Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Scott,
I seem to recall someone mentioning recently (not sure which thread) that sometimes the initial transmit audio (after first opening a new Thetis session) sounds very distorted or robot-like until one of the RingBuffer checkboxes is unchecked then rechecked. I've had that same problem for a long time, to the point that I just make it standard practice to uncheck/recheck a RingBuffer checkbox before my first transmission.
However, I've been playing with the Feedback Gain In and Out (just looking for any effect on the audio), and I noticed that when I reduced the In and Out values to .00000002 then the robot sound on initial audio transmit disappeared, without me ever having to manually reset the RingBuffer checkbox! I haven't researched in-depth any other settings yet but that's what I've noticed at a setting of .00000002.
As a sanity check I've shut down and restarted Thetis repeatedly, and even rebooted the computer and the 7000DLE a number of times, and haven't observed a single instance of the "robot" audio resurfacing since.
No idea what to make of that but I wanted to report my discovery. As a side note, the audio quality otherwise appears to be completely unaffected by the .00000002 Feedback Gain setting.
Mark
I seem to recall someone mentioning recently (not sure which thread) that sometimes the initial transmit audio (after first opening a new Thetis session) sounds very distorted or robot-like until one of the RingBuffer checkboxes is unchecked then rechecked. I've had that same problem for a long time, to the point that I just make it standard practice to uncheck/recheck a RingBuffer checkbox before my first transmission.
However, I've been playing with the Feedback Gain In and Out (just looking for any effect on the audio), and I noticed that when I reduced the In and Out values to .00000002 then the robot sound on initial audio transmit disappeared, without me ever having to manually reset the RingBuffer checkbox! I haven't researched in-depth any other settings yet but that's what I've noticed at a setting of .00000002.
As a sanity check I've shut down and restarted Thetis repeatedly, and even rebooted the computer and the 7000DLE a number of times, and haven't observed a single instance of the "robot" audio resurfacing since.
No idea what to make of that but I wanted to report my discovery. As a side note, the audio quality otherwise appears to be completely unaffected by the .00000002 Feedback Gain setting.
Mark
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
K1LSB wrote:However, I've been playing with the Feedback Gain In and Out (just looking for any effect on the audio), and I noticed that when I reduced the In and Out values to .00000002 then the robot sound on initial audio transmit disappeared, without me ever having to manually reset the RingBuffer checkbox! I haven't researched in-depth any other settings yet but that's what I've noticed at a setting of .00000002.
As a sanity check I've shut down and restarted Thetis repeatedly, and even rebooted the computer and the 7000DLE a number of times, and haven't observed a single instance of the "robot" audio resurfacing since.
Mark
Hi Mark and Scott!
Both of your results comport with our experience while back-porting the NR0V resampler into OpenHPSDR/PowerSDR. In PowerSDR, the loop gain is set to 1.0e-06 (instead of the 4.0e-06 used in Thetis).
Only time will tell, but maybe 1.0e-06 would be closer to a "one size fits all" value for Thetis?
73
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
What your are describing is the startup/initialization stability problem I've been whining (whinging for RichieK1LSB wrote:Scott,
I seem to recall someone mentioning recently (not sure which thread) that sometimes the initial transmit audio (after first opening a new Thetis session) sounds very distorted or robot-like until one of the RingBuffer checkboxes is unchecked then rechecked. I've had that same problem for a long time, to the point that I just make it standard practice to uncheck/recheck a RingBuffer checkbox before my first transmission.
![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
You don't even need to test for a "robot voice". All you need to do is look in Setup > Audio > VAC1 and see the under and overflow counters spinning up. If you don't see them spinning then you are good to go. If you do see them spinning, as I've mentioned many times, it seems the best way to get the resampler re-initialize is the action you describe.
I agree that this definitely helps. The flip side is that you may need to wait a minute or three for the resampler to totally settle down. However this smacks of a band-aid and not a real solution, because, again as I've already described in this topic (you may want to re-read it), I can easily and reliably make the resampler go unstable, primarily by cycling the PortAudio manual checkbox, and also by resetting a device driver.However, I've been playing with the Feedback Gain In and Out (just looking for any effect on the audio), and I noticed that when I reduced the In and Out values to .00000002 then the robot sound on initial audio transmit disappeared, without me ever having to manually reset the RingBuffer checkbox! I haven't researched in-depth any other settings yet but that's what I've noticed at a setting of .00000002.
