 Nov 2015 14:48:32 -0800 (PST), makolber@yahoo.com wrote: >>>> >>>>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote: >>>>>> John Larkin <jjlarkin@highlandtechnology.com> writes: >>>>>> >>>>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux >>>>>> > <john@devereux.me.uk> wrote: >>>>>> > >>>>>> >> >>>>>> >>I have seen designs where there is a cutout in the groundplane under the >>>>>> >>power inductor in e.g. a buck converter. Inductor is nominally a >>>>>> >>"shielded" one. >>>>>> >> >>>>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do >>>>>> >>you think? >>>>>> >> >>>>>> >>>>>if it is a magnetically shielded inductor, then it should not matter >>>>> >>>>>if it is an open magnetic circuit inductor, then it could matter. >>>>> >>>>>Mark >>>> >>>> People take great liberties with the word "shielded", but most such >>>> surface-mount inductors leak more field out the top than the bottom. >>> >>>Yep, hence my wording "nominally sheilded". It is easy to probe the >>>leakage field around the system, with a small wire loop on the end of a >>>coax and a scope. (Someone suggested a SMT inductor instead of the loop, >>>might have been you). >> >> A lot of the surfmount "shielded" inductors are drum (H-shaped) >> ferrite cores, wound and then dropped into a ferrite tube. That makes >> a ring-shaped air gap at both ends and allows lots of field leakage. >> That's symmetric, except that the board end sometimes has a plastic >> base which, with the solder joint, spaces the part a bit above the >> board. I tried one, sitting on a PCB with a layer 2 ground plane. It >> lost a couple per cent of its free-space inductance on the board in >> its normal mounting position, maybe 5% if pushed onto the board upside >> down. > >Hey you beat me to it, thanks. I found ~3% with a SRU1048 22uH. If I >drill a 10mm hole under it the change is about 1.5%. So it makes a bit >of a difference but does not eliminate the effect. > >This was all at 500kHz. > >The indicated Q went from 24 to 22 as it was brought up to the plane. > >Cool. That might matter if you need a super-efficient switcher. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
Reply by ●November 6, 20152015-11-06
On Friday, November 6, 2015 at 3:53:27 PM UTC-5, John Larkin wrote:> On Fri, 06 Nov 2015 07:59:52 +0000, John Devereux > <john@devereux.me.uk> wrote: > > >John Larkin <jjlarkin@highlandtechnology.com> writes: > > > >> On Thu, 05 Nov 2015 07:59:36 +0000, John Devereux > >> <john@devereux.me.uk> wrote: > >> > >>>John Larkin <jjlarkin@highlandtechnology.com> writes: > >>> > >>>> On Wed, 4 Nov 2015 14:48:32 -0800 (PST), makolber@yahoo.com wrote: > >>>> > >>>>>On Wednesday, November 4, 2015 at 12:55:50 PM UTC-5, John Devereux wrote: > >>>>>> John Larkin <jjlarkin@highlandtechnology.com> writes: > >>>>>> > >>>>>> > On Wed, 04 Nov 2015 16:33:30 +0000, John Devereux > >>>>>> > <john@devereux.me.uk> wrote: > >>>>>> > > >>>>>> >> > >>>>>> >>I have seen designs where there is a cutout in the groundplane under the > >>>>>> >>power inductor in e.g. a buck converter. Inductor is nominally a > >>>>>> >>"shielded" one. > >>>>>> >> > >>>>>> >>I assume it is to prevent some kind of "shorted turn" effect? What do > >>>>>> >>you think? > >>>>>> >> > >>>>>> > >>>>>if it is a magnetically shielded inductor, then it should not matter > >>>>> > >>>>>if it is an open magnetic circuit inductor, then it could matter. > >>>>> > >>>>>Mark > >>>> > >>>> People take great liberties with the word "shielded", but most such > >>>> surface-mount inductors leak more field out the top than the bottom. > >>> > >>>Yep, hence my wording "nominally sheilded". It is easy to probe the > >>>leakage field around the system, with a small wire loop on the end of a > >>>coax and a scope. (Someone suggested a SMT inductor instead of the loop, > >>>might have been you). > >> > >> A lot of the surfmount "shielded" inductors are drum (H-shaped) > >> ferrite cores, wound and then dropped into a ferrite tube. That makes > >> a ring-shaped air gap at both ends and allows lots of field leakage. > >> That's symmetric, except that the board end sometimes has a plastic > >> base which, with the solder joint, spaces the part a bit above the > >> board. I tried one, sitting on a PCB with a layer 2 ground