Spectrum analyzers are changing to accommodate the requirements of modern communications systems. RF envelope amplitude scaled to read Vrms, long the standard format of a spectrum analyzer’s measurement results, is being augmented by direct power measurement. Vector signal analysis is adding in-phase/quadrature (IQ) demodulation to the usual AM and FM capabilities.
New types of spectrum analyzers also have been developed. Some provide real-time signal acquisition, capturing all relevant information about the signal as it varies in time. Others address terahertz dense wavelength division multiplexed (DWDM) systems that modulate multiple wavelengths of light in a single fiber-optic cable.
Power
In all these cases, power measurements are key indications of a communications system’s operation. According to Brian Daly, product marketing engineer at Hewlett-Packard, “The most important measurements in modern communications systems are channel power, adjacent channel power (ACP), and modulation quality. Many spectrum analyzers will make these measurements with built-in functions which optimize the analyzer settings for the particular measurement.”
For example, code division multiple access (CDMA) systems use a dynamic control mechanism that keeps power output at a minimum. One purpose of operating with low power is to increase battery life for mobile transmitters. The main reason, however, is to reduce interference caused by having many users within the same frequency band. Second-generation (2G) CDMA can have up to 64 users transmitting simultaneously within a 1.25-MHz wide frequency band.
Rho, defined as the ratio of correlated power to total power, is the measure of modulation quality. As rho decreases from 1.0, it means that more transmitted power is appearing as interference. Unfortunately, as interference increases in a channel, the other CDMA transmitters operating within that band will increase their output power to maintain an acceptable bit error rate (BER), adding to the background noise.
For average and peak measurements, power meters are fine. But to determine how much power is transmitted in which channels, you need a spectrum analyzer.
ACP or ACPR (R for ratio) compares the power within a CDMA or wideband CDMA (W-CDMA) channel to that in adjacent and alternate channels. The adjacent channels are on either side and immediately next to the occupied channel.
For W-CDMA, adjacent-channel center frequencies are offset 5 MHz from the center of the active channel—5 MHz being the channel width and spacing. The first alternate channels are the next channels immediately after the adjacent channels, offset 10 MHz from the occupied channel.
Specifications, especially for W-CDMA transmission, are very stringent. ACPR for a base station must be less than 55 dB in the adjacent channel and less than 70 dB in the first alternate channel. Figure 1 shows a similar ACPR measurement on an IS-95 2G CDMA system. Because the spectrum analyzer’s own noise will add to the measurements, dynamic range should include a margin of several decibels to maintain accuracy, especially if linear instead of logarithmic amplification is used. See the sidebar for additional ACP considerations.
“A spectrum analyzer’s phase noise, third-order intermodulation distortion, and noise floor will determine the maximum dynamic range for an ACP measurement,” commented Mr. Daly of HP. “The accuracy of an ACP measurement will be determined primarily by the spectrum analyzer’s scale fidelity. If the analyzer has an overall absolute amplitude accuracy specification, make sure it applies for settings appropriate for a channel power measurement.”
Power also can be viewed on a per-user basis in a W-CDMA system. Code-domain power (CDP) measurements determine the power associated with each user. ACP tests measure the aggregate power within a frequency band simultaneously occupied by many users. CDP measures power associated with each active code and is primarily used for base-station transmitter testing.
CDMA signals are pseudorandom in nature, so additive white gaussian noise (AWGN) often is used as the input signal during component and subassembly development testing. The statistical power distribution of AWGN and the actual signal are similar, so developing an understanding of noise power measurement difficulties highlights problems that also affect CDMA power measurements.
Traditional spectrum analyzers detect the level of the RF input signal envelope and scale this peak voltage to represent rms, assuming a sine wave. Unfortunately, this simple approach doesn’t work well for noise-like signals.
