Figure 2-1 is a simplified block diagram of a superheterodyne spectrum analyzer. "Heterodyne" means mixing, that is, converting frequencies, and "super" refers to super-audio frequencies or frequencies above the audio frequency range. From the figure, we can see that the input signal first passes through an attenuator, then through a low-pass filter (you will see later why the filter is placed here) to the mixer, and then mixed with the signal from the local oscillator (LO).
Figure 2-1. Block diagram of a typical superheterodyne spectrum analyzer
Since the mixer is a nonlinear device, its output contains not only the two original signals, but also their harmonics and the sum and difference signals of the original signals and their harmonics. If any of the mixed signals falls within the passband of the intermediate frequency (IF) filter, it will be further processed (amplified and possibly logarithmically compressed). The basic processing steps are envelope detection, filtering with a low-pass filter, and display. The ramp generator produces the horizontal movement from left to right on the screen, and it also tunes the local oscillator so that the change in the local oscillator frequency is proportional to the ramp voltage.
If you are familiar with a superheterodyne radio that receives ordinary AM broadcast signals, you will find that its structure is very similar to the block diagram shown in Figure 2-1. The difference is that the output of the spectrum analyzer is a screen instead of a speaker, and its local oscillator is tuned electronically instead of by a front panel knob.
Now that the output of a spectrum analyzer is an XY trace on the screen, let's see what information we can get from it. The display is mapped onto a dial consisting of 10 horizontal grids and 10 vertical grids. The horizontal axis represents frequency, and its scale increases linearly from left to right. Frequency setting is usually a two-step process: first adjust the frequency to the center line of the dial using the center frequency control, and then adjust the frequency range (span) across the 10 grids using the frequency span control. These two controls are independent of each other, so the span does not change when the center frequency is changed. Also, we can set the start and end frequencies instead of the center frequency and span. In either case, we can determine the absolute frequency of any displayed signal and the relative frequency difference between any two signals.
The vertical axis is scaled by amplitude. You can choose a linear scale in voltage or a logarithmic scale in decibels (dB). Logarithmic scales are more often used than linear scales because they can reflect a larger range of values. Logarithmic scales can simultaneously display signals with amplitudes that differ by 70 to 100 dB (voltage ratios of 3200 to 100,000 or power ratios of 10,000,000 to 10,000,000,000), while linear scales can only be used for signals with amplitudes that differ by no more than 20 to 30 dB (voltage ratios of 10 to 32). In both cases, we use a calibration technique1 to give the absolute value of the level on the highest line of the dial, the reference level, and determine the values of the other positions on the dial based on the ratio corresponding to each small grid. In this way, we can measure both the absolute value of the signal and the relative amplitude difference between any two signals.
The frequency and amplitude scale values are annotated on the screen. Figure 2-2 is a typical spectrum analyzer display.
The first part of the analyzer is the RF attenuator. Its function is to ensure that the signal is at the appropriate level when it enters the mixer, thereby preventing overload, gain compression, and distortion. Since the attenuator is a protection circuit for the spectrum analyzer, it is usually set automatically based on the reference level value, but the attenuation value can also be manually selected in steps of 10 dB, 5 dB, 2 dB, or even 1 dB. Figure 2-3 shows an example of an attenuator circuit with a maximum attenuation value of 70 dB in 2 dB steps.
The DC blocking capacitor is used to prevent the analyzer from being damaged by DC signals or DC bias of the signal, but it will attenuate low-frequency signals and increase the minimum usable starting frequency of some spectrum analyzers to 9 kHz, 100 kHz or 10 MHz.
In some analyzers, an amplitude reference signal can be connected as shown in Figure 2-3. This provides a signal with precise frequency and amplitude for periodic self-calibration of the analyzer.
The function of the low-pass filter is to prevent high-frequency signals from reaching the mixer. This prevents out-of-band signals from mixing with the local oscillator and generating unwanted frequency responses at the intermediate frequency. Microwave spectrum analyzers or signal analyzers use preselectors instead of low-pass filters. Preselectors are adjustable filters that can filter out signals at frequencies other than the frequency we are concerned about. In Chapter 7, we will introduce in detail the purpose and method of filtering the input signal.
Analyzer tuning
We need to know how to tune the spectrum analyzer or signal analyzer to the frequency range we want. The tuning depends on the center frequency of the intermediate frequency filter, the frequency range of the local oscillator, and the frequency range of the external signal allowed to reach the mixer (allowed to pass through the low-pass filter). Among all the signal components output from the mixer, there are two signals with the largest amplitude that we want most. They are the signal components generated by the sum of the local oscillator and the input signal and the difference between the local oscillator and the input signal. If we can make the signal we want to observe higher or lower than the local oscillator frequency by an intermediate frequency, one of the desired mixing components will fall into the passband of the intermediate frequency filter, and then it will be detected and produce an amplitude response on the screen.
In order to tune the analyzer to the required spectrum range, we need to select the appropriate LO frequency and IF. Assuming the required tuning range is 0 to 3.6 GHz, the next step is to select the IF frequency. If we select an IF of 1 GHz, which is within the required tuning range, we can get a 1 GHz input signal, and since the output of the mixer contains the original input signal, the 1 GHz input signal from the mixer will have a constant output at the IF. So no matter how the LO is tuned, the 1 GHz signal will pass through the system and give a constant amplitude response on the screen. The result is a blank area in the frequency tuning range where measurements cannot be made, because the signal amplitude response in this area is independent of the LO frequency. So we cannot select an IF of 1 GHz.
That is, we should select the IF at a higher frequency than the tuning band. In a Keysight X-Series signal analyzer that can be tuned to 3.6 GHz, the first LO frequency ranges from 3.8 to 8.7 GHz, and the IF frequency selected is approximately 5.1 GHz.
Now we want to tune from 0 Hz (actually from some lower frequency since a 0 Hz signal cannot be observed with this configuration of instrument) to 3.6 GHz.
