Friday, September 2, 2016

Collins T-368 T-195 VFO to solid state

This started when I read a user's group post by Rick Campbell KK7B about these Collins VFOs that were at least at one time easily available from hamfests at low prices. And furthermore, they could be converted from tube operation to solid state without much difficulty. The advantages are low drift and low phase noise, among other things.
So I started looking around and found one at my local hamfest for $10. The total package is actually an exciter. The VFO covers 1.5 to 3.0 MHz and the chassis includes multipliers to give x2, x4, and x8 ranges. You can see why it needs to be stable if you're going to multiply by up to eight times.
I figured out how to energize just the oscillator section (two tubes) B+ and filaments and had some fun playing with it and measuring the drift. Then into the attic with it for several years, with conversion to solid state a back of the mind notion for "some day", which just arrived this week.
The magic is in that white soup can. It houses the frequency determining circuits. Between it and the front panel you see two tube sockets for the oscillator and buffer.
I found Army manuals for the thing on line and other resources in the form of a web page by John Seboldt K0JD giving lots of good information on doing just what I want to do:

That led to a good article on the subject "Transistorizing Surplus VFOs", QST February 1989, p 45.
The QST article used two dual-gate MOSFETs. K0JD used a JFET oscillator and DG MOSFET buffer. I fooled with that in LTspice but didn't like it so I used a JFET for the oscillator and a BJT (NPN) for the buffer as an emitter follower to give me a 50 ohm output and fairly decent looking sine wave.
It was essentially a Hartley I had built earlier on an SMT board with little SOT-23 active devices, SMT versions of the 2N4416A and 2N3904. I reused that board design although it's much larger than it needs to be since the frequency determining components aren't required.
As seen in the photo, mine is a temporary lash-up although I don't know if I'll ever do it up right. I didn't want to go through the mechanical details of removing the oscillator to get to the bottom of the tube sockets, so I just stuck wires into the correct pins from the top. It worked out well that the nodes I needed to access are accessible on the tube pins.
After getting it all hooked up I've done some checking on stability. It has gone two hours without moving a single Hertz and at other times might move one or two Hertz in a 30 minute period. Actually, I'm not sure my frequency counter is stable enough for this measurement.
I want to mention linearity too, but first I'll put in my schematic. Note that I didn't do any connections to the buffer's tube socket. And to the oscillator I just connected to the top of the tank, the coil tap, and ground.

I'm not an expert on military radio equipment, but with some of the stuff I've looked at it appears that no expense was spared in design or construction. Just first class. 
This unit has a mechanical counter to indicate the frequency. I took readings at intervals of 100 from 1500 to 3000 and plotted them in Excel. The linearity is very impressive although with the vertical scale of the plot it appears a little better than actual. It has to deviate by several kHz from the straight line before you can see it. There's also an offset of about 110 kHz which I can fix after I figure out how to disengage the shaft from the counter and reset it.

That white covering over the soup can is a heater. I thought I might turn it on and get even better stability, but it turns out that the setpoint is 32F! So you can surmise that the stability results from careful selection of components with regard to their temperature coefficients and the heater is just to keep the unit within the range that they can handle.
One drawback is the frequency range of 1.5 to 3.0 MHz. You've got 160, but double it and you only add 80. You'd need to double again for 40. Or heterodyning would be another approach.
If I were going to use this thing "for real", I'd probably add another stage of amplification to get to +7 dBm or more. 
Note that K0JD added a varactor offset tuning circuit. That could be necessary in some applications, but I didn't want to adversely affect stability at this point in my playing.
I haven't considered trying to make any of the multiplier stages functional with solid state components thus far.

OK, fun project.  If you see one, pick it up and have some fun playing with it.

Nick, WA5BDU

Saturday, July 23, 2016

Yet another Arduino and AD98xx DDS Project

These days Arduino Nano boards are as low as $3 each and AD9850 or AD9851 DDS modules can be found in the $5 to $15 range, so the temptation to do something with them is strong. The question is, what is that "something"?

Like most homebrewers, I have DDS units coming out my ears. But I could think of a few specific things I'd like to have in this implementation:

1) Battery powered to make it portable and easy to grab and use with a minimum of fuss.

2) Sealed up as tightly as possible. A current DDS unit I have leaks so much signal out via the power and USB cables and slipshod case that I can't hope to attenuate the output way down for small signal receiver testing.