I'm hoping that Walter's efforts here will bear some fruit
![Smile :)](./images/smilies/icon_e_smile.gif)
My results have not been as 100% perfect as yours. Note there is no reason/need to restart the PC to test this. Merely restarting Thetis is sufficient.As a sanity check I've shut down and restarted Thetis repeatedly, and even rebooted the computer and the 7000DLE a number of times, and haven't observed a single instance of the "robot" audio resurfacing since.
BREAK
Bryan--yes, totally agree. Somewhere between 1.0e-06 and 2.0e-06 seems to be a better starting spot, particularly for those who do not wish to make their own adjustments, unless and until additional tweaks can be made to the algorithm.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
I just finished a 12 hour run. I bumped the feedback gain from 1.0e-06 to 2.0e-06, because, with the advent of Richie's cool graphical readouts, it seemed that the lower feedback gain was not always keeping up with the rate of change in the sample rate differential thereby leading to additional under/overflow counts.
More importantly, I bumped the buffers just a hair up from all zeroes, as shown.
Results, scaled by a factor of 1.5 to match my previous 18 hour test, as posted above, show roughly a 4x improvement in under/overflow performance.
More importantly, I bumped the buffers just a hair up from all zeroes, as shown.
Results, scaled by a factor of 1.5 to match my previous 18 hour test, as posted above, show roughly a 4x improvement in under/overflow performance.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Ah nice one Scott. Whilst very simple, those graphs do make it somewhat easier to visualise what is happening. I am glad that they are of some benefit, I was unsure after implementing them ![Wink ;)](./images/smilies/icon_e_wink.gif)
Ref Walter's post up above requesting the additional values exposed, and initialisation of VAR. Warren has raised the following as possible values that could be exposed :
prop_ringmax
prop_ringmin
ff_ringmax
ff_ringmin
ff_alpha
Warren did let me know that slew was not really that important and probably a 'red herring' as it only has an effect when under/over running. It was only exposed because it was highlighted as changed in Walter's initial 'supercritical' message. I'll have a look at exposing those above, but it will probably take a while as free time is short at the moment and I will probably have to liaise with Warren.
R.
![Wink ;)](./images/smilies/icon_e_wink.gif)
Ref Walter's post up above requesting the additional values exposed, and initialisation of VAR. Warren has raised the following as possible values that could be exposed :
prop_ringmax
prop_ringmin
ff_ringmax
ff_ringmin
ff_alpha
Warren did let me know that slew was not really that important and probably a 'red herring' as it only has an effect when under/over running. It was only exposed because it was highlighted as changed in Walter's initial 'supercritical' message. I'll have a look at exposing those above, but it will probably take a while as free time is short at the moment and I will probably have to liaise with Warren.
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
W4WMT wrote:Purely for the future benefit of those searching this forum for possible solutions: Please don't forget that one can always take the lazy way out (as I have done) and simply clock your radio and your sound device from a GPSDO, thus eliminating clock domain skew and need for a variable resampler.
At least two manufacturers (PreSonus and Lynx Studio) offer sound adapters that accept an external clock source.
You can use a dual-ported GPSDO (like the Leo Bodnar) with one port set for 10 MHz (for the radio) and the other for 48 kHz (for the sound device). Or you can simply use two el-cheapo GPSDOs, one set for 10 MHz and the other set for 48 kHz. Works fine
73, Bryan W4WMT
This is not as trivial as it sounds (hah, a pun
![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
Both Lynx and PreSonus offerings with master clock inputs (what the audio engineers call a "word clock", if anyone is researching this themselves), while relatively inexpensive compared to more serious studio stuff, nevertheless are going to set you back between $500 and $1000, depending.
Probably the least expensive route you can take is to find a used PreSonus Studio 192 (rack mount or mobile/desktop flavor). These are discontinued, but are still supported by PreSonus and appear on eBay periodically. I got mine new on sale for $300 when they were being discontinued. Unfortunately, because the leap up into the PreSonus Quantum line is so expensive, the Studio 192's are still commanding high prices even on the used market, between $300 and $400.
So, at the end of the day, interface plus GPSDO is a $600 solution, give or take.
