Increased data-rate demands have driven the change from simplemodulation schemes such as FM, used on ProfessionalMobile Radio (PMR) equipment, first-generation cellphones and microwave links, tomore complex schemes such as GMSK,CDMA and N-QAM . CDMA signals, such as theIS95 “narrow” band CDMA standardorthe 3GPP W-CDMA standard, have a considerable amount of amplitude.
Typically, engineers expect a minimum of 10dB peak-to-average powerand a peak- to-average ratio of up to 16dB. Such amplitude variationmakes conventional CW linearity-corrected diode sensor unsuitable forthese measurements.
Radio links have adopted NQAM, typically 64-QAM or 256QAM, toincrease the data rate. Other systems such as newer, higher WLANstandards are also adopting 64-QAM for their fastest data rate. Symbolrates for these systems are higher than the bandwidth of commonlyavailable peak power meters. Moreover, an rms measurement canaccurately and economically indicate the power of the system.
Three types of power sensor design
The three types of power sensor design are thermistors, diodes andthermopiles or Seebeckeffect devices. Thermistors were traditionallyused for standard transfer. However, these resistors were not used fornormal measurements on systems and equipment due to its limitedpower-handling capability.
Diode-based sensors have two available formats – the square lawonly-based sensors and the linearity-corrected wide dynamic rangesensors. Recently, a third type has been introduced, the multiplediode-based sensor. Thermopile or Seebeck effect sensors operate like athermocouple and rely on the heating effect of the input signal (Figure 1 below ).
|Figure1. A thermopile sensor is best for measuring the true rms power ofcomplex waveforms like NQAM.|
This makes it suitable for measuring the true RMS power (root mean square)of complexwaveforms like N-QAM. These sensors always respond to the true rmsvalue of the input waveform regardless of the modulation on thecarrier. With a good return loss, thermopile sensors can reduce theuncertainty of a measurement.
However, they tend to be limited in dynamic range and are relativelyslower square law regions of three diode paths to make a true rmssensor that spans the dynamic range of 20dBm to -60dBm. There are twochangeover points between the diode pairs. Note that the firstchangeover is at approximately -3.5dBm and the second at -23.5dBm.
|Figure2. The diode square law extends from -70dBm to about -20dBm.|
Figures 3 and 4 below showthe system diagram and the physical layout of a universal sensor. Thepath for detector A has 40dB of attenuation and the detector isselected when the input power ranges from 20dBm to -3.5dBm. Hence, thesignal level on the than a diode sensor. Conventional diode detectorsoperate in the square law region, and are thus limited to the 50dBdynamic range (Figure 2 above ).
|Figure3. Operating three diode pairs can enable a sensor with a true rmsrange of 80dB.|
|Figure4. A three-path universal sensor is useful for measuring W-CDMA signalsproduced by UEs that cover a wide dynamic range.|
They can also use linearity correction techniques to extend theirdynamic range. However, the speed of the power meter limits suchtechniques, making them unsuitable for applications where thesystem-transmitted symbol rate exceeds the sampling rate of the powermeter.
The diode square law extends from -70dBm to about -20dBm. Theuniversal sensor uses the diode varies from -20dBm to -43.5dBm.Detector B has 23dB of attenuation and is selected when the input poweris between -3.5dBm and -23.5dBm.
The signal level at the diodes varies from -26.5dBm to 46.5dBm. Thefinal diode pair, detector C, has 6dB of attenuation and operates whenthe input level falls below -23.5dBm. Thus, the signal level on thediodes varies between -29.5dBm and -66dBm.
Operating three diode pairs can enable a sensor with a true rmsrange of 80dB. This is useful for measuring W-CDMA signals produced byuser equipment that covers a wide dynamic range. It is possible to havea similar sensor using just two diode paths. Note that the square lawregion of the diodes is 50dB, so using two paths operating over a rangeof 40dB each can produce an 80dB dynamic range sensor.
However, as the two-path sensor reaches the midway changeover pointof -20dBm, the input power hits -60dBm on the diode, where noise canseriously influence measurement. For the three-path sensor, the lowestsignal at either changeover point is -46dBm. This results in a betterSNR than the two-path approach, thus enabling faster and more accuratemeasurements.
The power meter, calibrator, sensor and properties of the DUT such asmatch and spurious signal output can influence measurement errors.
In a modern power meter (Figure 5,below ), the incoming signal is amplified and converted by theADC and then processed by the DSP. In traditional power meters, each ofthe range settings of the amplifiers was dedicated to a decade range.Wide dynamic-range power sensors required the use of ADCs that cancover ranges above 10dB.
|Figure5. Wide dynamic-range power sensors demanded the use of greaterdynamic-range ADCs to cover ranges above 10dB.|
The instrumentation accuracy of a power meter is below 0.5 percentand can be treated as a universal error. This is the performance of apower meter when considered as a baseband voltage measurement system.Some of the parameters that used to affect this figure, such asquantization error and zero carryover, have been reduced by theadoption of ADCs with greater resolution.
Often, the lowest gain range has the largest dynamic range. Let'sexamine the effect of quantization on this gain range, which isconsidered the most significant. The maximum input voltage to the ADCon this range is 4.5V. The 16bit ADC delivers a resolution of68.6µV per bit.
The smallest signal that the range has to deal with is approximately80mV. This corresponds to approximately 1,200bits on the ADC. So thequantization error is less than 0.09 percent and need not be treatedseparately. Other amplifier ranges have a much smaller dynamic range,hence quantization error is considerably smaller. Zero set and drift isresidual from the zeroing process and its drift over one hour, measuredwith maximum averaging.