3) A minimal user interface - no display, PC connection or rotary encoder. This is partially in the interest of compactness, partly to minimize signal leakage and partly just for the challenge inherent in designing such a beast.

4) Lots of bells and whistles as long as they're consistent with #3 above. One model is my Elecraft XG3 signal generator, which has some nice bits above and beyond its basic functions.

First a little about the hardware. Jim Giammanco N5IB was really the guy who kicked me off on this project with his DDS and Arduino Nano Experimenter's Board. It's a very nice board designed to handle the interconnections, power distribution and filtering for an eBay DDS module with an Arduino Nano in a stacked configuration. Unused pins are brought out for the programmer to play with.

Info on the Experimenter's Board is found on the PHSNA Yahoo Group site and also on Jim's page linked here:

User interface

The user has to be able to communicate with the system and vice-versa. I went with three little SMT pushbuttons for user input and with a miniature speaker for the system to talk back via Morse code. So the ham who is fluent in Morse is at a bit of an advantage with a UI of this nature.

In its simplest, top level operation, the user taps switch S1 to have the frequency announced and S2/S3 to step the frequency Up/Down.

But we want to be able to do a lot more so a menuing system is needed. I went with a system I'd seen in Steve Weber's ATS-3 transceiver and in the NORCAL / Dan Tayloe Stinger Singer frequency counter. The user holds S1 down and the menu options are played in a loop, a single letter for each option. The switch is released after the desired item is heard.

Features & functions

  • Tap a button to have the frequency announced in Morse
  • Tap Up/Down buttons to step the frequency. Hold for rapid stepping.
  • Change frequency step size, 1 Hz to 1 MHz in 10x increments
  • Change bands. Step through ham bands 160, 80, 40, 30, 20, 17, 15, 12 and 6 meters.
  • RF On/OFF - can turn off RF output without powering down the unit
  • Send CW - a test CW message is sent repetitively via the RF output
  • Send RTTY - a test RTTY message at 60 WPM, 170 Hz shift sent repetitively
  • Save current frequency to scratchpad EEPROM memory
  • Return to frequency stored in scratchpad EEPROM memory
  • Announce power supply or battery voltage in Morse (audio)
  • Update (save) current frequency / band & step to EEPROM so subsequent start-ups will start there.
There's also continuous battery monitoring and a low battery alarm if it drops below a threshold.

Power supply

The experimenter's board has provision for an LM7805 regulator and the Nano board can also accept 12 V supply voltage and use it's on-board regulator. But after exploring various battery options I decided to go with six (6) NiMH AA cells, which gives me a supply range of about 6.3 V to 8.4 V.  I definitely wanted batteries because opening and closing the box with wires running everywhere isn't fun.

I decided to go with a buck switching module with 5 Volts output as can be found on eBay for a dollar or two apiece. This one is the size of a postage stamp and is about 89% efficient. It doesn't drop out until the input reaches < 5.5 V. So I'm operating both the DDS and the Nano board from 5 VDC. I was surprised that my current draw was only 100 mA with this system, so I can get about 18 hours of operation from my 2000 ma-hr batteries. I put a 2.1 mm charging jack on the back of the box.

Here's the regulator:

And here's the box with the Arduino / DDS stack on one side and the batteries and regulator on the other:

I wanted to use a die-cast box but didn't have one the right size so I used a Radio Shack aluminum box that is reasonably tight and overlaps on most of the edges.

Trying it

One of the first things I wanted to do was to box it up, put on a 50 ohm terminator and see if I could hear it in my K3 with my big ham antennas connected. I could not hear it at all on 40 or 20 meters. On 6 meters I hear it faintly, but it doesn't move the S-meter. Of course, plugging two or three feet of wire into the BNC makes it loud and clear in the receiver. I'm not sure if I'll be able to attenuate it down to 1 uV with external attenuators or not. We'll see. But S9 shouldn't be a problem.

Some specifications

The output is a sine wave of about -7 dBm into 50 ohms which of course varies a couple dB over the wide frequency range.

The frequency range is 1.8 to 54 Mhz. Actually, it should go to 60 MHz and I found on the lower end I could go to 100 kHz before output dropped too much. A larger coupling capacitor should help those interested in going all the way down to audio. Oh, there's a transformer too, so more mods would be needed to emphasize audio frequencies.

