Digital Signal Processing With Matlab Examples Volume 3 Model Based Actions and Sparse Representation 1st Edition Jose Maria Giron-Sierra (Auth.)
Digital Signal Processing With Matlab Examples Volume 3 Model Based Actions and Sparse Representation 1st Edition Jose Maria Giron-Sierra (Auth.)
Digital Signal Processing With Matlab Examples Volume 3 Model Based Actions and Sparse Representation 1st Edition Jose Maria Giron-Sierra (Auth.)
https://textbookfull.com/product/conceptual-digital-signal-
processing-with-matlab-keonwook-kim/
https://textbookfull.com/product/understanding-digital-signal-
processing-with-matlab-and-solutions-first-edition-edition-
poularikas/
https://textbookfull.com/product/digital-signal-processing-a-
primer-with-matlab-1st-edition-samir-i-abood-author/
https://textbookfull.com/product/real-time-digital-signal-
processing-from-matlab-to-c-with-the-tms320c6x-dsps-third-
edition-morrow/
Smartphone based real time digital signal processing
Third Edition Abhishek Sehgal
https://textbookfull.com/product/smartphone-based-real-time-
digital-signal-processing-third-edition-abhishek-sehgal/
https://textbookfull.com/product/digital-signal-processing-and-
spectral-analysis-for-scientists-concepts-and-applications-1st-
edition-silvia-maria-alessio-auth/
https://textbookfull.com/product/matlab-signal-processing-
toolbox-user-s-guide-the-mathworks/
https://textbookfull.com/product/digital-image-processing-a-
signal-processing-and-algorithmic-approach-1st-edition-d-
sundararajan-auth/
https://textbookfull.com/product/from-algebraic-structures-to-
tensors-digital-signal-and-image-processing-matrices-and-tensors-
in-signal-processing-set-1st-edition-gerard-favier-editor/
Signals and Communication Technology
Digital Signal
Processing with
Matlab Examples,
Volume 3
Model-Based Actions and Sparse
Representation
Signals and Communication Technology
More information about this series at http://www.springer.com/series/4748
Jose Maria Giron-Sierra
123
Jose Maria Giron-Sierra
Systems Engineering and Automatic Control
Universidad Complutense de Madrid
Madrid
Spain
MATLAB® is a registered trademark of The MathWorks, Inc., and is used with permission. The
MathWorks does not warrant the accuracy of the text or exercises in this book. This book’s use or
discussion of MATLAB software or related products does not constitute endorsement or sponsorship by
the MathWorks of a particular pedagogical approach or particular use of the MATLAB software.
This is the third book of a trilogy. As in the other books, a series of MATLAB
programs are embedded in the chapters for several purposes: to illustrate the
techniques, to provide implementation examples, to encourage for personal
exploration departing from a successful start.
The book has two parts, each having just one chapter. These chapters are long
and have a considerable number of bibliographic references.
When using a GPS on a car, sometimes it is not possible to keep contact with
satellites, like for instance inside tunnels. In this case, a model of the car motion—a
dynamic model—can be used for data substitution. The adequate combination of
measurements and models is the key idea of the Kalman filter, which is the central
topic of the first part of the book. This filter was formulated for linear conditions.
There are modifications for nonlinear conditions, like the extended Kalman filter, or
the unscented Kalman filter. A new idea is to use particle filters. These topics are
covered in the chapter under an important perspective: Bayesian filtering.
Compressed sensing has emerged as a promising idea. One of the intended
applications is networked devices or sensors, which are becoming a reality. This
topic is considered in the second part of the book. Some experiments that
demonstrate image denoising applications were included.
For easier reading of the book, the longer programs have been put in an
appendix. And a second appendix on optimization has been added to support some
contents of the last chapter.
The reader is invited to discover the profound interconnections and common-
alities that exist behind the variety of topics in this book. This common ground
would become surely the humus for the next signal processing future.
