Step attenuator linearity

User avatar
w-u-2-o
Posts: 5540
Joined: Fri Mar 10, 2017 1:47 pm

Step attenuator linearity

Postby w-u-2-o » Thu Feb 27, 2020 6:12 pm

This is a re-post of an analysis I posted to the old Yahoo Group back in December 2019:

Summary:

After some extensive business travel in December, I've finally had an opportunity to revisit this subject and to try to reproduce the results that Paul reported earlier in the month. I was unable to do so, and can find no evidence of any non-linear responses in the receive signal path at any reasonable level of PureSignal feedback input power level.

Recap of Paul's Results:

Back at the beginning of December, Paul, W9AC, proposed that it was possible to overdrive the step attenuator and thereby cause non-linear behavior that PureSignal could not reconcile, thereby causing significant degradation of PureSignal linearization performance.

Paul, using a home built coupler with a reported coupling factor of 50dB (and reporting step attenuator values from the auto-attenuate algorithm commensurate with a 50dB coupling factor so there is good confidence in it), experienced very poor PureSignal linearization performance at 1.85MHz at power levels above 1KW into a dummy load. As a result, he added 3dB of additional attenuation, and then 6dB, obtaining much better performance at each step. His original spectrum analyzer plots are attached.

My Analysis:

After some research and correspondence with Mini-Circuits engineers, Figure 1 in the latest step attenuator (Mini-Circuits DAT-31A-SP+) data sheet was identified as showing both the 0.2dB compression point vs. frequency as well as being the maximum safe input level vs. frequency for the step attenuator. Mini-Circuits further advised that this data is valid for both the older and now obsolete DAT-31-SP+. Figure 1 is attached.

As the figure shows, the curve at 1.85MHz gives approx. +13dBm as the max. allowable, 0.2dB compression point input signal level to the step attenuator.

Under Paul's test conditions, at 1KW the output of the coupler would be 60dBm - 50dB, i.e. 10dBm. This is 3dB less than the equivalent point on Figure 1, therefore it does not seem reasonable that there should be a problem at this input power level.

As a result of this I postulated that perhaps there was some other non-linearity at play in Paul's configuration at feedback input. And I wanted to try some similar measurements for myself to see if I could duplicate Paul's results and perhaps identify what else might be causing the issue.

My Experimental Results:

I used an original version AC2IQ coupler with a claimed coupling factor of approx. 44dB. This same coupler is now available commercially from HRO as the Xtronic XDC-4SO. The amplifier is a KPA-500 running at 500W into a dummy load. The RF unit is an ANAN-8000DLE. Firmware is Protocol 2 "pre1". Software is Thetis 2.6.9 (b6). I measured my coupler and it has an actual coupling factor of approx. 47dB, not unreasonable given that it is a essentially a prototype of the commercial product.

At 500W, i.e. 57dBm, the output of the coupler is 57dBm - 47dB, i.e. 10dBm, which is exactly commensurate with the level Paul was achieving. And, as with Paul's example, the step attenuator value arrived at via the auto attenuation algorithm confirmed the 47dB coupling factor.

However, under these conditions the achieved PureSignal linearization performance was quite good, as shown in the attached file "0dB atten warm.JPG". I repeated these measurements with an additional 6 and 10dB of attenuation (results attached) provided by inserting a JFW rotary adjustable attenuator inline. Step attenuation values changed by the proper amount of dB each time. There was no significant difference in performance at any of the three feedback signal levels as measured by the "DUP" display mode on the radio.

Note that the word "warm" appears in the filenames as it was necessary to allow the amplifier to warm for 20 to 30 seconds in order to remove thermally related memory effects. I.e. the FETs in the KPA-500 need to obtain a reasonably stable thermal state to get good, consistent results. This issue should not be overlooked with other amplifiers or other testing. It is important to have a good dummy load that can absorb the necessary power for the necessary amount of time!

