filter and noise

Noise is a common and pervasive problem in electronics design and debug. At one time or another, almost every designer and debugger of electrical circuits will spend some time dealing with noise—either finding its source to fix it or reducing its impact on measurements. Noise can come from countless internal or external sources, often obscuring signals of interest.
Noise can make it difficult to perform measurements on a millivolt-range signal, such as in a radar transmission or heart-rate monitor. Another noise-related challenge is finding the true voltage of a signal. Moreover, noise may increase jitter, making it difficult to perform timing measurements.
Or there are times when you need a clean, noise-free trace to focus on the intended signal in a design. Other times, a clean trace is useful for reports and documentation to clearly show how a design works. Noise is pervasive in nearly all electrical design and debug work, though, so modern oscilloscopes offer filters to help reduce unwanted noise in measurements—and each filtering approach has its own pros and cons.
Before you can analyze a signal, you must have a stable display. This can be a problem if the signal is noisy, which makes it difficult to create a stable trigger. Oscilloscopes employ a number of strategies to assist with this problem.
Typically, the first step in creating a stable trigger is to evaluate which trigger-coupling mode will work best. These can include high-frequency (HF) reject, low-frequency (LF) reject, and noise reject trigger-coupling options, each of which can be used to create a stable trigger for a signal.
HF reject performs a low-pass filter on the trigger path, attempting to ignore any high-frequency instability or noise. LF reject performs a high-pass filter on the trigger path, attempting to exclude low-frequency signals from causing triggers. Noise reject increases the required trigger hysteresis and prevents random noise from causing triggers. It may be difficult to predict how these modes will affect a particular signal, so it’s usually best to try each one until a stable trigger is attained.
Trigger systems include a holdoff control that only allows triggers after a user-specified delay timer. If the signal is repetitive, adjusting the holdoff will help the oscilloscope ignore false triggers. If triggers are still unstable, a bandwidth-limit filter that passes the signal through a low-pass filter can be helpful.
The low-pass filter usually has only a few frequency settings available and often goes no lower than 20 MHz. For many applications, such as debugging power-supply issues, this may not be low enough. But by trying a different bandwidth setting, a stable trigger can be achieved.
After obtaining a stable trigger, the next step is to further adjust the display of noise on the oscilloscope. Several tools available can do this, including bandwidth-limit filter, average acquisition, high-resolution acquisition, and variable low-pass filter. The table highlights the advantages and disadvantages of these different approaches.
The bandwidth-limit filter reduces the oscilloscope bandwidth to the frequency that’s selected. This means that frequencies higher than the selected level will be attenuated or removed completely from the trigger path, the acquisition path, and the display path.
Further, the bandwidth-limit filter can be used both to attain a stable trigger and reduce the amount of noise displayed. It offers one of the simplest ways to reduce noise, and it works well if the undesirable noise is at frequencies above the fixed cutoff. However, any potential high-speed glitches will also be removed.
Oscilloscopes typically offer a very limited set of bandwidthlimit filter settings. Standard selections generally include 250 MHz and 20 MHz.
In examining an example of the default acquisition and display of a small-voltage sine wave, one can see some 30 mV of noise on the signal (Fig. 1a). Setting the bandwidth-limit filter to 20 MHz greatly reduces the amount of noise (Fig. 1b). This indicates that some amount of noise is greater than 20 MHz in nature, but there’s still some lower-frequency noise.
As the name implies, the average acquisition filter addresses the noise problem by taking several complete acquisitions and averaging them point by point to obtain the average voltage at each time sample in the acquisition. The number of acquisitions included in the average is user-adjustable. Noise is typically random from acquisition to acquisition— sometimes up and sometimes down. When these random variations are averaged over enough acquisitions, they will cancel out to create a stable signal on the screen.
It’s important to note that average acquisition filtering works only for repetitive waveforms. Non-repetitive waveforms or single-shot events can’t be averaged. The average acquisition approach reduces all kinds of uncorrelated signals and random noise, even at very low frequencies. It also works across all oscilloscope time/division settings. If we return to the previous example of a small-voltage sine wave (Fig. 1, again), the application of average acquisition filtering with 32 averages yields a very clean sine wave with almost no noise (Fig. 2).
