Even Better PureSignal Performance on your Apache Labs radio

[K1VF]

IMD SUMMARY 7000DLE MKII K1VF_Rev13.zip

Added Appendix 3 RL-2 Isolation Measurement

(22.21 MiB) Downloaded 420 times

Want up to 10 dB lower IMD when using PureSignal?  Read on!

PureSignal operates by using feedback from the PA output to calculate the gain and phase characteristics of the PA. Then, the input to the PA is ’pre-distorted’ to compensate for the PA’s gain and phase non-linearities.

If this critical feedback signal is itself, in any way, distorted between the PA output and the A/D input where it’s inserted back into the digital processing chain, this will clearly produce a sub-optimal PureSignal correction result.

Attached please find a summary of our recent work that has has explored one significant mechanism by which this feedback signal can be distorted in at least some Apache Labs radio models.

Thanks to Joe K1WPO, Tom KJ7TEA, Warren Pratt NR0V and Rob Sherwood NC0B for their measurements, input and advice while conducting this work.

Vince, K1VF

[w-u-2-o]

Interesting reading, Vince.

You went through far, far too much work to prove your point!

This is kind of a blast from the past. Anyone who remembers the very first days of PureSignal will be quite familiar with the ramifications of crosstalk and RF isolation affecting performance.

You might be surprised to know that the first implementations of PureSignal intentionally relied on the rather poor, dare I say horrible, internal crosstalk in the early Apache designs to obtain the feedback signal! This would be on the 100, 100D, etc. To reduce these problems Apache added an internal shield between the PA/RF board and the SDR board (which made disassembly/reassembly a real challenge!) These early Apache designs did not have any RX/feedback switching or internal couplers to obtain the feedback signal. Folks who were serious about achieving good PureSignal performance modified their Apache hardware to add a high quality, high isolation RF relay to switch between the receive and feedback paths and provisioned an external coupler to obtain the feedback signal. Dow-Key microwave coaxial relays and the very first versions of the AC2IQ couplers were extremely popular. A lot of folks documented these mod's, for example W1AEX.

Similarly, a high premium has been put on maximizing feedback levels without exceeding the no-damage input level to the hardware. The greater the feedback level for a given operating condition the better the feedback-to-crosstalk level becomes, thereby helping minimize input errors to PureSignal. Many of my old posts in this forum discuss this issue.

Thanks for finding this latent RF isolation problem. No doubt folks who want to have both improved PureSignal and transverter capability can simply use the same approach as in the early days and implement a high isolation switch in the micro-coax path. Those Dow-Key microwave coax relays are always available on eBay for very short money (<$50) and they typically provide >85 dB of isolation at HF frequencies.

I have some more feedback (ha, a pun ), but I need to organize those thoughts a bit more first before posting them.

[w4bf]

Going to try this. I had noticed when I first tested my 7000 Mk II that the IMD on a separate SDR rx didn't match what I saw on the Thetis screen. I blamed it on the SDR but maybe not. Will see, looks as simple as disconnecting a mini coax cable.

[K1VF]

Thanks Scott: Re>. 'You went through far, far too much work...'

Any thorough analysis requires more than a single data point, and thus we now have sufficient data to explore and use, at least for these three models. Having that data show trends that otherwise might not be apparent.

Your comments are appreciated, but if you review the material carefully, you'll see that your comments are somewhat of an apples and oranges comparison. As it turns out, yes, I'm familiar with the early 100D era Gen 1 radios. I've had a 100D since 2015 and worked through many of those early issues. That was a different form of crosstalk, whereby the PA output coupled in unintended ways and was at 100x the level we're dealing with here. I was more interested, at that time, in contributing to what ended up as the Auto-Attenuate feature that Warren developed and added to Protocol 1 that greatly enhanced the performance and auto-lock stability of PureSignal.

In this case, all of those Gen 1 issues from a decade ago have long since been resolved in recent and current production models starting from the 8000 up through the current G2. This is a different issue.

Finally, the point of having sufficient isolation of a wanted low-level pre-distortion correction signal is greatly preferred to poor isolation and 'swamping' via brute force application of over-range levels of the wanted FB signal to achieve better S/N, as is true in most small signal analysis.