Paul further proposed that the DUP display of the radio could not be trusted to provide proper results. This assertion is also not consistent with my results. Using an HP 8560 spectrum analyzer on the second forward output port of the AC2IQ coupler, and a tablet mirroring my main station computer display, one can see the results are quite comparable (photo attached). Note that the 8560 is using an un-attenuated output with levels exactly as you would expect (each tone 3dB down from the 10dBm total power output level) while the radio was still seeing an additional 10dB of attenuation. Also, I did not bother to match up the resolution bandwidths of the two measurement systems. No doubt had I bothered to match levels and match measurement settings the two displays would most likely match almost exactly. Regardless, there are no gross differences between the measurements, and no reason to distrust the DUP display mode of the radio. Indeed, the radio is probably a superior instrument compared to a nearly 25 year old spectrum analyzer!

73,

Scott

T44-160m-1KW-PSon-0 dB.jpg
T44-160m-1KW-PSon-0 dB.jpg (217.92 KiB) Viewed 11055 times


T44-160m-1KW-PSon-3 dB.jpg
T44-160m-1KW-PSon-3 dB.jpg (204.74 KiB) Viewed 11055 times


T44-160m-1KW-PSon-6 dB.jpg
T44-160m-1KW-PSon-6 dB.jpg (202 KiB) Viewed 11055 times


DAT-31A-SP+ Figure 1.JPG
DAT-31A-SP+ Figure 1.JPG (48.07 KiB) Viewed 11055 times


0dB atten warm.JPG
0dB atten warm.JPG (531.64 KiB) Viewed 11055 times


6db atten warm.JPG
6db atten warm.JPG (518.28 KiB) Viewed 11055 times


10dB atten warm.JPG
10dB atten warm.JPG (512.28 KiB) Viewed 11055 times


Analyzer vs Radio.jpg
Analyzer vs Radio.jpg (2.9 MiB) Viewed 11055 times
User avatar
w9ac
Posts: 290
Joined: Sun Apr 09, 2017 4:01 pm

Re: Step attenuator linearity

Postby w9ac » Thu Feb 27, 2020 8:07 pm

Scott,

My DUP display does track with the S/A until the non-linearity occurs and PureSignal can no longer linearize. You have not yet reached a point of non-linearity with the HP measurement. As such, I would expect your result.

It would be helpful if you varied coupling to the point where IMD is reached and PureSignal no longer linearizes. If that point nears or reaches the maximum allowable MCL input power so be it, but please measure it.

If you and I have reached different conclusions, then we need others to validate the testing and outcome.

Paul, W9AC
User avatar
w-u-2-o
Posts: 5540
Joined: Fri Mar 10, 2017 1:47 pm

Re: Step attenuator linearity

Postby w-u-2-o » Thu Feb 27, 2020 9:52 pm

w9ac wrote:It would be helpful if you varied coupling to the point where IMD is reached and PureSignal no longer linearizes. If that point nears or reaches the maximum allowable MCL input power so be it, but please measure it.

I'm loathe to do this because I'm already banging up against the maximum safe input level of +13dBm at the rear panel (well above the level at which you say 160M IMD occurs in the feedback path).

If you and I have reached different conclusions, then we need others to validate the testing and outcome.

I believe both sets of measurements are correct. And I also believe that there is a condition in your station that causes this problem. What I don't agree on is that it has anything to do with the step attenuator.

You may have already done this, but if you have access to an Xtronic coupler I wonder if that would make any difference in the measurements?

Also, do you have return loss data on all three ports of the toroidal sampler? If so, what does it look like?

73,

Scott
User avatar
w9ac
Posts: 290
Joined: Sun Apr 09, 2017 4:01 pm

Re: Step attenuator linearity

Postby w9ac » Thu Feb 27, 2020 10:20 pm

w-u-2-o wrote:You may have already done this, but if you have access to an Xtronic coupler I wonder if that would make any difference in the measurements?


