Benefits of Real-Time Analytics and Measurements for Wireless Communications

Real-time signal processing definition

The term "real time" originated from digital simulations of physical systems. If the speed of the simulation system matches the speed of the real system it simulates, we consider the digital system to be real time. To analyze a signal in real time, it means that the execution speed must be fast enough to accurately process all signal components in the relevant frequency band. If real-time components are to be processed, first the input signal must be sampled fast enough to satisfy the Nyquist theorem. This means that the sampling frequency must be more than twice the signal bandwidth. Secondly, all calculations must be performed continuously at a fast enough speed that the analysis output can keep up with any changes in the input signal.
Spectral analysis, also known as Fourier analysis, is the analysis of signals in the frequency domain. When using DSP, this means performing a discrete Fourier transform (DFT) on the time-sampled data.

The use of DSP for Fourier analysis is shown in Figure 1. After the input analog signal passes through the A/D and then the DFT engine, the FFT spectrum of the input signal is obtained. It can be seen that there are some time gaps between each FFT sampling in this figure. The entire spectrum analysis process in the above figure is equivalent to letting the signal first pass through a group of bandpass filters in the figure below, where the bandwidth and center frequency of each filter are separated according to the DFT basic unit. For each frequency domain unit, the I and Q or amplitude and phase complex envelopes are calculated. If the complex envelope is sampled, when the sampling rate is equal to the speed of the FFT of the signal in the above figure, the two results are exactly the same. The above shows a non-real-time FFT spectrum analysis process. For real-time spectrum analysis, two standards must be met: 1. The input signal must be sampled fast enough to meet the Nyquist theorem, and the sampling signal speed must be greater than twice the signal bandwidth. 2. The DFT calculation must be performed fast enough so that each DFT frequency domain unit meets the Nyquist standard.

We define the shortest time to capture a non-repeating event 100% of the time as the duration of capturing the narrowest rectangular pulse. To process all the signal information of interest in real time, first, there must be enough capture bandwidth to support the signal of interest. Second, a high enough ADC clock rate exceeds the Nyquist criterion. Third, there must be a long enough capture interval to support the narrowest resolution bandwidth (RBW) to analyze the signal of interest. Fourth, a high enough DFT conversion rate exceeds the Nyquist criterion for the RBW of the signal of interest. In today's communication systems, there are many narrow pulse communications. In order to measure and troubleshoot to find problems, it is very important to find, trigger and analyze these narrow pulses.

Windowing and DFT frame overlap

The role of the window function in DFT is to reduce spectrum leakage in FFT spectrum analysis, but sometimes it just happens to block effective information. In Figure 2, the leftmost figure shows the situation where there is a time gap during DFT processing, the middle figure shows that the DFT is adjacent and there is no time gap. The right figure uses frame overlap technology, that is, the second frame data will share the first frame data. As can be seen in the left figure, a short pulse signal appears in the time domain, but because the pulse signal is just between two DFTs in the time domain, it is directly lost when the transformation is performed. As can be seen from the middle figure, since the appearance of the short pulse is just at the edge of the adjacent window function, it is also suppressed. In other words, the signal is sampled, but it is minimized when digital processing is performed later. In the right figure, overlapping DFT technology is used. Since the first frame and the second frame overlap, the FFT is overlapped and calculated, and the spectrum of the pulse signal in the second frame is easily displayed. Therefore, when performing spectrum analysis, the use of frame overlap technology greatly enhances the signal transient processing capability.

The evolution of signal processing technology[page]

From the conventional swept spectrum analyzer in the 1960s to the vector signal analyzer introduced in the 1990s, and the spectrum analyzer introduced by Tektronix today, signal processing technology has gone through three stages. The spectrum testers in the 1960s were mainly used for military and communication systems. The test signals were mainly analog signals, and most of them were steady-state signals. At that time, the requirements for test instruments were mainly low noise floor and relatively high dynamic range. In the 1990s, complex digital modulation and the rapid development of communication technology made the test signals mainly digital modulation signals. Today, with the unlimited development of DSP software, the emergence of adaptive modulation signals, transient signals, and frequency hopping and variable frequency radar communications have made the test requirements for transient signals higher and higher. Redundant time correlation analysis, seamless capture, and arbitrary position triggering of frequency domain events are all very good tools for analyzing such transient signals.
In order to better understand the working principle of real-time spectrum analyzers, we can roughly look at the simple block diagrams of the three popular types of analyzer structures (Figure 3). Although there are many similarities in the analyzers, such as input attenuators, there are also many differences.

