Maximizing the FFT on an Oscilloscope

Keywords: FFT, DDC, digital signal processing, application-specific integrated circuit, resolution bandwidth, overlap

1. What is FFT on an oscilloscope
? 2. What problems can FFT on an oscilloscope solve
? 3. FFT on an oscilloscope often becomes useless to users. What is the problem?
4. How did we achieve the ultimate in spectrum analysis on oscilloscopes?
5. Development trend of spectrum analysis on oscilloscopes

1. With a digital oscilloscope, our processing of waveforms is no longer simple. It is no longer just to look at the waveform shape, and no longer satisfied with just measuring a few parameters. We always want to do more processing on the collected data. The oscilloscope has a more accurate understanding. It is more like a waveform analyzer. It is the dissatisfaction of engineers that gives us the motivation to constantly pursue the limit, because we often underestimate our potential. Where is the limit? Who was the first to use FFT (Fast Fourier Transform) in a digital oscilloscope? There are many opinions. It seems that suddenly, everyone has found that there is an FFT function on the oscilloscope, and it is a standard configuration. Although all have this function, the results are very different, and the speed and indicators are also different. The initial stage of anything is the same. First pursue it, then talk about differentiation. Besides, the oscilloscope itself is a qualitative tool. Who cares about the accuracy of the oscilloscope's indicators in the frequency domain, except our lovely R&D engineers. The situation is changing. Often users want to solve all problems with one instrument, because to be honest, many engineers do not have the conditions to put a potentiometer, spectrum analyzer, oscilloscope, or vector network on their desks. In most cases, the oscilloscope uses software FFT calculations to convert the collected time domain data samples into frequency domain samples, and then displays the frequency domain samples through data reorganization.

The ability of FFT depends on the following indicators: memory size, software operation speed, dynamic effective bit ENOB, and
noise floor. Because these indicators directly determine the refresh speed, dynamic range, sensitivity, and resolution bandwidth RBW after FFT.
Second, what problems can the FFT of an oscilloscope solve?
Limited by the tools at hand (all engineers dream of having the most advanced oscilloscope and spectrum analyzer on their desks), and many times when engineers debug circuits, they need to make qualitative observations first, so FFT becomes a good tool for viewing the spectrum. To be honest, many manufacturers' FFT functions are not satisfactory. There are only two reasons. One is that they do not have the ability to do it well. To do spectrum analysis well still requires many DSP experts and RF technical strength; the other is that they can do it well, but subjectively do not want to make FFT too strong or too good, so how can I sell my spectrum analyzer? There is an opportunity cost issue here. However, FFT can still solve some problems, such as looking at the spectral range, harmonic components, harmonic proportion, and roughly looking at spectrum interference, etc., but it often brings some embarrassing problems. For example, when the sampling chip is composed of multiple pieces, the overlapping spectrum will be exposed, the processing speed is too slow and it will make people collapse, the background noise is a bit too outrageous, and the jitter component proportion is a bit messy. Of course, some good methods will be thought of to avoid these problems, such as limiting the FFT analysis samples so that it will not freeze when storing FFT for a long time, such as waveform averaging to reduce the background noise, etc.
3. Is the FFT of the oscilloscope useless?
It must be said that sometimes it is really useless. The processing speed is too slow, and a slightly larger sample is almost like freezing. The RBW is too outrageous, the harmonic suppression ratio is very poor, the noise often drowns the harmonics, and the dynamic range is also very poor. But in fact, in many of our occasions, if the FFT function is good enough, it is not useless, but chicken legs. For example, testing the impulse response (characteristic curve) of filters and systems, distinguishing and locating noise interference sources, determining spurious radiation, jitter analysis, harmonic power analysis, and EMI analysis. From this perspective, FFT has a lot of uses.
4. How do we maximize the spectrum analysis function on the oscilloscope?

First of all, we need to increase the speed of spectrum analysis and refresh it in real time, so you no longer have to endure the oscilloscope freezing when performing FFT conversion. Secondly, we have increased the RBW to 1Hz, which is almost only possible with a spectrum analyzer. Our interface design is exactly the same as the operation of the spectrum analyzer, including the center frequency, spectrum range, start spectrum, cutoff frequency, RBW setting, and window function setting. Almost all the settings of the spectrum analyzer have been transplanted.
The following four aspects demonstrate how we can achieve the ultimate FFT function:
1. Dedicated digital down converter DDC.
The traditional approach is that the oscilloscope collects signal samples and then performs software calculations through software algorithms, which is very slow. Our approach is to use a dedicated hardware acceleration integrated circuit (ASIC) to hand over the FFT function to this hardware circuit, which is so fast that it hardly affects the refresh rate of the original waveform. Of course, this ASI requires a lot of money to develop. The core comparison uses a dedicated DDC circuit. Let's see how traditional oscilloscopes perform FFT.

