An advanced question - sampling an oscillator's signal for analysis

[Vgkid]

Thinking about this , couldn't you measure the difference in time between the 2 sources , doing this at 1pps will be a lot easier.
I need to go to bed , I will edit / reply when rested.

[dnessett]

I have been inactive on this thread for a while because I found the phase noise measurement device that I was looking for. I purchased an HP11729c carrier noise test set unit from ebay for $299 + $69 shipping. It supports phase noise measurements down to 1 Hz, although I probably won't push it that close to the carrier for the foreseeable future. I have spent most of my hobby time in the past several weeks reading its documentation and some supporting application notes gearing up to use it. It arrived three weeks ago and I have been familiarizing myself with its operation in preparation for testing.

I document the test setup to support two objectives: 1) to get some constructive feedback, and 2) to make it visible so when I publish test results those reading them are aware of how they were made. For those uninterested in how the tests are made and wish only to see the results, skip this and the next two of my posts.

During my reading, it became clear that much of the test setup and procedures required to determine phase noise are applicable to both the HP11729c and the AD8302. At some point I may compare measurements made with the AD8302 to those made with the HP11729c in order to see how accurate the AD8302 is in measuring phase noise.

Figure 1(a) shows the configuration of a generic Heterodyne Phase Detector. The Reference and DUT outputs both feed a mixer. The output of the mixer is passed through a low pass filter, which produces the phase difference signal.

Figure 1

The AD8302 (Figure 1(b)) uses log amplifiers before the mixer. However, the AD8302 does not itself provide a low pass filter for the mixer output. That must be provided externally.

The HP11729C is configured to provide two phase difference signals. Figure 1(c) presents a simplified view of how the first of these signals is created. This signal has a bandwidth of 10 Hz to 10 MHz and varies between 1 and -1 Volt. The mixed signal is first processed by a 15 MHz low pass filter, which feeds a 40 dB low noise amplifier. The result is presented at a connector on the unit, which has an output impedance of 50 ohms.

The second phase difference signal (Figure 1(d)) provided by the HP11720C has a bandwidth of 1 Hz to 1 MHz. Its signal path has the mixer and 15 MHz filter in common with the 10 Hz to 10 Mhz signal. However, it is not amplified before output. Instead, it is filtered a second time by a 1.5 MHz filter. The result is then presented at a second connector on the unit's body with an output impedance of 600 ohms. It varies between +10 and - 10 Volts.

Figure 2 shows a generic version of the test setup architecture. A concrete version of this architecture may contain other components and may elide some of those shown. Also, the signals feeding the phase detector may be sourced through the coupled port of the directional coupler, rather than the out port, in which case the out port is connected to the oscilloscope. Finally, the inputs to the phase detector may in some cases require the insertion of an attenuating pad to ensure the phase detector is not over-driven.

Figure 2

Figure 2 also illustrates how the phase detector signal is analyzed. Two (not necessarily exclusive) options are shown. The first is to display the spectrum of the phase detector signal on a low frequency spectrum analyzer in order to read the phase noise spectrum directly. This option is useful for measuring short-term oscillator stability. Normally, the spectrum analyser output is captured in a file in csv format and transfered to a machine for subsequent analysis. The output of the phase detector may also be connected to a data acquisition system that archives a digitized record of the phase detector signal for processing at a later time. This option is necessary for characterizing medium- to long-term oscillator stability.

The generic architecture of the oscillator test setup is presented to provide context for the plan to characterize oscillators for short-, medium- and long-term stability. The next post presents the first instance of the generic architecture, which is used to measure short-term oscillator stability. Specifically, it presents the details of a test setup using the HP 11729C as a frequency discriminator (delay line or one oscillator) phase noise test harness. A subsequent post describes the procedures used to turn the spectrum data into a phase noise plot.

It is unlikely that the casual reader will deep dive into the details of the next two posts. I provide these details only for those who are interested and to concretely document the measurement techniques used to characterize the phase noise information that I will publish on the EEVblog subsequently.

[dnessett]

Documented in this and the next post is a detailed description of the test setup and procedures used for measuring phase noise using the HP11729C as a frequency discriminator (one oscillator configuration). This post describes the test setup and the next post the procedures used to obtain phase noise characteristics for several hobbyiest oscillators.

I am providing this information with two objectives in mind. First, I invite constructive criticism of these details in order to improve the accuracy of the measurements I will make and publish. Second, putting these details on record allows those who read the phase noise measurements to know how those measurements were made.

Figure 1 shows the general architecture of the test setup. The box labeled "Phase Detector" is the HP11729C. This architecture follows the instructions given in the HP11729C operation manual. I will provide the set up information in stages, described by the headings prior to the textual description.

Figure 1

HP11729C Mechanical Configuration

For this test setup, the HP11729C has the following mechanical configuration:

• 50 ohm terminator on the 640 MHz output. Since the oscillators tested are all 10 MHz, the 640 MHz signal (which generates down converting frequencies to put the oscillator signal into the 10-1028 MHz range) is unused.
• 50 ohm terminator on the unused noise spectrum output. Sometimes this is the 1Hz-1MHz output and other times it is the 10Hz-10Mhz output.
• The Mode selector is set to phi(phase), CW and the Local selector is on.

Device Under test

The Device Under Test (DUT) must output a signal having power within the range of the HP11729C input limits. For 10 MHz signals this range equals -5 dBm to +10 dBm. All the oscillators tested have power greater than - 5 dBm (0.3556 V_p-p). Some however exceed 10 dBm (1 V_p-p). In this case a suitable attenuator pad (not shown in Figure 1) is placed between the DUT and the HP11729C signal input.

Directional Couplers

The HP11729C inputs terminate the coaxes that are connected to them. However, to monitor the signals during a test, two directional couplers are used to tap the oscillator and delay line inputs. The signals from these taps are displayed on an oscilloscope (Rigol 1104Z), which allows the use of the Delay Device to bring the two signals into rough quadrature. The Phase Lock indicator on the HP11729C is then used to bring the signals into tight quadrature. The insertion loss of a coupler, if necessary, is considered when determining which attenuation pad to use between the DUT and the directional coupler. Unless otherwise noted, the directional couplers used are MiniCircuit ZDC-10-1 devices.

