Once again, oscilloscopes come to the rescue by applying sophisticated analysis techniques to instantaneous frequency-versus-time measurements.
Faster clock and data signals, along with more closely spaced signal lines on board layouts, have pushed the need for spread-spectrum clocking (SSC) to minimize radiated emissions. Because the spectral spreading is implemented as clock modulation, designers demand a method to verify the speed and shape accuracy of the modulation. Using the powerful tracking capability of real-time oscilloscopes, a frequency parameter track that's time-synchronous with the carrier allows for time-domain analysis of spread-spectrum modulation.
Like any acquired waveform, the track waveform can be digitally filtered to reduce noise. Three techniques exist to lower the track noise level to provide a very clean view of the signal's instantaneous frequency variation. The criterion for choosing which technique to use is a function of speed versus flexibility.
The persistence trace mean technique enables the establishment of the SSC main mode and provides the most rapid recovery of SSC with the least steps required. The combination of sparse and enhanced resolution discards noise to permit a sparse Gaussian low-pass filtered SSC response. Inline digital filtering techniques can be used to eliminate the noise riding on SSC, and choosing the filter enables the most flexibility in recovering spread-spectrum clock demodulation.
These techniques result in a new and powerful method to view modulation frequency and amplitude in the time domain, view the modulation separated from the carrier, qualitatively evaluate the shape of the modulation, measure the modulation's amplitude and frequency, and identify modulation anomalies.
Before delving into these new measurement techniques, it's a good idea to start with the basics. The goal of traditional clock design has been to produce a repetitive clock edge with optimal frequency stability. But the stable frequency of a standard clock produces a large peak power spike within a narrow frequency band. When the magnitude of this fundamental frequency's peak power is large enough, it can result in electromagnetic interference (EMI) on adjacent signals. This EMI can cause crosstalk, modulation, coupling, and in some cases, device failure. So, high-speed circuit designers use SSC to reduce radiated emissions. Distributing the fundamental frequency of the clock across a wider frequency range reduces peak spectral power and results in fewer radiated emissions.
Traditionally, oscilloscopes haven't been used to test SSC. Because spread-spectrum clocking is a frequency-domain effect, a spectrum analyzer would commonly be used for this purpose. The spectrum analyzer displays frequency-domain information as amplitude versus frequency, where amplitude (usually in dB) is displayed on the Y-axis and frequency is displayed on the X-axis.
With the SSC of the device under test turned off, the spectrum analyzer will show a large, narrow spike at the fundamental frequency of the carrier. With SSC turned on, the spectrum analyzer view will show a wide plateau of frequency spreading around the fundamental frequency of the carrier. The spectral magnitude of each frequency in this wide plateau is much flatter than the tall narrow spectral peak when SSC is off. This is because the spectral energy is distributed across a wider frequency range, lowering the peak power of the fundamental frequency. This view is very clear on a spectrum analyzer. Incidentally, this view is also available using an oscilloscope.
A NEW VIEW
Originally designed for time-domain analysis, oscilloscopes measure the instantaneous value of a particular event and typically display waveform information as amplitude versus time. This time-domain view is ideal for detecting physical characteristics such as rise time, pulse width, overshoot, and ringing. Oscilloscopes can also display information in the presentation format of a spectrum analyzer where, once again, amplitude (usually in dB) is displayed on the Y-axis and frequency is displayed on the X-axis.
Thanks to a recent technological advance, a third view now has become the most powerful for SSC analysis. It displays instantaneous frequency versus time—the instantaneous frequency on the Y-axis and time on the X-axis. This shouldn't be confused with a spectrum analyzer's spectrogram or waterfall display, which shows the evolution of frequency content versus time as a third axis.
The new oscilloscope view of instantaneous frequency versus time shows the instantaneous frequency of each time-domain unit interval versus time from a single acquisition. So this view can consider frequency events as if they were timing events, allowing rise time, pulse width, overshoot, and ringing of a frequency transition to be characterized as frequency changes and modulates throughout a single acquisition.
