Subject: Re: Determining bjt noise parameters for Spice models?
Date: Sat, 14 Dec 2002 23:27:50 +0100
Organization: Wanadoo, l'internet avec France Telecom
References: <3DF6537E.B0FC9DBE@worldonline.fr> <3DF90203.C9691374@worldonline.fr> <3DF94AE1.firstname.lastname@example.org>
NNTP-Posting-Date: 14 Dec 2002 22:27:51 GMT
X-Mailer: Mozilla 4.79 [en] (X11; U; Linux 2.4.20 i686)
Robert Baer wrote:
> "A.Iakovlev" wrote:
> I have some old but useful references concerning noise as a function
> of frequency and source resistance.
> Keithly: "electrometer measurements" 1972
> Princeton Applied Research Corp : "How to use Noise Figure Contours"
> Michael Steffes: "Understanding and Improving Op Amp Noise and
> Distortion" date unknown and company unknown; presumed date around 1970.
> I may have others; these weer handy at a moments notice.
> If you are interested, i could scan at 200 DPI to keep the sizes down
> and e-mail them on request.
Thank you. At this time, I would like just to understand the basic things about
the bjts' noise.
> Start with a differential bipolar transistor pair, emitters connected
> together and driven by a current source; bases are inputs and equal
> resistors on collectors with signal testing across them.
> That is a crude description of the input stage to the uA709; a rather
> low noise op amp of those days (ignore the popcorn which was due to
> impurities in the process chemicals and environment).
> For a given current, one can calculate the noise voltage, and then
> measure actual noise voltage; they would agree within measurement
> accuracy (in the "flat" band of noise) over a rather wide range.
> This appears to work over a fairly wide range of source impedances,
> but as the source R increases, that "flat band" decreases materially -
> and one has to add in (noise current)*(source R).
> I have done this, but did little work to see about hi R noise
> correlation (measured VS calculated).
Do you mean the correlation between the different noise sources? The thermal
noise of RB,RC,RE, shot noise of Ic and Ib?
> The PAR cover has a graph, log freq horiz VS log Rin at 290 degrees
> Kelvin (temperature is *very* important).
> Looks somewhat like a topo plot of an oblong valley in the middle;
> lowest regions following a downward line (to right).
> Every system has its own characteristic curves, which greatly depends
> on the devices used, their operating current, and the topology
> I do not quite understand what you meant by "..(using) the standard
> 2SC3329 model..could not reach..much lower levels obtained.."
> Sort of sounds like you got better noise levels than the reference.
Yes, obtained lower noise levels than what said the datasheet.
> And that may be correct; were your conditions identical (circuit,
> operating current, temperature)?
Not sure about the circuit, but the datasheet does not specify much. I used
common emitter configuration, a resistance as collector load, the Vce, Ice,
Temperature as specified in the ds.
> The Rb on most bipolars is so low that it can be ignored in most
> cases; Rbb' (base spreading resistance) is in a similar class.
> The Re resulting from the biasing current is the major En contributor,
> and this is where one gets very good correlation measured VS calculated
> (take the simple single transistor case).
> Now when Ie is "large" to make for low Re, then Rb and Rbb' must be
> added; most data sheets will not give you any information about these
> So, measure noise with low Ic and with high Ib (low source R); in both
> cases, the noise is very close to the vector sum of Rs, Re, Rb and Rbb'.
> However, at the lower current, Re is higher and the base resistances
> can be ignored enough to approximate (closely) their combined
> (effective) value.
> Armed with that info, one can calculate NF at any reasonable frequency
> (say 1Khz to near Ft) and reasonable current; there is excellent
> correlation in AF and RF measured values.
The question: how to calculate re, rb, rbb', rc?
> Sorry about the long-winded mess above; hope is of some help.
> > Doing that with the standard 2SC3329 model and varying RB from 2 to 10 Ohm
> > could not reach the 0.6nV/rtHz specified by the datasheet - much lower
> > levels obtained; the datasheet announces rb about 2Ohm.