Oscilloscope memory depth: when bigger is not always better

Scope Architecture
What makes one scope “designed” for deep memory while another has to default its memory to 10K to remain responsive?  A lot of it comes down to the oscilloscope architecture.  In some scopes, the CPU system is an integral block in the oscilloscope architecture (“CPU-based architecture”), so much so that it is actually the gating item in how fast the scope can process the information and display it to the screen.  If the CPU system isn’t up to the task of handling deep memory acquisition records, it will lengthen the time it takes to process and display the data, therefore lowering the update rate of the scope (sometimes dramatically).  See Figure 1 for an example of this architecture.



Fortunately, there is another way. The scope designed for deep memory uses a custom ASIC that eliminates the need for the oscilloscope’s CPU to be an integral part of the architecture.  Is there still a CPU system?  Of course, but it is now used for peripheral crunching of data, which allows the scope to focus on what it does best: display waveforms.  Figure 2 shows an example of this architecture in the DSO-X 3000-series from Agilent which uses an ASIC (called MegaZoom IV) to provide fast update rates while maximizing memory and sample rate.  



Memory and the oscilloscope’s architecture are so intertwined that there are some things that even defaulting to a base memory depth of 10K can’t fix.  For example, one of the best enhancements in scopes in the last 15 years is the addition of digital channels, but not all digital channels are implemented the same.  In the CPU-based architecture discussed above, turning on the digital channels will actually cause such a slowdown in the oscilloscope that the update rate will never get above 135 waveforms per second regardless of the memory depth or time base setting.  That is orders of magnitude slower than the maximum update rate the manufacturer specifies.  Why is that?  Again, it goes back to the oscilloscope architecture.  As you can see in Figure 1, the MSO channels are not well integrated in to the CPU-based architecture, which requires the CPU system to have a larger role in plotting them.  With the MegaZoom architecture (Figure 2), the digital channels are an integral part of the custom ASIC that is doing the plotting and displaying of all the channels.  In the MegaZoom architecture the system will not slow down due to turning on digital channels.  Other common things like Sinx/x interpolation can also slow down a CPU-based system – so much so that you’ll see dramatic drop offs in update rate when moving between time-base settings as the scope switches on and off Sinx/x interpolation.  

Responsiveness of the scope is another drawback to a CPU-based system.  Have you ever changed the time-base setting on your deep memory scope and then waited for it to catch up?  Or tried to update a setting only to have it respond slowly and you accidentally click past the setting you were trying to get?  That is because the CPU system is trying to crunch through the data. The same issue that causes the update rate to slow is also causing the scope’s responsiveness to slow.

So far all we’ve discussed are modes when the oscilloscope is running and being used for something like debug. If you are just looking at a single-shot acquisition, deep memory is better again right?  You don’t need a fast update rate with a single-shot acquisition, and the responsiveness of the scope should be better once it is all captured and displayed.  Again, this would seem like a logical conclusion, and in some cases this is true.  But what if you are looking at a signal that has bursts of information with a significant amount of idle time in between (like a radar pulse or a serial bus sending frames/packets)?  With a traditional deep memory oscilloscope, you would be using all that deep memory to digitize the idle time and the bursts – this is not the best use of the memory since you probably only care about the signal bursts themselves.  Some oscilloscopes offer a memory system that has an option called “segmented” memory.  Segmented memory allows you to digitize just the portion of the waveform that you care about so you can make more efficient use of your deep memory.