Delay Line and Delay Device

The Delay Line is 400'+2*25'+2x50' = 550' of RG-58 coax. The total delay of the signal between the IF output and the mixer input varies depending on the delay value selected for the Delay Device. The Delay Device is described in this EEVBlog topic. Normally, the signal is delayed about 875 ns (8 full periods plus an extra 75 degrees) by the combined Delay Line and Delay Device (the Delay Complex).

Figure 2 shows a Tracking Generator trace of the Delay Complex (with a typical setting to bring two 10 MHz signals into quadrature) from 1 MHz to 20 MHz. As is apparent, the delay characteristic is linear in this frequency region.

Figure 2

Figure 3 shows an oscilloscope display of the input and output of the Delay Complex with a 10 MHz/1 V_p-p input generated by a Rigol DG1022. The measurement cursors show the amplitude of the input signal.

Figure 3

Figure 4 shows the same display with the cursors measuring the amplitude of the output.

Figure 4

The data in these two images shows that the Delay Complex reduces the power of the input by 8.268 dB. The calculation is:

Input: 1.088V_p-p = 4.712 dBm
Output: 420 mV_p-p = -3.556 dBm
Difference: 8.268 dB

As stated in the next post on procedures, the maximum sensitivity of the frequency discriminator occurs when the delay complex attenuates the input signal by 8.7 dB. While the 8.268 dB value provided by the Delay Complex is somewhat less than this optimal value, the need to put the two signals into quadrature and to utilize coaxes that were available necessitated the use of this slightly non-optimal value.

Low Frequency Spectrum Analyzer

The output of the HP11729C is the output of the mixer after it is passed through a 15 MHz filter and low noise amplifier. This signal carries the phase noise information of the oscillator under test (DUT). The bandwidth of the signal used for the foreseeable future is 10 Hz - 10 Mhz. Some of the most interesting information is at very low frequencies, specifically those less than 9 KHz. The spectrum analyzer used in these tests is a Siglent SSA3021X (hacked to elevate it to the capabilities of a SSA3032X). This spectrum analyzer has a lower frequency bound of 9 KHz, which means much of the interesting information in the phase noise signal is inaccessible.

I am currently investigating ways to capture the spectral phase noise data below 9 KHz. For the present, however, this is the lower bound for phase noise measurements using the test setup documented here.

[dnessett]

Given the test setup architecture documented in my last post, here are presented the procedures used to measure phase noise with a HP11729C in a frequency discriminator configuration for the hobbyiest oscillators targeted by the project. The procedures themselves are derived from the HP11729C operators manual and suitably modified. Mathematical support for these procedures are documented in the appendices of that document as well as in two HP application notes (especially their appendices), specifically, Phase Noise Characterization of Microwave oscillators - Frequency Discriminator Method and Phase Noise Characterization of Microwave oscillators - Phase Detector Method Method.

Warning: This post contains more than a normal amount of mathematics and is probably of no interest to the casual reader. I provide it only to document and justify the procedures I use to derive the phase noise data for various hobbyiest oscillators. Unless you like mathematics and are interested in these procedures, I suggest you stop reading at this point and move on to the next post.

Critical to understanding the mathematical justification of the procedures sketched below is the derivation of the phase discriminator equation, which applies to both the Phase Detector (two oscillator) Method and the Frequency Discriminator (one oscillator) Method. This derivation defines two key constants, the phase discriminator constant (\$K_{d}\$) and the phase detector constant (\$K_{\varphi}\$) that control the output of the Phase Detector.

The phase discriminator constant depends on the phase detector constant, which is, in fact, the mixer conversion gain/loss constant (in the HP11729C case it is a loss). These two constants are related according to the equation:

\$\nu(t)=K_{d}\varphi(t)\$, where \$K_{d}=\frac{K_{\varphi}V_{R-AMP}V_{DUT-AMP}}{2}\$, \$\nu(t)\$ is the voltage output of the Phase Detector after low-pass filtering, \$V_{R-AMP}\$ is the amplitude of the reference oscillator (or the Delay Complex output in the case of the Frequency Discriminator Method), and \$V_{DUT-AMP}\$ is the amplitude of the DUT. To get the units right (see this discussion (12-2-18 - corrected the missing URL)), \$K_{\varphi}\$ is specified in units of Volts^-1.

This equation is dervied in appendix B of Phase Noise Characterization of Microwave oscillators - Frequency Discriminator Method. However, it is not particularly well argued. I have attached a pdf file to this post that presents a more general deriviation. It loosely follows the material in section 8.9.1 of Time and Frequency: Theory and Fundamentals. I decided not to include this math inline because: 1) there are probably few who are interested in it, and 2) this post is already very long.

The output of an HP11729C when in a Frequency Discriminator (single oscillator) configuration is a signal having the following time dependent voltage (see appendix A of Phase Noise Characterization of Microwave oscillators - Frequency Discriminator Method for the mathematical justification):

\$\nu(t)=K_{\varphi}2\frac{\Delta f}{f_{m}}\sin(\pi f_{m}\tau_{d})\sin(2\pi f_{m}(t-\tau_{d}/2))\$

where \$f_{m}\$ is the modulation frequency; \$\Delta f\$ is the peak modulation deviation; \$\frac{\Delta f}{f_{m}}\$ is the modulation index; and \$\tau_{d}\$ is the total time delay of the Delay Complex.

(Note to the fastidious: Appendix A cited above is generally a good derivation. However, the author gets sloppy when dealing with \$K_{\varphi}\$. In particular, the two input signals are specified as V_R(t) and V_L(t) where each is defined as v*cos(angle stuff). When the mixer output is displayed, it is given as V_m(t)=\$K_{\varphi}\$*(lots of sinusoids). That would imply \$K_{\varphi}\$ = v²/2, which has units Volts². But this isn't compatible with the eventually derived equation, which has units V/Hz, not V²/Hz. For those interested in this question, I again suggest reading this discussion (12-2-18 - corrected the missing URL).)

