Delft University of Technology

Faculty of Electrical Engineering,

Mathematics and Computer Science
Circuits & Systems Group

(L)WDF Toolbox
for
MATLAB

User’s Guide

Version 1.0

Ing. H.J. Lincklaen Arriëns

February 2006
 (L)WDF Toolbox for MATLAB User’s Guide
© H.J. Lincklaen Arriëns 2006

The author assumes no responsibility whatsoever for use of the software by other parties, and makes no
guarantees, expressed or implied, about its quality, reliability, or any other characteristic.
Acknowledgement if the software is used is appreciated.

MATLAB is a registered trademark of The MathWorks, Inc.

APLAC is copyrighted by APLAC Solutions Corporation.

ii (L)WDF Design Toolbox for MATLAB User’s Guide

Table of Contents

Introduction. ................................................................................................................................................ 2
Installation and requirements..................................................................................................................... 2
Designing in the Continuous-time Domain. ............................................................................................... 3
Comparing Butterworth Transfer Functions..................................................................................... 3
Comparing different filter approximation methods........................................................................... 6
Cauer Filter Parameter Conversion. .................................................................................................. 8
Frequency Transformations ............................................................................................................... 9
Ladder Circuits.................................................................................................................................. 10
Some 3rd order Vlach Low-pass Transfer Functions. ....................................................................... 13
Vlach Band-pass Transfer Functions. .............................................................................................. 15
Designing in the Discrete-time Domain. ................................................................................................... 16
Just for fun: Comparing a Cauer with a FIR Filter. ........................................................................ 18
Designing Wave Digital Filters (WDFs). ......................................................................................... 19
Lattice Wave Digital Filters (LWDFs)............................................................................................. 26
Bireciprocal LWDF designs............................................................................................................... 29
Vlach Band-pass LWDF Designs. ..................................................................................................... 31
Transmission Line Filter Designs. ............................................................................................................ 32
GUI’s that cover all the above. .................................................................................................................. 37
Interface to Scheduling Toolbox................................................................................................................ 39
Epilog.......................................................................................................................................................... 41
References. ................................................................................................................................................. 42

(L)WDF Design Toolbox for MATLAB User’s Guide 1

Introduction.

This Toolbox contains two GUIs and a number of functions to design continuous-time and discrete-
time filters, featuring the design of Wave Digital and Lattice Wave Digital filters. Although (nearly) all
functions are accessible through the GUIs, we start with giving a short description of some of the
separate functions to indicate our way of thinking and to clarify some of the notations used.

We will not go into the details of filter theory since numerous excellent books have already been
written about this subject. We assume the reader’s acquaintance with common denominations like
pass-band (ripple), transition-band, stop-band, etc. We also assume a reasonable experience with the
MATLAB interfaces.
Even though, the toolbox presented here will turn out to be easy to use, even without an in-depth
knowledge of filter theory.

At this moment the toolbox includes functions for designs based on a number of classical filter
approximations, viz. Butterworth, Chebyshev, Inverse Chebyshev and Cauer designs.
Next to these, an approximation method with more freedom for tailoring the stop-band, described by
Jiri Vlach [Vla69], has been added. This method also enables the design of filters using unit elements,
in which case these unit elements will contribute to a better approximation of the ideal filter
characteristic.
The filters above are all classified as Infinite Impulse Response (IIR) filters. With the toolbox, two
types of discrete-time structures can be created that show excellent performance for IIR realizations.
These structures, Wave Digital Filters (WDFs) and Lattice Wave Digital Filters (LWDFs) [Ant79]
[Gaz86][Law90][Nou79], have been derived using the less-known wave digital theory. For an
exhaustive description of these (L)WDFs, the reader is referred to the (invited) paper by Fettweis
[Fet86].

Unlike filter theory in its early days where usually attenuation functions where used as a reference,
we prefer to work with transfer function: for plots these are restricted to magnitude and phase
characteristics, although it is of course possible to also work with phase delays and group delays, but
this is left to the user. Being MATLAB functions, there are no reasons not to extent the number of
functions with more elaborate or esoteric approximations, plot-functions, etc.

Included with the Toolbox are a number of example m-files that can help in getting acquainted with
the syntax and the possibilities of the toolbox. Most of the following designs can be found in one of the
example files.

In the following chapters we will briefly show how the toolbox functions work, and indicate what can
be done with them, by means of listing snippets of code and the resulting pictures and command
window answers. For more detailed information and help text, the reader is referred to the “(L)WDF
Toolbox Reference Guide” and the several reports and test cases listed on and available from
http://ens.ewi.tudelft.nl/~huib/mtbx/

Installation and requirements.

Installation is very simple: just unzip the Toolbox’s zip-file to a directory of your choice. Either set
MATLAB’s Current Directory to this directory or add it to MATLAB’s search path, with e.g.

>> addpath(‘Your_Directory’);

Only basic MATLAB functionality is needed: there are no dependencies on other MATLAB Toolboxes.

2 (L)WDF Design Toolbox for MATLAB User’s Guide

Designing in the Continuous-time Domain.

We will describe a continuous-time filter with a transfer function, H (s ) , which is the quotient of two
f (s)
polynomials in the complex frequency variable s, viz. H ( s ) = .
g ( s)
The magnitude of the transfer function is given by H ( jω ) , i.e. the modulus of the complex function,
and the phase angle arg ( jω ) .

Usually we start by approximating a normalized low-pass design with a function, the magnitude of
which approximates the ideal low-pass characteristic as good as possible, viz. showing
• a pass-band for which the input frequencies are passed (nearly) unaltered,
• a stop-band for which input frequencies are attenuated as much as possible,
• and in between a transition-band as narrow as possible.

For a normalized low-pass filter the cut-off frequency that indicates the edge of the pass-band is
located at a radian frequency equal to 1. In the literature, the normalized angular frequency is
commonly denoted with the symbol Ω.

transfer gain cut-off frequency

1.0

pass-band stop-band

Figure 1. The ideal low-pass magnitude

transfer function. Ω
0
1.0

Comparing Butterworth Transfer Functions.

In Figure 2 a number of transfer functions that use the Butterworth approximation method have been
plotted. The approximations differ from each other in the order of the transfer function.
The commands, partly basic MATLAB functions, partly functions from this toolbox, to construct this
picture, are:

Hs = struct([]);
for i = 1:5
Hs = [Hs; nlpf('butter',i)];
end
plotHs(Hs,2,1,[0.01 1000],1,1000, ...
'Butterworth Characteristics', ...
[ 'N = 1'; 'N = 2'; 'N = 3'; 'N = 4'; 'N = 5'] );
figure(1)
axis([ xlim -80 10])
xlabel('Normalized Frequency');
figure(2)
axis([ xlim -460 10])
xlabel('Normalized Frequency');

(L)WDF Design Toolbox for MATLAB User’s Guide 3

To be more exact, the magnitude and phase transfer functions for the 1st order upto and including the
5th order Butterworth low-pass filters, are calculated and shown.

Figure 2 clearly shows that increasing the filter order results in an increasingly better approximation
of the ideal, rectangular low-pass transfer function, at the expense of larger phase-differences between
in- and output signals.

See the Reference Guide for a detailed explanation of the syntax to be used for the Toolboxes’
functions.

Continuous-time Butterworth Characteristics

10
N= 1
0 N= 2
N= 3
-10 N= 4
N= 5
-20
Magnitude in dB

-30

-40

-50

-60

-70

-80
-2 -1 0 1 2 3
10 10 10 10 10 10
Normalized Frequency

Continuous-time Butterworth Characteristics

0
N= 1
-50 N= 2
N= 3
-100 N= 4
N= 5
Phase angle in degrees

-150

-200

-250

-300

-350

-400

-450
-2 -1 0 1 2 3
10 10 10 10 10 10
Normalized Frequency

Figure 2. Magnitude and phase transfer functions

for Butterworth low-pass filters.

4 (L)WDF Design Toolbox for MATLAB User’s Guide

All 5 transfer functions are combined in one structure array, Hs:

Hs is 5x1 struct array with fields:

poly_fs
poly_gs
ident
roots_fs
roots_gs

from which we can access all individual transfer functions, e.g. the one for N= 5:

>> Hs(5)
ans =
poly_fs: 1
poly_gs: [1.0000 3.2361 5.2361 5.2361 3.2361 1.0000]
ident: 'LP PROTOTYPE: 'butter',5,1'
roots_fs: []
roots_gs: [ -1.0000
-0.8090 - 0.5878i
-0.8090 + 0.5878i
-0.3090 - 0.9511i
-0.3090 + 0.9511i ]

f (s) and g (s) for this H (s ) are directly available in MATLAB’s polynomial notation, as
f ( s) = 1.0
g ( s ) = s 5 + 3.2361 s 4 + 5.2361 s 3 + 5.2361 s 2 + 3.2361 s + 1.0000

f (s) thus has no roots, while the relationship between g (s) and roots_gs is given by:
N
g (s) = ∏ { s − roots_gs(i)}
i =1

According to the theory, this 5th order polynomial results in one real root, and two pairs of complex
conjugated roots.