An AWGN signal band-limited by a spectrum analyzer’s intermediate frequency (IF) filter has a Rayleigh probability density function (PDF) with a mean value of 1.253 × s (standard deviation). The average of the square of the Rayleigh PDF is 1.05 dB higher than the square of the average of the envelope detector voltage. In other words, power measurements of noise that are derived from the spectrum analyzer’s envelope detector output are 1.05 dB too low.
Logarithmic amplifiers provide less gain for larger signals than for smaller ones. Averaging the output of a logarithmic amplifier creates a -2.51-dB error in noise power measurements. This error includes the -1.05-dB effect of the envelope detector.1
Finally, the definition of noise-power bandwidth involves ideal, straight-sided rectangular filters. The IF filters used in Hewlett-Packard analyzers have a noise power bandwidth between 1.05 and 1.13 times the ideal. This means that a higher than actual power level will be measured between 0.21 dB and 0.53 dB.
About 2.0 dB needs to be added to the displayed noise power measurement to compensate for all three error sources. A similar factor will apply to other manufacturer’s spectrum analyzers as well. Because the errors contributed by these sources are predictable, they can be corrected automatically in an analyzer.2
However, the size of the required correction varies for noise-like CDMA signals depending upon how many and which Walsh codes are used. The amplitude distributions of the signals resemble AGWN, but they are different so the errors will be different. For noise power measurement, the use of the spectrum analyzer’s envelope detector following logarithmic amplification has been assumed.
Another approach to power measurement is to use a true-rms detector. Bob Buxton, product marketing manager for the Tektronix Wireless Communications Test product line, said, “The use of trace averaging can lead to problems caused by the statistical nature of a digitally modulated signal. Instead, use a signal detection or averaging method that is independent of signal type.”
This solution has been implemented in the Rohde & Schwarz Model FSE and FSIQ spectrum analyzers. The output from a high dynamic range linear detector is sampled and digitized by a high-speed, high-resolution ADC. A dedicated ASIC incorporates each sample in an rms calculation. All samples taken between actual display points are averaged to determine the next power value to be presented on the screen.
Time
You also can make time-domain and time-related measurements with a spectrum analyzer. For example, test instruments must be synchronized correctly to measure the power bursts during the 577-µs time slots of the global system for mobile communications (GSMs). Many analyzers now offer delayed and gated sweep modes that can be synchronized to an external trigger.
If you do not require the highest speed sweeps or the very fine time resolution of a real oscilloscope, then you may benefit from a spectrum analyzer’s zero-span scope style of display. Zero-span operation fixes the local oscillator (LO) frequency and displays the envelope of the input voltage as a function of time.
In this mode, the pulse response of the instrument is set by the video filter and the narrowest filter among the several IF stages. As a result, a wide resolution bandwidth (RBW) must be chosen to accommodate harmonics associated with fast pulse edges. Otherwise, time-domain pulse response will be distorted. The video filter can smooth out a noisy display, but it cannot correct a distorted spectrum caused by a small RBW.
Miscellaneous
In addition to gated sweeps, tracking generators, counters, and in-phase/quadrature signal demodulation are available for many models. Relative frequency measurements can be quite accurate in a spectrum analyzer, but absolute accuracy may be lacking. If you need to measure absolute frequency, a built-in frequency counter with an accuracy of 1 part in 106 or better may be the answer.
Tracking generators provide a swept output frequency that is phase-locked to the LO. Characterization of amplifiers and filters is an ideal application for a tracking generator. The spectrum analyzer is tuned precisely to the frequency input of the device being tested.
Alternatively, some tracking generators incorporate a frequency offset that is useful for IF-to-RF conversion measurements. Of course, the degree to which you can view spurious responses or harmonics depends upon the chosen RBW. See sidebar.
Signal demodulation in a typical spectrum analyzer is limited to AM and FM signals. Conventional analyzers are scalar instruments because they only measure signal amplitude. More complex modulation schemes such as binary phase shift keying (BPSK) and 16-state quadrature-amplitude modulation (16QAM) require information about both phase and amplitude.