By choosing the LO frequency to start at the intermediate frequency (LO - IF = 0 Hz) and tune up to 3.6 GHz above the intermediate frequency, the mixing products of LO - IF can cover the required tuning range. Using this principle, the following tuning equation can be established:
If you want to determine the local oscillator frequency required to tune the analyzer to a low-, medium-, or high-frequency signal (such as 1 kHz, 1.5 GHz, or 3 GHz), you first transform the tuning equation to obtain fLO:
Figure 2-4. To produce a response on the display, the LO must be tuned to fIF + fs
Figure 2-4 illustrates the tuning process of the analyzer. In the figure, fLO is not high enough to cause the fLO -fsig mixing product to fall into the IF passband, so there is no response on the display. However, if the ramp generator is adjusted to tune the local oscillator to a higher frequency, the mixing product will fall into the IF passband at some point in the ramp (sweep), and we will see a response on the display.
Since the ramp generator can control both the horizontal position of the trace on the display and the LO frequency, the horizontal axis of the display can be calibrated according to the frequency of the input signal.
We haven't quite solved the tuning problem yet. What happens if the input signal frequency is 9.0 GHz? When the LO is tuned in the range of 3.8 to 8.7 GHz, when it reaches the IF (3.9 GHz) far from the 9.0 GHz input signal, it will get a mixing product at the same frequency as the IF and generate a response on the display. In other words, the tuning equation can easily become:
This formula shows that the structure of Figure 2-1 can also achieve a tuning range of 8.9 to 13.8 GHz, but only if signals in this range are allowed to reach the mixer.
The role of the input low-pass filter in Figure 2-1 is to prevent these high-frequency signals from reaching the mixer. As mentioned earlier, we also require that the intermediate frequency signal itself does not reach the mixer, so the low-pass filter must be able to effectively attenuate signals in the 5.1 GHz and 8.9 to 13.8 GHz range.
In summary, it can be considered that for a single-band RF spectrum analyzer, the selected IF frequency should be higher than the highest frequency of the tuning range, so that the local oscillator can be tuned from the IF to the upper frequency of the tuning range plus the IF, and a low-pass filter is placed in front of the mixer to filter out frequencies below the IF.
In order to distinguish signals that are very close in frequency (see the "Signal Resolution" section later), some spectrum analyzers have IF bandwidths as narrow as 1 kHz, some as narrow as 10 Hz or even 1 Hz. Such narrowband filters are difficult to implement at a center frequency of 5.1 GHz, so additional mixing stages (usually 2 to 4 stages) must be added to downconvert the first IF to the final IF. Figure 2-5 is an IF conversion chain based on a typical spectrum analyzer structure.
It can be seen that it gives the same result as the simplified tuning equation using only the first IF. Although only the passive filter is shown in Figure 2-5, there is actually amplification of the narrower IF stage. Based on the design of the spectrum analyzer itself, the final IF structure may also include other components such as logarithmic amplifiers or analog-to-digital converters.
Most RF spectrum analyzers allow the LO frequency to be as low as the first IF, or even lower. Because the isolation between the LO and the mixer's IF port is limited, the LO signal will also appear at the mixer output. When the LO frequency is equal to the IF, the LO signal itself is processed by the system and appears on the display as if a 0 Hz signal was input. This response is called LO feedthrough, and it masks low-frequency signals. Therefore, not all spectrum analyzers have a display range that includes 0 Hz.
IF Gain
Looking back at Figure 2-1, the next part of the block diagram is a variable gain amplifier. It is used to adjust the vertical position of the signal on the display without affecting the signal level at the mixer input. When the IF gain changes, the reference level value will change accordingly to maintain the correct value of the displayed signal indication. Usually, we want the reference level to remain unchanged when adjusting the input attenuation, so the RF attenuator and IF gain settings are linked.
When the input attenuation changes, the IF gain is automatically adjusted to offset the effect of the input attenuation change, so that the signal position on the display remains unchanged.
Signal resolution
Following the IF gain amplifier comes the IF section which consists of analog and/or digital resolution bandwidth (RBW) filters.
Analog Filters
Frequency resolution is the ability of a spectrum analyzer or signal analyzer to clearly separate the responses of two sinusoidal input signals. Fourier theory tells us that sinusoidal signals only have energy at a single frequency, so it seems that we should not have any resolution problems. No matter how close the two signals are in frequency, they should appear as two lines on the display. However, the signal response presented on the display of a superheterodyne receiver has a certain width.
The output of the mixer consists of the two original signals (the input signal and the local oscillator) and their sum and difference. The intermediate frequency is determined by a bandpass filter, which selects the desired mixing component and suppresses all other signals. Since the input signal is fixed and the local oscillator is swept, the output of the mixer is also swept. If a mixing component happens to sweep through the intermediate frequency, the characteristic curve of the bandpass filter will be plotted on the display, as shown in Figure 2-6. The narrowest filter bandwidth in the chain determines the total display bandwidth. In the structure shown in Figure 2-5, the filter has an intermediate frequency of 22.5 MHz.
Figure 2-6. As the mixing products are swept through the IF filter, the filter characteristics are traced on the display.
Therefore, the two input signal frequencies must be spaced far enough apart, or the traces they form will overlap on top of each other, making it look like there is only one response. Fortunately, the resolution (IF) filter in a spectrum analyzer is adjustable, so you can usually find a filter with a bandwidth narrow enough to separate closely spaced signals.
The data sheet of Keysight's spectrum analyzers or signal analyzers lists the 3 dB bandwidth of the available IF filters to describe the spectrum analyzer's ability to resolve signals. These data tell us how close two equal-amplitude sine waves can be to each other and still be resolved. At this point, there is a 3 dB dip at the peak of the two response curves generated by the signals, as shown in Figure 2-7, and the two signals can be resolved. Of course, the two signals can be closer until their traces completely overlap, but 3 dB bandwidth is usually used as an empirical value for resolving two equal-amplitude signals.
Figure 2-7. Two equal-amplitude sinusoidal signals can be resolved with a spacing equal to the 3 dB bandwidth of the selected IF filter.
If the standard (normal) detection mode is used (see "Detection Types" later in this chapter), sufficient video filtering needs to be used to smooth the signal traces, otherwise there will be smearing due to the interaction of the two signals. Although the smeared trace indicates the presence of more than one signal, it is difficult to determine the amplitude of each signal. Spectrum analyzers with the default detection mode of positive peak detection do not show the smearing effect, but it can be observed by selecting the sample detection mode.
More often than not, we encounter unequal amplitude sine waves. It is possible for a smaller sine wave to be buried by the sidebands of the larger signal response curve. This phenomenon is shown in Figure 2-8. The top trace looks like one signal, but it actually consists of two: one at 300 MHz (0 dBm) and another at 300.005 MHz (-30 dBm). The smaller signal is only visible after the 300 MHz signal is removed.
