Knowledge of pulse-timing and VNA measurement bandwidths can lead to accurate testing according to the burst-signal conditions of modern communications standards.
Burst signals are now commonly used in commercial communications systems, such as Global System for Mobile Communications (GSM) cellular networks. Such signals are essentially pulsed RF waveforms characterized by long pulse widths and long periods (typically hundreds of microseconds to a few milliseconds long). Many components, particularly power amplifiers (PAs), must be tested with such signals to evaluate them under actual operating conditions. Critical measurements include tests of S-parameters, harmonics, and intermodulation distortion (IMD), which can be challenging under burst-signal conditions. Fortunately, a vector network analyzer (VNA) can be a useful tool for rapidly extracting this information since it has fast-moving receivers (Rxs) and, assuming it has an appropriate per point analog triggering function, can make all of the measurements listed above among others. Some guidelines for performing burst measurements with a VNA follow.
A time-division-multiplexed communications system will normally have a pulsed RF waveform associated with it. The waveform for a single GSM channel (e.g., ) is shown in Fig. 1 and some other systems have parameters within an order of magnitude of those shown. Relative to pulsed radar systems, a measurement challenge from earlier years, the pulse widths for GSM and similar systems are large, duty cycles are fairly large and periods are long.
Many PAs used in such communications systems are heavily optimized for efficiency and spectral purity for a given output power in this pulsed format. As such, many do not operate correctly, if at all, with a CW waveform that would typically be used for a measurement. In addition, the amplifier (or other DUT) may exhibit unusual behavior at the start of the pulse or some other time subset. It may also be of interest to measure the DUT during a certain subinterval of the pulse (termed 'profiling').
Some PAs, particularly in handsets, can operate CW but they may be put in an inactive state (via a control voltage) during the off-periods of the cycle. How the amplifier responds, in a transient sense, at the beginning and end of the pulse period (being turned active and inactive) are very important and represent another example of profiling.
S-parameter and harmonic measurements are of somewhat obvious interest. Power sweeps of gain and output power, subsets of the S-parameter measurement class, are often of greatest interest for PAs. In terms of executing the burst measurement, however, there is little difference from the general case. Relative to the control-voltage profiling example, the power and frequency are usually constant; it is a pure transient response of the S-parameter or other quantity that is of interest.
Among other PA measurements is intermodulation distortion (e.g., refs. 2 and 3) although its related cousin, adjacent-channel power rejection (ACPR) (e.g., refs. 4 and 5) is increasingly important. Both of these linearity measurements have the characteristic of some strong applied signal with a resulting distortion signal at a different, yet close, frequency. For a highly linear DUT, this is then a high-dynamic-range measurement with spot power levels below 70 dBc sometimes of interest.
One approach to doing this measurement would be with a spectrum analyzer and just consider the main lobe of the response generated by the pulse train. For IMD measurements with small separations, this may become impractical. There are also limitations in flexibility and amplitude accuracy, even with triggered measurements (similar to that discussed in the next section), with such a setup. For S-parameter measurements, the spectrum-analyzer approach cannot offer error correction, which further impairs accuracy. Previous techniques involving CW filtering of the main lobe of the pulse spectrum6 appear impractical due to the duty cycles and pulse widths involved in most relevant communications systems. This leaves direct data sampling during the pulse-on period as a potentially accurate and flexible approach.
In the case of examining transient amplifier behavior during a control pulse, direct data sampling is perhaps the only choice. Any kind of Rx can potentially make this measurement if it is scalar (e.g., gain) and the Rx can take samples fast enough. If error correction is required (S-parameters) or IMD measurements are needed, then a network-analyzer approach with direct data sampling is a valid option.
The measurement of a single signal using the direct-data-sampling technique is illustrated in Fig. 2 for both the (a) RF burst scenario and the (b) CW transient measurement. A trigger pulse is used to trigger the measurement and it must be synchronized with the RF burst (often by using a dual-pulse generator). The trigger pulse orders the VNA or vector-network measurement system (assuming the instrument has been set to accept external triggering) to perform the measurement and the instrument's analog-to-digital converters (ADCs) begin converting after a certain latency delay. The pulses must be arranged so that the burst is settled by the time the latency ends. For the setups to be discussed here, the pulse settling is very short (< 1 µs) but the DUT behavior early in the pulse may be of interest. The sampling must not continue on too long or it will overrun the burst and the resulting composite data will probably be corrupt. Since the measurement time is constrained by the GSM pulse definition (assuming no more than one burst is needed for a single tone measurement), reducing the averaging will not speed overall measurement throughput.
Figure 2 shows the measurement of one tone (as would be appropriate for an S-parameter or harmonic measurement) while a complete IMD measurement typically requires the measurement of four tones (one for each of the two main tones, one for the lower product and one for the upper product). While conceivably this could be done within one burst, there may not be enough time to get enough samples for the dynamic range required of the measurement. Most modern Rxs use some form of digital intermediate-frequency (IF) filtering so narrower IF bandwidths (and hence more dynamic range) require more samples. Thus the measurement configuration of Fig. 3 will be used where one tone is measured per burst. The picture shown would then repeat for the next frequency or power point (by this we mean both main tones move to new frequencies or new powers assuming both tones are sweeping3).
For a control-response-transient measurement (CW and constant power, a control voltage to the DUT is synchronized with the pulse-timing pattern), it is simpler to just continuously sample data. Each set of N samples (dependent on IFBW) represents one time point allowing a time-trace to be taken. This allows maximum time resolution of the DUT's response. In the case of IMD measurements, it is perhaps most accurate to measure one tone on each subsequent control pulse.
