Full Advances in Information and Communication Networks: Proceedings of The 2018 Future of Information and Communication Conference (FICC), Vol. 1 Kohei Arai Ebook All Chapters
Full Advances in Information and Communication Networks: Proceedings of The 2018 Future of Information and Communication Conference (FICC), Vol. 1 Kohei Arai Ebook All Chapters
com
https://textbookfull.com/product/proceedings-of-the-future-
technologies-conference-ftc-2018-volume-1-kohei-arai/
textbookfull.com
https://textbookfull.com/product/proceedings-of-the-future-
technologies-conference-ftc-2018-volume-2-kohei-arai/
textbookfull.com
https://textbookfull.com/product/blended-using-disruptive-innovation-
to-improve-schools-1st-edition-michael-b-horn/
textbookfull.com
Antifa The Antifascist Handbook 2nd Edition Bray Mark
https://textbookfull.com/product/antifa-the-antifascist-handbook-2nd-
edition-bray-mark/
textbookfull.com
https://textbookfull.com/product/mathematical-and-statistical-
applications-in-food-engineering-1st-edition-surajbhan-sevda-editor/
textbookfull.com
https://textbookfull.com/product/changing-the-church-transformations-
of-christian-belief-practice-and-life-1st-edition-vladimir-latinovic/
textbookfull.com
https://textbookfull.com/product/wittgenstein-religion-and-ethics-new-
perspectives-from-philosophy-and-theology-mikel-burley/
textbookfull.com
https://textbookfull.com/product/practical-skills-in-chemistry-3rd-
edition-prof-john-dean/
textbookfull.com
When Good Gardens Go Bad Earth Friendly Solutions to
Common Garden Problems First Edition Judy Barrett
https://textbookfull.com/product/when-good-gardens-go-bad-earth-
friendly-solutions-to-common-garden-problems-first-edition-judy-
barrett/
textbookfull.com
Advances in Intelligent Systems and Computing 886
Kohei Arai
Supriya Kapoor
Rahul Bhatia Editors
Advances in
Information and
Communication
Networks
Proceedings of the 2018 Future of
Information and Communication
Conference (FICC), Vol. 1
Advances in Intelligent Systems and Computing
Volume 886
Series editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
e-mail: [email protected]
The series “Advances in Intelligent Systems and Computing” contains publications on theory,
applications, and design methods of Intelligent Systems and Intelligent Computing. Virtually all
disciplines such as engineering, natural sciences, computer and information science, ICT, economics,
business, e-commerce, environment, healthcare, life science are covered. The list of topics spans all the
areas of modern intelligent systems and computing such as: computational intelligence, soft computing
including neural networks, fuzzy systems, evolutionary computing and the fusion of these paradigms,
social intelligence, ambient intelligence, computational neuroscience, artificial life, virtual worlds and
society, cognitive science and systems, Perception and Vision, DNA and immune based systems,
self-organizing and adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics including
human-machine teaming, knowledge-based paradigms, learning paradigms, machine ethics, intelligent
data analysis, knowledge management, intelligent agents, intelligent decision making and support,
intelligent network security, trust management, interactive entertainment, Web intelligence and multimedia.
The publications within “Advances in Intelligent Systems and Computing” are primarily proceedings
of important conferences, symposia and congresses. They cover significant recent developments in the
field, both of a foundational and applicable character. An important characteristic feature of the series is
the short publication time and world-wide distribution. This permits a rapid and broad dissemination of
research results.