plane. It > >> lost a couple per cent of its free-space inductance on the board in > >> its normal mounting position, maybe 5% if pushed onto the board upside > >> down. > > > >Hey you beat me to it, thanks. I found ~3% with a SRU1048 22uH. If I > >drill a 10mm hole under it the change is about 1.5%. So it makes a bit > >of a difference but does not eliminate the effect. > > > >This was all at 500kHz. > > > >The indicated Q went from 24 to 22 as it was brought up to the plane. > > > > > > Cool. That might matter if you need a super-efficient switcher.Since the Q fell by roughly the same factor as the inductance, I take that as indicating minimal added dissipation. Q = xL/R, etc. That's encouraging, actually. Cheers, James Arthur
Reply by ●November 7, 20152015-11-07
dagmargoodboat@yahoo.com writes:> On Friday, November 6, 2015 at 8:34:36 AM UTC-5, John Devereux wrote: >> RBlack <news@rblack01.plus.com> writes: >> >> > On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmargoodboat@yahoo.com) >> > said: >> >> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote: >> >> > I have seen designs where there is a cutout in the groundplane under the >> >> > power inductor in e.g. a buck converter. Inductor is nominally a >> >> > "shielded" one. >> >> > >> >> > I assume it is to prevent some kind of "shorted turn" effect? What do >> >> > you think? >> >> > >> >> > Might a continuous plane be better? It could help to shield any field >> >> > leakage and reduce emissions and circuit noise. >> >> > >> >> > I am asking generally, but say 1A, 500kHz.[...]> I pulled up the old artwork for a look and thought it worth sharing > some more details of my experience. > > It was a buck SMPS, 12V input, 5V output @2.4A, ~460KHz. > > As received, > Layer 1 was needed to connect the inductor, > Layer 2 had non-critical signals, > Layer 3 was solid ground plane. > > > Layer 1 > .---------------------------// > .------/-----------. > | |-----. | .-+++-. +5V output > | | | | | C1 | > | '-----'-----|-|- - -|-------// > | .--|-|- - -|----------> > | | .| '-xxx-'. . . . . > | L1 |. | . . . . ..-------> > | | .|. . . . . | > | .-----. |. | . GND . .| > | | | | .|. . . . . | > | '-----' |. | . . . . .| > '-------\----/--|--'. .TP1. . | > | | /. . . (). . .| > > > TP1 is a solid through-hole metal post, connected to the GND trace and > to the ground plane on layer 3. > > You can see that the original ground run ran under L1, parallel > to L1's winding progression, creating a transformer-coupling in the topside > GND above TP1 (a solid metal grounded test post at the lower bound) and > prior to the bulk filter cap C1's negative terminal. > > So, no doubt there was significant coupling into that trace despite a > solid plane two layers deeper. > > That rendered C1 (and additional bypasses) ineffective at r.f. The > top-right portion of the GND trace supplied a ribbon connector, providing > a path for wicked radiation and conduction of the induced GND signal.I guess the very low impedance is what makes the cap ineffective.> The revised layout routed layer 1's GND well around L1 rather than under, > and nailed it to layer 3's ground plane at both ends of the run, and in- > between. > > The replacement inductor(*) was also a drum cemented into a ferrite box > like the first unit, however, the new unit's down-facing gap was tighter, > and filled with what appeared to be a ferrite-loaded cement. (The original > unit had a larger downward-facing gap, air-filled, IIRC.)I suppose I need to buy a few types.> * Bourns SRR1240-150M > > The improved gap configuration wasn't obvious on the datasheet--I had to > get a sample in hand. > > So, I can't say for sure that an inductor directly over a solid ground plane > would be problematic from this experience.Thank you very much for the detailed write-up James.> I can say there was significant transformer coupling due to fringe flux > from the downward-facing gap.I think if I build the whole SMPS section on a local ground "island" over the overall groundplane this will work OK. Any magnetic leakage through the island is only going to be at lower frequencies. As you pointed out my "Q" measurement indicates that efficiency should not be affected significantly. Thanks, -- John Devereux
