On 5/24/2010 2:12 AM, John Larkin wrote:> On Sun, 23 May 2010 21:45:06 -0700 (PDT), dagmargoodboat@yahoo.com > wrote: > >> On May 23, 9:44 pm, Phil Hobbs >> <pcdhSpamMeSensel...@electrooptical.net> wrote: >>> Winfield Hill wrote: >>>> dagmargoodb...@yahoo.com wrote... >>>>> This shunt filter only needs 200mV headroom: >>> >>>>> FIG. 2 >>>>> R1 >>>>> +15V>--+------------------/\/\/\--------+--> Vout 14.8v >>>>> | 5 | >>>>> | | >>>>> | .-------+------+--------+ >>>>> | | | | | >>>>> | | | R6 | >>>>> | | | 1k | >>>>> | R3 R5 | |<' Q3 >>>>> | 2.7M 10K +------| 2n3906 >>>>> | | | | |\ >>>>> | | | |/ Q2 | >>>>> | | +----| 2n3904 | >>>>> | | | |>. | >>>>> | C1 | |<' | | >>>>> '---||---+----| Q1 '--------+ >>>>> 10uF |\ 2n3906 | >>>>> | R4 >>>>> | 4.7R >>>>> | | >>>>> ------+----------------+---- >>> >>>> Nice ASCII art. Is fig 2 from your feverish brain? >>> >>>> I see your idea, invert the ripple and subtract it out. >>>> Good. To do that the cancellation amplifier needs to >>>> be biased class A, so it can work over the entire ripple >>>> range. It should continuously draw current from the >>>> supply through R1, and superimpose the inverted ripple >>>> signal on top of that. R4 can be trimmed to optimize. >>>> The new R7 should be sized to handle the p-p ripple. >>> >>>> Then John's delicate C-multiplier filter can follower, >>>> with all the heavy lifting having been done. >>> >>>> +15V>--+-----------------/\/\/\--------+--> Vout 14.8v >>>> | 5 | >>>> | | >>>> | .------+------+--------+ >>>> | | | | | >>>> | | | R6 | >>>> | | | 1k | >>>> | R3 R5 | |<' Q3 >>>> | 2.7M 10K +------| 2n4403 >>>> | | | | |\ >>>> | | | |/ Q2 | >>>> | C1 | +----| 2n3904 | >>>> '---||---+ | |>. | >>>> 10uF | |<' | | >>>> +----| Q1 '--------+ >>>> | |\ 2n3906 | >>>> R7 | R4 >>>> TBD 27k | 4.7R >>>> | | | >>>> --+------+---------------+---- >>> >>> The Kanner Kap uses an audio power amp to do this, applying a small >>> amount of positive feedback to multiply the value of a BFC. Works OK, >>> but it isn't worth paying royalties on. >>> >>> Cap multipliers are magic--especially two-pole ones. It's 0.7 volts >>> well spent IMO. If Early is a worry, use a slower transistor--the >>> ripple rejection is basically C_CE/C_BFC, with some degradation due to >>> Early voltage and capacitor ESR. >> >> >> Yep, two-pole--that's the stuff I was fiddling with whilst you guys >> were posting...(ASCII takes time!) >> >> Fig. 3 >> ====== >> Q1 Q2 >> 2n3904 2n3904 >> +15V>--+--------. .----+---. .--+---> +13.3V >> | \ ^ | \ ^ | >> R1 ----- R3 ----- | >> 100R | 100R | | >> | R2 | | | --- C4 >> +--/\/\/----+ +------' --- 100uF >> | 100R | | | >> C1 --- C2 --- --- C3 | >> 100uF --- 100uF--- --- 100uF | >> | | | | >> === === === === >> GND GND GND GND >> >> >> Output ripple is LT-Spice undetectable. Zout ~= 2ohms. >> >> >> Fig. 4 >> ====== Q1 >> 2n3904 >> +15V>--+----------------+----. .----+-----> +13.3V >> | | \ ^ | >> R1 | ----- | >> 3.3k | | --- C3 >> | R2 |/ Q2 | --- 100uF >> +--/\/\/---+---| 2n3904 | | >> | 3.3K | |>. | | >> | | | | === >> C1 --- C2 --- +-------' GND >> 10uF --- 10uF --- | >> | | R3 >> === === 10k >> GND GND | >> === >> GND >> >> Buffer Q2 eliminates loading on filter R1C1-R2C2, greatly improving >> transient response& recovery. >> >> The output at Q2(e) is super-clean, but changes in load current >> modulate Re(Q1) and the drop across it, so output ripple is somewhat >> worse than the reference. Zout is the same as Fig. 3. >> >> Early effect isn't as noticeable as Re, so far. >> >> I did a version following Fig. 4 with a one-pole C-mult stage, biased >> off a divider from Q2(e), and a Sziklai PNP across the whole thing. >> That means the single-pole stage operates as a cascode and sees no >> d(Vce) to speak of. 1KHz ripple disappears, and Rout drops to about >> 0.25 ohms. >> >> This version is silly with parts. There needs to be an op amp in >> there somewhere to greatly simplify things, but it's time for me to >> turn in. Hopefully these musings will inspire John to continue the >> fight. > > > This is what I have so far: > > ftp://jjlarkin.lmi.net/P14_reg.gif > > The wall wart is prefiltered by a C-L-C filter that should buy me > about 40 dB at the switcher frequency. Then this thing should be good > for maybe 100 more. Then I have some more RCs before the photodiodes > and a couple of other critical things. > > This regulates to 13.4 to allow some headroom here and there. > > The LM8261 has about 10 nv/rthz noise, which is a whole nother story. > > John >That's limited by the CMR of the op amp, though, which blows the whole thing out of the water--no? That's the point of using the BJT in the first place. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