Nevertheless, I may give this a shot because I'm already heavily invested with a Chinese BG7TBL GPSDO at the radio ($150) and a Presonus Studio 192 Mobile ($300) in the studio. I can get the Bodnar mini-GPSDO from Airspy in the US for $145. The only worry is that it has a 3.3V, 50Ω output and the studio audio word clock standard is a 5.2V, 75Ω square wave. So there's a chance it might not work or that I'll have to build a driver circuit for it.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Helmut,
It's wonderful that you have a $1300USD interface, but most people are not going to spend that kind of money.
You should probably post your PureSignal problem in the main 2.8.11 topic, it does not belong here.
It's wonderful that you have a $1300USD interface, but most people are not going to spend that kind of money.
You should probably post your PureSignal problem in the main 2.8.11 topic, it does not belong here.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
w-u-2-o wrote:I can get the Bodnar mini-GPSDO from Airspy in the US for $145. The only worry is that it has a 3.3V, 50Ω output and the studio audio word clock standard is a 5.2V, 75Ω square wave. So there's a chance it might not work or that I'll have to build a driver circuit for it.
You may luck out anyway. My 2-port Bodnar drives my Studio 192 Mobile wordclock input just fine. Perhaps they use the same output network in the mini?
That's a bummer about the Studio 192 Mobile being discontinued! Seems like our options are shrinking rather than expanding
![Sad :-(](./images/smilies/icon_e_sad.gif)
73
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
When you are unable to sleep.... code
So many adjustments. Getting there, VAC1 mostly done.
R.
![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
So many adjustments. Getting there, VAC1 mostly done.
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
I loved the name of the jpeg: scared1.jpg ![Smile :-)](./images/smilies/icon_e_smile.gif)
![Smile :-)](./images/smilies/icon_e_smile.gif)
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
W4WMT wrote:I loved the name of the jpeg: scared1.jpg
hah yep
![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
For those testing Slew and reporting that no changes are observed - well, due to a slight oversight, that is 100% correct, as it does nothing in the current version.
![Embarrassed :oops:](./images/smilies/icon_redface.gif)
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
w-u-2-o wrote:I can get the Bodnar mini-GPSDO from Airspy in the US for $145. The only worry is that it has a 3.3V, 50Ω output and the studio audio word clock standard is a 5.2V, 75Ω square wave. So there's a chance it might not work or that I'll have to build a driver circuit for it.
My standard Bodnar dual-port GPSDO has been working fine with a Studio Mobile 192, 7000DLE, and VMP software for nearly two years. No input level, nor drop-out issues. It's one of the more stable pieces to the puzzle. Port A = 10 MHz; Port B = 48 kHz.
As I recall, the Bodnar Mini model is available with only one output port.
Paul, W9AC
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
I only need a single 48KHz output. The studio/office where the Presonus is is two floors above the radio rack. The rack already has a 10MHz output GPSDO.
So we'll see how well time transfer and phase lock works across two GPSDOs. It should be fine. I've done this many times before on the job. But that is with more sophisticated and expensive equipment. It remains to be seen how this inexpensive stuff performs.
So we'll see how well time transfer and phase lock works across two GPSDOs. It should be fine. I've done this many times before on the job. But that is with more sophisticated and expensive equipment. It remains to be seen how this inexpensive stuff performs.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
w-u-2-o wrote:I only need a single 48KHz output. The studio/office where the Presonus is is two floors above the radio rack. The rack already has a 10MHz output GPSDO.
So we'll see how well time transfer and phase lock works across two GPSDOs. It should be fine. I've done this many times before on the job. But that is with more sophisticated and expensive equipment. It remains to be seen how this inexpensive stuff performs.
Works just fine here with two GPSDOs! My guess is it will work well there also.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Got the Bodnar unit.
The Presonus reports a lock at any output level from 8 to 32ma.
Performance is no better than with the Presonus internal reference, which is quite disappointing. Say 3 or 4 counts per hour in the worst case under/over-flow counter.
With the force option checkboxes checked, under and overflow counters accumulate counts at rates similar to that achieved with the internal reference. Also, the resampler appears to continue to run in parallel, and the var ratio, ringbuffer level and graphs all show the same type of activity one would see with the Presonus internal reference. I would have expected any var ratio measured to be very, very small and that buffer levels would be pinned near 50%. Not even close.
Unchecking the force boxes and letting the resampler act on the Bodnar reference, and the GPSDO at the 8000, again no real difference in behavior noticed.
