tx buffer underrun and clicking

[Alan Hopper]

Hi Steve and group,

I wonder if the radio should stay in transmit if the tx buffer underruns, it could send out zero signal and ramp down from the last value to avoid too much spurious stuff, I suspect the output would be cleaner than the current situation when network problems occur and there would be less stress on the relays and pa. There is still the watchdog to capture bigger problems.  Problems should still be visible on the panadapter assuming duplex is used.  Just a thought, I guess there are good reasons for it to be as it is.

I've been a bit slow adding tx buffer size control and reporting to Spark as I've never had a problem or any specific reports, It is getting near the top of the list though.

73 Alan M0NNB

[ron.ni...@gmail.com]

Hi Alan,

Have you tried testing Spark while connecting to the HL2 over WiFi?  I find that testing HL2 Tx using a laptop computer or mobile device, and just moving some distance away from the WiFi access point, reliably causes problems (ep6 errors and relay clicking).  Not sure what other factors may be involved (Mac vs. Pi vs. PC, 2.4G vs 5G, how many SSIDs are active, BLE devices, RFI, etc.)

73,

Ron

n6ywu

[Alan Hopper]

  Hi Ron,

I use it quite a lot over wifi without issues but as you pointed out before I'm in a quiet wifi area. At some point I shall try using at longer range to see if I can optimise further.

73 Alan M0NNB

[Matthew]

An interesting proposal and I'm not sure of the best way forward.

Over the summer I did a moderate amount of experimenting to understand some of these buffer issues in real-terms (with various buffer sizes and experimental gateware with larger buffer sizes). Sometimes I thought I had the tx buffer set just right, then occasionally I would get 1 short blip. If this was when using a data mode (e.g. ft8) it would cut out and stop for that transmission? It would then depend on how the SDR software and WSJT-X handle external PTT being turned off. I guess on the next transmission cycle it would pick up again, but it occurs that if every transmission cycle you have 1 blip (likely), this might mean your data message never gets through. Whereas this short blip may not affect the decode at the other end?

It feels like software should detect these (protocol tells us when they have occurred) and act upon it (frequency of occurrence based decision?) rather than the HL2 gateware make a decision that might not suit everyone?

Like you, I am starting to look at software making dynamic buffer size control decisions to better accommodate network latency/jitter.

73 Matthew M5EVT.

[Steve Haynal]

Hi Alan,

Have you set PTT or CW hang time? These implement exactly what you describe - how much time after an underrun the HL2 remains in TX in hopes of more TX data. The PTT hang time is for TX that is started with the MOX bit. The CW hang time is for TX that is started via an external CW key or CWX. PTT can be set up to 31ms. CW hang time up to 1023ms.

73,

Steve

kf7o

[Alan Hopper]

Hi Steve,

I've only just started looking at this and have never personally had any relay clicking. I really just wondered why under run should ever turn off tx, (other than the watchdog), my initial thought was that it did more harm than good. I'll look at the ptt hang time.

73 Alan M0NNB

[Alan Hopper]

Hi Steve,

I may have found a reason that people don't report relay clicking with spark, In adding control of tx latency and ptt hangtime I discovered that I had been setting rx13 vfo on these c&c bits ( for cva9 HL1 use), this defaults to 10MHz which I think results in 0ms tx buffer latency and 22ms ptt hang time. I'm a bit suprised it works with these settings but I've spent many hours transmitting and staring at the duplex feedback in tweaking fm and am recently and not seen any blips.

73 Alan M0NNB

[Steve Haynal]

Hi Alan,

That is very interesting. I would expect the high PTT hang value to prevent relay clicking as you stay in transmit, but the 0 TX buffer latency to result in TX drops. Do you think this only affects SparkSDR? I did test TX buffer latency and PTT hang with Quisk. I could reduce the TX buffer latency until I started seeing and hearing TX drops and relay chatter, and then increase PTT hang so that the relay chatter was masked but the TX drops were still present. My advice to users with these types of problems is still to find a good TX buffer latency and PTT hang time setting for your network. These values can be adjusted with hermeslite.py if your preferred software doesn't implement this.