Advisory Board
Chairman
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
e-mail: [email protected]
Members
Rafael Bello Perez, Faculty of Mathematics, Physics and Computing, Universidad Central de Las Villas, Santa
Clara, Cuba
e-mail: [email protected]
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
e-mail: [email protected]
Hani Hagras, School of Computer Science & Electronic Engineering, University of Essex, Colchester, UK
e-mail: [email protected]
László T. Kóczy, Department of Information Technology, Faculty of Engineering Sciences, Győr, Hungary
e-mail: [email protected]
Vladik Kreinovich, Department of Computer Science, University of Texas at El Paso, El Paso, TX, USA
e-mail: [email protected]
Chin-Teng Lin, Department of Electrical Engineering, National Chiao Tung University, Hsinchu, Taiwan
e-mail: [email protected]
Jie Lu, Faculty of Engineering and Information, University of Technology Sydney, Sydney, NSW, Australia
e-mail: [email protected]
Patricia Melin, Graduate Program of Computer Science, Tijuana Institute of Technology, Tijuana, Mexico
e-mail: [email protected]
Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro, Rio de Janeiro, Brazil
e-mail: [email protected]
Ngoc Thanh Nguyen, Wrocław University of Technology, Wrocław, Poland
e-mail: [email protected]
Jun Wang, Department of Mechanical and Automation, The Chinese University of Hong Kong, Shatin,
Hong Kong
e-mail: [email protected]
Rahul Bhatia
Editors
Advances in Information
and Communication
Networks
Proceedings of the 2018 Future of Information
and Communication Conference (FICC),
Vol. 1
123
Editors
Kohei Arai Rahul Bhatia
Faculty of Science and Engineering The Science and Information
Saga University (SAI) Organization
Saga, Japan Bradford, UK
Supriya Kapoor
The Science and Information
(SAI) Organization
London, UK
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
v
vi Preface
Hope to see you in 2019 in our next Future of Information and Communication
Conference but with the same amplitude, focus and determination.
Kohei Arai
Contents
vii
viii Contents
1 Introduction
Frequency modulation (FM) is a nonlinear encoding of information on a carrier
wave. It can be used for interferometric, seismic prospecting, telemetry and many
more applications, each with its own statistics, dominated by the underlying
generating process. However, its widest use is for radio broadcasting, which is
commonly used for transmitting audio signal representing voice.
Communication transmission channel is subject to various distortions, noise
conditions and other impairments. Those impairments severely degrade FM
demodulator performance when a critical level is exceeded. As a result thereof,
the intelligibility and quality of the detected speech decreases significantly. This
phenomenon is known as the Threshold Effect.
Long Short-Term Memory (LSTM) recurrent neural networks [8] are powerful
models that can capture long range dependencies and non-linear dynamics. In
This research was supported by the Intel Collaborative Research Institute for Compu-
tational Intelligence (ICRI-CI).
c Springer Nature Switzerland AG 2019
K. Arai et al. (Eds.): FICC 2018, AISC 886, pp. 1–11, 2019.
https://doi.org/10.1007/978-3-030-03402-3_1
2 D. Elbaz and M. Zibulevsky
many signal estimation tasks, the advantage of recurrent neural network becomes
significant only when there is a statistical dependency between the examples.
This paper introduces FM demodulator based on Long Short-Term Memory
recurrent neural network.
The main contributions of this work are as follows:
• Utilizing the LSTM abilities to capture the temporal dynamics of speech
signals and take advantage of the prior statistics of the speech to overcome
transmission channel disturbances.
• Taking an end-to-end learning based approach for filtering both acoustical
disturbances, modeled as phase noise and transmission channel disturbances,
modeled as amplitude noise. In this approach, the LSTM learn to map directly
from the modulated baseband signal to the modulating audio that had been
applied at the transmitter, thus creating a baseband to speech mapping.
We demonstrate this method by applying it to Frequency modulation (FM)
decoding in varying levels of amplitude and phase noise and show it has a superior
performance over legacy reception systems in low SNR conditions.
In [14], author addressed the analog FM problem, but the approach taken
was to imitate the way a conventional FM demodulator works by implementing
different neural network for each building block separately. It used memoryless
(or very short memory) feed-forward neural network with only one input at
some intermediate blocks, and therefore it did not take into account the prior
knowledge of the transmitted speech. Moreover, demodulation was performed
directly on the passband signal so the input of the neural network needed to
be sampled with very high sampling rate in order to detect the change in the
input, resulting in several samples, most of which are redundant, and a very
large network for actual sampling rates that is very difficult to train and not
suitable for practical use.
As for the problem of speech enhancement, several neural networks based
solutions were suggested and shown to give good performance, for example [4,
10,22] and in [9] an LSTM based model was suggested.
While the listed works have applied neural networks to the task of radio
demodulation and for speech enhancement separately, neither of these works
suggested the task of radio transmission decoding with the prior information
of transmitted speech messages. In this sense, our project is entirely novel as
our network exploits the prior knowledge of the speech signal to overcome both
acoustical disturbances and noise in the communication channel and performs
audio reconstruction by directly operating on the baseband representation of the
modulated data.
