JP2003124812A - Trellis type noise shaping modulator - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a modulator of which the performance such as stability, S/N in a signal band, the quality of generated noise or the like, is improved although a conventional noise shaping modulator has problems concerned with the above performance. SOLUTION: The trellis type noise shaping modulator is equipped with a recursive network for categories of outputtable states that has a means for selecting a set of several output candidates included in each element of a set of categories of outputtable states before each sampling time T recursively as to the sampling time T on the basis of the distance between an input and an output candidate and a means for generating final outputs one after another from the selected set of the output candidates.

Description

Description: BACKGROUND OF THE INVENTION [0001] 1. Field of the Invention [0002] The present invention relates to improvements in the performance of a noise shaping modulator such as the stability and the S / N ratio in a signal band, and more particularly to an A / D converter. And a converter for converting a multi-bit digital signal into a 1-bit digital audio signal such as DSD (Direct Stream Digital), a digital power amplifier, and the like. 2. Description of the Related Art A delta-sigma modulator is a typical noise shaping modulator. Although the configuration of the delta-sigma modulator is various, all of them can be regarded as having the same value as the block diagram shown below. FIG. 8 is a configuration diagram of a conventional digital sigma modulator. In the figure, 1
3 is an adder / subtracter, 14 is a quantizer, and 15 is a feedback function. The function H (z) in the feedback function 15 shown in the feedback loop of FIG.
This is called TF (noise transfer function)
By adjusting the NTF, the frequency distribution of the quantization noise can be changed. The function H (z), which is an NTF, cannot be freely selected because feedback is performed only based on past information, and H (∞) = 1
Must be satisfied. [0003] The delta-sigma modulator has many problems in practical use. One of the problems is "stability" and "S / S in a signal band." N ratio "
Problem. Here, "stability" refers to the degree to which the delta-sigma modulator does not "oscillate" with respect to various input signals.In the "oscillation" state, the output noise increases and the output signal in the signal band is increased. The correlation with the input signal is lost, and the modulator does not operate normally. In such a state, it is necessary to take measures such as resetting the circuit. Generally, in order to improve the “S / N ratio of the signal band”, it is necessary to transfer the noise to the outside of the signal band.
The gain at t of band) must be increased.
As a result, the input to the quantizer exceeds its input range,
This tends to lead to an "overload" condition, which increases the non-linearity and thus the stability of the delta-sigma modulator, especially for large amplitude signals. Therefore,
“Stability” and “S / N ratio in a signal band” are in a conflicting relationship. In the delta-sigma modulator, when the number of output bits is small, particularly when the output has a one-bit width, the stability is remarkably impaired. In the case of the converter, NT outside the signal band
Even if the gain of F is relatively low, it easily overloads, so that an oscillation state is likely to occur and the S / N of the signal band is increased.
This is because the ratio is deteriorated. In practice, when a delta-sigma modulator having a 1-bit width output and a high-order function H (z) is used, a "3 dB gain rule" has been proposed as a design guide. It is an empirical rule that the gain of the NTF outside the band must be suppressed to 3 dB in order to secure “stability” (for example, STEVEN R. NORSWORTHY, RICHARD SCHREIERand
GABOR C. TEMES, "Delta-Sigma Data Converters, The
ory, Design, and Simulation, IEEE PRESS, 1997). While in the case of the implementation can take the gain of the NTF outside the signal band a little more to increase the input level does not oscillate, for example, SACD definition of 0dB of (Super Audio CD) (energy Δ ^2/32 per sample) Basically, the value of 3 dB is a realistic value if it is desired to secure the degree. However, if the gain is suppressed to 3 dB, the number of bits used for transmitting information of the signal band is suppressed, and the bit use efficiency deteriorates. As a method for solving the conflict between the stability at the time of the low output bit and the S / N ratio of the signal band, M
There are cascaded delta-sigma modulators such as an ASH (multi-stage noise-shaping) delta-sigma modulator, and this modulator can increase the gain outside the NTF signal band while maintaining stability. In that respect, half of the problems with delta-sigma modulators can be solved,
The output format is an increase in the number of bits or an oversampled format, which is not improved in terms of improving bit usage efficiency. The sampling frequency for dropping (requantizing) data into the DSD format (1 bit, 64 × 44.1 kHz) used for the SACD is fixed. No improvement effect was expected for applications such as streaming. Another practical problem of the delta-sigma modulator is the quality of generated noise. For example, a phenomenon that a specific input has a strong spectrum at a specific frequency (herein referred to as a tone) is observed, and it is pointed out that this causes a problem in hearing. One solution to this is a so-called dither (di)
injection) is performed. However, in order to improve this effect sufficiently, it is necessary to inject dither with a certain amount of energy, and the resulting side effects such as noise mixing in the output and narrowing of the input dynamic range increase with the increase in dither energy. Resulting in. On the other hand, in the information communication field such as decoding of convolutional codes and turbo codes, even when a code containing noise mixed in the middle of information transmission is received, the code is “closest in shape” (hereinafter referred to as the closest distance). ) Viterbi algorithm is one of the effective methods to consider and reconstruct a code as the most probable candidate.