As said in the preface of the other books, our particular expertise on signal
processing has two main roots: research and teaching. I belong to the Faculty of
Physics, University Complutense of Madrid, Spain. During our experimental
research on autonomous vehicles, maritime drones, satellite control, etc., we
practiced main methods of digital signal processing, for the use of a variety of
sensors and for prediction of vehicle motions. From years ago, I teach Signal
Processing in a Master of Biomedical Physics, and a Master on New technologies.
v
vi Preface
The style of the programs included in the book is purposively simple enough.
The reader is invited to typeset the programs included in the book, for it would help
for catching coding details. Anyway, all programs are available from the book web
page: www.dacya.ucm.es/giron/SPBook3/Programs.
A lot of different materials have been used to erect this book: articles, blogs,
codes, experimentation. I tried to cite with adequate references all the pieces that
have been useful. If someone has been forgotten, please contact me. Most of the
references cited in the text are available from Internet. We have to express our
gratitude to the public information available in this way.
Please, send feedback and suggestions for further improvement and support.
Acknowledgments
Thanks to my University, my colleagues and my students. Since this and the other
book required a lot of time taken from nights, weekends and holidays, I have to
sincerely express my gratitude to my family.
vii
viii Contents
Figure 1.1 Keeping the car at a distance from the road border . . . . . . . . . 7
Figure 1.2 Prediction (P), measurement (M) and update (U) PDFs . . . . . . 9
Figure 1.3 Variation of K in function of σ y =σ x . . . . . . . . . . . . . . . . . . . . 10
Figure 1.4 The algorithm is a cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Figure 1.5 A two-tank system example . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Figure 1.6 System outputs (measurements) . . . . . . . . . . . . . . . . . . . . . . . . 19
Figure 1.7 System states, and states estimated by the Kalman filter . . . . . 19
Figure 1.8 Error evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Figure 1.9 Evolution of the Kalman gains . . . . . . . . . . . . . . . . . . . . . . . . 23
Figure 1.10 Evolution of the state covariance . . . . . . . . . . . . . . . . . . . . . . . 24
Figure 1.11 The prediction step, from left to right . . . . . . . . . . . . . . . . . . . 25
Figure 1.12 The measurement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Figure 1.13 Estimation of the next state . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Figure 1.14 Bayes net corresponding to Kalman filter . . . . . . . . . . . . . . . . 27
Figure 1.15 Satellite position under disturbances . . . . . . . . . . . . . . . . . . . . 33
Figure 1.16 Example of nonlinear function: arctan(). . . . . . . . . . . . . . . . . . 34
Figure 1.17 Original and propagated PDFs . . . . . . . . . . . . . . . . . . . . . . . . . 35
Figure 1.18 Propagation of a PDF through nonlinearity . . . . . . . . . . . . . . . 36
Figure 1.19 Propagated PDFs for sigma ¼ 0:7; 1; 2 . . . . . . . . . . . . . . . . . . 37
Figure 1.20 Propagation of a shifted PDF through nonlinearity . . . . . . . . . 38
Figure 1.21 Basic linear approximation using tangent . . . . . . . . . . . . . . . . 42
Figure 1.22 Falling body example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Figure 1.23 System states (cross marks) . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
Figure 1.24 Distance measurement and drag . . . . . . . . . . . . . . . . . . . . . . . 45
Figure 1.25 The three non-zero components of the o f=o x Jacobian . . . . . 47
Figure 1.26 Propagation of ellipsoids (state N=43 -> 44)s . . . . . . . . . . . . . 49
Figure 1.27 System states (cross marks) under noisy conditions. . . . . . . . . 51
Figure 1.28 Distance measurement. Drag . . . . . . . . . . . . . . . . . . . . . . . . . . 52
Figure 1.29 System states (cross marks), and states estimated
by the EKF (continuous) . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 54
xi
xii List of Figures
Figure 2.1 The line L and, a the ball B1=2 , b the ball B1 ,
c the ball B2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Figure 2.2 An example of the solution paths obtained with LARS . . . . . . 159
Figure 2.3 Solution paths using LARS for diabetes set . . . . . . . . . . . . . . 162
Figure 2.4 Solution paths using LASSO for diabetes set . . . . . . . . . . . . . 162
Figure 2.5 Soft-thresholding operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
Figure 2.6 Application of ISTA for a sparse signal recovery
example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Figure 2.7 Evolution of objective function along ISTA iterations . . . . . . . 172
