results of an AD device application in the frequency range of the testing signal ... and a short time amplitude and/or frequency instability of testing signal generators .... of the old and new AD transfer standards measured with DS 360 and Audio.
Evaluation of ADC Testing Systems Using ADC Transfer Standard Vladimír Haasz, Jaroslav Roztočil, David Slepička Dept. of Measurement, Faculty of Electrical Engineering, Czech Technical University in Prague Technická 2, CZ-16627 Praha 6, Czech Republic Phone: +420-2-2435-2186, Fax: +420-2-3333-9929, Email: [haasz, roztocil, slepicd]@fel.cvut.cz
Abstract – A transportable high-stable reference AD device was designed and built to compare the systems for testing the dynamic quality of ADCs or AD modules. Three different input modules can be used in the frequency range up to 5 MHz. First, this paper refers to the results of an AD device application in the frequency range of the testing signal up to 100 kHz, where four ADC testing systems were compared in different laboratories, and a short time amplitude and/or frequency instability of testing signal generators were evaluated. Second, it describes the first experience of its application in the frequency range from 100 kHz to 5 MHz.
I. INTRODUCTION There is no simple method for evaluating parameters of the systems that are used for testing the dynamic quality of the high resolution and/or high speed ADCs. The ADC testing methods are usually based on sampling a spectrally pure sine wave signal. However, the warranted parameters of the best low-distortion and low-noise generators manufactured at present are comparable with the corresponding parameters of the tested ADCs (THD, spurious components, noise floor etc.). The actual parameters are usually better than those guaranteed by the manufacturer; nevertheless, it is necessary to verify them for each single generator. For this purpose, the method of the rejection of the fundamental harmonic component using a notch filter is applied [1]. However, this method does not detect further parameters of generators. It can further influence the results during ADC testing. It particularly concerns both a short-time amplitude and frequency instability. The credibility of the achieved results can be degraded not only by the quality of the testing signal, but also by the sampling synchronization, signal processing, EMC conditions etc. Unfortunately, there is no reference digitizer or testing method developed to perform the testing with significantly higher precision. In general, comparative methods are used in metrology in the case, when there is no standard testing device with sufficient precision. A high-stable transfer standard is applied to compare various testing devices in different laboratories and to evaluate
their quality. Therefore, the comparative method was chosen for the evaluation of the systems designed for ADC dynamic testing. It enables to estimate real precision of testing systems and to compare advantages and disadvantages of applied methods. To make the comparative measurements possible, a transportable high-stable reference AD device (AD transfer standard—ADTS) has to be at disposal. So, the first prototype (ADTS1) was designed and built at the Dept. of Measurement, FEE CTU in Prague [2]. The prototype was designed to have high stability and to be disturbance-resistant. Its usability was tested and proved by the comparison of the measurements made in the ADC T&M Laboratory of the Dept. of Measurement of FEE CTU in Prague and three other different European laboratories within one year. Some limitations and imperfections were found out (i.e. problems with coherent sampling and data transmission, an influence of a power supply source). Therefore, a new improved version of the transportable reference AD device ADTS2 [3] was made (see Fig. 1). The functionality and universality of the device were increased and some imperfections of the first version eliminated.
Fig. 1. AD transfer standard—innovated version To achieve greater flexibility and to minimize internal disturbances, a modular solution was proposed. The designed device consists of two boxes. The ADC box includes a high-quality AD converter, signal conditioning circuits in a separate input module, and control circuits for an ADC timing and a data transfer to a computer or logic analyzer. The power supply box contains transformers, stabilization part and smart battery recharges, which could even work during the measuring. The device can be manually controlled from the front panel or by means of a computer using a special software. The internal shielding case including the input module was modified to enable an easy replacement of various input modules. The input module is temperature stabilized at adjustable temperature, galvanicly insulated from the FPGA circuit and supplied from batteries to be protected from external influences.
