[ABE-L] Seminário - Captura e processamento eficiente de sinais

[Hugo Carvalho]

Prezados,

Gostaria de lhes convidar para assistir a um seminário com um tema super
interessante e atual, ministrado pelo Felipe Pagginelli. Com muito orgulho
e satisfação, apresento meu primeiro aluno de Iniciação Científica
(estudando Filtro de Kalman) e de Mestrado (dissertação
<http://objdig.ufrj.br/11/teses/926242.pdf> sobre o Lema de
Johnson-Lindenstrauss), hoje doutorando na Universidade Técnica de Munique
sendo orientado por Felix Krahmer (TUM) e Ayush Bhandari (Imperial College
London). Ele voou longe após a conclusão do Mestrado, está passando umas
semanas no Brasil, e gostaria de nos trazer as novidades de sua pesquisa
por lá. Portanto, segue abaixo informações sobre a apresentação.

Abraços, e nos vemos lá!

--

*DATA:* 07/03/2023 (terça-feira)

*LOCAL:* Presencial, no Instituto de Matemática da UFRJ (Sala C-116 ou
C-119, informação a confirmar na segunda-feira)

*HORÁRIO:* 14h

*PALESTRANTE:* Felipe Pagginelli

*TÍTULO:* Next generation digital acquisition using modulo nonlinearities

*RESUMO: *One of the flagships of the Third Industrial — so-called
"Digital" — Revolution is a mathematical result called Nyquist-Shannon's
Theorem that allows for the perfect recovery of a bandlimited analog
time-signal from a digital representation of it via sampling and
quantization when samples are taken with at least with a minimum sampling
rate (Nyquist's rate).

Despite not being a requirement of that result, the electrical circuits —
called analog-to-digital converters (ADCs) — that are used to implement it
in real life are limited in the range of amplitudes — called Dynamic Range
(DR) — of the signals that they can acquire. That is a major bottleneck in
the area of Signal Processing, since Dynamic Range limitations yield
information loss since the very acquisition of the input.

The state-of-art methods can be summarized by the "acquired now, process
process later" procedure, that are either: purely algorithmic-based, by
"guessing" or combining various acquisitions of the same signal made by
ADCs calibrated in different ranges; or purely hardware-based, by designing
ADC with wider ranges. Nevertheless, both of those procedures are based on
calibration techniques and lack theoretical recovery guarantees, since the
ground-truth is unavoidably unknown.

In contrast, the Unlimited Sampling Framework (USF) was recently proposed
as the first paradigm that connects the hardware and algorithm parts to
overcome the information loss. Namely, on the hardware side, the use of an
ADC that folds the input signal inside its Dynamic Range (Self-reset ADC)
via a (nonlinear) modulo operation is proposed; on the algorithmic side,
they deal with the problem of unfolding those modulo samples.

On their pioneering work, the authors were able to: provide a sufficient
guarantee and an algorithm for perfect recovery in the noiseless setting by
leveraging on oversampling — i.e., sampling above the Nyquist's sampling
rate — and on the bandlimitedness and smoothness of the input signal. Also,
that novel approach is single-shot — that is, requires just one acquisition
of the input signal and admits inputs of unlimited amplitude.

In our talk, we intend to discuss how that first algorithm behaves in the
noisy setting, the theoretical reasons for that, and also to provide a
sensibility analysis for it. Finally, we intend to share further approaches
to that problem and to invite for future contributions.

*ORIENTADORES:* Felix Krahmer (TUM), Ayush Bhandari (Imperial College
London)
*BOLSA:* JADS-IGSSE

-- 
Hugo Tremonte de Carvalho

• Assistant Professor @
 - Federal University of Rio de Janeiro (*UFRJ <https://ufrj.br/>*)
 - Institute of Mathematics (*IM <http://www.im.ufrj.br/>*)
 - Department of Statistical Methods (*DME <http://www2.dme.ufrj.br/>*)

• Member of the MusMat Research Group <https://musmat.org/>

• Coordinator of the Specialization in Data Science
<http://www.im.ufrj.br/index.php/pt/ensino/pos-graduacao/pos-graduacao-do-im/cursos-lato-sensu/ciencia-de-dados>

Personal website:
*im.ufrj.br/~hugocarvalho/ <http://im.ufrj.br/~hugocarvalho/>*
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <http://lists.ime.usp.br/pipermail/abe/attachments/20230303/3c55c7b8/attachment.htm>