Decoders applying (generalized Viterbi algorithm) are widely used. The function of the noise shaping modulator is also similar to the decoder to which the Viterbi algorithm is applied, in that the input signal and the input signal in a sense are found to have an output that is the closest in shape (the difference is smaller than the original). Thus, it makes sense to apply the Viterbi algorithm to a noise shaping modulator. However, the Viterbi algorithm has not been applied to a noise shaping modulator. As described above, according to the present invention, by applying the Viterbi algorithm to the noise shaping modulator, the output format is low and the sampling frequency is fixed while sufficiently securing the stability of the noise shaping modulator. Even in this case, the used bits can be used almost completely for transmitting information of the effective signal band, and as a result, the S / N ratio of the signal band is improved, and
Injecting a smaller amount of dither effectively reduces tone,
As a result, a noise shaping modulator is provided that also reduces the side effects of dither. Means for Solving the Problems In view of the above, the present inventor has solved the problem by the following means. (1) In a noise-shaping modulator having an input and an output for each sampling time T, for each element of the possible output state before the sampling time T defined for each sampling time T, Means for selecting a set of several included output candidates recursively with respect to the sampling time T based on the distance between the input and output candidates, and sequentially generating a final output from the selected set of output candidates A trellis type noise shaping modulator, characterized in that the trellis is of an output enable type. Embodiments of the present invention will be described below in detail with reference to mathematical formulas and drawings. An embodiment of the present invention provides an input in
(For example, input in (T) given for each sampling time T) and each sampling time T (where
T takes a natural number as a value by normalizing the time unit, and outputs out (T) (for example, △ / 2 and −
In a noise shaping modulator having a 1-bit stream that can take a binary value of Δ / 2), the types of output possible states before the sampling time T defined for each sampling time T are as follows: [0013] However, for example, when there are a plurality of candidates whose distances hardly change, these candidates are left at the same time, or the stochastic behavior is required to reduce unnecessary tones included in the output. In some cases, it is effective to apply an application such as leaving one at random from among them. [Equation 3] If output candidates are generated recursively in this way, the output can be determined as follows. [Equation 4] [Equation 5] [Equation 6] [Equation 7] As long as the distance d _T used here satisfies the above properties, it can be freely selected according to the purpose of the user using this technique. Normally, NTF
In many cases, an output is required such that the energy of the noise normalized by the inverse function of (i) becomes as small as possible (the square norm of the noise becomes as small as possible), so the square norm seems to be general. However, in terms of easiness of calculation, the first-norm is more advantageous, so the first-norm can be employed. In that case, the energy of the noise generated by the noise shaping modulator naturally increases. Further, in selecting the distance d _T , it is necessary to consider the required amount of memory. From this and to reduce the amount of calculation, d
_It is convenient to select T that can perform recursive calculation on T. The square or norm norm after normalization with the above inverse function of NTF can be calculated recursively. However, the selection of the distance d _T is not limited to the one that allows recursive calculation. Classification of possible output states before sampling time T, characteristic of the Viterbi algorithm Therefore, the new noise shaping modulator is called a trellis type. The above items will be described in detail. First, the definition of a 1 × oversampling standard Nth-order trellis type 1-bit output noise shaping modulator as a typical format of the trellis type 1-bit output noise shaping modulator will be described below. (Equation 9) [Mathematical formula-see original document] [Equation 11] An embodiment will be described below with reference to the drawings.