Figure 2.8 A BP sparsest solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Figure 2.9 Evolution of objective function k xk1 along ADMM
iterations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Figure 2.10 Sparse solution obtained with OMP . . . . . . . . . . . . . . . . . . . . 180
Figure 2.11 Evolution of the norm of the residual . . . . . . . . . . . . . . . . . . . 181
Figure 2.12 The CS scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Figure 2.13 An sparse signal being measured . . . . . . . . . . . . . . . . . . . . . . . 186
Figure 2.14 Recovered signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
Figure 2.15 Evolution ofkxk1 along iterations . . . . . . . . . . . . . . . . . . . . . . 187
Figure 2.16 A phase transition curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Figure 2.17 Original image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Figure 2.18 (right) Chan-Vese segmentation, (left) level set . . . . . . . . . . . . 197
Figure 2.19 A patch dictionary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Figure 2.20 The dictionary problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Figure 2.21 Original picture, and image with added Gaussian noise. . . . . . 207
Figure 2.22 Patch dictionary obtained with K-SVD . . . . . . . . . . . . . . . . . . 207
Figure 2.23 Denoised image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Figure 2.24 A synthetic image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Figure 2.25 Example of figure during MCA process . . . . . . . . . . . . . . . . . 211
Figure 2.26 The original composite signal and its components . . . . . . . . . . 212
Figure 2.27 Visualization of banded matrix using spy() . . . . . . . . . . . . . . . 216
Figure 2.28 Visualization of the Bucky ball matrix structure . . . . . . . . . . . 217
Figure 2.29 Visualization of the bucky ball graph . . . . . . . . . . . . . . . . . . . 217
Figure 2.30 Visualization of HB/nnc1374 matrix using spy() . . . . . . . . . . . 218
Figure 2.31 Visualization of heat diffusion example . . . . . . . . . . . . . . . . . . 220
Figure 2.32 Effect of Gaussian diffusion, original on top . . . . . . . . . . . . . . 222
Figure 2.33 Effect of Gaussian anti-diffusion, original on top . . . . . . . . . . . 224
Figure 2.34 The diffusion coefficient and the corresponding
flux function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
Figure 2.35 Denoising of image with salt & pepper noise,
using P-M method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Figure 2.36 Example of Bregman distance . . . . . . . . . . . . . . . . . . . . . . . . . 228
Figure 2.37 ROF total variation denoising using split Bregman . . . . . . . . . 233
Figure 2.38 Evolution of nuclear norm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Figure 2.39 A test of reconstruction quality . . . . . . . . . . . . . . . . . . . . . . . . 236
xiv List of Figures
xv
xvi Listings
1.1 Introduction
Consider the case of satellite tracking. You must determine where is your satellite,
using a large antenna. Signals from the satellite are noisy. Measurements of antenna
angles have some uncertainty margins. But you have something that may help you:
the satellite follows a known orbit, so at time T must be at position P. However this
help should be taken with caution, since there are orbit perturbations.
Satellite tracking is an example of a more general scenario. The target is to estimate
the state of a dynamic system, a state that is changing along time. The means you
have are measurements and a mathematical model of the system dynamics. These
two means should be combined as best as possible.
Some more examples could be useful to capture the nature of the problem to be
considered in this chapter.
According with the description given in a scientific conference, a research team
was developing a small UAV (unmanned aerial vehicle) to fly on their university
campus. They used distance measurement with an ultrasonic sensor, for flight altitude
control. Sometimes the altitude measurements were lost or completely wrong. In
such moments, these measurements were substituted by a reasonable value. At least
a simplistic (or even implicit) model should be used here for two reasons: to determine
that there is a wrong measurement, and to obtain a reasonable value.
Nowadays a similar problem is found with vehicular GPS. In certain circumstances
of the travel—for instance, a tunnel—the connection with satellites is lost, but the
information to the driver should continue.
Another example is the case of biosignals. When measuring electroencephalo-
grams (EEG) or electrocardiograms (ECG), etc., sometimes artifacts appear due to
bad electrodes contact, eye blinking, interferences, etc. These bad measurements
should be correctly identified as outliers, and some correction procedure should be
applied.