The second sine on the right hand side of the equal sign in the above expression is a time varying signal, while the expression preceding it is the time-independent amplitude of that signal. Thus, the change of amplitude as a function of a change of (Fourier) frequency is expressed as (see above referenced appendix A):

\$\Delta V=2\pi K_{\varphi}\tau_{d}\Delta f\frac{\sin(\pi f_{m}\tau_{d})}{\pi f_{m}\tau_{d}}\$

For \$f_{m}<\frac{1}{2\pi\tau_{d}}\$ the approximation \$\frac{\sin(\pi f_{m}\tau_{d})}{\pi f_{m}\tau_{d}}\backsimeq1\$ holds, in which case we can use the approximation: \$\Delta V\backsimeq K_{d}\Delta f\$, where \$K_{d}=2\pi K_{\varphi}\tau_{d}\$. \$K_{d}\$ has units Volts/Hz.

The condition \$f_{m}<\frac{1}{2\pi\tau_{d}}\$ controls the maximum phase noise frequency this method can measure. As specified in the test setup architecture post, the delay created by the Delay Complex is approximatley 875 ns. Using this value in the above expression yields a maximum phase noise frequency of about 180 KHz. Attempting to resolve phase noise frequencies above this value will yield erroneous results.

The lower bound of phase noise frequency resolvable by the test setup depends on both the HP11729C and the spectrum analyzer. The 10Hz - 10 MHz output of the HP11729C resolves phase noise down to 10 Hz, whereas the 1Hz - 1 Mhz output resolves phase noise down to 1 Hz. However, the Siglent SSA3000X has a lower Fourier frequency bound of 9 KHz. Until I figure out how to implement a data recording low frequency spectrum analyzer (or buy one), that is the lower limit of phase noise measurements this setup supports.

Calibration and Measurement

The procedures documented here yield the function: \$\mathscr{L(\mathcal{\mathrm{f}})}\$, where

\$\mathscr{L(\mathcal{\mathrm{f}})=\frac{\mathcal{P_{SSB}}(\mathrm{f})}{\mathcal{P_{\mathrm{Carrier}}}}}\$

\$\mathcal{P_{\mathrm{Carrier}}}\$ is the power of the (oscillator) carrier signal. \$\mathcal{P_{SSB}}(\mathrm{f})\$ is the power of the frequency discriminator output at the Fourier frequency f. The spectrum analyzer captures \$\mathcal{P_{SSB}}(\mathrm{f})\$ for the phase noise frequencies within the bound alluded to above. However, the mixer output does not include the carrier frequency power. In fact, the carrier frequency power is never directly measured. Instead, \$\mathcal{P_{SSB}}(\mathrm{f})\$ is expressed logrithmically and the log value of \$\mathcal{P_{\mathrm{Carrier}}}\$ set during calibration is subtracted from it. The first procedure (documented in the HP11729C operators manual starting at pg 3-22) is the calibration used to accomplish this.

In brief, a signal generator (in my case a Rigol DG1022) capable of producing FM modulated signals is first connected to the spectrum analyzer. The signal produced is a 10 MHz carrier at -10 dBm modulated by a 10 KHz sinewave. The frequency deviation is set to 200 Hz. This results in sidebands at 10 MHz +/- 10 Khz. The difference between the power of the 10 MHz carrier and the positive 10 KHz sideband is recorded in dB. The signal generator is then connected to the HP11729C input and the Delay Complex is used to interconnect the HP11729C IF output to its 5-1280 MHz input (the other input to the HP11729C mixer). The HP11729C 1Hz-10MHz output is connected to the spectrum analyzer. The Delay Device is then set so that the signal generator input and Delay Complex input are in quadrature. The power of the 10 KHz sideband is noted.

The signal generator is disconnected from the HP11729C. The oscillator DUT output is attenuated so that it is in the range -3dBm - 2 dBm. This is then connected to the input port of the HP11729C. The resulting spectrum is recorded (in csv file format) and saved to a USB stick.

Corrections

The values of the noise spectrum generated using the procedures sketched above need correction in order to produce the correct value of \$\mathscr{L(\mathcal{\mathrm{f}})}\$. This is accomplised using Octave. The csv file is moved to a suitable computer.

Corrections are made by applying to every point in the spectrum representing \$\mathscr{L(\mathcal{\mathrm{f}})}\$ the following adjustments (see Appendix A of the HP11729C operators manual):

• Convert the data from 10 Hz RBW to 1 Hz equivalent noise bandwidth. The Siglent SSA3000X datasheet shows a RBW to noise bandwidth correction factor of -10 dB for frequencies between 1 MHz and 3.2 GHz whether the preamp is on or off. Consequently, 10 dBm must be subtracted from each data point in the spectrum.
• The phase noise data need correction according to the values used and observed during calibration. First, the carrier power used during calibration (-10 dBm) is added to each phase noise datum, effectively reducing its value by 10 dBm. Then the observed carrier to sideband separation in dB is subtracted from each data point. This converts the data units to dBc/Hz.
• The phase noise data does not yet represent the values of \$\mathscr{L(\mathcal{\mathrm{f}})}\$. Converting them to \$\mathscr{L(\mathcal{\mathrm{f}})}\$ requires a correction factor that depends on the phase noise Fourier frequency Specifically: \$-20\log\left(\frac{f_{off}}{f_{cal}}\right)\$ is added to each phase noise datum, where \$f_{off}\$ is the Fourier frequency corresponding to the phase noise datum (offset from the carrier) and \$f_{cal}\$ is the calibration frequency (10 KHz). The justification for this correction is given in Appendix A of HP11729C operators manual starting on page A-3 in the section titled, "Frequency Discriminator Correction Factor". Notice that for the phase noise datum corresponding to 10 KHz, this results in a log(1)=0 correction factor, i.e., the value of the 10 KHz datum does not change.
• The Siglent, as do most modern spectrum analyzers, uses logrithmic averaging, which introduces errors when measuring noise. This is explained in Keysight Technologies Application Note: Spectrum and Signal Analyzer Measurements and Noise (see pages 6-8). This requires adding 2.51 dBm to each data point.