73,

Steve

kf7o

[Alan Hopper]

  Hi Steve,

As far as I can remember I think it was only spark that supported 32 receivers.  I guess it is possible other software sends stuff on these bytes as it would go undetected but I'm not aware of it, I'll have a play with my emulator and see.

I'll do some tests with changing these settings. I'm also surprised it appears to work with latency at zero, I'll double check that I really am sending 0.

73 Alan M0NNB

[Chris Gerber]

Hi All

Maybe this mail I sent to Martin if of your interest?

Hello Martin

One problem I have seen on all other programs, like HL2 Beta6 as well as Simons Console also exist

on Thetis to some extend. I add a picture so you can see your self. It a cracking noise on the RF when

starting transmit it can be Tune or any other modulated mode

I use mostly Mixw2 but also FLdigi and all create those starting cracks at the beginning (Not Relay's)

I tested it with Several Windows10, 8  installs, 64 and 32 bits also have two HL2 here, same thing. Its rebroducable

and was heard first from a qso partner as  I was not aware of it, till i started investigating it with monitoring

on a second RX as well as scope RF onto a dummy load.

I am pusseld, what you think about it, no other SDR Radios as Sunsdr2 OR Anan10, or any other have those effects.

Test was simple, pressing Tune with about 1.5W ino dummy load and picked up RF for Storage scope.

I aaded two pics. The Frequency was at 14.1Mhz.

It can be heard underneath the Audio as well. I have no idea where it comes from but checked all changed

PSU's. Shielded LAN cable, all what I tought could cause it.

73 Chris HB9BDM

--
You received this message because you are subscribed to the Google Groups "Hermes-Lite" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hermes-lite...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/3c22866a-929f-49ed-b9c2-069ee72875ebn%40googlegroups.com.

[Alan Hopper]

Hi Steve and Chris,

Chris I did not see this when I was trying to get cw latency down in Spark but that was a while ago so I'll get the scope out and try again.

Steve, I've been playing with latency and ptt hang controls, I suspect the lack of relay clicking in Spark is partly luck on the value I was sending to ptt hang and partly low jitter.  I've noticed that if a very low latency is set  the reported buffer position/latency soon climbs to a similar value to a high setting once an underrun has occurred.  I've been loading my network to stress this and noticed that underruns are often followed by overruns (always 4 in a row), I guess this is because delays are more common than dropped packets so the buffer corrections have to be undone as all the expected packets do eventually turn up in a rush and the original timing is resumed. I'm just thinking out loud here and have not looked at the rtl yet.  I'm assuming the under and overrun bits reset themselves after being sent.

73 Alan M0NNB

[Chris Gerber]

Hi Jonas

Its not that simple. First I use direct connection LAN. Even exchanged Lan Cards, cables

tried all LAN Bios setup setings, to hope to solve it. No chance and that and two full

independet system, They behave the same way. Also tried all paras what could have an effect

on it. Must say its not allwys present but at about 60 to 70% of TX starts.

Good old Hermes on Aanan 10 never shown such signs. I will proceed with tests

on Spark, also with FT8 programms to Console, HPsdr Beta6 and new Thetis as I

am well equipped with testing gear here.

73 Chris HB9BDM

Am 11.11.2020 um 19:31 schrieb Jonas Sanamon:

Hi Chris,

Could it be network packet loss, or buffer underruns?  

Is this over LAN or with direct ethernet cable HL2 -> PC? 

If not direct, test that?

Regards, Jonas

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/ece6b4f9-6982-28ab-f15c-27e4a84754f0%40swissonline.ch.

[Chris Gerber]

Hi Alan

Here the starting phase on Spark 2. which look identical on both my systems.

I use Win10/64, Intel I7 8 Core here. Sparks runs well, but came aware
of the

very odd starting phase just when I found those carack I reported on
other programs

I press Tune button, with set to about 3W and see a first starting Pulse
appearing,

then a Pause, then a steady signal with same signal value as on start,
but now cont.

I have done a lot of such work many years back togehter with Peter
Martinez G3PLX

invoking Amtor ARQ, as at those days Transceivers where not fast enough for

a fast 10 to 20ms switching time needed for Amtor ARQ.

