course 10 example application of random signals - oversampling and noise shaping
DESCRIPTION
about quantizationTRANSCRIPT
![Page 1: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/1.jpg)
Review of Quantization Error
8-bit quantization error
In a heuristic sense, the assumptions of the statistical model appear to be valid if the signal is sufficiently complex and the quantization steps are sufficiently small, so that the amplitude of the signal is likely to traverse many quantization steps from sample to sample.
![Page 2: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/2.jpg)
Review: Assumptions about e[n]
e[n] is a sample sequence of a stationary random process. e[n] is uncorrelated with the sequence x[n]. The random variables of the error process e[n] are
uncorrelated; i.e., the error is a white-noise process. The probability distribution of the error process is uniform
over the range of quantization error (i.e., without being clipped).
The assumptions would not be justified. However, when the signal is a complicated signal (such as speech or music), the assumptions are more realistic. Experiments have shown that, as the signal becomes more
complicated, the measured correlation between the signal and the quantization error decreases, and the error also becomes uncorrelated.
![Page 3: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/3.jpg)
Review: Quantization error analysis
2/][2/ ∆≤<∆− ne e[n] is a white noise sequence. The probability density
function of e[n] is a uniform distribution
![Page 4: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/4.jpg)
Review: quantization error vs. quantization bits
The mean value of e[n] is zero, and its variance is
Since For a (B+1)-bit quantizer with full-scale value Xm, the noise
variance, or power, is
121 22/
2/
22 ∆=
∆= ∫
∆
∆−
deeeσ
BmX
2=∆
122 22
2 mB
eX−
=σ
![Page 5: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/5.jpg)
Review: Quantization error analysis
A common measure of the amount of degradation of a signal by additive noise is the signal-to-noise ratio (SNR), defined as the ratio of signal variance (power) to noise variance. Expressed in decibels (dB), the SNR of a (B+1)-bit quantizer is
Hence, the SNR increases approximately 6dB for each bit added to the word length of the quantized samples.
−+=
⋅=
=
x
m
m
xB
e
x
XB
XSNR
σ
σσσ
10
2
22
102
2
10
log208.1002.6
212log10log10
![Page 6: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/6.jpg)
We consider the analog signal xa(t) as zero-mean, wide-sense-stationary, random process with power-spectral density denoted by and the autocorrelation function by .
To simplify our discussion, assume that xa(t) is already bandlimited to ΩN, i.e.,
Oversampling vs. quantization
)( jwxx e
aaφ
)(τφaaxx
,0)( =Ωjaaxxφ ,|| NΩ≥Ω
Ωj
![Page 7: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/7.jpg)
We assume that 2π/T = 2MΩN. M is an integer, called the oversampling ratio.
Oversampling
Oversampled A/D conversion with simple quantization and down-sampling
Decimation with ideal low-pass filter
![Page 8: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/8.jpg)
Using the additive noise model, the system can be replaced by
Its output xd[n] has two components, one due to the signal input xa(t) and one due to the quantization noise input e[n].
Denote these components by xda[n] and xde[n], respectively.
Additive noise model
Decimation with ideal low-pass filter
![Page 9: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/9.jpg)
Goal: determine the ratio of signal power εxda2 to
the quantization-noise power εxde2, ε. denotes
the expectation value. As xa(t) is converted into x[n], and then xda[n], we
focus on the power of x[n] first. Let us analysis this in the time domain. Denote φxx[n] and φxx[ejw] be the autocorrelation and power spectral density of x[n], respectively.
By definition, φxx[m]= εx[n+m]x[n].
Signal component (assume e[n]=0)
![Page 10: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/10.jpg)
Power of x[n] (assume e[n]=0)
Since x[n]= xa(nT), it is easy to see that
That is, the autocorrelation function of the sequence of samples is a sampled version of the autocorrelation function.
The wide-sense-stationary assumption implies that εxa2(t)
is a constant independent of t. It then follows that
for all n or t.
][][][ nxmnxmxx += εφ)())(( nTxTmnx aa += ε
)(mTaaxxφ=
)()(][ 222 txnTxnx aa εεε ==
![Page 11: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/11.jpg)
Power of xda[n] (assume e[n]=0)
Since the decimation filter is an ideal lowpass filter with cutoff frequency wc = π/M, the signal x[n] passes unaltered through the filter.
Therefore, the downsampled signal component at the output, xda[n]=x[nM]=xa(nMT), also has the same power.
In sum, the above analyses show that
which shows that the power of the signal component stays the same as it traverse the entire system from the input xa(t) to the corresponding output component xda[n].
