Classifications

• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
  • H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
  • H03M7/3002—Conversion to or from differential modulation
  • H03M7/3004—Digital delta-sigma modulation
  • H03M7/3006—Compensating for, or preventing of, undesired influence of physical parameters
  • H03M7/3011—Compensating for, or preventing of, undesired influence of physical parameters of non-linear distortion, e.g. by temporarily adapting the operation upon detection of instability conditions

Definitions

• the present invention relates to signal processing and more particularly relates to look- ahead delta-sigma modulators having quantization error based output selection to minimize quantization noise.
• FIG. 1 depicts a prior art signal processing system 100 having a look-ahead delta- sigma modulator 102.
• the signal source 102 provides an input signal to pre-processing components 104.
• Preprocessing components include an analog-to-digital converter ("ADC") and oversampling components to generate a k-bit, digital input signal x(n).
• ADC analog-to-digital converter
• x(n) generally represents a signal sampled at 44.1 kHz times an oversampling ratio, such as 64:1.
• Look-ahead modulator 106 quantizes input signal x(n) and shapes the quantization noise so that most of the quantization noise is moved out of the signal band of interest, e.g. approximately 0-20 kHz for audio applications.
• Each output signal y(n) (also referred to herein as an "output value”) generally has one of two values selected from the set ⁇ + ⁇ /2, - ⁇ /2 ⁇ with " ⁇ " representing the full swing of y(n). (For convenience, ⁇ /2 will be represented as +1, and - ⁇ /2 will be represented as - 1.).
• the output signal y(n) can be processed further and, for example, used to drive an audio sound system or can be recorded directly onto a storage medium.
• Figure 2 depicts a schematic representation of prior art look-ahead delta-sigma modulator 106 with a look-ahead depth of M.
• Table 1 describes an embodiment of the symbols used in Figure 2.
• the look-ahead depth M refers to the dimension of each delayed output candidate vector Yoi used to determine output signal y(n).
• a negative delayed output candidate vector - Y D i ⁇ ⁇ 0, 1 ,2, ... , N- 1 ⁇ and the input vector X t are inputs to noise shaping filter 202(i).
• each of the N delayed output candidate vectors contains a unique set of elements.
• Each noise-shaping filter 202(f) of look-ahead delta-sigma modulator 106 uses a common set of filter state variables for time t during the calculations of respective cost value vectors Cj.
• Filter 202 maintains the actual filter state variables used during the calculation of each y(n). The state variables are updated with the selected y(n) output value. Loop filter 202 processes X; and -Y; to produce an error value, which in this embodiment is referred to as cost value vector Cj. Cost value vector Cj , and, thus, each element of cost value vector Cj is a frequency weighted error value. In some embodiments of look-ahead delta-sigma modulator 106, input signal vector x and delayed output candidate vectors Y D , are also used as direct inputs to filter 202(i)
• Quantizer error and output generator 203 includes two modules to determine y(n).
• the cost function minimum search module 204 computes the cost value power, (2) , of each cost value vector Cj in accordance with Equation 1, and determines the minimum cost value power at time t.
• t M c i - ⁇ [ c t P Equation 1.
• t l
• the cost function minimum search module 204 of quantizer 203 attempts to minimize the energy out of loop filter 202. Minimizing the energy out of loop filter 202 effectively drives the input to a small value, which effectively results in a relatively high loop gain for look-ahead delta-sigma modulator 106 and, thus, modifies the noise shapping transfer function in an undesirable way.
• the y(n) selector module 206 selects y(n) as the leading bit of Yj where Cj (2) m j n represents the minimum cost value power.
• the second primary thread of look-ahead modulator research involves pulse width modulation ("PWM”) amplifiers based on delta-sigma modulators combined with digital PWM modulation stages.
• PWM pulse width modulation
• the principal researchers have been Peter Craven and John L. Melanson.
• U.S. Patent No. 5,784,017 entitled “Analogue and Digital Converters Using Pulse Edge Modulations with Non-Linear Correction” inventor Peter Craven (“Craven”), which is incorporated herein by reference in its entirety, Craven described the use of look-ahead in delta- sigma modulators.
• the purpose of Craven was to ensure stability in alternating edge modulation, an inherently difficult modulation mode to stabilize.
• the delta-sigma modulator is operating at a low oversampling ratio (typically 4-16), and quantization noise is a special problem.
• Figure 3A depicts a model of a models filter 202 as a composite of two (2) transfer functions H ⁇ (z) and H 2 (z).
• the feedback of output candidate vector Y DJ introduces quantization noise 302 into the feedback loop of look-ahead modulator 106 but not into the input of look- ahead modulator 106.
• loop filter 202 can be modeled as having two separate transfer functions, Hi (z) and H 2 (z).