Our oscilloscope fft principle

From the comparison of the above figures, we can see that a DDC process is performed before the window function. The user sets the center frequency, initial and cutoff frequencies. The result of the process is that only the frequency band of interest, or the set frequency band, is processed. The traditional method must perform FFT operations on all frequency bands, and then select a frequency to display. The amount of data calculated is very large. On the contrary, our principle is to only process the frequency band you are interested in or the initial frequency and cutoff frequency range you choose. Of course, in extreme cases, the entire frequency band is also selected for processing. In this way, there is an opportunity to reduce the amount of processing and concentrate the processing power in the range after DDC.
The following two figures more clearly tell the difference between the traditional method and our method.

This approach brings two benefits:
a) Faster speed, as frequency conversion to baseband processing results in higher update rates and faster processing speed, saving processing time.
b) Better resolution bandwidth, as a better magnification factor can be used.

2. Use of hardware accelerators
In traditional solutions, software processing has always been used to implement functions such as statistical histograms, template test functions, and fft functions. In RS oscilloscopes, all functions are implemented using dedicated hardware circuits, freeing up the processor. Therefore, when performing histogram functions, template test functions, or fft functions that consume resources abnormally, the refresh rate is still very high, usually exceeding 60,000 times/s. This speed exceeds the refresh rate of all oscilloscopes on the market when they do not perform any operations. This ensures that the refresh rate is still very high when performing complex waveform analysis, and the high refresh rate ensures the fast display rate of the real-time spectrum.
3. Application of overlapping fft algorithms
The traditional oscilloscope fft calculation method is to collect a section, process a section, then collect again, and then process.

Therefore, although data is collected and processed continuously, the spectrum of sporadic signals can be easily lost and cannot be detected.

RS oscilloscopes process the collected samples in segments, dividing the collected signal into many small segments for processing, so that the changes in the spectrum content in a single acquisition can be seen. However, optical segmentation processing cannot avoid loss, because before the FFT operation, there is already window function processing, and it is inevitable that spectrum information will be lost between two adjacent frames. Therefore, we have adopted another more innovative method, using the FFT overlap algorithm, which greatly improves the influence of the window function and the loss of abnormal spectrum.

With the help of the display of simulated persistence, the display of real-time spectrum is more reliable and confident.

Benefits summary:
a) It is good for detecting abnormal signals
b) It can show rare events that occur in a short period of time
c) It can improve the refresh rate of the spectrum (because the FFT of a new frame starts before the FFT of a frame is completed)
d) It can distinguish multiple spectrum events in one FFT frame

4. Control interface and operation method similar to traditional spectrum analyzers
The previous oscilloscope operation method was nothing more than adjusting the acquisition time to affect the resolution bandwidth, and then selecting the frequency band of interest for observation. Now the method is to first select the center frequency, or select the start and end frequencies, and adjust the spectrum observation method by directly adjusting the RBW, so that users who are used to spectrum analyzers can also get used to oscilloscopes.

There is also a table to help understand which window function to use in which situation.

5. Use the template method to achieve the trigger setting in the frequency domain
Many people who are used to using oscilloscopes like the trigger function of oscilloscopes, and use various trigger methods to isolate various events, stabilize the display, and observe abnormalities. It is difficult to achieve triggering on traditional spectrum analyzers, but when we find the template trigger method of the oscilloscope, it is very easy to do it. The real-time spectrum of the time domain waveform is changed to the frequency domain for observation. With the help of some small tools of MASK test, it is easy to set and trigger. Because the shape of the template can be freely edited and the triggering action can be freely combined, such waveform analysis has completely transcended the usage habits of the time domain and frequency domain, and completely integrated the thinking methods of the time domain and frequency domain for signals.

Red template area trigger instance

5. Development trend of spectrum analysis on oscilloscopes
The analysis speed of oscilloscopes is getting faster and faster, the algorithms are becoming more scientific, and the storage depth is getting larger and larger. The FFT function is no longer dispensable as before. The ability of spectrum analysis depends on the FFT ability, the dynamic range, and the noise level. The spectrum analysis done by the principle of the oscilloscope needs to increase the dynamic range, which is nothing more than doing some time domain averaging before FFT to reduce noise, or increasing the storage depth, increasing RBW, and reducing asynchronous noise to achieve the purpose of increasing the dynamic range.
In addition to doing the FFT function well, oscilloscope manufacturers must also have such a broad mind and regard technological integration and technological progress as opportunities. The driving force of innovation always brings new extremes. Whether you can defend a piece of territory depends on whether users buy it. The goal is to continuously push the limit and continuously create new value for customers.