This chapter deals with optimal state estimation for dynamic systems. The
Bayesian methodology provides a general framework for this problem. Along time,
a set of important practical methods, which can be seen as Bayesian instances, have
been developed, such as the Kalman filter and the particle filter.
Given a dynamic system, the Bayesian approach to state estimation attempts
to obtain the posterior PDF of the sate vector using all the available information.
Part of this information is provided by a model of the system, and part is based
on measurements. The state estimation could be done with a recursive filter, which
repeats a prediction and an update operation.
Denote the posterior PDF at time step k as p(xk | Yk ), where Yk is the set of all
previous measurements Yk = { y j , j = 1, 2, . . . , k}.
The prediction operation propagates the posterior PDF from time step k − 1 to k,
as follows:
p(x | Y ) = p(x | x ) p(x | Y ) dx (1.1)
k k−1 k k−1 k−1 k−1 k−1
A B C
where A is the prior at k, B is given by the system model, and C is the posterior at
k − 1.
The update operation takes into account the new measurement yk at k:
⎛ ⎞
where w(k) is the process noise, and v(k) is the observation noise.
The first equation of the system model can be used for the term B, and the second
equation for the term L.
If some cases, with linear dynamics and measurement, the system model can be
a Gauss-Markov model:
Notice that, in order to align with the notation commonly employed in the Bayesian
filters literature, we denote vectors in boldface letters (in previous chapters we used
bars over letters).
1.1 Introduction 5
The Kalman filter [55] offers an optimal solution for state estimation, provided
the system is linear, and noises are Gaussian. It is a particular case of the Bayesian
recursive filter.
In more general cases with non-linear dynamics, the system model is based on
the two functions f(·) and h(·). A linearization strategy could be applied to still use
the Kalman filter algorithm, and this is called ‘Extended Kalman Filter’ (EKF). An
alternative is to propagate a few special points (‘sigma points’) through the system
equations; with these points it is possible to approximate the prior and posterior PDFs.
An example of this alternative is the ‘Unscented Kalman Filter’ (UKF). For non-
linear/non-Gaussian cases that do not tolerate approximations, the ‘Particle Filter’
could be used; the filter is based on the propagation of many points.
The chapter starts with the standard Kalman filter, after some preliminaries. Then
nonlinearities are considered, and there are sections on EKF, UKF, and particle
filters. All these methods are naturally linked to the use of computers or digital
processors, and so the world of numerical computation must be visited in another
section. Smoothing is another important field that is treated in Sect. 1.10. The last
sections intend to at least introduce important extensions and applications of optimal
state estimation.
Some limits should be admitted for this chapter, and therefore there is no space for
many related topics, like H-infinity, game theory, exact filters, etc. The last section
offer some links for the interested reader.
The chapter is mainly based on [8, 15, 19, 40, 72, 96].
In general, many problems related with state estimation remain to be adequately
solved. The field is open for more research.
1.2 Preliminaries
• Scalar case:
mean : μx = E(x(n))
• Two processes:
f(y) = E(x|y)
This is an example taken from [95]. It is the case of driving a car, keeping a distance
x from the border of the road (Fig. 1.1).
In this example is useful to take into account that:
with:
μ = Σ Σ1−1 μ1 + ΣΣ2−1 μ2
(1.8)
Σ −1 = Σ1−1 + Σ2−1
(it is usual in the literature to denote the Gaussian PDF as N (μ, Σ)).
At time 0, a passenger guess that the distance to the border is y(0). It could be
considered as an inexact measurement with a variance σ 2y0 . One could assume a
Gaussian conditional density f (x|y(0)) with variance σ 2y0 . In this moment, the best
estimate is x̂(1|0) = y(0).
At time 1, the driver also takes a measurement y(1), which is more accurate.
Assume again a Gaussian conditional density f (x|y(1)), with σ 2y1 < σ 2y0 . The joint
distribution that combines both measurements (product of Gaussians) would have:
σ 2y1 σ 2y0
μx = y(0) + y(1) (1.9)
σ 2y0 + σ 2y1 σ 2y0 + σ 2y1
1 1 1
= 2 + 2 (1.10)
σx2 σ y0 σ y1
The new estimate of the distance would be μx . The corresponding variance, σx2 , is
smaller than either σ 2y0 or σ 2y1 . The combination of two estimates is better.
