Coded data generation or conversion – Analog to or from digital conversion
Reexamination Certificate
1999-11-30
2001-05-15
Young, Brian (Department: 2819)
Coded data generation or conversion
Analog to or from digital conversion
C341S155000, C341S144000
Reexamination Certificate
active
06232899
ABSTRACT:
This invention relates to digital to analogue convertors, and to analogue to digital convertors.
In one type of digital to analogue convertor, a multi-bit digital signal is employed to control the width of pulses output by a pulse width modulator. The pulses are then integrated or otherwise low pass filtered over a period of time to provide analogue output levels. The pulse rate of the pulse width modulator, and hence the rate at which digital samples are supplied thereto, clearly needs to be well above that of the pass band of the low pass filter for this to be successful, however; since the pass band at the low pass filter is half the sampling frequency for the signal to be converted, the rate at which digital samples are supplied is therefore generally well in excess of the sampling frequency for the signal to be converted.
Another type of digital to analogue convertor, described for example in GB2183115, employs oversampling—that is, receives digital data samples at a significant multiple of the Nyquist frequency for the signal represented. The accuracy of the signal is then reduced by a quantizer (or more accurately, a re-quantizer) having a small number of levels. The output of the quantizer is fed back and subtracted from the input to the quantizer, and the result (which represents the error due to the quantizer, hereinafter referred to as the quantizer noise) is subtracted from the next sample input to the quantizer.
If the input signal level is constant (that is, the input signal contains no frequencies above zero hertz or DC) the quantizer error for one sample is thus taken account of to some extent in the following sample and if the quantizer output is averaged over a sufficiently large number of samples, the quantizer noise is eliminated. Reducing the number of quantization levels increases the magnitude of the quantization noise and consequently increases the number of samples over which the quantizer output must be averaged; and hence the sampling rate of the convertor (and its oversampling ratio).
However, the quantizer noise increases as the frequency of the input signal rises, leading to errors at higher frequencies in the output of the convertor. To reduce this effect, a filter may be placed in the path so as to filter the quantizer noise before it is subtracted from the next sample. The filter has a prediction characteristic so that when the path is added the spectral distribution of the quantizer noise is “shaped”, to reduce the noise level at frequencies below the Nyquist frequency of the signal to be converted, and consequently increase the noise at higher frequencies outside the signal band.
Alternatively, a similar effect may be achieved by placing a different low pass filter in the signal path at a point following that at which the quantizer noise has been introduced by an overall feedback path from the quantizer output.
The quantizer may, as described in GB2183115, have only two levels (in other words, it may be a one bit quantizer). In this case the output of the quantizer may (after suitable buffering or amplification) be integrated or otherwise low pass filtered using an analogue filter to provide an analogue signal corresponding to the digital input signal. Alternative types of convertor employ a quantizer having a greater, but still small, number of levels and consequently producing output signals comprising a small number of bits (for example, 3). However, since the output of such quantizers remains a multi-bit digital signal some form of conversion to an analogue signal is still required. It has therefore been proposed to employ, following the quantizer, a pulse width modulator controlled by the digital output of the quantizer to produce a pulse of constant height but of length determined by the quantizer output. This is advantageous because, whilst control of the precise height of a pulse requires high precision analogue circuits, control of the pulse lengths (with constant height) requires only a single amplitude source and an accurate timer, both of which are commercially available.
A digital to analogue convertor of this general type is disclosed in “Seventeen bit oversampling D to A conversion technology using multi stage noise shaping”, Matsuya et al, IEEE Journal of Solid State Circuits VOL 24 No. 4 August, 1989. Although that reference shows the use of the so-called MASH or multiple stage noise shaping structure, in which the quantization error is itself further re-quantized, a structure employing only a single quantizer with a noise shaping filter is equally possible.
The widespread use of digital audio technology, caused by the availability of low cost digital storage devices such as digital audio tape and compact disc, has led to a requirement for greater accuracy in digital to analogue and analogue to digital conversion. Seventeen and eighteen bit digital to analogue convertors utilizing oversampling and noise shaping and having a band width around 20 kilohertz are already known; the above referenced IEEE paper claims such a performance. A summary of known devices is given in HI-FI CHOICE, DECEMBER 1990, P54-59, “Keeping in Shape” (P Miller). However, there is evidence that the human ear can be responsive to quantization errors even using eighteen bits.
It is stated in the above referenced IEEE paper that resistance mismatching of P and N channel MOS devices causes second order harmonic distortion in the output signal. Accordingly, a differential or “push-pull” output structure is adopted in which the output of the quantizer controls two separate pulse width modulators; one producing longer pulses for higher signal levels, and the other producing shorter pulses for higher signal levels. The outputs of the two pulse width modulators are then fed to a differential amplifier which consequently produces pulses of lengths proportional to the quantizer output with reduced second order harmonic distortion.
We have realized, however, that there is another cause of distortion which is directly due to the use of pulse width modulation itself; the non-linear distortion due to the pulse width modulator can bring the quantizer noise back into the audio band. At resolutions of seventeen or eighteen bits, and employing the differential output configuration above, the effects of such distortion are not noticeable and hence, it is believed, have not been recognized, but at accuracies of 22 or 24 bits the inaccuracy inherent in the pulse width modulation stage would limit the performance of the convertor as a whole.
In “Multibit oversampled &Sgr;-&Dgr; A/D convertor with digital error correction”, Larson et al, Electronics Letters, Aug. 4, 1988, pages 1051-1052, and “Digitally corrected multi-bit &Sgr;-&Dgr; data convertors”, by the same authors in Proc 1989 IEEE Int Symp on Circuits and Systems (1989, pages 467-650), there is disclosed a method of correcting for non-linearity in a sampled system by providing a corresponding non-linearity in a feedback loop at a stage prior to the non-linear stage; the feedback non-linearity is provided as a ROM look up table. However, it is acknowledged that this technique is only suitable for errors which occur at sampling instants in the sampled system, and not for errors such as “settling errors” or random noise.
GB 2176070 provides a pulse width type modulator in which the edges of the pulses are allowed to have stepped amplitudes so as to reduce the non-linear distortion produced by the modulation. This would, of course, require several high accuracy reference amplitude sources. Further, it does not solve the problem of the reintroduction of quantizer noise into the audio band due to the remaining non-linearity.
When pulse width or like modulations are used, the pulse edges lie at variable points between the sampling instants of the sampled system, and so do the corresponding errors, which consequently cannot be corrected by the above disclosed method. In this application, the term “Pulse Edge Modulator” will be used to describe modulations effected by moving the edge of a pulse in time in this way, and encomp
Cirrus Logic Inc.
Nixon & Vanderhye P.C.
Young Brian
LandOfFree
Analogue and digital convertors does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Analogue and digital convertors, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Analogue and digital convertors will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2516734