Another technical specification for resolution filters is bandwidth selectivity (also called selectivity or shape factor). Bandwidth selectivity determines the ability of a spectrum analyzer to resolve sinusoidal signals of unequal amplitude. The bandwidth selectivity of a Keysight spectrum analyzer is usually specified as the ratio of 60 dB bandwidth to 3 dB bandwidth, as shown in Figure 2-9. The analog filter in a Keysight analyzer has four poles and is designed to be frequency tuned, with a characteristic curve that resembles a Gaussian distribution4. The bandwidth selectivity of this filter is approximately 12.7:1.
So, assuming the bandwidth selectivity is 12.7:1, how should the resolution bandwidth be selected if two signals with a frequency difference of 4 kHz and an amplitude difference of 30 dB need to be resolved?
Figure 2-9. Bandwidth selectivity: ratio of 60 dB bandwidth to 3 dB bandwidth
Some older spectrum analyzers or signal analyzers use 5 poles for the narrowest resolution bandwidth filter to improve bandwidth selectivity to 10:1. Newer analyzers can achieve better bandwidth selectivity by using digital IF filters.
Since we are interested in the rejection of larger signals when the analyzer is tuned to smaller signals, we do not need to consider the entire bandwidth, but only the frequency range from the center frequency to the edge of the filter. To determine how much the filter sidebands are reduced at a given frequency offset, use the following equation:
Figure 2-10. A bandwidth of 3 kHz (upper trace) cannot resolve the smaller signal, but a bandwidth of 1 kHz (lower trace) can.
Digital Filter
Some spectrum analyzers use digital technology to implement resolution bandwidth filters. Digital filters have many advantages, such as greatly improving the bandwidth selectivity of the filter. Keysight Technologies' PSA series and X series analyzers have fully digitized resolution bandwidth filters. In addition, Keysight's ESA-E series spectrum analyzers use a hybrid structure: analog filters are used when the bandwidth is larger, and digital filters are used when the bandwidth is less than or equal to 300 Hz.
Remaining FM
The minimum usable resolution bandwidth is usually determined by the stability and residual FM of the local oscillators (especially the first local oscillator) in the analyzer. Early spectrum analyzer designs used unstable YIG (yttrium iron garnet) oscillators, which typically had residual FM of about 1 kHz. Since this instability is passed to the mixing products associated with the local oscillators, it is meaningless to reduce the resolution bandwidth below 1KHz because it is impossible to determine the exact source of this instability.
However, modern analyzers have greatly improved residual FM. For example, Keysight's high-performance X-Series signal analyzers have a residual FM of 0.25 Hz (nominal); the PSA Series spectrum analyzers have 1 to 4 Hz; and the ESA Series spectrum analyzers have 2 to 8 Hz. This allows the resolution bandwidth to be reduced to 1 Hz. Therefore, any instability that appears on the analyzer is caused by the input signal.
Phase Noise
No oscillator is absolutely stable. Although we cannot see the actual frequency jitter of the spectrum analyzer's local oscillator system, we can still observe obvious manifestations of local oscillator frequency or phase instability, which is phase noise (sometimes called noise sidebands).
They are all affected to some extent by frequency or phase modulation of random noise. As mentioned earlier, any instability of the LO is transmitted to the mixing products formed by the LO and the input signal, so the modulation sidebands of the LO phase noise appear around those spectral components whose amplitudes are much larger than the system broadband noise floor (Figure 2-11). The amplitude difference between the displayed spectral components and the phase noise varies with the stability of the LO. The more stable the LO, the smaller the phase noise. It also varies with the resolution bandwidth. If the resolution bandwidth is reduced by a factor of 10, the displayed phase noise level will be reduced by 10 dB5.
Figure 2-11. Phase noise is only visible when the signal level is much greater than the system noise floor.
The shape of the phase noise spectrum depends on the design of the analyzer, especially the structure of the phase-locked loop used to stabilize the local oscillator. In some analyzers, the phase noise is relatively flat in the bandwidth of the stable loop, while in others, the phase noise decreases with the frequency offset of the signal. Phase noise is expressed in dBc (dB relative to the carrier) and is normalized to a 1 Hz noise power bandwidth. Sometimes it is specified at a specific frequency offset, or a curve is used to show the phase noise characteristics over a range of frequency offsets.
Normally, we can only observe the phase noise of a spectrum analyzer at narrow resolution bandwidths, where it smears the edges of the response curves of these filters. Using digital filters as described earlier does not change this effect. For filters with wider resolution bandwidths, the phase noise is buried under the sidebands of the filter response curve, as in the case of the two unequal amplitude sine waves discussed earlier.
Some modern spectrum analyzers or signal analyzers allow the user to select different local oscillator stability modes to achieve the best phase noise in various measurement environments. For example, the high-performance X-Series signal analyzers offer three modes:
– Optimize for phase noise less than 140 kHz offset from the carrier. In this mode, the LO phase noise near the carrier is optimized, while the phase noise outside 140 kHz is not optimized.
– Optimize for phase noise greater than 160 kHz offset from the carrier. This mode optimizes the phase noise at offsets greater than 160 KHz from the carrier.
– Optimize for fast tuning. When this mode is selected, the LO characteristics are compromised for phase noise in all frequency ranges less than 2 MHz from the carrier. This allows for maximum measurement throughput in the shortest measurement time when changing center frequency or span.
Figure 2-12b. Detailed display at 140 kHz offset from carrier
The high-performance X-Series signal analyzers can also be set to automatic mode for phase noise optimization, which sets the instrument to the best speed and dynamic range for different measurement environments. The analyzer selects fast tuning mode when span > 44.44 MHz or resolution bandwidth > 1.9 MHz. In addition, the analyzer automatically selects the best near-end carrier phase noise when the center frequency is < 195 kHz or when the center frequency is ≥ 1 MHz and the span is ≤ 1.3 MHz and the resolution bandwidth is ≤ 75 kHz. In other cases, the analyzer automatically selects the best far-end phase noise.
In any case, phase noise is the ultimate limiting factor in the ability of a spectrum analyzer or signal analyzer to resolve signals of unequal amplitude. As shown in Figure 2-13, based on 3 dB bandwidth and selectivity theory, we should be able to resolve the two signals, but it turns out that the phase noise masks the smaller signal.
