The community has many discussions about the SSDR DSP functions and many of these are qualitative posts saying that ANF, for instance, needs to be fixed. How about tightening focus?
So, lets take ANF. For me it seems to work okay and not having anything to compare I’m satisfied. But then I’m mostly on CW. Nevertheless, what exactly is the spec for the ANF? We have a sense of what it is supposed to do, but within what limits? Its a software subsystem tightly integrated into SSDR so Flex must have some specs to test their design. That spec they must hold close to not reveal intent and there isn’t much detail in paragraph 15.4 of the users guide nor specific performance specs.
That said, you can’t fix that which you cannot measure. We all know this.
Here’s an attempt at a provisional ANF performance spec, ultimately, in some fashion a “users’” desired specification:
Modes of operation: SSB, AM
Attenuation of interfering audio tone: 30 – 50dB
Effect on desired signal in passband: ? %THD?
Number of simultaneous tones notched in the passband: 5
Passband effectiveness: 1.8 to 5 khz
Number of slices using ANL: 1 to 8
Attack time: less than 100 msec
Effect on slice noise floor: none seen on spectrum
Effect on slice IMD: none (how to measure)
Then, measure SSDR with with our “users’” specification? These are sort of in order, my order, of importance. Users might consider more or less spec items. Users might consider different required values. Quantitative results might be a challenge.
All of the above is my idea of what to measure, I’m certain it can be improved.
By looking closely at the ANF’s desired spec a technical discussion can be had that may in fact be useful to Flex engineering. There are no hardware limitations as far as I know.
My thought is to move away from Flex flogging and into the tech specifics that we hams can debate. Field performance of the ANF (also for any comparable rig) can be measured using user accepted provisional specs as a grading rubric. Getting to this point is not easy in a public forum, but, perhaps it needs to be done to move forward.
Similar quantitative specs and answers for NB. NR and WNB could be considered.
So, for ANF, what exactly is it that we want?
First-- IMO, Spot on. Keeping issues on point, specific and with supporting information either about the way it acts differently then desired, or specifically how it is not perceived as working correctly and how to duplicate the issue is the best way to get the visibility of the team. It speeds re-creation of the issue both in alpha testing and in the programming lab, and is our best chance to get certain issues corrected. Keeping on point while being respectful is simply the right way to do it and a win win for everyone in the flex community.
Is it possible to determine at what, in passband signal strength does the effect become apparent? Is it influenced by bandwidth? What is the variation? What other features, NR, NB, WNB are in use? Do they have any effect on ANF performance? One would expect not and therefore could be dropped out of the ANF discussion, if the answer is "no." If yes, well, oh boy.....engineering would probably want to disentangle.
It seems a "notch depth variation with signal strength" is an expectation that could reasonably be included in specifying ANF performance. Or, as posted, perhaps the intended limit is zero, as is the case with TNF's. I suggest the effect is also related to, or identifiable with, receive signal distortion or degradation that I seem to recall reading in the community.
There isn't much field data to capture the effect nor to agree upon limits, except to desire "zero.". Some field data is going to be needed, a significant challenge.
ANF, I think can, as someone mentioned, be considered a black box, in general.
I would be interested in technical details of the ANF performance disappointments. Is notch depth variation as a function of signal strength the central issue? The only issue?
So, reported ANF performance issues, at this moment:
1. Notch depth varies with signal strength. (Performance Spec would be how much variation is acceptable and under what conditions). Someone implied that how engineerring achieves the spec is not relevant. I agree.
What else is wanted for ANF performance?
It is actually kind of easy to do. For example, I purchased ANAN radios specifically to check what other hams told me regarding the superiority of ANAN's recovered audio. The ANAN NR and NR2 features basically eliminate noise without having a noticeable effect on the audio. The NB and SNB features remove various periodic noise without having a noticeable effect on the audio. The ANF feature removes carriers, period--no effect on the audio and carriers are completely removed. Any honest comparison of the recovered audio, in real operating conditions, of a Flex to that of an ANAN quickly reveals the superiority of ANAN's DSP functionality.
In the company I operated, we regularly benchmarked ourselves against our competitors--this is a key method for product improvement. My suggestion is that FlexRadio honestly benchmark themselves against the competition and use that information to develop the specifications for improved DSP functionality.
When FlexRadio introduced the "knob" versions of the 6400 and 6600, Flex entered the mainstream transceiver market. Hams in the "knob" radio market space are used to noise mitigation features that work well and expect a transceiver in the FlexRadio price range to have these features.
The relatively poor noise mitigation features of the Flex 6000 series have been repeatedly discussed at length over several years without any substantial improvement. I don't don't think it is unreasonable to ask FlexRadio to address these issues.
I want it to work as good or better than my 20 year old Icom 706 MKIIG. I don't think that is too much to ask. Almost any knob and button radio made in the last 20 years has a better ANF than my 2018 Flex 6400.
Go to 40, 80 or 160 SSB any night and turn it on. Tell me what you think? You can't miss it failing miserably with +S9 signals..