For example, the wanted FB signal is ~0dBm at RX1, the unwanted coupled signal is -54dBm thru RL-2 (see measurement attached). Increasing the wanted FB to +10dBm improves the S/N to 64dB, eliminating it increases it to 90dB. No question which method is preferred. Not for nothin', BTW, +10dBm may overload the ADC (or come close to it) and the auto attenuate function would be offset close to the rail. Before knowing what the RL-2 isolation actually was by measurement, Rob had the foresight to ask Warren what level of isolation might be required, to which Warren estimated that 80dB or better is likely required to prevent contamination that would degrade PS correction. Those nuanced questions and answers demonstrate the detailed understanding of the subject matter, that far exceeds my abilities. Hardware, on the other hand, has always been my comfort zone. Subsequently, we measured it and found RL-2 only provided 54dB of isolation...those measurements are attached.

Now, I hope what happens here is that via an interest in learning, other folks with different radios than what we were able to test, as well as future yet to come Apache Labs radios can benefit from similar efforts without having to do the work we have already done. That should, I hope, generate content on this members site that directly addresses what I assumed to be part of its' mission. For example, Bill Diaz has already noted that the 8000, a very popular radio and still in high demand, has a very similar layout and is likely to benefit from the same adjustment.

Attached are measurements of RL-2 Isolation in a G2 manufactured in 2024 as a function of frequency from 2-30 Mhz. Also attached is the functional block of the 8000 showing that it too is likely to benefit by this simple but effective improvement.

Regards,
Vince
K1VF

[w9ac]

Attached are measurements of RL-2 Isolation in a G2 manufactured in 2024 as a function of frequency from 2-30 Mhz. Also attached is the functional block of the 8000 showing that it too is likely to benefit by this simple but effective improvement.
K1VF

Well done...

I suspect only a small percentage of owners actually use the transverter port. It may have been better for Apache to have shipped transceivers without the jumper installed and instead, place it in an accessory packet with instructions for those who need it.

RL-2 begs for PIN diode switching, even if it's ~ 0.5 dB of switching loss. With good board layout technique, an additional 20 dB of isolation should be easily achievable. With the Orion board's extreme component density, a user-modification would take a Herculean effort to get it right while not causing collateral damage.

Paul, W9AC

[w-u-2-o]

Thanks Scott: Re>. 'You went through far, far too much work...'

Any thorough analysis requires more than a single data point, and thus we now have sufficient data to explore and use, at least for these three models. Having that data show trends that otherwise might not be apparent.

Your comments are appreciated, but if you review the material carefully, you'll see that your comments are somewhat of an apples and oranges comparison. As it turns out, yes, I'm familiar with the early 100D era Gen 1 radios. I've had a 100D since 2015 and worked through many of those early issues. That was a different form of crosstalk, whereby the PA output coupled in unintended ways and was at 100x the level we're dealing with here.

There is no difference in kind. Crosstalk is crosstalk, RF isolation is RF isolation. Again, well done on finding this more subtle instance. Perhaps there are other places in the design that might deserve the same attention. For example, there is ample opportunity to improve TX and RX path isolation on the PA/RF board.

In this case, all of those Gen 1 issues from a decade ago have long since been resolved in recent and current production models starting from the 8000 up through the current G2. This is a different issue.

Perhaps we'll just have to agree to disagree. Same issue, different source.

Finally, the point of having sufficient isolation of a wanted low-level pre-distortion correction signal is greatly preferred to poor isolation and 'swamping' via brute force application of over-range levels of the wanted FB signal to achieve better S/N, as is true in most small signal analysis.

Those are some very negative ways to describe common, acceptable, and relatively easy practices for controlling SNR in any RF design. Perhaps not commonly seen in amateur radio equipment, but as the size and complexity of the RF design increases such things are very necessary. Think cellular base station radios, military electronic warfare systems, etc.

Of course the openHPSDR hardware designs don't approach that level of complexity. Nevertheless, you have already noted as to how the requirements for performing adaptive predistortion linearization are quite a bit more stringent than the average amateur transceiver might otherwise require.

For example, the wanted FB signal is ~0dBm at RX1,

It would be more accurate to say that the desired feedback signal level is 0 dBm at the ADC input. This was something I noted in your write-up and I hope that's what you meant. 0 dB at the RX1 input (board edge, not rear panel) will result in either +20 or +14 dBm at the ADC input as the (non-defeatable) preamp prior to the ADC is either +20 or +14 dB depending on the vintage of the hardware (the lower gain preamp came into use with the advent of the 8000).

the unwanted coupled signal is -54dBm thru RL-2 (see measurement attached). Increasing the wanted FB to +10dBm improves the S/N to 64dB, eliminating it increases it to 90dB. No question which method is preferred.