Recapitulating:
1. 0
H ( s) =
s 5 + 3.2361 s 4 + 5.2361 s 3 + 5.2361 s 2 + 3.2361 s + 1.0000

(L)WDF Design Toolbox for MATLAB User’s Guide 5

Comparing different filter approximation methods.

With the following instructions a Butterworth, a Chebyshev, an Inverse Chebyshev and a Cauer
approximation – all of them of order 3 – are drawn in one plot.

N = 3;
Hs_btw = nlpf('butter',N);
Hs_che = nlpf('cheby',N,1,1);
Hs_ich = nlpf('invcheby',N,40,1);
Hs_cau = nlpf('cauer',N,1,40,'a',1);
plotHs([Hs_btw;Hs_che;Hs_ich;Hs_cau],2,1,[0.1 100],0,1000, ...
'Transfer Characteristics', ...
[ ' butterworth '; ' chebyshev '; ' inverse cheb'; ' cauer ' ] );
axis([ xlim -60 10])
xlabel('Normalized Frequency');

For a better comparison of the filter shapes, the normalization of the cut-off frequencies is chosen such
that all magnitude transfer function show a 3dB attenuation at Normalized Frequency = 1.

Continuous-time Transfer Characteristics

10
butterworth
chebyshev
0
inverse cheb
cauer

Figure 3. Magnitude transfer functions -10

for 3rd order Butterworth, Chebyshev,

Magnitude in dB

Inverse Chebyshev and Cauer low-pass -20

filters.
-30

-40

-50

-60
-1 0 1 2
10 10 10 10
Normalized Frequency

Impulse responses of 3rd order normalized low-pass filters

0.4
Output Levels for a Unit Impulse

0.3 butterworth
chebyshev
Figure 4. Impulse response in the time inverse cheby
domain for the filters of Figure 3. 0.2 cauer

0.1

-0.1
0 2 4 6 8 10 12 14 16 18 20
Time [s]

6 (L)WDF Design Toolbox for MATLAB User’s Guide

From Figure 3, the phase plot is omitted. To obtain a phase plot, follow the previous example: it
should show that the phase characteristics are less smooth then the Butterworth’ one.

In Figure 4, the impulse responses of the same four filters are shown, obtained using MATLAB’s
impulse.m function (from the Control Toolbox). It can be seen that these filters belong to the class of
Infinite Impulse Response (IIR) filters.

The parameters for the Cauer filter are now:

>> Hs_cau
Hs_cau =
poly_fs: [0.0638 0 0.4121]
poly_gs: [1 0.9015 1.0560 0.4121]
ident: 'LP PROTOTYPE: 'cauer',3,1,40,'A',1,1'
roots_fs: [2x1 double]
roots_gs: [3x1 double]
>> Hs_cau.roots_fs
ans =
0 - 2.5420i
0 + 2.5420i
>> Hs_cau.roots_gs
ans =
-0.4826
-0.2094 - 0.9000i
-0.2094 + 0.9000i
>>
0.0638 s 2 + 0.4121
from which can be read that H ( s) =
s 3 + 0.9015 s 2 + 1.0560 s + 0.4121

The locations of the roots of g (s ) in the complex plane are shown in Figure 5 and can be seen to be
connectible with an elliptical function.

Pole Locations

0.5
Imaginary Part

-0.5

-1

-1 -0.5 0 0.5 1
Real Part

Figure 5. Location of the poles of the 3rd order

Cauer filter from this example.

(L)WDF Design Toolbox for MATLAB User’s Guide 7

From filter theory it is known that odd order Cauer approximations are usually realizable as lumped
element ladder circuits (see also later on). This is not so for even order transfer functions, which end
up with negative component values and thus are not realizable as ladder structures. The cause is that
for even order approximations, both at zero as at infinite frequencies, the transfer gain should have a
distinct value instead of 1 or zero. Skwirzynski [Skw65] described a solution for this problem, viz. to
either only shift the highest notch in the stop-band to infinity (Type ‘B’), or to also shift the first peak
in the pass-band to zero frequency (Type ‘C’). Both Type B and C transfer functions result in
realizable ladder structures.
The unaltered function will be denoted as Type ‘A’.

The following functions are plotted in Figure 6:

nlpf('cauer',4,1,45,'A',1)
nlpf('cauer',4,1,45,'B',1)
nlpf('cauer',4,1,45,'C',1) Continuous-time 4th Order Cauer LP-Filter Characteristics
10

-10
Type A
0.5 -20 B
Magnitude in dB

C
0
-30
-0.5
-40
-1

-1.5
-50
0 0.2 0.4 0.6 0.8 1

-60

-70

-80
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Figure 6. 4th Cauer filters, types A, B and C with Normalized Frequency
zoomed-in pass-band.

Cauer Filter Parameter Conversion.

If you are more accustomed with specifying Cauer filters in terms of reflection coefficient ρ and
modular angle θ , then function cc2pars.m can help with translating ρ and θ into the parameters used
by the Toolbox.
A filter, identified in a catalog classification as e.g. ‘C06 B 25 48’ can be translated with
>> [N,rp,rs,ftype,Wn,normtd] = cc2pars(6,25,48,'b');
and then computed with e.g.
>> [Hs,wp] = Hs_cauer(N,rp,rs,ftype,Wn,normtd);
or translated into a ladder structure using about the same parameters (see later on).
Functions ripple2rho and rho2ripple are simple conversion routines with self-explanatory names.

8 (L)WDF Design Toolbox for MATLAB User’s Guide

Frequency Transformations.

With the toolbox the following well-known frequency transformations are possible:

nlp2lp: converts a normalized low-pass to low-pass with an arbitrary cut-off frequency,

nlp2hp: creates a high-pass transfer function,
nlp2bp: obtains a band-pass filter, and
nlp2bs: transforms the normalized low-pass into a band-stop filter.

Below follows an example of the nlp2bp function, given our previously calculated 3rd order, type ‘A’,
Cauer filter:

Hs_cau_bp = nlp2bp(Hs_cau,1,1);
Hs_cau_b1 = nlp2bp(Hs_cau,1,2);
Hs_cau_b2 = nlp2bp(Hs_cau,10,10);
plotHs([Hs_cau_bp;Hs_cau_b1;Hs_cau_b2],2,1,[],0,1000, ...
'3rd order Cauer lp-bp Transformations', ...
[ ' fc = 1, BW = 1'; ' fc = 1, BW = 2'; ' fc = 10, BW = 10 ' ] );
axis([ xlim -60 20])
xlabel('Normalized Frequency');

Continuous-time 3rd order Cauer lp-bp Transformations

20
fc = 1, BW = 1
10 fc = 1, BW = 2
fc = 10, BW = 10

-10
Magnitude in dB

Figure 7. Three 6th order Cauer filters, each -20

obtained with a normalized low-pass to band-
pass transformation. -30

-40

-50

-60
-2 -1 0 1 2
10 10 10 10 10
Normalized Frequency

Notice that the transformation causes the band-pass filter to be arithmetically symmetric around the
center-frequency in Figure 7 (thus geometrically symmetric if plotted with a logarithmic frequency
scale).

To obtain the transfer function of a low-pass filter with a -3dB cut-off frequency at 15 kHz, e.g.
>> Hs_cau_15k = nlp2lp(Hs_cau,2*pi*15000);

Note that if one is only interested in low-pass filter designs, it is also possible to use the functions
Hs_butter, Hs_cheby, Hs_invcheby, Hs_cauer and/or Hs_Vlach which allow a parameter
cutOffFrequency to be passed directly, instead of the combination nlpf and nlp2lp.

(L)WDF Design Toolbox for MATLAB User’s Guide 9

Ladder Circuits.

The normalized 3rd order Cauer filter from a previous example can be realized as a ladder circuit:

>> nlpLadder = nlp_ladder('cauer',3,1,40,'a',1); L2

RS L2
L L2
RS 2
RS RS
Rs 1.00000 Ohm
C01 2.07190 F in shunt arm C2
C1 C2
C C2 C3 RL
L02 0.98116 H, parallel with 2
C02 0.15773 F in series arm
a) C1
C1 C1 C3
C3
RL
C3 RL RL
C03 2.07190 F in shunt arm
RL 1.00000 Ohm

or, if its dual realization form is wanted, starting with an inductor in a series arm (use ‘z’):

>> nlpLadder_d = nlp_ladder('cauer',3,1,40,'a',1,'z');

RS L1 L3
Rs 1.00000 Ohm
L01 2.07190 H in series arm
L02 0.15773 H, in series with L2
b) RL
C02 0.98116 F in shunt arm
L03 2.07190 H in series arm C2
RL 1.00000 Ohm

Figure 8a and b. Two functionally equivalent ladder circuits.

It was indicated above that even order Cauer filters of type ‘A’ are not realizable as ladder structures
with positive element values. Below we show 4th order designs, realizing a Type B and a Type C
transfer function. The ladder structure is the same for both Types, only the component values differ.
Because of the fact that for a Type B function the transfer gain at zero frequency differs from 1, the
source and load resistances cannot be equal. If this should be a problem, then a Type C will be needed.

4th order Cauer low-pass, type B: different source and load resistance values are needed.

