Design of a mixed-signal oscilloscope

Hewlett-Packard Journal, April, 1997 by Matthew S. Holcomb, Stuart O. Hall, Warren S. Tustin, Patrick J. Burkart, Steven D. Roach

The Decimation Problem

Even with a million points of sample memory, at the slower time base settings, there will still be more samples across the screen than can be stored. We need to prune (decimate) the available data down to the memory size, then further thin the stored data down into 4000-point records for rapid display. In these situations, when the ADCs are sampling faster than the data can be used, the oscilloscope is said to be oversampling. There are a variety of techniques available to perform this decimation process, each portraying a different representation of the input signal. Some methods will tend to highlight the subtle variations in the signal, others show the signal extremes, and others hide the signal extremes. As an example, suppose that the sample rate of the ADC is a factor of 100 greater than the desired store rate. Fig. 3 shows various decimation techniques for 100:1 oversampling.

[FIGURE 3 ILLUSTRATION OMITTED]

Simple Decimation. One sample is stored and the next 99 ignored; then the next sample is stored and the next 99 ignored, and so on. This approach is illustrated in Fig. 3b. Since it is the simplest to implement, it is the most common decimation technique in digital oscilloscopes. Unfortunately, because of the very regular spacing of the stored samples, it is exceptionally prone to aliasing. Therefore, the HP 54645A/D rarely use this technique (only when calculating FFTs, when exact sample spacing is called for).

Intra-Acquisition -Dithering. Rather than store the first sample during the sample interval, this patented technique stores a randomly selected sample for each interval. In Fig. 3c, sample #33 is stored during the first interval, then #69 during the second, then #2 in the third, and so on. This technique, used in all of the 546xx oscilloscopes, is remarkably effective against aliasing. The stored samples are not evenly spaced, so it becomes much more difficult for the sampled signal to "lock" to the samples. The HP 54645A/D oscilloscopes use this technique with an improved folding pseudorandom number generator. It is used whenever the instrument is in the normal display mode. For more information on this sampling technique, see the article on page 26.

Peak Detection. Another common data compression technique is to store the minimum and maximum values of all of the samples in the sample interval, as shown in Fig. 3d. Widely used in digital oscilloscopes, this technique is called peak detection. When this technique is in use, rare, infrequent excursions are never ignored (lost), as they might be in the preceding two methods. This approach, however, tends to over-emphasize noise on the displayed record. Peaks are exaggerated and baselines become fatter at the expense of signal details. For example, an AM signal, peak detected, would show the modulation envelope quite clearly, but would lose the shape of the carrier wave. Statistical information about where the signal spends most of its time between the minimum and maximum values is lost.