)(][][ 222 txnxnx ada εεε ==
![Page 12: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/12.jpg)
Power of the noise component
According to previous studies, let us assume that e[n] is a wide-sense-stationary white-noise process with zero mean and variance
Consequently, the autocorrelation function and power
density spectrum for e[n] are,
The power spectral density is the DTFT of the autocorrelation function. So,
12
22 ∆=eσ
][][ 2 mm eee δσφ =
white noise
ππσφ <<−= we ejw
ee ,)( 2
![Page 13: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/13.jpg)
Power of the noise component (assume xa(t)=0)
Although we have shown that the power in either x[n] or e[n] does not depend on M, we will show that the noise component xde[n] does not keep the same noise power. It is because that, as the oversampling ratio M increases, less of the
quantization noise spectrum overlaps with the signal spectrum, as shown below.
![Page 14: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/14.jpg)
Illustration of frequency and amplitude scaling
Since oversampled by M, the power spectrum of xa(t) and x[n] in the frequency domain are illustrated as follows.
![Page 15: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/15.jpg)
Illustration of frequency for noise
By considering both the signal and the quantization noise, the power spectra of x[n] and e[n] in the frequency domain are illustrated as
![Page 16: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/16.jpg)
Noise component power
Then, by ideal low pass with cutoff wc=π/M in the decimation, the noise power at the output becomes
Mdwne e
M
M e
2/
/
22
21][ σσπ
επ
π== ∫−
![Page 17: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/17.jpg)
Powers after downsampling
Next, the lowpass filtered signal is downsampled, and as we have seen, the signal power remains the same. Hence, the power spectrum of xda[n] and xde[n] in the frequency domain are illustrated as follows:
![Page 18: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/18.jpg)
Noise power reduction
Conclusion: Thus, the quantization-noise power εxde2[n]
has been reduced by a factor of M through the filtering and downsampling, while the signal poweer has remained the same.
For a given quantization noise power, there is a clear tradeoff between the oversampling factor M and the quantization step ∆.
MMdw
Mx ee
de 1221
2222 ∆
=== ∫−σσ
πε
π
π
![Page 19: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/19.jpg)
Oversampling for noise power reduction
Remember that
Therefore
The above equation shows that for a fixed quantizer, the noise power can be decreased by increasing the oversampling ratio M.
Since the signal power is independent of M, increasing M will increase the signal-to-quantization-noise ratio.
BmX
2=∆
22 )2
(12
1 Bm
deX
Mx =ε
![Page 20: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/20.jpg)
Tradeoff between oversampling and quantization bits
Alternatively, for a fixed quantization noise power,
the required value for B is From the equation, every doubling of the oversampling ratio
M, we need ½ bit less to achieve a given signal-to-quantization-noise ratio.
In other words, if we oversample by a factor M=4, we need one less bit to achieve a desired accuracy in representing the signal.
22 )2
(12
1 Bm
dedeX
MxP == ε
mde XPMB 2222 loglog2112log
21log
21
+−−−=
![Page 21: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/21.jpg)
Previously, we have shown that oversampling and decimation can improve the signal-to-quantization-noise ratio.
The result is remarkable, but if we want to make a significant reduction, we need very large sampling ratios. Eg., to reduce the number of bits from 16 to 12 would
require M=44=256. The basic concept in noise shaping is to modify the A/D
conversion procedure so that the power density spectrum of the quantization noise is no longer uniform.
Further reduction of quantization error Oversampling and Noise Shaping
![Page 22: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/22.jpg)
The noise-shaping quantizer, generally referred to as a sampled-data Delta-Sigma modulator, is roughly shown as the following figures. Analog form
Delta-Sigma Modulator
![Page 23: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/23.jpg)
Oversampled Quantizer with Noise Shaping
Accumulator (like an integrator): Sigma
Can be represented by the discrete-time equivalent system as follows: Discrete-time form
Minus the delayed feedback: delta
![Page 24: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/24.jpg)
Modeling the quantization error
As before, we model the quantization error as an additive noise source.
Hence, the above figure can be replaced by the following linear model:
additive noise
![Page 25: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/25.jpg)
This linear system has two inputs, x[n] and y[n]. According to the linearity, we can get the output y[n] by
1. set x[n]=0, find the output y[n] w.r.t. e[n] 2. set e[n]=0, find the output y[n] w.r.t. x[n] 3. add the above two outputs.
Output of a linear system
![Page 26: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/26.jpg)
Consider the output in the z-domain. We denote the transfer function from x[n] to y[n] as Hx(z) and from e[n] to y[n] as He(z).
When e[n]=0:
Transfer functions
X(z) X(z)-Z-1Y(z)
(X(z)-Z-1Y(z))/(1-z-1)
Z-1Y(z)
0
![Page 27: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/27.jpg)
Output when e[n]=0
1
1
1)(][][ −
−
−−
=z
zYzzXzY
)(][][][ 11 zYzzXzYzzY −− −=−
We have
So
That is when E[z] is zero.
][][ zXzY =
![Page 28: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/28.jpg)
When x[n]=0:
Transfer functions when x[n]=0
0 -Z-1Y(z)
E(Z)
Z-1Y(z)
-Z-1Y(z)/(1-Z-1)
![Page 29: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/29.jpg)
Output when x[n]=0
1
1
1)()(][ −
−
−−=
zzYzzezY
)(][][][][ 111 zYzzEzzEzYzzY −−− −−=−
We have
So
That is when X[z] is zero.
