Beruflich Dokumente
Kultur Dokumente
One-bit quantization
Quantization noise shaping
A/D DSP
2002 B. Boser 1
Oversampling
Nyquist rate ADCs
Sample at fs around 2x bandwidth Resolution set by number of decision levels of quantizer
Oversampled ADCs
Sample at fs >> bandwidth (16 500x) Use few quantization levels (typical 1-Bit) Employ DSP to reduce quantization error
A/D DSP
2002 B. Boser 2
One-Bit Quantization
Lets examine some properties of one bit random sequences
Values in the sequence are constrained to be either +1 or 1
By picking +1s and 1s at random (using MATLABs rand.m random number generator), we generate a sequence with zero mean
The sequence values model outputs of a hypothetical 1-Bit, 1MHz ADC Nothing stops us from doing a DFT of this sequence
A/D DSP
2002 B. Boser 3
-30
-60
-90
100
200
300
400
500
Frequency (kHz)
A/D DSP
2002 B. Boser 4
One-Bit DFTs
For 1-Bit DFT plots, well use a different normalization scheme
The dBWN (dB White Noise) scale sets the 0dB line at the noise/bin of a random +1, -1 sequence From the energy theorem, N-1 N-1
n=0
1 an = N = N
2
Am = N
m=0
(+1)2 or (1)2 = 1
A/D DSP
2002 B. Boser 5
Am (dBWN)
-30
-60 -90
100
200
300
400
500
Frequency (kHz)
A/D DSP
2002 B. Boser 6
2002 B. Boser 7
-30
-60 -90
100
200
300
400
500
Frequency (kHz)
A/D DSP
2002 B. Boser 8
-30
-60 -90
100
200
300
400
500
Frequency (kHz)
A/D DSP
2002 B. Boser 9
A/D DSP
2002 B. Boser 10
fS 11 fS 2 1 fS 2 11
Am (dBWN)
-30
-60 -90
100
200
300
400
500
Frequency (kHz)
A/D DSP
2002 B. Boser 11
A quantization noise model is completely inappropriate for this type of sequence Lets deliberately increase the quantization error a little and attack its periodicity
A/D DSP
2002 B. Boser 12
3 Pattern Options
Generate another 1-Bit sequence by concatenation of the following sequences:
[1 +1 1 +1 1 +1 1 +1 +1] [1 +1 1 +1 1 +1 1 +1 1 +1 +1] [1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 +1]
Selection of the above 9, 11, and 13 term sequences at random yields an average of +1/11 and the following DFT
A/D DSP
2002 B. Boser 13
-30
-60
-90
100
200
300
400
500
Frequency (kHz)
A/D DSP
2002 B. Boser 14
3 Pattern Options
The randomized concatenation of longer patterns certainly breaks up periodic error
Some concentration of energy near fs/11 and its harmonics (especially 5fs/11) is still visible
Noise below 50kHz is significantly lower than that of the sample-by-sample random sequences of slide 9
The noise shaping obtained with 3 pattern options benefits low frequencies at the expense of increased quantization error at high frequencies
A/D DSP
2002 B. Boser 15
average of 30 spectra
Am (dBWN)
-30
-60
-90
100
200
300
400
500
2002 B. Boser 16
Frequency (kHz)
A/D DSP
5 Pattern Options
Lets add two more patterns to our set of pattern options:
[1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 1 +1 +1] [1 +1 1 +1 1 +1 +1]
Selection of 7, 9, 11, 13, and 15 term sequences at random preserves the +1/11 average
Think of this process as a sort of time-variant dither DFT follows
A/D DSP
2002 B. Boser 17
Am (dBWN)
-30
-60 -90
3 options 5 options
100 200 300 400 500
Frequency (kHz)
A/D DSP
2002 B. Boser 18
The model that quantization error is uncorrelated with the input signal becomes reasonable with only 5 pattern options
Such reasonableness is required for small-signal analysis of the sigma-delta modulator
A/D DSP
2002 B. Boser 19
Loop filter
A/D DSP
1b Quantizer (a comparator)
2002 B. Boser 20
Sigma-Delta Modulators
The loop filter H can be either a SC or continuous time SCs are easier to implement and scale with the clock rate Continuous time filters provide anti-aliasing protection Can be realized with passive LCs at very high frequencies
fs + _
vIN
H(z)
dOUT +1 or -1
A/D DSP
2002 B. Boser 21
A/D DSP
2002 B. Boser 22
DAC
dOUT +1 or -1
-/2 or +/2
Properties of the first-order modulator:
Analog input range is the dout times the DAC reference The average value of dOUT must equal the average value of v IN +1s (or 1s) density in dOUT is an inherently monotonic function of vIN linearity is not dependent on component matching Alternative multi-bit DAC (and ADCs) solutions reduce the quantization error but loose this inherent monotonicity
EECS 247 Lecture 19: Oversampling 2002 B. Boser 23
A/D DSP
Simulation
1 X z -1 1-z-1 Sine Wave Discrete Filter Comparator 2 Q 3 Y
1.5
X Q Y
0.5 Amplitude
-0.5
-1
see L19_level1
A/D DSP
2002 B. Boser 24
A/D DSP
2002 B. Boser 25
Looks familiar?
