Spectrum analyzers are among the most versatile of RF/microwave measurement tools, with signal power among the most common measurement made with the instruments. Traditionally, the combination of a power meter and power sensor has been the measurement tool of choice for its well-characterized traceability path back to reference standards at national standards laboratories. But modern spectrum analyzers have made dramatic improvements in amplitude accuracy, with levels approaching (but not exceeding) the power meter and sensor. Understanding the error terms associated with a spectrum analyzer’s relative and absolute amplitude accuracies can help an engineer interpret the analyzer’s specifications when selecting a measurement tool and balancing price/performance tradeoffs.

The earliest spectrum analyzers were fully analog, and even an operator’s skill in reading and recording the measurement results impacted amplitude accuracy. The first digitaldisplay spectrum analyzers were introduced in the 1970s, with the HP 8566A and HP 8568A models from Hewlett-Packard Co. (now Agilent Technologies) among the most accurate and popular instruments of that time. They featured digital displays and digital marker readouts for improved accuracy.

The signal at the end of the analog chain was the filtered (and unipolar) result of detecting the intermediatefrequency (IF) signal level and therefore directly proportional to the amplitude of the signal in the selected resolution bandwidth (Fig. 1). It was called the “video” signal because it had driven the Y-axis video deflection plates in previous all-analog spectrum analyzers. As the block diagram shows, the analog signal was processed by an analog-to-digital converter (ADC) for storage and display without using cathode-raytube (CRT) persistence, allowing for communication with remote users. Because of this architecture, and the fact that a user no longer needed to interpret the results on the display, improved accuracy was possible compared to the all-analog spectrumanalyzer architecture.

Unfortunately, the improved architecture still suffered from gain (amplitude) drift in the IF circuitry. The frequency stability was relatively low as well, with some drift in the center frequency and filter bandwidths. The logarithmic amplifier that allowed decibel-scaled displays suffered considerable errors as well.

When the HP 8560A spectrum analyzer was introduced in 1989, it marked the first general-purpose swept-frequency spectrum analyzer where the ADC moved forward in the signal processing chain, to digitize the IF signal rather than the detected magnitude (the video signal). In this instrument, filtering, detection, and logarithmic conversion were performed digitally, but only in the narrowest resolution bandwidths (1 Hz through 300 Hz), and only with Fast- Fourier-Transform (FFT) processing.

Advances in ADC and signal-processing technology during the 1990s eventually brought an all-digital IF structure to some swept spectrum analyzers, beginning with those at the high end of price and performance. For example, the Agilent PSA series spectrum analyzers, introduced in late 2000, included digital processing for all resolution bandwidths. Digital signal processing (DSP) provided 160 choices for resolution bandwidth (1 Hz through 8 MHz) in swept and FFT analysis modes. These digital advances have recently been provided more economically in analyzers such as the Agilent X-series (MXA in 2006 and EXA in 2007).

Although the consistency of alldigital processing has improved IF specifications an order of magnitude compared to the previous generation of analog instruments, the RF signal path in a spectrum analyzer still has gains that drift with time and temperature. Fortunately, the effects can be minimized, and the amplitude accuracy further improved with an internal 50-MHz amplitude reference.

The primary driver of improvement in spectrum-analyzer accuracy has been the all-digital IF. But another driver is background alignments. The RF and especially IF analog-signalprocessing elements can be characterized with reference signals regularly, such as during retrace (the time the LO resettles after a sweep in preparation for the next sweep).

There are two kinds of amplitude accuracy: absolute and relative. The difference between the two seems confusing, given their names, because absolute accuracy is also relative. Absolute accuracy is the accuracy relative to a standard kept by a national standards laboratory, such as the National Institute of Standards and Technology (NIST), formerly the National Bureau of Standards (NBS). Relative accuracy is the accuracy of the ratio of two measurements, irrespective of individual accuracy traceability to NIST.

Absolute accuracy is useful when measuring devices against established requirements. For example, an RF power amplifier might be specified for a set of error requirements when delivering a certain amount of absolute power. Relative accuracy is often an excellent substitute for absolute accuracy. For example, in a test system with cables and switches, the absolute accuracy at a system port must be characterized due to the losses in the system, making the absolute accuracy of the spectrum analyzer itself moot, and the stability of the analyzer response the essential characteristic.

The accuracy of a spectrum analyzer is at its best when the signal being measured is at the same level and frequency as the analyzer’s built-in amplitude reference oscillator, which is often called the calibrator. The accuracy can be optimized by performing both sets of measurements under the same reference conditions—i.e., the same instrument settings [such as resolution bandwidth (RBW), video bandwidth (VBW), and sweep time].

For example, when new measurements are made with a different signal level, the change in response relative to the reference condition is called the “scale fidelity” error. Other settings that can change are the RBW (with errors referred to as “RBW switching uncertainty”), input attenuation (attenuator switching uncertainty), reference level (reference-level accuracy or IF gain uncertainty), and display scale (display-scale switching uncertainty).

Reference-level and display-scale uncertainties can be rendered zero with an all-digital IF. Another feature of some all-digital IF spectrum analyzer designs is that the scale fidelity can be made dependent on the level at the input mixer (input power minus attenuation) and independent of the reference level. This allows accuracy to be independent of display conveniences and user preferences.

One challenge in achieving excellent absolute amplitude accuracy lies in tracing the results back to a reference standard, such as that maintained by NIST. That traceability is closely tied to the capabilities of RF/ microwave power meters and power sensors. Understanding the traceability of an RF power meter and its sensors will help clarify the limitations of spectrum analyzers when striving for absolute amplitude accuracy.

Production environments may typically contain a number of RF/ microwave power meters and sensors that are used for power measurements and are regularly calibrated. In such an environment, calibration consists of comparing the results of the power sensor to another measuring device of known accuracy. This other device might be in the production facility’s metrology laboratory or calibration lab, or calibrations might be performed by another organization. The power sensor is calibrated against a reference standard or “transfer standard,” which itself is calibrated against a NIST standard, which is a primary standard. Each step away from the NIST standard usually involves less time-consuming (and less expensive) calibration practices; the tradeoff of each extra step is adding uncertainty. In contrast, regularly calibrating a spectrum analyzer, with its higher capital value, weight, and size, is impractical. As a result, spectrum analyzers are typically calibrated against power meters and sensors, although this adds one level of removal to the uncertainty of the traceability.

Continue to page 2