Transcript
Digital signal processing What is it and why use it ? “When I use a word”, said Humpty Dumpty “it means just what I choose it to mean” Lewis Carroll, “Through the Looking-Glass”.
D R Campbell
School of Computing
University of Paisley 1
Compact Disc Playback System Reading Head
Encoded Compact Disc
D R Campbell
Decode Electronics
Steering Digital Controls
School of Computing
Digital Filtering DAC
Audio Reconstruction Out Filtering
University of Paisley 2
1
Comparison of audio recording specifications Feature
LP Record
Analogue Tape
CD
DAT
Frequency Response
30 Hz – 20kHz +/- 3dB
20Hz – 20kHz +/- 3dB
20Hz – 20kHz +/- 0.5dB
20Hz – 20kHz +/- 0.5dB
Dynamic Range
70 dB
70 dB
> 90 dB
> 90 dB
Signal to Noise ratio
60 dB
~ 70 dB
> 90 dB
> 90 dB
Harmonic Distortion
1–2%
< 0.5 %
0.004 %
< 0.05 %
Channel Separation
25 – 30 dB
40 – 60 dB
> 90 dB
> 80 dB
0.03 %
0.03 %
Non-detectable
Non-detectable
Low
Low
High
High
Wow and Flutter Robustness
After IEEE ASSP Magazine Oct. 1988
D R Campbell
School of Computing
University of Paisley 3
A Typical Digital Signal Processing System Band limiting Low-pass filter
Signal in
Sampling and A to D conversion
Binary Numbers 0010 1110 1010
ADC E.g. Filter,
Analogue side Signal out
Low-pass filter Smoothing
D R Campbell
Digital side
Digital Pitch warp, Signal Echo Processing operations
DAC 1001 0111 0011
D to A conversion School of Computing
University of Paisley 4
2
DSP System Needs Input and output filtering Analogue to digital, and digital to analogue conversion Digital processing unit So why accept this apparent complexity?
D R Campbell
School of Computing
University of Paisley 5
Why use Digital Processing ? (1) 1. Precision In theory the precision of Digital Signal Processing systems is limited only by the conversion process at input and output (A to D and D to A). In practice, sampling rate (number of samples per second) and word length restrictions (number of bits) modify this. However the increasing operating speed and word length of modern digital logic is allowing many more areas of application. D R Campbell
School of Computing
University of Paisley 6
3
Why use Digital Processing ? (2) 2. Robustness Digital systems use two-level signals. Due to protective margins (noise margins), they are inherently less susceptible than analogue systems to : a) electrical noise (pick-up) b) component tolerance variations Adjustments for electrical drift and component ageing are essentially removed; important for complex systems. Practically inappropriate component values can be avoided e.g. very large capacitors or inductors. D R Campbell
School of Computing
University of Paisley 7
Why use Digital Processing ? (3) 3. Flexibility Programmability allows upgrading and expansion of the processing operations, without necessarily incurring large scale hardware changes. Practical systems with desired Time Varying and/or Adaptive characteristics can be constructed.
D R Campbell
School of Computing
University of Paisley 8
4
ADC configurations 1 Control
V in
Start/End Conversion, Range Select, Word Length select
Parallel out
ADC
Serial out V ref
D R Campbell
School of Computing
University of Paisley 9
ADC configurations 2 Control S/H V in
D R Campbell
S/H
ADC
School of Computing
Parallel out
University of Paisley 10
5
ADC configurations 3 V0
Multiplexer S/H
V7
Parallel out
ADC
S/H Mux address
D R Campbell
Control
School of Computing
University of Paisley 11
Analogue to Digital Conversion error noise x(t) Sampling x(nT) x(nT) + e(nT) + v(nT) ADC switch b bit quantiser e(nT) = Quantisation error (decreases by 6dB for each 1 bit increase in wordlength). v(nT) = electrical noise due to ADC (controlled by design). Other possible sources of error: Nonlinearity: Quantisation levels may not be exactly equally spaced. Timing jitter: Nominal sampling period may vary by small amounts. Finite acquisition time: Signal may change while taking a sample. N.B. ADC errors cannot be corrected by later processing. D R Campbell
School of Computing
University of Paisley 12
6
Case Study PC Sound Card Architecture
D R Campbell
School of Computing
University of Paisley 13
Sound Card ADC configuration Audio in Left
Sampler
S&H
Audio in Right
D R Campbell
Sampler
ADC
Control
ADC
School of Computing
Parallel out
Control in/out
Parallel out
University of Paisley 14
7
Simple Sound Card Architecture 2
Mic in (~10 mV)
Stereo Amp
Line in (~0.5V)
2
Line out (~0.5V)
2
+
2
AAF
Stereo Amp
2
ADC
16
Buffer 16 Memory
Sample rate control
RF Spk out 2 (~5mW)
Analogue to Digital Convertor
Anti-Alias Filter
2
DAC
16
Buffer 16 Memory
Bus I/F
PC Bus
Reconstruction Digital to Analogue Filter Convertor
Midi I/F
Midi in/out D R Campbell
School of Computing
University of Paisley 15
Sound Card Sampling Rates Not all sound cards support a free choice of sampling frequency. Cheaper cards and motherboard sound chips, particularly those in notebook PCs, are often limited to the relatively standard set of sampling frequencies (Fs): Fs kHz
8
Bandwidth kHz 4
11.025 16 22.05 32 32.075 44.1 48 5.5
8
11
16 16
22
24
Quality “Telephone” “Radio” “CD” “DAT” D R Campbell
School of Computing
University of Paisley 16
8
Sound Card Word Length Most modern sound cards support a 16 bit word length coding of the quantised sample values. This allows representation of 216 (65536) different signal levels within the input voltage range of the card. For example, if the voltage range of a particular input connection is +/- 5V, then the range is 10V and a 16 bit system will have a quantisation step size of Q = 10/65536 V = 0.15 mV which is the smallest voltage difference which can be represented in this example system. D R Campbell
School of Computing
University of Paisley 17
Dynamic Range The ratio of the largest signal amplitude to the smallest, is known as the dynamic range. Since a 16 bit word length allows 216 (i.e. 65536) different signal levels the dynamic range (DR) is calculated as DR= 20log([Voltage range]/[Quantisation step size]) dB = 20log(216) dB = 96 dB The human ear has a dynamic range of > 120dB therefore even “CD quality” reproduction involves some compromise.
D R Campbell
School of Computing
University of Paisley 18
9
