JP3167472B2 - Method for measuring SN ratio of analog-to-digital converter - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an analog-to-digital converter (hereinafter referred to as "ADC") capable of obtaining an SN ratio accurately and at high speed.

[0002]

[Technical background] Digital technology in signal analysis introduced
As they evolve, ADCs for converting analog signals to digital signals are also remarkable.
Because the characteristics of the ADC determine the limits of signal analysis,
When performing high-precision signal analysis, AD
Strict precision is required for the characteristics of C. In particular, in recent years,
There is an increasing tendency to attach importance to ADC characteristics close to the actual operation state, which are called dynamic characteristics.

[0003] In this dynamic characteristic evaluation, an SN ratio (Signal / Noise Ratio) measurement is generally performed. Theoretically, the SN ratio and the effective resolution (effective number of bits) of the ADC are: SN ratio = a · (the number of effective bits) + b (where a and b are constants determined by specifications and the like).

[0004] Since the SN ratio directly indicates the performance of the ADC, it is a useful evaluation guideline from the standpoint of actually using the ADC. To perform the SN ratio evaluation, as shown in FIG. 3, a signal source 11 for generating a high-purity sine wave is prepared, a signal from the signal source 11 is supplied to the ADC under measurement 12, and a clock generator 13 The ADC 12 is operated by the sampling clock. Then, the sampling data (time axis data) is stored in the buffer memory 14, and an arithmetic unit 15 performs a fast Fourier transform (FFT) operation on the data to obtain an S / N ratio. Unit 16). But,
Conventionally, the measurement of the SN ratio has two major problems described below.

(1) In the process of calculating the S / N ratio, the components other than the fundamental wave and the DC component are measured as noise components, so that distortion due to distortion of the signal source or the input analog part of the ADC, conversion characteristics, etc., is also noise. Is counted as
The SN ratio cannot be determined correctly.

(2) The input level of the sine wave signal to the ADC should be at the full scale of the ADC. This is because the noise related to the measurement includes a component irrelevant to the input level, and when a full-scale input is used (in this case, the signal component in the SN ratio is the largest), the result that the SN ratio is the best is obtained. Because it becomes. However, there are a gain error, an offset error, and the like in the input section of the ADC. Even if a specified full-scale input value is input, the error causes the top of the sine wave to be clamped and the distortion to increase. Or fall short of full scale.

To solve the above problem (1), a signal source shown in FIG. 3 having a smaller distortion is used to reduce the inclusion of the distortion in the noise to some extent. However, recent signal sources for evaluating high-resolution, high-speed ADCs have a frequency range of -80 to -90 dB at several MHz, and therefore cannot meet the above-mentioned requirements. The realization of such a signal source is considerably difficult with the current technology, and even if such a signal source is provided, it is not suitable for practical use in terms of cost. There is also a method of allowing a certain amount of distortion to be included in the noise, and excluding from the calculation a portion that is considered to be distortion based on the result of the FFT operation. However, it is unclear how far the distortion should be. For this reason, the noise component cannot be accurately obtained by this method. Furthermore, in the case of the distortion of the ADC to be measured itself, since the component and the level change each time the ADC changes, the calculation process is changed each time, or only the specific distortion is removed by knowing that the ADC includes an error. Must be performed. As a result, a different calculation must be performed each time, and the time required for the calculation becomes enormous. Conversely, if the calculation time is reduced, the error increases.

With respect to the problem (2), the expected values of the gain error and the offset error are taken into consideration, and the sine wave is not clamped even if these errors occur (usually 90%).
%), The input level is lowered to perform the evaluation. By taking such measures, it is possible to prevent a sine wave from being clamped due to a gain error and an offset error. However, since a signal component is suppressed to a low level, deterioration of the evaluation value of the SN ratio is promoted. ADC
There is a method of calculating the gain error and the offset error from the output of, and re-adjusting the output level of the signal source, or automatically adjusting the output level by incorporating a feedback circuit. Instead of being able to obtain it, the measurement efficiency is reduced.