Flex doesn't need specific engineering input. They know it's broke and should know how to fix or they wouldn't be good software based electrical engineers.. It's not like a bug that just happened. Every hardware DSP chip made in last 20 years works decent. Sad, but fixable...someday. I can wait.
AGC on slow througout.
ex:1 signal, LSB 7233, S9, carrier 7230 S9. BW 3.3Khz. ANF seems to have no effect at any setting.
ex: 2 tuned to 7230.950 LSB BW 2.4 kHz, many overlapping SSB. carrier at S9 on 7230, ANF setting has an effect but very small and carrier still audible.
ex: 3 tuning around 7230 to 7233 BW at 2.4, location of the offending carrier has an effect on the ANF performance. As the carrier is closer to the the tuned frequency the ANF performance is better but still does not knock out the carrier.
ex: 4 moved to 7300 SW station, BW 2.4khz, LSB, carrier at +20db over, tuned to 7301, so 1 Khz tone, ANF knocks the heterodyne completely out with some variation on modulation peaks. Variation is minor.
ex: 5 moved to 7315 SW station with carrier at +30 to +40 QSB, tuned to 7316 and ANF eliminates the heterodyne- AT ANY SETTING. Once again, modulation peaks cause ANF performance to vary, but again, minor.
ex: 6 on 7270.7 an S6 carrier, slightly above background noise, at 7269.7, ANF had no effect at any setting. BW 2.4Khz.
It's just data and a few scenarios. But I would say that ANF performance is a function of the intended signal, and it's modulation peaks, AND the background atmos noise.
Can anyone replicate and/or add to the data?
As a one off, I tried the ANF on CW signals, LSB mode two or three signals in the passband. The ANF takes out one but not all and notch effect varies with keying, that is there is a time to attack the heterodyne and time to release. No real measurement here, but the release time seems shorter than the attack time. ANF can take out one heterdyne only based on this informal test.
I feel this is a minor step in characterizing the ANF in real world scenarios and the results, though far from definitive, do seem to substantiate the comments posted by others in the thread.
In PSDR for the ANF there are settings we can custom. Taps, Gain, Delay, leak.
I wonder if they are making the same changes in the SSDR code mush as the same way in PSDR?
Or is there much more involved.
NB. NR, WNB off, AGC slow, AGCT set at approx knee.
Radio China just came up on 7285 so tuned to 7286 LSB and RC carrier showing about -70 dBm and I have a 1,000 hz heterodyne.
Using Spectrum Lab and ANF OFF the 1,000hz showed about -26 dB.
ANF On at 50% -45 dB
ANF On at 100% -57 dB.
All of these are +/- due to QSB.
But interestingly, the Radio China modulation did not seem to have a major "audible" effect. That is there was very little change that resulted in me hearing the tone. On SL I could see the amplitude of the 1000Hz tone moving up and down with modulation, as expected, but the attenuation from the ANF stayed about the same.
I did a similar run using WWV at 10Mhz and offset by 1000 hz and even with the various voice announcements and time ticks and tones I was still seeing 35 - 40 dB notch on the 1000 hz tone. It never was audible.
This seems to contradict what I found last night, but conditions are different at 2000z versus 0100z with a lot more noise,, signals, and QSB.
I'm going to try Spectrum Labs on the later evening 40m conditions and try to get some meaningful data.
So, yes, the ANF works, and it is really pretty darn good with nulls in some cases of 40 dB. But it seems affected by numerous factors, and using a haphazard search for test cases leaves a lot to be desired in identifying those factors. My guess, the very high level of noise. QRM in the passband is too much data for the ANF to process. That said, I have zero knowledge of how they handle the ANF process; so maybe I'm way off.
The ANF zeroes in, apparently, on the strongest heterodyne and perhaps with multiple heterodynes (40m at 0100z + for example) at near the same amplitude it loses the ability to process the data.
I guess information can be good with a proper dose of skepticism.
If anyone else wants to try Spectrum Lab, it's free, and the input can easily be selected from DAX. It has a bit of a learning curve - it's sort of a Swiss army knife of audio analysis.
It would be very interesting to have more data from users reporting significantly poor ANF performance to see if their scenario can be replicated.
According to Tim Ellison on the FRS Facebook page, when I posed the question to him, if the ANF issues would receive some needed attention, his response was:
"James Whiteway If it wasn't in 3.0.19 and the upcoming maintenance release for 3.0.19, then the answer is no. If we decide to do a 2.6, it might, but that is just speculation on my part. There is no talk at this time about doing a 2.6
I find his answer disappointing, but at least it is a clear, and honest, answer. It gives me an idea as to what, if any, options, I might look for to address this problem for myself.
I can either upgrade to v3.0.10 and hope someday, this issue, along with others, finally receive the attention they need. Or, I can look towards other manufacturers of SDR radios that already have a basic features working to their full potential. ( along with some other, useful to me features)
Time will tell.
Oh, Bill. It appears.you were spot on, with your statement regarding the .v2.50 update. If I came across as
a bit harsh in my response to you, please accept my sincere apologies