[dnessett]

In order to test the HP 11729C Phase Noise measurement procedures for correctness, I analyzed an FEI FE-5650A. The raw (uncorrected) results from the Siglent SSA3032X spectrum analyzer are shown in Figure 1.

Figure 1

This data is provided only to show that it remained above the noise floor of the Siglent.

Figure 2 is a graph of the corrected data.

Figure 2

Three data points are of interest:

10 KHz : -124 dBc/Hz
100 KHz: -169 dBc/Hz
180 KHz: -178 dBc/Hz

The 10 KHz figure is in rough agreement with the FE-5680A phase noise measurements posted on John Miles's KE5FX website. (Thanks to Skip Withrow of RDR-Electronics for pointing me to this result.) The 5680A and 5650A are virtually the same device, where the latter is repackaged in a smaller enclosure. However, the figures for 100 KHz are different. I measured -169 dBc/Hz and the graph on the web page shows around -133 dBc/Hz. It isn't clear (at least to me) whether this is a difference between the FE-5680A and FE-5650A or is an artifact of the HP11729C measurement discipline. In regards to the latter, it is interesting to note that as the frequency offset gets larger, the correction factor of \$-20\log\left(\frac{f_{off}}{f_{cal}}\right)\$ drives the corrected phase noise value lower and lower. So, it may be that there is a limit to the separation between the calibration frequency and the offset frequency, since one would expect the noise plot to stablize and become non-decreasing once it becomes pure white noise. As far as I can tell, this is not mentioned in the operator manual or application notes.

[hendorog]

Great job documenting all of that.

I'm unclear on how you can tell that your measurement is above the _phase noise_ floor of the SSA?
I thought that you need a low frequency SA which itself has better phase noise performance than the DUT to do a PN measurement using a delay line.

The only way I know of to measure below the PN floor of the SA is to use NFE to subtract a pre-measured PN floor. I'm not an expert by any means so correct me if I'm wrong.

[dnessett]

Great job documenting all of that.

Thanks.

I'm unclear on how you can tell that your measurement is above the _phase noise_ floor of the SSA?
I thought that you need a low frequency SA which itself has better phase noise performance than the DUT to do a PN measurement using a delay line.

The noise floor of the Siglent is documented in its data sheet as -126 dBm at 10 MHz with a RBW of 10 Hz with the preamp off and -144 dBm at 10 MHz with a RBW of 10 Hz with the preamp on (which I used for the measurements). The raw data (Figure 1) I obtained from the SA is higher than -100 dBm, so the data is not corrupted by the noise floor of the SA.

You are right, I need a low frequency SA to get to most of the interesting data. I am working on that right now. Also, I am trying to understand why at 100 KHz and 180 KHz the corrected data is obviously wrong. I have put a question out to a group that specializes in old HP/Agilent/Keysight equipment to see if they can help.

The only way I know of to measure below the PN floor of the SA is to use NFE to subtract a pre-measured PN floor. I'm not an expert by any means so correct me if I'm wrong.

I don't want to measure below the noise floor of the SA (the PN floor of the SA isn't relevant, since the signal from the HP11729C is a voltage that the phase discriminator equation maps to frequency fluctuations). I just wanted to check that the data wasn't being corrupted by the SA's noise floor.

[hendorog]

Thanks!

Reading up about it in the pdf you linked - note there is a extra character at the start and end of your link to this pdf.

http://hpmemoryproject.org/an/pdf/pn11729C-2.pdf

[dnessett]

Thanks!

Reading up about it in the pdf you linked - note there is a extra character at the start and end of your link to this pdf.

http://hpmemoryproject.org/an/pdf/pn11729C-2.pdf

Thanks for the heads-up on the bad hyperlinks. I usually test all of the links before posting, but I obviously didn't for those. I have fixed them (there were three).

[dnessett]

I asked the HP/Agilent/Keysight interest community on groups.io about the suspcious results I obtained for the phase noise of an FE-5650A using a HP11729C. There were several suggestions why the HP11729C might not provide the correct values, but John Miles (KE5FX), who also contributes to this forum, provided the correct analysis. To ensure this post is self-contained, I repeat the phase noise measurements I obtained:

10 KHz : -124 dBc/Hz
100 KHz: -169 dBc/Hz
180 KHz: -178 dBc/Hz

While the 10 KHz value was in the range of values shown for the FE-5680A (which is supposedly the same electrical device in a different physical package), the value for 100 KHz was much lower than that shown on John Miles's website (see the phase noise plot). The phase noise graph did not display a value for 180 KHz.

John had used the HP11729C in the past and remembered that for frequencies near the low end of its input range, the input signal sum component in the output of the mixer/LPF drives the LNA into saturation. This means the amplifier cannot properly produce the correct phase noise output.

In the case under consideration, the mixer produces a 20 MHz output component of considerable power, which represents the sum of the two inputs into the mixer (10 MHz). The 15 MHz LPF doesn't sufficiently suppress this sum. In addition, looking at the input to the LNA (which is available as the aux noise output), an even stronger 10 MHz component exists in the mixer output. This component will not be suppressed by the 15 MHz LPF. So, both the 10 MHz and attenuated 20 MHz components drive the LNA into saturation. It isn't clear why the mixer is producing a 10 MHz component, since classically only sum and difference products should appear.

Figure 1 shows the signal output of the 15 MHz LPF (the aux noise output). Notice the significant 10 MHz and 20 MHz frequency components.

Figure 1

Figure 2 shows the output of the LNA. There are significant extraneous spurs visible, which is evidence of LNA saturation. Note: the power of the 10 MHz and 20 MHz components is about 50 dB higher than the aux noise output. The LNA is spec'd at 40 dB gain, so this is additional evidence of the LNA going into compression.

Figure 2

I thank John for this analysis. To be honest, it is not something that would have occurred to me.

To work around this problem, I decided to use the 1Hz-1MHz output of the HP11729C. This output does not use the LNA and has an extra 1.5 MHz LPF that will eliminate the 10 MHz and 20 MHz outputs of the mixer.