Rs 1.00000 Ohm
C01 1.84916 F in shunt arm
L02 0.86103 H, parallel with
C02 0.41910 F in series arm
C03 2.65202 F in shunt arm L2
RS L4
L04 0.83131 H in series arm
RL 0.37598 Ohm
C2
same specifications as above, but now a type C version: C1 C3 RL
equal source and load resistances, both 1.0 Ω .

Rs 1.00000 Ohm
C01 1.40699 F in shunt arm
L02 1.29079 H, parallel with
Figure 9. Ladder circuit for both 4th order
C02 0.24945 F in series arm
C03 1.50477 F in shunt arm types B and C Cauer low-pass filters.
L04 1.62097 H in series arm
RL 1.00000 Ohm

10 (L)WDF Design Toolbox for MATLAB User’s Guide

The same transformation methods that we mentioned before are also applicable on ladder structures,
called resp. nladder2lp, nladder2hp, nladder2bp and nladder2bs here.

With nladder2bp, the ladder of Figure 8a will be transformed into a structure that realizes the
transfer function of Hs_cau_bp (shown in blue in Figure 7).

>> bpLadder = nladder2bp(nlpLadder,1,1);

>> showLadder(bpLadder,3,'after 3rd order Cauer lpladder-bpladder Transformation');

L2 L3
RS

C2 C3
L1 C1 L4 C4 RL

Rs 1.00000 Ohm
L01 0.48265 H, parallel with
C01 2.07190 F in shunt arm
Figure 10. 6th order Cauer band-pass ladder circuit,
L02 0.67867 H, parallel with
resulting from a transformation of the ladder in Figure 8a.
C02 0.17663 F in series arm
L03 5.66148 H, parallel with
C03 1.47347 F in series arm
L04 0.48265 H, parallel with
C04 2.07190 F in shunt arm
RL 1.00000 Ohm

Another example:
a high-pass filter with a cut-off frequency of 8 kHz, with source and load resistance values of 2700 Ω .

This is a two-step process (transform, then scale), for we first have to execute the transformation:

>> nlpLadder = nlp_ladder('cauer',3,1,40,'a',1,'z');

>> hpLadder = nladder2hp(nlpLadder,2*pi*8000); % 2*pi for realistic frequencies!

RS C1 C3
Rs 1.00000 Ohm
C01 0.00001 F in series arm
L02 0.00002 H, in series with L2
RL
C02 0.00013 F in shunt arm a)
C03 0.00001 F in series arm C2
RL 1.00000 Ohm

Notice that every time a ladder structure is computed, its magnitude transfer function will be reconstructed from
the ladder’s topology and will be compared with the original one.

(L)WDF Design Toolbox for MATLAB User’s Guide 11

 Transfer function reconstructed from ladder topology
10
b)
0

-10

Figure 11a and b. High-pass ladder -20

Magnitude in dB
circuit and its Magnitude Transfer function -30
for the component values in Table 1.
-40

-50

-60

-70

-80
0 2 4 6 8 10 12 14 16 18 20
Frequency in kHz

In the Toolbox as is, there is no special function to do impedance scaling, since the design of real
lumped element filters is not really the goal of this toolbox.
However, since the structure of the ladder is given (see the Reference Guide for a detailed description),

hpLadder =
elements: 'rcScR'
values: [6x1 double]

the following code will do the job (also applicable to band-pass and band-stop structures, but no unit
elements allowed):

function scVals = impScale(Ladder,R)

scVals = Ladder.values;
elements = strrep(lower(Ladder.elements),'s','lc');
elements = strrep(elements,'p','lc');
isC = (elements == 'c');
scVals( isC) = scVals( isC)/R;
scVals(~isC) = scVals(~isC)*R;

For hpLadder just calculated, and the given R = 2700 Ω , we can find:

Table 1. Rs 2k7 Ω
C01 3.6 nF
L02 54.7 mH Check in the plot that the notch is really at the
1
C02 46.7 nF calculated frequency of = 3.147 kHz.
2π L02 ⋅ C 02
C03 3.6 nF
RL 2k7 Ω

Note:
Several books, like [Zve67] and [Chr66], contain numbers of tables to calculate the values of ladder
structures with. When comparing those values with the ones calculated with the Toolbox, be sure to
use the same frequency normalization method with the Toolbox as has been done by the author(s).

12 (L)WDF Design Toolbox for MATLAB User’s Guide

Some 3rd order Vlach Low-pass Transfer Functions.

In [Vla69], J. Vlach described an approximation method that we have implemented as the ‘Vlach-
method’. The basic idea had been published by Sharpe [Sha54], but was extended, made more
accurate, and was also applied to band-pass approximations by Vlach.
Next to that, this method has the advantages that
• notches can be inserted in the stop-band (stop-bands for bandpass filters), and that
• unit elements that contribute to the filter approximation can be inserted.

When no notches are applied, a Vlach low-pass equals a Chebyshev approximation. This will be shown
in the next example for a 3rd order filter. Also, we will investigate what will be the result when we
position the notch at the exact frequency as where a 3rd order Cauer approximation (same pass-band
ripple and same frequency normalization method) would have located its notch.

% First find the frequency of the notch in the stop-band for a 3rd order Cauer filter.
% But the relation ws=1/wp holds true for the 'symmetric' Cauer transfer functions.
[Hsx,wp] = Hs_cauer(3,1,40,'a',1,-1);
wp2 = wp(2); % skip wp at w=0
ws2 = 1/wp(2);
% Again the Cauer filter, but now with its cut-of frequency at the -3dB point.
[Hs_cau,wp] = Hs_cauer(N,1,40,'a',1,1);
ws = wp(2)/wp2 * ws2; % This will be the notch frequency for our cut-off
% frequency reference method (-3 dB)

% Four Vlach transfer functions:

Hs_vla_a = Hs_Vlach(3,1,1,[],0,1); % no notch
Hs_vla_b = Hs_Vlach(3,1,1, 3 , 0,1); % notch at w = 3
Hs_vla_c = Hs_Vlach(3,1,1, ws , 0,1); % notch at Cauer’s ws
Hs_vla_d = Hs_Vlach(3,1,1, 2 , 0,1); % notch at w = 2

Continuous-time 3rd order Vlach Characteristics

10
no notch
fnotch = 3.0
0
fnotch eq cauer(1,40)
Figure 12. Four 3rd order Vlach filters. fnotch = 2.0
-10
Magnitude in dB

-20

-30

-40

-50

MATLAB shows that -60

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Normalized Frequency
>> ws
ws =
2.54202809474093

(L)WDF Design Toolbox for MATLAB User’s Guide 13

Alternatively, we could have realized that the roots of f (s ) , at which frequency or frequencies the
transfer function H (s ) becomes zero, immediately could have given us the value of ws.
Indeed, we can find that

>> Hs_cau.roots_fs
ans =
0 - 2.54202809474093i
0 + 2.54202809474093i

from which we see that the roots are purely imaginary and equal to ± j ws .

To be exact: in fact ws equals

>> fs = conv( [1 -Hs_cau.roots_fs(1)], [1 -Hs_cau.roots_fs(2)] );
>> ws = sqrt( fs(end) );

Now compare Hs_che with Hs_vla_a.

The poly_fs’s and poly_gs’s are −apart from accuracy matters in the computation− almost exactly
equal, i.e. a Vlach approximation without additional notches equals the Chebyshev approximation.

Hs_che =
poly_fs: 0.37434096543120
poly_gs: [1 0.90270351281101 1.03309585058947 0.37434096543120]
ident: 'LP PROTOTYPE: 'cheby',3,1,1,1'
Hs_vla_a =
poly_fs: 0.37434096541660
poly_gs: [1 0.90270351279928 1.03309585056261 0.37434096541660]
ident: 'LP PROTOTYPE: 'vlach',3,1,1,[],1,0'

Also, there is a very large similarity between Hs_cau and Hs_vla_c (the ‘simulated’ Cauer version).
This means that if the notch is positioned at the same frequency where the Cauer approximation
would have located it, then the Vlach and Cauer transfer functions will be identical.

Hs_cau =
poly_fs: [0.06377455880975 0 0.41210525743689]
poly_gs: [1 0.90152483145585 1.05600921457728 0.41210525743689]
ident: 'LP PROTOTYPE: 'cauer',3,1,40,'A',1,1'
Hs_vla_c =
poly_fs: [0.06377455885052 0 0.41210525770037]
poly_gs: [1 0.90152483163557 1.05600921500788 0.41210525770037]
ident: 'LP PROTOTYPE: 'vlach',3,1,1,[2.542],1,0'

Note: Given a filter order N, the maximum number of notches that can be inserted is floor(N/2).

14 (L)WDF Design Toolbox for MATLAB User’s Guide

Vlach Band-pass Transfer Functions.

The Vlach approximation is perfectly suitable for designing band-pass filters directly (without a
transformation), since the cut-off frequencies of the pass-band can be chosen separately and without
affecting each other, while the position of notches (if any) in the stop-bands can be tuned to be exactly
at the frequencies of disturbing signals.

