Reply-To: "Kevin Aylward"
From: "Kevin Aylward"
Subject: Re: Determining bjt noise parameters for Spice models?
X-Newsreader: Microsoft Outlook Express 6.00.2800.1106
X-Inktomi-Trace: public1-pete2-5-cust33.pete.broadband.ntl.com 1039558385 1168 220.127.116.11 (10 Dec 2002 22:13:05 GMT)
Date: Tue, 10 Dec 2002 22:13:04 -0000
NNTP-Posting-Date: Tue, 10 Dec 2002 22:13:06 GMT
> I wanted to predict noise properties of some bjt-based circuits using
> Spice. Then I've noticed that the bjt models (almost) never include
> the noise parameters (AF, KF ). How can those be determined from the
> data sheets?
> I've tried the advice found in a Spice manual: set AF=1 and then find
> KF such that it gives the requested noise figure; I've tried this for
> 2SC3329. Ok, I've found something like KF=5e-15 to get the specified
> NF at 1 kHz with Ri=20k.
Then I've varied the collector current but
> have not observed the specified NF changes! I've then looked at the
> NF at 10Hz. Apparently, AF=1 does not fit this transistor; I've
> tweaked it to AF=1.13 to get the noise figures ressembling those in
> the datasheet, but without knowing what I am doing!
This won't really work. KF is the flicker noise co-coefficient, or 1/f
noise contribution. A low noise transistor will have a 1/f corner such
that it won't contribute much noise at 1Khz. Secondly, you should use a
zero source resistance. So, look at the data sheet graph, and set KF to
match the data sheet at a low frequency, say 10 Hz. Do this in voltage
noise terms (nv/sqrthz), NF is pretty much useless.
And... AF is the flicker noise exponent. i.e.
Vn^2 = KF/f^AF in V/sqrthz
This noise is added to the normal white noise of the transistor.
> - would it be possible that someone explains the meaning of AF and KF
> (refer to an on-line doc, etc.) and how to calculate them from the
> - in general, is the bjt model in Spice capable of precisely enough
> modelling the noise behaviour of all the bjts out there (in
> particular: 2SC/2SA Toshiba/Hitachi low-noise series)?
Just about. You need to make sure the base resistance is set right. This
is the main variable that distinguishes transistor noise form exh other.
Other noise contributions are inherently fixed by Ic and hfe. So, after
the 1/f curve flattens out, set rbb until the flat portion reads
correctly. Do this at a highish collector current, e.g 10ma, also with r
source set to zero.
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.