The advanced digital oscilloscope
A 5 GS/s oscilloscope with 10,000 points of memory samples at 1 MS/s when it is set to 1 ms per division. The memory, not the converter, decides what you see.
01Three numbers that define a digital oscilloscope
The digital oscilloscope of Measurements I was described by its bandwidth and its sampling rate. An advanced instrument is judged by three numbers that interact: the analog front-end bandwidth, the maximum sampling rate of the converter and the memory depth, the number of samples it stores in one acquisition. With sin(x)/x interpolation, it is necessary to sample at about 2.5 times the bandwidth to reconstruct a full-bandwidth signal well.
The memory is the one that tends to be forgotten. The screen shows ten horizontal divisions; to fill them, the oscilloscope needs to store samples during that whole time:
A 500 MHz, 5 GS/s oscilloscope. As the timebase is lengthened, the actual sampling rate depends on the memory. A 100.03 MHz signal is observed: if half the sampling rate falls below it, an alias appears.
02Advanced triggering and segmented memory
Finding a rare event is harder than seeing it. Advanced oscilloscopes trigger on conditions: a pulse narrower or wider than a value, a runt pulse that does not reach a logic level, a violation of setup and hold times between data and clock, or a particular field of a decoded serial frame. Segmented memory divides the memory into thousands of short acquisitions, one for each trigger, and stores the time of each one: it is useful for recording bursts separated by long dead times without wasting memory.
Another important specification is the update rate, in waveforms per second. Between one acquisition and the next the instrument processes and does not see the signal: with 1000 waveforms per second and a 1 μs window, it is blind 99.9% of the time. High-rate instruments, with hundreds of thousands per second, find intermittent faults much sooner.
03Measurements, FFT and vertical resolution
The automatic measurements (rise time, frequency, RMS value) are calculated on the samples, and their accuracy depends on the number of points per edge. The FFT converts the record into a spectrum with a frequency resolution equal to the inverse of the recorded time: deep memory means fine resolution. The 8-bit converter gives 256 levels; in high-resolution mode, the instrument samples faster than necessary and averages neighboring samples, gaining half a bit for each doubling of the averaging, at the expense of bandwidth. The effects of sampling were studied in aliasing.
04In the laboratory
Observe a 50 MHz signal with minimum memory and increasing timebases. Note the sampling rate reported by the instrument and the point at which the waveform stops being the real one.
Capture bursts of a serial communication with segmented memory and measure the time between them using the time stamps.
Generate an occasional runt pulse with an arbitrary function generator and find it with the runt-pulse trigger and with infinite persistence.
05Common mistakes
- Looking only at the maximum sampling rate. With long timebases, the memory reduces it.
- Trusting the screen without looking at the actual sampling rate. An alias can look like a stable signal.
- Hunting for rare faults with a low update rate. The instrument does not see the signal most of the time.
- Using high resolution with fast signals. Averaging reduces the bandwidth.
06Self-assessment
What sampling rate does a 1 M-point oscilloscope use at 100 μs/div?
\( 10^6/(10 \cdot 10^{-4}) = 1 \) GS/s.
What resolution does the FFT of a 10 ms record have?
100 Hz.
How many effective bits are gained in high resolution by averaging 16 samples?
About 2 bits: half a bit for each doubling.
What is segmented memory for?
For recording many short events separated by long dead times, without spending memory on the dead times.
07Further reading
- Tektronix. XYZs of Oscilloscopes. Tektronix, introductory note. The concepts of sampling, memory and triggering explained by a manufacturer.
- Keysight Technologies. Evaluating Oscilloscope Sample Rates vs. Sampling Fidelity. Application note, Keysight. Why sampling at twice the bandwidth is not enough.
- Clyde F. Coombs Jr. (ed.). Electronic Instrument Handbook. 3rd ed., McGraw-Hill, 1999. The digital oscilloscope and its acquisition modes.