In [20], demodulation is viewed as a problem of inference and learning and
it was suggested to use a demodulation process that can be shaped by user-
specific prior information. We adopt this approach in our work and suggest a
neural network based solution for this problem.
3 Background
3.1 Signal Modulation
Frequency Modulation is the process of modulating a sine wave with an infor-
mation message xm (t) in the following manner:
t
y (t) = Ac cos 2πfc t + 2πfΔ xm (τ ) dτ
0
where xm (t) is the data signal, which is typically a speech signal, xc (t) =
Ac cos(2πfc t) is the sinusoidal carrier, fc is the base frequency of the carrier,
Ac is the amplitude of the carrier and fΔ is the frequency deviation, which
represents the maximum shift away he carrier’s base frequency.
Sinusoid with frequency modulation can be decomposed into two amplitude-
modulated sinusoids that are offset in phase by one-quarter cycle (π/2 rad). The
amplitude modulated sinusoids are known as in-phase and quadrature compo-
nents or the I/Q components. By using simple trigonometric identities the gen-
eral expression representing the transmitted signal can be expressed as follows:
4 D. Elbaz and M. Zibulevsky
t
y (t) = Ac cos (2πfc t) cos 2πfΔ xm (τ ) dτ
0
t
− Ac sin (2πfc t) sin 2πfΔ xm (τ ) dτ
0
and we can represent the modulated signal with its I/Q components in the
following way:
y (t) = I (t) cos (2πfc t) − Q (t) sin (2πfc t)
This signal has a bandpass spectrum centered around the carrier frequency fc .
It is common to analyze communication systems by using the low pass equiv-
alents, also referred to as baseband (or I/Q components) of the original band
pass signals.
In communication systems, the statistical model for each of the above noise
models is usually assumed to be white Gaussian noise. For clarity a Fig. 1
presents a diagram depicting the communication system and its elements.
where, σ is the logistic sigmoid function, i, f, o and c are respectively the input
gate, forget gate, output gate and cell activation vector cells at time t. xt is the
input feature vector, ht is hidden output vector, bi , bf and bo are the bias terms
and Whi , Whf , Who , Wxi , Wxf and Wxo are the weight matrices connecting the
different inputs and gates with the memory cells.
4 Method Description
4.1 Dataset and Training Procedure
4.2 Architecture
We utilize the abilities of the LSTM network to capture the temporal dynam-
ics of speech signals for the problem of source signal estimation from noisy
frequency-modulated measurements. As dictated by the underlying generating
speech, future samples are also related to current samples. To exploit this depen-
dency we introduce a small delay of 100 samples, this makes the system slightly
non causal. However, it enables us to use bidirectional LSTMs [16], which are
trained using input information from the past as well as from the future of a
specific time frame. This is achieved by processing the data in both directions
with two separate hidden layers. For combining multiple levels of representations
of the modulated speech signal we use deep architectures. Deep RNNs can be
created by stacking multiple LSTM layers on top of each other, with the output
sequence of one layer forming the input sequence for the next. The stacking of
multiple recurrent hidden layers have proven to give state-of-the-art performance
for acoustic modeling [5,11]. For the above stated reasons we have decided to
End to End Deep Neural Network Frequency Demodulation 7
5 Experimental Results
In order to evaluate the performance of the DNN demodulator we used both the
Mean Squared Error (MSE) objective measure. We compare the performance of
proposed LSTM demodulator against the performance of conventional demodu-
lator implementation from Matlab communication toolbox, which is based on [7].