For the earliest analyzer design, the swept spectrum analyzer, the signal first passes through a relatively narrow tunable preselector filter, then downconverts, then passes through a resolution bandwidth filter RBW for detection, and then passes through a video bandwidth filter to display the spectrum on the screen. The corresponding local oscillator change process is that the local oscillator performs a continuous frequency sweep in the tuned state within the frequency span. The second-generation vector signal analyzer (VSA) structure also downconverts the spectrum, but uses a local oscillator stepping method. Continuous frequency coverage is achieved by digitizing the time domain signal, and the width of the signal is determined by the intermediate frequency bandwidth of each local oscillator step. The VSA stores the data in memory and calculates its spectrum through fast Fourier transform and displays it on the spectrum. Although many structures of the real-time spectrum analyzer (RTSA) are similar to the VSA architecture, the most important difference lies in the hardware difference in real-time digital signal processing: that is, in the real-time spectrum analyzer, the signal is digitized by the ADC, then processed by the DSP for digital signal processing, and then there is an ultra-fast real-time FFT hardware circuit for calculation.

As we all know, the FFT process requires a lot of calculations. The length of the FFT calculation time depends on the number of points required for the FFT change and the speed of the calculation execution. If the execution time of an FFT calculation is less than the sampling time of each frame, we consider such FFT processing to be real-time. If the FFT processing time exceeds the time required for a frame of sampling, we consider it to be non-real-time. In Figure 4, the upper part of the figure is not calculated in real time, so the second frame of the time signal is lost, that is, there is no time to perform FFT processing, so the spectrum is not displayed in real time. In the lower part of the figure, because ultra-fast FFT processing is performed, and the time of each FFT processing is less than the frame sampling time, the spectrum of all signals will be analyzed in real time at this time.

In the usual sense, "real-time" refers to seamless capture and continuous real-time processing of data, that is, data input and output must be continuous and uninterrupted, and it cannot be stored first and then processed. Real-time spectrum analysis uses a storage-while-processing method, and the processing speed is very fast (greater than the sampling speed of each frame). For example, FM radio converts FM signals into sound, and there will be no interruption in the sound signal here, so we think it is real-time. In addition, the TV converts the RF signal into a dynamic image without interruption until we turn it off, so we think it is also real-time. The real-time spectrum analyzer converts the RF signal into a frequency-to-power trajectory, and there is no interruption in the middle, or what we call dead time.

So what is non-real time? For example, traditional signal scanning analysis, or single capture and processing like vector signal analysis. A less obvious non-real time method is continuous re-capture and processing, which causes the problem of dead time between two consecutive captures.

Real-time signal processing applications

Traditional swept spectrum analyzers are passive test signals, and the characteristic of the measured signal is that the carrier does not change with time. However, real-time spectrum analyzers actively detect transient signals, using DPX digital phosphor technology, which can detect transient signals in real time, implement capture analysis through frequency template triggering, and finally perform real-time spectrum analysis.

Tektronix patented DPX digital phosphor technology

When performing spectrum analysis, after the RF signal enters the ADC and is digitized, the sampled point set passes through the DFT engine and generates a DFT spectrum. Since more than 48,000 DFT operations can be performed per second, and each DFT spectrum will pass through the pixel buffer memory and display the number of times on the pixel statistics chart, it would be very unintuitive if the statistical number of these pixels appearing is directly displayed on the spectrum, so we use color temperature technology to represent the frequency of the signal in the frequency domain. If the frequency is high, it will be warm (red), and if the frequency is low, it will be cold (blue). Now superimpose every 1,400 DFTs on a spectrum screen, and reflect the frequency of the signal through color. According to the principle of human visual persistence, when more than 25 frames of images are played continuously per second, the picture seen by the human eye is continuous, so by playing 33 frames of spectrum screens per second, and each spectrum screen is composed of more than 1,400 DFT superpositions, so 48,000 vivid and real-time video signals per second can be easily displayed.