For some analyzers, vector signal analysis options are available. With this addition, the phase of the signal can be determined. Signal phase and magnitude then can be translated into an IQ representation which can be demodulated. Mr. Buxton of Tektronix commented that built-in modulation analysis allows testing of modulation quality in terms of phase error, error vector magnitude (EVM), and rho. These quantities can be tested to the relevant standards requirements.
Real-time Spectrum Analysis
Traditional spectrum analyzers detect the output of the IF stage following a mixer fed with the input signal and a swept LO. The display on these instruments is locked to the LO sweep so that the detected mixer products are displayed synchronously.
The LO can be tuned to only one frequency at a time. This basic observation is not important when dealing with CW or simply modulated, stable signals. It has become important because the characteristics of digitally modulated signals vary at a high rate compared to an analyzer’s sweep time. Transients can be missed.
Real-time spectrum analyzers avoid this problem. The LO is set to a fixed value, and a wide bandpass filter is placed ahead of the sample detector and its ADC. All the signal activity within the filter bandwidth is converted to data samples that can be used in time-domain analysis or as the input to an FFT.
In an instrument with continuous acquisition, spectrographs presenting frequency vs amplitude vs time can be developed as shown in Figure 2. This technique relies on high-speed, high-resolution digitizing.
“Real-time capture is required to analyze the effects of changing the signal level driving transmitter power amplifiers,” explained Mr. Buxton of Tektronix. “Large changes in power can lead to a temporary increase in phase error or EVM.” Because these events are transient in nature, a conventional analyzer cannot capture them.
Optical Spectrum Analysis
Speed certainly is the overriding consideration when dealing with dense wavelength division multiplexing (DWDM) fiber-optic signals. Thanks to erbium-doped fiber-optic amplifiers (EDAs), the distance over which fibers can be used is very large. EDAs can provide from 20 to 30 dB of gain, but only within the 1,530- to 1,565-nm wavelength band.
Unfortunately, an additional signal also is present in the amplifier’s output known as amplifier spurious emission (ASE). The effect of ASE is to raise the noise floor within the 35-nm passband. As a result, individual carriers within the band may be riding on pedestals caused by previous amplification. This effect complicates accurate SNR measurement.
DWDM literally multiplies the data-handling capacity of fibers by 8, 16, or more within the 35-nm passband. Slightly different wavelengths of light are separately modulated with independent information and multiplexed onto a single fiber.
The telecommunications unit of the International Telecommunications Union (ITU-T) proposes 100-GHz channel spacing (0.8 nm) in G.692 “Optical Interfaces for Multichannel Systems with Optical Amplifiers.” This spacing limits the number of channels within the EDA frequency range to about 40. The trend is toward even more channels with 50-GHz spacing.
Parameters to be measured include individual carrier power, channel wavelength and channel spacing, overall power, and crosstalk. Measuring overall power is not a problem, but distinguishing one signal from another requires a frequency-dependent filter.
References
1. “Spectrum Analyzer Measurements and Noise,” Application Note 1303, Hewlett-Packard, 1998, p. 8.
2. “Spectrum Analysis,” Application Note 150, Hewlett-Packard, 1989, pps. 31-33.
3. Ibid. p. 14.
Additional Resources
1. “Understanding Dense WDM,” Wandel & Goltermann, 1998.
2. Wolf, J. and Tiepermann, K., “Measuring ACPR of W-CDMA Signals with a Spectrum Analyzer,” Rohde & Schwarz, Proceedings of Wireless Symposium, Feb. 22-26, 1999.
3. “ACP Measurements on Amplifiers Designed for Digital Cellular and PCS Systems,” Tektronix Application Note, 1998.
