Digital Signal Processing +

Transcription

1 Digital Signal Processing + Nikil Dutt UC Irvine ICS 212 Winter Material adapted from Tony Givargis & Rajesh Gupta Templates from Prabhat Mishra ICS212 WQ05 (Dutt) DSP 1

2 Introduction Any interesting embedded system has to process some input signals and generate some output signals We use the term signal in a general way Digital devices process signals in digital form A uniformly sampled stream of data spread in time (e.g., audio) or space (e.g., image) ICS212 WQ05 (Dutt) DSP 2

3 Signals and Systems Signal: set of information and data E.g., audio, video, radio, Continuous vs. discrete-time Analog and digital System: entity that transforms input signals to output signals Linear and non-linear Constant parameter and time-varying parameter Causal and non-causal Digital Signal Processing (DSP) Digitize signals and process them in the digital domain In software (e.g., DSPs, ASIPs, GPPs) or hardware (ASIC, FPGA) ICS212 WQ05 (Dutt) DSP 3

4 Analog vs. Digital ICS212 WQ05 (Dutt) DSP 4

5 General DSP Architecture Environment Embedded System Sensors f(t) A/D f n µp Memory Actuators u(t) D/A u n ICS212 WQ05 (Dutt) DSP 5

6 Sensors and Actuators Sensors: Capture physical stimulus (e.g., heat, light, sound, pressure, magnetism, or other mechanical motion) Typical generate a proportional electrical current May require analog interface solenoid mic speaker Actuators Convert a command to a physical stimulus (e.g., heat, light, sound, pressure, magnetism, or other mechanical motion) May require analog interface laser diode/transistor compass dc motor accelerometer ICS212 WQ05 (Dutt) DSP 6

7 Sampling Sampling: how often is the signal converted (sampled) to represent original signal? Twice as high as the highest frequency signal present in the input As much as 10 to 20 times for even better results Typical Sampling Rates ICS212 WQ05 (Dutt) DSP 7

8 How Fast should ADC be? ICS212 WQ05 (Dutt) DSP 8

9 Quantization Sample precision - the resolution of a sample value Quantization depends on the number of bits used to measure the height of the waveform. 16 bit CD quality quantization results in 64K values. Audio formats are described by sample rate and quantization. Voice quality - 8 bit quantization, 8000 Hz mono(8 Kbytes/sec) 22kHz 8-bit mono (22kBytes/s) and stereo (44Kbytes/sec) CD quality - 16 bit quantization, Hz linear stereo (196 Kbytes/s) ICS212 WQ05 (Dutt) DSP 9

10 Quantization and Sampling Sample Height samples ICS212 WQ05 (Dutt) DSP 10

11 Aliasing Aliasing: erroneous signals, not present in analog domain, but present in digital domain Use anti-aliasing filters Sample at higher than necessary rate 1.5 Signal freq: 5.6 Hz Sampling freq: 9 Hz ICS212 WQ05 (Dutt) DSP 11

12 Signal Processing Sample Height Digital signal S 0, S 1, S 2 S n-1 samples What can we do with it? Transpose: e.g., Z i = S i + K Amplify: e.g., Z i = S i α Compose: e.g., Z i = (S 1 i α 1 + K 1 ) + (S 2 i α 2 + K 2 ) Filter: e.g, Z i = (S i + S i+1 ) / 2 Compress: e.g., using Huffman codes Archive, match against database, etc. Or, process after converting to frequency domain Spectral analysis ICS212 WQ05 (Dutt) DSP 12

13 Time Domain Any continuous periodic time varying signal can be represented as the sum of cosine functions of different amplitude and frequency E.g., input signal captured as the sum of 4 cosine functions Captured as a Trigonometric Fourier Series (continuous) ICS212 WQ05 (Dutt) DSP 13

14 Time and Frequency Domain Signal can be represented in frequency domain Discrete values, representing amplitude at each frequency for the periodic signal How about aperiodic signals? Linear function of infinite number of sinusoidal functions ICS212 WQ05 (Dutt) DSP 14

15 (Continuous) Fourier Transform (FT) Represent periodic functions Input: h(t) continuous time domain signal Output: H(f) continuous frequency domain signal H(f) is represented by complex number a+b j Amplitude: (a 2 +b 2 ) Phase-shift: arctan(b/a) H ( f ) = h( t) = e jω h( t) e H ( f ) e j 2πft j2πft dt df = cos( ω) + j sin( ω) ICS212 WQ05 (Dutt) DSP 15

16 Discrete Fourier Transform (DFT) Input: h t discrete time domain signal Output: H f discrete frequency domain signal H f is represented by complex number a+b j Amplitude: (a 2 +b 2 ) Phase-shift: arctan(b/a) DFT is computationally expensive There is a fast software implementation (FFT) H h e t f = jω 1 = N N 1 f = 0 N 1 t= 0 H e he 2πtf j N = cos( ω) + f t 2πft j N jsin( ω) DFT IDFT Euler s Eq. ICS212 WQ05 (Dutt) DSP 16

17 Discrete Cosine Transform A special version of the DFT Used in JPEG and MPEG compression Single dimension version DC component Average signal value Cosine of differing frequency components Multiply input signal with different cosine functions 2-dimensional version Similar in nature ICS212 WQ05 (Dutt) DSP 17

18 General DSP Architecture Environment Embedded System Sensors f(t) A/D f n µp Memory Actuators u(t) D/A u n ICS212 WQ05 (Dutt) DSP 18

19 Summary Introduction to DSP Analog and digital domain conversions Sampling rate Quantization Aliasing Digital Signal Processing Transformations Time and frequency domain conversions FFT and DCT Good introductory reference for DSP: Richard Lyons: Understanding DSP Addison Wesley 1997 ICS212 WQ05 (Dutt) DSP 19