- Buyers Guide
Network Analyzer Measurements
RF and microwave production testing and ways to reduce its contribution to manufacturing time while maintaining accuracy
Network Analyzer Measurements
Hewlett-Packard Co., Microwave Instruments Division
Santa Rosa, CA
Increased manufacturing throughput is one of the keys to achieving success in high volume, low cost markets such as wireless communications. As part of the manufacturing process, component and subsystem testing can either increase or decrease throughput, depending on the efficiency with which it is performed. This article concentrates on RF and microwave production testing and ways to reduce its contribution to manufacturing time without sacrificing accuracy.
An overall measurement process must be evaluated before ways to increase measurement throughput become apparent. For example, when a vector network analyzer is used to manually tune filters, the filters are connected to the test equipment, applicable instrument settings are recalled, the filters are tuned and filter characteristics are viewed on the analyzer's screen for the results of the tuning. Several different instrument settings may be needed depending on the specified filter characteristics.
Even when automated parts handlers are used, the device under test (DUT) must be installed either manually or automatically into a test fixture, a network analyzer instrument state recalled and measurements performed. In an automated system with an external computer, data must be transferred to the computer for calculations or storage. If instrument setups are stored in an external computer, time is required to transfer these setups and control commands from the computer to the analyzer.
When evaluating a measurement process, the relative importance and value of improving each measurement step must be considered. Enhancing one aspect, such as sweep speed, may not always provide the best improvement in throughput. For manual measurements such as filter tuning, faster sweep speeds are necessary. However, beyond an update rate that the user perceives to be continuous (approximately 30 updates per second), faster sweep speeds will not increase throughput when filters are tuned manually. (On the other hand, faster sweep speeds will impact automated measurements.) Some factors (such as data transfer rates) are not relevant to manual tuning, whereas the time required to set up or recall instrument states from memory is relevant to both manual and automatic measurements.
Actual test time is often a minor portion of the total time required for a measurement process. When testing multiport components, for example, an operator may need 1.5 minutes to install a DUT in a test fixture for a measurement that takes a vector network analyzer only 10 seconds to perform. Cutting the analyzer's test time in half saves five seconds - a five percent reduction in the total test time (to 95 seconds from the original 100 seconds). Cutting the interconnection time by one-third reduces test time to 70 seconds from its original 100 seconds (an improvement of 30 percent).
A measurement process often involves actions (such as calibration) that are not part of every test but can affect the overall throughput. Calibration time may range from a few minutes for a single one-port calibration to several hours for a series of two-port calibrations of a high-dynamic-range multiport DUT. Improving measurement throughput requires careful consideration of the trade-offs involved in optimizing different elements of a test process, such as sweep speed, amount of averaging and type of calibration used, measurement bandwidth and required measurement accuracy.
Sweep speed for a network analyzer generally is defined as the time required to tune the test signal source across a defined measurement range of frequency, power or a combination of the two. Sweep speed is often specified in terms of time per data point, although this ratio is not a true estimate because it does not include all elements of the measurement. A more representative measure of the time required for a sweep is called cycle time. Cycle time includes sweep setup time, band-switching time, data acquisition time, retrace time (the time to move from one end of a sweep to the start of the next sweep), data calculation and formatting times, and analyzer display update rate. Error correction time must be added as well since with a two-port calibration an analyzer will sweep in both the forward and reverse signal directions to update a data trace.
Published specifications for sweep speed or cycle time may not be representative of typical tests since data sheet specifications often list times for best-case conditions, such as using wide IF bandwidths and taking a single sweep within one band. An analyzer's total frequency span is usually broken into many bands, and switching between bands can add to the total cycle time. Actual sweep speed is dependent upon an analyzer's setup parameters, including the number of data points, frequency range, and required dynamic range and accuracy.