The specification for this parameter is that the error term is below0.5 percent of full scale on the most sensitive range. For the twosensors discussed here, the most sensitive ranges cover 10dB. The zeroset is 0.05mW for the fast thermal sensor and 0.05nW for the universaldiode sensor. The effect of the zero set and drift becomes moresignificant as the power level decreases on the bottom range. For asignal at the lowest end of the published dynamic range, thecontribution is less than five percent.
The power reference provides the power meter with a traceable 0dBmreference level for calibrating the sensors. Calibration of thereference adheres to standards and can be considered accurate to within±1.2 percent peak, or 0.9 percent RSS in a year. Another errorthat must be accounted for is the mismatch between the sensor undercalibration and the reference. The reference has a VSWR of less than1.04, which reduces this error. For the two sensors underconsideration, this error term is 0.31 percent.
The power sensor contributes five factors to the uncertainty budget:
1) Linearity. The sensor hasa linearity specification, which is the measurement deviation from anideal power measurement device.
2) Temperature coefficient. Both the thermopile and the diode elements have a temperaturecoefficient. Some modern sensors are individually calibrated fortemperature drift and have a small thermistor located on the substrate,which the power meter uses to calculate the correction. Typically,there is less than 1 percent residual error for a temperature range.
3) Mismatch uncertainty. This occurs during the measurement between the sensor and the DUT.Often, this hugely impacts the error budget, even with a relative goodmatch of the sensors.
4) Cal factor uncertainty. This is a function of the mismatch between the sensor and the calfactor calibration system, and is influenced by the sensor beingtested. So, a fast thermal sensor at 38GHz has a cal factor uncertaintyof 3.62 percent and the universal sensor at 2.2GHz has 0.6 percent.
5) Noise. This is dependenton the type of sensor and the signal level applied. For a thermopile,the noise contribution increases as the signal level decreases. For auniversal sensor, engineers must consider the increase in noise on eachset of diodes toward the range-change points. After the range change,SNR improves. The power meter signal channel contributes relativelylittle to the overall noise performance of the sensor. Meanwhile,averaging can have a facility that allows users reduce noise.
Mismatch can contribute significantly to the error budget duringmeasurement. It is caused by the different impedances between thesensor and the source (Figure 6, below ).
|Figure6. Mismatch due to the different impedances between the sensor and thesource causes measurement uncertainty.|
Assume the worst
Sensors, which are passive terminations, tend to have a better matchthan active sources. The reflected wave combines vectorially with thetransmitted wave to produce a standing wave. The sensor will detectthis, but it is not possible to locate the maxima and minima.Consequently, engineers should always take the worst case whenconsidering mismatch error. This equation describes mismatch:
%mismatch uncertainty = 100L(1'”s”/)2 -1,
where s is the source and L is the load, which in this caseis the sensor.
An attenuator improves mismatch mismatch error. Some power metershave a facility that allows users to enter a table with attenuatorvalues that can be applied to the measurement. A precision attenuatorcan be calibrated to 0.05dB or 1.15 percent. If a nonprecisionattenuator is used, the calibration error can be larger than themismatch improvement required.
Harmonics and spurious signals also cause errors on powermeasurement. Square law sensors will add the powers of all the signalswithin its passband. For most finished systems compliant withinternational specifications, the influence of these signals on themeasurement is negligible.
However, for measurements made on uncompleted systems or parts ofsubsystems where there is no filtering, these signals can induce extraerrors.
Consider a local oscillator that is leaking through a mixer and isonly 20dB down on the main signal. The sensor will add the two powerstogether, causing an additional 1 percent error due to the presence ofthe two signals.
Meanwhile, for an amplifier running into compression, the harmonicoutput maybe 10dB down from the carrier. Such will add an extra 10percent on the reading, which is more significant compared with therest of the errors in the system. This property of true rms sensors canbe used during multicarrier tests.
For two carriers spaced a few megahertz apart, the peak voltage is2V. A diode-based peak power meter reads this as 4xPower, while thetrue rms sensor will correctly identify the combined signals as 2P.
Let's consider the influence of these errors on the two measurementscenarios involving a 2.2GHz W-CDMA signal measured with a universalsensor at 10dBm and a thermal sensor measuring a 38GHz radio link at10dBm. In both cases, we will assume that the source VSWR is 1.5, andthat the signal spurious output and the effect of zero noise drift areboth negligible.
|Table1. Linear summation assumes that the worst-case errors will always add,while RSS summations claim that errors are from different sources.|
The linear summation (Table 1 above )assumes that the worst-case errors will always add. In RSS summation,it is considered that errors come from different physical mechanisms,so it assumes that on average, they will not add in the worst case.Many companies and uncertainty schemes adopt this approach when dealingwith the summation of non-physically related uncertainties. Using 3Dgraphs can best demonstrate these uncertainties.
|Figure7. With averaging, noise at changeover points is reduced to negligible.|
Figure 7 above shows the sumof the uncertainties at room temperature for the universal sensor withno averaging applied. Worst-case addition has been used. The influenceof noise on each diode path is also detailed in the graph. With amodest amount of averaging, noise at the changeover points can bereduced to negligible levels.
|Figure8. The 2.5 percent uncertainty (lowest in the graph) is just over±0.1dB while the 8 percent uncertainty (highest) is 0.33/-0.36dB.|
Shown in Figure 8 above isthe uncertainty surface for the thermopile sensor across its operatingfrequency range. The source match is at 1.2, to reduce the uncertaintydue to mismatch. The uncertainties have been added as RSS terms. Theincrease in uncertainty at low power levels is mainly due to theinfluence of the zero set parameter. The frequency-related ripple isdue to the cal factor uncertainty varying across the range.
Michael Osoba is ProductMarketingEngineer at Anritsu Ltd.