In both cases, DNN and conventional demodulator, the modulated signal sample
rate is set to 240 kHz and the frequency deviation is set to 75 kHz (United States
standard). In order to boost the performance of the conventional FM receiver
and compensate FM characteristic in that it amplifies high frequency noise and
degrades the overall signal-to-noise ratio, we used matlab FM broadcasters. FM
broadcasters insert a pre-emphasis filter prior to FM modulation to amplify the
high-frequency content. The FM receiver has a reciprocal de-emphasis filter after
the FM demodulator to attenuate high-frequency noise and restore a flat signal
spectrum. For clarity the full FM broadcast system is depicted in Fig. 2. We start
with the noise free case, i.e. neither phase nor amplitude noise were added to the
modulated signal. For the noise free case we get output SNR of 36.56 dB, these
results indicate high quality reconstruction. Figure 3 shows the spectrogram of
the original audio and spectrogram of the LSTM demodulator reconstruction
for noise free case. We investigated the performance of the proposed receiver, by
comparing the audio reconstruction quality in experiments employing FM mod-
ulation in the presence of various levels of additive white Gaussian (AWGN)
noise in both the amplitude and phase.
The comparison between the proposed LSTM demodulator and conventional
demodulator is presented in Figs. 4 and 5. Figure 4 show the speech reconstruc-
tion MSE for various levels of AWGN amplitude noise. Figure 5 show the speech
reconstruction MSE for various levels of AWGN phase noise.
Transmited audio
8000
6000
4000
2000
0
0 0.5 1 1.5 2
DNN reconstruction
8000
6000
4000
2000
0
0 0.5 1 1.5 2
Fig. 3. Spectrogram of the original audio signal and DNN demodulator reconstruction.
Fig. 4. Audio reconstruction quality comparison- Conventional Vs. DNN based demod-
ulator for various levels of amplitude noise.
The conducted experiments show that the proposed receiver has a clear
advantage over the conventional receiver in noise conditions, this is mostly due
to the fact that the proposed LSTM demodulator takes advantage of the statis-
tics of the generating speech signal. We prove this point by limiting the memory
of the network to only one time step, as in theory we can map the FM signal
back to audio with almost no memory. Though we were able to reconstruct the
audio for noise free case with rather small reconstruction error of SNR 17.76 dB,
however for low SNR conditions, of 0 dB amplitude noise, reconstruction was not
possible without using memory, and the demodulation failed. This experiment
shows that indeed in order for quality reconstruction to take place under noise
conditions the statistics of the generating speech signal must be accounted for.
Next we compare the reconstruction quality of the proposed LSTM demodulator
Visit https://textbookfull.com
now to explore a rich
collection of eBooks, textbook
and enjoy exciting offers!
End to End Deep Neural Network Frequency Demodulation 9
Fig. 5. Audio reconstruction quality comparison- Conventional Vs. DNN based demod-
ulator for various levels of phase noise.
6 Conclusion
References
1. Amini, M., Balarastaghi, E.: Universal neural network demodulator for software
defined radio. Int. J. Mach. Learn. Comput. 1(3), 305–310 (2011)
2. Fan, M., Wu, L.: 2017 International Conference on Communication, Control, Com-
puting and Electronics Engineering (ICCCCEE) (2017)
3. Garofolo, J.S., Lamel, L.F., Fischer, W.M., Fiscus, J.G., Pallett, D.S., Dahlgren,
N.L.: DARPA TIMIT acoustic-phonetic continuous speech corpus CD-ROM.
NASA STI/Recon Technical report N, 0, pp. 1–94, January 1993
4. Goehring, T., Bolner, F., Monaghan, J.J.M., van Dijk, B., Zarowski, A., Bleeck,
S.: Speech enhancement based on neural networks improves speech intelligibility
in noise for cochlear implant users. Hear. Res. 344, 183–194 (2016)
5. Graves, A., Mohamed, A., Hinton, G.: Speech recognition with deep recurrent
neural networks. In: ICASSP, no. 3, pp. 6645–6649 (2013)
6. Graves, A.: Generating sequences with recurrent neural networks. preprint.
arXiv:1308.0850 (2013)
7. Hatai, I., Chakrabarti, I.: A new high-performance digital FM modulator and
demodulator for software-defined radio and its FPGA implementation. Int. J.