I would have expected this to be a slam dunk, easy day, zeroes across the board move, but it hasn't been for me so far.
Still testing at every possible Bodnar drive level. I'm on my last one now, 24ma. Perhaps that will do better. Maybe I should put a o'scope on it.
BREAK
Bryan--what do you see for var ratio and buffer levels with your force options checked and unchecked?
The Presonus reports a lock at any output level from 8 to 32ma.
Performance is no better than with the Presonus internal reference, which is quite disappointing. Say 3 or 4 counts per hour in the worst case under/over-flow counter.
With the force option checkboxes checked, under and overflow counters accumulate counts at rates similar to that achieved with the internal reference. Also, the resampler appears to continue to run in parallel, and the var ratio, ringbuffer level and graphs all show the same type of activity one would see with the Presonus internal reference. I would have expected any var ratio measured to be very, very small and that buffer levels would be pinned near 50%. Not even close.
Unchecking the force boxes and letting the resampler act on the Bodnar reference, and the GPSDO at the 8000, again no real difference in behavior noticed.
I would have expected this to be a slam dunk, easy day, zeroes across the board move, but it hasn't been for me so far.
Still testing at every possible Bodnar drive level. I'm on my last one now, 24ma. Perhaps that will do better. Maybe I should put a o'scope on it.
BREAK
Bryan--what do you see for var ratio and buffer levels with your force options checked and unchecked?
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
I think the var ratio will always calculate and update. The part of the code that uses it gets given the 'forced' value instead if forced is checked.
R.
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
ramdor wrote:I think the var ratio will always calculate and update. The part of the code that uses it gets given the 'forced' value instead if forced is checked.
This is correct, as confirmed to me by Warren NR0V via Email.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
w-u-2-o wrote:Bryan--what do you see for var ratio and buffer levels with your force options checked and unchecked?
Hi Scott,
With the 7000 and the sound device both clocked by GPSDO (force=1), my var ratio will slowly hunt around in the last two significant digits. This is completely normal (confirmed by NR0V) because of "jitter" in the sample streams (imposed by imperfect thread schedule timing in the Windows CLR) creates a little bit of noise in the matching loop filter. But with force set to 1, the var ratio is ignored and the code reverts to a normal fixed ratio resampler.
But even with clock domains synced, the ring buffers still have to be big enough to handle the OS imposed jitter (mentioned above). In your case, 3 or 4 overflows/underflows per hour sounds like a reasonable compromise. Given enough cores, unused by competing applications running on the machine, this can by driven to zero.
73
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
First of all, and perhaps most importantly, the results achieved here are not so dramatically better than using the resampler with the normal internal references. That is high praise for the resampler algorithm. If a means to make the resampler more stable on startup could be found, and to more gracefully handle the error condition of an external driver reset, a GPS derived external reference on the sound interface really isn't necessary.
Pictured below are overnight (approx. 10 hours) results with the GPS reference working in concert with the resampler. I chose this approach because yesterday initial short term testing did not yield good results with Force = 1.0 turned on.
While the PC is not very busy overnight except with backups, this is still the best overnight test I've ever achieved.
In order to achieve these results I had to set the drive on the Bodnar unit to 24ma, which gave the best performance, and set the VAC feedback gain to 1E-06. This reliably kept the var ratio from exceeding 1±1E-5, i.e. no var ratio greater than 1.000010 or less than 0.999990.
In further testing today I'm going to investigate the effects of an even lower feedback gain, and also the effects of changing to Force = 1.0.
Pictured below are overnight (approx. 10 hours) results with the GPS reference working in concert with the resampler. I chose this approach because yesterday initial short term testing did not yield good results with Force = 1.0 turned on.
While the PC is not very busy overnight except with backups, this is still the best overnight test I've ever achieved.
In order to achieve these results I had to set the drive on the Bodnar unit to 24ma, which gave the best performance, and set the VAC feedback gain to 1E-06. This reliably kept the var ratio from exceeding 1±1E-5, i.e. no var ratio greater than 1.000010 or less than 0.999990.
In further testing today I'm going to investigate the effects of an even lower feedback gain, and also the effects of changing to Force = 1.0.
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Some additional results from today. Running approx. 4 hours time blocks just to get some better data. Better performance equates to lower under/over-flow count rates.
