Outline. J-DSP Overview. Objectives and Motivation. by Andreas Spanias Arizona State University

Transcription

1 Outline JAVA-DSP () A DSP SOFTWARE TOOL FOR ON-LINE SIMULATIONS AND COMPUTER LABORATORIES by Andreas Spanias Arizona State University Sponsored by NSF-DUE-CCLI New NSF Program Award Starts Jan Dec involves 5 universities Also core software used in an NSF CRCD PRESENTED AT SMU NOV. 18, 2004 Overview Web-based DSP simulations On-line laboratories in ASU senior-level DSP class Capabilities for simulations in speech, image, controls, communications, and TF representations Advanced Functionality (DNA, HMM, MPEG Audio etc) Functions/Tools for Exposing High-School Students to DSP Echo and Reverberation Effects, Bass/Treble controls Shelving, Peaking Filters, and Graphic Equalizers MIDI and DTMF demos Embedding Interactive Simulations in Web Lectures Conclusion A. Spanias 2 Objectives and Motivation Overview Develop an on-line simulation tool for use in all DSP-related and Linear Systems courses. BASIC FUNCTIONALITY IN Motivate Students to take DSP electives Fundamental DSP Functions (FFT, IFFT, Windowing etc.) Buttons to select blocks Working area Expose High School Students to Electrical Engineering Provide hands-on DSP experiences to undergraduate/ graduate DSP students and distance learners. Accelerate leaning by exposing students to hands-on manipulation of signals and DSP systems. TOOLS FOR INSTRUCTORS Seamlessly embed simulations in web lectures; Demos for use in High-School Environments Basic Arithmetic Functionality, Multi-rate DSP Pole-Zero z-domain diagrams Frequency Response, Visualization Blocks, Digital Filtering, FIR/IIR Filter Design Spectral Estimation 3D Animations, Blocks Dialog windows Visualize DSP concepts! A. Spanias 3 A. Spanias 4 1

2 The On-line Labs A B C D E F Menu items Filter Blocks Section Permanent Blocks List Menu Selection (Existing) List Menu Selection (Planned) Disclaimer A B C D H E F J G LAB 1: Difference Equations and the Z-Transform This J lab introduces the students to the concepts of linear- time-invariant invariant (LTI( LTI) ) systems, Z-transforms, Z and the impulse response of LTI systems. In particular, students observe the filtering effects and become familiarized with the source-filter configuration. G H I J K L Interactive Visual Demos Simulation Flowgram Dialog window (PZ Placement) Plot Window to View Results Help Information Error messages I Error messages shown here L K LAB 2: Pole-Zero Plots and Frequency Responses This J lab deals with the effects of pole-zero locations on the magnitude frequency response. The students observe the variations in the frequency response by graphically moving the poles and zeros in the Z-domain. Z They also design low-pass and high-pass filters based on the pole- zero placement method. A. Spanias 5 On-line Labs (2) On-line Labs (3) LAB 3: FIR and IIR Filter Design This J lab deals with the design and analysis of various FIR and IIR filter design methods. These include filter design based on windowing, frequency- sampling, the Parks-McClellan algorithm, and IIR analog filter approximations. LAB 4: The Fast Fourier Transform (FFT) In this J lab, students gain familiarity with the estimation of DFT spectra, DFT spectral leakage, DFT resolution, Parseval s theorem for the DFT, and FFT properties and symmetries. LAB 5: Multi-rate Signal Processing and QMF Banks This J exercise deals with the effects and applications of sampling rate conversion and quadrature mirror filter (QMF) banks. In particular, students experiment with aliasing and imperfect reconstruction in decimation and interpolation of digital signals. s. LAB 6: Random Signal Processing Spectral Estimation This J lab teaches the basics of classical and parametric spectral estimation. Specifically, students become familiar with correlograms, periodograms, spectrograms, AR estimators, and linear predictive coding (LPC). Pole-Zero fie-ne\fie- 03\Prof_SPANIAS\ASSESSMENT\AV I\PoleZeroPlacement1.aviAnimation Lab Submission Procedure 2

3 Blocks for Filter Design Filter Design Modules in A A Window Method Kaiser Design Min. Max. Method IIR Filter Design - Analog Approximations Frequency Sampling Method A. Spanias 9 A. Spanias 10 Filter Design Using Window Method Kaiser Design Module in Window Method Window Method Kaiser Design Kaiser Design Min. Max. Method Min. Max. Method IIR Filter Design - Analog IIR Filter Design - Analog Approximations Approximations Frequency Sampling Frequency Sampling Method Method A. Spanias 11 A. Spanias 12 3

