I looked at this some more, and added a cap directly across the transistor from collector to emitter.
The attenuation curve is like a bathtub. The capacitor in the emitter affects the low frequency. The C-E cap and ESL change the high frequency side. These are rational effects, and tend to say SPICE is telling the truth. But this says the ratio of the C-E cap and the emitter cap have no bearing on the attenuation, since they affect the opposite ends of the spectrum.
One thing that had a dramatic affect on the attenuation is the emitter current. Going one order of magnitude up reduced the attenuation a great deal, and dropping the current to 1 mA finally gave close to -140dB. This says the low noise supply should be used only for the sensitive parts of a circuit, and as soon as the design permits, run the rest of the circuit on noisier supply voltages.
It's not clear what the emitter current is changing, but there has to be a measurable parameter for the transistor that indicates how well it will perform in this application. But if this analysis holds water, it means I have learned something I didn't know before.
CMRR and PSRR matter, certainly. But the transistor doesn't regulate, and doesn't seem to ripple reject super well at low frequencies, if the LT Spice 2N3904 model is to be believed. At high frequencies, ripple rejection of the transistor thing improves as the load cap and Re start making a lowpass. My 15 ohms is better than 2 ohms of Re, and drops less DC too. So the circuits really aren't all that different, but I'll have better low-frequency PSRR... clear down to DC.
But for photodiodes, you don't care too much about the low frequency stuff, since it's in series with the photodiode capacitance.
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
The Early effect causes problems at higher currents, because it adds a conductance that's more or less proportional to the emitter current. That makes a difference, especially if the base resistor is too big, so that the base current change causes an output voltage change.
Lots of Spice models don't handle it well IME. Slower, bigger, higher voltage transistors have higher Early voltages. Big fat ceramics (10 uF) are good for the output cap.
I've used MPSA14s to cut 50 mV of 10 kHz ripple down to a nanovolt or two in real circuits. I couldn't do a lot of what I do without cap multipliers.
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
Try JFETs. No Vbe offset = arbitrarily low dropout. In fact it's negative a lot of the time: even better. Easy to cascade/cascode. Use P or N channel, however you want.
Source terminal is squishy (low Gm). Solution: servo with op-amp, or if you want to be quirky, add a shunt regulator so the current draw is constant. You're only drawing like 15mA, right?
Which reminds me of another novel, useless circuit I invented, the shunt current source.
On my website,
formatting link
"This revolutionary (and impressively useless) circuit is the completion of an analogy. Consider: voltage sources are available in two flavors, shunt (e.g., TL431) and series-pass (e.g., LM7805). But current sources are only available in one style, series-pass. These simple circuits complete the analogy, providing a shunt current source. In both cases, a resistor provides a current greater than or equal to the desired output current over the rated range; a current sense resistor, voltage reference and voltage amplifier (VBE and a BJT in the left example; a TL431 and differential pair in the right example) adjust a shunt current to keep the output current constant."
Man, this whole thing smacks of audiophoolery. Sometimes, they'll put a CCS into a shunt regulator (even a rather noisy one like a glow discharge tube) just because they feel like it. Difference being, you can actually measure nanovolts.
Tim
--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms
"John Larkin" wrote in message
news:da4hv5pumif114u33dau1pqoh93sc11m8b@4ax.com...
>
>
>
> 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.
>
> I'm using the LT Spice 2N3904 model, which I take to be a sort of
> generic small-signal NPN. The 33r base resistor value doesn't seem to
> matter.
>
> There must be a better way, ideally one that doesn't throw away 0.7
> perfectly good volts.
>
> John
>
Here's a circuit that lurking in this discussion about voltage regulators inspired me to come up with over the weekend, speaking of useless circuits and audiophools. It's a voltage regulator that appears to have decent line regulation without any negative feedback. Cuz negative feedback is bad, right? It's also expensive!
formatting link
Can you see how it works? Or how I think it is intended to work? :) It doesn't really need a split supply, that's just for messing around. The PSRR is only as good as the output opamp, unfortunately I haven't found a way to get rid of it yet!
Press Ctrl-C to copy the selected block to the clipboard.
Open any plain ascii text editor such as NotePad or EditPad.
