Spectrum analyzers get real.
Spectrum analyzers have been around for a long time--see reference 1 for an example of a Hewlett-Packard 1964 microwave analyzer--but real-time instruments are significantly different. According to an Agilent real-time spectrum analyzer technical overview document, "In a spectrum or signal analyzer with a digital intermediate frequency (IF) section, real-time operation is a state in which all signal samples are processed for some sort of measurement result or triggering operation. ... In addition to gap-free analysis, a real-time RF analyzer may be defined as having four more key attributes: high-speed measurements, consistent measurement speed, frequency mask triggering, and advanced composite displays." (2)
Other companies have similar definitions, the distinguishing factor being the digital signal processing (DSP) that has replaced the final IF section.
Traditional, analog designs frequency-convert the input signal to a suitable IF with a bandwidth larger than the widest bandwidth IF filter--the resolution bandwidth (RBW) filter. In the frequency-conversion process, a mixer is driven by the RF input signal and a sweeping local oscillator (LO). When the output of the mixer falls within the IF bandwidth, it is amplified and applied to the RBW filter.
The purpose of the filter is to discriminate between frequencies that are close to each other, so the filter's rejection some distance away from the center frequency is important. Analog designs use a diode detector to convert the RBW filter output envelope to a baseband signal that may be further filtered by a video bandwidth (VBW) filter before display.
Most companies making spectrum analyzers gradually have replaced analog circuitry with digital for many reasons. Eliminating drift over time and temperature as well as improving accuracy are among the usual reasons that this general trend has occurred in instrument design. For spectrum analyzers, there are a large number of additional benefits.
Digital RBW Filters
The digital IF overview presented in an Agilent application note stated, "A partial implementation of digital IF circuitry is implemented in the Agilent ESA-E series spectrum analyzers. While the 1-kHz and wider RBWs are implemented with traditional analog LC and crystal filters, the narrowest bandwidths (1 Hz to 300 Hz) are realized using digital techniques." (3)
Rather than sweep the LO, in ESA-E series instruments, the LO steps in 900-Hz increments when narrow RBWs are selected. Sufficient steps must be taken to cover the required analysis span. The output from a mixer with a fixed LO remains a time-domain signal but centered at a different frequency. After sampling by an ADC, FFT filters produce the final RBW filtered data.
Analog implementations of narrow RBW filters present several problems that the digital approach solves. Analog filters contain energy-storage elements that must charge/discharge to accommodate transients. This means that sweeping the LO must be done slowly to avoid distortion. In addition, the selectivity of a narrow bandwidth analog filter is much less than that of an equivalent digital filter.
As described in reference 4, "the ... [R&S FSP and FSU analyzers were] equipped with FFT filters for the narrow bandwidths. Multiple narrowband FFTs were concatenated to a trace representing the selected frequency span. As the computing time for the FFTs was small compared to the settling time for narrow RBW filters, the FFT method provided a great speed advantage over the traditional sweep method." Further digital processing implements the equivalent of the VBW function and scaling before display.
Making the next step to a completely digital IF has become possible because of advances in high-speed ADCs and DSP hardware such as FPGAs. Figure 1 shows a block diagram of the all-digital IF in Agilent's PSA series instruments. This is a good example of the continuing relocation of the ADC to a position ever closer to the RF input in a swept-tuned instrument.
The PSA achieved a 10-MHz bandwidth using an ADC clocked at a 30-MHz rate. Although the ADC resolution was limited, the PSA implemented an autoranging scheme that took advantage of the time delay through the multipole anti-alias filter to vary analog pre-ADC gain as well as digital post-ADC gain. This scheme effectively increased the ADC dynamic range on the fly.
Returning to Figure 1, the Hilbert transform block following the ADC separates the samples into I and Q streams. Because the physical RF input and its sampled representation are analytic, the imaginary (I) part of the data is the Hilbert transform of the real (Q) part. After digital filtering, the I and Q signals are converted to a magnitude and phase representation. VBW filtered and scaled to correspond to power or voltage, and displayed.
For narrow RBWs, the LO is stepped rather than swept, providing a time-domain signal at the input to the RISC processor FFT block. RBW filtering and decimation for narrow spans are accomplished there.