C-multiplier again
Started by ●May 22, 2010
Reply by ●May 24, 20102010-05-24
Reply by ●May 24, 20102010-05-24
On 5/24/2010 8:09 AM, Mike wrote:> Phil Hobbs<pcdhSpamMeSenseless@electrooptical.net> wrote: > >> Cap multipliers are magic--especially two-pole ones. It's 0.7 volts >> well spent IMO. If Early is a worry, use a slower transistor--the >> ripple rejection is basically C_CE/C_BFC, with some degradation due to >> Early voltage and capacitor ESR. >> >> Cheers >> >> Phil Hobbs > > Phil, you mentioned earlier being able to reach 140 db in one stage: > > http://groups.google.com/group/sci.electronics.design/msg/143f77519fed66e8 > > That's a ratio of 10 million to one. How do you do it? > > Thanks, > > Mike >Like I said, it's basically C_CE/C_BFC. You pick a transistor with reasonable characteristics at frequencies you care about, drive its base from a really really filtered version of V_CC--with a resistor in series to make sure it doesn't oscillate and doesn't blow up if the input or output gets shorted--and put a BFC at the output. If the transistor has 10 pF C_CB and the BFC is 100 uF, that's 140 dB, provided you look after other stuff such as the Early voltage and the ESR of the output cap. Generally if your application needs more than 100 dB of ripple rejection, you have to be pretty careful. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
Reply by ●May 24, 20102010-05-24
On 5/23/2010 12:29 PM, Vladimir Vassilevsky wrote:> > > John Larkin wrote: > >> >> >> I need a super-low noise power supply. I have a 15 volt switching >> wall-wart input and want as close to 15 volts, regulated, as I can >> get; 14 would be nice, 13.5 is OK. >> >> The LDOs that I can find are all pretty noisy and have mediocre PSRR. > > > http://www.abvolt.com/misc/psrr.jpg > > The topology like this is stable and provides for ~100dB of PSRR. > > You can also simulate huge LC filter with gyrators. > > Oh, and the trivial solution would be cascading regulators one after the > other. > > > Vladimir Vassilevsky > DSP and Mixed Signal Design Consultant > http://www.abvolt.comThat's assuming that your op amp is running off batteries? Because otherwise its PSR is going to dominate above ~10 kHz. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
Reply by ●May 24, 20102010-05-24
Phil Hobbs wrote:> On 5/23/2010 12:29 PM, Vladimir Vassilevsky wrote: > >> >> >> John Larkin wrote: >> >>> >>> >>> I need a super-low noise power supply. I have a 15 volt switching >>> wall-wart input and want as close to 15 volts, regulated, as I can >>> get; 14 would be nice, 13.5 is OK. >>> >>> The LDOs that I can find are all pretty noisy and have mediocre PSRR. >> >> >> >> http://www.abvolt.com/misc/psrr.jpg >> >> The topology like this is stable and provides for ~100dB of PSRR. >> >> You can also simulate huge LC filter with gyrators. >> >> Oh, and the trivial solution would be cascading regulators one after the >> other. >> >> > That's assuming that your op amp is running off batteries?Opamp is running from the output. Autostart is no problem.> Because > otherwise its PSR is going to dominate above ~10 kHz.Even if it is, there is RC on the power rail anyway. VLV
Reply by ●May 24, 20102010-05-24