- Continuing with the approach I took in my post above, the GPS reference in concert with the resampler, but now with the feedback gain lowered to 1E-07, performance remained approximately the same.
- Activating force = 1.0: just slightly better than the above settings, but that is subjective. A more formal and length test is really necessary.
- With or without force turned on, with the GPS reference in use one can watch the var ratio be much better behaved overall. In fact, with force on, the feedback gain set to 1E-07 keeps the var ratio from ever varying more than ±9E-07, which is amazing. This tends to support the fact that the GPS reference really is working to keep much better time than the internal reference of the Presonus interface.
- The fact that the resampler keeps the difference, as measured in under/over-flow counts per hour, nearly the same under all conditions, continues to speak very highly of the resampler algorithm. It really tends to make a case for NOT buying a GPS reference for the audio side.
- Even after the var ratio has had a chance to settle down, it would appear that PC activity seems to perturb it quite often. One can watch obvious discontinuities appear in the var ratio. One wonders if there should not be some sort of low pass filter placed on the output of the loop that sets the var ratio.
- Most interestingly, with the GPS reference in use, either with the resampler or not, any buffer settings above zero make the under/overflow count rate dramatically higher. For instance, by setting all buffers to 10ms, within just a few minutes I had more than 10 counts. This I cannot explain.
73!
Scott
- Continuing with the approach I took in my post above, the GPS reference in concert with the resampler, but now with the feedback gain lowered to 1E-07, performance remained approximately the same.
- Activating force = 1.0: just slightly better than the above settings, but that is subjective. A more formal and length test is really necessary.
- With or without force turned on, with the GPS reference in use one can watch the var ratio be much better behaved overall. In fact, with force on, the feedback gain set to 1E-07 keeps the var ratio from ever varying more than ±9E-07, which is amazing. This tends to support the fact that the GPS reference really is working to keep much better time than the internal reference of the Presonus interface.
- The fact that the resampler keeps the difference, as measured in under/over-flow counts per hour, nearly the same under all conditions, continues to speak very highly of the resampler algorithm. It really tends to make a case for NOT buying a GPS reference for the audio side.
- Even after the var ratio has had a chance to settle down, it would appear that PC activity seems to perturb it quite often. One can watch obvious discontinuities appear in the var ratio. One wonders if there should not be some sort of low pass filter placed on the output of the loop that sets the var ratio.
- Most interestingly, with the GPS reference in use, either with the resampler or not, any buffer settings above zero make the under/overflow count rate dramatically higher. For instance, by setting all buffers to 10ms, within just a few minutes I had more than 10 counts. This I cannot explain.
73!
Scott
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
w-u-2-o wrote:Even after the var ratio has had a chance to settle down, it would appear that PC activity seems to perturb it quite often. One can watch obvious discontinuities appear in the var ratio. One wonders if there should not be some sort of low pass filter placed on the output of the loop that sets the var ratio.
Ditto. I see the same. However, your results still look very good and slightly better than what I see when using the VMP Virtual ASIO driver. With the Studio 192 Mobile ASIO driver, I see no over/underflows. In either case, buffer size is 128 and 48 kHz sample rate. All boxes are checked, meaning Force and Buffer Latency with those dialed down to zero just as you show. Any means to tweak anything in VMP for better results?
I tried changing my Bodnar's output drive strength from 32 mA to 24 mA. FWIW, no change.
Paul, W9AC
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Paul,
In the latest versions of Voicemeeter, under the channel name for each of the three virtual channels, you will see the Voicemeeter internal sample rate and buffer size shown. Right click on that and you can select smaller buffer sizes.
I had messed with this once, a long time ago. My own results were poor when going below the recommended 7168 sample buffer size. Also, it didn't seem to make a difference. I had not yet remoted my station to the basement so I was able to easily use an oscilloscope to measure transmit latency directly. Lowering the Voicemeeter buffer size made little difference at that time. It might be different now.
I don't think this will have any positive effects on under/over-flow count rates. Latency is a different topic; if you want to go on a latency witch hunt by all means start a separate topic about that
Certainly the complexity of my software audio chain is contributing to non-zero counts. I bet if I just go direct to/from the Presonus to Thetis VAC it would be all zeroes all the time. I'm not sure why my results would be better than yours. Consider that my chain looks like this:
Presonus (ASIO) Voicemeeter (ASIO) Reaper (ASIO) Voicemeeter (ASIO) Thetis
And that's in both directions, left to right for transmit, right to left for receive.