[0009]

SUMMARY OF THE INVENTION The present invention has been proposed to solve the above problems, and has as its object to provide an accurate and high-speed SN ratio measuring method.

[0010]

SUMMARY OF THE INVENTION In the SN ratio measuring method of the present invention, a sine wave signal for measurement is provided from a signal source to an ADC to be measured. The present invention basically allows the analog-to-digital converter under test to sample the sine wave signal for measurement, converts time-axis data obtained by the sampling into frequency-axis data, Of these, a noise component is obtained from components other than the DC component, the fundamental wave component, and the harmonic component, and S is calculated based on the noise component and the fundamental wave component.
It is characterized in that the N ratio is obtained. For example, the noise component can be obtained by a plurality of samplings in the same cycle, or the noise component can be obtained by performing a plurality of samplings over a plurality of cycles and averaging the measurement results of the noise component. .

In the present invention, a sine wave signal for measurement is given to the ADC to be measured over m periods (m is a positive integer of 2 or more), and the ADC performs sampling by m × 2x times (x is a positive integer). The S / N ratio can also be determined as follows. That is, an FFT operation and a DFT (Discrete Foe) are applied to the time axis data obtained by the sampling.
uri Transform) operation is performed, and the time axis data is converted into mx frequency axis data (spectrum). These frequency axis data are (p−
1) The (m + 1) -th (p = 1, 2,..., X) first frequency axis data group and the second frequency axis data not including the frequency axis data constituting the first frequency data group It is separated and stored in groups.

The first frequency axis data group includes the measured ADC.
Is equivalent to frequency axis data obtained when 2 × samplings are performed over one cycle. The first frequency axis data in the first frequency axis data group is a DC component, and the second frequency axis data (that is, m · 2 over m cycles)
The (m + 1) th frequency axis data when x samplings are performed is a fundamental wave component. Therefore,
Fundamental wave component (signal component E) in SN ratio
_signal )

[0013]

(Equation 3) Becomes Also, from the second frequency axis data group, a noise component (E
_noise )

[0014]

(Equation 4) Is calculated. In equations (1-1) and (1-2),
E (i) represents the ith frequency axis component. In addition,
Details of these equations will be described later. Thus, 20
log (E _signal / E _noise ), SN
A ratio is determined.

In the case where the sine wave signal is clamped, the fundamental wave component can be obtained from time axis data (sampling data) instead of the equation (1-1). In the FFT, the number of data to be converted is 2 ^N (N is a positive integer). Therefore, when converting the time axis data to the frequency axis data by FFT, m is usually 2 ^t (t is a positive integer) and x is 2 ^n-1 (n is a positive integer). On the other hand, in the DFT, the number of data to be converted is arbitrary, and thus there is no such limitation.
Arbitrary values such as 50 and 500 can be adopted as x, such as 5, 6 and the like.

[0016]

FIG. 1 is an explanatory diagram showing an embodiment of the present invention. In the figure, a signal source 1 generates a sine wave signal for measurement. This sine wave signal is provided to the ADC 2 to be measured. In this case, the sine wave may have a defined full-scale amplitude. The sampling clock of the ADC 2 to be measured is generated by the clock generator 3.
Since the time axis-frequency axis conversion operation is performed, the point of the time axis data (sampling data) is m · 2x (m is a positive integer of 2 or more, x is a positive integer).

Here, the frequency of the input sine wave is set so that the data obtained by digitally converting the input sine wave becomes an even number.
The frequency of the sampling clock has been determined. In the present embodiment, x = 512, m = 2, and the point of the time axis data is m · 2x = 2048. That is, sampling of 2048 points is performed over two cycles of the input sine wave. For example, if the frequency of the input sine wave is 1 kHz, the frequency of the sampling clock is 1.0
24 MHz.

