Tutorials are FREE.

Saturday, December 18, 2004

   

Time:  16:00 - 17:30

Overview on Band-Pass Sampling Approaches for On-Board Processing
Authors: L. Simone, G. Lippolis, M. C. Comparini, D. Gelfusa, V. Piloni, R. Novello,
Affiliation: Alenia Spazio, Rome, Italy

 

Sunday, December 19, 2004

   

Time:  16:00 - 17:30

Radar processing techniques and technologies: state of the art, hot topics and way ahead
Authors: C. Falessi ( AMS Technology Director), A. Farina ( AMS Scientific Director ) .
Affiliation: AMS, Rome, Italy

 

Monday, December 20, 2004

(Note: parallel tutorial sessions)   

 

Time:  16:00 - 17:30

•  On embedded system architectures for speech recognition applications: the gap between the status and the demand.
  Authors: L. Arnone, S. Bocchio, A. Rosti
  Affiliation: STMicroelectronics, Agrate Br. Italy
   

• On-Board Signal Acquisition for Deep Space Communications
  Authors: L. Simone*, M. C. Comparini*, G.C. Cardarilli**, A. Del Re**, M. Re**
  Affiliation: * Alenia Spazio, Rome, Italy- ** University of Rome “Tor Vergata”, Rome, Italy
   

Tuesday, December 21, 2004

 

End of Symposium

No tutorials

 

All Questions concerning tutorial should be directed to the tutorials chair:

                Professor Gian Carlo CARDARILLI

                “Tor Vergata”University, Rome, Italy

                Email: g.cardarilli@uniroma2.it

 

TUTORIAL DETAILS

 

                                                                                                                         Back to Top

Title: On embedded system architectures for speech recognition applications: the gap between the status and the demand.

Authors: L. Arnone, S. Bocchio, A. Rosti

Affiliation: STMicroelectronics, Agrate Br. Italy

 

Description: Speech provides a natural and direct method for human-machine interaction, since it is one of the principal forms of human communication. Speech has many advantages over other means of communication, which make recognition systems attractive. People can speak roughly 4-5 times faster than they can type, allowing for greater communication efficiency. Speech recognition accessories such as microphones and cell phones are small and simple, allowing mobile capability.

Until recently, research in speech recognition had largely focused on achieving high accuracy with continuous, speaker independent recognition on larger vocabulary sets, usually loosing speed and achieving complexity in return. To maintain high recognition rates as vocabulary sizes increases, research has concentrated heavily on modelling speech on two levels, acoustic and language.  Despite the increase in accuracy, the complexity of using acoustic and language modelling effectively has hindered recognition speed and computational cost. As speech research takes recognition to a larger scale, the speed, memory size, and area of the required computing hardware becomes problematic. When speech recognition is done in software running on a high-end processor, many of these concerns can be ignored. But for handheld devices, these problems limit the usability of recognizers for all portable or mobile applications. Implementing speech recognition in hardware/software co-design architecture allows to leverage on the flexibility of  the software solution and also allows to a better understand how speech recognition should be done to maximize performance, so that it can be ported to embedded mobile system. In a pure software approach, the trade-off among accuracy, speed, and computational cost are fixed. They depend on the algorithms running over the CPU

The constraints of complexity and speed hinder the use of large vocabulary speech recognition systems. Much research has been done to increase speeds and reduce complexity through software. This has met with some success, but for more aggressive speed and power improvements for recognition systems, the introduction of more hardware implementations seem inevitable. Recent speech recognition research has conventionally resolved performance issues through careful software design and tuning. In software, running speech on a high-end processor is still limiting because the memory structure of the processor is poorly optimized for speech workload. We believe that the reason why there has not been any significant speed improvement with each generation of new processors is because the memory bandwidth bottleneck. Speech requires large amounts of data, being transferred from memory. Because processors are general purpose, the memory structure is not optimal for the speech recognition application.

 

                                                                                                                          Back to Top

Title: Overview on Band-Pass Sampling Approaches for On-Board Processing

Authors: L. Simone, G. Lippolis, M. C. Comparini, D. Gelfusa, V. Piloni, R. Novello,

Affiliation: Alenia Spazio, Rome, Italy

 Description: Aim of this tutorial is to provide an overview on band-pass sampling approach for on-board processing applications.