The 1Hz-1MHz output has an output impedance of 600 ohm and a voltage range of +/- 10V. Fortunately, I had bought a 600-50 ohm impedance matching pad in case I had to use this output. This pad has an advertised insertion loss of 16.6 dB. Consequently, when processing the raw output, I included a correction that added 16.6 dB to each data point (in addition to the other corrections specified in a previous post). Since a voltage swing of +/- 10V represents a signal with maximum power of 30 dBm, with the 600-50 ohm matching pad in place, the maximum power of this signal would be 13.4 dBm. On the other hand, since the two inputs are in quadrature, it is likely the actual voltage swing of the output will be much less.

I ran the phase noise experiment. The processed spectrum (10Hz-200Hz) is shown in figure 3.

Figure 3

This yielded better results for 100Khz. Specifically,

10 KHz : -96 dBc/Hz
100 KHz: -122 dBc/Hz

Comparing these results with those on John Miles website reveals some puzzles. In particular, the published results are:

10 KHz: lower bound of around -125 dBc/Hz and upper bound of around -100 dBc/Hz (the data is in a graph and not presented numerically)

100 KHz: lower bound of around -133 dbc/Hz and upper bound of around -113 dBc/Hz

The 100 KHz result from the HP11729C is within the bounds of the published results, but the HP11729C 10 KHz result is about 4 dB higher than the upper bound of the published result.

I decided to back out the 16.6 dB correction for the impedance matching pad and see what happened. Figure 4 shows the result.

Figure 4

Numerically, the offsets of interest are:

10 KHz : -112 dBc/Hz
100 KHz: -138 dBc/Hz

Now the 10 KHz value is within the range of the published result, while the 100 KHz result is about 5 dB lower than the published result's lower bound.

What to do? After stewing on this for a while, I had an idea. Suppose the published figure for the impedance matching pad insertion loss was too high? If it was somewhat lower, both the 10 KHz and 100 KHz experimental results might conform to the published result.

So, I measured the insertion loss of the impedance matching pad. This was a bit tricky, since the only device I have with connectors at 600 ohm output impedance is the HP11729C. Furthermore, I couldn't use an input signal that would be filtered by the 1.5 MHz LPF in front of the 1Hz-1MHz output.

To begin with, I purchased a 600 ohm BNC terminator, which arrived last weeked. Needing only some signal of sufficient power coming from the 1Hz-1MHz output, I did the following. I input a 1 MHz/200 mV signal from my DG1022 to the 5-1028 MHz input (one of the mixer inputs) of the HP11729C. I then input a second 1 MHz/200 mV signal phase shifted by 90 degrees into the Microwave test signal input (the other mixer input). I connected the 1Hz-1MHz output to my scope using a 3' coax (which at 1 MHz should not have transmission line characteristics), first through a BNC-T terminated with the 600 ohm terminator and then through the impedance matching pad to the BNC-T terminated with a 50 ohm terminator.

Figure 5 shows the result without the impedance matching pad and Figure 6 shows the result with the pad.

Figure 5

Figure 6

The (rough) peak-to-peak voltages are 9 mV or -36.9 dBm and 1.1 mV or -55.19 dBm. This is a difference of 18.29 dB. The measurements on my scope (a Rigol 1104Z) using a crude cursor set up at the lower limit of the scope's voltage range are not definitive. Nevertheless, it doesn't seem like the insertion loss is less than 16.6 dB. So, this eliminated the hypothesis I was considering.

I then considered the possibility that the phase noise data published on John Miles's website was averaged over a significant interval of time. This seems reasonable, since the FE-5650A is intended as a component in a time-keeping device.

The sweep interval selected by my SA for 10KHz-200KHz at 10 Hz RBW was about 17 seconds. So, I decided to lengthen this interval to see if that brought the experimental data into line with the published results.

Unfortunately, the Siglent SSA3032X would not let me increase the sweep interval when the RBW is 10 Hz. I had to increase RBW to 1 KHz to execute this experiment.

Figure 7 shows the results from a 6 second sweep, whereas Figure 8 shows the results from a 300 second sweep.

Figure 7

Figure 8

A cursory examination of the plots shows that increasing the sweep interval actually increased the measured phase noise. For example, the (raw and uncorrected) value for 10 KHz from the 6 second sweep is ~ -49 dBm, whereas the (raw and uncorrected) value for 10 KHz from the 300 second sweep is ~ -44 dBm. In other words measured phase noise gets worse for longer averging times.

At this point, I decided to present the results I have so far obtained and ask for comments. As things stand now, I can think of 4 possibilities for the discrepancies between the experimental results I obtained and those published on John Miles's website:

• There is a problem with my measurement methodolgoy or its execution.

• There is a problem with the published results.

• The FE-5680A and FE-5650A are not identical except for packaging. The published results do not apply to the FE-5650A.

• The published results are for a freshly minted FE-5680A, whereas my experimental results are for a 15 year-old FE-5650A. Aging has deteriorated the phase noise performance of the latter.

I would be interested in comments addressing these possible explanations or other explanations for the discrepencies between the results I have obtained and the published results.

[KE5FX]

Maybe try a phase-locked measurement instead of a frequency discriminator measurement.  The calibration process for that should rule out any gain/loss problems in your test signal path. 

[dnessett]

Maybe try a phase-locked measurement instead of a frequency discriminator measurement.  The calibration process for that should rule out any gain/loss problems in your test signal path.

Thanks, John, for the suggestion. However, there are some problems with that approach.

First, the phase-locked method requires an tunable reference oscillator to keep it and the DUT in quadrature. One of the advantages of the frequency discriminator setup is once you obtain quadrature (by adjusting the delay complex), the inputs stay in quadrature. I have thought about this and have bought a couple of BLILEY 10 MHz sine wave OCXOs that can be adjusted in the vicinity of +/- 4 Hz. Somewhere down the road I plan on trying this out, but I have to think through the interface between the HP11729C and the adjustable OCXO. Of course, I could buy a used HP8662A with option 3 tunable source. However, on ebay, the cheapest used HP8662A with option 3 I could find goes for $2,700. Right now that is outside my hobby budget.