The Move to Real Time
Swept-tuned analyzers, whether they use a digitally implemented IF or not, can miss a transient burst occurring at a frequency far away from that being swept at any instant. Narrow analog RBWs aggravate the situation because of the associated slow sweep speed. Analyzers that implement narrow RBWs via FFT filters have much faster sweeps but still can miss transients.
Instead, real-time analyzers operate in a stepped fashion, dealing with blocks of time-domain data in bandwidths as wide as 160 MHz in Agilent's N9030AK-RT2 PXA signal analyzer, 110 MHz in Tek's RSA5000 and RSA6000 instruments, and 40 MHz in the R&S FSVR.
All of the usual analog problems must be addressed in the overall design of a real-time analyzer, such as the number and positioning of the mixers and IF frequencies, LO purity, preselection, and dynamic range. Assuming a good RF/microwave solution ahead of the final mixer, the real-time capabilities are based on the analog-to-digital conversion fidelity and the DSP that follows, in particular the FFT.
Compared to Figure 1, Figure 2 shows a very different view of the circuit functions from the ADC to the display found in Agilent's PXA real-time signal analyzer. Some simplified signal analyzer block diagrams depict the ADC followed by a large FPGA indicating that the high-speed hardware is implemented in that way. Some calculations help to explain a few of the details.
Agilent's ADC is sampled at a 400-MS/s rate to achieve the desired 160-MHz bandwidth. The minimum transient duration that has a 100% probability of intercept (P01) is 3.57 ps, which is about 150 ns longer than the reciprocal of the 292,968 FFTs/s rate, 3.413 ps. This means that the ADC has time to convert more than 1,000 samples for each FFT and that transients will be included in at least one FFT'.
A similar calculation for Tek's 110-MHz bandwidth RSA6000 with a 300-MHz clock rate and 292,969 FFTs/s yields the same 3.413 [micro]s per FFT, and the clock rate is high enough to acquire more than 1,000 samples. Minimum transient duration is 3.7 is, so again, at least one FFT will see a transient that long. For the R&S FSVR, the minimum signal duration is 24 [micro]s for the correct power to be displayed.
While National Instruments (NI) does not offer a dedicated real-time spectrum analyzer product out of the box, as Nikhil Ayer, product manager--RF/microwave test, explained, "Our modular PXI vector signal analyzers can be extended with user-defined FPGA processing to create such systems. NI FlexRIO FPGA modules Offer DSP-focused FPGAs for real-time signal processing, peer-to-peer data streaming for high-bandwidth baseband data movement, and abstract graphical programming through NI Lab VIEW FPGA software. When combined with one of our VSAs such as the PXIe-5665 or PXIe-5667, users can create a real-time spectral analyzer or address any other application which requires real-time signal processing. M vector signal transceivers take this a step further because they are software-designed instruments."
It's obvious that things have to operate quickly to catch transients, but what do you do with all the data? It's important to remember that the modulation on the frequency-shifted RF input is dynamic: it could be constantly changing. So, each successive FFT shows the state of the signal as it changes in time. This is where the various forms of color-graded persistence add tremendous value by helping you to distinguish among frequently and less frequently appearing signals and even see signals hidden behind others.
Tek's InstaVu oscilloscopes from the 1990s introduced color-graded persistence, subsequently enhanced and relabeled as DPX technology. The basic idea is to run successive short acquisitions at a very high rate--lots of waveforms/s. Data from each acquisition is added to a sample-number vs. signal-amplitude pixel map in which a counter at each location keeps track of how many times the waveform has gone through that point.
The map is read out at a 30-Hz rate to avoid display flicker, and the data is displayed as a range of colors corresponding to the accumulated count at each pixel. The same scheme suits display of high-rate, short FFTs and has been adopted by the major real-time signal analyzer manufacturers.
For a couple of reasons, representing very short events is challenging. If the event occurs partly in one data acquisition block and partly in the next, the FFTs on those two blocks will each show reduced amplitude results. Also, because a windowing function is applied ahead of the FFT calculation, the position of a short event within the block of ADC data matters. Events near the ends of the block will be reduced in amplitude.