Figure 1 shows a 1x oversampling standard second-order trellis type 1
FIG. 2 is a block diagram of an arithmetic circuit, FIG. 3 is a block diagram of a path metric adjuster, and FIG. 4 is a block diagram of a path selection, path metric addition, filter variable addition, and transmission detection block. FIG. 5 is a block diagram of an integrated type of path memory and path selector, FIG. 6 is a block diagram of module A shown in FIG. 5, and FIG. 7 is a module B shown in FIG.
FIG. First, a description will be given based on a block diagram of the 1 × oversampling standard second-order trellis type 1-bit output noise shaping modulator shown in FIG. Here, LSI
In addition to the function of the 1 × oversampling standard second-order trellis type 1-bit output noise shaping modulator, an oscillation detection / reset function which is important in actual mounting is provided. In the figure, 1
Is a distributor, 2 is an arithmetic circuit, 2a is an arithmetic circuit (00), 2
b is an arithmetic circuit (01), 2c is an arithmetic circuit (10), 2d
Denotes an arithmetic circuit (11), 3 denotes a path metric adjuster, 4 denotes a path memory, and 5 denotes a path selector. [Mathematical formula-see original document] The other three arithmetic circuits 2b to 2d have similar structures. The path metric adjuster 3 focuses on the fact that the path metric has a meaning in the relative value and has no meaning in the absolute value. The path metric adjuster 3 appropriately avoids a phenomenon in which the path metric monotonically increases with time and overflows. In addition to the function of uniformly subtracting the value of, it also has a reset function at the time of oscillation. The path memory 4 stores the history of the operation results of each of the operation circuits 2a to 2d. Next, the operation of each part will be described in more detail. First, filter variables used in the filter calculation in the arithmetic circuit 2 will be described. Equation 14 [Equation 15] For the sake of simplicity, the following description is based on the assumption that n is 4. FIG. 2 shows a block diagram of the arithmetic circuit (00) 2a as a representative of the arithmetic circuit 2 in FIG. FIG. 4 is a timing chart of input / output signals of each part of the arithmetic circuit (00) 2a of FIG. The arithmetic circuit (00) 2a shown in FIG.
In the block diagram, reference numeral 6 denotes a path selection / path metric addition / filter variable addition / oscillation detection, 7 a digital filter, 8 a buffer, and 9 a changeover switch. The arithmetic circuit (00) 2a has inputs of two path metric / filter variables PFIN 0 and PFIN 1 and a signal SIN input. The path metric filter variable input 1 PFIN 0, has a path metric P0 (T) corresponding to each sampling time T.
, Filter output U0 (T), filter variables V01 (T) to V
04 (T) is input. The same applies to the path metric / filter variable input 2 PFIN 1. [Equation 16] [Mathematical formula-see original document] (Equation 18) [Equation 19] [Mathematical formula-see original document] Next, the "path selection, path metric addition, filter variable addition, oscillation detection" block outputs v1 (T + 1), v2 (T + 1), v3 (T + 1 ), v4
Output in the order of (T + 1), u (T + 1). [Equation 21] After a short delay, the oscillation state OV (T + 1) and the path metric p (T + 1) are output to POUT. Oscillation state OV (T +
As 1), the following three types are distinguished and output (oscillation detection). [Mathematical formula-see original document] Next, the "digital filter" block will be described. The block of the “digital filter” 7 in FIG. 3 has v1 (T + 1), v2
(T + 1), v3 (T + 1), v4 (T + 1), u (T + 1) in order of FOUT input, V1 (T + 1), V2 (T + 1), V3 ( DFOUT is output in the order of T + 1), V4 (T + 1), U (T + 1). This digital filter block is given by: FIG. 4 shows a case where the difference (latency) between the timings of input and output is 1, for example, but the latency varies depending on the configuration of the filter. Next, the "buffer" block will be described. In the block of the “buffer” 8 in FIG. 2, the data input from DFOUT is simply rearranged and output as BFOUT. The timing is as shown in FIG. And the "changeover switch" of FIG.