Reconfigurable Comput. 2011 (2011)
8. Hochreiter, S., Schmidhuber, J.U.: Long short-term memory. Neural Comput. 9(8),
1735–1780 (1997)
9. Kolbaek, M., Tan, Z.-H., Jensen, J.: Speech enhancement using long short-term
memory based recurrent neural networks for noise robust speaker verification. In:
IEEE Workshop on Spoken Language Technology (SLT), no. 1, pp. 305–311 (2016)
10. Kumar, A., Florêncio, D.: Speech Enhancement In Multiple-Noise Conditions using
Deep Neural Networks. CoRR, abs/1605.0 (2016)
11. Li, X., Wu, X.: Constructing long short-term memory based deep recurrent neu-
ral networks for large vocabulary speech recognition. In: 2015 IEEE International
Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 4520–4524
(2014)
12. Önder, M., Akan, A., Doǧan, H.: Advanced neural network receiver design to com-
bat multiple channel impairments. Turkish J. Electr. Eng. Comput. Sci. 24(4),
3066–3077 (2016)
13. Pascanu, R., Mikolov, T., Bengio, Y.: On the difficulty of training recurrent neural
networks. JMLR.org (2013)
14. Rohani, K., Manry, M.T.: The design of multi-layer perceptrons using building
blocks (1991)
15. Rumelhart, D.E., Hinton, G.E., Williams, R.J.: Learning representations by back-
propagating errors. Nature 323(6088), 533–536 (1986)
16. Schuster, M., Paliwal, K.K.: Bidirectional recurrent neural networks. IEEE Trans.
Sig. Process. 45(11), 2673–2681 (1997)
17. Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., Salakhutdinov, R.:
Dropout: a simple way to prevent neural networks from overfitting. J. Mach. Learn.
Res. 15, 1929–1958 (2014)
18. Sutskever, I., Vinyals, O., Le, Q.V.: Sequence to sequence learning with neural
networks. In: Advances in Neural Information Processing Systems (NIPS), pp.
3104–3112 (2014)
19. Tieleman, T., Hinton, G.: Lecture 6.5-rmsprop: Divide the gradient by a running
average of its recent magnitude. In: COURSERA: Neural Networks for Machine
Learning (2012)
End to End Deep Neural Network Frequency Demodulation 11
20. Turner, R.E., Sahani, M.: Demodulation as probabilistic inference. IEEE Trans.
Audio Speech Lang. Process. 19(8), 2398–2411 (2011)
21. Wornell, G.W.: Efficient symbol-spreading strategies for wireless communication.
Research Laboratory of Electronics, Massachusetts Institute of Technology (1994)
22. Xu, Y., Du, J., Dai, L.-R., Lee, C.-H.: A regression approach to speech enhancement
based on deep neural networks. IEEE/ACM Trans. Audio Speech Lang. Process.
23(1), 7–19 (2015)
Performance Enhancement of MIMO –
MGSTC Using a New Detection and Decoding
Technique
Abstract. In this paper, the performance of Multi – Group Space Time Codes
(MGSTC) which is one of the Multiple Input Multiple Output (MIMO) com-
munication methods, was investigated and improved using a new signal
detection and decoding technique called parallel decoding algorithm. It is shown
that the new technique reduces the overall signal detection time, thus it increases
the speed of signal processing at the receiver. In addition, the new technique
prevents possible error propagation that may be present in serial detection
methods. Simulation results demonstrate the advantages of using parallel
decoding technique.
1 Introduction
diversity [7]. This implies increasing the transmission reliability of the MIMO system
with lower computational complexity. This is achieved without any bandwidth
expansion under assumption that the channel state information is available at the
receiver (CSIR) only [8].
The key purpose of the spatial multiplexing is to increase the data transmission rate
without requiring the bandwidth expansion by transmitting multiple independent data
streams over multipath channels. This technique exploits the multipath propagation
between the transmitter and the receiver [1, 9]. In general, the basic concept of a spatial
multiplexing is layered space time (LST) coding, where a layered indicates a data
stream from a single transmit antenna.
In the following sections, the performance of Multi – Group Space Time Codes
MGSTC-MIMO communications system is investigated using either the standard
decoding algorithm or the proposed parallel decoding algorithm. The results show that
the new decoding algorithm enhances the performance of this system.
Multi – Group Space Time Codes (MGSTCs) is a MIMO communications method that
achieves both spatial diversity and spatial multiplexing simultaneously. In 1999, the
first example of MGSTCs was published by Tarokh et al. [10].
Nstream ¼ Nt ð1Þ
These Nstream data streams split into q groups. Each group consists of B1 ; B2 ; . . .; Bq
bits, respectively as shown in Fig. 1.