Figure 2-13. Phase noise prevents resolution of unequal amplitude signals.
Scan time
Analog resolution filter
If resolution is the only criterion for evaluating a spectrum analyzer, it would seem that the resolution (IF) filter of the spectrum analyzer should be designed to be as narrow as possible. However, resolution affects the sweep time, and we pay great attention to sweep time because it directly affects the time required to complete a measurement.
The reason for considering resolution is that the IF filter is a band-limited circuit that requires a finite amount of time to charge and discharge. If the mixing product sweeps through the filter too quickly, it will cause a loss of displayed amplitude as shown in Figure 2-14. (For other methods of dealing with IF response time, see "Envelope Detector" later in this chapter.) If we consider the time that the mixing product stays in the IF filter passband, this time is proportional to the bandwidth and inversely proportional to the sweep per unit time (Hz), that is:
Figure 2-14. Sweeping too fast causes a drop in displayed amplitude and a shift in the specified frequency.
The key conclusion we have drawn is that changes in resolution have a significant impact on sweep time. Older analog analyzers typically offer step values in the 1, 3, 10 pattern or in a ratio roughly equal to the square root of 10. So, for each change in resolution, the sweep time is affected by a factor of about 10. Keysight X-Series signal analyzers offer bandwidth steps of up to 10% to achieve a better compromise between span, resolution, and sweep time.
The spectrum analyzer will generally automatically adjust the sweep time according to the settings of the span and resolution bandwidth, and maintain a calibrated display by adjusting the sweep time. If necessary, we can set the sweep time manually without using automatic adjustment. If the required sweep time is shorter than the maximum available sweep time provided, the spectrum analyzer will display "Meas Uncal" in the upper right corner of the grid line to indicate that the display result is not calibrated.
Digital resolution filter
The digital resolution filters used in Keysight spectrum analyzers or signal analyzers have a different impact on sweep time than the analog filters described previously. For swept analysis, digitally implemented filters can sweep 2 to 4 times faster without more intensive processing.
X-Series signal analyzers equipped with Option FS1 can programmatically correct for the effects of too fast a sweep speed for resolution bandwidths between approximately 3 kHz and 100 kHz. As a result, sweep times can be reduced from seconds to milliseconds, depending on the specific setup. See Figure 2-14a. The sweep time, excluding the correction process, is 79.8 seconds. Figure 2-14b shows a sweep time of 1.506 seconds with Option FS1 installed. For these widest resolution bandwidths, the sweep times are already very fast. For example, at k = 2, 1 GHz span, and 1 MHz resolution bandwidth, the formula gives a sweep time of only 2 milliseconds.
For narrower resolution bandwidths, Keysight spectrum analyzers or signal analyzers use a fast Fourier transform (FFT) to process the data, so the sweep time will also be shorter than the formula predicts. Different analyzers will have different performance because the signal being analyzed is processed in multiple frequency ranges. For example, if the frequency range is 1 kHz, then when we choose a resolution bandwidth of 10 Hz, the analyzer is actually processing the data simultaneously in 1 kHz units through 100 adjacent 10 Hz filters. If digital processing could be instantaneous, then we would expect the sweep time to be reduced by a factor of 100. In practice, the reduction is less, but still very significant.
Figure 2-14b. Full span sweep speed with 20 kHz RBW and Option FS1
Envelope Detector
Older analyzers often use an envelope detector to convert the IF signal to a video signal.7 The simplest envelope detector consists of a diode, a load resistor, and a low-pass filter, as shown in Figure 2-15. The IF link output signal in this example (an amplitude-modulated sine wave) is fed to the detector, and the detector output response varies with the envelope of the IF signal, rather than the instantaneous value of the IF sine wave itself.
For most measurements, we choose a resolution bandwidth narrow enough to resolve the individual spectral components of the input signal. If the LO frequency is fixed and the spectrum analyzer is tuned to one of the spectral components of the signal, the IF output will be a stable sine wave with a constant peak value. The output of the envelope detector will then be a constant (DC) voltage with no variations that the detector needs to track.
However, sometimes we intentionally make the resolution bandwidth wide enough to include two or more spectral components, while in some cases there is no choice because the frequency interval between these spectral components is smaller than the narrowest resolution bandwidth. Assuming that there are only two spectral components in the passband, the two sine waves will affect each other to form a beat note, as shown in Figure 2-16. The envelope of the intermediate frequency signal will change with the phase change between the two sine waves.
The bandwidth of the resolution (IF) filter determines the maximum rate at which the envelope of the IF signal can change. This bandwidth determines how far apart the two input sine waves can be in frequency so that they can both fall within the filter passband after mixing. Assuming a final IF of 22.5 MHz and a bandwidth of 100 kHz, two input signals spaced 100 kHz apart will produce mixing products at 22.45 and 22.55 MHz, thus meeting the above criteria, as shown in Figure 2-16. The detector must be able to track the envelope changes caused by these two signals, not the envelope of the 22.5 MHz IF signal itself.
The envelope detector makes the spectrum analyzer a voltmeter. Let's consider again the case of two equal-amplitude signals in the IF passband. The power meter will indicate a level 3 dB higher than either signal, which is the combined power of the two signals. Assume that the two signals are close enough that the attenuation caused by the filter roll-off when the analyzer is tuned to the center of them is negligible. ( For this discussion, we assume that the filter has an ideal rectangular characteristic.)
The analyzer display will then vary between a voltage value of twice the signal level (greater than 6 dB) and 0 (negative infinity on a logarithmic scale). Remember that the two signals are sinusoidal signals (vectors) of different frequencies, so their phases with respect to each other are constantly changing, sometimes they are exactly in phase, adding their amplitudes, and sometimes they are exactly out of phase, subtracting their amplitudes.
Therefore, the envelope detector changes according to the changes in the signal peak value (rather than the instantaneous value) from the IF link, resulting in a loss of signal phase, which gives the spectrum analyzer the characteristics of a voltmeter.
The resolution bandwidth filter implemented by digital technology does not include an analog envelope detector, but uses digital processing to calculate the square root of the sum of the squares of the I and Q data, which is numerically the same as the output of the envelope detector.