On May 24, 2:39=A0pm, Phil Hobbs <pcdhSpamMeSensel...@electrooptical.net> wrote:> On 5/24/2010 8:09 AM, Mike wrote: > > > > > > > Phil Hobbs<pcdhSpamMeSensel...@electrooptical.net> =A0wrote: > > >> Cap multipliers are magic--especially two-pole ones. =A0It's 0.7 volts > >> well spent IMO. =A0If Early is a worry, use a slower transistor--the > >> ripple rejection is basically C_CE/C_BFC, with some degradation due to > >> Early voltage and capacitor ESR. > > >> Cheers > > >> Phil Hobbs > > > Phil, you mentioned earlier being able to reach 140 db in one stage: > > >http://groups.google.com/group/sci.electronics.design/msg/143f77519fe... > > > That's a ratio of 10 million to one. How do you do it? > > > Thanks, > > > Mike > > Like I said, it's basically C_CE/C_BFC. =A0You pick a transistor with > reasonable characteristics at frequencies you care about, drive its base > from a really really filtered version of V_CC--with a resistor in series > to make sure it doesn't oscillate and doesn't blow up if the input or > output gets shortedYeah a bit of resistance is needed. I make these with 2N3904's.. and at certain voltage/ current levels they would sing at 300MHz or so. (Actually I now use 2N4401's and 4403's, these don't seem to sing.. but I've kept the base R. ) I've never measured the rejection ratio, but you can make a ~1nV/rtHz power supply able to deliver several hundered mA's of current, driven from a switcher. George H. --and put a BFC at the output. =A0If the transistor has> 10 pF C_CB and the BFC is 100 uF, that's 140 dB, provided you look after > other stuff such as the Early voltage and the ESR of the output cap. > Generally if your application needs more than 100 dB of ripple > rejection, you have to be pretty careful. > > Cheers > > Phil Hobbs > > -- > Dr Philip C D Hobbs > Principal > ElectroOptical Innovations > 55 Orchard Rd > Briarcliff Manor NY 10510 > 845-480-2058 > hobbs at electrooptical dot nethttp://electrooptical.net- Hide quoted tex=t -> > - Show quoted text -
Reply by ●May 24, 20102010-05-24
On 24/05/2010 12:11 PM, dagmargoodboat@yahoo.com wrote:> On May 23, 6:56 pm, David Eather<eat...@tpg.com.au> wrote: >> On 24/05/2010 8:45 AM, John Larkin wrote: >> >>> On Mon, 24 May 2010 08:28:03 +1000, David Eather<eat...@tpg.com.au> >>> wrote: >> >>>> On 24/05/2010 8:07 AM, John Larkin wrote: >>>>> On Sun, 23 May 2010 13:26:26 -0700 (PDT), dagmargoodb...@yahoo.com >>>>> wrote: >> >>>>>> On May 23, 11:29 am, John Larkin >>>>>> <jjlar...@highNOTlandTHIStechnologyPART.com> wrote: >>>>>>> On 23 May 2010 04:28:01 -0700, Winfield Hill >> >>>>>>> <Winfield_mem...@newsguy.com> wrote: >>>>>>>> John Larkin wrote... >> >>>>>>>>> I need a super-low noise power supply. I have a 15 volt switching >>>>>>>>> wall-wart input and want as close to 15 volts, regulated, as I can >>>>>>>>> get; 14 would be nice, 13.5 is OK. >> >>>>>>>>> The LDOs that I can find are all pretty noisy and have mediocre PSRR. >> >>>>>>>>> So I thought about using a Phil Hobbs-ian c-multiplier transistor, an >>>>>>>>> R-C lowpass and an emitter follower, with a slow opamp loop wrapped >>>>>>>>> around it for DC regulation. It looks fine on paper, simple loop to >>>>>>>>> stabilize, but I figured I may as well Spice it and be sure. >> >>>>>>>>> What I'm seeing is mediocre PSRR. Stripping out the opamp and such, I >>>>>>>>> have... ftp://jjlarkin.lmi.net/C-multiplier.gif >>>>>>>>> which has psrr of about 70 dB at low frequencies, improving as the >>>>>>>>> output cap finally kicks in at around 5 KHz. The transistor equivalent >>>>>>>>> seems to look like the expected dynamic Re of about 2 ohms, with a C-E >>>>>>>>> resistor of around 6.6K. Reducing Vb (and Vout) doesn't help much. >> >>>>>>>> You're complaining about a 70dB improvement? There is a simple >>>>>>>> way to use your 0.7 volts, well maybe 0.8 volts, to get even >>>>>>>> more rejection: change your simple NPN follower into a Sziklai >>>>>>>> connection (AoE page 95). The base resistor across the added >>>>>>>> PNP creates a relatively-fixed collector current for your NPN, >>>>>>>> which means a fixed Vbe, for improved AC ripple rejection. >> >>>>>>> Since the problem is the Early effect, namely the effective C-E >>>>>>> resistance bleeding ripple through, it didn't seem to me like the >>>>>>> Sziklai thing would help. The PNP doesn't insulate the NPN from the >>>>>>> ripple. So I spiced it. If the LT Spice transistor models are to be >>>>>>> trusted, it's actually worse. The optimum value for the PNP's b-e >>>>>>> resistor is zero. >> >>>>>>> John >> >>>>>> Win's idea looks pretty decent to me, IIUIC: >> >>>>>> FIG. 1 (View in fixed font) >>>>>> ====== >> >>>>>> Q1 >>>>>> 2n3906 >>>>>> Vin>--+----. .