• the noise transfer function (“NTF”) equals l/[l+z -1 *H 2 (z)].
• the signal transfer function (“STF”) equals H ⁇ (z)/[H-z _1 *H 2 (z)].
• HI and H2 are identical. In the general case, HI and H2 differ. The choice of implementation affects mainly the STF. For most purposes, the NTF is the critical design criteria, making the choice of HI less critical.
• Figure 3B represents a typical fifth (5 th ) order noise-shaping filter 300 with input x(n) and feedback y(n).
• Filter 300 represent one embodiment of filter 202 and can be applied to non- look-ahead and look-ahead delta-sigma modulators.
• the scale factors k and gain factors g are design choices.
• the ki scale factor provides a unity gain for the first integrator stage 302. Equation 2 represents the transfer function H ⁇ (z):
• Equation 3 represents the transfer function H 2 (z):
• Figures 3C and 3D depict the respective NTF and STF pole-zero plots of filter 300.
• the NTF is a high-pass function with multiple zeros in the signal band
• the STF is an all-pole low-pass function.
• Figure 4 depicts the transfer function of H 2 (z) when the NTF is set to be a seventh (7 th ) order Butterworth filter with added NTF zeros.
• the transfer function H 2 (z) 402 does not maintain a smooth transition as 0 dB is approached. Rather a spike 404 occurs, which represents noise. This noise will be in the near out-of-band frequency region, which is undesirable as it makes reproduction more difficult.
• a look-ahead delta-sigma modulator includes a digital filter to filter data derived from input signal data and respective elements of delayed output candidate vectors to generate filter output vectors, wherein the digital filter includes state variables that are updated via feedback of a selected output value.
• the look-ahead delta-sigma modulator further includes a quantization error generator, coupled to the digital filter to receive the filter output vectors, to determine a set of quantization error vectors from each set of M element modulator output candidate vectors and the filter output vectors, wherein M is greater than one and each element in the output candidate vectors is a potential output value of the delta- sigma modulator.
• the look-ahead delta-sigma modulator also includes an output generator to select from each set of quantization error vectors a quantization error vector associated with a modulator output candidate vector and to select an output from the associated modulator output candidate vector.
• a method of determining output values of a delta-sigma modulator using quantization error vectors includes filtering data derived from input signal data and respective elements of M-element delayed output candidate vectors to generate filter output vectors. The method further includes generating quantization error vectors for each set of M element modulator output candidate vectors and each set of M element filter output vectors, wherein M is greater than one and each element in the output candidate vectors is a potential output value of the delta-sigma modulator. The method also includes selecting from each set of quantization error vectors a quantization error vector associated with one of the modulator output candidate vectors and generating an output from the associated modulator output candidate vector.
• a signal processing system includes an M-depth delta-sigma modulator delta-sigma modulator to determine output values from respective sets of M element modulator output candidate vectors using quantization error vectors, wherein M is greater than one, each element in the output candidate vectors is a potential output value of the delta-sigma modulator.
• a method of determining an output signal using an M-depth delta-sigma modulator and quantization error vectors, wherein M is greater than one and each element in the output candidate vectors is a potential output value of the delta-sigma modulator includes processing an input signal vector and a set of delayed output candidate vectors. For each processed input signal vector and delayed output candidate vector, the method further includes computing a quantization error vector from the processed input signal vector and delayed output candidate vector and processing each quantization error vector to identify the output candidate vector that best matches the input signal vector. The method also includes selecting an output from the output candidate vector that best matches the input signal vector.
• a method of processing a signal using a delta-sigma modulator includes determining output values of an M-depth delta-sigma modulator from respective sets of M element modulator output candidate vectors and an M element modulator input vector using quantization error vectors, wherein M is greater than one, each element in the output candidate vectors is a potential output value of the delta-sigma modulator.
• Figure 1 depicts a signal processing system with a conventional look-ahead delta-sigma modulator.
• Figure 2 depicts example data used by the look-ahead delta-sigma modulator of Figure 1 to determine an output signal.
• Figure 3 A depicts a model of the look-ahead modulator of Figure 1 that includes a model of a noise shaping filter.
