Modern spectrum analyzers have improved dramatically in recent years in terms of low-frequency analog precision and a tremendous boost from digital-signal-processing (DSP) components, especially in the front-end intermediate-frequency (IF) filters. While spectrum analyzers can provide even better accuracy at RF, their accuracy at microwave frequencies has not generally been improving. Because of the demanding applications for modern spectrum analyzers, including analysis of signals with advanced wideband digital modulation formats, higher accuracy is difficult to realize with even the best spectrum-analyzer designs. Fortunately, a combination of careful hardware design and clever measurement algorithms has been incorporated in the new MXA signal-analyzer platform from Agilent Technologies (www.agilent.com) to help users maintain the best accuracy without requiring a specific kind of input signal or external test signal.
Microwave spectrum analyzers employ tunable preselector filters to improve performance by removing unwanted mixer images and responses to local-oscillator (LO) harmonics. Unfortunately, these preselectors have instabilities and must be retuned frequently, and proper preselector tuning traditionally has required a signal of nearCW statistical distribution at the frequency of interest. In the new MXA signal analyzers, an integral noise source is used as a tuning signal for the preselector filter, helping to ensure filter accuracy as part of an automatic routine in the instruments.
Modern spectrum analyzers operating to 26.5 GHz have a "low band" and a "high band" signal path, as shown in the simplified block diagram (Fig. 1). The low band typically operates to 3 GHz or higher. In the low band, the signal is frequency upconverted to a high IF near 4 GHz or more, then downconverted to a lower IF near 300 MHz. This double-conversion scheme can dramatically reduce mixer image responses.
"High-band" frequency ranges cannot practically be built with the same block diagram as low-band ranges because the first IF amplifier would have to work at a frequency where the amplifier noise and distortion would inevitably be unsatisfactorily high for operators. The alternative block diagram involves a single-conversion step to the IF shown in Fig. 1. In this block diagram, image responses in the first mixer are spaced by only twice the IF, or about 600 MHz. Such images are unacceptable in a general-purpose spectrum analyzer. Thus, the tunable preselector filter (a bandpass filter) is used to remove the images.
To achieve the rejection and tuning bandwidth required at microwave frequencies, the preselector filter is based on yttrium-iron-garnet (YIG) technology. The action of YIG spheres controlled within a precise magnetic field creates the filter passband resonances needed to remove unwanted images and responses from the spectrum analyzer's signal path.
YIG preselectors usually have a passband width of about 40 to 80 MHz that can be tuned across a wide range of microwave frequencies. When used at frequencies up to 26.5 GHz, the required quality factors (Qs) of the resonators are very high, resulting in sharp cutoff frequencies but also accompanied by amplitude and frequency instabilities.
Post-tuning drift is one form of instability in a YIG-tuned bandpass filter. The magnet used to tune the resonant frequency of the YIG spheres heats up or cools down as the selected frequency is changed. The temperature change of the magnet affects the dimensions of the magnet and the magnetic field strength, and thus the frequency of the filter tuning. Mechanical aging of the magnet/sphere structure works in the same way to cause further instabilities.
Also, the relationship between the tuning current and the filter center frequency is not perfectly modeled by any simple algebraic function. Therefore, even without tuning instabilities, there are tuning errors. The result is that frequency tuning errors lead to amplitude errors (Fig. 2).
Figure 2a shows a typical YIG filter response. The x-axis represents frequency, but because the frequency of a YIG filter is nearly proportional to tuning current, the x-axis can also be thought of as YIG filter tuning current. In this example, small tuning current errors map to amplitude errors proportional to the slope of the passband shape at the operating point. The design operating point is the midpoint between the 4-dB response points, because this design is highly robust regarding tuning errors.
A YIG filter can be adjusted by means of measurements with a modern spectrum analyzer. A user can adjust operating current directly, or execute a "preselector center" operation. Because the spectrum-analyzer amplitude response was factory calibrated under the condition of centering the preselector tuning, centering is the preferred operation. Note that peaking the preselector would result in poorer amplitude accuracy.