σ 2y0
x̂(1|1) = μx = y(0) + (y(1) − y(0)) = x̂(1|0) + K (1)(y1 − x̂(1|0))
σ 2y0 + σ 2y1
(1.11)
where:
σ 2y0
K (1) = (1.12)
σ 2y0 + σ 2y1
Equation (1.11) has the form of a Kalman filter; K () is the Kalman gain. The equation
says that the best estimate can be obtained using the previous best estimate and a
correction term. This term compares the latest measurement with the previous best
estimate.
Likewise, the variance could be written as follows:
The term w is random perturbation with variance σw2 . The lateral velocity is u, set
equal to zero. After some time T , the best estimate (prediction) would be:
The increase of variance is not good. A new measurement is welcome. Suppose that
at time 2 a new measurement is taken. Again the product of Gaussians appears, as
we combine the prediction x̂(2|1) and the measurement y(2). Therefore:
where:
σx2 (2|1)
K (2) = (1.19)
σx2 (2|1) + σ 2y2
1.2 Preliminaries 9
0.2
P
0.15
M
0.1
0.05
0
-10 -5 0 5 10 15
x
The philosophy of the Kalman filter is to obtain better estimates by combining pre-
diction and measurement (the combination of two estimates is better). The practical
procedure is to update the prediction with the measurement.
Figure 1.2 shows what happens with the PDFs of the prediction (P), the measure-
ment (M) and the update (U). The update PDF has the smallest variance, so it is a
better estimation.
0.8
0.6
K
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
sigy/sigx
Another aspect of the Kalman filter philosophy is that, in order to estimate the
present system state, the value of K () is modulated according with the confidence
offered by measurements or by prediction. Suppose that the prediction variance is
constant, then, if measurement uncertainty increases, K decreases: it could be said
that the correction exerted by K becomes more ‘prudent’.
Figure 1.3 shows how K depends on σ y /σx . The figure has been generated with
the Program 1.17.
This subsection focuses on the propagation of mean and covariance. The basis of the
next study is an important lemma, which applies to a partition of a set of Gaussian
variables:
1.2 Preliminaries 11
x1
x =
x2
With:
μx1 S11 S12
μx = ; Sx =
μx2 S21 S22
Lemma: the conditional distribution of x1 (n), given a x2 (n) = x∗2 (n) is Gaussian
with:
−1
mean = μx1 + S12 S22 (x2 − μx2 )
−1
cov = Σ11 − Σ12 Σ22 Σ21
x(n + 1)
(1.22)
y(n)
Taking into account the model, the four covariance components are:
Σx x = A Σ(n) A T + Σw (1.23)
Σ yx = C Σ(n) A T + Σwv
T
(1.25)
Σ yy = C Σ(n) C T + Σv (1.26)
Thus, the mean and the variance of the partitioned process are the following:
A x̂(n) B
μp = + u(n) (1.27)
C x̂(n) 0
where:
−1
K (n) = Σx y · Σ yy = [A Σ(n) C T + Σwv ] · [C Σ(n) C T + Σv ]−1 (1.31)
The last three equations provide a one-step prediction. The term K (n) is called the
Kalman gain.
And the important point is that since we know the conditional mean, then we have
the minimum variance estimate (MVE) of the system state: this is the Kalman filter.
This subsection links with subsection (5.2.5) where a simple example of Wiener
filter was described. In that example, no model of x(n) was considered. A recursive
estimation was obtained, with the following expression:
As it was said in the second book, the above expression has the typical form of
recursive estimators, where the estimation is improved in function of the estimation
error.
1.2 Preliminaries 13
In the preceding subsection the Kalman filter was derived by studying the propa-
gation of means and variances. Notice that x̂(n + 1) was obtained using y(n). This
is prediction.