Paste the selected text into the word processor with Ctrl-V.
Use "Save As" to save the file to a suitable folder. Use the file extension ".asc" to run it in LTspice.
Repeat if there is a plot file, using the ".plt" extension.
If you have installed LTspice and associated the ASC file extension, all you need to do is select the file in MS Explorer and it will load the file into LTspice.
If you have not associated the file extensions, then load LTspice and use the Open File command to locate and run the file.
Thanks, that new info makes a big difference. When you said to use a BFC capacitor on the emitter, I thought you meant a big capacitor, so I looked at electrolytics above 1000uF.
These have a series resonance somewhere around 100KHz, and it made the C-E capacitance only affect the high side of the bathtub curve. There was no capacitance ratio effect that you mentioned earlier.
However, a 10uF ceramic has a much higher series resonance frequency, and this changes the entire picture. I'm including a fairly long LTspice file that describes it. Here's a description of the output signals:
Vout1 : This uses the Fairchild spice model for the MPSA14. It shows that spice cannot model the conductance properly in a bjt series pass element, as you stated.
Vout2 : Using the numbers from your previous posts, the attenuation is 20*log(2e-9/50e-3) = -147.95 dB. Using a 10uF emitter capacitor, the MPSA14 can be modeled as a simple 60 megohm resistor in parallel with a 400 femtofarad capacitor. This produces a shelf at -146.95 dB from about 10KHz to just over 1 MHz. The attenuation is the ratio of the two capacitors, just as you said earlier.
Vout3 : Now that we have a more realistic model for the MPSA14, it is useful to try it with a large electrolytic. Low ESR caps are now very common. I used a Nic Components Corp. aluminum electrolytic NRE-HL332M16V12.5x35F, 3300uf, 0.020 Ohm @ 100KHz, listed in page 3 of
formatting link
This shows a dramatic improvement of about 50 dB for frequencies below 10KHz. (Ignore the -190dB at 6KHz:)
Vout4: The 10uF ceramic works well for the higher frequencies, but it doesn't do much for frequencies below 10KHz. Here's the effect of using a large electrolytic in parallel with the 10uF ceramic. There is a huge reduction below 10KHz, and a significant reduction all the way up to 300KHz.
So for the price of a plain electrolytic you can probably find in your junkbox, you can get a major improvement in attenuation. This might also apply to the reference filter.
Of course, as Joerg always recommends, add diodes to protect against hard shorts.
Regards,
Mike
Here's the file. There might be prolems due to line wrap:
Version 4 SHEET 1 1140 1108 WIRE -256 -416 -272 -416 WIRE 16 -416 -176 -416 WIRE 112 -416 16 -416 WIRE 144 -416 112 -416 WIRE 384 -416 144 -416 WIRE 480 -416 384 -416 WIRE 576 -416 480 -416 WIRE 672 -416 576 -416 WIRE 768 -416 672 -416 WIRE 864 -416 768 -416 WIRE 16 -384 16 -416 WIRE 384 -352 384 -416 WIRE 576 -352 576 -416 WIRE 768 -352 768 -416 WIRE -48 -336 -112 -336 WIRE 144 -336 144 -416 WIRE 480 -336 480 -416 WIRE 672 -336 672 -416 WIRE 864 -336 864 -416 WIRE -272 -304 -272 -416 WIRE -112 -304 -112 -336 WIRE 32 -288 16 -288 WIRE 80 -288 32 -288 WIRE 32 -224 32 -288 WIRE 48 -224 32 -224 WIRE 144 -224 144 -240 WIRE 144 -224 128 -224 WIRE 240 -224 144 -224 WIRE 256 -224 240 -224 WIRE 384 -224 384 -272 WIRE 432 -224 384 -224 WIRE 480 -224 480 -272 WIRE 480 -224 432 -224 WIRE 576 -224 576 -272 WIRE 624 -224 576 -224 WIRE 672 -224 672 -272 WIRE 