I have noted that I can set the Presonus buffer size to 128 or even 64 when Reaper is not involved, but Reaper audio gets glitchy at less than 256. Perhaps you might try bumping the buffer to 256. Make sure to match that in Thetis VAC setup, it does not pull the buffer size automatically from the driver like Voicemeeter and Reaper do.
73,
Scott
In the latest versions of Voicemeeter, under the channel name for each of the three virtual channels, you will see the Voicemeeter internal sample rate and buffer size shown. Right click on that and you can select smaller buffer sizes.
I had messed with this once, a long time ago. My own results were poor when going below the recommended 7168 sample buffer size. Also, it didn't seem to make a difference. I had not yet remoted my station to the basement so I was able to easily use an oscilloscope to measure transmit latency directly. Lowering the Voicemeeter buffer size made little difference at that time. It might be different now.
I don't think this will have any positive effects on under/over-flow count rates. Latency is a different topic; if you want to go on a latency witch hunt by all means start a separate topic about that
![Smile :)](./images/smilies/icon_e_smile.gif)
Certainly the complexity of my software audio chain is contributing to non-zero counts. I bet if I just go direct to/from the Presonus to Thetis VAC it would be all zeroes all the time. I'm not sure why my results would be better than yours. Consider that my chain looks like this:
Presonus (ASIO) Voicemeeter (ASIO) Reaper (ASIO) Voicemeeter (ASIO) Thetis
And that's in both directions, left to right for transmit, right to left for receive.
I have noted that I can set the Presonus buffer size to 128 or even 64 when Reaper is not involved, but Reaper audio gets glitchy at less than 256. Perhaps you might try bumping the buffer to 256. Make sure to match that in Thetis VAC setup, it does not pull the buffer size automatically from the driver like Voicemeeter and Reaper do.
73,
Scott
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
w-u-2-o wrote:I have noted that I can set the Presonus buffer size to 128 or even 64 when Reaper is not involved, but Reaper audio gets glitchy at less than 256. Perhaps you might try bumping the buffer to 256. Make sure to match that in Thetis VAC setup, it does not pull the buffer size automatically from the driver like Voicemeeter and Reaper do.
Scott
Thanks, Scott. I'll give those suggestions a try.
Paul, W9AC
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
First of all, I have to say I really like the FF Alpha value is now exposed. I haven't messed with anything but that and the Feedback Gain, but by slowing both of those settings down things seem much more stable. The values I've settled on are:
Feedback Gain: 0.0000001 (1E-07)
FF Alpha: 0.005
The only downside is you have to wait a few minutes for stability to occur. You can actually watch the accumulation of under/over flows slow down and then ultimately stop, and watch the neat graph that Richie has provided us flatten out.
RESULTS
I gave three different configurations each 12 hours to "soak" and accumulate under/over flow counts. I zeroed each counter after 5 minutes, which seemed an adequate settling time as described above. PC activity during each 12 hour period was light and typical. Most of the time the PC was left alone.
In every case the Var Ratio settled down and stayed within ±2E-06 of the final value.
In every case the stabilization took approx. 5 minutes. This was even true when Force was turned on. @ramdor this implies that there is a resampler related function still working even with Force set on. I watched the under/over flow counters running and the count rate slowly decrease. That should not be the case if Force really equals "fixed".
In the following results counts are shown as OUT Over/Under and IN Over/Under.
1) Presonus internal reference with resampler: OUT 17/2 IN 1/11. Total counts of any type: 31. Counts per hour: 2.6.
2) Presonus GPS reference with resampler: OUT 9/5 IN 3/4. Total counts of any type: 21. Counts per hour: 1.8.
3) Presonus GPS reference Force=ON: OUT 18/7 IN 6/7. Total counts of any type: 38. Counts per hour: 3.2.
So something weird is going on. We get the expected performance improvement going from configuration 1 to 2, but we take a BIG step back in the configuration that supposedly ought to work the best, configuration 3. Combined with the unexpected appearance of resampler-like settling activity noted in configuration 3, that begs the question of whether or not the Force checkbox is having the desired effect.
Thanks!