4 Parks-McClellan Optimum FIR Design Filter Design Modules in The Parks-McClellan design is based on Min-Max This class of methods involve minimizing the maximum error between the designed FIR filter frequency response and a prototype E( e min { max ( ) } jω E e { h ( i ), i = 0,1,..., L} jω ) = W ( e jω where )( H d ( e jω ) H ( e jω )) Window Method Kaiser Design Min. Max. Method IIR Filter Design - Analog Approximations Frequency Sampling Method A. Spanias 13 A. Spanias 14 IIR Design - Applying the Bilinear Transformation IIR Filter Design in specification Prewarping Design bilinear transformation Ω ω ω Window Method Kaiser Design Min. Max. Method IIR Filter Design - Analog Approximations Frequency Sampling Method A. Spanias 15 Ω A. Spanias 16 4

5 Filter Design using Frequency Sampling Filter Design using Frequency Sampling Ideal frequency response (a) 0 π/4 π/2 π Sampled ideal frequency response N=16 samples (b) 0 π/4 π Interpolated frequency response 0 π/4 π (c) Frequency Sampling Method Demo A. Spanias 17 A. Spanias 18 Extensions from DSP to other Systems Courses Advanced Functionality Analog and Digital Communications Hidden Markov Model (HMM) Training Control Systems Perceptual Audio Coding Techniques Image and 2D Signal Processing Genomic Signal Processing Speech Analysis and Synthesis Adaptive Signal Processing and Beam- Time/Frequency Representations forming Applications A. Spanias 19 A. Spanias 20 5

6 for use in High Schools MIDI Functionality Developed a series of functions that are high school friendly. These functions are categorized as: Tone-generators, MIDI, DTMF Echo and Reverberation Effects. Simulates a piano keyboard and generates MIDI sounds at the frequencies described by the MIDI standard. Generate a sequence of pre-recorded tones. MIDI DEMO These functions are complemented with simple exercises (tone, echo, echo, etc) as well as advanced simulations (vocoders, MP3, etc). Developed pilot materials a for dissemination to high-school students. The MIDI block can generate a single tone of length: 256 (1 frame), 1280 (5 frames) and 8192 (32 frames) samples. y = cos(2 π fnt) where f is taken from a MIDI standard table [ FFT A. Spanias 21 A. Spanias 22 DTMF Functionality Digital Audio Filters (1) Generates dual-tone-multifrequency (DTMF) tones used in landline telephony applications. DTMF DEMO Echo Effects The echo effect is obtained by mixing the input signal with its delayed version. The tones can be played back using the provided sound player, and used in a DSP simulation. The proportion of the delayed signal to the "clean" original signal determines how obvious the echo is, and the delay signifies the echo period. y = cos(2 π f nt) + cos(2 π f nt) 1 2 where f1 and f2 are chosen from the tone frequencies (697, 770, 852, 941, 1209, 1336, 1477 (Hz)). The sampling frequency is 8 KHz, i.e., T = 0.125ms FFT y(n)= x(n) + b. x(n-r) R = the number of echo delay in samples. In order to have a distinguishable echo, R should be relatively large. b is the attenuation constant ( b < 1). Echo A. Spanias 23 A. Spanias 24 6

7 Digital Audio Filters (2) Digital Audio Filters (3) Reverberation Effects Tone Control Shelving & Peaking Filters Reverberation is obtained by mixing the input signal with the delayed versions of its feedback. The effect of the feedback results in multiple echoes. In order to introduce high school students to filtering and frequency responses we used the bass/treble controls as used in high-fidelity audio systems. SHELVING DEMO y(n) = x(n) + b. y(n-r) R = feedback delay in samples. b is the attenuation constant ( b < 1). Reverb In addition to bass and treble controls, the Peaking Filter block allows students to attenuate audio signal components outside a specified frequency range. A. Spanias 25 A. Spanias 26 Digital Audio Filters (4) On-line Labs for HS Students Graphic Equalizer Input Simple Simulations Advanced Simulations Cascaded peaking filters as shown. By dividing the audible frequency spectrum into several frequency bands. It alters the frequency response of each band independently by varying the corresponding peaking filter s gain. Peaking Filter #1 Peaking Peaking Peaking Filter Filter Filter #2 #N-1 #N Output Frequency domain representations and simple MIDI and DTMF simulations. The Echo and Reverberation concepts leading to FIR and IIR filtering concepts. The Low-pass and high-pass filtering concepts in conjunction with the shelving filters LPC compression schemes used in Cell phones. Simulation of some of the basic elements in an MPEG-1 Layer-III (MP3) encoder. DEMO Introduction to the compression techniques employed in a typical JPEG compressor. GRAPHIC EQUALIZER DEMO Band-pass filtering concepts in conjunction with the peaking filters and graphic equalizer. A. Spanias 27 A. Spanias 28 7

