Digital Signal Processing A host of technologies touching

Digital Signal Processing A host of technologies touching many interdisciplinary areas What is a signal? a. b. c. An agreed upon message that triggers action or carries information A message that conveys notice or warning An electronic message conveyed by telephone, radio, radar, or television

A measurable physical quantity (e.g. voltage, current, magnetic field) What is a digital signal? An array of measurements a. Examples of Signal Processing? Stock market prices, streaming of video, data mining, natural language processing, data compression, image processing, earthquake prediction, medical diagnosis, etc. b. Interdisciplinary aspects: Physics, Engineering, Psychology, Mathematics, Computer Science, Cognition, Biology, etc. c. Processing algorithms: breaks down to twiddling single and multi

dimension arrays; often algorithms result in small amounts of code, which is heavy in mathematics/statistics d. Difficulties: Signals are often imprecise and have many redundancies Digital Signal Processing (DSP) Radar and Sonar What signal comes back? Image Processing Transmit & reconstruct

Linguistics Recognize & Synthesis Algorithms Compress and encrypt, noise removal, filter Music Synthesis and edit Military:

Secure channels, laser bombs Exploration: oil, minerals, ocean mapping Climate: Earthquake, weather patterns Grand Challenges Real world simulations

Medical Medical Resonance Imaging Applications: Cell Phones, voice mail, phone support, web translation, and many more. Signal Redundancy Continuous signal (virtually infinite) Sampled representations of speech Mac: 44,100 2-byte samples per second (705kbps) PC: 16,000 2-byte samples per second (256kbps) Telephone: 4k 1-byte sample per second (32kbps)

CELP Audio Compression Algorithm: 8kbps Research: 4kbps, 2.4 kbps Military applications: 600 bps Human brain: 50 bps Definition: Sampling Rate is the number of measurements per second Noise Definition: Random electrical or acoustic activity that obscures communication Noise impact on speech

Noise degrades the speech recognition effectiveness Consider construction noise outside the window How much noise makes normal speech unintelligible? What do human beings do to compensate? Automated Solution Apply appropriate filters to eliminate noise from signals Effective filters must not destroy essential signal data

Natural Language Processing The Noisy Channel Computational Linguistics 1. Replace the vocal articulators with a synthesizer 2. Replace the ear with a receiver 2. Replace the brain with the computer Could a computer process this? I cdnuolt blveiee that I cluod aulaclty uesdnatnrd what I was rdgnieg.

The phaonmneal pweor of the hmuan mnid Aoccdrnig to rscheearch at Cmabridgde Uinervtisy, it deosn't mttaer in what oredr the ltteers in a word are, the olny iprmoatnt tihng is that the frist and lsat ltteer be in the rghit pclae. The rset can be a taotl mses and you can still raed it wouthit a problem. This is bcuseae the huamn mnid deos not raed ervey lteter by istlef, but the word as a wlohe. Amzanig huh? Yaeh and I awlyas thought slpeling was ipmorantt! Discrete versus Continuous

Continuous (analog): A signal, typically in found in nature, whose infinite values are represented by a mathematical function. That is given any point in time, if we know the initial conditions, we can compute a value at that point. Discrete (digital): A sequence of measurements, each one separated by a fixed time interval. Digital processing of continuous signals Computers deal with digital signals, because memory, though large, has a finite size. Input devices measure signal strength thousands of times a

second (e.g. audio: 44,100 is common). These values translate into a huge arrays in computer memory. Signal Dimensionality Note: Even higher dimension signals can be reduced to a single dimension Single Dimension (audio) Double Dimension (images)

Three Dimensions (video) Higher Dimension (Climate Change Simulations) Signal Power (Energy) Energy is the square of the amplitude Decibels: ratio between power of two signals. Dimensionless, like percent. SPL (Sound Pressure Level) is the ratio of

a sound to the threshold of human hearing Note: doubling the power increases decibels by three. Increase by 10 decibels increases power by 10. Note: 10 log(P1/P2) = 20 log(A1/A2) Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Sound dB

TOH 0 Whisper 10 Quiet Room 20

Office 50 Normal conversation 60 Busy street

70 Heavy truck traffic 90 Power tools 110 Pain threshold 120 Sonic boom

140 Permanent damage 150 Jet engine 160

Cannon muzzle 220 Working with Signals Special purpose languages: MatLab, Octave Matlab is very expensive, but available on campus Octave, is largely compatible (but not entirely), and is free Free signal visualization programs: GraphCalc Sound Editing tools: ACORNS Sound Editor, Audacity. Sound Editor source is available to the class