A signal with a frequency range from zero (DC) to some higher frequency determined by circuit components. Early analog display technology used this signal to directly drive the vertical deflection of a CRT, so it was called a video signal.
show
Until the mid-1970s, spectrum analyzer displays were purely analog. The displayed trace showed the continuously changing signal envelope with no information loss. But analog displays had their own disadvantages, the main problem being the long sweep times required to process narrow resolution bandwidths. In extreme cases, the displayed trace became a slowly moving spot on a cathode ray tube (CRT) screen, with no actual trace. So, the long sweep times made the display meaningless.
Keysight Technologies (then part of Hewlett-Packard) pioneered a variable persistence memory CRT on which the fading rate of displayed information could be adjusted. If adjusted properly, a new trace would appear to update the display just as the old trace had just disappeared. The display was continuous, flicker-free, and free of the confusion caused by overlapping traces. It worked fairly well, but the brightness and fading rate had to be readjusted for each new measurement state.
In the mid-1970s, digital circuits were developed and were soon used in spectrum analyzers. Once a trace was digitized and stored in memory, it was permanently available for display. It became simple to refresh the display at a flicker-free rate without causing the image to blur or fade. It became simple to refresh the display at a flicker-free rate without causing the image to blur or fade.
Figure 2-17. When digitizing an analog signal, what value should each point show?
Detector Type
With a digital display, we need to decide what value should be used to represent each displayed data point. Regardless of how many data points we use on the display, each data point must represent the signal present within a certain frequency range or time interval (although time is not usually used when discussing spectrum analyzers).
This process is like putting the data of a certain time interval into a signal collection unit (bucket), and then applying a certain necessary mathematical operation to extract the information bits we want from this signal collection unit. Then the data is put into the memory and written to the display. This method provides great flexibility.
Here we will discuss 6 different types of detectors.
In Figure 2-18, each signal collection unit contains data of span and time frame determined by the following formula:
Figure 2-18. Each of the 1001 trace points (signal collection units) covers a frequency span of 100 kHz and a time span of 0.01 ms.
Frequency: Width of signal collection unit = Span width/(Number of trace points – 1)
Time: Width of signal collection unit = sweep time / (number of trace points – 1)
The sampling rate varies from instrument to instrument, but reducing the span and/or increasing the sweep time can yield higher accuracy, since either increases the number of samples contained in the signal collection unit. For analyzers using digital IF filters, the sampling rate and interpolation characteristics are designed to be equivalent to continuous-time processing.
The concept of "signal collection unit" is important, it can help us distinguish these 6 types of display detectors:
– Sampling detection
– Positive peak detection (peak detection for short)
Figure 2-19. Trace points stored in memory based on different detector algorithms
The first three types of detection (sample, peak, and negative peak) are relatively easy to understand, as shown intuitively in Figure 2-19. Normal, average, and quasi-peak detection are more complex and are discussed later.
Let us return to the previous question: How can we use digital technology to display analog systems as faithfully as possible? Let us imagine the situation described in Figure 2-17, that is, the displayed signal contains only noise and a continuous wave (CW) signal.
Sampling detection
As the first method, we only select the instantaneous level value at the middle position of each signal collection unit (as shown in Figure 2-19) as the data point. This is the sampling detection mode. In order to make the display trace look continuous, we designed a system that can depict the vector relationship between each point. Comparing Figures 2-17 and 2-20, we can see that we have obtained a reasonable display. Of course, the more points on the trace, the more realistic the analog signal can be reproduced. The number of available display points of different spectrum analyzers is different. For the X-Series signal analyzer, the number of sampling display points of the frequency domain trace can range from a minimum of 1 point to a maximum of 40001 points. As shown in Figure 2-21, increasing the sampling points can indeed make the result closer to the analog signal.
Although this sampling detection method can well reflect the randomness of noise, it is not suitable for analyzing sine waves. If a 100 MHz comb signal is observed on a high-performance X-Series signal analyzer, the analyzer span can be set to 0 to 26.5 GHz. Even if 1001 display points are used, each display point represents a 26.5 MHz frequency span (signal collection unit), which is much larger than the maximum resolution bandwidth of 8 MHz.
As a result, when using the sampling detection mode, the amplitude of the comb signal can only be displayed when the mixing product is exactly at the center of the intermediate frequency. Figure 2-22a is a display with a bandwidth of 750 Hz and a span of 10 MHz using sampling detection. Its comb signal amplitude should be basically consistent with the actual signal shown in Figure 2-22b (using peak detection). It can be concluded that the sampling detection method is not suitable for all signals and cannot reflect the true peak value of the displayed signal. When the resolution bandwidth is smaller than the sampling interval (such as the width of the signal collection unit), the sampling detection mode will give incorrect results.
Figure 2-22b. Actual comb signal obtained using (positive) peak detection in a 10 MHz span.
(Positive) Peak Detection
One way to ensure that the true amplitude of all sine waves can be recorded is to display the maximum value that appears in each signal collection unit. This is the positive peak detection method, or peak detection, as shown in Figure 2-22b. Peak detection is the default detection method of many spectrum analyzers because it ensures that no sine signal is lost regardless of the relationship between the resolution bandwidth and the width of the signal collection unit. However, unlike the sampling detection method, since peak detection only displays the maximum value in each signal collection unit and ignores the actual randomness of the noise, it is not ideal in reflecting random noise. Therefore, spectrum analyzers that use peak detection as the first detection method generally also provide sampling detection as a supplement.
Negative peak detection
Negative peak detection displays the minimum value in each signal collection unit. Most spectrum analyzers offer this detection method, although it is not as common as the other methods. For EMC measurements, negative peak detection can be useful to distinguish CW signals from pulsed signals. Later in this application note, we will see that negative peak detection can also be used to identify signals when making high-frequency measurements using an external mixer.
Normal detection
To provide a better visual display of random noise than peak detection and to avoid the problem of signal loss in the sample detection mode, many spectrum analyzers also provide a normal detection mode (commonly known as Rosenfell9 mode). If the signal has both rising and falling edges as determined by positive and negative peak detection, the algorithm will classify the signal as a noise signal.
Roesnfell is not a person's name, but a description of an operation method used to test whether the signal within the signal collection unit represented by a given data point is rising or falling. It is sometimes written as rose'n'fell.
In this case, odd-numbered data points are used to display the maximum value in the signal collection unit, and even-numbered data points are used to display the minimum value. This is shown in Figure 2-25. The normal detection mode and the sampling detection mode are compared in Figures 2-23a and 2-13b. (Since the sampling detector is very effective in measuring noise, it is often used in noise cursor applications. Similarly, a detection type that can provide results without any tendency is required in channel power measurement and adjacent channel power measurement. Peak detection is suitable for this. For spectrum analyzers without average detection function, sampling detection is the best choice.)
