Using oscilloscope filters for better measurements
Most oscilloscopes are fitted with filters to help improve measurement by reducing overall noise in an acquisition to improve signal-to-noise ratio (SNR). Even the most basic oscilloscopes include an analog 20 MHz low pass filter in the input channel signal path. High end oscilloscopes with a GHz or better bandwidth generally offer multiple input low pass filters, some analog and some digital. Noise or enhanced resolution (ERES) digital filters are digital low pass filters used to increase the amplitude resolution of the oscilloscope trading bandwidth for improved SNR. Beyond these input band limiting filters, oscilloscopes often include optional digital filter software providing more general filter types for more complex filtering needs.
This article will deal with how to use all of these filtering tools.
Input band limit and noise filters
Oscilloscopes with bandwidths greater than 100 MHz generally include a 20 MHz band limit filter intended to reduce the oscilloscope’s bandwidth to lessen broadband noise in low frequency measurements. This filter is usually enabled in the input channel setup similar to the one shown in Figure 1.
Figure 1 The input channel setup for an oscilloscope with the bandwidth limit filter selection highlighted in yellow and the noise filter highlighted in orange. Source: Arthur Pini
The oscilloscope in this example has a full bandwidth of 4 GHz and input band limit filters of 20 and 200 MHz. Oscilloscopes with higher bandwidths usually offer more band limit filter selections. It also has the setup for a noise filter with user selected bandwidths.
These filters improve SNR by reducing the measurement bandwidth. The amount of improvement depends on the frequency distribution of the noise signals. If the noise source were spectrally flat Gaussian white noise, the SNR improvement would be proportional to the square root of the bandwidth reduction. So, reducing the bandwidth by a factor of four would cut the noise level in half for a spectrally flat noise signal.
Using the band limit filter
As an example of how the oscilloscope’s filters can be applied, let’s look at the ripple voltage on the 5-volt bus of a circuit card to measure ripple due to circuit loading effects and see the effect of the band limit low pass filters on the measurement. We’ll evaluate a 20 MHz band limit filter (Figure 2).
Figure 2 The measurement made at full bandwidth is shown on top. The lower trace was acquired using the 20 MHz band limit filter with a reduction in high frequency noise. Source: Arthur Pini
The upper trace was acquired at full bandwidth using AC coupling and a high-impedance probe with a ground spring to reduce stray pickup on wire ground leads. The low frequency pulse-like waveform is the ripple caused by circuit loading as various devices on the PC board turn on and off. This desired signal is obscured by higher frequency noise. The waveform acquired using the 20 MHz filter shows a significant reduction in the high frequency noise, but it has not eliminated it. The 20 MHz filter has had little effect on the low frequency ripple components, which have a very low frequency below 20 MHz. Comparing the peak-to-peak amplitudes in the segments between 2 and 4 ms, where the high-frequency noise is the main component, the filter has reduced the peak-to-peak noise from 50.4 down to 13.1 mV. The noise that remains has frequency components that are lower than 20 MHz.
Looking more closely at the full bandwidth acquisition using zoom and the fast Fourier transform (FFT) to take a closer look at the acquired signal and evaluate the frequency distribution of the noise. The signal acquired at full bandwidth shows why the 20 MHz filter left a fair amount of noise on the signal (Figure 3).
Figure 3 The zoom expansion and the FFT of the noisy waveform show additional details of the noise components.
Expanding the time signal horizontally (second trace from the top), the high frequency noise appears as narrow capacitively coupled impulses. The fast edges on these noise components have high frequency components that will be spectrally spread. Looking at the FFT (third trace from the top), we see a broad spectrum caused by these noise elements. Also, the spectral peaks representing periodic signal components have the highest density below 15 MHz. Voltage variations due to circuit loading are rectangular pulse-like low frequency variations. In the FFT spectrum, these appear at the extreme left, below 50 kHz. The bottom trace is the FFT of the low frequency components from 0 to about 1 MHz. The 20 MHz input band limit filter attenuates those noise components above 20 MHz but leaves the other spectral components unattenuated.
Reducing the noise further requires reducing the measurement bandwidth further. That can be accomplished using the noise filter, allowing the user to select one of six possible reduced bandwidths.
Enhance resolution noise filters
The noise filters are also known as ERES filters because they increase the effective number of bits of resolution of the oscilloscope. These filters are also available via the input channel setup, as is seen in Figure 1, and also as a math function in the oscilloscope used in this example. The ERES filter processes ‘n’ samples at a time from the acquired input and weights them to produce a finite impulse response (FIR) filter with a Gaussian low pass frequency response. The Gaussian low pass filter has no side lobes in the frequency domain, and it never causes overshoot, undershoot, or ringing in the time domain, maintaining signal integrity. The ERES filter uses any of six sample lengths: 2, 5, 11, 25, 52, and 106 taps to achieve resolution enhancement of 0.5 to 3.0 bits in steps of half a bit each. The low pass filter’s cutoff frequencies depend on the acquisition sampling rate. Table 1 shows the noise filter characteristics of all six steps for the sample rate of 100 mega-samples per second (MS/s) which was used during the acquisition.