The overlapping of FFTs solves both problems. Rather than computing the FFT of one data block and then the next, the second block may contain the last .80% of the first block and only 20% new data. Successively repeating this process causes the position of the captured event to move through the center of the windowing function and eliminates errors caused by events split between data blocks.
Real Time vs. Swept Tuned
Swept Analyzers Remain Relevant
Real-time signal analyzers clearly have the edge for difficult signals such as those that hop and occupy a large bandwidth. And, transient bursts that interfere with another signal may be very hard to identify without a real-time analyzer. Nevertheless, they aren't always the best solution.
According to Darren McCarthy, A&D technical marketing manager at R&S, "The continued use of swept analysis has an important mission to support legacy defense ATE equipment and the continued code emulation of test methods written many years ago that were based on analog detection and timing. There are practical everyday uses of swept analyzers that are not readily replaced by real-time analysis:
* Tracking generators and source control of swept signal generators are not integrated into real-time analysis methods.
* Video outputs typically used the analog detectors--many real-time analyzers no longer support video output measurement methods used in some EMC test methods and raster scanning applications.
* [Swept analyzers support] visual detection of pulse recurrence frequency (PRF) rates--while pulse measurements are typically available using digital acquisition, the viewing of the 'picket fences' created by a swept analyzer could give a trained user an instantaneous understanding of PRF rates, staggered pulse rate interval, or other timing artifacts across die spectrum of a pulse."
Rigol Technologies does not make a real-time analyzer but has taken advantage of ADC and FPGA advances to develop low-cost swept-tuned spectrum analyzers. Chris Armstrong, general manager at Rigol Technologies USA, explained, "In building a modern swept analyzer, we started by re-envisioning the IF section in an all digital approach. The cutting-edge digital components that are available allowed our engineering team to build a digital FPGA-based platform to control the front end and generate high-speed, accurate frequency analysis with an impressive cost advantage. Additionally, the FPGA foundation allows us to incrementally add features and improve every aspect of the instrument over time since we are not limited by any expensive-to-change ASIC hardware."
Armstrong added that swept spectrum analyzers are used where wideband analysis or a tracking generator is needed. The amplitude response of a wideband RF filter is shown in Figure 3. The single sweep extends for more than 80 MHz and was made with Rigol's DSA815-TG 1.5-GHz spectrum analyzer and built-in tracking generator.
Agilent's Richard Overdorf, applications engineer and product planner, said that swept analysis was preferred for virtually all cases in which a static signal environment existed. For constant or CW signals, swept operation gives the user more flexibility with wider spans and RBW choices, and real-time generally is of less value, he said.
For example, Overdorf continued, "A very common measurement that can become significantly more time-consuming and exhibit degraded dynamic range with a [real-time] stitched FFT is wide-span spurious measurements. ... [In a real-time analyzer], high-order image-reject filters generally are used with associated amplitude and phase ripple."
The vendors participating in this article agreed that aerospace and defense, surveillance, and radar applications often involved short time-duration pulses. These challenging signals require an instrument with both good low-level sensitivity and the benefits of real-time operation.
If input signals are accepted across the entire frequency range of a spectrum analyzer, it's virtually certain that unwanted signals will interfere with those in the band being analyzed. To avoid this problem, conventional swept analyzers use a tunable YIG bandpass filter ahead of the first mixer stage. Not only does this approach eliminate interference, it also minimizes radiation from the LO source through the input connector--an important consideration in secure applications.
Unfortunately, YIG filters are limited to about 50-MHz bandwidth, which is far too narrow for the very wideband analysis required for today's signals. Even within the 50-MHz band pass, the phase changes significantly near each edge. Further, because of the hysteresis associated with the magnetic tuning, the filter center frequency cannot be shifted to a different value and then returned exactly to the initial setting. A Tek article (5) considers 35 to 40 MHz as the practical YIG preselector bandwidth. Agilent's PXA signal analyzer uses a YIG preselector with a bandwidth that varies from 46 to 74 MHz depending on the analyzer bandwidth and the tuned center frequency.
Bypassing the preselection filter has been the only option available for making very wide band measurements. For example, option 067 for Anritsu's MS2830A provides preselector bypass that supports 125-MHz wideband measurements to 43 GHz on an appropriately configured instrument.
