Analog-to-Digital Converters (A/D converters) are capable of quite impressive noise and distortion performance. However, even the best signal generators and synthesizers produce signals that are not spectrally pure enough to evaluate noise and distortion performance of A/D converters. Signal sources can exhibit both harmonic distortion and noise levels that are worse than the distortion and noise performance of the A/D converter, making the A/D converter performance appear worse than it really is. The old adage of "Garbage In - Garbage Out" certainly applies here. In this article, we will see the definite need to use an input filter when evaluating A/D converter performance.
The plots of Figure 1 show the difference between two signal sources (the KronHite model 2400B and HP model 8644B). Note the dynamic performance parameters at the top right of each plot. Both generators were set for an output frequency of 9.6 MHz and the sample rate was 40 Msps. The same channel of a single dual 10-bit A/D converter (A/D converter10D040 from National Semiconductor) and evaluation board was used in both cases. Both of these signal sources are very good pieces of equipment, but they cannot produce a signal that is "clean" enough to adequately evaluate today's monolithic A/D converters.
Figure 1. FFT plots of the output of a 10-bit A/D converter with two different signal sources and no A/D converter input filter, Using KronHite model 2400B Signal Generator (top) and Using HP model 8644B Signal Generator (bottom).
Now note the results seen in Figure 2 when using these same generators, but with a 10 MHz, 5-pole elliptical bandpass filter between the signal source and the A/D converter input. All dynamic performance parameters (SINAD -- Signal to Noise And Distortion, SNR -- Signal to Noise Ratio, THD -- Total Harmonic Distortion and SDFR -- Spurious-Free Dynamic Range) are improved with the use of an adequate A/D converter input filter.
Note also that, while the filter improves the apparent A/D converter performance with either signal source, there are still differences between the performances noted with the two signal sources. While the need for the filter is obvious, the filter did not eliminate all noise and distortion products contained in the signals. The need for a good signal source and a bandpass filter becomes obvious when we compare Figures 1 and 2.
Figure 2. FFT plot of the output of the A/D converter10D040, a dual 10-bit A/D converter; with two different signal sources and a 10 MHz input filter, Using KronHite model 2400B Signal Generator (top) and Using HP model 8644B Signal Generator (bottom).
A bandpass filter is used rather than a low pass filter as any noise below the lower filter cutoff frequency is thus attenuated, enhancing SNR.
These plots were all taken with National Semiconductor's evaluation board for the A/D converter10D040 dual 10-bit, 40 Msps Analog-to-Digital Converter and National Semiconductor's WaveVision software. This software is available as a free download from National Semiconductor's web site at http://www.national.com/adc.
There are a few things to note from the plots here:
1. All dynamic parameters at the A/D converter output are better with the use of the A/D converter input filter.
2. The outputs of both signal sources contain many harmonics and spurs that the filter reduces.
3. While spurs from the generator are reduced with the use of an A/D converter input filter, they may not be completely eliminated with that filter.
4. Even with the use of an input filter, the performance noted with the two signal sources is different. The filter cannot reduce phase noise.
5. The ENOB (Effective Number of Bits) using the linear generator went from 6.9 without the filter to 9.5 with the filter, while the ENOB of the synthesizer went from 8.1 without the filter to 9.4 with the filter.
The irony in this last point is that the filter made the signal from the "worst" generator improve so much that it is better than the signal of the "better" source. While the THD observed with the two generators are about the same when using the filter, the observed SNR with the linear generator and filter is better than with the synthesizer and filter. This can result from higher phase noise in the synthesizer than in the linear generator. A filter cannot eliminate phase noise.
The outputs of even the best signal sources available exhibit more noise and distortion than does the A/D converter, so the signal source output needs to be cleaned up before being presented to the A/D converter under test. ALWAYS use a bandpass filter before an A/D converter when evaluating it. The higher the A/D converter resolution, the more important is this filter. The plots here show that there are harmonics, spurs and noise that will reduce the apparent performance of the A/D converter. Without an adequate A/D converter input filter, you may be looking more at the noise and distortion characteristics of the signal source than that of the A/D converter.
Nicholas "Nick" Gray is currently working on the development of new products in the area of high speed/high resolution Analog-to-Digital Converters. His career includes video circuit design and 19 years as an applications engineer for data conversion products.