My unit uses an AD9851 module. It can be built with an AD9850 module which would put the top end somewhere above 30 MHz. BTW, I followed the PHSNA hardware guidance and replaced the filter on the DDS module, said to be inferior, with an external one.

What's good, what's missing?

I said I used the Elecraft XG3 as a model. A big advantage it has is internal attenuators giving you four selectable output levels of -107, -73, -33 and 0 dBm. Switchable internal attenuators was more than I wanted to take on at this time.

But the XG3 can't tell you what frequency it's on.  It has band indicating LEDs but you have to jot down actual frequencies or use the USB to PC interface and software to be reminded of what they are. Also, it does not allow adjusting the frequency in steps.

The XG3 has a square wave output. It's credited for giving harmonic marker signals up into the GHz range. But I didn't really like the square waves for stuff I'm doing.

Where's the source code, schematics and other good info?

I put the source code and a "user's manual" kind of document in the files area of the PHSNA group.

And Jim's PDF describing the experimenter's board is found both there and on his site, linked earlier.

For those who don't want to sign up for the PHSNA group, I'll be glad to email the files or make them available somewhere else.

That's it.

Nick, WA5BDU

Sunday, June 12, 2016

Honda eu2000i generator waveform

I was testing out the generator two weeks before Field Day so I decided to do something I've wanted to do, which is look at the waveform on my oscilloscope. Is it close to a sine wave?

I approached this with a little anxiety because, what if I accidentally hook the scope probe's ground lead to the hot side of the line?  So I took some precautions.  First, I connected the generator's ground terminal to the ground wire at my meter. Inside the house, I plugged an extension cord into the wall and verified that I knew which side is the "hot" side on the other end. I verified 120 VAC from that side to system ground (the case of a piece of equipment with 3-wire plug). I looked at the utility's waveform on my scope and it looked OK. I didn't need to hook up the grounded side of the probe, since all grounds are common.

Next I unplugged the cord from the wall and took it outside and plugged it into my Honda eu2000i generator. I also plugged in a 60 W lamp, for just a bit of load.

I came back inside and re-checked the voltage on the hot lead to ground with my DMM. Here's where I got a surprise. Instead of 120 VAC, I read 60 VAC. So I moved the probe to the neutral lead and it's also 60 VAC.  What's going on?  It looks like the ground terminal on the generator is at the center point of the output, not on one side. Unless I'm missing something.

Here's what the waveform looked like:

Pretty good!  It actually looked a bit more sine-like than what was coming out of the wall.


Nick, WA5BDU

Saturday, July 18, 2015

Si570 revisited - flexible Arduino controller

Back in 2009, I did a blog post about my then new Si570 synthesizer board being controlled by my then new Arduino MCU board.  I got a surprising number of requests for my source code, which was a little embarrassing because my programs mainly sent pre-calculated register values to the Si570 to program discrete frequencies.  At best, one program used the "ratio" method to tune plus or minus 30 kHz or so from the programmed center frequency, as is allowed by the chip design.

I wanted to do an "any frequency, right now" control scheme, but doing that on the Si570 is much more difficult than it is on the AD9850 family of DDS synthesizers.  The Si570 uses one 3-bit divider register, one 7-bit divider register  and one 38-bit divider register plus the crystal frequency to program its output.  And it's not deterministic -- multiple sets of register values will work and many others will not work, in accordance with rules defined in the data sheet.  Whew!  It's fairly simple to crank out a set of registers for a given frequency "off line" and send them to the chip.  It's quite another thing do to it in real time as fast as you normally turn a rotary encoder knob.

A new Arduino Si570 control program

The best way to approach a programming problem is to find someone who has solved it before.  Craig Johnson AA0ZZ did a very nice Si570 control program for the PIC16F88 MCU and described it in a July/August 2011 QEX article. His is all done in assembly language. The methodology is very important to the solution.  Craig found sets of two of the three registers that would work in each of 24 "bands" from 10 MHz to 157 MHz and saved them in tables.  The actual crystal reference frequency (see "Calibration") was used with those register values to  create another band table of pre-calculated constants.  Now, much of the heavy number crunching needed for a frequency change is already done, although there is plenty of 64 bit integer math left that must be done efficiently for each frequency change.

