ABCD To Z And Back
A little while back I was reminded of the ABCD matrix during a discussion
on the synth-diy mailing list. Their power and simplicity is perfect for
design and analysis of passive filter networks. On this page I bring together
the theory of the ABCD matrix, and show how it can be used to solve seemingly-complex
passive networks using nothing more than 2x2 matrices. And along the way we'll use
computers to do the tedious algebraic crunching.
A To D Via B And C
The ABCD Matrix (or Chain Matrix or Transmission Matrix) describes
a two-port network: a box with a two-wire port on the left and a two-wire port on the right.
One peculiarity of the ABCD matrix as compared to the S, T, Z and H matrices is the
direction of the current flow on the second port is reversed (those familiar with
two-port networks would have spotted the "wrong" sign of I2 already).
The importance of this
subtle distinction comes to light when we cascade - or chain - two networks
whereupon the output current of the first network becomes the input current of the second
network without tedious sign-changing to worry about.
The A, B, C and D refer to the names of the four components of the ABCD matrix. First
we write out the matrices in full:
From here we can write down the expressions for V1 and I1:
By setting V2 or I2 to zero (short-circuit or open-circuit respectively)
we can derive simple expressions for the four terms:
Components A and D are dimensionless ratios, B is in Ohms, and C is in Siemens.
A network is said to be symmetrical (input impedance equals output impedance) if A = D.
A network is reciprocal if AD - BC = 1; that is, the voltage at port 2 when a current is
applied to port 1 is the same as the voltage at port 1 when the same current is applied to port 2.
Now consider two networks chained together:
Starting from the left side of the circuit we have
Chaining the two we get
Note that the order of multiplication is important in matrix math so make
sure you follow the order as described by the circuit.
There are just two basic networks to consider - what we might call the canonical forms.
These are the series element and the shunt element.
The series element has one impedance - Z - in series with two of the port connections.
The convention is as shown below although there is no reason it cannot be in the bottom
If we now consider the four matrix terms by examining the circuit with the
right-hand port open- and close-circuited then we find the corresponding matrix
Similarly, the shunt element has one impedance connected across, or shunting,
the two sets of connections:
From which we can derive the matrix as:
Worked Example: RC Low Pass Filter
To illustrate how ABCD matrices are applied we show how the tranfer function
of the well-known RC low-pass filter can be derived from a chain of a series resistor
and a shunt capacitor.
After matrix multiplication we get
If we assume that the output of the resulting network connects to a high-impedance
circuit (e.g., op-amp buffer) then we can use the A term as simply the inverse of the transfer
function of the network:
which is the well-known transfer function of an RC low-pass filter.
Running In Parallel
ABCD matrices work fine for chaining two or more networks in series, but are rather
difficult for parallel networks (e.g., twin-T notch filter). The solution is to use
Y (admittance) networks as an intermediate step, converting the two ABCD matrices
into Y matrices, adding them, and then converting the resulting Y matrix back into an
4-Stage RC Low Pass Filter
Given that we have just seen how the classic RC filter can be analysed with the
help of the ABCD matrix let's take a look at a more involved example.
Consider the 4-stage unbuffered RC filter in the paper by Electronotes/Hutchins
(ELECTRONOTES 210 Vol. 22 May 2012).
The filter in question is reproduced below:
This circuit comprises four RC sections in cascade. According to the author:
I don't know how many times I have started out to actually solve the unbuffered case (A).
I always gave up, even after dropping back to the case where all the R's and all the C's were
equal. It clearly was not impossible - just excessively tedious.
We previously computed
the ABCD matrix for a single RC section, so the solution is the multiplication
of four RC sections:
Expanding out the multiplication would be rather tedious by hand, so we use
maxima to do the
algebraic gymnastics for us.
Setting all resistors equal and all capacitors equal simplifies the transfer
function to the following expression:
Factorising this we can determine the locations of the poles:
The first term is the familiar RC low-pass filter we derived above. The second term is rather more complex.
If we rewrite the term substituting p = sCR then we have a simple cubic expression
p3 + 6p2 + 9p + 1. Using any one of the
online cubic solvers
we find the three values for p = -0.121, -2.347 and -3.532.
Notice that they are all real, i.e., lie on the real axis.
We can now write down the three poles as: (p + 0.121)(p + 3.532)(p + 2.347). Substituting back we get
(sCR + 0.121)(sCR + 3.532)(sCR + 2.347). Finally we can write down the transfer function:
For the plot below we set R to 1k0,
C to 100n and sweep over the range 1Hz to 100kHz.
To confirm these results an AC simulation was run in
LTSpice confirming the theoretical filter behaviour:
Some interesting observations can be made from the spice plot. Firstly, note that at around 2kHz
the phase (dashed line) reaches 180° of phase shift. In other words a 2kHz sine wave is inverted,
albeit reduced in amplitude by about 26dB (roughly 1/20th of the input level).
Secondly, the phase shift continues heading towards 360° - by
the time the frequency reaches around 1MHz the phase shift is very close to 360°, although
this far from the cutoff point the attenuation is around 200dB. Things get rather theoretical
down at this level!
The classic twin-T filter is a purely RC circuit that can produce a very sharp
notch response. It (or its close cousin the bridged-T) is often used in audio analysers to remove
the fundamental signal. The circuit is remarkably simple, yet theoretically can achieve very
As can be seen the circuit comprises two T-sections (one RCR the other CRC) wired in parallel.
On one branch the current leads, and on the other the current lags. As the frequency approaches
the notch frequency the phase shifts approach ±90° effectively cancelling each other
out at the summing node. If all the components were ideal then the cancellation would be perfect.
Analysis proceeds by considering each of the T-sections in turn.
To facilitate the operations in Maxima I use these two definitions to convert between ABCD
and Y matrices:
abcd_to_y(m) := matrix([m[2,2]/m[1,2], (m[1,2]*m[2,1] - m[1,1]*m[2,2])/m[1,2]],[-1/m[1,2], m[1,1]/m[1,2]]);
y_to_abcd(y) := matrix([-y[2,2]/y[2,1],-1/y[2,1]],[(y[1,2]*y[2,1] - y[1,1]*y[2,2])/y[2,1],-y[1,1]/y[2,1]]);
The first section, comprising R1, C1 and R2, is described by three chained matrices:
The second section, comprising C2, R3 and C3, produces
For the next step we compute the ABCD matrices for the two sections,
The result is rather complicated, so we won't show it here.
What we are interested in is the transfer function, which is:
Still quite a handful, so we apply the usual simplification setting
R1 = R, R2 = R, R3 = R/2 and C1 = 2C, C2 = C, C3 = C.
That looks more tractable. If we then expand out s = jw we get
For a notch filter (band reject) we are interested in the frequency
at which the transfer function is zero (infinite attenuation). For this
we need only find when the numerator is equal to zero:
Finally we can determine the relationship between f, C, and R:
Which is the expected result.
In this section we have seen how the ABCD matrices can be used together
with automated tools (maxima in this case) to analyse the behaviour
of passive component networks. While intermediate results may look
scary, in practice we are primarily interested in the transfer function
which is readily extracted from the ABCD matrix.
As an aside, there may be readers thinking "Ah yes, but what if there
is an inverting op-amp on the output, so we can't assume that I2
is zero. What then smarty-pants?". In that specific case you can model
the input resistance of the inverting amplifier stage as a final shunt
resistor and then you can either consider the output voltage, as before,
or determine the current through the resistor as the current into the
transimpedance (current-in-voltage-out) amplifier.