Also, more or less geometrically symmetric filters are obtainable, as in shown in the next example:

>> Hs = Hs_bpVlach(10,0.1,[1.5 2.5], [0 1.0 3.0],0,1);

>> plotHs(Hs,1)

Continuous-time Characteristics
10

-10

-20

-30
Magnitude in dB

-40

-50

-60

-70

-80

-90

-100
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency

Figure 13. Vlach 10th order band-pass design with one notch
in the lower and one notch in the upper pass-band.

As can be seen from Figure 13, the transfer function between the notches is (almost) symmetrical
f c _ lower + f c _ upper
around 2.0, i.e. (the geometrical mean).
2

We will reveal more about the capabilities of the Vlach functions when treating microwave filter
designs.

Note:
For a band-pass filter to be realizable there should be at least one notch at frequency 0.
Given a filter order N, the maximum number of notches (notches at zero are treated differently
because of the theoretical symmetry with the negative frequency part of the spectrum) is given by
(number of notches in 0) + 2*(number of notches not in 0) <= N.

(L)WDF Design Toolbox for MATLAB User’s Guide 15

Designing in the Discrete-time Domain.

We assume that the reader is familiar with the theory on the z-transform and knows about the
methods to translate a continuous-time function into a discrete-time function.

In this Toolbox, we will only use the bilinear transformation, mainly because of the fact that the
theory of Wave Digital Filters is also based on this method.

2 z −1
The transformation uses the substitution s= ⋅
T z +1

In here, we will always set T = 2 , the result of which is that the continuous-time Normalized
Frequency value 1 always transforms into the discrete-time value 0.25 (i.e. a frequency value equal to
¼ of the Sample-frequency) and vice versa. The complete relationship between the continuous-time
frequency scale (from 0 to ∞ ) and the discrete-time relative frequency scale (0 to 0.5) is given by an
tan −1 -function, as shown in Figure 14.
The discrete-time transfer function is denoted as H (z ) .

discrete-time plot
0.5 0.5
Frequency / Sample Frequency

0.4 0.4

0.3 0.3
bilinear relation
0.2 0.2 θ = atan(F)

0.1 0.1

0 0
0 -20 -40 -60 -80 0 1 2 3 4 5
Magnitude in dB

0
Magnitude in dB

Figure 14. Domain conversion based on -20 continuous-time plot

the bi-linear transformation, which is
-40
commonly indicated as ‘warping’.
-60

-80
0 1 2 3 4 5
Normalized Frequency

The functions fs2fz, fz2fs, Hs2Hz, Hz2Hs and plotHz are specially meant for the design of discrete
filters and for switching between the domains.

16 (L)WDF Design Toolbox for MATLAB User’s Guide

The actual design of a discrete-time filter is a two step process, i.e. the design of a continuous-time
filter followed by the bilinear transformation.
Since the relation between frequencies in the two domains is exactly known, it is possible to pass ‘pre-
warped’ discrete-time frequencies to the continuous-time design, that, after transformation will result
in the correct discrete-time frequencies.

>> Hs = Hs_Vlach(7,0.01, fz2fs(0.1), fz2fs([0.15 0.2 0.35]), 0,0);

>> Hz = Hs2Hz(Hs);
>> plotHz( Hz,1 );

Discrete-time Characteristics
10

-10

-20
0.01
-30

Magnitude in dB
0.005

-40
0

-0.005 -50

-0.01 -60

-0.015
0 0.02 0.04 0.06 0.08 -70
0.1 0.12

-80

-90

-100
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

Figure 15. A discrete-time domain Vlach low-pass filter with its cut-off frequency
and three notches specified in the discrete-time domain.

The complete transfer function H(z) of this example then turns out to be

H ( z) =
(
10 −2 ⋅ 0.1212 + 0.0463 z -1 + 0.1212 z -2 + 0.1498 z -3 + 0.1498 z -4 + 0.1212 z -5 + 0.0463 z -6 + 0.1212 z -7 )
1.0 − 4.8352 z -1 + 10.6717 z - 2 − 13.7583 z -3 + 11.1224 z - 4 − 5.6162 z -5 + 1.6363 z - 6 − 0.2119 z - 7

(L)WDF Design Toolbox for MATLAB User’s Guide 17

Just for fun: Comparing a Cauer with a FIR Filter.

It is interesting to compare toolbox designs with the familiar FIR filters.

Shown here is a 7th order Cauer filter (with a pass-band ripple of 0.1 dB, a stop-band ripple of 50 dB,
freqNormMode set to 0) next to a 128 taps FIR filter that has been designed using the Remez
algorithm from the Signal Processing Toolbox (the original function remez.m has been renamed to
firpm.m).

In MATLAB’s notation:

>> n = 127;
>> f = [ 0 0.3 0.34102 1 ];
>> m = [ 1 1 0 0 ];
>> b = firpm( n, f, m ); % Parks-McClellan optimal equiripple FIR filter design.

In here, vector f is adjusted such that the stop-band ripple of the FIR filter also shows peaks at
maximally −50 dB.

7th Order Cauer Filter versus 128 Taps FIR Filter

10

128 Taps FIR Filter

-10
7th Order Cauer Filter

-20
Magnitude in dB

-30

-40

-50

-60

-70

-80
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

Figure 16. Comparing the Magnitude Transfer functions of a FIR filter

and a 7th order discrete-time Cauer filter.

Note that the FIR filter needs 128 multiplications (or 64, depending on the implementation method)
while if the Cauer filter is implemented as a (Lattice) Wave Digital Filter, it only needs 7 or 8
multiplications.

18 (L)WDF Design Toolbox for MATLAB User’s Guide

Designing Wave Digital Filters (WDFs).

A Wave Digital Filter is essentially a translation of a ladder structure into the discrete-time domain.
See [Fet86] and all other literature on WDFs to learn about all the benefits of this kind of filters.
Unfortunately, there exists very little software to design WDFs. With this Toolbox it is possible to
translate all the ladder structures that can be designed with the Toolbox’s functions into WDFs.

For the translation, the circuits have to be described using the so-called wave variables, A and B,
instead of the common voltages and currents. This way of describing circuits has originally been
developed for microwave and high-frequency circuitry.
According to the theory, components like inductances and capacitors translate into Delay Elements,
while the ‘wiring’ needed to connect these components translates into ‘adaptors’. These adaptors
contain computational elements to perform additions, sign-inversions (or subtractions) and one or two
multiplications with constant coefficients. The ladder circuit determines the structure of the WDF,
while the values of the lumped elements in the ladder determine the multiplication coefficients.

An overview of the adaptors supported by this Toolbox are given in Figures 18 and 19, with an
additional explanation of the (commonly used in Signal Processing) symbols in Figure 17.

x 1 [n ] add: y [n] = x1 [n] + x2 [n]

(constant) multiplier: y [n] = k ∗ x [n]

+
x2 [n] y [n] x [n ] y [n]

k
ALU functions
x 1 [n ]

x2 [n]
-+ y [n] x [n ]
T
y [n ] = x [n − 1 ]

subtract: y [n] = x1 [n] − x2 [n]

one clock cycle delay

x [n] y [n ] = − x [n − 1 ]
-T

Figure 17. Explanation of the basic elements.

(L)WDF Design Toolbox for MATLAB User’s Guide 19

 B2 A2 A2

A1 B3
A1 2 B3
1 3

B1 A3
B1 A3

B2
Figure 18a. Symbol and Block scheme for a ‘matched 3 port parallel adaptor’

A2
B2 A2
A1 −1 −1 A3

A1 2 B3
1 3

B1 A3
B1 B3

B2

Figure 18b. Symbol and Block scheme for a ‘3 port parallel adaptor’

A2
B2 A2

A1 B3
A1 2 B3
1 3

B1 A3 B1 A3

B2

Figure 18c. Symbol and Block scheme for a ‘matched 3 port serial adaptor’

A2
B2 A2
A1 A3

A1 2 B3
1 3

B1 A3
B1 B3

B2

Figure 18d. Symbol and Block scheme for a ‘3 port serial adaptor’

20 (L)WDF Design Toolbox for MATLAB User’s Guide

 a)
A1 B2
1 2

B1 A2

b) c)
A1 -1 A2 A1 -1 A2

-α1
α
α = 1 − α1

B1 B2
B1
B2

Figure 19. Symbol a), and two functionally equivalent realization schemes b) and c) for a
‘2 port adaptor’

Given a structure xLadder that has been determined before, the syntax to obtain a WDF is simply
>> WDF = ladder2WDF(xLadder)

Here WDF will also be a structure, containing the fields

WDF.wdaStruct
WDF.wdaNo
WDF.mulFacs

WDF.wdaStruct describes the WDF block diagram, where the block diagram is represented with 2
strings, one describing the adaptors in the signal path (bottom row), the second one (top row)
describing the elements or adaptors connected to the serial or parallel ports of the first mentioned
adaptors.
So, the bottom row can only consist of the following codes
's' - for a reflection free 3-port serial adaptor,
'p' - a reflection free 3-port parallel adaptor,
'S' - a 3-port serial adaptor with two coefficients,
'P' - a 3-port parallel adaptor with two coefficients,
'm' - an output inverter or scaling factor, if needed.
Some rules have been formulated for connecting the adaptors to each other:
For all these adaptors, port 1 is the input, port 3 the (reflection free) output, and port 2 the interface
to the top row elements.
Each element in the top row string is connected to port 2 of the adaptor in the same position in the
bottom row string. Possible codes are:
'+' - a single delay element (translation of a capacitance),
'-' - a delay element in series with an inverter (inductance),
's' - a reflection free serial adaptor (series LC resonator),
'p' - a reflection free parallel adaptor (parallel LC resonator),
'x' - for an empty slot.