][)1(][ 1 zEzzY −−=
![Page 30: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/30.jpg)
In fact, feedback systems have been widely used (serves as a fundamental architecture) in control engineering.
Generally:
Formula:
Remark: feedback system
G
E Y X
H
;)()(1
)()()(
zHzGzG
zXzY
+=
)()(11
)()(
zHzGzXzE
+=
![Page 31: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/31.jpg)
From the feedback system formula, we can also obtain that
Another way of derivation
1
11
11
)()(1)(
)()()(
1
1
1=
−+
−=+
==
−
−
−
ZZZ
zHzGzG
zXzYzH x
1
1
1 1
11
1)()(1
1)()()( −
−
− −=
−+
=+
== Z
ZZzHzGzE
zYzHe
![Page 32: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/32.jpg)
Hence, in the time domain, we have
Therefore, the output y[n] can be represented equivalently as
The quantization noise e[n] has been modified as
Time domain relation
][][ nxnyx =
]1[][][ˆ][ −−== nenenenye
][ˆ][][][][ nenxnynyny ex +=+=
][ˆ ne
![Page 33: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/33.jpg)
To show the reduction of the quantization noise, let’s consider the power spectral density of
Since we have the input-output relationship between e[n] and as
In the frequency domain, we have
Power spectral density of the modified noise
][ˆ ne
][ˆ ne
)()1()( 1 zEZzYe−−=
)()1()()(ˆ jwjwjwe
jw eEeeYeE −−==
![Page 34: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/34.jpg)
Equivalent system
![Page 35: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/35.jpg)
The power-spectral-density relation between the modified and original quantization noises is thus
Power spectral density of the modified noise
)(||1||)( 2ˆˆ
jwee
jwjwee eee φφ −−=
22
22
22
22
))2/sin(2(
))cos(22())(2(
)11()1)(1(
||1||
e
eejwjw
ejwjw
ejwjw
ejw
wweeeeee
e
σ
σσ
σσ
σ
=
−=+−=
+−−=−−=
−=
−−
−−−
−
p.s.d. of the modified noise p.s.d. of the original noise
Model the original quantization error as white noise with this variance
![Page 36: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/36.jpg)
Remember that the downsampler does not remove any of the signal power, the signal power in xda[n] is
The quantization-noise power in the final output is See the following illustration for its computation
Quantization-noise power
∫−=π
πφ
π)(
21 jw
xxde ePdede
)(][][ 222 txnxnxP adada εεε ===
![Page 37: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/37.jpg)
Before decimation Power spectral density of modified noise
The modified noise density is non-uniform and lower in the effective band region
![Page 38: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/38.jpg)
After decimation
Down-scaled by M and also stretched by M
Downscaled by M
![Page 39: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/39.jpg)
Hence, the quantization-noise power in the final output is
Assume that M is sufficiently large, we can approximate that
Quantization-noise power
∫−=π
πφ
π)(
21 jw
xxde ePdede
Mw
Mw
2)
2sin( ≈=
∫−∆
=π
ππdw
MM2
2
))2
2sin(2(122
1
![Page 40: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/40.jpg)
With this approximation,
For a (B+1)-bit quantizer and maximum input signal level
between plus and minus Xm, ∆= Xm/2B. To achieve a given quantization-noise power Pde, we have
We see that, whereas with direct quantization a doubling of the oversampling ratio M gained ½ bit in quantization, the use of noise shaping results in a gain of 1.5 bits.
Bits and quantization tradeoff in noise shaping
3
22
361
MPde
π∆=
mde XPMB 2222 loglog21)6/(log
21log
23
+−+−= π
![Page 41: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/41.jpg)
![Page 42: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/42.jpg)
The noise-shaping strategy can be extended by incorporating a second stage of accumulation, as shown in the following:
Second-order noise shaping
![Page 43: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/43.jpg)
In the two-stage case, it can be derived that
In general, if we extend the case to p-stages, the corresponding noise shaping is given by
Second-order (i.e., 2-stage) noise shaping
42ˆˆ ))2/sin(2()( we e
jwee σφ =
21)1()( −−= zzHe
pe
jwee we 22ˆˆ ))2/sin(2()( σφ =
![Page 44: Course 10 example application of random signals - oversampling and noise shaping](https://reader033.vdocuments.us/reader033/viewer/2022060110/555cdfa0d8b42aeb2c8b56c8/html5/thumbnails/44.jpg)
By evaluation, with p=2 and M=64, we obtain almost 13 bits of increase in accuracy, suggesting that a 1-bit quantizer could achieve about 14-bit accuracy at the output of the decimator.
Although multiple feedback loops promise greatly increased quantization-noise reduction, they are not without problems. Specifically, for large values of p, there is an increased potential for instability and oscillations to occur.