In designing the control part, different interfaces and timing of various types of ADCs were also considered. The applied AD modules are based on the following professional AD kits: 16-bit successive approximation ADC AD977A; 200 kSa/s; 16-bit sigma delta ADC AD7723; 1.2 MSa/s; 14-bit cascaded ADC AD9240; 10 MSa/s. The control part is based on the FPGA circuit, which enables the real-time control of the whole device. Its maximum data frequency was increased due to high-speed galvanic insulators (maximum frequency of 20 MHz). An internal crystal oscillator or an external signal can be used as a source of a sampling clock, whose frequency can be divided by adjustable ratio of 1 – 232. Measured data are sent to TTL or differential LVDS drivers, or, in case of histogram mode or RS-232 transfer (up to 115 kbit/s with hardware handshake), they are saved into the SRAM (up to 256 kSa) first and then transmitted. The microprocessor provides the LCD and keyboard control as well as the serial RS-232 communication. To prove that the AD transfer standard does not change its parameters during the transport, testing measurements were executed both before the transport and after of it. Both measurements were compared and their results were nearly identical, which confirmed the stability of the AD transfer standard. II. ACHIEVED RESULTS A. Results of comparison using ADTS1 The ADC module with the AD976A 16-bit successive approximation ADC (maximum sampling frequency of 200 kSa/s) was used in ADTS1. The basic results of the FFT test (SINAD, THD and SNR) are summarized in Table I. The results were computed from the 32 kSa records, which is the maximum length of the data record from one of the laboratories. To eliminate leakage, all the values were computed by means of the FFT test and Blackman-Harris 7 term window. This window provides a sufficient suppression ratio of its side lobes for testing 16-bit ADC, and the main lobe is still not too wide. Due to the window length, 6 bins adjacent to the fundamental harmonic are eliminated for the SINAD (SNHR) computation. Possible instability greater than 6 frequency bins around the fundamental harmonic is considered to be noise. It appeared in case of the AP Syst. II generator, which seems to be worse in SINAD (SNHR). Nevertheless, this generator has an excellent THD. It is necessary to mention that all the measured values are influenced not only by the properties of evaluated testing systems, but also by the quality of the transportable reference AD device. The results shown in Table I can serve for the comparison of the tested
parameters at the evaluated generators, nevertheless they say nothing about their absolute parameters. Further, it should be stated that the measured spectrum for the fundamental frequency above 30 kHz can be partly influenced by the passive antialiasing filter, used in the input module of the AD device. Not only the basic parameters (see Table I), but also the short-time instability of generators can be estimated. A method of its detection is mentioned bellow. Providing the very good short-time stability of the gain of the input amplifier and the sampling frequency of the AD transfer standard, the best sine wave curve fit test can also be used for the evaluation of the short-time instability of the testing signal. After applying this method it is easy to compute residuals (differences between measured and fitted waveforms) and to observe their time dependence corresponding to the instability of the used generator. The typical results are shown in Fig. 2. There is no visible instability in case of Fig. 2a. Even though the results in Fig. 2b indicate instability, it is not possible to evaluate its type (either the amplitude or the frequency).
a) Generator without visible instability
b) Generator with visible instability
Fig. 2. Values of residuals of all samples
The type of instability can be identified after recalculating the residuals of all samples to one period of the testing signal. The positions of the residual maxima depend on the type of the generator instability [5]. If the residuals achieve their maxima in the location of both positive and negative maxima of the testing signal, the amplitude instability is detected. If the maxima of
Table I. Values of SINAD, THD and SNR measured with DS 360, B&K 1049, R&S UDP and AudioPrecision System II generators; sampling rate 156 kSa/s, data record 32 kSa, THD calculated from the first 20 harmonics (modified according the remarks from PTB)
frequency (kHz) 10.333 20.333 50.333 100.333
DS 360 86 85 81 73
SINAD B&K R&S 1049 UDP 85 84 77 73
85 84 80 74
AP Syst.II
DS 360
86 85 82 –
–98 –99 –87 –80
THD B&K R&S 1049 UDP –97 –90 –78 –73
–97 –97 –85 –77
AP Syst.II
DS 360
–98 –97 –99 –
86 86 82 75
SNR B&K R&S 1049 UDP 86 86 84 81
86 84 81 76
AP Syst.II 86 86 82 –
the residuals are located near the zero crossing of the testing signal, the frequency instability is detected. The residuals displayed in Fig. 3a correspond to the signal with the amplitude instability, while the residuals displayed in Fig. 3b correspond to the frequency instability. The described method enables an easy detection of small short-time instabilities, which could be hardly discovered by means of other methods.
a) Generator with amplitude instability
b) Generator with frequency instability
Fig. 3. Residuals of all samples recalculated to one period of the testing signal.