I tried with Flex together some years ago on their Flex5000 to work fast
ARQ modes, and only

with setting  Buffer sizes to smallest possible range we finally made
it. But the filtershape was

then all gone.

73 Chris HB9BDM

[Alan Hopper]

Hi Chris,

yep on tune you will get an odd respose because of this https://groups.google.com/g/hermes-lite/c/qnselynGLYE/m/j8g-N6rFAQAJ something todo with setting the tuner bit.

73 Alan M0NNB

[Steve Haynal]

Hi Chris and Group,

If the tuner is enabled, you will see some signal drops in the tune signal as the tuner goes through the FSM. Try with the ATU feature disabled:

https://github.com/softerhardware/Hermes-Lite2/tree/master/hardware/enclosure/endcaps/kf7o/hl2_40b#powersdr

Which gateware version are you running? Please try with the latest 72p5. Later versions have double the TX buffer and more range for TX buffer latency and PTT hang time settings.

What values are you setting for TX buffer latency and PTT hang time? How are you setting them? If not supported by the software you are using, have you set them via hermeslite.py?

Which software (exact version numbers) are you using? If PowerSDR, are you using the version from Reid?

Have you tried Quisk, LinHPSDR and piHPSDR?

The Sunsdr2 OR Anan10 may not be optimizing for latency and run different gateware. Software which sends batches of packets to the HL2 can be problematic. Quisk and SparkSDR and possibly other software do not do that.

I will try to write a simple hermeslite.py function to gather TX buffer data for people who experience this problem. We need to see exactly what is happening in your situation. Be prepared to run hermeslite.py.

 https://github.com/softerhardware/Hermes-Lite2/tree/master/software/hermeslite

73,

Steve

kf7o

[Chris Gerber]

Hi

Well I shall try the newest  Gateware,  but I have been carefully using all newest  avaiable software

before coming out with the mail, Geirs Beta6 version even new Thetis version with the extra .exe file

Simons latest Console !

No Tuner attached here.

I did adjust all what could have an effect on timing as I am no beginner on Digi connections, all failed, I shall give a chance and load up new Gatware

Dont have Hermeslite.pi installed here, only Spark2. I dont sent bursts. I use mostly Olivia 1000/32 mode or also RTTY

but as you can see on Pics the scope shot was taken with only using TUNE no other program installed to any

of the checked Programs like HPSDR, from Geir, or the original 3.5 version also Simons Console

One HL2 has an old Gateware version the other one 71.3.

Its only during the TX startup. Checkes Latency on my computers all green.

I just want to use HL2 as any other of my SDR's as a SW Transceiver to communicate toghter with my

good fast Windows10 PC system.

I do know all those used Digi programs well and know where to set Paras.

But the scope pics speaks for itself.

Next step here to find new Gateware and see, if ok fine, if not I might dispose HL2, as so far its cost less in price

but a enormous amout of time and  nerves.

73 Chris HB9BDM

--
You received this message because you are subscribed to the Google Groups "Hermes-Lite" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hermes-lite...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/4e7304e5-820b-45dd-8e0e-e1fcf061c786n%40googlegroups.com.

[Alan Hopper]

Hi Chris,

when Steve mentioned sending bursts he was refering to some sdr software sending the udp packets out in bursts, powersdr certainly used to do this, this is not something you have control of.

The tune button on spark currently sets a bit to trigger a tuner, I had not realised quite what it did, this causes the .5s ish delay you saw, It should not be there with the normal ptt, your other images look like this is not the problem with other software.

As you say this is nothing todo with external digi programs. I don't see it here with Spark unless I really stress my network, this maybe because the ptt hang time is accidentaly set quite high in spark.

I'm sure we can get to the bottom of this.

73 Alan M0NNB

[Joe LB1HI]

Hi All,
Also tested with Key down CW emission. Effect is the same as with tune. Maybe a little less, but there is also a distorted signal and clicking relays.

It occurs to the smallest extent (almost never)in OpenHPSDR-PowerSDR mRX PS v3.4.9 control software. 

after new installation of the widows system.