Anyway, using Craig's article, the data sheet, and my HP48G calculator, I was able to eventually grasp all the nuances and program an Arduino Nano to do essentially the same task: Take a frequency as input, generate the six registers required and send them to the Si570 board via the I2C bus.

My program right now

I decided to put on the brakes at the point where I have all the Si570 control routines working, but before I began to "personalize" the program with menus, LCD display, rotary encoder control and so on.  I'll add those things and publish the source shortly (?), but I want to put out the basic "kernel" of functions right now for readers who are programmers and want to add their own bells and whistles and user interface schemes.

Right now the program is functioning in a "demo mode".  It does a calibration, then goes to 14.025 MHz and tunes up and down in ten 10 Hz steps forever.  That's so a user can hear that it is working (changing frequency) without having to tune the receiver to follow it.

Routines in place are the all important one that accepts a frequency to 1 Hz resolution, calculates the registers, and send them to the Si570.  Also included are step size functions and step-up and step-down functions.

Invisible to the user but important are functions to detect band change (so the correct values will be selected from the band data tables) and to detect movement more than 3500 PPM plus or minus the last center frequency.  When this happens, the Si570 requires a "freeze" operation to load a new "center" frequency.  This is all automated in the software.

Also in the program are some serial routines to allow it to talk to the PC via the Serial Monitor in the Arduino IDE.  That's mostly for troubleshooting and development so you could delete all the serial stuff and save some memory.

Currently, the calibrate() routine sends a bunch of information to the user over the serial link.  After uploading the program, press control-shift-M to open the serial monitor and see start-up registers, crystal frequency and so on.


This is not like the calibration you are accustomed to with AD98xx synthesizers where you calibrate to an external standard. Silicon Labs custom calibrated your Si570 at the factory and stored values in all the registers to cause it to start up on the frequency specified at purchase time.  This is to correct for expected small variations in the 114.285 MHz internal crystal reference.

So when the Arduino does its calibration, it is expected to "know" the specified start-up frequency.  Then it downloads the register values programmed into non-volatile memory by Silicon Labs and uses those to "back calculate" the actual value of the reference crystal, which is then used to generate the constants in the band tables.

My software currently calibrates every time it starts up.  Craig's software does it if you hold in a button when powering the PIC up. I can see why you might want it to be "dealer's choice" so I will make it an option in my full featured version.

Some hardware notes

Connection between the Arduino and Si570 is pretty simple - two wires plus ground.  But it can't be quite that simple if you have a 3.3 V Si570 and 5 V Arduino.  You need bi-directional logic level shifters between the two. I used two BS170 MOSFETs plus two resistors as Craig showed in his article.  You can get the same on a tiny board from eBay if you want to go that way.

Another hardware issue I encountered was unexpected - RFI.  When I first tried the program with the hardware, I got a lot of errors in I2C transmissions.  When I plugged the RF output from the Si570 into a power meter with 50 ohm input, most of it went away.  So it seems that reflections from the output can cause problems - terminate your RF output!  I also added a 4.7 uF tantalum capacitor from my 3.3 volt reference on the MOSFET shifters to ground. That may have helped a bit.

I'm getting out about +14 dBm, BTW.  That's a lot of RF and I'm glad to have it.

Speaking of hardware issues, the Si570 is a pretty clean RF source with low phase noise.  Generally better than most AD98xx units and better than the new Si5351 part, although I have and use both of those.

Another hardware caveat:  I've been burned a couple of times by inaccurate Arduino Nano documentation.  This time I spent half a day troubleshooting I2C communications before finding that SDA and SCL are actually on pins A4 and A5, not D4 and D5 as shown on some drawings.

What about that startup frequency?  That's important to the calibration.  I lot of Si570 boards out there were bought with the Sotfrock project in mind.  In those circuits, the VFO operates at 4x the operating frequency and a lot of Si570s were purchased with a startup frequency of 14.080 MHz in mind, meaning they start up at 56.320 MHz.  Mine is one of those.  Others, possibly taking the default start-up frequency, may start up at 10.000 MHz.  If you don't know, power up your Si570 and check it with a frequency counter.  If it's other than 56.320 (or thereabouts), change this line in the source code to suit:

  #define STARTUP_FREQ 56320000UL

Finally, where do you get those Si570s mounted on a plug-in board?  I see several people (mostly hams) on the web offering them.  Mine was made by WA6UFQ. You might start your search from a Softrock site.