Sidebar
Relationships Among Spectrum-Analyzer Specifications
The bandwidth of the narrowest IF filter determines the resolution of a conventional spectrum analyzer. A small RBW integrates less noise and improves dynamic range. A narrow RBW also increases sweep time according to the relationship
ST (sweep time) = k (Span / RBW2) (1)
where: Span = total bandwidth of the displayed spectrum
k = a constant that ranges from about 2 to 15 depending upon the shape of the IF filter proportional to the product of rise time and bandwidth.3
Equation 1 shows that ST is inversely proportional to the square of RBW. So, although many spectrum analyzers boast very fast 5-ms sweep times, this time is associated with a wide RBW. These instruments also provide RBWs as low as 1 to 10 Hz and corresponding STs from 1,000 to 2,500 s.
Shape Factor
The usual logarithmic voltage vs linear frequency display of a spectrum analyzer results in a nearly straight skirt between an IF filter’s -3- and -60-dB points. Shape factor (SF) is the ratio of the filter bandwidth at these two attenuation levels. See Figure 3.
IF filters provide sufficiently sharp roll-off so that signals close to each other can be resolved, but without excessive phase distortion. Filters with a Gaussian characteristic achieve linear phase response, and those used in several HP analyzers have an SF of about 11:1. For the Rohde & Schwarz FSEA/B analyzers, the shape factor is specified as less than 12 or 15, depending upon model. A 1-kHz filter is, by definition, 1 kHz wide between the -3-dB points. It will be about 11 kHz wide between the -60-dB points.
For RBWs below 1 kHz down to 1 Hz, digital filters are used. These filters can have an SF as low as 6:1 or less. They also allow much faster STs.
Example 1
To resolve two signals that differ by 1 kHz in frequency and 20 dB in amplitude, assume that the larger signal is centered within the RBW. A 1-kHz filter with an SF of 11:1 has an offset of 500 Hz at its -3-dB point and 5,500 Hz at -60 dB. A 1-kHz offset corresponds to -8.7 dB. The filter is too wide, and the smaller signal will be obscured beneath its skirt.
A 300-Hz filter with a 6:1 SF has only a 900-Hz offset at -60 dB, so the smaller signal will be resolved very easily. Generally, the effect of choosing a smaller RBW is more gradual and linear because only the bandwidth changes, not both the bandwidth and the SF.
Example 2
W-CDMA modulation occupies a 4.096-MHz bandwidth within the 5-MHz channel. Adjacent channel power is specified to be at least 55 dB down compared to the reference channel power. Equation 2 relates SF, RBW, and approximate attenuation for a given offset beyond the filter’s -3-dB point.
attenuation (dB) = 114 × offset / [RBW (SF-1)] + 3 (2)
The difference between the modulation bandwidth and the channel width is 904 kHz. The offset to be used in Equation 2 is half this or 452 kHz.
Assuming that SF = 11:1, a 100-kHz RBW will provide about 54-dB attenuation—not quite enough. Because most spectrum analyzers provide filters in a 1-, 3-, 10-RBW sequence, 30 kHz is the next choice. From Equation 2, the approximate attenuation is over 170 dB.
The RBW filters used in the Rohde & Schwarz FSIQ7 spectrum analyzer have a further SF specification of about 16:1 at -80 dB. Using this information, the attenuation of a 30-kHz RBW filter at a 452-kHz offset is 150 dB. Neither 150 dB nor 170 dB is likely to be achieved by the actual physical filter, but a 30-kHz RBW should be a good choice for this application.
“The wider channel bandwidths of W-CDMA stretch the limits of spectrum analyzers for ACP measurement,” said Mr. Buxton of Tektronix. “Having to integrate in-band power across a 5-MHz bandwidth may overdrive the analyzer. This will create intermodulation products within the spectrum analyzer which will fall into the adjacent channels. Additionally, integrating over 5 MHz in the adjacent channels increases the effect of the noise floor and further limits the dynamic range available for ACP measurement.”