The time axis data sampled by the ADC 2 to be measured is temporarily stored in the buffer memory 4,
It is taken in by a time axis-frequency axis converter (FFT arithmetic unit 5 in the figure). Here, m and x are selected so that the number of data to be converted (time axis data) is 2 ^N (N is a positive integer) so as to enable FFT conversion (in this case, m = 2 ¹ , x = 2 ⁹ ). If the time axis-frequency axis converter is an arithmetic unit such as a DFT arithmetic unit which is not restricted by the above ^2N condition, m is any positive integer of 2 or more, and x is any positive integer. be able to. The arithmetic unit 5 converts the time-axis data into frequency-axis data.
Perform FT operation. The result of such an operation is usually output separately for a real part and an imaginary part. Here, it is assumed that the frequency axis data is output as data having only a size (that is, spectrum).

By the FFT operation, half of the number of time axis data, that is, mx (1024 in this embodiment) frequency axis data can be obtained. Of these frequency axis data, (p−1) · m + 1th (p = 1, 2,...)
., X) (in the present embodiment, the first, third,..., 1
The 023rd frequency axis data constitutes a first frequency axis data group. The frequency axis data other than the first frequency axis data group, that is, the second, fourth,..., 1024th frequency axis data become a second frequency axis data group.

The first frequency axis data group is stored separately in the memory 7A, and the second frequency axis data group is stored separately in the memory 7B. The distribution of the frequency axis data to the memories 7A and 7B is performed by the data switch 6.
After all the frequency axis data are stored in the memories 7A and 7B, the arithmetic unit 8 performs calculations based on the above-described equations (1-1) and (1-2). In the figure, a display or the like is displayed on a display 9). (1-
Substituting m = 2 and x = 1024 into the equations (1) and (1-2) gives the signal component E in the SN ratio.
_The expression (1-1) representing the _signal is represented as follows.

[0021]

(Equation 5)

[0022]

(Equation 6) Becomes From the above equations (2-1) and (2-2), S
An N ratio is determined.

[0023]

(Equation 7)

Hereinafter, the above operation will be described in detail. Considering an ideal ADC, m · 2 over m periods as described above
The time axis data when sampling x times is 1
This is equivalent to sampling 2x times over a period and arranging m sets of this time axis data. That is, in the case of the above embodiment, the time axis data having two periods at 2048 points is the time axis data having one period at 1024 points.
It is equivalent to a set.

When this result is converted from time axis to frequency axis,
The second frequency axis data group is always 0. That is,
The number of time axis data when m × 2x samplings are performed over m periods (referred to as “case I” for convenience) is 2x times sampling over one period (referred to as “case II” for convenience) ), The frequency resolution by the time axis-frequency axis conversion calculation becomes 1 / m. However, in this case, the time axis data is sampled 2x times per cycle, and the contents are substantially the same as the time axis data obtained by taking this for m cycles. In other words, the time axis data in case I does not include information different from the time axis data in case II. Therefore, the result of the time axis-frequency axis conversion in case I and the result of the time axis-frequency axis conversion in case II should be exactly the same.

When this logic is applied to the above embodiment, as shown in the spectrum diagrams of FIGS. 2A and 2B, in case I (when sampling at 2048 points over two cycles), the odd number of the frequency axis The frequency axis data (FIG. 2A) appearing second and the frequency axis data (FIG. 2B) sequentially appearing on the frequency axis in case II (when sampling at 1024 points over one cycle) are the same. . FIG. 2C shows a state in which 1024 samplings are performed per cycle. Note that FIG.
The first frequency axis data E in (A) and (B)
(1) is a DC component. In addition, 3 in FIG.
The second frequency axis data E (3) and the second frequency axis data E (2) in FIG. 2B are fundamental wave components. Further, the frequency axis data E (5) and E (7) appearing at the fifth and seventh positions in FIG.
And the fourth and fourth frequency axis data E (3), E
(4) is a distortion component.

In FIG. 1, noise at the input stage of the ADC 2 and noise due to the quantization error exist independently of the measurement time of the ADC 2. Therefore, the actual ADC
In the measurement of 2, the frequency axis data is (p−1) · m + 1th (p = 1, 2,..., X) frequency axis data (first
The frequency axis data (second frequency axis data group) other than the frequency axis data group (2) is not always zero. The noise component can be expressed by the general formula:

[0028]

(Equation 8) (No distortion is included in these noise components). In the case of the above embodiment,

[0029]

(Equation 9) Becomes FIG. 2A shows noise components E (2) and E (2).
(4),..., E (1024).