Regarding power the Si570 ... It uses more than 100 mA so don't use a 78L05 regulator and I wouldn't try to have the Arduino supply the power either. I used a 78M05 with a small heat sink.  it's rated at 500 mA. Why am I talking about 5 V regulators? My Si570 board has a 3.3 V regulator on it and wants 5 V to the board.

Let's wrap it up

Where's the source code?

Right here:

I won't be revising this file except to fix any errors that may be lurking in it.  Later I hope to add a link to a version with LCD, rotary encoder and so forth.

Here's a link to Craig Johnson's page, which includes the QEX article:

If you don't want to fight through creating an Arduino based controller, Craig's PIC based card works very well and allows LCD, rotary controller and so forth.  (I have one.)


Nick, WA5BDU

Wednesday, April 22, 2015

Q - steel wire at RF

Some discussions we had on the QRP-L list March 4 through 13th of 2015 on the subject of RF resistance of steel conductors merged in well with my recent attempts to improve my ability to measure coil Q.

In order to compare copper with steel and also Copperweld, I'd need to wind some coils that were physically identical and test them.  I have a ceramic form I picked up at Dayton last year, which has grooves for the conductor, assuring constant spacing and diameter.  I also need to have the same diameter and type (solid conductor) samples for a good comparison.

Below is my coil with ten turns on it, bright steel #18 bare wire:

And below is the same coil with #18 solid copper, insulated:

OK, I'll skip photos of the other two since they all look the same.

You see my shunt measurement "fixture" (BNC tee with receptacle on 3rd port) and a silver mica capacitor chosen to resonate in the vicinity of 7 MHz.

I had two kinds of iron or steel wire, one bright and shiny and the other dark in color.  The shiny one may have been galvanized but I doubt it.  I used the dark one as well because I'm sure it's not coated in any way.  However, it was a standard increment smaller in diameter.

I also wound one coil with #18 Copperweld.  I was glad to have that form because that stuff is super springy and fights you every step of the way.  I probably got this piece from Burstein-Applebee back in the 60s:

Wow, 250 feet for $1.89!  That's a lot of dipoles.


In the table below I list the results for each sample.  The coils took about 8 feet of wire, so I extrapolated the effective resistance out to 100 feet.  I also show the ratio of RF resistance for each sample to that of copper.

What should we make of this?  Well, I'd say iron or steel conductors aren't good for antennas (especially long ones) or inductors.  (Unless "broadband" is your thing.)  We probably already knew that.  

One thing that kicked off this study was data from a web site showing huge RF resistances for steel.  The calculations were based on skin effect and the effect of permeability in making the skin depth very thin.  I wondered if the permeability of these materials remained high even at RF.

Interpolating from the data on the web site, the ratio of steel to copper RF resistance at ~7 MHz is 33.3.  I got ratios of 13 and 15.  So, somewhat less but still pretty damning for steel conductors at RF.

What about Copperweld?  One alarming potential outcome of RF resistance being primarily determined by permeability was that the skin layer of Copperweld might be very thin as well, resulting in high RF resistance, despite the copper cladding.  This didn't turn out to be the case, with Qs almost equal to copper and RF resistance only 24% higher for Copperweld.

Another observation ... with the steel samples the inductances I measured with my AADE meter were 10% or so more than values calculated from the resonant frequency with a known capacitor.  No doubt this is due to the low measurement frequency of the AADE.

I was a bit surprised that the darker steel (iron) wire had about the same Q as the bright shiny wire did, despite having a smaller diameter.  Maybe the bright stuff was alloyed with something that raised its resistivity as happens with stainless.  I don't think this wire would be classified as stainless though.

Error Factors:

I don't know the composition of either of my steel or iron samples.  (I called the dark colored one "iron" just to differentiate the two.) Both strongly attracted a magnet.

I think my Q measurement methods are reasonably accurate these days.  But this isn't NBS work by any means - just the efforts of an amateur experimenter.

It would have been useful if I'd measured the DC resistance of the samples, but I wasn't geared up at the moment for accurate small resistance measurements.

Nick, WA5BDU


Wednesday, April 8, 2015

Q measurements with shunt fixture

In the previous post I discussed both shunt and series connected unknown circuits and formulas for each, but the results I showed were only for my 1-ohm fixture and series connected unknowns.