Figure 2b shows the importance of YIG filter centering. Point A shows the plot used for factory calibration of the YIG filter's frequency response. This point is on the response curve of a new analyzer at room temperature. Its horizontal location is at the midpoint between the 4 dB (relative to peak) response frequencies.
Point B is on a congruent curve, displaced vertically to indicate the expected total system response changes when the ambient temperature is changed. The influences of post-tuning drift and aging, in addition to ambient temperature changes, could result in the curve with point F on it. In this case, the amplitude error can be quite large. This error is indicated as length E, which is the difference between the response points B and F.
To improve amplitude accuracy, the analyzer's YIG preselector filter should be re-centered. Dimension C shows the location of the –4 dB points relative to the peak of the new tuning curve. Re-centering results in the response as shown with point G. The new error is now represented by the distance D, which is much smaller than the original error distance E.
The "preselector centering" algorithm measures the filter response in order to optimize the tuning. It assumes a CW-like input signal, and observes the relative response as the tuning current is swept. It searches this relative response curve for the –4 dB points, and centers the tuning between these points.
One of the assumptions in this preselector centering algorithm is that the input signal to the filter exhibits excellent amplitude stability during the duration of the sweeping portion of the centering operation. This stability must be well under 1 dB so that amplitude variations are not mistakenly treated as filter passband shape variations. The signal must have minimal frequency modulations as well, for similar reasons. While modulations that are well under 1 MHz in width are acceptable, wideband digital modulations can cause tuning errors. Finally, the signal must have a good signal-to-noise ratio (SNR).
These constraints on the input signal can be a problem in practical measurement applications. Centering cannot occur when measuring low-level harmonics, for example. Centering cannot be used to measure spot noise density. It can't work with increasingly common digital communications signals such as orthogonal frequency-division multiplex (OFDM), wideband code-division-multiple-access (WCDMA), or time-division-multiple-access (TDMA) formats. Similarly, it can't work with most radar signals.
Thus, while spectrum-analyzer specifications for amplitude accuracy only apply following a preselector center operation, such operations are often not feasible. This leads to measurements with significantly reduced accuracy.
If the analyzer integrated a full-range CW signal generator dedicated for preselector centering, a user would never need to provide a suitable signal. Unfortunately, such a capability isn't economically feasible; however, by employing a broadband noise generator and a new centering algorithm, centering without strict signal requirements can be accomplished economically. The block diagram for this new approach is shown in Fig. 3.
At first glance, it might appear that a noise source wouldn't help in tuning a YIG preselector filter. After all, the amount of noise coming out of the filter is nearly independent of the tuning current. But the frequency distribution of that noise does change. The amount of noise that makes it to the IF will vary with the tuning current. By analyzing the curve of Fig. 4, a preselector centering algorithm can center the passband using just the noise source.
Centering can be invoked at every frequency of interest to the user. But the built-in noise source also allows the factory calibration for the overall tuning curve to be implemented in the analyzer and be rerun at will, instead of only on a return to a fully equipped facility. Thus, the effects of aging and (to a much lower level of significance) ambient temperature changes can be removed by running the "characterize preselector" operation occasionally. With a well-characterized preselector, centering is rarely even needed. Not needing to re-center is even more convenient than a centering operation that doesn't require CW-like signals.
In summary, with the addition of a noise source and a new tuning algorithm, the Agilent MXA signal analyzer can meet its microwave amplitude accuracy specifications in the measurement of all signal types, not just high-amplitude, nearly-CW signals. For example in measuring a 40 MHz wide digitally modulated signal at a frequency of 6 GHz, the measurement accuracy without effective preselector centering is not specified and might be worse than –10 dB, while proper centering improves the figure to the warranted ±1.5 dB.