• Figure 3B represents a noise shaping filter of Figure 3A.
• Figure 3C represents a pole-zero plot of a noise transfer function of the filter of Figure 3B.
• Figure 3D represents a pole-zero plot of a signal transfer function of the filter of Figure 3B.
• Figure 4 depicts an example loop filter transfer function of the look-ahead modulator of Figure 1.
• Figure 5 depicts a signal processing system having a look-ahead delta-sigma modulator with quantizer input values representing quantization error.
• Figure 6 depicts a model of the look-ahead modulator of Figure 5.
• Figure 7 depicts an example loop filter transfer function of the look-ahead modulator of Figure 5.
• Figure 8 depicts a signal processing system having a look-ahead delta-sigma modulator with weighted quantization error vectors.
• Figures 9A-9F depict weighting windows with a downward weighting trend.
• Figure 10 depicts a signal processing system that includes a look-ahead modulator of Figure 5, an output device and process, and an output medium.
• Figure 11 depicts post-processing operations in an embodiment of the signal processing system of Figure 8.
• Equation 1 It has been proposed in conventional publications that the power output of a delta-sigma noise-shaping filter is a good metric for optimization, i.e. see Equation 1.
• Equation 1 has at least two disadvantages: (1) Equation 1 is non-monotonic in magnitude frequency response and (2) Equation 1 is not appropriate for more than two quantization levels.
• Equation 1 is non-monotonic in magnitude frequency response and (2) Equation 1 is not appropriate for more than two quantization levels.
• the approach of Equation 1 with look- ahead delta-sigma modulators can decrease the signal to noise ratio relative to conventional delta-sigma modulators.
• the signal processing systems described herein include a look-ahead delta-sigma modulator that processes multiple output candidate vectors and an input vector to determine a quantization error vector for each output candidate vector.
• the quantization error vector represents a difference between a cost value vector and an input candidate vector.
• Look-ahead delta-sigma modulator output values are selected using the quantization error vectors by, for example, determining the minimum power quantization error vector for each input vector X and selecting the output value from the input candidate vector associated with the minimum power quantization error vector.
• Quantization error vectors can also be weighted using a non-uniform weighting vector.
• the look-ahead delta-sigma modulators of the signal processing systems described herein include a loop filter with a noise transfer function (“NTF”) and a signal transfer function (“STF").
• NTF noise transfer function
• STF signal transfer function
• the filter is designed around optimizing the NTF quantity l/[l+z _1 *H 2 (z)] ("H 2 (z) is illustrated in Figure 6).
• Optimizing "l/[l+z "1 *H 2 (z)]” is effectively the same as minimizing z + H 2 (z).
• Minimizing z + H 2 (z) effectively minimizes a difference between the input and output values of the look-ahead delta-sigma modulator quantizer.
• the difference between the input and output values of the quantizer represent quantization error.
• the look-ahead delta-sigma modulators of the signal processing system described herein selects an output based on quantization error values.
• the look-ahead delta-sigma modulator output value is selected from Yj, where [Cj - Yi] (2) m j n represents the minimum quantization error power.
• Figure 5 depicts one embodiment of a digital signal processing system that includes a look-ahead delta-sigma modulator for selecting output values using quantization error.
• "Best" can be defined as closest matching in the signal band of interest.
• “Closest matching” can be defined, for example, in a power sense (lowest distance), in a minimum/maximum sense, in a psycho- acoustically weighted sense, other desired measure.
• the "best" output signal pattern Y[n] is the pattern Y[n] such that [H(X-Y D )- Y] has the lowest power.
• a "signal band of interest” is, for example, a frequency band containing a signal with data of interest. For example, an audio signal band of interest is approximately 0 Hz to 25 kHz.
• look-ahead delta-sigma modulator 500 represents an embodiment of a look-ahead delta-sigma modulator that uses quantization error to determine output values.
• look-ahead delta-sigma modulator 500 can be reduced using pruning or other computation reduction methods. Pruning techniques include eliminating or reducing processing of redundant cost value vectors and eliminating or reducing redundant arithmetic calculations.
• Look-ahead delta-sigma modulator 500 performs noise shaping on the input data, input vector X t and each negative delayed output candidate vector -Y D ⁇ , in accordance with respective loop filter 502(i) transfer function.
• the state variables used by each loop filter 502(i) are identical as depicted by loop filter 502 and connecting lines indicating the common use of the state variables 501.
• Multiple quantization error vectors are determined by changing the values of each output candidate vector. For example, for an M-depth output candidate vector Y l5 if each element of Y, has two possible values, then there are up to 2 2 possible output candidate vectors, Yi, Y 2 , Y 3 , and Y 4 .