Here I'm showing some results for shunt connected unknowns.  BTW, in both cases the unknown is a series connected L-C circuit.

Recall that I discovered that my 1-ohm fixture had enough loop reactance (from inductance) to affect the measurement significantly, but in the series configuration it could be calibrated out.  For the shunt case it wasn't so simple. So the "fixture" for shunt measurements needed to minimize the loop created by connection of the unknown.  The simplest seemed to be just a BNC Tee with a BNC receptacle connected to the open end.

The two female ports go to the PHSNA DDS and the AD8307 measurement circuit.

I made some measurements with the type 2 and type 7 toroids again and with the molded choke, and I also added a couple of air wound B&W type coils - one small and one quite a bit larger.  In all cases I selected resonating silver mica capacitors to give resonance around 7.5 to 8.5 MHz.  Just for consistency and to be somewhat "midband HF".  I also added series resistances in several instances, to see if my measured Rx would increase by the amount I added, as a method of checking accuracy.

BTW, there are no transformers or minimum loss pads here, so this is a 50 ohm environment.

Below are some results from my spreadsheet.

If the data table doesn't fit the column, click it to make it full size.

I feel like I'm getting pretty decent results here, consistent with my expectations.  

BTW, elsewhere in my spreadsheet I enter my dBm measurement with the unknown port open and again with the unknown network connected to give me the attenuation caused by the unknown.  I also enter the environment resistance (50 ohms), and it gives me Rx, the series resistance of the network, presumed to belong mainly to the inductor.  

I also enter the measured inductance of the coil and the resonant frequency and my spreadsheet gives me the Q of the circuit.

NRK 4/8/2015

Wednesday, April 1, 2015

Q measurement with PHSNA

[Edit #1:  Some of my images (graphs & schematics) shrunk to fit the column width.  Just click on images that seem to be truncated to see them full sized.]

Measuring Q is an interesting challenge.  With the PHSNA (a DDS calibrated source plus AD8307 RF power measurement instrument plus analysis software), it seems like it should be a piece of cake.  I'm already doing it with crystals, after all.

But then I remember that Qs of 100 to 200 or more, even with toroids, aren't unusual, and inductors with reactances in the 50 to 200 ohm range are common in RF / HF designs. Do the math and you see that we may be measuring fractional ohm loss resistances.  A bit of a challenge in a 50 ohm system.

One way to increase sensitivity is to put the component being tested in a lower resistance fixture, for example, 12.5 Ω is typical for crystals.  Going to extremes, a 1 Ω fixture does a lot.  A problem is that two 50 to 1 Ω minimum loss pads back to back have a combined attenuation of 45.9 dB.  So we need a strong source or sensitive detector or both. My PHSNA system has about 0 dBm RF output and the detector should go down to say, -60 dBm without much trouble.  So I probably don't need an amplifier in line.

I also got into looking at Q measurement methods in EMRFD (page 7.36).  Typically, one puts a series resonant circuit for which Q is desired in series between the source and detector.  In this example, the test circuit is put in the shunt configuration between source and detector.  I wondered how that differs in sensitivity form the series method.  So the first thing I did was do the math to solve for loss resistance in both configurations.

I derived the formula in general terms for source (and detector) resistance Rs instead of just for 1 Ω.  Rx is the value of the loss in the coil (or coil + capacitor).  For attenuation A, I get:

The ultimate purpose of this exercise though is to find Rx.  So solving the above for Rx for the shunt configuration:

The procedure is to measure power to the detector without the tested circuit installed, then measure it again with it installed in shunt (to ground, between source and detector).  'A' is the dB difference in the two measurements and here is it is a negative value.

To compare this method with the series method, I need to derive the equations for that configuration.

Series configuration:

and solving for Rx:

I was somewhat surprised when this result didn't look like the equation I've been using for crystal Rloss measurements. I finally realized it was because my math began with attenuation as a negative number but the other equation entered it as positive.  So they are equivalent.
I put the equations into an Excel spreadsheet and plotted the attenuation A against Rx for both configurations.  My thought is that the method that produces the greatest change in A per ohm change in loss resistance is the more sensitive.

1 Ω test fixture
Rx on horizontal axis
dB attenuation on vertical axis

From the above, the series method seems better for "all around" measurements, but the shunt method looks like it would be better for high Q / low loss resistance items.  In a one ohm environment.