What happens when a sinusoidal signal is encountered? We know that when the mixing component passes through the intermediate frequency filter, the characteristic curve of the filter will be depicted on the spectrum analyzer display. If the filter curve covers many display points, the following situation will occur: the displayed signal will only rise when the mixing component is close to the center frequency of the filter, and will only fall when the mixing component is far away from the center frequency of the filter. In either case, positive peak and negative peak detection can detect the amplitude change in a single direction, and according to the normal detection algorithm, display the maximum value in each signal collection unit, as shown in Figure 2-24.
What happens when the resolution bandwidth is narrower than the bin? The signal now rises and falls within the bin. If the bin happens to be odd, everything works fine and the maximum value within the bin will be plotted directly as the next data point. However, if the bin is even, the minimum value within the bin will be plotted. Depending on the ratio of the resolution bandwidth to the bin width, the minimum value may be partially or completely different from the true peak value (the value we want to display). In the extreme case where the bin width is much larger than the resolution bandwidth, the difference between the maximum and minimum values within the bin will be the difference between the signal peak and the noise, as is the case in the example in Figure 2-25. Looking at the 6th bin, the peak in the current bin is always compared to the peak in the previous bin, and when the bin is odd (such as the 7th bin), the larger value of the two is displayed. This peak actually occurs in the 6th bin, but is not displayed until the 7th bin.
Figure 2-24. When the value in the signal collection unit only increases or decreases, the normal detection shows the maximum value in the unit
Normal detection algorithm:
If the signal value rises and falls within a signal collection unit: the even-numbered signal collection unit will display the minimum value (negative peak value) within the unit. The maximum value will be recorded, and then the peak value within the current unit will be compared with the peak value of the previous (recorded) unit in the odd-numbered signal collection unit and the larger value (positive peak value) will be displayed. If the signal only rises or only decreases within a signal collection unit, the peak value will be displayed, as shown in Figure 2-25.
This process may cause the maximum value of the data point to be displayed too far to the right, but this offset is usually only a small percentage of the span. Some spectrum analyzers, such as the high-performance X-Series signal analyzers, compensate for this potential effect by adjusting the start and stop frequencies of the local oscillator.
Another error is to display two peaks when there is only one peak. Figure 2-26 shows an example of how this can happen. When using a wider resolution bandwidth and peak detection, two peak outlines are displayed.
Therefore, peak detection is most suitable for locating CW signals from noise, sampling detection is most suitable for measuring noise, and normal detection is most appropriate when both signals and noise need to be observed.
Figure 2-26. Normal detection shows two peaks when there is only one
Average detection
Although modern digital modulation schemes have noise-like characteristics, sampling detection cannot provide all the information we need. For example, when measuring the channel power of a W-CDMA signal, we need to integrate the RMS value of the signal. This measurement process involves the total power of the signal collection unit within a certain frequency range of the spectrum analyzer, and sampling detection cannot provide this information.
Although a general spectrum analyzer collects amplitude data multiple times in each signal collection unit, sampling detection only retains one value of the data and ignores other values. Average detection uses all the data in the signal collection unit within the time (and frequency) interval. Once the data is digitized and we know the environment in which it is implemented, we can process the data in a variety of ways to obtain the desired results.
Some spectrum analyzers call the detection that averages power (based on the RMS value of voltage) rms (root mean square) detection. The average detection function of the Keysight X-Series signal analyzer includes power averaging, voltage averaging, and signal logarithmic averaging. Different averaging types can be selected individually by pressing a button:
Power (rms) averaging is the average of the root mean square level of the signal, which is obtained by taking the square root of the sum of the squares of the voltage values measured in a signal collection unit and dividing it by the input characteristic impedance of the spectrum analyzer (usually 50 Ω). Power averaging calculates the true average power and is most suitable for measuring the power of complex signals.
Voltage averaging is the averaging of the linear voltage values of the signal envelope measured in a signal collection unit. This method is often used to measure narrowband signals in EMI testing (this part will be discussed further in the next section). Voltage averaging can also be used to observe the rise and fall of AM signals or pulse modulation signals (such as radar signals, TDMA transmission signals).
Log power (video) averaging is the averaging of the logarithmic amplitude values (in dB) of the signal envelope measured within a signal collection unit. It is best used to observe sinusoidal signals, especially those close to noise. 11
Therefore, using an average detector with power as the average type provides the true average power based on the rms voltage value, and a detector with voltage as the average type can be regarded as a general average detector. There is no other equivalent method for a detector with logarithmic average type.
The use of average detection to measure power is an improvement over sampling detection. Sampling detection requires multiple scans to obtain enough data points to provide accurate average power information. Average detection changes the measurement of channel power from the sum of signal collection units within a certain range to the synthesis of time intervals representing a certain frequency range of the spectrum analyzer. In a fast Fourier transform (FFT) spectrum analyzer 12, the value used to measure channel power changes from the sum of displayed data points to the sum of FFT transformation points.
In both sweep and FFT modes, this synthesis captures all available power information, rather than just the power information at the sampling point as with sample detection. So when the measurement time is the same, the average detection results are more consistent. In sweep analysis, the stability of the measurement results can also be improved simply by extending the sweep time.
EMI detectors: average and quasi-peak detection
An important application of average detection is to detect the electromagnetic interference (EMI) characteristics of equipment. In this application, the voltage averaging method described in the previous section can measure narrowband signals that may be obscured by broadband pulse noise. The average detection used in EMI test instruments will take the envelope to be measured and pass it through a low-pass filter with a bandwidth much smaller than the RBW. This filter integrates (averages) the high-frequency components of the signal (such as noise). To implement this type of detection in an older spectrum analyzer that does not have voltage average detection, set the spectrum analyzer to linear mode and select a video filter whose cutoff frequency must be less than the minimum PRF (pulse repetition frequency) of the signal to be measured.
Quasi-peak detection (QPD) is also used in EMI testing. QPD is a weighted form of peak detection, and its measurement value decreases as the repetition rate of the measured signal decreases. That is, a pulse signal with a given peak amplitude and a pulse repetition rate of 10 Hz has a lower quasi-peak value than another signal with the same peak amplitude but a pulse repetition rate of 1 kHz. This signal weighting is achieved by a circuit with a specific charge and discharge structure and a display time constant defined by CISPR.