Rare, intermittent events aren't visible to a spectrum analyzer, which relies on spectral repetition to form its spectral magnitude lines. A rare or one-time event may not have enough spectral energy to surpass the noise floor. In addition, it's impossible to detect characteristics such as the shape of a modulation curve using a spectrum analyzer.
By using the oscilloscope's hybrid time-domain frequency-domain information, events such as frequency anomalies, SSC asymmetry, frequency-slope changes, and other events invisible to a spectrum analyzer are shown clearly on the oscilloscope display. The bottom line is that today's rapid advances in real-time oscilloscope technology have uniquely positioned the oscilloscope as the most versatile and comprehensive instrument for SSC analysis.
In Figure 2, the persistence trace mean (green waveform) shows the main mode of the frequency track (blue waveform), clearly revealing the spread-spectrum clock modulation of the 1.5-Gbit/s Serial ATA waveform (brown waveform). Parametric measurements show the minimum frequency (1.495 GHz), maximum frequency (1.500 GHz), and peak-to-peak frequency deviation (4.8 MHz) of the 300,000 data cycles acquired (4 million sample points).
Frequency-at-level measurement of the track reveals that the SSC modulation frequency is 29.8 kHz. These time-domain measurements, when applied in a frequency-versus-time mode, provide both quantitative and qualitative insight into SSC behavior, which had never been available previously.
Filtering is the traditional method for reducing high-frequency noise. With the advent of digital signal processing, digital filters were incorporated into real-time oscilloscopes. Today, filter types such as low pass, high pass, bandpass, and band stop are commonly used to filter input signals. Applying this digital filtering technique to a frequency track, instead of applying digital filters to the input waveform, can reduce or eliminate the high frequencies and spurious noise residing in the frequency-track modulation.
Using digital filters to lower track noise is the most versatile method of the three, because exact filter cutoff coefficients can be specified to include or exclude known sources of modulation noise. The digital-filtering-package (DFP) technique gives users extensive control and accuracy in varying low- and high-frequency cutoffs, filter taps, stop-band attenuation, and pass-band ripple. Yet it also requires the greatest processing time of the three methods.
Combining sparsing with enhanced resolution is the third alternative for lowering track noise. Because the track contains many high-frequency fluctuations, the use of the sparse function will decimate the track, leaving N out of every M points remaining. The SSC modulation occurs slowly compared to the frequency of the carrier.
Therefore, decimating the extra frequency track values will leave a sparsed SSC structure intact. The shape of this structure is then further smoothed with enhanced resolution. Combining sparse and enhanced resolution is straightforward to implement and lets users view the intermediate steps involved in this implementation. It's faster than DFP and more flexible than persistence trace mean.
In Figure 3, the blue waveform shows the frequency track of the 4-Mpoint, 2.5-Gbit/s PCI Express serial bitstream (brown waveform) with SSC. The green waveform performs a 75:1 sparsing of the track, and the red waveform performs a Gaussian, 3-bit, low-pass filter enhancement of the sparse. Note how clean the red waveform is compared with the green and blue waveforms.
This combination of sparse and enhanced resolution can reveal the SSC modulation separated from the carrier. Parametric measurements applied to the track show a minimum (2.4858 GHz), maximum (2.5017 GHz), and peak-to-peak frequency deviation (15.9 MHz) of the 500,000 data cycles acquired (4 million sample points), while the track's frequency-at-level measurement reveals an SSC modulation frequency of 31.3 kHz.
The fast Fourier transform (FFT) can be applied to the input waveform of both a normal and an SSC-modulated waveform. Overlapping the FFT spectra on a single grid shows the reduction in peak power levels due to spectral spreading. Measurements determine the relative power difference between the normal and SSC inputs.
The FFT can also be applied to the persistence trace mean, DFP, or enhanced resolution sparse trace to display the frequency content of the track. This would reveal any multitonal effects and determine relative strength of the SSC. When the FFT is applied to the track, persistence trace mean, eres, or DFP output, then the frequency-domain view displayed shows the frequency content of the modulation, separated from the carrier. This way, the effects of SSC are completely isolated from the waveform itself, and the full analysis capability of the scope is applied directly to the SSC.