The preferred method is "do both". Then the total improvement is 100 dB.

Not for nothin', BTW, +10dBm may overload the ADC (or come close to it)

It will not overload the ADC. This calculation has been done over and over again going all the way back to 2013, and probably before then as well (I joined the community in 2013 when I bought a used 100D). See this topic.

and the auto attenuate function would be offset close to the rail.

Yes it will. And why is this a bad thing? Indeed, doing it this way provides another benefit, which is to allow PureSignal some dynamic range of adjustment. When feedback levels are configured for +10 dBm at the rear panel input at whatever maximum RF power output the station is capable of, there are still a few dB of margin in the step attenuator, there is maximum feedback-to-noise ratio under the conditions that need it most (QRO), and there is at least 13 dB or more of dynamic range available for good PureSignal performance without changing coupling factors, i.e. PureSignal results remain good from, for example, 1500 down to 75 Watts RF output.

[dj1yr]

Hi,

what is the IMD free dynamic range of the Rigol?
How did you rule out that you are not measuring the IMD of the Rigol?

We measured the Orion MK2 boards IMD many years ago (we also found a construction error), the output IMD of a REV 5.2 is partly worse than their measurements without PD, we measured 5 MK2 boards at that time, all behaved similarly.

The problem you describe has been known for many years, at least among people who have built with Orion MK2 themselves.

Unfortunately, Apache made some design mistakes in the first versions of the 10-100D and the 7000 is not perfect either.

[w-u-2-o]

A minor error (I think) on Page 5--in the first paragraph Figure 1 should be Figure 2. And it might be more clear to specifically reference Figure 1 in the second paragraph.

[K1VF]

In an effort to keep my original post on track, I'd like to reiterate:

The summary points of the presentation are:

You can achieve even better IMD performance with PureSignal with the XVTR Jumper removed:

1. The Thetis Panadapter display (DUP Mode) and an external S/A now agree to w/in 1dB on all bands, at all power levels, on all three models tested;

2. With the XVTR Jumper removed:
3rd Order IMD improves with PureSignal by as much as 12dB, with 3rd order IMD products down by as much as -60dBc, at all power levels; and

3. With the XVTR Jumper removed:
When used to correct an external RF Amplifier, PureSignal corrects to even lower IMD levels: up to -60dBc on a Flex PG XL LDMOS amplifier (a 35dB improvement to 3rd order IMD) and -64dBc on an Alpha 77DX amplifier (a 28dB improvement).

All of these measurements are provided in the writeup. We've verified S/A measurements using both sampling type spectrum analyzers (Rigol) and also with hp and Tektronix analog spectrum analyzers, all of which agree.

The ADC level is inconsequential. Both the wanted [0dBm] signal and the unwanted signal from RL-2 [-54dBm]are equally affected by the subsequent step attenuator and preamp ahead of the ADC. Therefore, isolation is still 54dB, no difference.

Scott, thanks for finding the typo, yes Fig 1 and Fig references got mis-typed. I'll fix those an a few other minor typos and post a new revision shortly.

[K1VF]

Hi René:

Thank you for your post. You asked "How did you rule out that you are not measuring the IMD of the Rigol?"

That is a very good question. First, We applied the measurement principles taught in the Rhode & Schwartz white paper on IMD measurements, here:

https://www.rohde-schwarz.com/us/applic ... 15423.html

We applied external attenuation such that both the f0 and 3rd/5th/7th order products tracked up/down as attenuation was stepped down/up, until we noticed that the IMD degraded relative to f0 as input levels to the S/A were increased to too high a level, then we stepped back 6-10dB below that threshold level.

Second, we compared the two Rigol S/A's to two different analog S/A, an hp3585A and a Tektronix 492AP, all measurements agreed within about 1dB.

Third, as you know, when IMD measurements degrade -due to excess signal levels input to the S/A- the IMD measurement degrades because it includes the S/A's IMD, and that data is not usable. In the case of our measurements, the IMD measurements are actually better, not worse, and all five instruments agree to w/in ±1dB across a series of hundreds of measurements.

That is how we knew that we were not measuring IMD of the Rigol [or the other Spectrum Analyzers].