I repeated the above plot for a 12.5 ohm fixture and again for a 50 ohm fixture and present results below:

12.5 Ω test fixture
Rx on horizontal axis
dB attenuation on vertical axis

Above, the series configuration gives a fairly constant slope of about 0.3 dB per ohm.  The shunt configuration gives much better sensitivity, up to the source resistance of 12.5 Ω which is the crossover point again.

Finally for the 50 ohm fixture case:
50 Ω test fixture
Rx on horizontal axis
dB attenuation on vertical axis

(Above) Again the change per ohm with the series method is constant but it’s down to 0.08 dB/Ω.  With the series method, much more sensitivity is achieved with about 8 dB/Ω at 1 Ω, decreasing to 0.36 dB/Ω at 15 Ω.  It's clear that with higher source / detector resistances, the shunt method is better.
It just occurred to me that if you already have a 50/50 system, by using the shunt method you don't have to go to the trouble of building a lower resistance jig.  So EMRFD rules again.
Some practical results:
I wanted to go with the 1 ohm fixture first.  Below is a schematic of one I got from Bill Carver, W7AAZ.  It uses SMT precision resistors from Mouser. Also shown is a 3 dB attenuator.  Cut off at the right is matching to the HYCAS amplifier.  Bill was showing me how to use it to measure crystal parameters.

And below, we have it in physical form:

An unexpected problem!

I did some series method measurements with the tested item installed (as shown) and with a short across the two alligator clips.  What a shock to find, in some cases, more power measured through the tested device than with the short. Negative attenuation!

This baffled me for a while but I began to suspect the inductance of the loop formed by the two alligator clips and the shorting wire between them.  I measured the attenuation of a loop about that size on my AADE L/C meter and get somewhere between 0.045 and 0.075 uH.  At 8.2 MHz, that’s 2.5 to 3 Ω of reactance.  And in a 1 Ω environment, that’s significant.  Using LTSpice, it seems to add about 7 dB attenuation over a “real” short.

But when I have my series L/C circuit under test installed, the small loop inductance gets absorbed into my test coil’s inductance and cancelled when I find the resonant peak.  That’s why the circuit under test actually has a higher power value than the so-called shorted fixture.
I saw a couple of ways around this.  One was to resonate out the stray inductance when I did my "shorted fixture" measurements.  It's best to do this reasonably close to the measurement frequency, so I needed about 5800 pF.  That worked -- I was able to see the peak and measure available power with strays cancelled out.
Another method would be to assume the calculated 45.89 dB attenuation of the fixture is accurate, measure power with source plugged right into detector, and take 45.89 dB off of that for my "shorted fixture" power.  The method of resonating out strays seemed to give slightly better accuracy.
Oh, I did use a lowpass filter following the generator.  The DDS-60 is pretty well filtered, but notes on measuring Q emphasize the need for very good harmonic suppression.
Now some measurements.  I had a couple of iron power toroids wound with turns as noted.  One was a T68-7 and the other a T50-2.  I measured inductance with my AADE meter and chose a resonating capacitor (silver mica) to resonate at something over 8 MHz.
I made a measurement of the L/C resonant circuit and a second one with some series resistance added, to see if the delta came out close to the resistor value.  I also tested a type 61 toroid and a miniature molded choke, just to do some samples with lower anticipated Q.  I plug the attenuation and source resistance values as well as inductance and frequency into an Excel spreadsheet and have it crank out loss resistance Rx and Q.

The method is  to measure the shorted fixture power as discussed above, then hook up the test specimen and do a response sweep with PHSNA.  It finds peak and minimum values.  In this case, I want the peak.  I take a peek at the plot just to make sure it's not a "false peak".  Take the difference of the two dBm readings to get dB attenuation.
So I think my accuracy is probably decent, but could be improved.  That may be my next post.
What about the shunt method?
I decided not to try that with the fixture I have now, because dealing with the stray inductance is not so simple.  It will not be absorbed into the test specimen's inductance but instead will combine with it in a more complex way.  Probably making another fixture with minimal strays is a good solution. Also, based on the graphs, the 50 ohm method might be the best overall for the shunt method.
EMRFD is the ARRL book Experimental Methods in RF Design
PHSNA can be found at:
... and in an article in the spring 2014 QRP Quarterly magazine.
Nick / WA5BDU