• the number of output candidate vectors can be pruned by, for example, eliminating or reducing redundant arithmetic operations. Selection of output values for the look- ahead delta-sigma modulator can be based upon any of a number of techniques.
• loop filter 202 can be modeled as having two separate transfer functions, H ⁇ (z) and H 2 (z).
• the noise transfer function (“NTF”) equals l/[l+z _1 *H 2 (z)].
• the signal transfer function equals H 1 (z)/[l+z "1 *H 2 (z)].
• the filter 502(i) is designed around optimizing the NTF quantity l/[l+z _1 *H 2 (z)]. Optimizing "l/[l+z _1 *H 2 (z)]" is effectively the same as minimizing z + H 2 (z).
• look-ahead delta-sigma modulator 500 selects an output value y(n) using quantization error values.
• Search module 504 computes quantization error power (C . - Y t ) (2) by determining a sum of squares in accordance with Equation 4: -y t ) 2 Equation 4
• the y(n) selector module 508 selects y(n) as the leading bit of Y,. from [C, - Y ⁇ ] (2) m ⁇ n- [ - Y ⁇ ] (2:) m ⁇ n represents the minimum quantization error power.
• This example can be extrapolated to cover any look-ahead depth and number of output candidate sets. To maintain causality, only earlier quantization values are fed back to the filter.
• the feedback vector choices are the 16 possible sets of values for times 10, 11, 12 and 13.
• the filter output at time 12 will be a function of all output choices up to time 11, but not of the output choice for time 12.
• the filter output at time 13 will depend on choices up to time 12.
• the quantization error will depend on the current filter output and the current output candidate.
• the quantization error at time 13 therefore depends on the value of choices at times 10 throughl3.
• the four quantization errors used are those at times 10 throughl3. This differs from the behavior when using C to create the cost function. In that case, the four filter outputs used for the optimization will be 11 through 14, as 10 is unaffected by the choice of the current vector.
• the selector module 508 can produce r output values for each input by providing as an output the r leading bits of the output candidate vector having the minimum quantization error power value, where generally 1 ⁇ r ⁇ M (the depth of the output candidate vector) and, in one embodiment r equals 2.
• the r outputs are then fed back to the noise shaping filter 502 and the state variables are updated r times.
• minimum quantization error search and output generator 506 has a defined gain. Additionally, the demonstrated signal-to-noise ratio of look- ahead delta-sigma modulator 5O0 exhibits in some cases a lOdB improvement over conventional look-ahead delta-sigma modulators.
• Figure 7 depicts a comparison of the transfer function of H 2 (z) for look-ahead delta- sigma modulator 106 with l+z _1 H 2 (z) of delta-sigma modulator 500 when the NTF is set to be a seventh (7 th ) order Butterworth filter with added NTF zeros.
• the plot 702 obtained using look- ahead delta-sigma modulator 5O0 demonstrates a smooth transition where the spike 404 occurs.
• plot 702 demonstrates quantization noise shaping improvement by reducing noise in a signal bandwidth of interest (e.g. 0Hz - 25kHz) by look-ahead delta-sigma modulator 500 relative to conventional look-ahead delta-sigma modulators.
• look-ahead delta-sigma modulator 800 represents an embodiment of a look-ahead delta-sigma modulator using time weighted, frequency weighted error values and quantization error to determine output values.
• the look-ahead delta-sigma modulator 800 is identical to look-ahead delta-sigma modulator 500 except for minimum quantization error search and output generator 806.
• the non-uniform weighted minimum quantization error noise search module 804 minimizes the scalar, power of a weighted quantization error vector Cj - Yj, for i D ⁇ 0,1,2, ... , N-l ⁇ .
• Vector W represents one embodiment of a non-uniform weight vector.
• W » [C ( - Y ⁇ ] (2) mm represents the minimum weighted, quantizer input/output difference power.
• the y(n) selector module 808 selects y(n) as the leading bit of output candidate vector Y,.
• the non- uniform weight vector W includes, for example, at least one non-zero weight element that is different from another weight element value or at least one non-unity, non-zero weight value.
• Weighting the quantization error vector can be accomplished by determining the dot product of the quantization error vector and the time-domain weight vector.
• the look-ahead delta-sigma modulator 800 represents one embodiment of look-ahead delta-sigma modulators that weight quantization error vector elements with at least one non-zero, non-unity weight (i.e., not equal to 1).
• Non-uniform weighting can allow a delta- sigma modulator to obtain higher signal-to-noise ratios than conventional look-ahead delta-sigma modulators while maintaining linearity associated with the conventional look-ahead delta-sigma modulators.
• the elements of the non-uniform weight vector W are a matter of design choice and are generally chosen empirically to minimize output signal noise.
• the elements of the weight vector trend downward in the time-domain.
