DSP performs magic in scopes, but don’t ignore the tradeoffs
An expert viewpoint brought to Electronic Design by Agilent Technologies, Inc.
When I set up a measurement and look at my scope, the first thing that comes to my mind isn't that the display might have some inaccuracies. I intuitively trust that my name-brand instrument delivers accurate readings. However, with scopes spec'ing bandwidths as high as 6 GHz and beyond, we must acknowledge that they're pushing the boundaries of what's possible with today's components. Fortunately, scope manufacturers have come up with signal-processing methods that compensate for the shortcomings of real-world components. Some engineers are uncomfortable with the idea that the scope is "massaging" the data prior to display, but you simply need to understand what the DSP algorithms are doing and the tradeoffs they involve.
In a perfect world, realtime scopes would have infinitely fast sample rates and bandwidth, perfectly flat frequency responses, linear phase responses and no noise. With ideal components we could achieve these goals, but welcome to the real world. To some degree, especially at very high frequencies, a scope's hardware will distort the signal in some way. Last month's column looked at the effect of probes and what instrument designers have done to improve signal acquisition at the front end. This month I'll describe some of the considerable activity going on within the scope itself, much of it software based.
This concept isn't brand new. In fact, for more than 20 years manufacturers of digital scopes have provided software assistance to improve readings. For instance, waveform-reconstruction filtering is relatively common. While equivalent-time scopes use repetitive sampling to fill in datapoints to improve timing resolution and waveform details, real-time models employ sin(x)/x software reconstruction to fill in datapoints between evenly spaced realtime samples. With this second approach, though, the input must not have any frequency components beyond the Nyquist frequency (Fs/2), otherwise measurements of digitized points can be in error by literally orders of magnitude. And because scopes don't have brickwall filters to remove all traces of frequencies higher than the Nyquist, you must further reduce the bandwidth to stay in the safe range. In practice, the bandwidth of a scope that samples at 20 GHz while using waveform reconstruction shouldn't exceed 8 GHz.
Another type of software filter performs phase correction to address problems when scope hardware adds unwanted phase shifts to higher order components of high-speed signals. A phase shift even in the 5th harmonic of a high-speed digital signal induces excessive overshoot and produces high measurements of rise and fall times. Phase-correction filtering can compensate for these hardware-induced errors.
Most engineers understand and appreciate such software filtering, but they sometimes have trouble accepting other methods, especially those that manipulate the scope's frequency response. Why would this be necessary? A scope's hardware never gives a perfectly flat response, especially at higher frequencies. Instrument designers compensate for irregular responses by adding hardware that induces amplification or attenuation in certain regions, but even this doesn't result in a completely flat response. With software filters that mirror hardware characteristics, however, scopes can fine-tune this performance.
Some manufacturers aren't content with a flat response just in the standard range, so they take the mirror-image filtering concept even further to extend the effective bandwidth of high-end scopes beyond what the hardware is designed for. In a 6-GHz scope, for instance, this method can almost magically extend its bandwidth out to 7 GHz. At the same time, the filter also generates a sharper rolloff characteristic, which reduces hf noise and helps eliminate aliasing when testing signals with out-of-band components.
When they hear of this feature, many users become suspect: how can you suddenly recreate with software a signal that the scope's hardware has already attenuated out? This all sounds too good to be true, so there must be tradeoffs, right? Absolutely - in this case the big tradeoff being that the instrument also amplifies noise along with the signal, and the S/N ratio degrades considerably at higher frequencies. Thus users generally have the option to turn on this feature depending on their measurement goals.
These software enhancements are only the start. Scope manufacturers will continue to leverage new component technology to further increase the native capabilities of their instruments. And their software engineers will always find ways to improve and extend these capabilities. Be sure to take advantage of both.
To learn about the advantages and disadvantages of using DSP filtering on oscilloscope waveforms, download the new, free Application Note 1494 at
www.agilent.com/find/curveol-aug04
ADVERTISEMENT