[page]

Figure 5 shows the signal generated by the demo board, which simulates a signal that loses lock every 1.28 seconds. The left picture is tested with an ordinary spectrum analyzer. It is difficult to find the problem in free-running mode. Next, it uses the maximum hold mode. We can see that some transient clutter signals occasionally appear in the spectrum, but it is difficult to find out what signal it is. In the right picture, because the real-time spectrum analyzer has a 100% detection probability, the transient spectrum of each signal loss can be clearly observed.

Tektronix patented frequency mask trigger technology

Using DPX can only find faults at best, but if these signals are found but cannot be captured, quantitative analysis cannot be performed. Traditional signal analyzers can only use power level triggering to trigger a transient signal, that is, setting the center frequency and frequency span, and then setting a threshold for triggering. It cannot achieve frequency selection triggering in the frequency domain. Tektronix's patented frequency template triggering technology solves this problem. It defines a green template in the frequency domain, which contains two-dimensional information of frequency and power. It can easily capture small signals that appear next to large signals in the frequency domain, and can also trigger based on the existing spectrum waveform of the signal.

As shown in Figure 6, the level trigger of the ordinary spectrum cannot trigger the low-power signal under the high-power signal. Now we only need to draw a black template. When the transient small signal appears from the white area on the black template, the real-time spectrum analyzer can use this as a condition to capture the signal for a period of time in real time.

By effectively storing the captured signal, it can be replayed through the three-dimensional spectrum. In Figure 7, the horizontal axis represents frequency, the vertical axis represents time, and the color represents the power of the signal. It can be seen that the green transient interference spurious appears very briefly, only appears in two frames of the spectrum, and the color is also relatively light. The three-dimensional spectrum can clearly reflect the trajectory of the transient signal and analyze it.

Tektronix's time-correlated multi-domain analysis technology

Tektronix's analysis of each domain is not independent of time, but takes time as a benchmark. The five domains of RF analysis, including frequency domain analysis, time domain analysis, modulation domain analysis, code domain analysis and statistical domain analysis, can also be combined with multiple instruments to perform cross-domain time-related analysis.

Figure 8 has four small pictures. The upper left picture reflects the relationship between signal time and amplitude, the upper right picture reflects the signal spectrum, the lower left picture shows the three-dimensional spectrum diagram, and the lower right picture shows the time preview window. Ordinary signal analysis does not have a time preview window, so it is impossible to arbitrarily analyze pulse signals. In the lower right picture, you can see a blue horizontal line, which is actually an analysis window that can analyze 6 pulses at a time; red refers to the spectrum window, and its corresponding spectrum is displayed in the spectrum window above; there is an MR cursor point in each observation domain. When the MR cursor is pulled, the values ​​in all domains will change, so we call it multi-domain correlation analysis.

Figure 9 shows the changes in the spectrum core at the moment when WLAN is interfered. Since both Bluetooth and WLAN signals work at 2.4G, there will be co-frequency interference sometimes. If you do not use the three-dimensional spectrum diagram to observe the information changes in each domain, it is difficult to find the reason for the error in the constellation diagram. From the figure, it can be seen that at the first moment, since the WLAN signal does not conflict with the Bluetooth signal, the spectrum and constellation diagram are very good, but at the next moment, since the Bluetooth signal and WLAN signal appear at the same frequency, the power is superimposed, and a peak appears on the spectrum, and the constellation diagram is wrong at this time. Through this three-dimensional diagram, it is easy to see that the reason why the signal is not demodulated correctly is not due to external interference, but interference from the Bluetooth signal.

In addition, the spectrum analyzer can also be combined with an oscilloscope and a logic analyzer to test the RF part, analog part, and digital baseband part of the same signal circuit board.