B. First experience with ADTS2 The first comparative measurements using the new AD transfer standard ADTS2 with the AD module—AD977A evaluation board (16 bit, 200 kSa/s)—were executed at DS 360 generator and AudioPrecision System II generator. In this case both AD transfer standards were used. The results of the basic FFT test (SINAD, THD and SNR) using 128 kSa records are shown in Fig. 4. They confirm that the parameters measured by the ADTS2 are the same or a little bit better than those measured by the old one. The imperfections of the first prototype, e.g. problems with coherent sampling, data transmission and influence of a power supply source, were not proved. Especially, the frequency instability has bigger influence on the results because of higher record length (see section A). Besides the traditional measurements mentioned above, an analysis of undesirable components was performed. For this purpose, 16 data files were measured for each frequency. Then the spectrum for each file was calculated and all 16 spectra were averaged (see Fig. 5). The undesirable components can be divided into two groups: first, harmonic and spurious components, second, random noise. Random noise is partly suppressed (by ~6 dB in this case) by averaging, and the harmonic
components corresponding to nonlinearities and spurious components are stressed. The averaged spectrum could simplify the distortion analysis but it does not enable to determine e.g. SINAD or ENOB of the ADC or the generator.
88
-75 DS360 - ADTS1 DS360 - ADTS2
-80
AP-Syst.II - ADTS1
84
AP-Syst.II - ADTS2
THD (dB)
SINAD (dB)
-85
80
-95
DS360 - ADTS1 76
-90
DS360 - ADTS2 -100
AP-Syst.II - ADTS1 AP-Syst.II - ADTS2 72
-105 10
30
50
70
90
110
10
f (kHz)
30
50
70
90
110
f (kHz)
a) SINAD
b) THD
88
SNR (dB)
84
80
DS360 - ADTS1 76
DS360 - ADTS2 AP-Syst.II - ADTS1 AP-Syst.II - ADTS2
72 10
30
50
70
90
110
f (kHz)
c) SNR Fig. 4. Frequency dependence of SINAD, THD and SNR of the old and new AD transfer standards measured with DS 360 and Audio Precision System II generators
Fig. 5. Averaged frequency spectrum of the input signal with frequency fin = 50.333 kHz
C. The first experiments with 14-bit 10 MSa/s AD module The AD module with AD9240 ADC (14 bit, 10 MSa/s) was also applied in the ADTS2. First tests showed a certain problem concerning the basic level of noise probably caused by the noise of the voltage reference. Since no commercial low-distortion generator with the frequency range up to several MHz is suitable for 14 bit ADC testing, the generators SR DS360 and HP33120A (or Agilent 33250A) were used in the first experiments. Both of them are based on the principle of the direct frequency synthesis. The SR DS360 is the low-distortion generator convenient for ADC testing (see above), still with the frequency range up to 200 kHz only. HP33120A generates the sinewave with the frequency up to 15 MHz; nevertheless the distortion of the output signal (arisen due to a low resolution of an output DAC and a standard low-pass filter) is too high for 14 bit ADCs testing. The results of the first experiments are summarized in Fig. 6. Non-coherent sampling, 7-term Blackman-Harris window and sampling rate of 10 MSa/s were used. It is evident that the objective of the device is just to compare the testing systems, not the absolute measurements. Comparisons of some properties of the testing signal generators can be made from the frequency spectra shown in Fig. 7.