Reply by ●November 7, 20152015-11-07
On Sat, 07 Nov 2015 13:10:43 +0000, John Devereux <john@devereux.me.uk> wrote:>dagmargoodboat@yahoo.com writes: > >> On Friday, November 6, 2015 at 8:34:36 AM UTC-5, John Devereux wrote: >>> RBlack <news@rblack01.plus.com> writes: >>> >>> > On Thu, 5 Nov 2015 19:38:24 -0800 (PST), (dagmargoodboat@yahoo.com) >>> > said: >>> >> On Wednesday, November 4, 2015 at 11:33:34 AM UTC-5, John Devereux wrote: >>> >> > I have seen designs where there is a cutout in the groundplane under the >>> >> > power inductor in e.g. a buck converter. Inductor is nominally a >>> >> > "shielded" one. >>> >> > >>> >> > I assume it is to prevent some kind of "shorted turn" effect? What do >>> >> > you think? >>> >> > >>> >> > Might a continuous plane be better? It could help to shield any field >>> >> > leakage and reduce emissions and circuit noise. >>> >> > >>> >> > I am asking generally, but say 1A, 500kHz. > >[...] > >> I pulled up the old artwork for a look and thought it worth sharing >> some more details of my experience. >> >> It was a buck SMPS, 12V input, 5V output @2.4A, ~460KHz. >> >> As received, >> Layer 1 was needed to connect the inductor, >> Layer 2 had non-critical signals, >> Layer 3 was solid ground plane. >> >> >> Layer 1 >> .---------------------------// >> .------/-----------. >> | |-----. | .-+++-. +5V output >> | | | | | C1 | >> | '-----'-----|-|- - -|-------// >> | .--|-|- - -|----------> >> | | .| '-xxx-'. . . . . >> | L1 |. | . . . . ..-------> >> | | .|. . . . . | >> | .-----. |. | . GND . .| >> | | | | .|. . . . . | >> | '-----' |. | . . . . .| >> '-------\----/--|--'. .TP1. . | >> | | /. . . (). . .| >> >> >> TP1 is a solid through-hole metal post, connected to the GND trace and >> to the ground plane on layer 3. >> >> You can see that the original ground run ran under L1, parallel >> to L1's winding progression, creating a transformer-coupling in the topside >> GND above TP1 (a solid metal grounded test post at the lower bound) and >> prior to the bulk filter cap C1's negative terminal. >> >> So, no doubt there was significant coupling into that trace despite a >> solid plane two layers deeper. >> >> That rendered C1 (and additional bypasses) ineffective at r.f. The >> top-right portion of the GND trace supplied a ribbon connector, providing >> a path for wicked radiation and conduction of the induced GND signal. > >I guess the very low impedance is what makes the cap ineffective. > >> The revised layout routed layer 1's GND well around L1 rather than under, >> and nailed it to layer 3's ground plane at both ends of the run, and in- >> between. >> >> The replacement inductor(*) was also a drum cemented into a ferrite box >> like the first unit, however, the new unit's down-facing gap was tighter, >> and filled with what appeared to be a ferrite-loaded cement. (The original >> unit had a larger downward-facing gap, air-filled, IIRC.) > >I suppose I need to buy a few types. > >> * Bourns SRR1240-150M >> >> The improved gap configuration wasn't obvious on the datasheet--I had to >> get a sample in hand. >> >> So, I can't say for sure that an inductor directly over a solid ground plane >> would be problematic from this experience. > >Thank you very much for the detailed write-up James. > >> I can say there was significant transformer coupling due to fringe flux >> from the downward-facing gap. > >I think if I build the whole SMPS section on a local ground "island" >over the overall groundplane this will work OK. Any magnetic leakage >through the island is only going to be at lower frequencies. As you >pointed out my "Q" measurement indicates that efficiency should not be >affected significantly. >The ground island my help here but it's not the reason to do a ground island. I almost always use the powdered metal type of inductors (similar to Vishay IHLP series) but also ground islands to keep the loop currents local to the SMPS. Only the input and output capacitors and low-side switch/inductor get connected to this local plane, though. The feedback, compensation, and any "analog" ground pins connect to the over-all ground. The ground connections of the input and output capacitors should be close to each other on this island, too. Most SMPS recommended layouts follow this strategy.