Yes, I have repeated the 160m results using three couplers:

Coupler #1: Xtronic XDC-1
Coupler #2: 60dB coupling coefficient capacitive voltage divider (later changed to 50 dB)
Coupler #3: 50 dB coupling coefficient T-44 toroid sampler.

The first notion that something was amiss occurred after constructing the toroid sampler. I initially blamed the inability to linearize (on 160m at 1KW) on core saturation of the T44 core. It wasn't until I inserted a Weinschel programmable attenuator that the couplers could be compared while looking at the IMD threshold where PureSignal won't linearize. I see zero difference in any of the couplers with a 56 dB total coupling coefficient. That means all samplers have fixed pads. The 50 dB couplers have 6 dB fixed pads, the Xtronic started with 20 dB and later replaced with a 10 and 3dB pad in series.

w-u-2-o wrote:Also, do you have return loss data on all three ports of the toroidal sampler? If so, what does it look like?


No, I did not measure return loss. only swept insertion loss (IL). I first measured the Weinschel's IL and discovered a bad cell. After replacement, I used my N2PK VNA in S21 mode and swept IL through 60 MHz at 0, 1, 3, 6, and 10 dB. I did this to prove that the attenuator was performing as expected. Like your attenuator, the Weinschel uses purely passive cells.

As Warren pointed out on the Apache list, it doesn't take much IMD at the receiver's input to destroy PureSignal's ability to linearize. That's why I would like others to repeat the test and get a comprehensive set of data. Your results and mine alone may not be enough to form an accurate conclusion. Here's Warren's response:

"I did want to point out that 0.2dB attenuator compression (compression from zero amplitude to maximum amplitude of the signal) is quite significant as far as linearization with PureSignal is concerned. 0.2dB is more than 2% voltage change which will be seen by PureSignal as additional non-linearity of the PA. When we’re trying to get IMD to, say, -60dBc, a 2% error is quite significant...It’s unfortunate that the non-linearity persists when the attenuator is set to 0dB – if it did not, at least there would be a complete workaround."

Finally, we shouldn't just depend on datasheets and math. It's a necessary start, but measurement is still required.

Paul, W9AC
User avatar
w-u-2-o
Posts: 5540
Joined: Fri Mar 10, 2017 1:47 pm

Re: Step attenuator linearity

Postby w-u-2-o » Fri Feb 28, 2020 12:28 am

Would it be difficult to measure the return loss of your coupler? Just curious to see if there is an impedance problem lurking.
User avatar
w9ac
Posts: 290
Joined: Sun Apr 09, 2017 4:01 pm

Re: Step attenuator linearity

Postby w9ac » Fri Feb 28, 2020 3:53 pm

w-u-2-o wrote:Would it be difficult to measure the return loss of your coupler? Just curious to see if there is an impedance problem lurking.


I can do it as time permits, but keep in mind I've made IMD measurements using three different RF samplers that yield the same result. The only RL measurement that I had considered earlier is an S11 reflection sweep into the 7000's Bypass port during TX. For a problem to exist, it would mean all three samplers are going non-linear into what's predominantly a passive input circuit.

Concerning the MCL DAT-31A-SP+ datasheet and the "Figure 1" graph, what I believe is shown (and suggested by Don Jackson, W5QN) is the maximum safe input level. It does not mean that the device isn't going non-linear to some small degree at levels below the curve.

It would be interesting to conduct some out of circuit testing of the DAT-31A-SP+. I'm not motivated to do that level of testing at this time. Frankly, MCL should provide more detailed analysis of input power by frequency vs. IMD products on their datasheet.

Paul, W9AC
User avatar
w9ac
Posts: 290
Joined: Sun Apr 09, 2017 4:01 pm

Re: Step attenuator linearity

Postby w9ac » Thu Jun 11, 2020 1:04 pm

"As the figure shows, the curve at 1.85MHz gives approx. +13dBm as the max. allowable, 0.2dB compression point input signal level to the step attenuator...Under Paul's test conditions, at 1KW the output of the coupler would be 60dBm - 50dB, i.e. 10dBm. This is 3dB less than the equivalent point on Figure 1, therefore it does not seem reasonable that there should be a problem at this input power level."