Cycle Time Optimization
Sweep speed/cycle time can be optimized by using the widest possible IF bandwidth, using only the dynamic range required for a given test, minimizing the number of averages and data points, and using the fastest possible calibration method for a given level of accuracy. The widest possible IF bandwidth should be used without compromising dynamic range. Wider IF bandwidths result in faster sweep speeds but with more trace noise and a higher noise floor (decreased dynamic range). Typically, a 10-fold reduction in IF bandwidth produces a 10 dB reduction in the trace noise, as shown in Figure 1 for an HP 8753E vector network analyzer. Averaging can reduce noise and improve dynamic range, but too many averages will slow down a measurement. Therefore, the minimum number of averages should always be used. In addition, dynamic range can be extended by using the maximum source power that does not overload the analyzer's front end or DUT.
To optimize sweep speed, band switching should be minimized. This goal can be achieved by using the smallest possible frequency range needed to test the DUT and by carefully selecting the span to avoid the analyzer's band-switching points. In some cases, using the minimum number of test points can result in less time per sweep, although network analyzer sources have a practical limit on speed. Below this limit, a reduction in the number of swept points will not produce any practical reduction in sweep time.
Customizing the Sweep
Sweep speed can also be optimized by using a user-defined list of frequencies instead of a full series of data points to sweep across a range. Points are used where data are needed (such as in a filter's passband and rejection bands) rather than on the filter skirts. Different IF bandwidths and source-power levels can be chosen for each segment depending upon the test requirements. For example, higher source-power levels and smaller IF bandwidths might be used for better dynamic range in a filter's rejection bands.
When measuring a device such as a bandpass filter/amplifier assembly that has a broad dynamic range, the swept-list function can provide increased throughput compared to the standard (linear) sweep mode, as shown in Figure 2 . In both cases, 201 data points were measured. In the linear sweep mode, a fixed IF bandwidth of 300 Hz and a power level of -10 dBm were used to improve dynamic range in the rejection band without overdriving the receiver in the filter's passband. In swept-list mode, variable IF bandwidths were used with variable source power levels. The linear sweep required 676 ms to span 525 to 1275 MHz. Over the same frequency range, the swept-list sweep required only 349 ms. This savings resulted from optimizing the spread of data points and omitting data points over spans where information was not needed. In addition, the linear sweep did not correctly measure the filter's rejection, but the swept measurement did.
The choice of calibration can impact sweep speed. For most analyzers, a simple response calibration provides measurement throughput that is approximately the same as that for uncorrected measurements. However, twice as much time may be needed for measurements based on a full two-port calibration since sweeps in the forward and reverse directions are required to update all four S parameters for error correction (even when only one S parameter is being displayed).
Reverse sweeps should be minimized to optimize measurement speed with full two-port calibration. Some analyzers offer a function called fast two-port mode or tune mode in support of this approach. Normally, an analyzer in full two-port mode will switch the output power sequentially between ports 1 and 2 in order to measure all four S parameters. With fast two-port mode, the operator can specify the number of forward sweeps before power is switched to port 2 for the reverse sweep. The analyzer will then update the trace on every forward sweep (using data from the last reverse sweep) until it takes the next reverse sweep. This mode provides an improvement in speed that lies between a response calibration and a full two-port calibration, and can be especially effective for tuning. All data are fully error corrected immediately after the reverse sweep is taken.
The method by which input channels are swept can also determine throughput. An analyzer normally makes measurements in a chopped mode (in which both input ports are measured during a sweep) by alternating between the two samplers point by point during the sweep. This mode is faster than the alternate-sweep mode in which only one sampler is measured during a sweep and the other sampler is turned off to minimize crosstalk and enhance dynamic range. The other sampler is then turned on and measured for the next sweep. Again, the trade-off is speed for dynamic range. For a given dynamic range, the use of the alternate-sweep mode can yield faster results than the chopped mode with a narrower IF bandwidth.
Another way to reduce sweep time is by turning off unnecessary analyzer functions. Turning off all nonessential features such as markers spares update time and increases measurement throughput. Even turning off the display can save time in some automated systems.
The time required to recall an instrument state can add to total measurement time. An instrument state is a particular set of stimulus and response parameters that controls how an analyzer makes a specific measurement. It includes the frequency range, number of points, IF bandwidth, source power and other front-panel settings. It may also include calibration data and memory traces. The fastest instrument-state recalls take place from the analyzer's internal memory in contrast to recalling an instrument state from an analyzer's disk drive or from an external computer.