Now, let us establish a second version of the filter where x̂(n + 1) is obtained using
y(n + 1). This is filtering. A scalar Gauss-Markov example will be considered. The
derivation of the Kalman filter will be done based on minimization of estimation
variance. This is a rather long derivation borrowed from [8]. Although long, it is an
interesting deduction exercise that exploits orthogonality relations.
Let us proceed along three steps:
1. Problem statement
The scalar system is the following:
where A and C are constants (they are not matrices, however we prefer to use
capital letters).
Assumptions:
x(0) = 0; w(0) = 0
(it will reach the typical form, after the coming development)
The estimation variance is:
∂ Σ(n + 1)
= 2 E {(x̂(n + 1) − x(n + 1)) · y(n + 1) } = 0 (1.38)
∂ k(n + 1)
Another random document with
no related content on Scribd:
itself. The details of the process have not, however, been well
ascertained.
The egg of this Distome thus gives rise to a larva which enters the
tissues of one particular Mollusc. Here it becomes a branched
sporocyst within which the sexual worms are formed, apparently
each from a single embryonic blastomere ("Keimzelle"), by a process
comparable with the development of a parthenogenetic ovum, and
the whole cycle has been termed Alloiogenesis, i.e. alternation of
sexual and parthenogenetic generations (Grobben).[80] Leuckart[81]
and Looss,[82] however, consider that what was once a
metamorphosis of an individual (as in the Holostomatidae) has now
become, by maturation of the Cercaria in the comparatively modern
warm-blooded bird, a metamorphosis extending over two or more
generations.
The preventive measures seem to be: (1) Destruction of the eggs and of the
manure of rotten sheep; (2) slaughter of badly fluked sheep; (3) adequate
drainage of pastures; (4) an allowance of salt and a little dry food to the sheep;
and (5) dressings of lime or salt on the ground to destroy the embryos.[86]
The worm is found usually in couples, which have been proved to be male and
female individuals (Fig. 34), often in considerable numbers in the veins of the
pelvic region, chiefly the veins of the bladder and of the large intestine, and it is
tolerably certain that Bilharzia enter these vessels from the portal vein. Their
long slender bodies enable them to penetrate into the finer vessels, which get
partially or entirely choked up, and the circulation accordingly impeded. But the
most serious consequences are observed in the urinary bladder. The mucous
membrane is swollen and inflamed here and there, chiefly on the dorsal surface,
the capillaries appear varicose and covered with mucus, mixed with blood-
extravasations in which Bilharzia-eggs are noticeable. The eggs also cause
numerous swollen knots in the submucous tissue. Should the disease not pass
beyond this stage (and such is usually the case, especially in South Africa), a
temporary haematuria ensues. The urine, which is only expelled with great effort,
accompanied by intense pain, is mixed with blood, mucous clots, and masses of
Bilharzia-eggs, from which some of the embryos have already hatched out. The
symptoms, however, may gradually pass away, and a more or less complete
recovery accomplished. The disease may indeed be of a far less severe
character, and may not interfere with the usual occupations of the patient; but,
on the other hand, a far more extensive thickening of the wall of the bladder
sometimes occurs; hard masses of eggs, uric acid crystals, and other deposits,
may lead to the formation of stones, degeneration of the substance of the ureter,
and eventually to that of the kidney itself. The stone, indeed, has long been
known to be a prevalent disease in Egypt, and it is now known to arise from
concretions formed round masses of Bilharzia eggs. From the portal vein, again,
other Bilharzia may gain access to the rectum, or the liver, and it has also been
found in the lungs, and may give rise to most serious complications, if indeed the
patient lives.
Of the other Trematodes of man and domestic animals there is not room to
speak fully. Distomum pulmonale, which occurs in the lungs of the cat, tiger, and
dog, as well as in man, is especially common in Japan, China, Corea, and
Formosa. D. sinense and D. rathouisi have been also found in inhabitants of
these countries.
CHAPTER III
CESTODA
INTRODUCTION—NATURE OF CESTODES—OCCURRENCE OF CESTODES—THE TAPE-
WORMS OF MAN AND DOMESTIC ANIMALS—TABLE OF THE LIFE-HISTORIES OF THE
PRINCIPAL CESTODES OF MAN AND DOMESTIC ANIMALS—STRUCTURE AND
DEVELOPMENT OF CESTODES—TABLE FOR THE DISCRIMINATION OF THE MORE
USUAL CESTODES OF MAN AND DOMESTIC ANIMALS—CLASSIFICATION.