Scott, I went back and looked at your computation and you assumed an output power level of 1000W, not 1500W. As shown on my QRZ page, the testing level is 1500W. With a 50 dB coupling factor that's (1500log10*10+30 = 61.76 dBm). Correcting for this, we're nearly at 12 dBm, not 10 dBm as you incorrectly state. If +13 dBm is the maximum input level of the MCL device, then at 1500W with 50 dB coupling, we're really close to the edge, particularly when Warren states that just 0.2 dB of compression is significant for good linearization. There's also production variability of the MCL attenuator to consider.

So, at a +10 dBm level out of the sampler, you see about -55 dBc of distortion products (I'm looking at your HP SA image). Compare that with my 53 dB plot. At 1500W (not 1KW), that's +9 dBm into the MCl attenuator with 53 dB of coupling. Now our sample power numbers are getting close. Moreover, my 53 dB coupling spectra plot show about -50 dBc. It's not too far from what you're seeing. That 5 dB spread can result from several things, different model of ANAN transceivers, different model amps, etc. You measured a 8000DLE and I have a 7000DLE.

Next, you and I show different results when adding more sample attenuation beyond this point. I see distortion products bottoming out at a 55 dB coupling factor at 1500W on 160m (hence a 6 dB pad to get to 56 dB). I believe your description states that you saw no improvement with added sample attenuation. My testing is with an Alpha 77Dx triode amp with already excellent IMD. I believe your testing was close to the edge where you would have seen rapid changes in PS had you increased the sampling level a bit more. As time permits, I'll go back and test at 1KW.

Referring back to the 31 dB attenuator, MCL specs the DAT-31A+ absolute input level at +30 dBm. Looking at the MCL power curve that makes sense for the upper end of the spectrum. Again, that does not mean small amounts of non-linearity aren't occurring near or even below the curve at 1.8 MHz when I'm 1 dB away (at 1500W) from the curve's maximum +13 dBm level.

What I'm saying is that a 56 dB coupling factor results in: (1) good PS linearization at full 1.5KW power on 160m; and (2) the ability to linearize when an external amp is in bypass. With 56 dB of coupling, I can linearize down to less than 50W. The GUI spectrum display and Anritsu SA completely track at any power level between these extremes.

Obviously, your 50 dB coupling suggestion is fine for power levels that are about 1KW or less on all bands, including 160m.

Paul, W9AC
User avatar
w-u-2-o
Posts: 5540
Joined: Fri Mar 10, 2017 1:47 pm

Re: Step attenuator linearity

Postby w-u-2-o » Thu Jun 11, 2020 3:16 pm

Paul,

I chose the 1KW test condition based on your original post to the old Yahoo group, as well as the labeling of your posted figures, not what you posted on your QRZ page. Sadly the Yahoo group archives are gone so I can't go back and verify my reading comprehension ;)

I would greatly appreciate it if you would go back and retest at 1KW. That may be important data to have. Alternately, simply change your coupling factor to 53, or better yet 52dB, and see what happens. I keep a 10dB rotary step attenuator with 1dB steps in my quiver of test tools for just that sort of testing.

73,

Scott
User avatar
w9ac
Posts: 290
Joined: Sun Apr 09, 2017 4:01 pm

Re: Step attenuator linearity

Postby w9ac » Thu Jun 11, 2020 4:54 pm

I can tackle this again in the next week or two and post the results here. It would be great to reach a conclusion that we can all agree on!

Paul, W9AC
User avatar
w9ac
Posts: 290
Joined: Sun Apr 09, 2017 4:01 pm

Re: Step attenuator linearity

Postby w9ac » Sat Feb 13, 2021 6:20 pm

While making some station wiring changes, I ran across the root cause of the 160m IMD disparity. Have a look at the attached screen captures from my Anritsu S/A. As before, the measurement is taken from a Bird inductive coupling loop at the input of a Bird TermaLine load.

