Electrical computers and digital data processing systems: input/ – Input/output data processing – Input/output addressing
Reexamination Certificate
1999-03-29
2002-03-19
Smith, Matthew (Department: 2825)
Electrical computers and digital data processing systems: input/
Input/output data processing
Input/output addressing
Reexamination Certificate
active
06360283
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates generally to the design and fabrication of signal processing circuits. More particularly, the present invention relates to the development of a library of operators, along with a system and method for using the operators, to design and fabricate customized integrated circuits which implement digital signal processing in a delta sigma domain.
2. State of the Art
General Signal Processing
Those skilled in the art will appreciate that signal processing can be performed in an analog domain (analog signal processing) or in a sampled domain (digital signal processing). In the analog domain, signals are infinitely variable, while in the digital domain signal variations occur in discrete intervals, both in time and amplitude with the length of the intervals, being a function of the desired granularity.
Although analog signal processing is performed on input signals in their native domain, it is difficult to implement analog circuit components with high precision. That is, analog signal processing circuits involve the use of components which cannot be precisely matched. Further, analog circuits are susceptible to signal drift and to environmental noise. These circuits therefore require substantial overhead (e.g., filtering) to maintain signal integrity. Despite these drawbacks, analog signal processing is still used where, for example, real time processing is required.
For applications where the signal processing delays of digital signal processors can be tolerated, digital signal processing is used because of its non-susceptibility to drift and the influences of environmental noise, and because of its ability to provide high precision results. With conventional multibit digital signal processing, a multibit sample of an analog signal is obtained at each of a plurality of the discrete sampling instances. Each multibit value approximates the level which the analog signal possesses at that time.
One multibit technique for converting an analog signal into a sampled signal is known in the art as pulse code modulation (PCM). Here, a multibit binary number is used to represent the magnitude of the analog signal at each sampled interval. However, high resolution PCM digital signal processing can only be achieved with high circuit complexity and with relatively slow sequential digital processing.
Digital signal processing (DSP) systems such as PCM systems typically operate on an analog signal by using an analog-to-digital converter to produce the multibit values as an approximation of the analog signal. That is, the multibit levels constitute an approximation of magnitudes of the analog signal at discrete points in time. Any differences between the original analog signal and the sampled signal can be considered quantization noise which has been added to the signal (that is, quantization error).
PCM processing is performed by taking Nyquist sampling into account (i.e., sampling is performed at a minimum of twice the highest frequency of the analog signal) to ensure that the original analog signal can be reconstructed. When sampling at the Nyquist frequency, the quantization noise can only occupy space in the frequency domain at which the original signal exists. In the frequency domain, noise from the quantizer spreads from DC to one-half of the Nyquist rate as described, for example, in a document entitled “Principles of Sigma-Delta Modulation for Analog-to-Digital Converters” available from Motorola, Inc. (see, for example, page 8). However, because the band, within which the quantization noise is spread, corresponds to the band from which the original analog signal is to be retrieved, it is difficult to distinguish and retrieve the original signal. Because the number of levels of the analog-to-digital converter is proportional to the precision or accuracy with which the signal can be represented, such converters tend to be complex and expensive. Further, conventional digital signal processors may be inappropriate for real time applications since they are based on relatively slow sequential signal processing and the use of a centralized digital signal processor.
More recently, a delta sigma (also referred to as a “sigma delta”) approach has been used to implement analog-to-digital conversion. With a typical delta sigma converters, rather than going directly to a multilevel representation of an analog signal, an intermediate stage is used to implement oversampling. Such oversampling is performed at a rate much higher than the necessary Nyquist frequency, but with a reduced number of quantization levels.
The oversampling of a delta sigma approach spreads the quantization noise over the band from DC up to one-half of the oversample rate (which is relatively high in comparison to the Nyquist frequency). Because oversampling is performed at a relatively high frequency, the quantization noise is now spread over a band which extends to a limit much higher than the signal band of interest. Therefore, the amount of noise in the signal band of interest is reduced. Such a configuration permits a (digital) filter to easily separate the signal of interest, and is widely used for high resolution analog-to-digital converters. Conventional analog-to-digital converters typically include a digital decimation filter that subsequently passes the signal band and rejects the noise, thereby providing a multibit word output at the Nyquist rate (i.e., a rate sufficient to fully represent the analog signal).
A similar reverse technique can be used to provide digital-to-analog conversion. After further processing with multibit digital signal processing components, the output from the digital signal processor can be converted back to an analog signal by using a filter to interpolate between the Nyquist samples. The resultant signal which is now smooth is resampled at a much higher rate, adding quantization noise to produce a bit stream. This bit stream is then supplied to a continuous time analog filter to reconstruct the original analog signal.
The reduction of noise in the signal band of interest can be more significantly reduced by including the quantizer in a closed loop. A filter can be included in the loop to shape the quantization noise. If the forward gain of the path has high DC gain, it suppresses the quantization noise at the lower frequencies.
Having explained noise shaping with respect to a first order modulator wherein noise rises proportional to frequency, it will be appreciated that different order modulators can be used to adjust the noise shape. Higher order filters decrease the portion of the quantization noise present in a signal band of interest.
For example, with a second order modulator, for every doubling of the oversample ratio (the ratio of the oversample rate to the Nyquist frequency), 2.5 bits of enhanced resolution can be achieved. Thus, even if a single threshold quantizer which only gives a one bit result is used, oversampling at 1000 times greater than the Nyquist rate results in ten doublings and produces 25 bits of available resolution. The residual noise in the signal band would be represented by only one of the 25 bits or one part in 32 million. A document entitled “An Overview of Sigma-Delta-Converters” by Pervez M. Aziz et al (IEEE Signal Processing Magazine, January 1996), page 82, compares different resolutions achieved with different delta sigma modulators.
In sum, delta sigma modulation has been used as an intermediate step to producing multibit (PCM) digital signal processing, wherein the quantization noise of a bit stream is spread across a frequency band greater than the signal band of interest. However, because the delta sigma modulation is merely used as an intermediate step to multibit processing, the signal processing delays and the circuit complexities which result when changing from the filtering word lengths in the multibit domain are incurred.
Delta Sigma Processing
More recently, it has been realized that delta sigma signals can be processed in the bit stream domain, without conversio
Do Thuan
Duality Semiconductor, Inc.
Hunton & Williams
Smith Matthew
LandOfFree
Method and apparatus for producing signal processing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for producing signal processing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for producing signal processing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2884277