Java based resources: Java Sound, Tritonus GraphCalc (Freeware ) Good for creating and visualizing signals Communication of Signals Pulse code Modulation (PCM) An array of measured values Audio wav files: PCM with a header, defining big versus small endian, sample rate, bytes per sample, number of channels

Compression techniques to reduce bandwidth Transmit differences between adjacent values Dont transmit runs of the same value Algorithms (e.g. Linear prediction) to predict next values from previous, and transmit the prediction errors Coding techniques to transmit infrequent values using more bytes than those that are common Processing first step: translate compressed values to PCM Understanding Sine Waves Sine is the ratio of the height to the hypotenuse Many phenomena in nature occur in sine wave patterns

Basic Terminology periodic signals Amplitude The distance from zero to the maximum height Period The time it takes for a sine wave to complete one cycle Frequency (Hz) The repetitions or cycles per second (1/period) Rope Impulse Waves A hand jerk is an impulse starting the wave to travel. Speed of hand jerk determines frequency Rope stiffness determines wave velocity Reverse waves starts at the fixed end, which can resonate if forward

waves correlate, or cancel if they dont. Eventually a steady state is reached and we perceive no motion Traffic Waves Complex Wave Patterns Sound waves occupying the same space combine to form a new wave of a different shape. Harmonically related waves add together and can create

any complex wave pattern. Harmonically related waves have frequencies that are multiples of a basic frequency. Note: All frequency waves do not have to start at zero, they can be out of phase. The amount of shift in degrees is called their phase angle. Complex Wave Examples Time vs. Frequency Domain

Time Domain: A composite wave summing different frequencies Frequency Domain: Split time domain into component frequencies Historical Note Jean Baptist Joseph Fourier (1768-1830) Paper submitted: Academy of Science in Paris, 1807 Claim: All signals decompose into a sum of sine waves The review committee Laplace (1749-1827) voted to accept Lagrange (1736-1813) voted to reject. The claim did not

account for waves with sharp corners The paper never got published Results Fourier was correct if we use an infinite series of waves Lagrange was correct if we use a finite series of waves Fourier analysis is the foundation for Digital Signal Processing (DSP) Time and Frequency Understanding Digital Signal Processing, Third Edition, Richard Lyons

(0-13-261480-4) Pearson Education, 2011. Non Periodic or Quasi-periodic Many signals change their characteristics over time Fourier showed that within smalls time windows, we can analyze signal chunks as if they were periodic Processing involves 1. Chopping signals into overlapping frames of some millisecond width 2. Window the frames, which means apply algorithms to smooth the abrupt edges 3. Perform feature extraction algorithms on windowed frames to

obtain set of numbers that functions like a frames fingerprint 4. Use the feature sets to perform additional processing Quasi-periodic: Signals, that are somewhat periodic, but whose features slowly change over time Sample Sound Signal (Sound Editor) Download and install from ACORNS web-site Top: this is a demo Bottom: A goat . A coat

Time Domain: x-axis: time, y-axis: amplitude measurement Frequency Domain x-axis: time, y-axis: frequency, darkness = amplitude Narrow Band Spectrogram (horizontal lines emphasizes frequency bins) Wide Band Spectrogram (Vertical lines alignment of pitch harmonics) Signal Power or Energy Power is the square of the amplitude, often measured in decibels

Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Temporal Features Advantages Minimal processing Easy to understand Examples

Zero-crossing rate Autocorrelation and autocorrelation peaks Pitch periods Loudness contour Maximum and minimum distance between audio positive and negative amplitude Degree of voice in sounds (voicing quality) Variance between Speakers

Signal Filter Modify the signal input in some way to produce a desired output Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Filter Example Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011.

Moving Average Filter Useful Examples: A Crude low pass filter Smoothing a speech pitch contour Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Signal processing Operations

Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Linear Systems If x1[n] ---> system ---> y1[n] and x2[n] ---> system ---> y2[n] where system manipulates the input values in some way A system is linear must be Homogeneous (output is the sum of its parts) c1*x1[n] + c2*x2[n] ---> system ---> c1*y1[n] + c2*y2[n]

Time invariant (delay of input delays the output) x1[n] = x[n+k] ---> system ---> y1[n] = y[n+k] Linear Systems A Linear System is one that is homogenous and additive Homogeneity Multiply input to multiply the output If (x(n) y(n) Then c * x(n) -> c*y(n)

Additivity Summing input signals sums the output signals If x1(n)y1(n) & x2(n)y2(n) Then x1(n)+x2(n)y1(n)+y2(n) Time Invariance Not required to be linear, but highly desirable Shift the input signal to identically shift the output Assume: x(n)y(n)

Then: x(n+) y(n+) Time Invariant Example Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Sinusoidal Fidelity Feed a sine wave to the input Linear Systems Output will also be a sine wave Output phase can shift

Output amplitude can be different Non-linear systems Not likely to demonstrate sinusoidal fidelity It is possible, however, for a non-linear system to generate sine waves that dont correlate to the input in a linear fashion Linear System Example Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011.