I don't know when, but at some point I purchased a DX Engineering RG5000HD receiver guard and placed it in-line with the PS RF sample. When I began pulling cables today, I discovered it under the desktop. I don't even recall purchasing it or installing it but there it is.

Note that in my earlier measurements, I used a Weinschel programmable step attenuator with 1.0 dB steps to verify sample line insertion loss (IL). Well, the loss of the RG goes undetected because it's spec'ed at 0.15 dB (50 MHz). ARRL measured IL of only 0.01 dB at 1.8 MHz.

But DXE further states:

"The Receiver Guard is an advanced multi-stage receive signal limiting device that is virtually RF transparent at normal receiving levels, all the way up to -10 dBm,..With the Receiver Guard installed, when RF signal levels on the receive antenna line rise above the -10 dBm level, the specially designed circuit limits the output signal..."

However, have a look at the power limit chart in the QST review. The -10 dBm threshold is for the standard, non HD version. The same chart shows a soft knee with the onset of nonlinearity at about +5 dBm for the HD version I have. The chart is not frequency specific but the ARRL Lab does show additional limiting detail by frequency in Table 10.

The screen captures below show my Alpha 77Dx tube amp, tuned to 160m at 1500W output but I didn't optimize linearity. I simply adjusted plate loading for a typical condition where peak power is found then loading is advanced slightly with a reduction of grid current. I wanted to conduct the test in a way that I normally use the amp after a quick band change -- not where I am extracting the "nth degree" of linearity from the PA tuning adjustments.

A 50 DB coupling coefficient is used for both measurements. The only change between the two screen captures is that I bypassed the RG with a BNC barrel. I verified identical results with my capacitive divider and toroid samplers.

So, what does this mean? It means that I am revising my claim that a coupling coefficient of 50 dB is adequate when NOT using a receiver guard or any other device that may affect linearity in the sampled signal path. It also brings into question whether DXE's standard RG device (i.e, non HD version) should even be used with PS.

My measurements do show that with a coupling coefficient of 56 dB, the RG HD is transparent at 1500W from 160m through 6m. I suggest this level of coupling when used with the RG HD device.

I do apologize to Scott and others for not taking a deeper look into the sample line. I should have eye-witnessed the entire sample cable from end-to-end. It got missed due to the very low IL of the RG.
RG.jpg
RG.jpg (125.03 KiB) Viewed 5745 times
PS_1KW_160m_RG_Active.jpg
PS_1KW_160m_RG_Active.jpg (218.96 KiB) Viewed 5745 times
PS_1500W_160m_RG_Bypass.jpg
PS_1500W_160m_RG_Bypass.jpg (208.52 KiB) Viewed 5745 times

Paul, W9AC
User avatar
w-u-2-o
Posts: 5540
Joined: Fri Mar 10, 2017 1:47 pm

Re: Step attenuator linearity

Postby w-u-2-o » Sat Feb 13, 2021 10:18 pm

Hi Paul,

So happy to read that you sorted this out :)

Limiters are a great and wonderful tool for protection, but obviously only if properly specified for the purpose at hand. I use them quite extensively in my work (my real day job), where I design systems that live in very severe co-site environments. A typical limiting level in my designs is +30dBm. Like I said, a severe environment! I've seen competitor systems stop working many times as their LNAs become damaged.

I also use the DX Engineering limiter, but in a more conventional way. I have a dual antenna setup and a 2 x 2 switching matrix that always guarantees the TX ANT is on ANT1 and the other antenna is on ANT2, and the limiter sits on the ANT2 input.

Since the setup for PureSignal is a static case and should be configured so that no condition should cause a higher feedback signal than +13dBm, there should be no requirement for a limiter on that signal path.

Any how, glad a logical answer to this issue was finally discovered!

73!

Scott

Return to “PureSignal Operations (PowerSDR & Thetis)”