In the block 9, POUT and BFOUT are multiplexed at the timing shown in FIG. 4 and output as PFMID (path metric filter variable output). Next, the function and operation of the path metric adjuster will be described. Fig. 3 "Path metric adjuster" 3
FIG. In the figure, reference numerals 10a to 10d denote "path metric extraction" blocks of PFMID from the arithmetic circuits (00) to (11), and reference numerals 9a to 9d denote changeover switches. The path metric adjuster is given by: (2) A function of resetting a filter variable, a filter output, and a path metric value when the outputs of all the arithmetic circuits are oscillating, and performing a return operation. The function (1) can be realized, for example, as follows. [Formula 25] However, when p (T + 1) overflows, P (T + 1) also remains overflowing. (It should be noted that this feature does not guarantee that outputs from "all" arithmetic circuits are immune from overflow, and that it is not necessary to do so.) 2) The function of OV from all arithmetic circuits
When (T + 1) is case 1 or case 2, it is recognized that the entire noise shaper is in an oscillation state, and reset is performed. If this oscillation state is not detected, the path metric
The adjuster controls the output from all arithmetic circuits.
At timing 4, P (T + 1), U (T + 1), V ₁ (T + 1), V ₂ (T + 1), V
₃ (T + 1), V ₄ (T + 1) are output as they are. When this oscillation state is detected, the output from the arithmetic circuit in Case 1 is left as it is, and the output from the arithmetic circuit in Case 2 is P (T +
1), U (T + 1), V ₁ (T + 1), V ₂ (T + 1), V ₃ (T + 1) and V ₄ (T + 1) are all output as 0. Next, the functions and operations of the path memory 4 (FIG. 1) and the path selector 5 (FIG. 1) will be described. Path memory 4
Has a function of storing S (T) output from each operation circuit (operation circuits (00 to 11) to SOUT. In order to distinguish the output from each operation circuit, for example, 26] [Equation 27] As an example, when the path memory 4 and the path selector 5 are configured as follows, the operation can be performed efficiently.
FIG. 5 shows an integrated block diagram of the path memory and the path selector. In the figure, 11 is module A, 11a is module A # 1, 11b is module A # 2, 11c
Is module A # 1 _buffer / 2-1, 11d is module A # 1 _buffer / 2, 12 is module B, 12a is module B # T _buffer , and 12b is module B # T _buffer .
_buffer-1, 12c is module B # T _buffer / 2 +
Reference numerals 2 and 12d indicate module A # T _buffer / 2 + 1.
In the figure, a single thick arrow includes four connections, and the clock here is in accordance with the input of S (T).
That is, one sampling corresponds to one cycle of the clock. [Formula 28] There are two types of modules, called module A11 and module B12, respectively. The block diagram of the module alone is shown below. FIG. 6 is a block diagram of the module A alone. In FIG.
Indicates an output terminal, and CK indicates a clock input terminal. FIG. 7 is a block diagram of the module B alone. In the figure, A and B indicate input terminals from the preceding stage, Y indicates an output terminal, and A / B # indicates an input terminal from the module A. [Mathematical formula-see original document] Next, comparison data with the conventional type using the trellis type noise shaping modulator of the present invention are shown in the figure. FIG. 9 shows a signal band gain Gm = 1.
0th order, 1st order, 2nd order, 3rd order, 4th order, 6th order, 8th order when 75 fifth-order Butterworth high-pass filters are used
FIG. 6 is a graph showing input DC levels and time until the oscillation state is reached for each of the next, twelfth, and trellis orders. However, the sampling rate is 2.822 which is the same as SACD.
4 MHz. From this figure, the higher the trellis order,
It can be seen that oscillation becomes difficult for a high level DC input. Since the oscillation phenomenon is chaotic, the graph is not a continuous function with respect to the abscissa, so that this graph is considerably uneven. In fact, 0.12 at present
Although the drawing is performed in steps of 5 dB, the narrower this interval is, the more violently it goes up and down. However, reading trends can be done satisfactorily. Next, as NTF, the signal band gain Gm
= 57 dB DC signal (DSD) at the 0th, 2nd, and 8th trellis orders when a 5th-order Butterworth high-pass filter of 1.5 is used.