Note: Linear systems can be connected in series and the result is still a linear system. Flipping the order of components does not affect the result because linear systems are commutative. Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. Note: Sending an impulse to an unknown system can help determine its operational characteristics Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011.

Non Linear System Example Useful Examples Median of five is useful in computing a smooth pitch contour Speech signal are mostly linear, but have some non-linear components

Understanding Digital Signal Processing, Third Edition, Richard Lyons (0-13-261480-4) Pearson Education, 2011. How to Handle non-Linear Systems? They must be linear for known ways of analysis Approaches Ignore the non-linearity if it is small enough Example: low levels of noise Convert to small amplitudes.

Many systems appear linear with small amplitudes Apply a linearizing transform Assume a[n] = b[n] * c[n] Take logarithms (log(a[n] = log(b[n]) + log(c[n]) Synthesis and Decomposition Note: Algorithms often decompose a signal into pieces, perform some calculations, and then reconstruct Note: We can always reconstruct the by adding up the decomposed pieces

Decomposition: break signal into two or more additive components Synthesis: Combining signals using scaling and addition DSP with Synthesis and Decomposition Analyze in pieces to get result Numeric Example: 2041*4 = 2000*4+40*4+1*4 Decompose, Feed into linear

Systems, and Synthesize Output Even/Odd Decomposition Assumptions Even number of total points (N) End sample attached to beginning Decompose into even and odd signal Even Symmetry xE[n]= (x[n] + [N-n])/2 Point 0 and N/2 = signal value

Odd Symmetry xO[N]= (x[n]-x[N-n])/2 Point 0 and N/2 = 0 Interlaced Decomposition Decompose into even and odd signals Even signal contains zeroes in the odd

indices Odd signal contains zeroes in the even indices

Recently Viewed Presentations

  • Frindle Vocabulary Words

    Frindle Vocabulary Words

    a soft mud or slime, especially at the bottom of a pond or river or on the ocean bottom. ooze. cramped. debris. interior. ooze. robotic. sediment. sonar. material that settles to the bottom of a liquid. sediment. cramped. debris. interior....
  • Endocytosis - Sewanhaka High School

    Endocytosis - Sewanhaka High School

    What is the "goal" of the lab? Demonstrate how Darwin's Finches (those that he observed on the Galapagos Islands) have adapted new beaks yet remain similar to the shared common ancestor that most likely came from the mainland.
  • Hypotensive disorders in pregnancy

    Hypotensive disorders in pregnancy

    is an acronym for haemolysis, elevation of liver enzymes and low platelets. Women with HELLP syndrome typically present with epigastric pain, nausea and vomiting. Hypertension may be mild or even absent. HELLP syndrome is associated with a range of serious...
  • Equation of a Line -

    Equation of a Line -

    Equation of a Line. Slope formula based on coordinates: ... Different Ways to Write Equations of Lines. Slope-Intercept Form: y = mx + b (m = slope; b = y-intercept) Use when you know the slope and the . y-intercept...
  • Anonymized -

    Anonymized -

    The effect of Federal Public Service experience is essentially the same for visible minorities and all other candidates. ... Public Services and Procurement Canada. PSPC. Global Affairs Canada. GAC. Royal Canadian Mounted Police. RCMP. Immigration, Refugees and Citizenship Canada.
  • Edge Detection Enhancement Using Gibbs Sampler

    Edge Detection Enhancement Using Gibbs Sampler

    Edge Detection. Detecting Edges in images is a complex task, but it useful in other image processing problems. Feature detection/extraction. Segmentation. Picture enlargement. Many algorithms exist for edge detection. Used Canny Algorithm for its robustness (when set correctly, hard to...
  • Slayt 1 - Amazon S3

    Slayt 1 - Amazon S3

    A screw is an inclinedplanewrapped in a spiral arounda cylindricalpost. Examples: Screw, jarlids, corkscrew, spiral staircase. IMA of a screwdepends on thespacing of thethreads. If IMA is larger, moreturnsarerequiredtodrive it into a material.
  • Leucémie lymphoïde chronique

    Leucémie lymphoïde chronique

    Regardes au microscopie optique à fort grossissement . permet une étude cytologique des éléments figurés du sang GR , GB , PLT . Permet de confirmer les données hématimetriques . les globules rouges :des cellules anucléés de 8 µm de...