But after installing additional SDR software (OpenHPSDR Beta, Thetis, Red Pitaya/Charly25, Console V3, Quisk)   often causes disruptions to occur even in OpenHPSDR-PowerSDR mRX PS v3.4.9. But a database reset often helped.

73, Joe
PS. This problem not appear in Spark at all   

[Steve Haynal]

Hi Chris,

I do not experience these problems with my Quisk and WSJT-X setup. Thetis protocol1 support is new, and never written to support the HL2. It may take a little time to have all issues worked out. To get to the bottom of what you may be experiencing, we will need to collect some data with hermeslite.py. If you can't or don't want to do that, then I agree you should sell your HL2. This is an experimental project. I do not want you or others to feel like you are wasting your time or nerves.

73,

Steve

kf7o

[Chris Gerber]

Hi Steve

I changed Gateware to 72.5. The problem still exists, except the cracking noise shifted back

about 500ms, compare to former Gateware.

The starting squence stays now clean, but as you can hear on the added wave files, its still there.

Pic’s and wave files made with Powersdr v3.5 Beta7, and all from Steve paras set!!

Effect also occurring with Console V3 Program !!!

I spent hours or days to fix the observation on al parts, PC’s SW, HW, LAN, etc before

going to the net (Forum). But was not able to solve the problem.

Tested on 14.1Mhz mode USB, HL2 and Powersdr v3.5.Beta7., and Mixw2

Tested:  Pressing Tune Button, or start of Olivia mode on Mixw2

Observing starting  phase of RF signal from HL2

Monitored Sound on Icom7300. Storage Scope OWON

 

My setup here:

Windows 10 pro 64, Supermicro board X11SCA, Intel I7, 8 core

HL2 Gateware 72.5 direct connected to PC via LAN.

Installed Software for HL2: Console V3 Newest version, PowerSDR v3.5 Beta 7 from Geir

All setting for HL2 made. Also Thetis new (not in use anymore). Spark2

Digimode Programs: Mixw2, FLDigi, Multipsk, FT8 all in use and working perfect

Vac Program from MZY v4.60, Cat Eltima Virtual Serial port

 

Wishing list: Can someone of HL2 users also test it under the very same conditions?

There was on LB station who cared for!

I also need help to install Hermesllit.PI for my WIN10 system

Is there a windows software to look at setting on HL2?

If the problem only exist here, sorry, then I exuse mayself for all 'My fake News to the forum' !!

Added Wave files and Pic

 

73 Chris

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/2787bd4e-f7b8-45ae-a072-93ba51d4e1dfn%40googlegroups.com.

[Alan Hopper]

Hi Chris,

Have you tried the digiu mode in Spark, I understand it may not be the program you want to use but might give some clues as to where the problem is as I happen to be looking at this area at the moment. I've also just released a new version with the tune issue fixed and the ability to tweak the hl2 tx buffer and ptt hang.  It is only recently that Reid has started to tweak PowerSDR to work well with the HL2 (which is great) so I suspect it is not so good at sending udp packets with low jitter as software with more HL2 focus. There has been effort put into the HL2 to reduce latency (for remote cw) which may have challenged older software. Your images would be explained by udp packet jitter, either from the sdr sw or your pc or network .  I know Quisk and the hl2 linhpsdr builds have had effort put in to produce low jitter so it would be interesting to know if these or spark work.

73 Alan M0NNB

[Chris Gerber]

Hi Alan

Thanks for the mail. After I have sent off yday the mail to the forum I went again trough all my MS Installs which I use here.

All off them showing the same distortion effect on start. I can repeat it.

This are Spark2, Console, Powersdr Original v3.49, Powersdr v3.50 Beta7 from Geir, also new Thetis

They are all working fine here, except the cracks on starting on ALL of those programs.

I checked, changed all setup setting on each program which could have an effect creating the cracks.

No way, till I came to Geir's new HL2 Powersdr v3.5 beta7, their I found at Setup/HPSDR: TX BUFFER LATENCY of (10) and

PTT HANG of (4). This are the Default settings on install!

I changed those settings and Bingo ! The cracks where gone..

Then checked crack situation with all other installs, Cracks where gone.