The paper will focus on the following topics:

·        Frequency translations using mixing and multirate operations

q       Mixing for complex discrete-time signals

q       FIR filter with frequency translation

·        Real and I/Q sampling of bandpass signals

q       The image rejection concept

q       Zero-IF approach

q       Hilbert-filter approximation: second-order sampling, halfband filtering, interpolation filters

·        Nonidealities in sampling and Analogue-to-Digital Conversion

q       Aperture jitter effects

q       Quantization and saturation noise

q       Spurious-free dynamic range

q       Automatic Gain Control design for Analogue-to-Digital Conversion

 

                                                                                                                          Back to Top
 

Title: On-Board Signal Acquisition for Deep Space Communications

Authors: L. Simone^*, M. C. Comparini^*, G.C. Cardarilli^**, A. Del Re^**, M. Re^**

Affiliation: * Alenia Spazio, Rome, Italy- ** University of Rome “Tor Vergata”, Rome, Italy

 

Description: A big issue in Deep Space Communications is related to the on-board signal acquisition strategies.

 

In general, the carrier acquisition procedure is based on the on-ground sweeping, i.e.:

1.      The ground station transmits the up-link un-modulated carrier.

2.      The carrier frequency is swept around the estimated on-board receiver rest frequency.

3.      The sweep range is selected according to the uncertainty on the rest frequency estimation.

The sweep rate is selected according to the signal dynamics (Doppler rate), the acquisition time, the acquisition probability. It is strongly related to the input signal power: higher is the carrier-over-noise spectral density ratio, wider is the loop bandwidth and faster is the rate.

As an alternative, the on-board receiver can perform the sweeping function avoiding on-ground sweep (i.e. local sweep technique).

Another option is to accomplish up-link carrier acquisition relying on on-board spectral estimation process. The estimated up-link frequency is used to properly set the local oscillator rest frequency.

 

The digital techniques, largely used in the modern Deep Space Transponders, offer the possibility to implement high-order loop which ensures superior tracking performance with respect to the analogue counterparts. However, in case of receiver idling (i.e. only the noise is present at the receiver input) such kind of loops present frequency drift (in case of a 2nd – order, II type loop) or frequency drift rate (in case of a 3rd – order loop). These effects must be taken into account when designing the acquisition algorithm.

 

In order to optimize the carrier acquisition performance, two different approaches have been studied and implemented. In case of on-ground or on-board sweeping techniques, we propose the use of a Carrier Frequency Detection scheme, which allows to detect the carrier power when the up-link signal passes around the receiver rest frequency in order to close the loop only when the carrier is inside the pull-in bandwidth.

 

On the other hand, spectral estimation techniques based on Fast Fourier Transform (FFT) algorithm have been successfully implemented both in Software and in dedicated Hardware device. In this last case, an innovative approach based on the use of the COrdinate Rotation DIgital Computer (CORDIC) algorithm for complex samples phase rotation has been developed.

 

Aim of this paper is to provide an overview of the on-board signal acquisition techniques suited for Deep Space Transponders addressing also mission aspects and implementation issues.

                                                                                                                          Back to Top

Title: Radar processing techniques and technologies: state of the art, hot topics and way ahead

Authors: C. Falessi ( AMS Technology Director), A. Farina ( AMS Scientific Director ) .

Affiliation: AMS, Rome, Italy

 Description: List of contents:

 

Techniques :

- Evolution of adaptivity in radar systems

- Time domain

- Space domain

- Space-time domains

- Space-time-frequency domains

- Parallel processing to support real-time adaptivity

- Knowledge Based Radar: the way ahead for the beginning of 21st century

 

Technologies :

- COTS DSP and PowerPC, System COTS

- FPGA and Programmable ASIC

- Co-Sourcing Off The Shelf and Tile architecture

- SW, Middleware and Modelware

- Digital-Photonic  concurrent processing

- Emerging processing technologies

 

 

                                                                                                                          Back to Top