-------+---+------+--> +13.3v >>>>>> | V / | | | >>>>>> R1 ------ | R2 --- C1 >>>>>> 470 | Q2 | 1k --- 15uF >>>>>> | | 2n3904 | | | >>>>>> '------+---. / === === >>>>>> \ ^ >>>>>> ----- >>>>>> | >>>>>> R3 >>>>>> 33 >>>>>> | >>>>>> +14v>---' >> >>>>>> LT Spice says 31uV of the 50mV 1KHz ripple gets through (32dBv), >>>>>> and the load step is 340uV. That's a lot stiffer than the original, >>>>>> which >>>>>> had a 4.5mV load step (d(i) = 2mA for both). >> >>>>>> The Sziklai version has the same ripple; I don't quite understand >>>>>> how Early explains that--Early should wreck the load step response >>>>>> too, shouldn't it? >> >>>>>> FIG 1's load step is only 60uV if you replace R1 with a 5mA current >>>>>> source, >>>>>> the 1KHz ripple stays the same. >> >>>>>> This shunt filter only needs 200mV headroom: >> >>>>>> FIG. 2 >>>>>> ====== >>>>>> R1 >>>>>> +15V>--+------------------/\/\/\--------+--> Vout = 14.8v >>>>>> | 5 | >>>>>> | | >>>>>> | | >>>>>> | | >>>>>> | .-------+------+--------+ >>>>>> | | | | | >>>>>> | | | R6 | >>>>>> | | | 1k | >>>>>> | R3 R5 | |<' Q3 >>>>>> | 2.7M 10K +------| 2n3906 >>>>>> | | | | |\ >>>>>> | | | |/ Q2 | >>>>>> | | +----| 2n3904 | >>>>>> | | | |>. | >>>>>> | C1 | |<' | | >>>>>> '---||---+----| Q1 '--------+ >>>>>> 10uF | |\ 2n3906 | >>>>>> R2 | R4 >>>>>> 5k | 4.7R >>>>>> | | | >>>>>> === === === >> >>>>>> LT Spice says 20dBV rejection @ 1KHz, zero @ d.c., natch. >> >>>>> Only 100 dB to go! But I don't understand Q1s biasing. >> >>>> Improved ripple response (but I think a little defective - it only works >>>> when Vin drops). >> >>>> When Vin drops Q1 turns on via base current drawn out through C1. Q1 >>>> robs base current from Q2 turning it off, which in turn turns off Q3 and >>>> reduces the current flow and hence voltage loss through R1. >> >>> So what's the quiescent current of Q1? Of Q3? >> >>> John >> >> R2 is missing - from the base of Q1 to GND - I suggest a value of 18k >> but it is a weird circuit I think a ripple reduction of no more than 46db > > If by 46dB you mean power, i.e. 20log(Vin/Vout) = 46, yes, that's > easily possible--that implies 0.5% gain accuracy. > > If you mean 46dBv, i.e. 10log(Vin/Vout) = 46, i.e. Vout / Vin = 25ppm, > no, that ain't happening, not unless you use op amps and some mighty > fine resistors. > > -- > Cheers, > James ArthurUm, I thought power was 10*log(what ever / what ever else) and voltage was 20*log(what ever / what ever else) e.g. a reduction in ripple of 46 db = a factor of 200, no? (I still like my idea of lm334 feeding LM431/tl431 - 4 small components - 6 if you use 2 x 334)
Reply by ●May 24, 20102010-05-24
Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:> On 5/24/2010 8:09 AM, Mike wrote:[...]> Like I said, it's basically C_CE/C_BFC. You pick a transistor with > reasonable characteristics at frequencies you care about, drive its > base from a really really filtered version of V_CC--with a resistor in > series to make sure it doesn't oscillate and doesn't blow up if the > input or output gets shorted--and put a BFC at the output. If the > transistor has 10 pF C_CB and the BFC is 100 uF, that's 140 dB, > provided you look after other stuff such as the Early voltage and the > ESR of the output cap. Generally if your application needs more than > 100 dB of ripple rejection, you have to be pretty careful. > > Cheers > > Phil HobbsIs that spiceable? I made a simple circuit with a voltage source driving the base and a cap on the emitter. I tried various transistors such as 2N2222 and 2N2369, and various ESR and ESL values for the cap. The capacitance had little effect on the attenuation floor, but mainly moved the low frequency corner. No reasonable combination of transistors or cap values got below -120dB. The base resistance had little effect. Here's the file if you'd like to show me how it should work: Version 4 SHEET 1 1140 1108 WIRE -304 -448 -384 -448 WIRE -160 -448 -224 -448 WIRE -128 -448 -160 -448 WIRE -128 -400 -128 -448 WIRE -192 -352 -256 -352 WIRE -384 -320 -384 -448 WIRE -256 -320 -256 -352 WIRE -128 -288 -128 -304 WIRE -32 -288 -128 -288 