672 -224 624 -224 WIRE 768 -224 768 -272 WIRE 816 -224 768 -224 WIRE 864 -224 864 -272 WIRE 864 -224 816 -224 WIRE -272 -208 -272 -224 WIRE -112 -208 -112 -224 WIRE 256 -208 256 -224 WIRE 384 -208 384 -224 WIRE 576 -208 576 -224 WIRE 768 -208 768 -224 WIRE 864 -208 864 -224 WIRE 144 -160 144 -224 WIRE 256 -128 256 -144 WIRE 384 -128 384 -144 WIRE 576 -128 576 -144 WIRE 768 -128 768 -144 WIRE 864 -128 864 -144 WIRE 144 -64 144 -80 WIRE 256 -32 256 -48 WIRE 384 -32 384 -48 WIRE 576 -32 576 -48 WIRE 768 -32 768 -48 WIRE 864 -32 864 -48 WIRE 256 64 256 48 WIRE 384 64 384 48 WIRE 576 64 576 48 WIRE 768 64 768 48 WIRE 864 64 864 48 FLAG -112 -208 0 FLAG -272 -208 0 FLAG 144 -64 0 FLAG 112 -416 Vin FLAG 240 -224 Vout1 FLAG 256 64 0 FLAG 432 -224 Vout2 FLAG 384 64 0 FLAG 624 -224 Vout3 FLAG 576 64 0 FLAG 816 -224 Vout4 FLAG 768 64 0 FLAG 864 64 0 SYMBOL npn 80 -336 R0 SYMATTR InstName Q1 SYMATTR Value q1model SYMBOL voltage -272 -320 R0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V1 SYMATTR Value 15 SYMBOL voltage -112 -320 R0 WINDOW 123 0 0 Left 0 WINDOW 39 36 57 Left 0 SYMATTR SpiceLine Rser=10 SYMATTR InstName V2 SYMATTR Value 10 SYMBOL current 144 -160 R0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 WINDOW 0 31 14 Left 0 WINDOW 3 27 60 Left 0 SYMATTR InstName I1 SYMATTR Value 1ma SYMBOL voltage -160 -416 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 npn -48 -384 R0 SYMATTR InstName Q2 SYMATTR Value q1model SYMBOL res 368 -368 R0 SYMATTR InstName R4 SYMATTR Value 80e6 SYMBOL cap 240 -208 R0 SYMATTR InstName C1 SYMATTR Value 10µf SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 240 -48 R0 SYMATTR InstName R2 SYMATTR Value 3m SYMBOL ind 240 -144 R0 SYMATTR InstName L1 SYMATTR Value 2.5nh SYMBOL cap 368 -208 R0 SYMATTR InstName C2 SYMATTR Value 10µf SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 368 -48 R0 SYMATTR InstName R3 SYMATTR Value 3m SYMBOL ind 368 -144 R0 SYMATTR InstName L2 SYMATTR Value 2.5nh SYMBOL res 144 -240 R90 WINDOW 0 0 56 VBottom 0 WINDOW 3 32 56 VTop 0 SYMATTR InstName R1 SYMATTR Value 200e6 SYMBOL cap 464 -336 R0 SYMATTR InstName C3 SYMATTR Value 400f SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 560 -368 R0 SYMATTR InstName R5 SYMATTR Value 80e6 SYMBOL cap 560 -208 R0 SYMATTR InstName C4 SYMATTR Value 3300µf SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 560 -48 R0 SYMATTR InstName R6 SYMATTR Value 20m SYMBOL ind 560 -144 R0 SYMATTR InstName L3 SYMATTR Value 10nh SYMBOL cap 656 -336 R0 SYMATTR InstName C5 SYMATTR Value 400f SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 752 -368 R0 SYMATTR InstName R7 SYMATTR Value 80e6 SYMBOL cap 752 -208 R0 SYMATTR InstName C6 SYMATTR Value 3300µf SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 752 -48 R0 SYMATTR InstName R8 SYMATTR Value 20m SYMBOL ind 752 -144 R0 SYMATTR InstName L4 SYMATTR Value 10nh SYMBOL cap 848 -336 R0 SYMATTR InstName C7 SYMATTR Value 400f SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL cap 848 -208 R0 SYMATTR InstName C8 SYMATTR Value 10µf SYMATTR SpiceLine Rser=1u Lser=1n SYMBOL res 848 -48 R0 SYMATTR InstName R9 SYMATTR Value 3m SYMBOL ind 848 -144 R0 SYMATTR InstName L5 SYMATTR Value 2.5nh TEXT 8 -512 Left 0 ;'MPSA14 Darlington vs passive models TEXT 48 -472 Left 0 !.ac oct 100 1 4e6 TEXT -264 144 Left 0 !.MODEL Q1model NPN(IS=1.34E-14 BF=340 NF=1 VAF=
The lf rejection was about 3000:1. The hf rolloff corner corresponds to a 2 ohm Re and the 15u load cap. The load is only 14 mA, so Re is relatively high.