• weighting using selected nonuniform weights decreases the aliasing due to truncation of the sample set, and improves the signal-to-noise ratio of the look-ahead delta-sigma modulator 300.
• Figures 9A-9F depict various exemplary weight vectors applied by the non-uniform weighted minimum quantization error noise search module 804.
• Figure 4 depicts the weight vectors as windows that weight each set of cost value powers and depicts exemplary weighting windows with a downward weighting trend.
• any variety of windows can be used with look- ahead delta-sigma modulator 800. Exponential windows are generally the easiest to implement with Trellis delta-sigma modulators.
• a tapered weight vector applied before summing the power in the search module 804 solves the quality problem associated with unity weighting and improves both signal-to-noise ratio and the noise transfer shaping.
• Weighting windows similar to Figure 4C, having weight vector W [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, .92, .80, .70, .52, .24], perform well for a depth of 10.
• the window depicted in Figure 9C can be difficult to implement in a look-ahead system that re-uses prior cost computations, as Trellis and tree structures often do.
• An exponentially decaying window such as the window of Figure 9A can be exploited in these cases.
• Weight elements, w t , in a weight vector W can also be defined to be within particular ranges.
• the best window for a particular depth and application can be determined empirically.
• the Melanson Weighting Patent describes further embodiments and applications of weight vectors W
• signal processing system 1000 depicts one embodiment of a signal processing system that includes look-ahead modulator 500.
• Signal processing system 1000 is particularly useful for high-end audio applications such as super audio compact disk (“SACD") recording applications.
• Signal processing system 1000 processes an input signal 1004 generated by an input signal source 1003.
• the input signal 1004 may be digital or analog and may be from any signal source including signals generated as part of a recording/mixing process or other high end audio sources or from lower-end sources such as a compact disk player, MP3 player, audio/video system, audio tape player, or other signal recording and/or playback device.
• the input signal 1004 may be an audio signal, a video signal, an audio plus video signal, and/or other signal type.
• input signal 1004 undergoes some preprocessing 1006 prior to being modulated by look-ahead modulator 1002.
• pre-processing 1006 can involve an interpolation filter to oversample a digital input signal 1004 in a well-known manner.
• Pre-processing 1006 can include an analog-to-digital converter to convert an analog input signal 1004 into a digital signal.
• Pre-processing 1006 can also include mixing, reverberation, equalization, editing, out-of-band noise filtering and other filtering operations.
• pre-processing 1006 provides discrete input signals X[n] to look- ahead modulator 1102.
• Each discrete input signal x[n] is a K-bit signal, where K is greater than one.
• look-ahead modulator 500 processes input signals X[n] and candidates Y[n] to determine an output signal 1007.
• Output signal 1007 is, for example, a collection of one-bit output values. The output signal 1007, thus, becomes an encoded version of the input signal 1004.
• signal processing system 1000 typically includes postprocessing 1008 to post-process the output signal 1007 of look-ahead modulator 500.
• Postprocessing 1008 can include lossless data processing 1102.
• For SACD audio mastering there is a lossless data compression stage 1104, followed by a recording process 1106 that produces the actual pits that are burned into a master storage medium 1108.
• the master storage medium 1108 is then mechanically replicated to make the disks (or other storage media) 1112 available for widespread distribution.
• Disks 1112 are, for example, any variety of digital versatile disk, a compact disk, tape, or super audio compact disk.
• Playback/output devices 1010 read the data from the disks 1112 and provide a signal output in a format perceptible to users. Playback/output devices 1010 can be any output devices capable of utilizing the output signal 1007.
• the storage media 1108 and 1 112 include data encoded using signal modulation processes achieved using look-ahead modulator 500.
• look-ahead delta-sigma modulator 500 Many systems can implement look-ahead delta-sigma modulator 500.
• the weighting of look-ahead delta-sigma modulator 500 can be implemented using hardware and/or software.

Landscapes

• Physics & Mathematics (AREA)
• Nonlinear Science (AREA)
• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)

Priority Applications (2)

Applications Claiming Priority (10)

Publications (1)

Family

ID=34831499

Family Applications (1)

Country Status (3)

Family Cites Families (2)

• 2005
  • 2005-01-13 JP JP2006549595A patent/JP2007518376A/ja active Pending
  • 2005-01-13 WO PCT/US2005/001093 patent/WO2005074141A1/en active Application Filing
  • 2005-01-13 EP EP05705645A patent/EP1704643A1/en not_active Withdrawn

Non-Patent Citations (7)

Also Published As

Similar Documents

Legal Events