8 Seamlessly Embed Simulations in Web Content Interface with MATLAB FIVE SIMPLE STEPS 1. Prepare demonstration in DSP Tutorial 3 THREE SIMPLE STEPS 1. Prepare demonstration in Functionality being developed in 2. Export simulation in script. 3. Copy and paste script into an HTML file. 4. Add your own educational content 5. Deliver to students. HTML code SCRIPT Z transform laboratory exercise In this lab we use the Filter block of to <applet CODE="JDsp.class" invert the width="400" Z transform of various signals. As we height="250"> have seen in the previous lab, the Filter block in J- DSP can implement a filter transfer function of the <param name="numcommand" following value="15"> form: <!-- START PARTS --> <param name="0" value="b0-siggen(1,1)"> <param name="3" value="b3-pzplace(2,3)"> <param name="4" value="b4-freqresp(2,0)"> Start the editor to <!-- END PARTS --> see an example of a filter <!-- START CONNECTIONS --> with an impulse response <param name="5" value="c "> h(n) = 0.9 n u(n). Convert <param HTML name="6" code value="c "> the impulse response <param name="7" value="c "> equation to the equivalent <!-- END CONNECTIONS --> transfer function in the z- transform domain. 2. Export simulation in MATLAB script. 3. Copy and paste into MATLAB editor window. A. Spanias 29 A. Spanias 30 Assessment Assessment (2) General Assessment Concept-Specific Assessment (1) 3% 1% 0% How helpful are the features? Is it easy to get used to the? Will the students consider for simple DSP simulations? How beneficial is the tool for distance learning? as a multi-disciplinary simulation environment? Strongly Agree (%) Agree (%) Neutral (%) Disagree (%) Strongly Disagree (%) 2% 13% 0% 47% 24% 61% Understanding of general concepts of using FFT in signal analysis. 3% 0% 8% 42% 47% Learning of using window type for sharp transition (in Lab 4) 46% 3% 0% 50% 8% 42% Understanding of the concepts of the Z-transform (in Lab 1) Understanding of the concepts of FIR and IIR filter design 1% 1% 47% 7% 44% 3% 2% 11% Understanding of the concepts 29% of pole-zero and freq-response 55% Learning of generating a sinusoid with a digital filter 8

9 Concept-Specific Assessment (2) Yes 5% 1% No 95% Understand more clearly the relationship of the impulse response with the transfer function 99% Understand more clearly the spectral resolution of the FFT is limited by frame size window type and window size Assessment (3) % 17% 56% 30% 17% 53% Yes 37% 23% 40% Lab No. No 40% 13% 47% After the exercise you are more comfortable with the related topics Did not use Enough information in the help screen 7% 93% 1 6% 94% 8% 92% Lab No. 10% 90% Pre/Post Assessment (1) Our newest assessment instrument that focuses on evaluating whether learning of certain topics is attributed specifically to USING. Assess whether accelerated the learning curve in the DSP class. STATISTICS From EEE 407 Class: No. of Students: 64 (Spring 2003) A 20% average improvement can be noted after performing the Lab-1. Assessment (4) % of students who answered correctly Customized Specifically to Pre/Post-Lab Assessment Results Lab-1 Lab-2 Lab-3 Lab-4 Lab-5 Labs Pre-lab assessment Post-lab assessment % Improvement Assessment (5) Conclusions Pre/Post Assessment (2) The percentage improvement corresponding to Lab-2 was significant, i.e., 45%. This can be related to the fact that the lab-2 simulations involve seamless animations of pole-zero locations and frequency response computations. This assessment result was influential in re-designing most of the blocks to incorporate animations. % of students who answered correctly Customized Specifically to Pre/Post-Lab Assessment Results Pre-lab assessment Post-lab assessment % Improvement Developed several functions in that are intended to introduce HS students to DSP concepts as used in several exciting applications. The modules start from simple concepts using MIDI and DTMF encoders and then move to more involved simulations using filters and echo/reverberation systems. Java Scripting Capabilities to seamlessly embed simulations in web content Improvements of 22%, 10%, and 15% can be noted in the Labs 3, 4, and 5, respectively Lab-1 Lab-2 Lab-3 Lab-4 Lab-5 Labs Finally, some advanced simulations and demonstrations have been developed to familiarize HS students with the DSP concepts used in cellular telephony, JPEG compression, and MP 3 players. A. Spanias 36 9

10 : A Distance Learning Paradigm Existing Prototype Run simulation Evaluate lecture streaming video simulation Streaming video Questions Student notes and report Lecture notes in HTML HTML lecture interface Visit Related references links Related web page content: 1. Prof. Smith s web page 2. FFT dedicated web site Related books: 1. Classroom text 2. The FFT transform by notes and lab report Attach text Attach plots Submit laboratory report Lab 2 concentrates Load instructors on the Fast Fourier example Transform simulation (FFT). FFT lecture Consider the symmetries in the following signals. We want to see how these symmetries affect FFT spectra. for more information on Planned GUI for Comprehensive Delivery of Lectures/Simulations/Labs A. Spanias 37 A. Spanias 38 10