CISPR, the International Special Committee on Radio Interference, was established in 1934 by a number of international organizations to address radio interference. It is a non-governmental organization composed of members of the International Electrotechnical Commission (IEC) and many other international organizations, and its recommended standards often become the basis for statutory EMC test requirements adopted by government regulatory agencies around the world.
QPD is also a way to quantitatively measure the "interference factor" of a signal. Imagine that we are listening to a radio station that is being interfered with. If we only hear the occasional "hiss" caused by the noise every few seconds, then we can basically listen to the program normally. However, if the interference signal of the same amplitude appears 60 times per second, we can no longer listen to the program normally.
Smoothing
There are several different ways to smooth out the amplitude variations of the envelope detector output in a spectrum analyzer. The first method is average detection, which has been discussed previously. There are two other methods: video filtering and trace averaging14. These are described below.
Video Filtering
Identifying signals close to noise is not just a problem for EMC measurements. As shown in Figure 2-27, the spectrum analyzer display is the measured signal plus its own internal noise. To reduce the effect of noise on the displayed signal amplitude, we often smooth or average the display, as shown in Figure 2-28. The variable video filter included in the spectrum analyzer is used for this purpose. It is a low-pass filter that is located after the envelope detector and determines the bandwidth of the video signal that will later be digitized to generate amplitude data. The cutoff frequency of this video filter can be reduced to less than the bandwidth of the selected resolution bandwidth (IF) filter. At this point, the video system will no longer be able to follow the rapid changes in the signal envelope passing through the intermediate frequency link. The result is an averaging or smoothing of the displayed signal.
Figure 2-29. Smoothing effect when the VBW to RBW ratio is 3:1, 1:10, and 1:100
This effect is most noticeable when measuring noise, especially when high resolution bandwidth is used. As the video bandwidth is reduced, the peak-to-peak fluctuations in the noise are reduced. As shown in Figure 2-29, the degree of reduction (the degree of averaging or smoothing) varies with the ratio of the video bandwidth to the resolution bandwidth. When the ratio is less than or equal to 0.01, the smoothing effect is good, while as the ratio increases, the smoothing effect is less ideal. The video filter will not have any effect on an already smoothed signal trace (for example, a displayed sinusoidal signal that is already well distinguished from the noise).
If the analyzer is set to positive peak detection mode, two things can be noticed: First, if VBW > RBW, then changing the resolution bandwidth has little effect on the peak-to-peak fluctuations of the noise. Second, if VBW < RBW, then changing the video bandwidth appears to affect the noise level. The noise fluctuations do not change much because the analyzer is currently only showing the peak of the noise. However, the noise level appears to change with the video bandwidth, which is due to the change in the averaging (smoothing) process, which changes the peak of the smoothed noise envelope, as shown in Figure 2-30a. When the average detection mode is selected, the average noise level does not change, as shown in Figure 2-30b.
Figure 2-30b. Average detection mode: the noise level remains constant regardless of the VBW to RBW ratio (3:1, 1:10, 1:100).
Since the video filter has its own response time, when the video bandwidth VBW is smaller than the resolution bandwidth RBW, the change in scan time is approximately inversely proportional to the change in video bandwidth. The scan time (ST) is described by the following formula:
The analyzer automatically sets the corresponding sweep time according to the video bandwidth, span and resolution bandwidth.
Trace averaging
The digital display provides another option for smoothing the display: trace averaging. This is a completely different process from using an average detector. It averages by scanning two or more times point by point, and the new value of each displayed point is obtained by averaging the current value and the previous average value:
Therefore, after a number of scans, the display will gradually converge to an average value. By setting the number of scans for which the averaging occurs, the degree of averaging or smoothing can be selected, just like video filtering. Figure 2-31 shows the trace averaging effect obtained with different scan numbers. Although trace averaging does not affect the scan time, because multiple scans require a certain amount of time, the time required to achieve the desired averaging effect is roughly the same as the time used by video filtering.
Figure 2-31. The average trace effect when the number of scans is 1, 5, 20, and 100 (the corresponding trace position offset of each group of scans is from top to bottom)
In most cases, it does not matter which display smoothing method you choose. If the signal being measured is noise or a low-level sinusoidal signal that is very close to noise, the same effect will be achieved whether you use video filtering or trace averaging.
However, there is still a clear difference between the two. Video filtering averages the signal in real time, that is, as the scan progresses, we see the full average or smoothing effect of each displayed point on the screen. Each point is averaged only once, and the processing time on each scan is about 1/VBW. Trace averaging requires multiple scans to achieve full averaging of the displayed signal, and the averaging process at each point occurs over the entire time period required for multiple scans.
Therefore, for some signals, different smoothing methods will produce completely different effects. For example, when video averaging is used for a signal whose spectrum changes over time, different average results will be obtained for each scan. However, if trace averaging is selected, the result will be closer to the actual average value, as shown in Figures 2-32a and 2-32b.
Figures 2-32a and 2-32b show the effects of applying video filtering and trace averaging, respectively, to an FM broadcast signal.
A spectrum analyzer with time gating function can obtain spectrum information of signals that occupy the same part in the frequency domain but are separated from each other in the time domain. By adjusting the interval between these signals using an external trigger signal, the following functions can be achieved:
– Measure any of multiple signals separated from each other in the time domain (for example, you can separate the spectra of two wireless signals that are time-divided but have the same frequency)
– Measure the signal spectrum of a time slot in a TDMA system
– Eliminate the spectrum of interfering signals, such as removing transient processes on the edge of periodic pulses that only exist for a period of time
Why is time gating necessary?
Traditional frequency domain spectrum analyzers can only provide limited information when analyzing certain signals. These difficult signal types include:
– RF pulses
– Time multiplexing
– Time Division Multiple Access (TDMA)
– Spectrum interleaved or non-contiguous
– Pulse modulation
In some cases, time gating can help you make measurements that would otherwise be difficult, if not impossible, to make.
Measuring time division duplex signals
Figure 2-33a shows how to use time gating to perform complex measurements. It shows a simplified digital mobile signal consisting of radios #1 and #2 occupying the same frequency channel but time sharing. Each signal transmits a 1 ms pulse, then turns off, and the other signal transmits again for 1 ms. The key is how to measure the individual spectrum of each transmitted signal.
