From: "John Jardine"
Subject: Re: SPICE and amp stability
Date: Wed, 20 Nov 2002 01:40:44 -0000
NNTP-Posting-Date: 20 Nov 2002 01:29:16 GMT
X-Newsreader: Microsoft Outlook Express 5.00.2615.200
Walter Harley wrote in message
> I have a simple 100mW audio amp circuit that, naively implemented, is
> unstable at RF. The amp consists of an LM833 inverting stage with Av
> followed by a push-pull emitter follower biased as class AB. Feedback is
> taken from the output of the emitter follower.
> I can stabilize the amp by putting a small-value capacitor between the
> output and the inverting input - basically taking the emitter follower out
> of the loop at high frequencies. However, unless I use a very tiny value,
> this causes substantial intermodulation distortion as the emitter follower
> is not all that linear without feedback. If I do use a tiny value, I'm
> concerned that it won't be stable in real life.
> Rather than just try cap values at random, I would like to really
> what's going on. So, I'd like to model the phase response and gain
> bandwidth of the system, and find a way to compensate the circuit so that
> gain drops below unity by the frequency where there is 180 degree phase
> shift. But I'm not sure how to do this, either analytically or in PSpice,
> for four reasons that maybe y'all can help me with:
> 1) How do I figure out the phase response of the feedback network? The
> "capacitors" in the circuit are the input capacitance of the opamp, which
> not specified in the datasheets, and the various parasitic capacitances,
> whatever appears across the load (probably cable capacitance). Do I just
> 2) To simulate the system and calculate a Bode plot, should I model the
> parasitic capacitances in PSpice and then look at phase shift between the
> circuit input and the inverting input of the opamp? Or is there some more
> correct way to determine the loop characteristics?
> 3) Do the SPICE models for opamps accurately depict the bandwidth, phase
> shift, and input capacitance in general? If not, which of these
> do I need to externally model?
> 4) Does anyone know where to find a reasonable SPICE model for the LM833?
> can't find one anywhere, except for one on OnSemi's site that shows JFET
> inputs, which doesn't seem correct for an LM833.
> Thanks, everybody.
You have the perfect system model ready and waiting. Use it!. It is complete
with exact transistor and op-amp capacitances and a 100% perfect set of
parasitics including the nasty bits that can't be modelled in Spice such as
cross coupling, ground loops, physical locations, local interference etc.
First off, stick a 100ohm resistor in series with that emitter follower
base. These can be quite happy to merrily oscillate at R.F. by themselves.
It's not something that would be normally picked up by Spice.
To get the network response it is usual to open the loop gain up to the max
possible yet still just allowing the amp to function and then use the dual
channel scope to check the i/p o/p phase shift as i/p frequency is swept.
Based on this known (near)open loop phase shift v. gain, the feedback
components can then be set to allow stability at the wanted much lower
Surely, it's impossible to fault find or test a setup you've designed
without the most basic equipment such as 'scope and signal gen. Spice just
wont cut it here. The Spice is for use for 'what ifs' or for detailed
clarification in areas you already know the performance of.