(L)WDF Design Toolbox for MATLAB User’s Guide 21

With the 's' and 'p' adaptors, port 1 is connected to a single delay element (translation of the
capacitance), port 2 to a delay element in series with an inverter (the inductance), while the reflection
free port 3 is connected to port 2 of the corresponding bottom row adaptor.

WDF.wdaNo defines the numbering of the individual adaptors,

WDF.mulFacs lists the multiplication coefficients of the adaptors, starting from adaptor one. The very
last adaptor in a asymmetric or the middle adaptor in a symmetric structure, which is not reflection
free, needs two coefficients, while, if the bottom row string ends with an 'm', the last value will be the
scaling coefficient.

The next commands create a WDF from a 3rd order Butterworth low-pass:

>> nlpLadder = nlp_ladder('butter',3,'z');

RS L1 L3

Rs 1.00000 Ohm
L01 1.00000 H in series arm a) C2 RL
C02 2.00000 F in shunt arm
L03 1.00000 H in series arm
RL 1.00000 Ohm

>> lpLadder = nladder2lp(nlpLadder,fz2fs(0.15));

>> WDF = ladder2WDF(lpLadder)
wdfType not specified : '3p' assumed ...

Adaptor 1: 3p serial, p3 matched : alpha1 = 0.33754

Adaptor 2: 3p parallel, p3 matched : alpha1 = 0.07918
Adaptor 3: 3p serial : alpha1 = 0.14675
alpha3 = 0.62555

Adaptor port1 port2 port3

-----------------------------------------
1 A1/B1 -T_L01 3pA(2).p1
2 3pA(1).p3 T_C02 3pA(3).p1
3 3pA(2).p3 -T_L03 A3/B3
b)

WDF = -T T -T
wdaStruct: [2x3 char]
wdaNo: [2x3 double] 2 2 2
Ain Bfwd
mulFacs: [4x1 double]
1 3 1 3 1 3

Brev A3 = 0
A1 .α 1 A2 .α 1 A3 .α 1
A3 .α 3

Figure 20a and b. 3rd order Butterworth low-pass realizations as a ladder circuit
and as a Wave Digital Filter.

22 (L)WDF Design Toolbox for MATLAB User’s Guide

 WDFs actually show two outputs, denoted here as Bfwd and Brev. (see Figure 21a).
Bfwd is the ‘wanted’ or ‘forward’ output (in terms of scattering matrix notation also known as S 21 ),
while Brev is the ‘reflected’ or ‘reverse’ output ( S11 ), which shows a complementary characteristic
2 2
B fwd B 2 2
w.r.t. Bfwd, viz. + rev = 1 or B fwd ( jω ) + Brev ( jω ) = 1.
Ain Ain

If the filter is specified to be a low-pass, then the reflected output will show a high-pass behavior.
In many designs, however, this reflected output is hardly usable, showing particularly bad stop-band
behavior. Butterworth and bireciprocal Cauer designs (see later) can make good use of this feature.

Just like the passive ladder circuits can show resonance behavior with the result that on some of the
nodes voltages or currents can be larger than the input values, WDFs can internally also be more
sensitive for certain frequencies than initially expected. These effects should be taken into account
when implementing the filter in hardware with e.g. fixed-point arithmetic support.
This phenomenon has been made visible if Figure 21b) where the complete frequency responses for all
B-outputs of the 3rd order Butterworth WDF of Figure 20b have been plotted (Note: see the Reference
Guide which syntax for ladder2WDF to use for obtaining the impulse reponses for all B-outputs).
In the plots that are returned by ladder2WDF (or showWDF) the maximum levels of these responses
will be shown in the lower subplot (again Figure 21a).
When such large signal values can really
occur within a particular environment,
the WDF’s internal buswidth should be Reconstructed from the adapter (all three-ports) structure ...
10
adapted to prevent overflow errors.
Be warned that the other nodes inside the 0

adaptors are not monitored here. forward Bfw d

-10
reverse Brev
-20
Magnitude in dB

-30
a)
-40

-50
Figure 21a and b. Output information for
the WDF of Figure 20b. -60

-70
b)
-80
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency Response of All B-outputs
Frequency / Sample Frequency
4
B1rev
3.5 B2-A1
Maximum Signal Level in Frequency range

B3-A1
3
3 B1-A2
B2-A2
2.5
B3-A2
2.5
B1-A3
Signal Level

2
B2-A3
2 B3fw d
1.5

1.5 1

1 0.5

0
0.5 1 2 3
Adapter Number ( resp. B1,B2[,B3] )
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
frequency / Sample Frequency

(L)WDF Design Toolbox for MATLAB User’s Guide 23

For symmetric ladder circuits, like that in Figure 20a, it is also possible to translate them into
symmetric WDF structures. If also the component values are mirrored, this will result in mirrored
coefficients too, which can benefit the accuracy of the structure.

The preference for a specific structure can be made available to the ladder2WDF function:
>> ladder2WDF(lpLadder,'3p_sym');

Adaptor 1: 3p serial, p3 matched : alpha1 = 0.33754

Adaptor 2: 3p parallel : alpha1 = 0.14675
alpha3 = 0.14675
Adaptor 3: 3p serial, p3 matched : alpha1 = 0.33754

-T T -T

Ain 2 2 2 Bfwd
1 3 1 3 1 3

Brev A3 = 0
A1 .α 1 A2 .α 1 A3 .α 1
A2 .α 3

Figure 22. Symmetrical version of the WDF of Figure 20b.

The next structure is s translation of the band-pass ladder of Figure 10, again using 3-port adaptors.
>> ladder2WDF(bpLadder,'3p');

The parallel L-C resonators result in additional adaptors in the top row with two delay elements, one
for the L and one for the C connected to port 2 and port 1 respectively.

T T T T
A1 .α 1 A3 .α 1 A5 .α 1 A7 .α 1
1 1 1 1

2 -T 2 -T 2 -T 2 -T
3 3 3 3

−1 −1 −1 −1

Ain 2 2 2 2 Bfwd
1 3 1 3 1 3 1 3

Brev A3 = 0
−1 A2 .α 1 A4 .α 1 A6 .α 1 A8 .α 1
A8 .α 3

Figure 23. WDF version of the band-pass ladder of Figure 10.

24 (L)WDF Design Toolbox for MATLAB User’s Guide

Instead of using 3-port adaptors for implementing resonators, serial or parallel, it is also possible to
use translations with 2-port adaptors. This is shown in Figure 24 that will result in exactly the same
transfer functions as the structure from Figure 23, although with different intermediate maximum
signal levels.

The use of 2-port adaptors can be forced with:

>> ladder2WDF(bpLadder,'2p');

-T -T -T -T
A1 .α 1 A3 .α 1 A5 .α 1 A7 .α 1
2 2 2 2

1 1 1 1
a)

T T T T

Ain 2 2 2 2 Bfwd
1 3 1 3 1 3 1 3

Brev A3 = 0
−1 A2 .α 1 A4 .α 1 A6 .α 1 A8 .α 1
A8 .α 3

b)
Maximum Signal Level in Frequency range

2.5

1.5

0.5

0
1 2 3 4 5 6 7 8
Adapter Number ( resp. B1,B2[,B3] )

Figure 24. WDF version of the band-pass ladder of Figure 10 with 2-port
adaptors in the top row a), and b) the maximum transfer gains at
each B-output of every adaptor.

(L)WDF Design Toolbox for MATLAB User’s Guide 25

Lattice Wave Digital Filters (LWDFs).

Lattice WDFs are constructed with two parallel operating all-pass functions (made up with 2-ports),
i.e. transfer functions showing a gain factor of 1 for all frequencies but differing in their phase
responses. By adding the two outputs we get an overall magnitude transfer function with maximum
output levels of 2 at those frequencies where the output signals are in phase, and notches when the
signals are exactly in anti-phase. Subtracting the signals has the complementary effect.
Now the trick is to create phase functions such that the overall magnitude transfer functions shows
the desired characteristic.
It can be shown [Gazsi85] that odd order LWDFs can be used to create all odd order low-pass and
high-pass transfer functions that have been describes before, while even order LWDFs result in band-
pass and band-stop functions.

In the following example, we will design a high-pass LWDF:

>> [Hs,wp] = Hs_Vlach(5,0.1, fz2fs(0.2), fz2fs([0.29 0.35]), 0, 1);

>> Hshp = nlp2hp(Hs,1);
>> LWDF = Hs2LWDF(Hshp);
LWDF: ODD filter order, so LOW/HIGH pass filter assumed

Structure appears to be a lowpass/highpass filter.