A graph of the frequency distribution of output noise when a reference of 0 dB is equivalent to -9 dB on a full-range basis is shown below. The frequency distribution is obtained by averaging the power over 10 seconds using a DFFT at 64 k points in the Blackman window. FIG. 10 is a graph showing the frequency distribution of the output noise at the zeroth trellis order, FIG. 11 is a graph showing the frequency distribution of the output noise at the second trellis order, and FIG. 12 is the output noise at the eighth trellis order. FIG. Comparing the above graphs for the 0th, 2nd, and 8th trellis orders, as the trellis order increases, the tones seen near 1.4 kHz, 2.8 kHz, and 1.4 MHz decrease. You can see how it goes. According to the present invention, the following excellent effects are exhibited. According to the invention of claim 1 of the present invention, since it is a trellis-type noise shaving modulator having a recursive network of various types in which output is possible before a characteristic sampling time T in the Viterbi algorithm, a conventional delta-sigma type Compared to the 1-bit output noise shaping modulator (equivalent to the 1-time oversampling zero-order standard trellis type 1-bit output noise shaping modulator of the present invention), the 1-time oversampling N of the present invention
The following standard trellis type 1-bit output noise shaping modulator: 1. When the same NTF is adopted, the calculation amount exponentially increases as the order N increases, but for a larger input, However, oscillation is not performed, the operation becomes stable, the generation of tone itself is suppressed, and the removal effect by dither is enhanced. 2. As a result, N with a higher gain outside the signal band
Even if TF is adopted, stability can be maintained, and S / N
The ratio can be improved. Therefore, it is possible to approach the theoretical limit by increasing the order N as needed. In fact, This effect is so remarkable that the stability can be sufficiently ensured by increasing the order N.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of an embodiment of a 1 × oversampling standard second-order trellis type 1-bit output noise shaping modulator. FIG. 2 is a block diagram of an arithmetic circuit. FIG. 3 is a block diagram of a path metric adjuster. FIG. 4 is a diagram showing timings of path selection, path metric addition, filter variable addition, and an oscillation detection block. FIG. 5 is an integrated block diagram of a path memory and a path selector. FIG. 6 is a block diagram of a module A alone. FIG. 7 is a block diagram of module B alone. FIG. 8 is a configuration diagram of a delta-sigma modulator. FIG. 9 shows a signal band gain Gm = 1.7 as NTF.
0th, 1st, 2nd, 3rd, 4th, 6th, 8th using a 5th Butterworth high-pass filter of 5th
FIG. 7 is a graph showing input DC levels and time until an oscillation state is reached for each of the next, twelfth, and trellis orders. FIG. 10 is a graph showing the frequency distribution of output noise in the 0th-order trellis order. FIG. 11 is a graph showing the frequency distribution of output noise in the second trellis order. FIG. 12 is a graph showing a frequency distribution of output noise at an eighth trellis order. [Explanation of Symbols] 1: Distributor 2 (2a to 2
d): arithmetic circuit 3: path metric adjuster 4 (4a-4)
d): path memory 5: path selector 6: path selection / path metric addition / filter variable addition / oscillation detection 7: digital filter 8: buffers 9, 9a to 9d: changeover switches 10a to 10d: "path metric" from the arithmetic circuit Extraction "block 11: module A 11a: module A # 1 11b: module A # 2 11c: module A # 1 _buffer / 2-1 11d: module A # 1 _buffer / 2 12: module B 12a: module B # T _buffer 12b: module B # T _buffer +1 12c: module B # T _buffer / 2 + 2 12d: module A # T _buffer / 2 + 1 13: adder / subtractor 14:
Quantizer 15: Feedback function D: Input terminal Q: Output terminal A, B ,: Input terminal from previous stage Y: Output terminal A / B #: Input terminal from module A CK: Clock input terminal

Claims (1)

Claims: 1. A noise shaping modulator having an input and an output for each sampling time T, a sampling time T defined for each sampling time T.
Means for selecting, for each element of the previous class of possible output states, a set of several output candidates contained therein, recursively with respect to the sampling time T, based on the distance between the input and output candidates; Means for sequentially generating a final output from a set of selected output candidates. A trellis type noise shaping modulator characterized by comprising a recursive network (trellis) of an output enabled type.