Finally, in each of the three models tested, please note the right-most column under the header: PureSignal Improvement.

That is the difference in the Orion Rev 5.2 or Saturn Rev 2_1 boards with their respective PA with PureSignal ON versus OFF. Note that in every measurement made, on every band and at all power levels, the IMD is much better with PS ON –by a long shot– as is shown for all models on all bands.

The 7000DLE built in 2017 was 18-35dB better with PureSignal ON
The G2 was 28-41dB better with PureSignal ON
The 7000DLE MKII was 12-35dB better with PureSignal ON

I would love to hear from anyone with an 8000DLE model and a spectrum analyzer who might be able to provide a dataset with the XVTR Jumper J31-J32 installed and then removed with PureSignal ON and OFF on 20m at 50 and 100W, Anyone interested can PM or eMail me for an excel template. Phone support provided.

[w-u-2-o]

In further reviewing the write-up, it appears there is only data that directly compares jumper installed vs. jumper removed on the following three G2 related pages: 49, 58 and 67.

On page 49, with the G2 on 40M and at 50W, there is only a very small improvement in PureSignal performance, approx. 3 dB.

On page 58, with the G2 on 20M and at 50W, there is a significant improvement of PureSignal performance, approx. 12 dB.

On page 67, with the G2 on 80M and at 50W, there is effectively no improvement in PureSignal performance, maybe 1 dB.

Assuming that page 103 is with the offending jumper installed, the above results do not correlate well with the isolation data shown on that page. The isolation measurements are within 1 dB on 20M and 40M, yet 20M performance was significantly better with the jumper removed while 40M performance remained the same. Thus the presence or absence of the jumper alone cannot account for those results. Indeed, these results imply that there is another source of poor isolation (i.e. crosstalk) on 40M somewhere in the G2.

On 80M, there is approximately 5 dB better isolation than on 20M and 40M. Given that the results on page 67 show essentially no difference in performance, this implies that merely achieving 5 dB more isolation is enough to do the trick.

In summary, only 20M is shown to benefit in any significant way and only on the G2. There is no data presented that shows the improvement from removing the jumper on the ANAN-7000 cases. Could that data be made available?

[K1VF]

Regarding isolation of RL-2, since we can see from the initial dataset the effect is both frequency and power dependant, higher bands needed to be tested.

Shown in the attached summary chart, replacing the RL-2 jumper with a 30dB step attenuator demonstrates that ≥90dB of isolation is required to prevent degraded IMD performance of PureSignal, as well as degraded S/A-to-Thetis DUP measurement accuracy. Simply improving RL-2 with another 5dB of isolation isn't 'enough to do the trick', the jumper needs to be removed.

On 10m, RL-2 has 60dB of isolation (see previous post and measurement). By adding attenuation in place of the jumper (i.e., more isolation) in 10dB steps, we can measure both IMD as well as Thetis (DUP) to S/A differences. As the measurements summarized on the chart shows, the full 30dB of attenuation, in addition to the 60dB of RL-2 (i.e., >=90dB total) isolation is needed on 10m before the effects of contamination go away. This is in agreement with Warren's prediction that >80dB of isolation would be required. The higher the frequency and power level, the higher the isolation needs to be. It's clear that ≥ 90dB is required on 10m, and that IMD performance as well as agreement with an external Spectrum Analyzer both improve dramatically with the RL-2 jumper removed.

Regarding your question about the availability of additional data, yes, more data is available, on all bands from 160-10m, and has been summarized in the attached tables and charts. In an effort to summarize several hundred measurements in tables and S/A screen grabs, a visual summary has been provided instead, condensing all of the data on all bands in a few simple charts.

The attached plots do a better job of illustrating the improvement in both the IMD performance of PureSignal, as well as the accuracy of Thetis DUP mode (compared to the S/A) with the RL-2 jumper removed. IMD performance is improved significantly, by 10-20dB from 10-30Mhz, as well as S/A to Thetis agreement, also improving by as much as 20dB.

Vince, K1VF

Notes:
All measurements were made using an external Directional Coupler with 47dB of attenuation;
All measurements were at a 50W power level;
Nominal S-ATT level was -15±1dB;
Auto-attenuate was enabled.

For these models, that provides 0 dBm at the RX1 input, and +5 dBm at the ADC input, which is 6dB below the ADC overload level of 11dBm, or 2.25Vp-p, per the ADC specifications.