WIRE 16 -288 -32 -288 WIRE -128 -256 -128 -288 WIRE -32 -256 -32 -288 WIRE -384 -224 -384 -240 WIRE -256 -224 -256 -240 WIRE -32 -176 -32 -192 WIRE -128 -160 -128 -176 WIRE -32 -80 -32 -96 WIRE -32 16 -32 0 FLAG -256 -224 0 FLAG -384 -224 0 FLAG -128 -160 0 FLAG -160 -448 Vin FLAG -32 -288 Vout FLAG -32 16 0 SYMBOL npn -192 -400 R0 SYMATTR InstName Q1 SYMATTR Value 2N2369 SYMBOL voltage -384 -336 R0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V1 SYMATTR Value 15 SYMBOL voltage -256 -336 R0 WINDOW 123 0 0 Left 0 WINDOW 39 24 38 Left 0 SYMATTR SpiceLine Rser=1 SYMATTR InstName V2 SYMATTR Value 10 SYMBOL current -128 -256 R0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName I1 SYMATTR Value 20ma SYMBOL voltage -208 -448 R90 WINDOW 0 49 39 VRight 0 WINDOW 123 -48 40 VRight 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V3 SYMATTR Value2 AC 1 SYMATTR Value "" SYMBOL cap -48 -256 R0 SYMATTR InstName C1 SYMATTR Value 1000�f SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res -48 -96 R0 SYMATTR InstName R1 SYMATTR Value 100m SYMBOL ind -48 -192 R0 SYMATTR InstName L1 SYMATTR Value 10n TEXT -216 -536 Left 0 ;'BJT Ripple Filter TEXT -224 -504 Left 0 !.ac oct 100 1 1e7
Reply by ●May 24, 20102010-05-24
On Sun, 23 May 2010 18:01:24 -0700 (PDT), dagmargoodboat@yahoo.com wrote:>On May 23, 5:07�pm, John Larkin ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote: >> On Sun, 23 May 2010 13:26:26 -0700 (PDT), dagmargoodb...@yahoo.com >> wrote: >> >> >> >> >On May 23, 11:29�am, John Larkin >> ><jjlar...@highNOTlandTHIStechnologyPART.com> wrote: >> >> On 23 May 2010 04:28:01 -0700, Winfield Hill >> >> >> <Winfield_mem...@newsguy.com> wrote: >> >> >John Larkin wrote... >> >> >> >> I need a super-low noise power supply. I have a 15 volt switching >> >> >> wall-wart input and want as close to 15 volts, regulated, as I can >> >> >> get; 14 would be nice, 13.5 is OK. >> >> >> >> The LDOs that I can find are all pretty noisy and have mediocre PSRR. >> >> >> >> So I thought about using a Phil Hobbs-ian c-multiplier transistor, an >> >> >> R-C lowpass and an emitter follower, with a slow opamp loop wrapped >> >> >> around it for DC regulation. It looks fine on paper, simple loop to >> >> >> stabilize, but I figured I may as well Spice it and be sure. >> >> >> >> What I'm seeing is mediocre PSRR. Stripping out the opamp and such, I >> >> >> have... �ftp://jjlarkin.lmi.net/C-multiplier.gif >> >> >> which has psrr of about 70 dB at low frequencies, improving as the >> >> >> output cap finally kicks in at around 5 KHz. The transistor equivalent >> >> >> seems to look like the expected dynamic Re of about 2 ohms, with a C-E >> >> >> resistor of around 6.6K. Reducing Vb (and Vout) doesn't help much. >> >> >> > You're complaining about a 70dB improvement? �There is a simple >> >> > way to use your 0.7 volts, well maybe 0.8 volts, to get even >> >> > more rejection: change your simple NPN follower into a Sziklai >> >> > connection (AoE page 95). �The base resistor across the added >> >> > PNP creates a relatively-fixed collector current for your NPN, >> >> > which means a fixed Vbe, for improved AC ripple rejection. >> >> >> Since the problem is the Early effect, namely the effective C-E >> >> resistance bleeding ripple through, it didn't seem to me like the >> >> Sziklai thing would help. The PNP doesn't insulate the NPN from the >> >> ripple. So I spiced it. If the LT Spice transistor models are to be >> >> trusted, it's actually worse. The optimum value for the PNP's b-e >> >> resistor is zero. >> >> >> John >> >> >Win's idea looks pretty decent to me, IIUIC: >> >> >FIG. 1 � � (View in fixed font) >> >====== >> >> > � � � � � � Q1 >> > � � � � � �2n3906 >> >Vin >--+----. � �.