Because multiple waveforms must be acquired to create one averaged waveform, the display can be slow to update from a changing input signal or a front-panel knob change. This means infrequent glitches will likely be missed. Yet in some applications, average acquisition filtering is a better choice than the bandwidth-limit filter, since the full bandwidth of the oscilloscope is available to capture high-frequency repetitive events.
Similar to average acquisition filtering because it uses averaging to eliminate noise, the high-resolution acquisition filter performs a box-car average on each acquisition (Fig. 3). In this manner, it averages several adjacent samples within a single acquired waveform to create a single averaged sample.
High-resolution acquisition filtering has the effect of reducing high-frequency noise, because the average will cancel out the high-speed variance in voltages caused by the noise. It also reduces the sample rate because it converts many samples into one. As a result, high-resolution acquisition filtering is limited to slower time/division settings where the oscilloscope still has sufficient sample rate to represent the measured signal.
Unlike average acquisition filtering, high-resolution acquisition filtering can be used on non-repetitive and single-shot waveforms. Because only one waveform needs to be acquired, a scope’s highresolution acquisition mode provides a much faster update to the display after an input or front-panel setting change. Combining neighboring samples in time also reduces the chance of aliasing at slower time/division settings.
Because high-resolution acquisition filtering is a type of lowpass filtering, it may cause the user to miss high-speed glitches on the signal. Typically, no indication is given of what frequencies, if any, are being removed in high-resolution acquisition filtering. It may reduce some aliased frequencies from the display. Other aliased frequencies may still be present due to the poor frequency- selectivity nature of the high-resolution low-pass filter.
Some oscilloscopes offer post-processing DSP filters that remove certain frequencies of noise from the signal, giving the user complete control over the filter frequency. While these filters may be flexible, they often are slow and suitable only for singleshot or slow update-rate displays. There’s also a risk that the DSP will filter out interesting and important glitches or anomalies without the user’s knowledge.
Variable low-pass filters, which are coming on the market, are another method of removing unwanted noise. They allow the user to select a low-pass filter frequency to apply to the displayed acquisition. In addition to the low-pass filtered trace, the filters can ensure that the user doesn’t miss any unexpected high-frequency glitches or large-magnitude noise. It does so by providing a background trace showing the peak-detected (min/max sampled) raw acquisition underneath the clean filtered waveform (Fig. 4).
The low-pass filter cutoff frequency can be adjusted to control the amount of noise reduction. Filter-frequency readouts let the user characterize what frequencies of noise are on a given signal without the need to set up a more cumbersome fast Fourier transform (FFT).
A glitch that can be captured at the fastest time/division setting ideally would still be shown when inspecting the signal at the slowest time/division setting. This is where a peak-detect background trace capability is useful to capture peak excursions of the signal up to the oscilloscope’s bandwidth—even for singleshot waveforms.
In an example showing the use of a variable low-pass filter to capture the power-on of a switched-mode power supply, one can observe a small negative spike to the left of the display and oscillation to the right (Fig. 5a). By changing the filter-frequency cutoff to 550 kHz, the display shows that the oscillation has been removed from the main signal (Fig. 5b).
Thus, it’s apparent that the oscillation is between 550 kHz and 1.1 MHz. This analysis can be performed while stopped on the same single-shot capture. Note also that the spike is still shown in the glitch capture background, even though the foreground trace was filtered.
As with high-resolution acquisition filtering, variable low-pass filtering isn’t available at all time/division settings. At faster settings, the range of the filter is reduced. At the fastest time/division settings, no filtering is available because the low-pass filter works by reducing the number of sample points in the waveform.
At many time/division settings, the oscilloscope runs at a reduced sample rate and there are many extra points. When the oscilloscope runs at or near its full sample rate, fewer extra points exist, reducing the variable low-pass functionality. As such, average acquisition is a better choice for reducing noise at the fastest time/division settings.
Variable low-pass filtering can be used on repetitive, nonrepetitive, and single-shot waveforms. The filter-frequency adjustability allows the user to remove noise without rolling off the signal. Compared to the bandwidth-limit filter, variable low-pass filtering can handle lower frequencies (less than 1 MHz).
Another problem can be the effects of aliasing. Variable low-pass filtering deals with aliasing by passing no more than 1% of the high-frequency content that causes alias. This ensures that only the aliased frequencies are removed, not the signal of interest.