NIST Technical Note 1337: Characterization of Clocks and Oscillators

detail elsewhere [12]. The low pass filter sectionis used in order not to saturate the amplifiers withthe carrier feedthrough signal from the mixer. Onedc amplifier is used for phase noise measurementsfrom dc to 100 kHz from the carrier. One acamplifier is used for phase noise measurements from50 kHz to 32 MHz. This range is well matched for oneof our spectrum analyzers. The wideband ac amplifierhas a bandwidth of 50 kHz to 1.3 GHz and is used whenthe desired measurement bandwidth exceeds 32 MHz.The wide bandwidth spectrum analyzer also provides aconvenient way to observe the gross features of theoutput phase noise and to identify any major spuriousoutputs if present. In order to obtain the mostaccurate measurement of the phase noise it is,however, necessary to measure the amplitude of thefirst IF signal at about 21 MHz in order to avoid thevariations in gains of the log amplifiers withvarious enVironmental factors.the beat period and pretrigerred at about -2V inorder to accurately determine the slope of the outputin volts per radian. This calculation, shown in eq.(23) below, is typically accurate ± 2% or 0.2 dB.K = Volts/Second)(Period/(2~) (23)3) The two sources to be measured are phase lockedtogether with sufficient bandwidth that the phaseexcursions at the mixer are less than 0.1 radian.The necessary phase lock gain is calculated using anestimate for the noise of the oscillators and thetuning rate for the reference oscillator. This isthen verified by noting the peak to peak excursionsof the dc amplifier and using the measured conversionsensitivity measured in step 2 above. If the peakphase excursions are in excess of 0.1 radian, thenthe phase lock loop bandwidth is increased (ifpossible) in order to satisfy this condition.4) The modulator is driven by a reference frequency(typically at +7 to +10 dBm) which steps through theFourier frequencies of interest and the detected RMSvoltage recorded on the appropriate spectrumanalyzer. This approach accurately yields therelative gains of each amplifier and its respectivespectrum analyzer since it automatically accounts forthe effect of the phase lock loop and residualfrequency pulling as well as the termination of themixer and the variations in gain of the variousamplifiers with Fourier frequency. The amplitude ofthe phase modulation on the carrier is constant inamplitude to better than ± 1.5 dB (typically ±0.2dBfor f < 500 MHz) for reference frequencies dc toabout 10' of the carrier frequency or a maximum of 1GHz. Initial measurements of the prototypemodulator are shown in Fig. 14.Fig. 13. Generalized block diagram of the new NBSphase noise measurement systems. The phase noise ofcarrier frequencies from 1 MHz to 100 GHz can bemeasured by varying the components in the phaseshifters and mixers. Dedicated measurement systemscovering 5 MHz to 50 GHz are described in the text.IV. B Measurement Sequence1) The output power of the two sources to bemeasured is typically set to between +5 and + 13 dBmat the mixer. This takes into account the insertionloss of the phase modulator. If the oscillatorsdon't posses sufficient internal isolation to preventunwanted frequency pulling, isolators (or isolationamplifiers) are generally inserted between thesources and the mixer.2) The absolute sensitivity of the mixer and the dcamplifier for converting small changes in phase tovoltage changes is determined in a way similar to thetraditional method, namely by allowing the twooscillators to slowly beat. The output of the dcamplifier is recorded by the digitizer in the FFTconnected to the dc amplifier in order to accuratelydetermine the period of the beat. In test sets C andD an additional 50 MHz digitizer is used to averageand record the beat frequency. The time scale of thedigitizer is then expanded to approximately 10' ofThis measurement is then combined with themeasurement of the absolute mixer sensitivitymultiplied by the gain of the dc amplifier describedin step 2 above. The absolute gain of all theamplifiers shown in Fig. 13 can generally bedetermined to an accuracy of ± 0.3 dB (1.5 dB forFourier frequencies from 500 MHz to 1 GHz) over theFourier frequencies of interest.5) Next the spectral density function of the FFT isverified. The level of the noise determined by theFFT for the input of the noise source amplifiersequentially shorted to ground, connected to groundthrough a 100 kO metal film resister, and connectedto ground through 200 pF, is then recorded. Fromthese data one can determine the inherent noisevoltage and noise current of the noise sourceamplifier plus the FFT as well as the noise of theresistor to about ± 0.25 dB which is the statedaccuracy of the FFT. This primary calibration of theFFT can be carried out from about 20 Hz to over 50kHz. Above 50 kHz, the noise gain of the amplifierwe used contributes a significant amount of noise.With some compensation the noise is flat to within±0.2 dB from 20 Hz to 100 kHz. Next the noise sourceis switched into the output of the mixer and therelative noise spectrum of all the spectrum analyzersis calibrated by knowing their relative gains. Thisprocedure verifies the voltage references and noisebandwidths of the various spectrum analyzers.6) The noise voltage is recorded on the threespectrum analyzers over the Fourier frequencies ofinterest, generally the same one used in step 4439TN-136