Transcript RISC Processor Architecture

How NOT to Design Active Filters
Active Filter Design Software
is flexible, inexpensive and easy to use
But practical aspects of hardware design
frequently degrade the performance of
the very best theoretical circuits
This article talks about hardware – not software
How NOT to Design Active Filters
An Ideal Operational Amplifier
has infinite gain, bandwidth, slew rate, CMRR & PSRR
and zero offset, drift, bias current, noise, crosstalk (in duals & quads) and output impedance
Real op amps (even Analog Devices ones) are more messy!
How NOT to Design Active Filters
A Real Operational Amplifier
has finite gain, bandwidth, slew rate, CMRR & PSRR
and appreciable offset, drift, bias current, noise, crosstalk (in duals & quads) and output impedance
Active filter designs must take account of these realities
How NOT to Design Active Filters
The gain of a real op amp is not infinite
Neither is the bandwidth
Typical values of Aol for general purpose op amps are 105 - 107
Gain bandwidth products of such op amps rarely exceed 10 - 20 MHz
and are frequently much lower in high precision parts
If Aol is 106 and GBP is 1 MHz
Open Loop Gain at 20 KHz is only 50
Which is far from infinite!
How NOT to Design Active Filters
High-speed op amps have wider bandwidth
But, usually, much lower Aol
Typical values of Aol for high-speed op amps are 103 - 105
but are sometimes even lower
GBP of high-speed voltage feedback op amps may reach 350 MHz
Current-feedback (transimpedance) op amps
DO NOT HAVE a gain bandwidth product
To a first approximation their BW is not affected by the gain for a given Rfb
Their bandwidths can reach 1 GHz but they are not suitable for active filters
Active filters made with op amps should not be used at over 20 MHz -above this frequency passive LC filters with amplifiers providing
interstage gain and isolation will give more reliable performance
How NOT to Design Active Filters
When designing active filters
it is best to treat an op amp as an integrator
How NOT to Design Active Filters
But some op amps have a second pole*
at HF which can cause instability if ignored
(*His name is Frederic Chopin)
How NOT to Design Active Filters
“Spice” macro-models sometimes
omit HF poles & zeros
This is partly to permit reasonably rapid convergence
and partly because too complete a model enables
our competitors to deduce how the amplifier is designed
How NOT to Design Active Filters
Think about large-signal bandwidth
as well as small-signal bandwidth
For an op amp with a slew-rate of SR (V/Sec)
the relationship between full power bandwidth
(FPBW) and pk-pk output swing 2Vpk is:-
FPBW  SR
2Vpk
How NOT to Design Active Filters
Zout of an op amp reacts with Cload to produce
an additional pole which may cause instability
Op amps exist which are designed to drive capacitive loads without
instability, but such amplifiers are still slowed by load capacitance.
How NOT to Design Active Filters
Transimpedance or current-feedback
op amps oscillate with capacitive feedback
Current-feedback or transimpedance op amps
are a relatively new architecture of HF op amp
They have a low-impedance current input at their
inverting input and oscillate with capacitive feedback
Therefore they cannot be used in many
classical active filter configurations
How NOT to Design Active Filters
Adequate supply decoupling is essential –
this means that supplies must be short-circuited
at all frequencies above DC*
At low frequencies decoupling capacitors may be shared between several ICs,
but at HF each op amp must have its own decoupling.
HF decoupling capacitors must be low inductance types (ideally surface mount)
and must have short, wide, low inductance leads and PC tracks.
(* DC short-circuits are inadvisable.)
How NOT to Design Active Filters
Most people remember the offset voltage
many forget the bias current
When Ib flows in a resistance
it increases the effective Vos
When designing active filters there is a temptation to use large
resistances so that one can use small capacitors (which are
cheaper and more readily available at high accuracies)
This can cause high offsets
Sometimes this matters – sometimes it doesn’t
How NOT to Design Active Filters
Bias compensation can help
but only if Ib+ & Ib- are equal
Bias compensation resistor Rbc has the same resistance as the parallel combination of Rin and Rfb
(Decoupling Rbc ensures HF stability)
How NOT to Design Active Filters
Modern “single supply” & “rail-to-rail”
op amps often have higher bias current
than previous generations of op amps
This is because techniques to reduce Ib do not work if the input
common-mode range must include one or both supplies
op amps with FET Inputs
do not have this problem
How NOT to Design Active Filters
NEVERTHELESS
the use of FET input op amps do not
allow the use of very high resistances
because high resistance is associated
with high Johnson noise
All resistances have Johnson noise of
4kTR
V
Hz
T is the temperature in Kelvin, R is the Resistance,
k is Boltzmann’s Constant (1.38 x 10-23 Joules/K)
[It is rarely profitable to reduce the temperature, one can reduce the resistance,
but it is not possible to change Boltzmann’s Constant as Boltzmann is dead]
How NOT to Design Active Filters
AMPLIFIER NOISE
Every op amp contains three
uncorrelated noise sources
Voltage noise Vn
Current noise in the non-inverting input In+
Current noise in the inverting input In-
How NOT to Design Active Filters
When calculating the noise of an amplifier
it is necessary to consider the effects
of all three amplifier noise sources and
also the Johnson noise of all resistors used
The op amp current noise In generates voltage noise
when it flows in any impedance, resistive or reactive
But only resistances have Johnson noise
The diagram on the next slide shows only resistances
but a more general (and complex) diagram would show
reactive and resistive impedances
How NOT to Design Active Filters
How NOT to Design Active Filters
TO SUMMARIZE
Do not assume all Op Amp parameters are either zero or infinite
but actually consider the effects of finite non-zero Aol, GBP, Ib,
slew rate, crosstalk, noise (voltage & Current), CMRR, PSRR and Zout
&
RTFDS*
* Read The Friendly Data Sheet
(See article at http://www.analog.com/analog_root/static/raq/raq_caveat.html)