T
Top row all-pass sections :
1st degree section : y01 = -0.43626
γ3
2nd degree section : y02 = -0.78933
y03 = -0.33269
Bottom row all-pass sections : T T

2nd degree section : y04 = -0.40012

y05 = -0.53078 γ1 γ2
−1
1/2
>> Hz2 = LWDF2Hz(LWDF); + highpass
>> plotHz(Hz2,1); Input
Outputs

+ lowpass
a) 1/2

γ4

b)
T
Discrete-time Characteristics
10
γ5
0

-10
T
-20

-30
Magnitude in dB

-40

-50 Figure 25a and b. Lattice Wave Digital

high-pass filter, structure and transfer
-60
characteristics.
-70

-80

-90

-100
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

26 (L)WDF Design Toolbox for MATLAB User’s Guide

Notice that the specific frequency points were specified in the discrete-time domain directly (viz. at
0.2, 0.29 and 0.35 of the Sample Frequency), and that these point are mirrored by the low-pass to
high-pass transformation to respectively 0.5 − 0.2 = 0.3 , 0.5 − 0.29 = 0.21 and 0.5 − 0.35 = 0.15 of the
Sample Frequency.
Also, the peaks in the pass-band of the normalized low-pass function are returned in the variable wp
by Hs_Vlach, so the peaks in the pass-band of the high-pass characteristic can be calculated (zoom in
the real MATLAB plot to see it):
>> 0.5 - fs2fz(wp)
ans =
0.5000
0.3774
0.3217
Since low-pass and high-pass outputs are complementary functions, these are exactly the relative
frequencies where the low-pass characteristic shows its notches.

In Figure 26, the phase characteristics of the individual all-pass sections (top and bottom) are shown,
together with the resulting absolute phase difference. All characteristic points mentioned above are
clearly recognizable. Also very obvious is the relative weakness of this type of structures, viz. the very
small phase differences in the stop-band. In case the coefficients are not realized accurately enough,
e.g. due to quantizing effects, the stop-band cannot be guaranteed to be exactly as calculated.

0.5 -150

0.4
-155
0.3
-160
0.2
Phase angle in degrees

Phase angle in degrees

-165
0.1

0 -170

-0.1
-175
-0.2
top section -180
-0.3 bottom section
| phase difference |
-185
-0.4

-0.5 -190
0.1 0.15 0.2 0.25 0.3 0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.4
Frequency / Sample Frequency Frequency / Sample Frequency

100

-100
Phase angle in degrees

-200

top section
-300
bottom section
| phase difference |
-400

-500

-600
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

Figure 26. Phase characteristics describing the LWDF of Figure 25.

(L)WDF Design Toolbox for MATLAB User’s Guide 27

As said before, even order LWDFs are used to realize band-pass (band-stop) transfer functions:

>> Hs = nlpf('invcheby',5,45,1);
>> Hsbp = nlp2bp(Hs,fz2fs(0.15),fz2fs(0.1));
>> LWDF = Hs2LWDF(Hsbp); Hz2 = LWDF2Hz(LWDF); plotHz(Hz2,1,2);

T T T

γ2 γ4 γ6

a) T T T

γ1 γ3 γ5
−1
1/2
+ bandpass
Input
Outputs

+ bandstop
1/2

γ7 γ9

T T

γ8 γ10

T T

b)
Discrete-time Characteristics
10

-10

-20

-30
Magnitude in dB

-40

-50

-60

-70

-80

-90

-100
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

Figure 27a and b. Lattice Wave Digital band-pass filter obtained with a normalized
low-pass to band-pass transformation, structure and magnitude
transfer characteristics.

28 (L)WDF Design Toolbox for MATLAB User’s Guide

 Bireciprocal LWDF designs.

A bireciprocal filter has the property that it’s outputs, low-pass and high-pass, are exactly their mirror
images, being reflected around ¼ of the sample frequency. The LWDF structure in such a case will be
less complex then the structures that were shown before.
Not all filter types can be exactly mirrored, since there should be a distinct relationship between the
behavior of the transfer function in the pass-band and that in the stop-band. In fact, only Butterworth
and Cauer approximations (type ‘A’) will be usable. Butterworth low-pass filters with a cut-off
frequency of ¼ of the sample frequency are inherently bireciprocal, for Cauer filters only one of the
ripple values may be chosen while this fixes the other one.

The following example (re)calculates the "Cauer parameter (elliptical) bireciprocal low-pass filter" that
has been described in detail by L. Gaszi as Example 5 in [Gaz85].
The parameters copied from Gazsi, are the filter order N=19, and the stop-band loss value of 76.89 dB.

>> Hs = Hs_cauer_birec(19, 76.89);

>> LWDF = Hs2LWDF( Hs(1) );
>> plotHz( LWDF2Hz(LWDF),1,2 );

Discrete-time Characteristics
10

-10

-20 low-pass
high-pass
-30
Magnitude in dB

-40

-50

-60

-70

-80

Figure 28. Transfer characteristic and -90

structure of the 19th order bireciprocal Cauer -100

LWDF. 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

2T 2T 2T 2T

γ1 γ2 γ3 γ4
−1
1/2
T
+ highpass
Input
Outputs

+ lowpass
1/2

γ5 γ6 γ7 γ8 γ9
Figure 27a and b. Lattice Wave Digital band-pass filter obtained with a normalized
low-pass to band-pass transformation, structure and magnitude
2T 2T transfer2Tcharacteristics.
2T 2T

(L)WDF Design Toolbox for MATLAB User’s Guide 29

The coefficients are calculated to be:

LWDF: ODD filter order, so LOW/HIGH pass filter assumed

Structure appears to be a bireciprocal lowpass/highpass filter.

Top row all-pass sections :
single delay
single section, 2 delays : y01 = -0.22671
single section, 2 delays : y02 = -0.60341
single section, 2 delays : y03 = -0.84001
single section, 2 delays : y04 = -0.95112
Bottom row all-pass sections :
single section, 2 delays : y05 = -0.06417
single section, 2 delays : y06 = -0.42397
single section, 2 delays : y07 = -0.74221
single section, 2 delays : y08 = -0.90604
single section, 2 delays : y09 = -0.98481

Compare the results obtained here with Gazsi's Figures 14a and 14c. Gazsi’s coefficients are copied
below. Note that there is a difference in the numbering of the adaptors and coefficients.

Toolbox
Gazsi
Table 2. notation
y01 γ 3 = −0.226 119
y02 γ 7 = −0.602 422
y03 γ 11 = −0.839 323
y04 γ 15 = −0.950 847
y05 γ 1 = −0.063 978
y06 γ 5 = −0.423 068
y07 γ 9 = −0.741 327
y08 γ 13 = −0.905 567
y09 γ 17 = −0.984 721

Discrete-time Low-Pass Transfer Characteristic

-20
Magnitude in dB

-40

-60

-80

-100
0.0 4.0 8.0 12.0 16.0 20.0 24.0 28.0 32.0
Frequency in kHz

Figure 29. Magnitude transfer characteristic of the low-pass part of the 19th order
bireciprocal Cauer LWDF with the same absolute horizontal frequency
scale as used by Gazsi.

30 (L)WDF Design Toolbox for MATLAB User’s Guide

Vlach Band-pass LWDF Designs.

Of course it is also possible to create band-pass LWDFs based on the Vlach approximation:

>> Hs = Hs_bpVlach(6,1, fz2fs([0.13 0.17]), fz2fs([ 0 0.1 0.2]),0,1);

>> LWDF = Hs2LWDF(Hs);
LWDF: EVEN filter order, so BANDpass/stop filter assumed

Structure appears to be a bandpass/bandstop filter.

Top row all-pass sections :
T
2nd degree section : y01 = -0.88515
y02 = 0.58890
γ2
Bottom row all-pass sections :
2nd degree section : y03 = -0.94758
y04 = 0.49278 T

2nd degree section : y05 = -0.95050

y06 = 0.67555 γ1
−1
1/2
+ bandpass
Input
Outputs

+ bandstop
1/2

γ3 γ5

T T
Discrete-time Characteristics
10

0
γ4 γ6

-10

-20 T T

-30
Magnitude in dB

-40

-50

-60 Figure 28. Structure and Magnitude

-70
Transfer characteristics for a 6th order Vlach
band-pass LWDF.
-80

-90

-100
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Frequency / Sample Frequency

There are two possibilities to obtain H (z ) for the LWDF. The first is by translating the originating
H (s) with Hs2Hz, but this will only return the forward transfer function. In the LWDF examples,
therefore LWDF2HZ has been used, which returns a MATLAB structure with both the forward and
reverse function that are reconstructed from the LWDF parameters.

>> Hz1 = Hs2Hz(Hs); % only a single Hz

>> Hz2 = LWDF2Hz(LWDF); % Hz(1) and Hz(2), obtained from the LWDF data

(L)WDF Design Toolbox for MATLAB User’s Guide 31

Transmission Line Filter Designs.