Second, at present I am limited to phase noise measurements greater than 10 KHz because that is the lower limit of my SA. So, I am trying to figure out a way to implement a low frequency recording spectrum analyzer and investing a lot of my time on that. I can't say I am near solving this problem, but I don't want to go through all the learning it would take implement the phase-locked approach at the present by putting the low frequency spectrum analyzer on the back burner. I intend to do this sometime in the future, but right now I want to figure out how to measure phase noise less than 10 KHz using the frequency discriminator approach.

Third, I am new to phase noise measurement and one of the attractions of this project is to learn. If I just give up on the frequency discriminator approach without understanding what is going on, I have failed to meet this objective. In addition, while I won't say I never give up, when I do it leaves a bad taste in my mouth.

Dan

[dnessett]

I have been searching for a low frequency recording spectrum analyser in order to measure the phase noise spectrum of various hobbiest oscillators. This has taken a very long time and significant effort, but eventually paid off. Here I describe the unsuccesful part of my search in order to document some approaches that do not work. I have started a new thread to describe the approach that was successful. I use a different thread for this because a low frequency recording spectrum analyzer is an instrument that has many uses, not just the measurement of oscillator phase noise.

I needed an SA that measures signal spectra from 1 Hz to at least 200 KHz. I have a Siglent SSA3021X hacked to SSA3032X. However, this SA is limited to frequencies above 9KHz. My first attempt at creating a low frequence SA involved a Spyverter upconverter that shifted low frequencies to a band in the 120 - 180 MHz range. The marketing blather for the Spyverter indicated that it shifted "almost DC" to 60 MHz into this range. My plan was to upconvert the low frequencies and then analyze the spectra on my Siglent.

This did not work because the marketing information was (like most marketing drivel) dead wrong. Below about 1 KHz the Spyverter introduced significant insertion loss, which meant I would have to correct the value for each frequency bin in order to obtain correct spectrum data. (see this post). To be fair, the technical information I could find on the Spyverter (there wasn't much) indicated a lower bound of 1 KHz. While it would be possible to calibrate the Spyverter and correct the spectrum accordingly, it is likely the correction factior would depend not only on the fourier frequency, but also on the total signal power. Furthermore, there is no evidence that two Spyverters would generate the same correction data. So, I abandoned this approach.

My next attempt was to see if there was a low frequency spectrum analyzer on Ebay. The best deal was a used HP3580A, which typically costs about $750. However, this scope is not a recording instrument, so I couldn't postprocess the data I measured (which was a requirement for my application). Furthermore, it measures only 5Hz to 50 KHz. I needed an instrument that was capable of computing spectra from 1Hz - 200 KHz. So, I gave up on used instruments from Ebay.

I then read a post that suggested PicoScope USB products had a good FFT spectrum analyzer built in. After investigating its capabilities, I bought a PicoScope 4262 (I originally purchased a 4224, but its 12-bit ADC was insufficient, so I returned it and got the 4262, which is 16-bit). This approach worked, but there are some problems that require workarounds. The low frequency recording spectrum analyzer post describes them. The PicoScope 4262 costs more than I originally budgeted for a low frequency recording spectrum analyzer (it lists at $1235). Fortunately, TEquipment had one on sale for $915.51 (I got the last one).

Now that the capture of low frequency spectra is solved, the next step in using the frequency discriminator configuration of the HP11729C is calculating its noise floor. I will make the necessary measurements and calculations and report the results. I also need to change some of the post-processing calculations, as the original ones assumed the use of an analog spectrum analyzer, not an FFT-based instrument. I will publish the modified calculations after figuring out how to obtain the noise floor information.

[dnessett]

I have finally obtained the last piece of infrastructure necessary to analyze the phase noise of hobbiest oscillators. To ensure that phase noise measurements with the frequency discriminator configuration of the HP11729C are not simply displaying the phase noise of this instrument, I need to measure its noise floor. The approach I decided to take was to obtain a low phase noise oscillator and apply it to the frequency discriminator. The result would either be the phase noise floor of the oscillator or the phase noise floor of the frequency discriminator (or some piecewise composition of both). In any case, the result would be an upper bound on the phase noise of the frequency discriminator. As long as phase noise measurements stay above this floor, they should be accurate.

My search for an ultra low phase-noise oscillator took longer than I expected. I first tried to purchase a new OX-204 oscillator from Microsemi. This was a frustrating exercise, since they do not provide this product through a distributer. So, I had to contact the manufacturer and after several weeks of email conversations, it became apparent they were not set up to deal with sales to individuals.

So, I looked around for a used ultra-low phase noise oscillator. I found one - the Morion MV89. It had at least two advantages over the OX-204. First, its phase noise specs were significantlly better (those given below are for 5 MHz, but the literature leads one to believe that the 10 MHz version has similar characteristics):

Frequency Offset	dBc/Hz
1 Hz	-105
10 Hz	-130
100 Hz	-145
1 Khz	-150
10 KHz	-155

Table 1 - Phase noise of the Morion MV89

This is significantly better than the advertised phase noise of a new FE-5650, which is:

Frequency Offset	dBc/Hz
10 Hz	-100
100 Hz	-125
1 Khz	-145

Table 2 - Phase noise of the FE-5650

The GPSDO I intend to test has no published specs, so it is impossible to say whether the Morion MV89 specs are better or worse.

The second advantage of the MV89 is cost. The OX-204 has a new cost of $480 (in the configuration I attempted to buy). The used MV89 cost me $40 on ebay.

One disadvantage of the MV89 is its reputation for poor manufacturing quality control. A number of purchasers have reported its failure after ~30 days. However, they have also reported that if a unit last longer than this interval, it generally is reliable. Since the device costs so little, I bought 3 of them to ensure I had at least one that I could rely on.

Each unit I purchased is about 14 years old. So, it is likely they will not deliver their advertised new phase noise specifications. However, as long as the noise floor produced using them in the frequency discriminator configuration of the HP11729C is lower than the measured phase noise of the test units, they will have done their job.