My opamp circuit starts with the opamp's roughly 100 dB PSRR and then hits a 15 ohm + 120 uF lowpass, about 88 Hz corner frequency. That RC rolloff gets pretty far down before the opamp's PSRR starts to get bad. And I don't lose the 0.7 volts (or twice that for a darlington) which happens to matter in my immediate case.
Neither circuit is perfect. There's probably something really good lurking out in circuit space.
Yech, could've saved on a lot of mess by adding an OTA or two.
Considering the multiplier, I would hazard a guess at some overly complicated truncated-Taylor-series correction.
It's worth noting that, if Vo is the output, then all the other nodes supplied by it inherently have feedback. In particular, Vref will vary a small amount; Vbias will vary proportionally; I_R5 will vary proportionally; and there's early effect on all transistors, and PSRR in the multiplier.
With R15 and R19 so excessively large compared to the impedances on the other sides (R20 is shunted by D1, and R29 by R18), the OTA offsets will be huge, and proportional to I_R6 (hence, OTA). The first LT1014 sections seem to be doing I-to-V conversion, relative to Vbias (a "safe" value, given the OTA outputs will work somewhere between Vref and Vo, assuming Vo/2 > ~Vref).
The LHS OTA output is subtracted from Vbias, to which it is relative, so the multiplier gets something centered around 0. This is superfluous, as it has differential inputs to begin with. The RHS OTA gets the same treatment, and this zero-referenced value goes to the mult's add input. Output goes to inverting amp to Vo.
Now, LHS OTA has Vref on one side (assuming the zener is actually biased in breakdown), and
Just because V_GS < 0 doesn't mean V_DS < 0. JFETs fall off a cliff at low V_DS, typically well before they get to 0.7V. BF862s fall off starting above 1 V. They also have far worse Early effect (or the FET equivalent, whatever it's called).
formatting link
Fairly horrible tempco, though, IIUC.
You've never designed a high performance photoreceiver, have you? Noise on the bias supply comes in exactly like TIA voltage noise--they're on opposite ends of the same capacitor (the photodiode capacitance). Achieving nanovolt noise on the supply is very frequently the difference between success and failure.
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
I just downloaded a model of the LM13700 - I'll test it out.
Shhh! :)
You're right, I missed that. Those resistors are too large. Performance improves when I make them a value that's more in line with the impedance on the other side.
I had the differential amplifier there because I was messing around with scaling factors going into the multiplier. It turned out to not be necessary and the bias voltage can just be applied to pins 2 and 4.
Pretty close - the circuit is in essence an analog computer. My idea was that the transfer function of a differential pair is approximately Iout = (Io/2VT)*tanh(vid), where Io is the LTP current. By selecting the V-I converter resistor and Io appropriately you can cause the multiplying terms to drop out (at only one temperature, though) and you get tanh(vid). This is then squared by the multiplier to get tanh^2. The reference voltage is divided down 1000-1 and applied to a similar differential pair.
The idea is that the amplified output of the second differential pair will be approximately the _derivative_ of the tanh function with respect to the reference voltage, or sec^2h(vid), or tanh^2(vid) - 1, which is also inverted to get -tanh^2(vid) + 1. Then it's added in the multiplier to cancel the tanh^2 terms and get 1, or a voltage that's stable with respect to variations in the reference. In practice I'm not getting 100mv out as I expected but the multiplier does seem to put out a stable voltage, with 1V P-P at 1000 Hz bouncing on the supply a FFT on the output shows the first harmonic down -115 dB, which is about the PSRR of the output op amp.
Whether this is better than just two TL431s set to different voltages attached to the inputs of the same op amp as a differential amplifier, I do not know. :)
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.