AD converter fidelity test to verify purity

　　The ability to accurately digitize a sine wave is a sensitive test of the fidelity of a high-resolution AD converter. This test requires a sine wave generator with a residual distortion component close to 1ppm. In addition, a computer-based AD output monitor is required to read and display the converter output spectral content. If this test is to be performed at a reasonable cost and complexity, its components must be designed and performance verified before use.

　　summary

　　A schematic diagram of the system is shown in Figure 1. A low distortion oscillator drives the AD through an amplifier. The AD output interface formats the converter output and communicates with a computer that executes the spectrum analysis software and displays the resulting data.

　　Oscillator Circuit

　　The oscillator is the most challenging circuit design part of the system. To meaningfully test the 18-bit AD, the oscillator impurities must be extremely low and these characteristics must be verified by independent means. Figure 2 is essentially a "full inverting" 2kHz Wien bridge design (A1-A2) based on research done at Harvard University's Winfield Hill. The J-FET gain control of the original design is replaced by an LED-driven CdS phototransistor isolator, eliminating the errors caused by the J-FET conductivity modulation and the need for fine tuning to minimize these errors. Band-limited A3 receives the A2 output and a DC offset bias and provides an output through a 2.6kHz filter to drive the AD input amplifier. The automatic gain control (AGC) signal for the A1-A2 oscillator is taken from the circuit output ("AGC sense") by AC-coupled A4, which feeds rectifiers A5-A6. The DC output of A6 represents the AC amplitude of the circuit's output sine wave. This value is balanced with the LT-1029 reference by current summing resistors terminating in AGC amplifier A7. A7, driving Q1, closes the gain control loop by setting the LED current (and therefore the CdS phototube resistance), thereby stabilizing the amplitude of the oscillator output. Although this attenuates the band-limited response of A3 and the output filter, obtaining gain control feedback information from the circuit's output maintains output amplitude. It also places demands on the A7 loop closure dynamics. Specifically, the band-limiting of A3 combines with the hysteresis and ripple-rejection components of output filter A6 (in the base of Q1) to produce a significant phase delay. A 1μF dominant pole on A7, along with an RC zero, provides this delay, resulting in stable loop compensation. This approach replaces a tightly tuned high-order output filter with a simple RC roll-off filter, thereby minimizing distortion while maintaining output amplitude.

　　Eliminating the oscillator-related components from the LED bias is key to keeping distortion low. Any such residual noise will adjust the amplitude of the oscillator, thus introducing impurities. The band-limited AGC signal forward path is well filtered, and the large RC constant in the base of Q1 provides the final steep roll-off. As shown in Figure 3 (emitter current of Q1), the oscillator-related ripple is approximately 1nA (less than 0.1ppm) at a total current of 10mA.

　　The oscillator achieves its performance with only one trimming adjustment. This adjustment (which determines the center of the AGC capture range) is set according to the schematic annotation. [page]
　　Verifying Oscillator Distortion

　　Verifying oscillator distortion requires sophisticated measurement techniques. Attempting to measure distortion with a conventional distortion analyzer (even an advanced model) will encounter limitations. Figure 4 shows the oscillator output (Trace A) and its residual distortion indication at the analyzer output (Trace B). The oscillator's relative motion is blurred within the analyzer's noise and uncertainty floors. The HP-339A used in the test specifies a minimum measurable distortion of 18ppm; the instrument indicated 9ppm when this photo was taken. This is beyond the specification and is highly suspect, as distortion measurements at or near the performance limits of the device introduce significant uncertainty.2 If meaningful measurements of oscillator distortion are to be made, a sophisticated, professional analyzer with a very low uncertainty floor is necessary. The Audio Precision 2722, which specifies a 2.5ppm total harmonic distortion + noise (THD + N) limit (1.5ppm typical), provided the data in Figure 5. As shown in this figure, the total harmonic distortion (THD) is -110dB, or about 3ppm. Figure 6 (obtained using the same instrument) shows the THD + N to be 105dB, or about 5.8ppm. In the final test shown in Figure 7, the analyzer determined the spectral content of the oscillator (dominated by the third harmonic at -112dB, or about 2.4ppm). These measurements provide confidence in applying this oscillator to AD fidelity characterization.

　　AD Testing

　　The AD test sends the oscillator output to the AD through its input amplifier. This test measures the distortion components produced by the input amplifier/AD combination. The AD output is checked by a computer, which quantitatively indicates the spectral error components in the display of Figure 83. The display includes time domain information (which shows the biased sine wave centered within the converter's operating range), a Fourier transform (indicating the spectral error components), and detailed tabular readouts. The LTC-2379 18-bit AD/LT6350 amplifier combination tested produced a second harmonic distortion of -111dB (about 2.8ppm), while higher frequency harmonics were well below this level. This indicates that the AD and its input amplifier are operating correctly and within specifications. To achieve harmonic cancellation between the oscillator and the amplifier/AD, it is necessary to test multiple amplifier/AD samples to increase the confidence of the measurement4.