Figure 2-33a. Simplified digital mobile radio signal in the time domain
Unfortunately, a traditional spectrum analyzer cannot do this. It can only display the mixed spectrum of the two signals, as shown in Figure 2-33b. However, a modern analyzer, using the time gating function and an external trigger signal, can observe the spectrum of the separate wireless signal #1 (or #2) and determine whether it has the displayed spurious signal, as shown in Figure 2-33c.
Adjusting these parameters allows you to observe the signal spectrum of a desired time period. If there is only one gate signal in the time period of interest, you can use the level gate signal as shown in Figure 2-34. However, in many cases, the timing of the gate signal will not completely match the spectrum we want to measure. Therefore, a more flexible method is to use edge trigger mode in combination with the specified gate delay and gate pulse width to accurately define the time period of the signal you want to measure.
Figure 2-35. A TDMA signal with 8 time slots (in this case, a GSM signal), with time slot 0 being “off”.
Consider a GSM signal with 8 time slots as shown in Figure 2-35. Each burst is 0.577 ms long and the entire frame is 4.615 ms long. We may be interested in the spectrum of the signal only within a specific time slot. In this example, we assume that two of the 8 available time slots are used (time slots 1 and 3), as shown in Figure 2-36. When observing this signal in the frequency domain, see Figure 2-37, we observe that there are unwanted spurious signals in the spectrum. To resolve this issue and find the source of the interfering signal, we need to determine which time slot it appears in. If we want to observe time slot 3, we can set the gate trigger to the rising edge of the burst in time slot 3 and specify a gate delay of 1.4577 ms and a gate pulse width of 461.60 μs, as shown in Figure 2-38. The gate delay ensures that we only measure the spectrum of the signal in time slot 3 during the entire duration of the burst. Note that the gate start and stop values must be carefully chosen to avoid the rising and falling edges of the burst train, as some time is required for the RBW filtered signal to settle before measurement. Figure 2-39 shows the spectrum for time slot 3, indicating that the spurious signal is not caused by this burst.
Figure 2-39. The spectrum of time slot 3 shows that the spurious signal is not caused by this burst.
Gated FFT
Keysight X-Series signal analyzers have built-in FFT capabilities. In this mode, the analyzer captures data and performs FFT processing after the selected delay after the trigger is enabled. The IF signal is digitized and acquired in a time period of 1.83/RBW. The FFT is calculated based on this data acquisition to obtain the spectrum of the signal. Therefore, the spectrum exists at a specific time in a known period of time. This is the fastest gating technique when the analyzer span is narrower than the maximum width of the FFT.
To obtain the greatest possible frequency resolution, the smallest RBW available for the spectrum analyzer should be selected (its capture time is adapted to the time period to be measured). However, this is not always necessary in practice, and you can choose a wider RBW and reduce the gate pulse width accordingly. The minimum available RBW in FFT gating applications is usually narrower than the minimum available RBW for other gating techniques, because in other techniques the IF must be fully stable during the pulse duration, which requires longer time than 1.83/RBW.
LO gating
LO gating, sometimes called sweep gating, is another time gating technique. In LO gating mode, we sweep the LO by controlling the ramp voltage generated by the sweep generator, as shown in Figure 2-40. Like all spectrum analyzers, when the gating signal is turned on, the LO signal climbs in frequency. When the gating is turned off, the sweep generator output voltage is fixed and the LO stops rising in frequency. Because this technique can measure multiple signal collection units during the duration of each burst pulse signal, it is much faster than video gating. Let's take the GSM signal mentioned above as an example.
Figure 2-40. In LO gate mode, the LO sweeps only during the gate interval.
LO gating
LO gating, sometimes called sweep gating, is another time gating technique. In LO gating mode, we sweep the LO by controlling the ramp voltage generated by the sweep generator, as shown in Figure 2-40. Like all spectrum analyzers, when the gating signal is turned on, the LO signal climbs in frequency. When the gating is turned off, the sweep generator output voltage is fixed and the LO stops rising in frequency. Because this technique can measure multiple signal collection units during the duration of each burst pulse signal, it is much faster than video gating. Let's take the GSM signal mentioned above as an example.
It takes 14.6 ms to sweep a 1 MHz span with the standard non-gated X-Series signal analyzer, as shown in Figure 2-41. If the gate pulse width is 0.3 ms, the spectrum analyzer must scan within 49 (14.6 divided by 0.3) gate signal intervals; if the full frame length of the GSM signal is 4.615 ms, then the total measurement time is equal to 49 gate signal intervals multiplied by 4.615 ms, which equals 226 ms. This is a significant improvement in speed compared to the video gating technology described later. Both the X-Series signal analyzer and the PSA series spectrum analyzer have local oscillator gating capabilities.
Some spectrum analyzers (including Keysight 8560, 8590 and ESA series) use video gating signal analysis technology. In this case, the video voltage is turned off or "negative infinity" when the gating signal is in the cutoff state. The detector is set to peak detection, and the sweep time must be set to ensure that the gating signal appears at least once in each display point or signal collection unit, so as to ensure that the peak detector can obtain the real data in the corresponding time interval, otherwise there will be trace points without data values, resulting in an incomplete display spectrum. Therefore, the minimum sweep time = display points N x burst pulse time period. For example, in GSM measurement, the complete frame length is 4.615 ms. Assuming that the ESA spectrum analyzer is set to the default display point number of 401, the minimum sweep time for GSM video gating measurement is 401 x 4.615 ms = 1.85 s.
Some TDMA formats have cycle times as long as 90 ms, resulting in very long sweep times when using video gating techniques. Now that you know how a typical analog spectrum analyzer works and how to use some of its important features, the next step is to discuss how spectrum analyzer performance improves when digital technology replaces some of the analog circuitry.
Good, the introduction is very detailed. The basic skills required to design this kind of professional RF instrument must be very solid, including hardware and software.
Details
Published on 2022-6-12 13:14
Good, the introduction is very detailed. The basic skills required to design this kind of professional RF instrument must be very solid, including hardware and software.
Fred_1977 posted on 2022-6-12 13:14 Very good, the introduction is very detailed. The basic skills of designing such professional RF instruments must be very solid, including hardware and software.
Is it something that can be learned from technology? Very good
提升卡
变色卡
千斤顶