A lumped element ladder circuit has to be transformed first, before it can be realized in a microwave
technology. In such a high frequencies environment, it is no longer possible to work with floating
inductances or capacitors as occur in series arms. A transformation is usually accomplished by shifting
in Unit Elements from the input and/or the output into the circuit making use of Kuroda identities to
transform a “Unit Element − series arm element” into a “Unit Element − shunt arm element”
combination.
With the Vlach-approximation functions from the toolbox, the insertion of Unit Elements is
straightforward and done with the aid of a binary vector. A Unit Element can be inserted to the left of
every inductor or capacitor and changes an impedance function to its right describing the rest of the
circuit into a reactance function and vice versa. The examples below illustrate the process. A zero
means the absence, a one the presence of a Unit Element before the particular inductance or
capacitance.
A strong feature of this toolbox is that the Unit Elements actually contribute to the filter’s transfer
characteristic: each Unit Element increases the order of the approximation of the pass-band and adds
a small amount of attenuation in the stop-band (up to 7.7 dB at high frequencies).
RS L2 L4

without Unit Elements: C1 C3 C5 RL

>> nlp_ladder('vlach',5,1,[],[],[]);

RS L4 L6

One Unit Element, before the first inductor:

C1 UE2 C3 C5 RL
>> nlp_ladder('vlach',5,1,[],[],[ 0 1 ]);

RS L6

Two Unit Elements:

C1 UE2 C3 UE4 C5 C7 RL
>> nlp_ladder('vlach',5,1,[],[],[ 0 1 1 0 0 0]);

… or, to immediately obtain the complete filter in which only shunt capacitors are present:

>> nlp_ladder('vlach',5,1,[],[],[ 0 1 1 1 1 0]);

RS Figure 29.
Chebyshev low-pass filter with
4 additional Unit Elements..
C1 UE2 C3 UE4 C5 UE6 C7 UE8 C9 RL

32 (L)WDF Design Toolbox for MATLAB User’s Guide

with the following element values:

Rs 1.00000 Ohm
C01 2.17298 F in shunt arm
UE02 1.37838 Ohm
C03 3.07531 F in shunt arm
UE04 1.43977 Ohm
C05 3.14506 F in shunt arm Microwave Low-pass
UE06 1.43977 Ohm 10
C07 3.07531 F in shunt arm
UE08 1.37838 Ohm no Unit Elements
0
4 Unit Elements
C09 2.17298 F in shunt arm
RL 1.00000 Ohm -10

Magnitude in dB
-20

Continuous-time Characteristics -30

0.5
reference magnitude function
reconstructed from ladder topology
-40
1 3 9
0

-50
Magnitude in dB

-0.5 -60
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
frequency / Reference Frequency

-1
2 8

-1.5
0 0.2 0.4 0.6 0.8 1 1.2
Frequency

Figure 30. Magnitude Transfer functions of the 5th order low-pass Chebyshev(!) filter
without and including 4 Unit Elements. The horizontal scale has been
adapted to facilitate the comparison with the APLAC plots.

The plot of Figure 30 clearly shows the increased roll-off in the transition band and the increased
number of peaks and valleys in the pass-band (5th order with 4 UEs = 9, equivalent to 9th order).
The equations describing the transfer function can be obtained with
>> Hs4 = nlpf('vlach',5,1,[],4);
>> Hz4 = Hs2Hz(Hs4);
because of the similarities in the frequency behavior of discrete-time and microwave circuits.

The correctness of the computations above can be verified with the aid of a third party program like
the APLAC RF DESIGN TOOL. With APLAC it is possible to simulate transmission line filters.
The element values listed above have been entered in the APLAC input format in a file
'vlach_5_1_4ue.i' (see Figure 33), describing a transmission line filter with its cut-off frequency at
1 GHz. The resulting simulation output is given in Figure 31, and reveals a striking resemblance with
what the Toolbox predicts.
Figure 31 has been created with the latest release of APLAC at the time of writing this document, viz.
version 8.10 of the free downloadable Student Version (see http://www.aplac.com/ ).

(L)WDF Design Toolbox for MATLAB User’s Guide 33

 S-parameter Analyse
APLAC 8.10 Student version FOR NON-COMMERCIAL USE ONLY
10.00
|S21|
0.00
(dB)
-10.00

Figure 31. Results of the APLAC .i-file -20.00

listed in Figure 33 based on the values
-30.00
calculated for the example of Figure 29 with a
cut-off frequency of 1 GHz. -40.00
The plots use the same axes scaling as is used
in Figure 30. -50.00

-60.00
0.000 0.500 1.000 1.500 2.000
f/Hz
MagdB(S(2,1))

S-parameter Analyse
APLAC 8.10 Student version FOR NON-COMMERCIAL USE ONLY
0.50
|S21|
0.00
(dB)
-0.50

-1.00

-1.50
0.000 0.200 0.400 0.600 0.800 1.000 1.200
f/Hz
MagdB(S(2,1))

Although it was mentioned above that Unit Elements can be inserted to the left of the lumped
elements, this is not the complete story. Access UEs are padded to the right of the ladder, and if no
lumped elements should be present at all, a filter consisting of only UEs can be constructed.

RS L2

>> nlp_ladder('vlach',3,0.1,[],[],[ 0 0 0 1 1 ]); C1 C3 UE4 UE5 RL

>> nlp_ladder('vlach',0,1,[],[],[ 1 1 1 1 1 ]);

RS

UE1 UE2 UE3 UE4 UE5 RL

Figure 32. Examples of Unit Element insertions.

34 (L)WDF Design Toolbox for MATLAB User’s Guide

 $ Filename: vlach_5_1_4ue.i
$ see User's Guide, Figure ???
$ H.J. Lincklaen Arriens

Declare IVAR
+ C01 = 2.17298
+ RUE2 = 1.37838
+ C03 = 3.07531
+ RUE4 = 1.43977
+ C05 = 3.14506
+ RUE6 = 1.43977
+ C07 = 3.07531
+ RUE8 = 1.37838
+ C09 = 2.17298

$----------------------------------------------------
$ Transmission lines
$ type - name - node connections - electrical length - ref freq in GHz – impedance
$----------------------------------------------------
Tline line1 1 0 2 0 EL_LENGTH 90 FC 2 Z 1/C01
Tline line2 1 0 3 0 EL_LENGTH 90 FC 2 Z RUE2
Tline line3 3 0 4 0 EL_LENGTH 90 FC 2 Z 1/C03
Tline line4 3 0 5 0 EL_LENGTH 90 FC 2 Z RUE4
Tline line5 5 0 6 0 EL_LENGTH 90 FC 2 Z 1/C05
Tline line6 5 0 7 0 EL_LENGTH 90 FC 2 Z RUE6
Tline line7 7 0 8 0 EL_LENGTH 90 FC 2 Z 1/C07
Tline line8 7 0 9 0 EL_LENGTH 90 FC 2 Z RUE8
Tline line9 9 0 10 0 EL_LENGTH 90 FC 2 Z 1/C09

$----------------------------------------------------
$ 2-port definition
$----------------------------------------------------
DefNPort ideal_splane_filt 2 1 0 1.00000 9 0 1.00000

$----------------------------------------------------
$ Output commands
$----------------------------------------------------
Sweep "S-parameter Analyse"
+ LOOP 2000 FREQ LIN 0.0001 2
+ WINDOW 0
+ Y "|S21|" "(dB)" -60 10.0 GRID
+ WINDOW 1
+ X "f" "Hz" 0.0001 1.2
+ Y "|S21|" "(dB)" -1.5 0.5 GRID

Show
+ WINDOW 0 Y MagdB(S(2,1))
+ WINDOW 1 Y MagdB(S(2,1))
EndSweep

Figure 33. Listing of a transmission line implementation of the low-pass Chebyshev(!)

filter with 4 Unit Elements as shown in Figure 29.

(L)WDF Design Toolbox for MATLAB User’s Guide 35

Although of no practical use, the strength of this toolbox function is also demonstrated with the
following example in which just one Unit Element is inserted at an arbitrary location in the ladder
circuit. The MATLAB result is again verified in APLAC (only a part of the .i-file is shown)

>> nlp_ladder('vlach',5,1,[],[],[ 0 0 0 1 0 0]);

RS L2 L6
Declare IVAR
+ C01 = 2.14956
+ L02 = 1.10085 a) C1 C3 UE4 C5 RL
+ C03 = 3.04808
+ RUE4 = 1.41224
+ C05 = 2.92797
+ L06 = 0.80819

$----------------------------------------------------
$ Transmission lines
$----------------------------------------------------
Tline line1 1 0 2 0 EL_LENGTH 90 FC 2 Z 1/C01
Tline line2 1 4 3 3 EL_LENGTH 90 FC 2 Z L02
Tline line3 4 0 5 0 EL_LENGTH 90 FC 2 Z 1/C03
Tline line4 4 0 6 0 EL_LENGTH 90 FC 2 Z RUE4
Tline line5 6 0 7 0 EL_LENGTH 90 FC 2 Z 1/C05
Tline line6 6 9 8 8 EL_LENGTH 90 FC 2 Z L06

$----------------------------------------------------
$ 2-port definition
$----------------------------------------------------
DefNPort ideal_splane_filt 2 1 0 1.00000 9 0 0.37598

S-parameter Analyse
APLAC 8.10 Student version FOR NON-COMMERCIAL USE ONLY
10.00
|S21|
0.00
(dB)
Figure 34. APLAC plots showing the -10.00
simulation of the ladder circuit of a) if
-20.00
written as a transmission line filter in b).
-30.00

-40.00

-50.00

-60.00
0.000 0.500 1.000 1.500 2.000
b) f/Hz
MagdB(S(2,1))

S-parameter Analyse
APLAC 8.10 Student version FOR NON-COMMERCIAL USE ONLY
0.50
|S21|
0.00
(dB)
-0.50

-1.00

-1.50
0.000 0.200 0.400 0.600 0.800 1.000 1.200
f/Hz
MagdB(S(2,1))

36 (L)WDF Design Toolbox for MATLAB User’s Guide

GUI’s that cover all the above.