Spectrum Analyzer Products
Optical Spectrum Analyzer
The PC-based Model WG OSA-155 Optical Spectrum Analyzer measures wavelength, power, and optical signal-to-noise ratio on individual carriers in the 1,500- to 1,620-nm wavelength range. Up to 32 channels, each with +15 dBm power or up to 256 lower power channels, can be measured simultaneously. An extended 1,450- to 1,650-nm range includes optical supervisory channels. Power measurement accuracy is ±0.5 dB, sweeps take 2 s, and wavelength/frequency measurement accuracy is ±0.04 nm or ±5 GHz. A 9.4″ TFT color LCD with touch-screen capability controls the battery-powered, 19.6-lb instrument and displays results. From $34,000 to $40,000. Wavetek Wandel & Goltermann, (919) 941-5730.
Microwave Analyzer
The Model MS2668C Spectrum Analyzer has a frequency range of 9 kHz through 40 GHz and is targeted at microwave point-to-point and local exchange carrier message distribution system (LMDS) radio system test. The noise level in a 10-Hz bandwidth is -135 dBm at 1 GHz and -121 dBm at 40 GHz. The harmonic distortion is -90 dBc at 20 GHz. Measurement capabilities include transmit, adjacent channel, and noise power; spurious emissions; two-tone intermodulation distortion; harmonics; and carrier-to-noise ratio. External mixers to 110 GHz and spectrum monitoring software are optional. $40,000. Anritsu, (800) 267-4878.
Fiber-Optic Test System
The DWDM Fiber-Optic Component Test System comprises several 700 Series modules. Optical resolution, data acquisition rate, wavelength selectivity, and wavelength detection characteristics are determined by the modules. The system can test parts including fiber Bragg gratings, isolators, add/drop multiplexers, circulators, and fiber. It performs insertion loss vs wavelength, crosstalk, waveshape, bandwidth, polarization dependence, and return loss tests. The company’s FiberWORKS™ software automates the measurement process. Contact company for price. Rifocs.
Real-Time Analyzer
The Model 3086 Real-Time Spectrum Analyzer continuously samples signals within a user-selectable frequency band up to 30 MHz wide. This is wide enough to capture the highest proposed W-CDMA chip rate of 20.48 Mchips/s. Features include IQ input ports, a DC-to-3-GHz frequency range, and frequency mask triggering. The 12.1″, 256 color, 1,024 × 768 pixel LCD provides simultaneous viewing of correlated real-time, vector, error vector, and conventional spectrum displays. A 1,000:1 zoom mode and both waterfall and spectrogram display formats are included. $61,950. Tektronix, (800) 426-2200, code 1132.
Spectrum Analyzer Family
A family of five HP ESA-E general-purpose portable spectrum analyzers covers frequencies from 1.5 GHz (HP E4401B) to 26.5 GHz (HP E4407B). The full-span RF sweep time is 5 ms, and up to 28 measurements can be made per second. Accuracies are ±1% of span, ±101-Hz absolute frequency, and ±1-dB amplitude. An optional 10-Hz digital resolution bandwidth (RBW) filter supports a -151 dBm noise floor. Other options include GPIB, RS-232, and parallel printer interfaces; PC trace transfer software; an integral preamplifier; demodulation; a high-stability reference; a built-in tracking generator; and narrow RBW filters. HP E4401B: $11,950; HP E4407B: $28,550. Hewlett-Packard, (800) 452-4844, ext. 6340.
Quiet Signal Analyzer
The Rohde & Schwarz FSIQ Series of signal analyzers includes fully integrated spectrum and modulation analysis for testing to wireless communications standards such as GSM, IS-136, CDMA, and W-CDMA. Three models have high-frequency bandwidths of 3.5 GHz, 7 GHz, and 26.5 GHz, with a low frequency -3 dB point of 20 Hz. Phase noise of -152 dBc/Hz at 5-MHz offset and a -155-dBm thermal noise-floor result in a 75-MHz dynamic range for measuring 4.096 Mchips/s ACPR. Starts at $68,000. Tektronix, (800) 426-2200, code 1098.
Copyright 1999 Nelson Publishing Inc.
June 1999