One advantage of the MV89 is it has an electronically controlled frequency adjust input. The range of carrier frequencies produced by this adjustment is 10Mhz +/- 4 Hz. This means I should be able to use them to configure the HP11729C as a phase detector instrument. One problem is the frequency adjust voltage of the HP11729C is +/-5V, whereas the frequency adjust voltage that the MV89 expects is +/- 2.5V. I will have to design a simple resistor divider circuit to bring these two into compliance.

The next step is to obtain a noise floor measurement of the frequency discriminator configuration of the HP11729C and then start testing 10 MHz oscillators.

[jpb]

I have an MV89 I got from China years ago and it seems to work fine (though I don't have anything that will measure its phase noise though.

The main issue with them I seem to remember for Time Nuts postings is that the 10MHz devices are frequency doubled 5MHz devices:

https://www.mail-archive.com/time-nuts@febo.com/msg58269.html

The reliability issue is to do with a capacitor going bad but that shows up as a low level output I think.

It will be interesting to see what your measurements show.

[dnessett]

I have an MV89 I got from China years ago and it seems to work fine (though I don't have anything that will measure its phase noise though.

The main issue with them I seem to remember for Time Nuts postings is that the 10MHz devices are frequency doubled 5MHz devices:

https://www.mail-archive.com/time-nuts@febo.com/msg58269.html

The reliability issue is to do with a capacitor going bad but that shows up as a low level output I think.

It will be interesting to see what your measurements show.

Thanks for the info, jpb.

Given your experience with the MV89, I have a question. When I ran one of the MV89s for an hour or so, I noticed it became quite hot. I am still able to pick it up and hold it my hand, but it is on the borderline of that. When you were working with yours, did you have a heatsink on it? If so, how did you attach it (as there are no screw holes for this purpose on the top)?

[FriedLogic]

The reason that the case of an MV89 is quite hot is that it has a large outer oven very close to the case. There's a picture of one opened up with the outer oven top removed on this page:
http://www.rbarrios.com/projects/MV89A/

So definitely no heatsink required.

The temperature control on an oven oscillator can fail, and then it can get really hot. That's one of the reasons that you have to be careful if you ever cover them.

[dnessett]

I have an MV89 I got from China years ago and it seems to work fine (though I don't have anything that will measure its phase noise though.

The main issue with them I seem to remember for Time Nuts postings is that the 10MHz devices are frequency doubled 5MHz devices:

https://www.mail-archive.com/time-nuts@febo.com/msg58269.html

The reliability issue is to do with a capacitor going bad but that shows up as a low level output I think.

It will be interesting to see what your measurements show.

The reason that the case of an MV89 is quite hot is that it has a large outer oven very close to the case. There's a picture of one opened up with the outer oven top removed on this page:
http://www.rbarrios.com/projects/MV89A/

So definitely no heatsink required.

The temperature control on an oven oscillator can fail, and then it can get really hot. That's one of the reasons that you have to be careful if you ever cover them.

Since both of you seem to have experience with the MV89, I have a question about its operation. I have been beating my head against a wall for the past 2 weeks trying to get the adjust function to work. I have designed 3 circuits of increasing sophistication that translate -/+ 10 V to 0-5 V and then connected the output of these circuits to the adjust pin of the oscillator. With the latest circuit, I could only get the voltage on the adjust pin to move from 5 volts to about 4.1 volts.

My frustration then led me to do what I should have done in the first place, I used a resistor between the ref pin and the adjust pin to see if I could get the voltage to swing between 0 and 5 volts. I first started with 10 Kohm, then increased to 100 Kohm, 1 Mohm and then 10 Mohm. As the resistance increased, the available swing between the two pins increased.

Finally, I just tested the voltage with a short between the adjust and ref pins and leaving the ref pin open (effectively, infinite resistance). In the latter case, the adjust pin would only go down to ~2.5v. With the pins shorted, the voltage at the adjust pin is 5v.

The spec states that the adjust pin should be capable of a swing of 0-5v. But, I can't figure out how to get it below 2.5v. The information on frequency adjust is almost nil.

Even when I bring the adjust pin up to 5v, I see no effect on the output frequency. When I first turn on the oscillator, my frequency counter shows 9,999,995Hz or thereabouts. As the oscillator warms up it reaches 9,999,999.5Hz. So, my counter is able to see changes in Hz. (It is a very old discrete transistor piece of equipment built by a company called Kay Elemetrics Corp, called a Count-a-Marker, model 8323A. The date on the drawings is 2-6-1970, so we are taking about something almost 50 years old).

Even if the counter isn't accurate in regards to the absolute frequency, it should show changes as I change the voltage on the adjust pin. But, so far I cannot see any change whatsoever.

Any help you or others can provide on how to use the adjust pin would be greatly appreciated.

[FriedLogic]

Hi,

The frequency control pin  - 'Uin' in the data sheet - on the MV89A the I have is biased to half the reference voltage 'Uref'. In this case it's 4.96V for Uref and 2.48V on the Uin pin if nothing is connected to it. 

Connecting Uin to 0V with a 1K resistor pulled the voltage on it down to 64mV (so 64uA current), and connecting it to Uref with the 1K takes it up to Uref less 63mV. Like most frequency control inputs on oscillators, it's quite high resistance so is not hard to drive.

The frequency change was -3.53Hz and +3.47Hz. Data sheet spec is >+/-2.5Hz, so most counters should see it fine.

[dnessett]

Hi,

The frequency control pin  - 'Uin' in the data sheet - on the MV89A the I have is biased to half the reference voltage 'Uref'. In this case it's 4.96V for Uref and 2.48V on the Uin pin if nothing is connected to it. 

Connecting Uin to 0V with a 1K resistor pulled the voltage on it down to 64mV (so 64uA current), and connecting it to Uref with the 1K takes it up to Uref less 63mV. Like most frequency control inputs on oscillators, it's quite high resistance so is not hard to drive.

The frequency change was -3.53Hz and +3.47Hz. Data sheet spec is >+/-2.5Hz, so most counters should see it fine.

Sorry about the late reply. I have notifications turned on, but for some reason did not receive an email when your post appeared.