Fig. 36.—Echinobothrium affine Dies., from the intestine of Torpedo, × 43. hd,
Head; hk, hooks; hl, lobes of the head; ov, ovary; pe, penis; ps, penis-sheath;
te, testes; ut, uterus; vag, vagina; yg, yolk-glands. (After Pintner.[95])
Taenia solium, from man (Fig. 39, B), or Echinobothrium (Fig. 36), from an
Elasmobranch fish, is fixed to the mucous lining of the intestine of its host by
means of a radially-constructed apparatus of four suckers and a circlet of hooks
(Fig. 39), which are borne by the "head" or "scolex," being that part of the worm
which is directly derived from part of the larva, and which contains the central,
commissural portion of the nervous system. Firm adhesion to the host's intestine
is necessary, in order to avoid the loosening action of the peristaltic movements
of the intestine as the food passes along. The heads of different Cestodes
exhibit a marvellous variety of suckers and hooks, from a mere muscular
depression in Schistocephalus, to the compound proboscides of
Tetrarhynchus[96] which is found in Elasmobranchs. The jointed body, often of
enormous length (up to 20 yards in Bothriocephalus latus), is usually separated
from the head by a slender neck, from which the proglottides are segmented off
from behind forwards, and become more and more individualised as they recede
farther away from the neck by the intercalation of younger joints. Thus in Fig. 36
the mature, distal proglottis has passed through all the stages represented by
the other segments.
The longitudinal muscles, the nerves, and excretory vessels which supply the
proglottides are continuous throughout and with those of the head. Each joint
contains at first male genitalia comparable with those of a Trematode; then the
female organs develop, and finally self-fertilisation follows. The Cestodes feed
through their skin, probably by the aid of fine protoplasmic processes, which
penetrate the tough investing membrane and absorb the already digested food
which bathes them. When a proglottis of Calliobothrium is approaching maturity
it separates from the parent, the broken ends of muscles, nerves, and excretory
vessels speedily heal, and it is now capable of continued growth and of fairly
active movement if it remains in the intestine of the host. According to van
Beneden, it may even attain a size equal to, or exceeding, that of the whole
parent or "strobila."[97] These considerations led Leuckart, von Siebold, P. J. van
Beneden, and others, to Steenstrup's conclusion that a jointed tape-worm is
really a colony composed of two generations—the head and neck derived from
the larva, and the proglottides produced by the segmentation of the neck.[98]
This view of the colonial nature of jointed Cestodes was generally adopted from
1851 to 1880. During the last fifteen years, however, the varied interpretations of
the facts of the ontogeny of this group have led some authors to adopt the
monozootic view (that a Cestode is one individual), others are still of the older
opinion, and Hatschek (Lehrbuch, p. 349) and Lang take up intermediate
positions. Lang considers that the formation of the joints of a tape-worm from a
small fixed "scolex," is not only largely comparable with the strobilation of a
scyphistoma and the consequent formation of a pile of medusae, as in the life-
history of Aurelia, but that both processes have arisen from the power of
regenerating the necessary organs in each of the new segments. The result in
both cases is the rapid formation of a number of joints, which gradually separate
from the parent, to carry the eggs and young to new stations. Just as some
Coelenterata (Lucernaria) may be regarded as not having advanced much
beyond a scyphistoma stage, so there are unisegmental Cestodes (e.g.
Archigetes, Fig. 37) which have remained as a slightly altered but sexual scolex,
directly comparable with a Trematode, and, as all authors are agreed,
representing one generation only. Such monozootic forms are now classed as a
special family, the Cestodaria or Monozoa, of which Caryophylleus mutabilis,
from the intestine of various Cyprinoid fish, is the most abundant representative,
while Amphiptyches (Gyrocotyle) urna, from Chimaera monstrosa of the
northern hemisphere, is paralleled by A. rugosa, found in Callorhynchus
antarcticus of the southern seas.