Both swept classes of measurements (single measurements or multiple per frequency or power point) can be made with a variety of test setups. Two possibilities are shown in Fig. 4(a) for low power and higher power device testing. The first is limited to output power of well under 1 W typically due to VNA input limits (varies slightly from manufacturer to manufacturer) while the second, based on a VNA plus external test set, can go to higher power levels (usually 5 W for handset-amplifier testing and 50 to 100 W for base-station-amplifier testing).
The biggest difference from an ordinary measurement is the requirement for a pulse generator. A dual-pulse generator is usually preferred so that the actual RF bursting can be easily adjusted relative to the VNA trigger pulse. A fixed-delay generator to create the trigger pulse is also a possibility but there will be less measurement flexibility. The external trigger pulse is a TTL signal while the RF bursting signal will be dependent on switch choice.
A simple shunt PIN diode switch was used for the examples to follow and the drive was set for 1.5 V (current limited) to bias the diode (shutting off the RF) and around 0 V for RF on. In higher power testing, some other switching arrangement may be needed since most PIN switches cannot handle more than 0.5 to 1 W. If the input power levels (to the DUT) rise much above 0 dBm, the switching devices should probably be placed prior to the combiner (and any pre-amps) since they can generate their own intermodulation products. Generally the reverse signals are not burst since most DUTs would not be dependent on the waveform of a small reverse signal. This path could also be switched, however, if desired.
An example setup for transient measurements under RF CW conditions is shown in Fig. 4(b). In this case, an RF switching mechanism is obviously not needed but the pulse generator is generally still required to create the control pulse for the DUT. In many cases, the operating mode of the VNA may need to change to accommodate the fast sampling required for this measurement so GPIB may be used to trigger this mode and the sweep and to synchronize the control pulse with the sweep.
With the base setup in place, there are a number of measurement details to consider. The burst operation itself rarely limits bandwidth since very broadband switch assemblies (greater than 10-GHz bandwidth) are available. In the case of the IMD-compatible setups shown in Fig. 4, the combiner (often a Wilkinson structure) is usually the limiting factor for lower power setups. These are now available covering a few hundred megahertz to at least 5 GHz so reasonable bandwidth measurements are possible. In higher-power setups (for base-station devices for example), the passive components and the DUT itself will often be narrowband for power-handling reasons.
The test setups were designed to allow some flexibility in timing between the RF pulse and the measurement trigger pulse. There is some latency between when the trigger pulse occurs and when the measurement starts in the VNA (50 to 100 µs in the MS462XX family of VNAs) so there is no need for the RF pulse to start until about that delay after the trigger. A small amount of time (less than 10 µs usually) might be needed for the switch to settle. If DUT bias switching is being used, the time constants of that bias network are relevant.
The length and period of the RF pulse are set by the standard in question (GSM, for example) so those parameters are not free. The measurement time (i.e., number of samples) can be changed as long as the measurement is within the RF pulse. In the MS462XX, the maximum number of averages is about 17 in a 30-kHz bandwidth for a clean measurement on a GSM pulse. A smaller number can be used to profile the behavior of the DUT in a certain part of the pulse (as will be shown in an example).
Power levels are usually set by the DUT requirements. As discussed previously, the only new requirements are to keep the drive levels at the switch low enough to avoid overdriving them. In the case of transient measurements, the detailed shape of the control pulse (ramp speeds) are often of critical interest and may be specified by appropriate communications standards. This is usually of less interest in the case of swept (RF burst) measurements.
To gain confidence in the method, measurements on well-behaved amplifiers (wideband, shallow compression, no gain expansion, and no linearizers) and passive devices will first be presented with and without the switch present. Since the IMD level of such a device should not be dependent on the pulse form or timing, the measured results with and without switching should be the same. Examples confirming this are shown in Figs. 5 and 6.
Figure 6 also illustrates the available dynamic range with such a burst measurement. While these graphs suggest that the timing is being done correctly, it may be illustrative to view an example where it is not. Figure 7 shows a linear-amplifier power-sweep measurement where the measurement is continued beyond the end of the burst. The resulting noisy data indicates the measurement failure.
A more conventional S-parameter measurement of a 900-MHz PA was performed next. This will be a power-sweep measurement and the output power and gain are shown in Fig. 8. The gain expansion in this particular amplifier is obvious. Since the gain is a strong function of RF drive in this region, it follows that this amplifier's performance will be a function of the burst waveform and hence it is important that it be tested with the appropriate input signal.
Harmonic measurements of an amplifier under these measurement conditions may also be of interest. A single-frequency, single-power measurement is shown in Fig. 9 covering up to the sixth harmonic to illustrate that reasonable bandwidth is quite possible with a relatively low-powered (a few watts) DUT.
To see an amplifier being affected by the burst waveform, it is interesting to look at the performance at the beginning of the pulse relative to that in the middle. This particular 1-W, 800-to-1000-MHz amplifier has gain expansion prior to its desired operating point so one might expect some dynamics at the beginning of the pulse. The two measurements in Fig. 10 are for the data being taken right after the burst starts and then midway into the burst (300 µs). The 1-dB peak difference is interesting in that it is somewhat averaged (measurement over the first 65 µs of the pulse) so it under-represents the peak. It also suggests some potential spectral quality problems for the first portion of the burst cycle for certain power levels.
At very low power levels and closer to compression (the right and left extremes of Fig. 10), one sees a much-smaller difference, which might be consistent with DUT power being a less violent function of input power in these ranges. It is important to note that the measurement early in the burst was done after the input burst itself had settled so the difference is due to the DUT behavior.
Only gross profiling has been demonstrated here but it was done with simple equipment quickly and with high possible dynamic range. Very wideband Rxs can be used for more detailed profiling but there will generally be a time or dynamic-range penalty involved.