Coded data generation or conversion – Analog to or from digital conversion – Increasing converter resolution
Reexamination Certificate
2000-10-07
2001-12-18
Young, Brian (Department: 2819)
Coded data generation or conversion
Analog to or from digital conversion
Increasing converter resolution
C341S155000
Reexamination Certificate
active
06331831
ABSTRACT:
BACKGROUND
This invention relates to Analog to Digital (A/D) conversion.
A/D conversion involves converting a time-continuous and amplitude-continuous signal into a sequence of time-discrete and amplitude-discrete samples and encoding those samples. Conventionally, the signal is sampled in time with a sampling clock having frequency 1/&tgr;, and uniformly quantized in amplitude with step size q. The accuracy of the conversion depends, of course, both on the resolution in time and the resolution in amplitude. For most practical applications, the signal to be converted is band-limited, and the sampling frequency is at or above the Nyquist rate. Therefore, the time discretization is reversible. The amplitude discretization, however, introduces an irreversible loss of information. This is typically referred to as quantization error.
The prior art has shown that the expected quantization error is
E
(
&LeftBracketingBar;
s
(
t
)
-
s
T
(
t
)
&RightBracketingBar;
2
)
=
1
12
q
2
τ
τ
N
(
1
)
where s(t) is the time and amplitude-continuous input signal, S
T
(t) is a time and amplitude-continuous signal developed from the time discrete and amplitude discrete signal, &tgr; is the actual sampling interval, and &tgr;
N
is the Nyquist sampling interval.
The
τ
τ
N
quotient in equation (1) suggests that the conversion accuracy can be improved beyond the precision of the quantizer by introducing oversampling. Because the costs involved in building high-precision amplitude quantizers are relatively high, modem techniques for high accuracy A/D conversion are based on oversampling. Apart from this economical and technological consideration, however, using oversampling to improve the conversion accuracy is dramatically inferior, in rate-distortion sense, to refining quantization. For a given precision of the quantizer, unless some entropy coding is used, the bit-rate of oversampled A/D conversion is inversely proportional to the sampling interval, R=O(1/&tgr;); hence, the mean squared error decays only inversely to the bit-rate, E(|s(t)-S
T
(t)|
2
)=O(1/R).
Only recently has it been demonstrated that the accuracy of oversampled A/D conversion is better than suggested by equation (1), and that even without entropy encoding the quantized samples can be efficiently represented so that an exponentially decaying rate-distortion characteristic can be attained. A deterministic analysis shows that if the quantization threshold crossings of a signal s(t) form a sequence of stable sampling for the space of square intergrable bandlimited signals to which s(t) belongs, and s
T
(t) is a square intergrable function, in that same bandlimited space, for which the oversampled A/D conversion produces the same digital sequence as for s(t), then the squared error between s(t) and s
r
(t) can be bounded as
&LeftBracketingBar;
s
(
t
)
-
s
T
(
t
)
&RightBracketingBar;
2
<
c
s
(
τ
τ
N
)
2
(
2
)
uniformly in time. Moreover, representing the information about the quantization threshold crossings requires only a logarithmic increase of the bit-rate as the sampling interval tends to zero. Hence using this quantization threshold crossings based representation the rate distortion characteristic of oversampled A/D conversion becomes
|S
(
t
)−
S
T
(
t
)|
2
<a
s
e
−aR
(3)
Where &agr; is a positive constant, and &agr;
s
is proportional to the factor c
s
in equation (2).
While these results provide a radically new perspective on oversampled A/D conversion, some important issues remained unanswered. The &tgr;
2
conversion accuracy is established under the assumption that the quantization threshold crossings of the input signal constitute a sequence of stable sampling in an appropriate class of bandlimited signals. Although there are bandlimited signals which have this property, there also exist bandlimited signals for which the threshold crossings are too sparse to ensure this conversion accuracy; giving a precise characterization of these two classes of bandlimited signals is very intricate. Another problem is that the stronger error bound in equation (2) is not uniform on sufficiently general compact sets for simple oversampled A/D conversion in its standard form. Note also that the result about O(&tgr;) accuracy with the linear reconstruction is valid only in a small range of discretization parameters q and &tgr;, and that the error does not tend to zero along with the sampling interval but rather reaches a floor level for some finite &tgr;. So, there are basically no sufficiently general results about the accuracy of simple oversampled A/D conversion. Moreover, no explicit algorithms for reconstructing a bandlimited signal with &tgr;
2
accuracy are known, and the feasibility of local reconstruction with this accuracy is not clear.
SUMMARY OF THE INVENTION
An advance in the art is achieved with an A/D conversion scheme that employs a dither signal. Specifically, the signal to be converted is augmented with a deterministic, piecewise continuous, dither signal such that, a signal that results from combining the dither signal with the signal to be converted, has a zero-crossing within each interval I
j
=(jT,(j+1)T), where j is an integer, T=&pgr;/&lgr;B, B is the bandwidth of the signal to be converted, and &lgr; is a constant that is greater than 1, and those zero crossings in adjacent intervals I
j
are always separated by some minimal distance, &egr;, for all intervals I
j
. Zero crossings of the combined signal are detected, and the instances where those crossings occur are encoded; one zero crossing for each interval.
REFERENCES:
patent: 6049299 (2000-04-01), Lunacek
patent: 6268814 (2001-07-01), Kolsrud
Cvetikovic et al, Single-Bit Oversamples A/D Conversion With Exponential Accuracy in the Bit-Rate, Data Compression Conference, DCC 2000, Snow Bird, UT, Mar. 2000, pp. 343-352.
Cvetkovic Zoran
Daubechies Ingrid
AT&T Corp.
Brendzel Henry T.
Young Brian
LandOfFree
A/D conversion through dithering does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with A/D conversion through dithering, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and A/D conversion through dithering will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2591591