-------+---+------+--> +13.3v >> > � � � | � � V �/ � � � �| � | � � �| >> > � � �R1 � �------ � � � | � R2 � �--- C1 >> > � � �470 � � | � � Q2 � | � 1k � �--- 15uF >> > � � � | � � �| � 2n3904 | � | � � �| >> > � � � '------+---. � � / � === � �=== >> > � � � � � � � � � \ � ^ >> > � � � � � � � � � ----- >> > � � � � � � � � � � | >> > � � � � � � � � � � R3 >> > � � � � � � � � � � 33 >> > � � � � � � � � � � | >> > � � � � � �+14v >---' >> >> >LT Spice says 31uV of the 50mV 1KHz ripple gets through (32dBv), >> >and the load step is 340uV. �That's a lot stiffer than the original, >> >which >> >had a 4.5mV load step (d(i) = 2mA for both). >> >> >The Sziklai version has the same ripple; I don't quite understand >> >how Early explains that--Early should wreck the load step response >> >too, shouldn't it? >> >> >FIG 1's load step is only 60uV if you replace R1 with a 5mA current >> >source, >> >the 1KHz ripple stays the same. >> >> >This shunt filter only needs 200mV headroom: >> >> >FIG. 2 >> >====== >> > � � � � � � � � � � � � � � � R1 >> >+15V >--+------------------/\/\/\--------+--> Vout = 14.8v >> > � � � �| � � � � � � � � � �5 � � � � � | >> > � � � �| � � � � � � � � � � � � � � � �| >> > � � � �| � � � � � � � � � � � � � � � �| >> > � � � �| � � � � � � � � � � � � � � � �| >> > � � � �| � � � �.-------+------+--------+ >> > � � � �| � � � �| � � � | � � �| � � � �| >> > � � � �| � � � �| � � � | � � R6 � � � �| >> > � � � �| � � � �| � � � | � � 1k � � � �| >> > � � � �| � � � R3 � � �R5 � � �| � � �|<' Q3 >> > � � � �| � � �2.7M � � 10K � � +------| � 2n3906 >> > � � � �| � � � �| � � � | � � �| � � �|\ >> > � � � �| � � � �| � � � | � �|/ Q2 � � �| >> > � � � �| � � � �| � � � +----| �2n3904 �| >> > � � � �| � � � �| � � � | � �|>. � � � �| >> > � � � �| � C1 � | � �|<' � � � | � � � �| >> > � � � �'---||---+----| �Q1 � � '--------+ >> > � � � � � 10uF � � � |\ 2n3906 � � � � �| >> > � � � � � � � � � � � �| � � � � � � � �R4 >> > � � � � � � � � � � � �| � � � � � � � 4.7R >> > � � � � � � � � � � � �| � � � � � � � �| >> > � � � � � � � � � � � === � � � � � � �=== >> >> >LT Spice says 20dBV rejection @ 1KHz, zero @ d.c., natch. >> >> Only 100 dB to go! > >That's 20dBV, e.g., (ripple in)/(ripple out) = 100. Pretty good for >standard parts with no trimming, I thought. Better cancellation needs >more accurate parts, e.g. op amp + precision resistors.I'm used to dBV meaning 20*log(volts), namely 100 volts is +40 dBV. It's an absolute level thing like dBm or dBuw. If a voltage regulator has 1 volt p-p ripple at its input and 10 mV at its output, its PSRR is 100:1 or 40 dB.> >Do you really need that ratio >= 10^12 ? Or do you mean dB(power), >i.e. (ripple in)/(ripple out) >= 10^6?I want any switcher noise (or any other noise!) to be below 1 nV in any 1 Hz bandwidth referred to my input. The box is the size of a small sandwich and is full of stuff. A microvolt of switcher noise on a critical power rail would probably make a visible birdie in somebody's data spectrum. Like having roaches in the salad. If my switcher has, say, 100 mV of ripple, 1e6:1 reduction, 120 dB, to 100 nV, might just do. 140 dB would be better.> >> But I don't understand Q1s biasing. > >Nope, you understand it fine! R2 got overlooked in the ASCII-art >conversion--5k from Q1(b) to GND. > >> But I can replace all that stuff after R1 with a big polymer aluminum >> cap and get about the same rolloff, probably better at high >> frequencies. > >If you've got the room that sounds fine. > > >> >I used transistors because they're fast--for canceling wideband noise. >> >> >You could use op-amps or TLV431 or such for accuracy and get make a >> >shunt regulator / noise canceler with much better 1KHz rejection, plus >> >load regulation. >> >> >Silliness, but fun. >> >> *I'm* not having much fun. I've got a circuit that needs nV/fA noise >> levels and it's beseiged from all directions. Johnson noise. Shot >> noise. Power supplies coupling in through diode junctions. Switchers >> inches away. And I'm supposed to Gerber it tomorrow. >> >> Wish you were here. > >It sounds like a lovely challenge, with a few shields tossed in the >mix. Not unlike RF receivers, methinks--those might inspire.Milled aluminum blocks this time. John
Reply by ●May 24, 20102010-05-24