Instead, Tek has developed a switched filter bank to replace the YIG preselection filter. Although Tek doesn't specify details of the filters, the Tektronix Component Solutions group sells a nine-channel 8- to 22-GHz switched filter bank that was "developed by the same microwave technology group behind Tektronix' Real-Time Spectrum Analyzers" as stated in the datasheet.
The filters have between 1.5- and 2.0-GHz bandwidths and are claimed to provide very stable and repeatable performance. They are intended for "super-heterodyne systems where the mixer input band must be a subset of the input frequency band," according to the datasheet. Because the filter performance does not conflict with wide bandwidth measurements, they do not need to be bypassed.
NI also has developed a switched filter bank--the PXIe 5693 RF preselector module with 16 fixed-tuned suboctave bandpass filters that covers the frequency range from 20 MHz to 7 GHz. The module includes notch filters to suppress TV band (55 MHz to 75 MHz) and FM band (88 MHz to 108 MHz). Preamplifiers following the preselection filters allow for a low-noise figure without compromising distortion performance.
What Else Can It Do?
Having solved the basic signal measurement requirements digitally, many opportunities exist within a real-time signal analyzer for increased functionality. A major feature is frequency mask triggering, which compares a spectrum mask to successive FF Is as they are performed. This means that you can trigger the instrument when some predefined combination of frequencies occurs in the same way that an oscilloscope triggers on a time-domain waveform mask.
Tek's Matt Maxwell, product manager for the RTSA, said that you need to be able to "trigger on any signal that you can see on the DPX display within the real-time bandwidth. Not only is this a frequency-selective trigger, but this allows you to trigger on signals underneath other signals."
Digital techniques also have been applied to oscilloscope data processing, bringing scopes and signal analyzers much closer together. For example, the R&S RTO-K1 1 option permits I/Q data to be extracted from digitally modulated signals that the RTO has acquired. The I/Q data can be exported to analysis software such as Matlab or LabVIEW with an adjustable sampling rate. Similarly, Agilent's 89600 VSA software can provide very wideband analysis for signals captured on a high-speed oscilloscope.
Anritsu has used an application solution approach in its signal analyzer design. Many options are available including extensive recording and playback/signal generation. As explained by Patrick Weisgarber, business development manager at the company, "Combining the ... [features] along with a real-time vector signal analyzer in a single instrument provides engineers with an expanded analysis capability. It also has cost and space benefits because it eliminates the need for multiple instruments and their associated expenses and bench real estate."
R&S's McCarthy summed up the recent advances. "For several years, most all spectrum and signal analyzer platforms have had both swept and vector signal analysis capability. This has been a platform evolution from spectrum analyzer vendors for over a decade. While different bandwidths for vector signal analysis capability typically are available for time-domain vector analysis, with enough platform insight, modern designs can add real-time analysis to vector signal analysis hardware with a software option."
For More Information Agilent Technologies www.rsleads.com/306ee-185 Anritsu www.rsleads.com/306ee-186 National Instruments www.rsleads.com/306ee-187 Rigol Technologies www.rsleads.com/306ee-188 Rohde & Schwarz www.rsleads.com/306ee-189 Tektronix www.rsleads.com/306ee-190
(1.) Lecklider, T., "Spectrum Analyzers Digitally Enhanced," EE-Evaluation Engineering, June 2012, p 44.
(2.) Real-Time Spectrum Analyzer (RTSA) PXA X-Series Signal Analyzer N9030AK-RTI & N9030AK-RT2, Agilent Technologies, Technical Overview 5991-1748EN, February 2013, p. 3.
(3.) Spectrum Analyzer Basics, Agilent Technologies, Application Note 150, 5952-0292, August 2006.
(4.) Ramian, F., Implementation of Real-Time Spectrum Analysis, Rohde & Schwarz, White Paper, p. 3.
(5.) "The Growing Importance of Switched-Filter Preselection, Tektronix, January 2010.
by Tom Lecklider, Senior Technical Editor
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||SPECIAL REPORT - SPECTRUM ANALYZERS|
|Author:||Lecklider, Tom, Sr.|
|Date:||Jun 1, 2013|
|Previous Article:||The beat goes on: modeling the human heart.|
|Next Article:||Sensors get smarter.|