Scott
Feedback Gain: 0.0000001 (1E-07)
FF Alpha: 0.005
The only downside is you have to wait a few minutes for stability to occur. You can actually watch the accumulation of under/over flows slow down and then ultimately stop, and watch the neat graph that Richie has provided us flatten out.
RESULTS
I gave three different configurations each 12 hours to "soak" and accumulate under/over flow counts. I zeroed each counter after 5 minutes, which seemed an adequate settling time as described above. PC activity during each 12 hour period was light and typical. Most of the time the PC was left alone.
In every case the Var Ratio settled down and stayed within ±2E-06 of the final value.
In every case the stabilization took approx. 5 minutes. This was even true when Force was turned on. @ramdor this implies that there is a resampler related function still working even with Force set on. I watched the under/over flow counters running and the count rate slowly decrease. That should not be the case if Force really equals "fixed".
In the following results counts are shown as OUT Over/Under and IN Over/Under.
1) Presonus internal reference with resampler: OUT 17/2 IN 1/11. Total counts of any type: 31. Counts per hour: 2.6.
2) Presonus GPS reference with resampler: OUT 9/5 IN 3/4. Total counts of any type: 21. Counts per hour: 1.8.
3) Presonus GPS reference Force=ON: OUT 18/7 IN 6/7. Total counts of any type: 38. Counts per hour: 3.2.
So something weird is going on. We get the expected performance improvement going from configuration 1 to 2, but we take a BIG step back in the configuration that supposedly ought to work the best, configuration 3. Combined with the unexpected appearance of resampler-like settling activity noted in configuration 3, that begs the question of whether or not the Force checkbox is having the desired effect.
Thanks!
Scott
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Something i've just spotted, force flag is getting lost in the resampler if anything on the advanced tab is changed. Some of them call a function to init the matcher, this function also resets the force flag, and the forced value.
R.
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Scott,
FWIW, my latest VAC1 and Advanced settings appear to eliminate the need to uncheck/recheck the RingBuffer boxes after a Thetis session startup (at least for me).
I don't have any fancy clock hardware. And I've left VAC2 alone for now, focusing on VAC1.
Here are my current settings:
Mark
FWIW, my latest VAC1 and Advanced settings appear to eliminate the need to uncheck/recheck the RingBuffer boxes after a Thetis session startup (at least for me).
I don't have any fancy clock hardware. And I've left VAC2 alone for now, focusing on VAC1.
Here are my current settings:
Mark
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
So yep, bug with forced being undone, sorry to say Scott.
I will get a build out in a few hours when I have time to check implications.
R.
I will get a build out in a few hours when I have time to check implications.
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
Also, do you want var to return forcedVar if it is enabled. The get diags function from rmatch always return var, which is always processing regardless. When var comes to be used a check is made for the forced flag, if set fvar is used, and not then var. It would give the var diag display in setup (and the grapher) the forced value that was being used if set.
R.
R.
Richie - MW0LGE - https://www.qrz.com/db/mw0lge
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Latest Release [2.10.3.5] : https://github.com/ramdor/Thetis/releases/tag/v2.10.3.5
Latest Work In Progress : https://github.com/ramdor/Thetis/releas ... .3.6-dev_2
Re: Thetis 2.8.11-21+++ VAC advanced controls: results and discussion
ramdor wrote:So yep, bug with forced being undone, sorry to say Scott.
I will get a build out in a few hours when I have time to check implications.
R.
No worries, Richie! If it was easy then anyone could do it
![Very Happy :D](./images/smilies/icon_e_biggrin.gif)
I'll start what will no doubt be a short term experiment. I've gone back to VAC defaults and restarted with GPS reference and Force activated, just for grins.
Also, do you want var to return forcedVar if it is enabled. The get diags function from rmatch always return var, which is always processing regardless. When var comes to be used a check is made for the forced flag, if set fvar is used, and not var.
Not 100% sure I'm understanding this question. If you mean do I want the var ratio display to show "1.000" vs. a live var measurement, personally I'd prefer the live var measurement because it allows one to continue to study the performance of the resampler even while the var ratio is fixed.
This also begs a question: with respect to the buffer occupancy stat's, are those being developed and presented both numerically and graphically on the basis of the actual var in use? I.e, if force is enabled, those stat's are based on the forcedVar?
Thanks for looking at that so quickly!
Cheers!
Scott