I will respond in a day or two. I have had a family medical emergency that I am dealing with right now.

[dnessett]

[snip]

Connecting Uin to 0V with a 1K resistor pulled the voltage on it down to 64mV (so 64uA current), and connecting it to Uref with the 1K takes it up to Uref less 63mV. Like most frequency control inputs on oscillators, it's quite high resistance so is not hard to drive.

The frequency change was -3.53Hz and +3.47Hz. Data sheet spec is >+/-2.5Hz, so most counters should see it fine.

I have had a bit of time to work on experimenting with the Adjust pin of the MV89. Here are the results.

I measured the output impedance of the Ref Pin and the input impedance of the Adjust pin. Figure 1a shows the test setup for the former.

Figure 1

I first measured the voltage when R_out_test was open and then for the values 200 ohms and 400 ohms. The open voltage was 5.09V. The output impedance of the Reference pin is then:

Output Impedance = ((5.09/V_out_test)-1)*R_out_test

Here are the results:

R_out_test	V_out_test	Output Impedance
200.55	3.35	104.16
399.9	4.02	106.44

So, it appears the Output impedance of the Reference Pin is approximately 100 ohms.

Figure 1b shows the test setup for measuring the input impedance of the Adjust pin. Varying R_Input_Test and measuring the voltages V1 and V2 gives the current flowing through R_Input_Test, which then is used to estimate the input impedance of the Adjust pin.

I_in=(V2-V1)/R_Input_Test

Z_in=V1/I_in

I performed the test for three different values of R_Input_Test. Here are the results:

R_Input_Test	V2	V1	Iin	Zin
1.0001K	3.0095	2.9945	15uA	199633
10.255K	3.0095	2.9479	6.024uA	489320
100.39K	3.0095	2.7225	2.856uA	952305

This suggests the Adjust pin is approximately a current sink without a fixed input impedance.

I then tried the suggestion by FriedLogic of connecting a 1K resistor to a voltage source and connecting the output of the resistor to the Adjust pin. Here are the results:

Voltage Source	Frequency
5V	9999999.61
2.5V	9999999.58
0V	9999999.53

Unless I am reading the frequency counter incorrectly, I am only getting a fraction of a Hz variation in the output frequency for the full range of the Adjust pin specified input.

Anyone have an idea what might be happening (including an operator error on my part)?

[FriedLogic]

The generic data sheet specification is that for a 0-5V change in the control voltage you should get at least 5Hz change in the output frequency. You're not getting close to that, so the oscillator looks like it might be faulty. It would also be worth checking the connections and power supply, just in case.

The frequency control input on my one looks like the 2.48V on it is a voltage divider made up from two resistors of around 75K to divide the reference voltage by 2. Maybe somebody has opened up the ovens and knows what the circuit actually is.

[SoundTech-LG]

These MV89s seem of poor reliability in general. While they do perform well within spec, the failure rate seems high. I have one that I burned in for a couple of years. I then installed it, and within a few weeks it died. It seems the heater is no longer working. Maybe I'll heat it up with a blow torch and pop the guts out...

[dnessett]

Since others were getting frequency variation when the adjust pin was set to voltages between 0 and 5V and I was not and since this was true for all three MV89s that I had purchased, I began to suspect my frequency counter was faulty. Therefore I bought a used HP5335A from ebay. This was the quickest delivery of an ebay item that I have ever experienced. I ordered it on Feb 12 and it arrived in the evening of Feb 13. This certainly had something to do with the fact that the seller was located only 50 miles from my residence, but still I was completely amazed that purchase through shipment preparation and then shipment transport (by Fedex) took only 2 days.

Anyway, varying the voltage on the adjust pin of each MV89 yielded the following data on the HP 5335A:

Table 1 - MV89 - 1

Adjust Pin Voltage	Frequency
open	10,000,151
0V	10,000,147
2.5V	10,000,151
5V	10,000,154

Table 2 - MV89 - 2

Adjust Pin Voltage	Frequency
open	10,000,153
0V	10,000,149
2.5V	10,000,153
5V	10,000,156

Table 3 - MV89 - 3

Adjust Pin Voltage	Frequency
open	10,000,153
0V	10,000,150
2.5V	10,000,153
5V	10,000,157

This confirms that there is nothing wrong with the MV89s I bought. Frequency variation is approximately 7 Hz for each (I have left off the fractional hertz part of the measurement). This variation agrees with what FriedLogic specified in his post (-3.53 to +3.47Hz).

Now I have to figure out why the HP5335A is measuring ~150Hz greater than 10MHz for each oscillator. The fact that each oscillator is showing this suggests that the frequency counter is out of calibration. I confirmed this by measuring the output of an ebay 10 MHz GPSDO and observed its frequency to be 10,000,151 Hz.

The HP5335A was advertized to have option 10, which is the OCXO equipped version, but I haven't looked inside yet to confirm this. I also looked at the manual and could not find a way to trim the frequency of the instrument's internal frequency standard to correct this apparent anomaly. However, I can always use an external oscillator and bypass the internal oscillator.

Now that it is confirmed the MV89s can be adjusted properly, the next step is to build enclosures for each of them that takes the frequency adjust signal from the HP11729C and adjusts the frequency of the oscillator so I can use the HP11729C in a phase detector configuration.

[FriedLogic]

When I first turn on the oscillator, my frequency counter shows 9,999,995Hz or thereabouts. As the oscillator warms up it reaches 9,999,999.5Hz.

Oops, I didn't pay enough attention to what you said there. The frequency change during warm up should be in the hundreds of hertz (it's 310Hz low when cold at 20°C ambient on my one), so that was a giveaway.

The HP5335A frequency error sounds like a fault rather than a calibration issue.
As you mentioned, the first thing to try would be an external reference.
I think that the HP5335A normally used a 10811 OCXO, which has a hole on the top for the frequency adjustment trimmer capacitor in it, but this will not adjust it by 150Hz.
The 10811 that I have is about 210Hz low when cold at 20°C ambient, so if your one is 150Hz low it might be worth checking if the oven is heating up - if it's there at all!