For an easy access of the majority of the Toolbox functions, the wdf_GUI (Graphical User Interface),
shown below, and the more specific bpVlach_GUI (for LWDFs only) have been developed.
Given the examples and explanations in the previous sections, there shouldn’t be any problems with
the possibilities offered by and the parameters needed by the GUI’s.
Parameter fields that are not needed for a particular design are grayed-out while those that are
needed pop-up and need to be filled in. If a parameter may contain more than one value, like the
‘Stopband Zero Frequencies’ or the ‘Unit Element Positions’, just separate the individual values with
a space (or spaces).
Output values are written to the MATLAB console window and can be used for further processing.

Specific design properties, e.g. whether a bireciprocal LWDF structure can be used, are automatically
detected.

Figure 35. Screen shot of the wdf_GUI.

(L)WDF Design Toolbox for MATLAB User’s Guide 37

 Figure 36. Screen shot of the bpVlach_GUI for rapidly creating
LWDFs based on the Vlach approximation.

38 (L)WDF Design Toolbox for MATLAB User’s Guide

Interface to Scheduling Toolbox.

This toolbox contains two functions to translate the WDF and LWDF structures into text files in
which all operations are listed in a way that it can be read by the Scheduling Toolbox,
viz. WDF2cir and LWDF2cir.
Both function also return the component values in a structure that also can be recognized by the
Scheduling Toolbox.

When WDFs using 2-port adaptors are to be scheduled, there is no guarantee that when the delay
element between a 3-port and its corresponding 2-port adaptor is connected from port B2 of
Adaptor(n) to port A1 of Adaptor(n−1) – as has been done in all WDF-figures up to now– will result in
a faster circuit then when it should be connected from port B1 of Adaptor(n−1) to port A2 of
Adaptor(n). Therefore, the position of the delay elements can be specified in WDF2cir (and in showWDF,
see Figure 37).

NlpLadder = nlp_ladder('cauer',5,0.1,45,'a',1,'z');
WDF = ladder2WDF(NlpLadder,'2p_sym');
WDF2cir(WDF,'r'); % output to screen
showWDF(WDF,'r',2); % plot in figure 2

T T
A2 .α 1 A5 .α 1
2 2

1 1

-T T -T T -T
m3
Ain 2 2 2 2 2 Bfwd
1 3 1 3 1 3 1 3 1 3

Brev A3 = 0
A1 .α 1 A3 .α 1 A4 .α 1 AT6 .α 1 A7 .α 1 T
A4 .α 3 A2 .α 1 A5 .α 1
2 2

1 1

-T T -T T -T
m3
Ain 2 2 2 2 2 Bfwd
1 3 1 3 1 3 1 3 1 3

Brev A3 = 0
A1 .α 1 A3 .α 1 A4 .α 1 A6 .α 1 A7 .α 1
A4 .α 3

Figure 37. The location of the delay elements connecting to 2-ports can be
specified (default is the left-most connecting arm).

(L)WDF Design Toolbox for MATLAB User’s Guide 39

In the LWDF structure, pipeline registers can be inserted easily with LWDF_insRegs before calling
LWDF2cir:
supposed that the LWDF structure LWDF known here is the one from Figure 27a, then
>> LWDFp = LWDF_insRegs(LWDF,[ 1 1 ]);
>> showLWDF(LWDFp,'L')
will result in a structure like Figure 38 which allows more parallelism when scheduling.

T T T

γ2 γ4 γ6

T T T

γ1 γ3 γ5
−1
1/2
+ bandpass
RegT1 RegT2
Input
Outputs
RegB1 RegB2
+ bandstop
1/2

γ7 γ9

T T

γ8 γ10

T T

Figure 38. Pipelined version of the LWDF of Figure 27a.

Here it would also have been possible to position the delay elements between the adaptors in the right-
most arm instead of in the left arm by using showLWDF(LWDFp,’r’).

40 (L)WDF Design Toolbox for MATLAB User’s Guide

Epilog.

It is by no means expected that this Toolbox should ever cover all kinds of design wishes.
When assembling this documentation −some time after the functions have actually been written− lots
of omissions, improvements and extensions came to mind, but time goes on and other projects are
waiting ….
Since – except for the p-code files – the functions are normal MATLAB code, everyone can tailor the
code to his/hers own needs.
Intermediate results may be the starting point for implementations in specific techniques or
technologies.
An automatic generation of DSP code, whether or not with intervention of the Scheduling Toolbox
(which generates VHDL), is surely possible.
Also, the setup of the (MATLAB) WDF structure leaves enough possibilities to diverge from the simple
ladder structures shown here.

The Toolbox has been tested on Windows XP PC’s, and also with the R14 Student version on a Linux
system (KDE on Slackware). As expected, the GUIs on Linux will need some visual adjustments.

Although the majority of the functions initially were written in MATLAB Release 13, some
modifications and extensions were made using Release 14 features. Rewriting these parts of the code
(particularly used for nested functions) to be executable with Release 13 again, including the GUIs,
has not been considered and is left to the confirmed, for whatever reason, R13 user.

(L)WDF Design Toolbox for MATLAB User’s Guide 41

References.

[APLxx] APLAC, a Powerfull and versatile analog and RF Simulator

http://www.aplac.com/

[Abr72] "The Process of the Arithmetic-Geometric Mean."

from
Handbook of Mathematical Functions
with Formulas, Graphs, and Mathematical Tables
edited by M. Abramowitz & I.A. Stegun
Dover Publications, Inc. (1964-72)
9th edition, §17.6, pp. 571 and 598-599, 1972

[Ant79] A. Antoniou
Digital Filters: Analysis and Design
McGraw-Hill Book Company (1979)
ISBN 0-07-002117-1

[Bel68] V. Belevitch
Classical Network Theory
Holden-Day, San Francisco (1968)

[Bin64] J.A.C. Bingham

A New Method of Solving the Accuracy Problem in Filter Design
IEEE Trans. on Circuit Theory, CT-11,
No. 3, pp. 327-341 (September 1964)

[Chr66] E. Christian & E. Eisenmann

Filter Design Tables and Graphs
John Wiley & Sons, Inc. (1966)

[Fet86] A. Fettweis
Wave Digital Filters: Theory and Practice (Invited Paper)
Proc. of the IEEE, Vol. 74, No. 2, Februari 1986

[Gaz85] L. Gazsi
Explicit Formulas for Lattice Wave Digital Filters
IEEE Trans. on CAS, Vol. 32, pp. 68-88, Jan 1985

[Hel90] J. Helszajn
Synthesis of Lumped Element, Distributed and Planar Filters
McGraw-Hill Book Company (1990)
ISBN 0-07-707166-2

[Law90] S. Lawson & A.Mirzai

Wave Digital Filters
Ellis Horwood Ltd. (1990)
ISBN 0-13-946997-4

42 (L)WDF Design Toolbox for MATLAB User’s Guide

[Nou79] R. Nouta
Studies in Wave Digtal Filter Theory and Design
PhD Thesis, Delft University of Technology (1979)

[Rho76] J.D. Rhodes

Theory of Electrical Filters
J. Wiley & Sons Ltd. (1976)
ISBN 0-471-71806-8

[Sha54] Ch.B. Sharpe

A General Tchebycheff Rational Function
Proc. IRE, Vol. 42, pp. 454-457 (1954)

[Skw65] J.K. Skwirzynski

Design Theory and Data for Electrical Filters
Van Nostrand Company Ltd. (1965)

[Su_96] K.L. Su
Analog Filters
Chapman & Hall (1996)
ISBN 0 412 63840 1

[Tel52] B.D.H. Tellegen

A general network theorem, with applications
Philips Research Reports, Vol. 7, 1952, pp. 259-269

[Vla69] J. Vlach
Computerized Approximation and Synthesis of Linear Networks
John Wiley & Sons, Inc. (1969)
SBN 471 90870 3

[Wan99] L. Wanhammer
DSP Integrated Circuits
Academic Press, (1999)
ISBN 0-12-734530-2

[Zve67] A.I. Zverev

Handbook of Filter Synthesis
John Wiley & Sons, Inc. (1967)

(L)WDF Design Toolbox for MATLAB User’s Guide 43

 This page intentionally left blank

44 (L)WDF Design Toolbox for MATLAB User’s Guide