Memory

Some Key Questions Concerning the Value of Long Memory in Oscilloscopes:

(1) How much memory per channel is available for capture of signals?

(2) Will the scope automatically maximize the sample rate and thus record more data into memory as the user requests longer timebases?

(3) How long does it take for the scope to acquire a long record and display it (i.e. what is the screen update rate when capturing long data records)?

Digital scope technology is evolving rapidly.

Scopes have always been about showing you what is really going on with your signals. Sample rate and BW specifications used to tell the majority of the scope performance story. Then everything changed. Memory went from 50 k to 50 M, then to 100 M. Views of signals got longer than you would guess. To appreciate how long, think miles of scope screens, hundreds of miles, per trigger!

Deep Memory Measured in Miles? In an Oscilloscope?

Suddenly your new "signal viewing tool" isn't just showing you a few cycles - its showing you the variations in your critical measurements through data chunks that stretch out over 125 miles of contiguous scope screens per trigger. With 100 M samples of 20 GS/s data, the sweep is 5 milliseconds of single shot recording. That is about a 2 hour scroll @ 65 mph to view all the screens. Some things are obviously best done with fast CPUs. Inspecting 125 miles of scope data should be near the top of your list.

Fast Forensic DSP is Essential

Once your scope's front end specifications are clearly defined, performance is all about memory depth combined with smart and fast DSP algorithms to find and measure the exceptions from the ideal in your signals. With a high speed architecture, you receive insights to solutions in seconds instead of days, weeks, or never.

It used to be about looking at ONE fuzzy edge and wondering if 1,000 triggers were enough data for a jitter measurement (like looking 1000 times for a train on the tracks, seeing no trains, and deciding that no trains exist). Now defects can be captured and automatically detected and cataloged from extremely deep records (up to 100 Mega-Samples) of signal data per trigger. One shot could contain 15 million measurements per parameter each with complete statistics.

Different Memory Lengths = Different Displays

Here are 3 examples of scope data display determined by memory depth, each with their own characteristics.

(1) Cycle by Cycle - Traditional - Memory depth is from 50 to 5,000 samples and you are looking at 1 or more cycles on screen where each cycle is clearly defined. You can closely examine each cycle to verify heights, widths, and shapes visually and with automated measurements. Additional DSP routines could show FFT (spectral) or statistics on several acquisitions but limited to scope's trigger rate. Could take several minutes to hours to become statistically relevant as 1 to 5 million measurements are usually suggested.

(2) Small Block - Advanced - Memory depth is about 1 M and you are looking at 1 or more packets on screen where each packet is captured in the primary trace. You can closely examine each cycle of the packet with a ZOOM trace to verify heights, widths, and shapes visually and with automated measurements. After scrolling through a few of these, you'll enjoy pass/fail testing to find exceptions as scrolling and trying to find things manually redefines tedious!

(3) Large Block - Newest - Memory depth is about 10 to 256 M and you are looking at densely packed data on each screen of the primary trace. You could closely examine each cycle of the record with a ZOOM trace to verify heights, widths, and shapes visually and with automated measurements. But DSP derived waveforms can find your exceptions automatically. They include jitter tracks, parameter tracks, (trends of each instance of your measurement), histograms, unit interval bit cell slicing with software based clock data recovery, or sequences of bursts displaying overlayed, waterfalled, etc. segments.

The above trace contains 40 Million samples of a 500 MHz clock. Note there are almost 1 million measurements of period at level and rise times. Modern scopes can measure every instance in long arrays of signal data and present statistics. The histograms for each measurement reveal additional insights about your signal's true characteristics.
Large Block data records trump triggers per second in short memory scopes. Its about processing speed with deep memory arrays to get the clues that bring the real answers fastest. Compare DSP benchmarks. Processing Speed Rules!

Deep Memory Reveals Lower Frequency Issues

High speed clocks and data streams are most often effected by lower frequency phenomonena such as power supply noise. Capture 1 full cycle or more of the lowest frequency that modulates your signal and you can detect problems with a single trigger acquisition.

View

(1) Can I zoom into portions of a long waveform to see important details and at the same time still view the entire signal?

(2) If I record long/complex signals on all four channels of my scope, is it possible to simultaneously have a zoom portion displayed from each of the four inputs?

(3) Can I use the entire screen area of the scope to view my signal?

(4) Will the scope sort through all the data and find the max/min points for each portion of the waveform to display on the screen? Or will the view of the waveform be a randomly chosen set of points?

Measurements

(1) Will signal parameters be measured only on the first pulse width, period, amplitude, etc or does the scope use more of the data in the memory to make measurements?

(2) Will the FFT function measure the frequency content of all the data in memory, or of any selected portion? Or is it confined to measure only a small piece of the acquired data?

Analysis

(1) Are there worst case parameter measurements that use data in the long waveform to find worst case signal characteristics? Or do the parameter statistics only use data from a small piece of the waveform?

(2) Can I draw a bar chart (histogram) that gives an analysis of key signal characteristics?

(3) Can I draw a trend line that shows the time related behavior of signal parameters?

(4) Can I use Math functions on the entire waveform?

(5) Can I perform multilevel math (like squaring a signal then integrating to get the total power)?

Documentation

(1) Is it easy to print out a long landscape mode hardcopy that gives a full view of waveform details from a long record?

(2) How long does it take to send a long waveform from the scope to a computer?

(3) How long does it take the scope to convert a binary file to ASCII (for use in spreadsheet type programs)?

(4) Is there a portable storage device available for the DSO that allows long waveforms to be stored on a portable media that can then be used in a computer?

Caution!

Manufactures specify their largest numbers. Look out for some scope's acquisition memory values that divide by the number of active channels.

Caution!

Don't confuse acquisition memory with reference memory specifications. Reference memory is used for copies of waveforms recorded earlier and made available for comparison either by viewing or by mathematics. Some scopes have longer acquisition memories than reference memory. Compare.

Not all scopes have the same number of reference memories. More are better. Make sure they have the width to contain your processed data. You can't store 12 bit averaged waveform data in 8 bit reference memory. Compare.

Not all scopes have the same number of memories for front panel set-up/storage and recall. Compare.

Some scopes allow for Memory Segmentation where you can acquire multiple trigger events in a single sweep display. Some scopes will also record the time of each trigger event that occurred.
Return to Scope FAQ Main Menu