A/D DSP
2002 B. Boser 26
2002 B. Boser 27
2002 B. Boser 28
dOUT
A/D DSP
2002 B. Boser 29
q: = /2
0.4 0.5
A/D DSP
2002 B. Boser 30
First-order modulators are too prone to limit cycles to be of much practical use
They do provide the basis for higher-order s
A/D DSP
2002 B. Boser 31
vIN
+ _
z-1 1 z-1
dOUT
H(z) =
A/D DSP
2002 B. Boser 32
2002 B. Boser 33
q vIN + _
H(z)
dOUT
A/D DSP
2002 B. Boser 34
H(z)
dOUT
STF =
Dout GH ( z ) = Vin 1 + GH ( z )
NTF =
Q G = Vin 1 + GH ( z )
2002 B. Boser 35
H(z)
dOUT
STF =
Dout GH ( z ) = Vin 1 + GH ( z )
NTF =
Q G = Vin 1 + GH ( z )
2002 B. Boser 36
A/D DSP
2002 B. Boser 37
Integrated Noise
Just as we did for thermal noise, lets look at the integrated noise at the output of the modulator In the discrete time case, noise integrals are summations, but the result is called integrated noise nonetheless
A/D DSP
2002 B. Boser 38
Noise Summations
Integrated noise is computed from the energy theorem (N even):
arms =A0/ N , M=0 1 arms = N 1 arms = N
A/D DSP
A0 +
m=1
2Am
, 0<M<N/2
A0 + AN/2 +
N/2-1 m=1
2Am
, M=N/2
2002 B. Boser 39
Integrated Noise
30 Integrated Noise [dBV] 20 10 0 -10 -20 -30 -40 -50 -60 -70 -80 0 0.1 0.2 0.3 Frequency [ f/f ] s 0.4 0.5 Output Spectrum Integrated Noise
A/D DSP
Amplitude [ dBWN ] /
The total noise at the modulator output with no input sums to 0dB This is consistent with a binary signal
2002 B. Boser 40
Integrated Output
30 Integrated Noise [dBV] 20 10 0 -10 -20 -30 -40 -50 -60 -70 -80 0 0.1 0.2 0.3 Frequency [ f/f ] s 0.4 0.5 Output Spectrum Integrated Noise
The total power still sums to 0dB Only little quantization noise at low frequency
A/D DSP
Amplitude [ dBWN ] /
2002 B. Boser 41
ADCs which sample their inputs at much higher frequencies than the Nyquist rate minimum are called oversampling ADCs
A/D DSP
2002 B. Boser 42
A/D DSP
2002 B. Boser 43
H(z)
SQ ( f ) =
NTF (z ) = G 1 + GH ( z )
1 2 f s 12
B
SY =
S ( f ) NTF (z )
Q
2 z = e2 jfT
df
A/D DSP
2002 B. Boser 44
NTF (z ) = 1 z
SY =
B
fs
S ( f ) NTF (z )
Q
2 z = e2 jfT
df
fs
1 2 (2 sin fT )2 df f s 12
= 1 z 1 z + 1
1 2 3 M 3 12
2
2002 B. Boser 45
M
16 32 1024
DR
33 dB 42 dB 87 dB
2 1 2 3 M 3 12 9 DR = M3 2 2
A/D DSP
2002 B. Boser 46
Dynamic Range
DR increases 9dB for each doubling of M 1st order modulators require very high M for >10-Bit resolution
higher order filters improve this tradeoff substantially
Next time well design an oversampled audio ADC with better than 16-Bit resolution
A/D DSP
2002 B. Boser 47