Recall speed depends on the content of the memory register or instrument state that is being recalled. The most complicated instrument states take the longest time to recall. For example, for an HP 8753E network analyzer, the recall times for a single-channel measurement with no calibration and a two-channel measurement with full two-port calibration are approximately 0.5 and 0.9 seconds, respectively. Recall times are not uniform but will vary greatly, depending on the parameters needed for a particular application.
Published instrument-state recall times are sometimes specified for simply recalling setup information, rather than recalling the information and performing a single sweep. The two conditions may be considerably different since time is required for the analyzer to set up the source and for the receiver to perform a sweep. The time required to recall the setup information and perform a single sweep is a better approximation of actual measurement conditions.
Some analyzers have a feature called spurious avoidance (used to reduce low level spurious signals) that can be shut off before storing an instrument state to speed the subsequent instrument-state recall. Turning off this function bypasses calculations and setup details, making the recall much faster. Of course, the effects of shutting off the spurious avoidance function should be examined by performing measurements with and without the function to determine whether the measured results change. In some analyzers, turning off spur avoidance does not reduce recall time.
Recalling an instrument state might not be the fastest way to set up and make a new measurement. For example, it is possible to decouple the test channels of a two-channel analyzer to set up different instrument states on each channel. It may be faster to switch between test channels rather than recall a different instrument state.
For tests that consist of a series of measurements over different frequency ranges, the list-frequency mode may provide better total measurement speed than recalling different instrument states for the ranges. In the list-frequency mode, each frequency range can be set up as a segment of the frequency list with all of the segments calibrated at one time. Each segment then can be swept separately to focus on measurements in that frequency range without losing calibration.
Determining what parts of a measurement to automate and how to do it can affect total measurement time. For example, sometimes measurement data should be sent to an external computer for processing if the computer is significantly faster than the analyzer at doing so. If the computer is not faster, it probably makes more sense to process the data in the analyzer. When data must be stored and/or manipulated, the external controller may make the most sense. Still, decisions must be made about the choice of operating system, programming language or commercial software, and type of GPIB or HPIB card for the external computer to communicate with the network analyzer. In some cases, the analyzer's internal automation might be easier to integrate into a larger automatic test equipment system than an external controller.
Automation and data-transfer speeds can be improved by following a few guidelines: Single-sweep mode should be used to ensure that a measurement is complete before data are transferred. The fastest data transfer method should be used. For example, formats that can transfer an array of data as blocks (rather than byte by byte) can enhance data transfer in some situations. A data format and associated commands that provide the fastest transfer speeds for a given application should be used. An analyzer's internal data format usually is the fastest method although the data must be reformatted for use by a PC. The minimum amount of data needed should be transferred to increase automated test speeds. For example, it may be faster to transfer a trace with only a few points in it instead of using markers to read out data.
The degree of accuracy required for a measurement can impact measurement throughput. Calibration methods that correct more errors also take more time to perform since more calibration standards must be measured. A balance must be struck between the desired accuracy and measurement speed. Frequently recalibrating an analyzer can ensure good measurement accuracy but will add to the total measurement time. Many operators will check accuracy by measuring a verification device to avoid unnecessary recalibration.
The level of required accuracy depends on the application and DUT. Network analyzers with transmission/reflection test sets can offer economical measurement solutions with good throughput using a variety of simple calibration methods such as response and one-port calibration techniques. Since these calibration methods only correct for some of the errors that might be present in a measurement, they are best suited for certain types of DUTs. For example, DUTs that have good input and output match will be less affected by source and load match errors so response or one-port calibrations can yield good results with these devices. A device that has low insertion loss, however, will have its measurements affected by source and load match errors.
For better accuracy, a network analyzer with an S-parameter test set is required. Such a system can provide full two-port error correction such as short-open-load-thru calibration. This type of calibration corrects for 12 errors: reflection tracking, directivity, source match, transmission tracking, load match and crosstalk (in both the fo