So this default setting if not changed, sets HL2 to a MISS BEHAVING value, and it stays as that, as no other program

in use is changing or overwrting it !

Thats the reason that Steve or other are not seeing it, if someone does not have Geir's program installed, or as many user have

programs to set those values with Linux based programs manuelly, the problem is not present.

I wish their would be on all MS based programs as on Geir's version a Latency or PTT Hang  adjsutment, as we had many years back

on Flexs SmartSDR. We always had to adjust it, as those day's  PC's where so slow and different.

So it seems I have found and cured it, it took time to go trough all the setup pages on so many programs

I am still investigating and checking with different values in the program, so dont want to call it a fix!

73 Chris HB9BDM

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/30497fd9-d838-465e-92a9-82b20289435en%40googlegroups.com.

[Matthew]

Hi Chris,

Can I just check I understand this correctly, you are hearing clicks on your transmitted audio but you aren't actually hearing the relay clicking on the HL2?

I can replicate this same behaviour in linHPSDR when I haven't got my software transmit audio buffers big enough. I'm afraid I don't know enough about Windows, but in Linux, using PulseAudio "virtual audio cables" the operating system seems to add large buffers on these virtual connections, the result is bursts of audio data to the SDR software at once. Between these times the SDR software gets 0's from the buffer. This is entirely a fault in software and nothing to do with the HL2 hardware.

Perhaps you can convince a developer of one of your preferred SDR software packages to report if the software audio buffers are empty?

73 Matthew M5EVT.

[Alan Hopper]

Hi Chris,

good to hear you have something working.

It is a good point that software that does not set these values will inherit values left behind by other software as long as the HL2 is powered up. Powering down will reset to the default which I believe is the same as the powersdr beta 7 defaults.  Spark does write to these bits so should not have been affected by previously run software , versions prior to 2.0.3.9 wrote values that could have upset other software, the latest version allows setting both values.

73 Alan M0NNB

[Chris Gerber]

Hi Alan

Correct, once the value was set with Geir's program as default, (Not his fault, or attenion) it stays in the HL2, and would not reset by any other program. Working Powersdr v3.5 Beta7 again, even set the odd default again.

It changed only when I changed the default setting in Powersdr v3.5 bet7 HL2. Then it stayed as that and worked ok.

A question here, is their no one around which could write an MS based program or App, which is identical to Hermeslit.pi

for MS users, which is awfull to install on Windows. I think with hundreds of new Hermes Lite 2 sold and used with MS as me, it would be a great benefit to the HL2 community. As I know there are many straight MS Windows users, and if with one singe install of a good program like Powersdr v3.5 Beta7 HL2 a odd default setting can have such cracking noise effect, and stays stored as here. It would be wort it.

73 Chris

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/bf7f1d43-3322-4c08-87a9-a042cfbaab78n%40googlegroups.com.

[Chris Gerber]

Hi Matthew

Correct, read my last two mail to Alan and you will understand. Its as you  say for Window user its different as their is

no open access or reset possibility with Windows. I only found by accident those cracking noise problem, and I think I am on the track

to fix it, as by changing one or two single setup setting in one program it went away.

73 Chris HB9BDM

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/5feb4b87-3875-42a2-a7e5-a7d8cac2461an%40googlegroups.com.

[Steve Haynal]

Hi Chris and Group,

I'm glad you are making progress. Reid has exposed the TX buffer latency and PTT hang settings I've been talking about. This latency adjustment is actually an enhancement which only the HL2 supports on protocol1. It is nice to have low latency for CW and good networks when possible. It may be that the default values of 10ms latency and 4ms hang are too aggressive for some software. I can increase these in the next gateware if people like. Also, PowerSDR and probably Thetis send bunches of packets at a time instead of spreading them out, which can place additional demands on the TX buffer resources.

This weekend I did enable some additional debug options for the HL2. I have a setup where for every regular packet sent to the standard software, the HL2 also sends a discovery response to the computer registered for port 1025. I can then add additional debug data to the discovery packet. One application of this allows us to look at exactly what is happening with the TX buffer and TX relays inside the FPGA. I've only tried it with Quisk, but everything is working as expected with 10ms latency and 4ms hang on Quisk under Linux. No relay clicks during TX. No drops during TX. I will eventually try some other software. For anyone still experiencing relay click or TX drop problems, I will eventually post this test so that anyone can run it with any software. This will provide the information we need to debug and fix any remaining issues.