These noise components appear in all frequency axis data. In order to convert the data to all frequency axis data, generally, equation (4) may be multiplied by {m / (m-1)}, and in the case of the above embodiment, equation (5) may be multiplied by {2}. It may be multiplied (see equations (1-2) and (2-2)). In addition,
The greater the number of points in the time axis data (the smaller the resolution of the frequency axis data is smaller than the noise band),
Needless to say, the accuracy is further improved.

On the other hand, the (p-1) .m + 1 th (p = 1,
2, x) frequency axis data (first frequency axis data group), more specifically, the frequency axis data stored in the memory 7A in FIG. This is equivalent to the frequency axis data in the case where sampling is performed in (1). Therefore, the fundamental wave component is given by the (m + 1) th frequency axis data of the first frequency axis data group.

The fundamental wave component can also be obtained from the time axis data stored in the buffer memory 4. If the clamp of the measured sine wave is not large, there is almost no error even from the result of the time axis-frequency axis conversion,
If the clamp is large, it is preferable to calculate from the time axis data. The dotted line in FIG. 1 can be realized by an arithmetic circuit and control means.

[0033]

As described above, according to the present invention, the following effects can be obtained. (1) In the process of calculating the S / N ratio, noise can be measured separately from the signal source, the input analog part of the ADC, and the distortion generated by the distortion of the conversion characteristic, so that the S / N ratio can be measured with high accuracy. can do. (2) Even if the top of the sine wave is clamped at the input part of the ADC due to a gain error, an offset error, or the like, and the distortion increases, it does not affect the noise measurement as described above. Therefore, the necessity of adjusting the full-scale input is greatly reduced, so that more accurate and highly accurate S / N ratio measurement can be performed. (3) Due to the above effects, it is not necessary to use a high-precision signal source, so that hardware cost can be reduced. (4) By changing the control means of the arithmetic unit, the measurement method of the present invention can be implemented with existing circuits and equipment.

[Brief description of the drawings]

FIG. 1 is a diagram showing an example of a measurement circuit used for a measurement method of the present invention.

FIGS. 2A and 2B are diagrams for explaining the operation of the measuring method of the present invention, wherein FIG. 2A is a spectrum diagram when m = 2 and x = 512 in the present invention, and FIG.
FIG. 4C is a spectrum diagram when 1024 samplings are performed over a period, and FIG. 4C is a diagram illustrating a state where 1024 samplings are performed over one period.

FIG. 3 is a diagram for explaining a conventional SN ratio measuring method.

[Explanation of symbols]

 Reference Signs List 1 sine wave signal source 2 ADC to be measured 3 clock generator 4 buffer memory 5 FFT calculator 6 data switch 7A, 7B memory 8 calculator 9 display

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H03M 1/00-1/88 G01R 31/28

Claims (2)

(57) [Claims] A sine wave signal for measurement is supplied to an analog-to-digital converter to be measured, and the sine wave signal for measurement is applied to the analog-to-digital converter for measurement over m periods (m is a positive integer of 2 or more). 2 × times (x is a positive integer) sampled, the time axis data obtained by the sampling is converted to m × x frequency axis data, and the frequency axis data is (p−1) × m + 1 th (p = 1,
, X) and a second frequency axis data group other than these frequency axis data groups.
Is stored separately from the first frequency axis data group and the second frequency axis data group. (Equation 2) (Where E (i) is the value of the i-th frequency axis data), and the SN ratio is calculated based on the fundamental component and the noise component. 2. An analog according to claim 1 , wherein a fundamental wave component is obtained from time axis data obtained by said sampling, and an SN ratio is obtained based on said fundamental wave component and said noise component. -A method for measuring the S / N ratio of a digital converter.