-50 SR DS360 HP 33120A
THD (dB)
-60
-70
-80
-90 10
100
1000
10000
f (kHz)
Fig. 6. Measured values of THD (for the first 7 harmonic components) for generators SR DS360 and HP 33120A
a) Generator DS 360
b) Generator HP 33120A
Fig. 7. Measured frequency spectrum of input signal with frequency fin = 100.33 kHz
There is no antialiasing filter used in the applied AD module. It is proved by the increasing noise level on the high frequencies in the frequency spectrum at the low-distortion generator SR DS360. It corresponds to the results of the former measurements, which showed the significantly high frequency components in the output signal of this type of low-distortion generators [6]. Since the generator itself has a low harmonic distortion [7], the measured values of THD evidently correspond to the quality of this parameter of the AD transfer standard. On the other hand, the amount of higher harmonic components obviously increases with the frequency of the output signal at the HP 33120A generator. To make the identification of undesirable components easier, the spectrum averaging was applied. The effect is illustrated in Fig. 8. The averaging from 64 records decreases the noise floor by 8 dB. The input frequency was chosen, so that the higher harmonic components (from 8th) would overlap the previous harmonics (up to 7th). Under this condition it could be stated that
no spectral line is of the harmonic origin, except the seven harmonic lines. It is very probable that the non-harmonic spectral lines in Fig. 8b) are caused by the digital direct synthesis of the used generator as a side effect.
a) Without averaging
b) Averaging of 64 records
Fig. 8. Frequency spectrum of the input signal with frequency fin = 1333.333 kHz— generator Agilent 33250A (the actual level of fundamental harmonic component = 0 dB !)
III.
CONCLUSION
The comparison measurements by the transportable high-stable reference AD device enable both comparison and evaluation of different ADC testing systems. The measurements performed by now confirmed the applicability of this method, which is used for an international comparison of various calibration devices. The replaceable AD modules allow an application of the testing signal in the frequency range up to 5 MHz. The satisfactory results were also achieved for the testing signal in the frequency range up to 100 kHz using 16-bit successive approximation ADC AD976A (200 kSa/s). The first experiments with the AD module with the 14-bit cascaded ADC AD9240 (10 MSa/s) were also performed. This module enables to evaluate the systems using testing generators with an output signal frequency up to 5 MHz. First results proved its functionality. The present interest is to decrease its relatively high noise. ACKNOWLEDGEMENTS Authors would like to thank Dr. D. Dallet from University Bordeaux, Dipl. Ing. R. Ižak and Dr. V. Schulze from IMMS Ilmenau for their support in executing the comparative measurements. This work has been supported by the research project MSM 210000015.
REFERENCES [1]
J. Roztočil, J. Brossmann, V. Haasz: Advanced Method for ADC Testing Generators Verification. 7th Biennial Conference on Electronics and Microsystem Technology BEC’2000, Tallinn, October 2000, pp. 323-326.
[2]
V. Haasz, J. Fischer, J. Novák, J. Vedral: Transportable AD Box for Comparative Measurement. IEEE Instrumentation and Measurement Technology conference, Budapest, May 21-23, 2001, pp 671-674
[3]
J. Fischer, V. Haasz, D. Slepička, J. Vedral: Transportable Reference AD Device – New Innovated Version. Accepted paper for XVII IMEKO World Congress, Dubrovnik, June 22−27, 2003,
[4]
V. Haasz, J. Roztočil, D. Dallet, D. Slepička: Comparison of Parameters of Systems Used for AD Converters and Modules Testing. ADDA&EWADC 2002, Prague 2002, pp. 123-126
[5]
V. Haasz, J. Roztočil, D. Slepička: Evaluation of Short Time Instability of Generators Used for ADC testing, Accepted paper for XVII IMEKO World Congress, Dubrovnik, June 22−27, 2003,
[6]
V. Haasz, H. Schumny: Methods for Dynamic Quality Tests of AD-modules and Microcontrollers with Integrated AD Converters. Symposium IMEKO TC-4, Naples 1998, pp. 419-423
[7]
J. Roztočil, J. Brossmann, V. Haasz: Advanced Method for ADC Testing Generators Verification. In: BEC 2000 - Baltic Electronics Conference. Tallinn, 2000, pp. 323-326.