73,

Steve

kf7o

[Chris Gerber]

Hi Steve

 

Yes, seems I have fixed my distortion problem here, which must have been present here since my first Powersdr v3.5 Beta HL2 install, and was found accidentally.

I am used to those programs, as used  Flex’s SmartSDR for ages, then HPSDR Stuff from Phil at Perth.  So went trough all installed programs, changed settings, monitoring all the time, but have missed first Geir's Latency/PTT Hang setting, as guessed the cause was somewhere else. After that is was easy to see progress, on all installs.

Anyway I am satisfied, as all my programs running now clean as something, so am glad it’s looked at, from your side as well.

I will not go into details, but as stated in my mails, I am missing an HW/SW checking tool for HL2, on Windows.

I installed Python here and searched for Quisk, Hermeslit.pi, but could not find any thing on James homepage for MS.

I have two handicaps; I am no Linux user anymore, then my age 84.

But thanks anyway for the attention, am sure I was not the only one with it, but it was hard to reckon or see.

 

73 Chris HB9BDM

To view this discussion on the web visit https://groups.google.com/d/msgid/hermes-lite/64d005e7-e320-4381-82c7-b1562b270ca5n%40googlegroups.com.

[didier....@gmail.com]

Hello guys,

Just to provide some feedback regarding relay clicking and underrun issues with the Hermes Lite.

I am using the Hermes Lite locally in my shack connected directly to the Hermes lite with a cat 6 i guess cable . I am using quisk , Pihpsdr , a raspberry PI3 B+ . Basically i have no issue at all and few weeks ago i had some relay clicking on 20m band due to a bad grounding from my tube amplifier. During this time i was using the gateware 72p3

Same with the P3 gateware i am using frequently the Hermes Lite remotely through my wifi network with a Windows 10 computer and it was not the same story since my Wifi network is clearly not the best performer. i had most of the time clicking when i was using SPARK , Quisk , Thetis and powersdr . I tried to adjust many parameters as mentioned by you guys (Ptt latency and tx buffer) without any success. The only software working sometime without relay clicking was Console V3.

Few weeks ago when the 20201107_72p5 gateware was released , the situation change completely and of course locally the hermes lite still wortking like a charm but remotely the latency and buffer adjustements seems to work very well .From now i did some test with SPARK and the latest release for Powersdr beta 8 from Gi8TME/Mi0BOT and definitly it is working really smoothly with my Wifi network (i am using a repeater to access my shack and the signal level is almost at the maximum with a very low latency).

Currently my TX Latency buffer latency is 30 and the TX hang time is adjusted to 10. On Windows 10 computer all firewall are disable. according the router , nothing special and did not apply any QOS or funky parameters.

My 2 cents 

73s F5NPV

[Steve Haynal]

Hi F5NPV,

That is interesting. I can't think of anything which changed from 72p3 to 72p5 which might explain this. They both have the larger TX buffer. 

73,

Steve

kf7o

[ron.ni...@gmail.com]

Hi Chris,

So the default HL2 "TX BUFFER LATENCY of (10) and PTT HANG of (4)" settings did not work for you?

To what values did you have to change these settings to stop the Tx glitching?

Thanks.

73,

Ron

n6ywu

[didier....@gmail.com]

Hello Steve ,

I think i made a mistake when i mention the P3 gateware and i double check tonight and in fact it was the P2 . My sincere appologies.

So , what is the gap regarding the buffer between the p2 and P5 gateware ?

73s Didier

[Steve Haynal]

Hi Didier,

72p2 did have the smaller TX buffer size. 72p1 and 72p5 have the larger (doubled) TX buffer size. That should explain the difference you saw. I plan to keep the larger TX buffer size going forward as I didn't anticipate the range of latencies and jitter in people's networks. Also, I think I will make the default latency 20ms and PTT hang 12ms as not everyone knows how to change those.

73,

Steve

kf7o