Re: gEDA-user: Current source rail blocking

[Karel Kulhavy <clock@xxxxxxxxxxxxx>]

On Mon, Oct 30, 2006 at 06:40:13AM -0500, Dan McMahill wrote:
> Karel Kulhavy wrote:
> >gn Sun, Oct 29, 2006 at 07:09:46PM -0500, Dan McMahill wrote:
> >
> >>Karel Kulhavy wrote:
> >>
> >>>Hello
> >>>
> >>>In my broadband high sensitivity amplifier I have a rail where bases of 
> >>>all
> >>>current source transistors are hooked up. Each current source is a 
> >>>transistor
> >>>with an emitter resistor which compensates variance in aplification so 
> >>>that
> >>>even unmatched transistors produce matched currents.
> >>>
> >>>Do you have any recommendations how to block the current sources from 
> >>>picking
> >>>up some garbage from the air and causing oscillation? Should I block the 
> >>>rail
> >>>with a single big capacitor against the ground, or use individual 
> >>>capacitor for
> >>>each transistor placed between emitter and base?
> >>
> >>What sort of transistors are these?  Are you seeing oscillations in 
> >
> >
> >2N3904
> >
> >
> >>practice or is this just a concern prior to seeing any real hardware? 
> >
> >
> >In practice.
> >
> >
> >>I'd avoid using transistors which are faster than need be.  For example, 
> >>sticking a 20 GHz device in there may not be a good idea.
> >>
> >>If you can tolerate some additional noise, you can stick some resistance 
> >>in series with the base if the transistors themselves are oscillating. 
> >
> >
> >Why does this help? Slows down the transistor by forming an RC lowpass
> >with the inherent E-B capacitance of the transistor?
> 
> Calculate the impedance seen looking into the base of a common emitter 
> stage that has capacitance from emitter to ground.  If you have 
> capacitance to ground at the emitter, then there is an impedance 
> proportional to 1/j*w there.  Now refer that to the base side.  At 
> higher frequencies, the transistor current gain is falling off with 
> frequency so now you get a 1/(j*w)^2 term in the impedance looking into 
> the base.  When you actually work out the math, you'll get some 
> expression that will have a capacitive term along with a *negative* 
> resistance term.  Combine that with some inductance (from the board 
> layout, package leads, etc) and you have an oscillator.  The series R 
> serves to produce a net positive resistance.
> 
> >Is it possible to hook up a test circuit with the parasitics in gnucap and
> >run it and see it if it oscillates? Or is it just going to say "internal
> >node open" or fail to converge?
> 
> It should be very easy to simulate the negative input resistance.  This 
> is also easy to calculate by hand.  To simulate oscillations you'll need 
> to be sure you have a correct model for whatever parasitic inductance 
> you may have in the base circuit.
> 
> >
> >BTW how does the reality work that it always converges? Is it possible to
> >build a real electronic circuit that causes the universe to fail to 
> >converge
> >and be terminated with an error message?
> >
> >If not, why isn't the same calculation that is used to run the universe 
> >just
> >put into gnucap so it would converge every time?
> >
> >Does gnucap convergence failure indicate the circuit would oscillate?
> >Does an oscillator circuit in gnucap always cause convergence failure on
> >transient mode simulation?
> 
> Many of my convergence problems (not speaking of gnucap but simulators 
> in general) come from bad inputs.  Reality usually doesn't include an 
> ideal 1F capacitor or an inductor with no loss.  Reality usually doesn't 
> include a voltage coefficient on some element which causes a stable 
> operating point with internal voltages outside the supply.  Certainly a 
> simulator can mess up too, but it's pretty easy to feed them a model 
> which does not reflect reality.
> 
> >Well, I want to build a current source from a single 2N3904 (or a double 
> >one if
> >it's a current mirror where the driving half can be recycled for multiple
> >mirrors) that gives 2.5mA constant current and behaves like a current 
> >mirror
> >to as high frequencies as possible (i. e. not something that is nicely
> >stable but from 10kHz up it starts behaving like a capacitor instead of
> >current source). How would you do it?
> 
> When you say behaves like a current mirror to as high freq. as possible, 
> do you mean Iout/Iin is wideband or that the impedance looking into the 
> output is as high as possible?

The impedance looking into the output. What about putting a coil in series with
the collector?

> 
> Your basic approach you already described is fine.  I'd just pay 
> particular attention to the potential for instability.  This means don't 
> add extra C to ground from the emitter and maybe leave room in your 
> layout for some series R in the bases.

The bases of several transistors in the current supplies are wired together.  I
am concerned what happens if some load puts a surge of electricity into the
collector of one of them and it gets through the C-B capacitance on the
connected bases. Then the drive voltage of the current supplies goes up and all
supplies start producing more current. That extra current can reflect somewhere
inside the circuit and come the same way back, amplified. Then it would start
oscillating.

I want to make sure the current supplies are immune to this. If I block the
bases with 100nF to ground, the transistors start being common base
amplifier. That means if something gets induced on the emitter resistor, a
current of the same magnitude will be copied into the collector and result
in possibly high voltage difference if the current source is loaded with
high impedance.

Blocking the emitters to ground is complete nonsense.

So what remains is blocking individual transistors, putting 100nF over the E-B
junction. But if a current surge comes from emitter, it gets coupled through
the C-B capacitance and E-B in sync get moved up. The transistor has E-B
charged to constant voltage and that corresponds to the pre-set current. The
resistor in emitter now starts taking more current and it has to come from
somewhere - so current start flowing through the E-B 100nF capacitor and be
eaten from the connected bases. But they don't care as they are not sensitive
to momentary fluctuations.

Disadvantage of the latter approach is that it requires a separate capacitor
for each transistor. But what do you think is the best way? Putting the
capacitor between ground and base, or between emitter and base?

An artificial inductor also consists of a transistor with a big capacitor
between E and B, doesn't it?

CL<
> 
> -Dan


_______________________________________________
geda-user mailing list
geda-user@xxxxxxxxxxxxxx
http://www.seul.org/cgi-bin/mailman/listinfo/geda-user