On 23 May 2010 16:54:54 -0700, Winfield Hill <Winfield_member@newsguy.com> wrote:>dagmargoodboat@yahoo.com wrote... >> >> This shunt filter only needs 200mV headroom: >> >> FIG. 2 >> R1 >>+15V >--+------------------/\/\/\--------+--> Vout 14.8v >> | 5 | >> | | >> | .-------+------+--------+ >> | | | | | >> | | | R6 | >> | | | 1k | >> | R3 R5 | |<' Q3 >> | 2.7M 10K +------| 2n3906 >> | | | | |\ >> | | | |/ Q2 | >> | | +----| 2n3904 | >> | | | |>. | >> | C1 | |<' | | >> '---||---+----| Q1 '--------+ >> 10uF |\ 2n3906 | >> | R4 >> | 4.7R >> | | >> ------+----------------+---- > > Nice ASCII art. Is fig 2 from your feverish brain? > > I see your idea, invert the ripple and subtract it out. > Good. To do that the cancellation amplifier needs to > be biased class A, so it can work over the entire ripple > range. It should continuously draw current from the > supply through R1, and superimpose the inverted ripple > signal on top of that. R4 can be trimmed to optimize. > The new R7 should be sized to handle the p-p ripple. > > Then John's delicate C-multiplier filter can follower, > with all the heavy lifting having been done. > > +15V >--+-----------------/\/\/\--------+--> Vout 14.8v > | 5 | > | | > | .------+------+--------+ > | | | | | > | | | R6 | > | | | 1k | > | R3 R5 | |<' Q3 > | 2.7M 10K +------| 2n4403 > | | | | |\ > | | | |/ Q2 | > | C1 | +----| 2n3904 | > '---||---+ | |>. | > 10uF | |<' | | > +----| Q1 '--------+ > | |\ 2n3906 | > R7 | R4 > TBD 27k | 4.7R > | | | > --+------+---------------+----How about an opamp powered from Vout, with a resistor from the opamp output to ground? Let the opamp supply current fight the output ripple. That's thermally stable, simple, high gain, and tunable. (except I need regulation, too) John
Reply by ●May 24, 20102010-05-24
On Mon, 24 May 2010 10:45:27 -0700, Joerg <invalid@invalid.invalid> wrote:>John Larkin wrote: >> On Mon, 24 May 2010 08:19:40 -0700, Joerg <invalid@invalid.invalid> >> wrote: >> >>> John Larkin wrote: >>>> On Sun, 23 May 2010 21:45:06 -0700 (PDT), dagmargoodboat@yahoo.com >>>> wrote: >>>> >>>>> On May 23, 9:44 pm, Phil Hobbs >>>>> <pcdhSpamMeSensel...@electrooptical.net> wrote: >>>>>> Winfield Hill wrote: >>>>>>> dagmargoodb...@yahoo.com wrote... >>>>>>>> This shunt filter only needs 200mV headroom: >>>>>>>> FIG. 2 >>>>>>>> R1 >>>>>>>> +15V >--+------------------/\/\/\--------+--> Vout 14.8v >>>>>>>> | 5 | >>>>>>>> | | >>>>>>>> | .-------+------+--------+ >>>>>>>> | | | | | >>>>>>>> | | | R6 | >>>>>>>> | | | 1k | >>>>>>>> | R3 R5 | |<' Q3 >>>>>>>> | 2.7M 10K +------| 2n3906 >>>>>>>> | | | | |\ >>>>>>>> | | | |/ Q2 | >>>>>>>> | | +----| 2n3904 | >>>>>>>> | | | |>. | >>>>>>>> | C1 | |<' | | >>>>>>>> '---||---+----| Q1 '--------+ >>>>>>>> 10uF |\ 2n3906 | >>>>>>>> | R4 >>>>>>>> | 4.7R >>>>>>>> | | >>>>>>>> ------+----------------+---- >>>>>>> Nice ASCII art. Is fig 2 from your feverish brain? >>>>>>> I see your idea, invert the ripple and subtract it out. >>>>>>> Good. To do that the cancellation amplifier needs to >>>>>>> be biased class A, so it can work over the entire ripple >>>>>>> range. It should continuously draw current from the >>>>>>> supply through R1, and superimpose the inverted ripple >>>>>>> signal on top of that. R4 can be trimmed to optimize. >>>>>>> The new R7 should be sized to handle the p-p ripple. >>>>>>> Then John's delicate C-multiplier filter can follower, >>>>>>> with all the heavy lifting having been done. >>>>>>> +15V >--+-----------------/\/\/\--------+--> Vout 14.8v >>>>>>> | 5 | >>>>>>> | | >>>>>>> | .------+------+--------+ >>>>>>> | | | | | >>>>>>> | | | R6 | >>>>>>> | | | 1k | >>>>>>> | R3 R5 | |<' Q3 >>>>>>> | 2.7M 10K +------| 2n4403 >>>>>>> | | | | |\ >>>>>>> | | | |/ Q2 | >>>>>>> | C1 | +----| 2n3904 | >>>>>>> '---||---+ | |>. | >>>>>>> 10uF | |<' | | >>>>>>> +----| Q1 '--------+ >>>>>>> | |\ 2n3906 | >>>>>>> R7 | R4 >>>>>>> TBD 27k | 4.7R >>>>>>> | | | >>>>>>> --+------+---------------+---- >>>>>> The Kanner Kap uses an audio power amp to do this, applying a small >>>>>> amount of positive feedback to multiply the value of a BFC. Works OK, >>>>>> but it isn't worth paying royalties on. >>>>>> >>>>>> Cap multipliers are magic--especially two-pole ones. It's 0.7 volts >>>>>> well spent IMO. If Early is a worry, use a slower transistor--the >>>>>> ripple rejection is basically C_CE/C_BFC, with some degradation due to >>>>>> Early voltage and capacitor ESR. >>>>> Yep, two-pole--that's the stuff I was fiddling with whilst you guys >>>>> were posting...