Number of Bits
Number of Taps
Bandwidth (MHz)
0.5
2
25.00
1.0
5
12.05
1.5
11
6.05
2.0
25
2.90
2.5
52
1.45
3.0
106
0.800
Table 1 The number of taps and resulting bandwidth of the noise filter for the six possible filter bandwidth limit settings and a sample rate of 100 MS/s.
The ERES noise filter provides a range of low pass cutoff frequencies that decrease proportionately to the number of taps in the filter. Changing the acquisition sample rate will scale the cutoff frequencies proportionally, providing still greater choice in the cutoff frequencies. Selecting the 3-bit enhancement uses 106 samples to achieve an 800 kHz bandwidth. The result of applying the 800kHz low pass filter to the acquired signal is shown in Figure 4.
Figure 4 The 800kHz low pass noise filter eliminates most of the high frequency noise, allowing a detailed study of the lower frequency ripple components due to circuit loading. Source: Arthur Pini
The 800 kHz selection of the noise filter has removed much of the high frequency noise, and the voltage variations due to the circuit loading are more clearly visible.
Low pass filters can attenuate or eliminate high frequency noise. In some cases, you may want to be able to separate the low and high frequency components and study them independently. That requires the use of both a high and a low pass filter. This oscilloscope includes an optional digital filter package that offers various filter types and a broader range of cutoff frequencies than the standard noise filter.
General purpose digital filters
The digital filter package option broadens the offering of filters. It can create four types of filters: low pass, high pass, band pass, and band stop, as shown in Figure 5.
Figure 5 Examples of the frequency responses of low pass (yellow trace), high pass (red trace), band pass (blue trace), and band stop (green trace) filter types. Source: Arthur Pini
These filters can be created using FIR or infinite impulse response (IIR) topologies. IIR filters allow users to select digital filter types identical in response to well-known analog filters, including Butterworth, Bessel, Chebyshev, or Inverse Chebyshev, examples are shown in Figure 6.
Figure 6 Comparing the amplitude frequency responses of Bessel (red trace), Butterworth (yellow trace), Chebyshev (blue trace), and inverse Chebyshev (green trace) IIR low pass filters. Source: Arthur Pini
These are the most commonly used analog filter types. The Butterworth or ‘maximally flat’ filter has the flattest amplitude response of all the available filters. The Bessel filter is noted for its uniform phase response as
a function of frequency. If you need the fastest roll off, the Chebyshev and inverse Chebyshev filters have the narrowest transition region for a given number of stages. On the negative side, the Chebyshev filter has amplitude ripple in the passband, while the inverse Chebyshev filter exhibits a flat passband response but has ripple in the stop band. The filter package provides control of the cutoff frequencies, the filter order, the transition width, and the stop band attenuation of each filter. The filter option package also allows users to use a custom-designed filter.
Two different instances of a Butterworth filter are applied to the acquired signal separately to separate the power rail ripple’s low and high frequency components. The high frequency components are removed by using a sixth order Butterworth low pass filter with a cutoff frequency of 50 kHz. The low frequency components are separated out by applying a sixth-order Butterworth high pass filter with the same 50 kHz cutoff frequency. The results are shown in Figure 7.
Figure 7 Using a low pass and a high pass filter to separate the low and high frequency components of the ripple. Source: Arthur Pini
The 50 kHz cutoff frequency was selected to be below the nominal 61.7 kHz switching frequency of the power source so that the filter would reasonably attenuate signal components at that frequency due to the power switching. The acquired signal is shown in the upper left grid. The extracted low frequency components appear below it in the left center grid. The bottom left grid shows the FFT of the low pass filtered signal. The cursor marks the 61.7 kHz switching frequency, which is attenuated more than 30 dB below the low frequency maxima.
The high pass filter output appears in the upper right grid. Notice that the low frequency ripple due to circuit loading is gone. The zoom of that waveform, shown in the center right grid, shows the familiar noise spikes but without the load related ripple. The bottom right grid shows the FFT of the high pass filtered ripple signal with the cursor marking 61.7 kHz. Note that the spectral components below the high pass cutoff are attenuated.
With the high and low frequency ripple components separated, it is possible to measure them independently. For example, the measurement parameter P1 is set to measure the amplitude of the load related ripple. The ripple amplitude is 16.33 mV. The high pass filtered waveform can also be measured or studied to reveal the sources and effects of the high frequency ripple components.
More accurate measurements with filters
Oscilloscope filters provide users with several options to improve SNR to make more accurate measurements. They can reduce high frequency noise components or selectively separate high and low frequency noise mechanisms, allowing measurements of the selected elements. Spectrum analysis tools, like the FFT, help determine how to set the filter parameters to obtain the most accurate results.
Arthur Pini is a technical support specialist and electrical engineer with over 50 years of experience in electronics test and measurement.
Related Content
10 tricks that extend oscilloscope usefulnessv
10 More tricks to extend oscilloscope usefulness
Measure frequency response on an oscilloscope
View noisy signals with a stable oscilloscope trigger
googletag.cmd.push(function() { googletag.display(‘div-gpt-ad-native’); });
–>
The post Using oscilloscope filters for better measurements appeared first on EDN.