EDIT: 05 FEB 2025
Good catch Scott, thank you. Yes, the X-Y scatter plots are created and formatted manually, which is tedious, so duplicating just required an array to be cut-and-paste for the 2nd radio's data, which was interrupted with a call. That file has been corrected and the Summary PDF replaced with an edit note.

[w9ac]

It's interesting that leakage is similar between the 7000 and G2 models. As shown, an approximately 24 dB IMD improvement on 10m is a substantial numerical improvement. But even with the jumper in place - 50 dBc is still quite good, made even better with the jumper removal.

It also proves that the DUP can only be counted on for an approximate display of IMD change with PS engaged. There's no substitute for a spectrum analyzer placed directly at the test load.

Vince, thanks for putting the time into this. I think it's a very worthwhile exercise.

Paul, W9AC

[Warren_NR0V]

Vince,

Many thanks to you and the small team that worked with you to put this together!

This is an excellent, well-organized, summary of the data and of an approach Apache owners can use to significantly improve IMD performance and improve the accuracy of the DUP display while transmitting. While, when using PureSignal, IMD performance is very good without this modification, I would highly recommend this for those that are interested in exceptional performance.

Note: The accuracy of the DUP display should only be effected by the RL2 contamination while transmitting. Even without this modification, remember that, assuming proper frequency and amplitude calibration, the radio can be used as an excellent spectrum analyzer!

73,
Warren NR0V

[w-u-2-o]

The new data presentation is certainly much more complete and compelling than the initial effort. Thank you for putting it together. The improvement in IMD performance is indeed dramatic above 40M.

There does appear to be an error in the new data presentation. It seems the data for the S/A IMD with Jumper is identical for both G2's. This might just be a cut and paste mistake.

One thing that might be confusing to some is the use of the label "Delta". Many might expect that to be the difference in IMD performance with and without the jumper, a very important value. In this case it's the difference between the Thetis measurement of IMD vs. the spectrum analyzer, of course. Perhaps it would be more clear if both types of deltas are shown.

Here is the data plotted in a different way which might make it easier to appreciate the IMD performance improvement. The solid lines represent the jumper in place, and dashed lines represent the jumper removed. Again, the two G2's appear to have identical data sets for the jumper installed cases.

Capture.JPG (250.52 KiB) Viewed 9534 times

A similar plot could be done showing the spectrum analyzer accuracy difference data.

[K1VF]

EDIT: 05 FEB 2025
Good catch Scott, thank you. Yes, the X-Y scatter plots are created and formatted manually, which is tedious, so duplicating just required the arrays to be cut-and-paste for the 2nd radio's data, which was interrupted with a call. That file has been corrected and the Summary PDF replaced with an edit note. You might update your plot with the corrected array. It was the G2#2 K1VF G2 with the duplicate data.

[K1VF]

Hi Paul:

I understand that Thetis was originally written as a Spectrum Analyzer, project code name Cyclops.

If you look at the G2#2 data with jumper removed, the Thetis DUP display agrees with the external S/A to w/in 1dB everywhere except 12m, 1.7dB. That's actually w/in the absolute power level accuracy spec of the S/A in most cases (±1 or ±1.5dB), so PureSignal with Thetis DUP and the jumper removed is nearly a good as an external S/A, that was the impetus of the entire project. We already knew that Thetis in RX mode was already very close in accuracy to an external S/A, now it can also be relied upon in TX mode under 2-tone testing.

The other two radios, are w/in about 1-2 dB until 18-30Mhz, where they start to diverge approx 1-4dB, still not bad. I have some ideas about how to address the remaining small discrepancy from 18-30Mhz in TX 2-tone mode, and will be performing some experiments shortly. We'll report back if we make some headway in this regard. Thanks for your post.

Vince
K1VF

[K1VF]

If you update your chart, correct the G2 2024>1.8Mhz>No Jumper data point appears to have been transcribed as -55.9, it should be -55.0

[K1VF]

"A similar plot could be done showing the spectrum analyzer accuracy difference data."

Please do.

[w-u-2-o]

Here is the updated chart with corrections listed at the bottom.

If you would send me the actual spreadsheet with the Thetis and external spectrum analyzer data I'd be happy to make a similar chart as I'd rather not manually transcribe data a second time.

Capture.JPG (283.2 KiB) Viewed 9406 times