(ASCII takes time!) >>>>> >>>>> Fig. 3 >>>>> ====== >>>>> Q1 Q2 >>>>> 2n3904 2n3904 >>>>> +15V >--+--------. .----+---. .--+---> +13.3V >>>>> | \ ^ | \ ^ | >>>>> R1 ----- R3 ----- | >>>>> 100R | 100R | | >>>>> | R2 | | | --- C4 >>>>> +--/\/\/----+ +------' --- 100uF >>>>> | 100R | | | >>>>> C1 --- C2 --- --- C3 | >>>>> 100uF --- 100uF--- --- 100uF | >>>>> | | | | >>>>> === === === === >>>>> GND GND GND GND >>>>> >>>>> >>>>> Output ripple is LT-Spice undetectable. Zout ~= 2ohms. >>>>> >>>>> >>>>> Fig. 4 >>>>> ====== Q1 >>>>> 2n3904 >>>>> +15V >--+----------------+----. .----+-----> +13.3V >>>>> | | \ ^ | >>>>> R1 | ----- | >>>>> 3.3k | | --- C3 >>>>> | R2 |/ Q2 | --- 100uF >>>>> +--/\/\/---+---| 2n3904 | | >>>>> | 3.3K | |>. | | >>>>> | | | | === >>>>> C1 --- C2 --- +-------' GND >>>>> 10uF --- 10uF --- | >>>>> | | R3 >>>>> === === 10k >>>>> GND GND | >>>>> === >>>>> GND >>>>> >>>>> Buffer Q2 eliminates loading on filter R1C1-R2C2, greatly improving >>>>> transient response & recovery. >>>>> >>>>> The output at Q2(e) is super-clean, but changes in load current >>>>> modulate Re(Q1) and the drop across it, so output ripple is somewhat >>>>> worse than the reference. Zout is the same as Fig. 3. >>>>> >>>>> Early effect isn't as noticeable as Re, so far. >>>>> >>>>> I did a version following Fig. 4 with a one-pole C-mult stage, biased >>>>> off a divider from Q2(e), and a Sziklai PNP across the whole thing. >>>>> That means the single-pole stage operates as a cascode and sees no >>>>> d(Vce) to speak of. 1KHz ripple disappears, and Rout drops to about >>>>> 0.25 ohms. >>>>> >>>>> This version is silly with parts. There needs to be an op amp in >>>>> there somewhere to greatly simplify things, but it's time for me to >>>>> turn in. Hopefully these musings will inspire John to continue the >>>>> fight. >>>> >>>> This is what I have so far: >>>> >>>> ftp://jjlarkin.lmi.net/P14_reg.gif >>>> >>>> The wall wart is prefiltered by a C-L-C filter that should buy me >>>> about 40 dB at the switcher frequency. Then this thing should be good >>>> for maybe 100 more. Then I have some more RCs before the photodiodes >>>> and a couple of other critical things. >>>> >>>> This regulates to 13.4 to allow some headroom here and there. >>>> >>>> The LM8261 has about 10 nv/rthz noise, which is a whole nother story. >>>> >>> Try a low noise audio opamp and a depletion mode n-channel. >> >> Well, I can't spend the rest of my career working on this circuit! >> > >Yup, saw the thread too late, when you mentioned "Gerbers due on Monday". > > >> It would be interesting to power an opamp from the output; then >> there's no PSRR problem. A depletion fet would add enough offset so >> the thing (might) start up and work. But the fet would be acting >> almost resistive at low voltage, so the PSRR thing returns with a >> vengence. >> >> A bipolar PNP, with constant-current base drive (like Tim's circuit) >> is interesting, but still has Early feedthrough. A strategicly placed >> resistor could null out most of that. >> >> An RRO opamp, like in my circuit, incorporates the error amp and the >> PNP, all in one SOT23. The guys who designed it cared about PSRR and >> had a zillion transistors to work with. >> >> Any attempt to add regulation to a noise filter circuit must inject >> the noise of the regulator device into the loop; in my case, it's the >> 10 nv per of the LM8261. The loop comp parts can be tweaked to trade >> off noise injection versus transient load regulation. >> >> I *could* spend the rest of my career on this. Maybe I'll toss another >> 120 uF polymer cap after the first inductor, and move on with my life. >> > >If it works and is good enough, time to let it go. If not you could add >a simple one-transistor cap multiplier into the positive supply of the >opamp. Three parts, all tiny. > > >> If I keep changing this board, The Brat will take up patricide. >> > >Aha, I thought that it's the brat pushing you on this project :-) > >I think my dad would have had a hissy fit if I'd ever done that. He >occasionally said "You electrical guys, you don't have a lot of >patience, do ya?"A picosecond is a terrible thing to waste. John