Then each of the q group bits is mapped into the specific group of the transmit
antennas by space time coding, which is called a component code. Each component
code can differ from one another within the same encoder and can use either STBCs or
Random documents with unrelated
content Scribd suggests to you:
de l’iniquité, et qu’il est comme un sépulcre, non pas blanchi, mais
tout rouge.
« Donc, j’empoigne les montants du petit escalier de bois qui se
dressait le long de la colonne, et gravis les degrés. J’étais vêtu
comme tous les jours et selon mon caractère ; on pouvait voir que
j’étais pasteur d’une secte indépendante, un homme de Dieu. Et il se
fit un grand silence ! Et le sergent, qui attendait les signatures pour
les enrôlements, derrière la petite table, se leva ! Et il me tendit la
plume ! Je n’avais pas pensé à ça. C’était seulement quand on était
devenu soldat, ici, qu’on avait le droit de parler pour dire pourquoi on
se faisait soldat. Ce n’était pas une réunion contradictoire. Il fallait
signer, signer d’abord ! Et puis je te jure qu’à ce moment j’entendis
comme une voix qui insistait à mon oreille : « Signe ! mais signe
donc !! Pourquoi serais-tu venu, si ce n’est pour signer ? » Ils étaient
là dix mille, dont l’attente, le désir, dont l’émotion m’empoignaient. Et
c’était le Seigneur, évidemment, qui planait dans cette attente, dans
cette émotion, dans ce désir de dix mille âmes ! Je pris la plume et je
signai : Obedediah Rivett, ex-private aux Cornwall Rifles,
présentement pasteur de la confession wesleyenne, 150, Alexander
Road. Et je me sentis pénétré d’une joie, d’une joie ! Tout mon cœur
en éclatait, il me semblait que ma tête touchait les nuages. J’avais
les yeux secs et radieux, et pourtant c’était comme si j’eusse été
inondé de larmes à l’intérieur : des larmes divines et délicieuses.
J’étendis les bras, j’ouvris la bouche et je parlai ! Je ne me
souviens plus exactement de ce que j’ai dit, mais c’était beau !
Quelque chose je suppose comme : « J’arrive tel Athaniel, je quitte
la vigne de mon père, et la maison fraîche, et l’épouse, et le figuier
devant la porte et je prends le casque, et je ceins l’épée. Car il est
écrit : « Tu donneras ta vie pour ton peuple ! Tu n’auras pas de pitié
pour la race impure ! »
Voilà ce que j’ai dit, je crois, et beaucoup d’autres choses. Et je
m’aperçus que, par le petit escalier de bois, c’étaient des gens et
encore des gens qui montaient pour signer, pour s’enrôler comme
moi ! Je restai là, les bras étendus, je ne sais combien de temps.
Puis je retournai à Bloomsbury, tout étourdi, hors de moi-même. Je
trouvai ma femme dans le parloir :
« — Lizzie, je me suis engagé. Demain, je me présente au camp
d’Aldershot.
« Mais elle ne me crut pas :
« — J’ai été appelé, Lizzie. En vérité j’ai été appelé.
« Alors, elle comprit que c’était sérieux. Elle a pleuré, mais
maintenant elle m’envoie un colis toutes les semaines. C’est une
brave femme. »
— C’est drôle, fit Muffin en réfléchissant. Et il y en a pas mal qui
ont été « appelés » comme toi : des anarchistes, des pacifistes. C’est
drôle !
— Oui, reconnut l’ancien pasteur de la chapelle wesleyenne. Je
te dis que c’est rigolo !
TOM KETTLE, AUSTRALIEN
« C’est à partir de cet instant que j’ai répété sans cesse à ceux
qui doutaient : « Vous verrez, vous verrez ! Il n’y a pas de peuple qui
ait moins changé que le peuple de France. C’est quarante millions
d’aristocrates. Il étonnera le monde… »
FIN
TABLE DES MATIÈRES
Le Diable au Sahara 5
Le mammouth 51
Le manteau de plumes 75
L’ombre de Byron 91
Du berger à la bergère 101
Le parfum 111
La Mer.
La mine 127
Un gabier exceptionnel 135
Un cimetière 145
Les cachalots 153