Title: NR slightly too strong
Post by: SP9SOY on 11. March 2019, 16:09:47
Hi dear Developers !
Thanks for providing so nice software for the transceiver OVI-40/Sparrow/etc.!
I am using it over 3 months without any issues/problems.
The only thing, that might be considered to be changed, is IMO, the setting of Noise Reduction. It begins on level 5, and it is a bit too much. It would be better, if NR is weaker at level 5, or if setting could begin on "0" or "1" level.
I tried to set the NR action lower with Menu Setting "NR asnr" and it DOES work. But this setting is not saved into Eprom, so one must change it on the every switch on of the transceiver....
Thanks
Bests, Alex SP9SOY |
Title: Re:NR slightly too strong
Post by: DF5LI on 11. March 2019, 19:18:55
Bravo, Alex, dass du dieses Thema mal angesprochen hast !
Ich finde auch, dass die NR viel zu agressiv eingreift; selbst in der niedrigsten Einstellung klingen die Signale gurgelig wie aus der Badewanne ! Vor Monaten gab es mal eine FW-Version, wo die NR wirklich gut funktionierte, aber dann gab es eine "Verbesserung" und sie wurde unbenutzbar.
Die Entwickler sollten sich mal die NR eines IC7300 anhören. Bei kleinen Werten unterdrückt sie die Störgeräusche wirkungsvoll, ohne aus dem Signal eine Live-Übertragung aus einem wild plätschernden Wassereimer zu machen...
EDIT: Ich habe mal nachgesehen: Am 07.12.2017 hatte Frank die NR so modifiziert, dass sie wunderbar funktionierte und ich mich zu einem begeisterten Posting habe hinreissen lassen. Danach wurde die NR wieder kaputt optimiert >:( |
Title: Re:NR slightly too strong
Post by: DF8OE on 12. March 2019, 06:10:33
In der NR vom 07.12.17 war ein falscher Pegel für den Audiolevel. Wenn man die NR eingschaltet hat wurde der gesamte Pegel um ca. 15dB reduziert. Dies führte dazu dass subjektiv der Eindruck entstanden ist die NR würde das Rauschen gut unterdrücken und kaum Artefakte erzeugen. Dies war aber objektiv nicht der Fall. Wenn man den Pegel durch Drehen am Lautstärkeregler wieder soweit erhöht hatte dass der Pegel so war wie ohne NR dann gab es fast keine Unterdrückung des Rauschens...
Klar nehmen die Artefakte beim Einschalten einer NR zu - das ist technisch bedingt. Wir haben mit der "finalen Version" auch Audiovergleiche mit den NRs anderer (kommerzieller) Geräte durchgeführt und die NR auch auf der SDR-Akademie vorgeführt. Es war (ist) kein großer Unterschied zwischen diesen NRs zu hören (gewesen).
Ich habe eben mal testweise die NR bei "1" beginnen lassen. Die Unterdrückung des Rauschens nahm dann stark ab (was zu erwarten war) bei einer minimalen Reduzierung der Artefakte. Prozentual gesehen STEIGT der Pegel der Artefakte bei einer Reduzierung an... Ein guter Wirkungsgrad ist für mich subjektiv bei der vorhandenen Default-Einstellung von "50" zu hören. Ich habe immer noch die grausige "alte" NR aus früheren Zeiten des mcHF im Ohr. Die hat das Rauschen praktisch gar nicht unterdrückt - nur Artefakte erzeugt. Da sind wir doch einen Riesenschritt weitergekommen :D.
Allerdings gebe ich zu, dass ich die NR genauso selten benutze wie den Magnify (also fast nie)...
Hier ist Frank gefragt - eine simple Reduzierung des Startwertes bringt NICHT was ihr euch vorstellt.
EDIT:
Hier (https://www.youtube.com/watch?v=CLC4VWjHdfw&feature=youtu.be&t=2256) könnt ihr Audiobeispiele mit unserer aktuellen NR aus dem Vortrag bei der SDR-Akademie hören
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: DD4WH on 12. March 2019, 14:14:56
@Alex: thanks for your comments!
@DF5LI: well, we do not need Bravos for simple requests ;-). If you have longstanding problems with any feature of the radio, just post them here on the forum and we will see what we can do . . .
OLD MCHF noise reduction:
* was based on the LMS algorithm
* did only work with strong signals, i.e. high SNR
* was not able to pull weak signals out of the noise
* this is the algorithm implemented in most commercial transceivers
NEW MCHF noise reduction
* is based on state-of-the-art algorithms in the frequency domain (spectral noise reduction)
* pulls weak signals out of the noise
* ONLY works for speech (SSB), because it is designed specifically for speech
* NEVER use for CW or anything else
* ONLY works when filter bandwidth is lower than 2.7kHz
* not everybody likes the sound of this noise reduction
I do not own any commercial HAM RADIO transceiver. So if you want to convince me to modify the software to sound like something, you would need to post a sound example (or send me your IC7300 ;-)).
Often we are asked to provide the same software / sound functionality in UHSDR (for free = 0,- €) as high-end ham radio transceivers (IC7300, KenwoodTS590 etc., 1300,- € to 1700€) :-). I take this as a compliment for our work. But it is difficult to program a software as a voluntary work (often called hobby :-)) and at the same time be as good or better than a commercial product worth more than 1000,- € ;-).
Unfortunately my hardware does not work at the moment, so I cannot really provide quick help with the particular problem of tweaking the NR.
But I will keep this in my mind when I resume work on the UHSDR again.
Alex, I would be very interested in your preference on the exact settings of the NR for your purpose! Could you post them here?
Best 73s,
Frank DD4WH |
Title: Re:NR slightly too strong
Post by: DF8OE on 12. March 2019, 17:15:27
I have listened the last few hours using DSP NR @setting 50. It works like a charm - nothing what harms. Completely satisfying.
Sorry - I cannot confirm anything is not working perfectly.
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: DF5LI on 12. March 2019, 19:20:59
Hi Frank! You wrote:
* ONLY works when filter bandwidth is lower than 2.7kHz
That is a setting I did not know yet. I will try it soon and I hope, I will be completely satisfied, too ;D
EDIT: I have tried it with 2,5 kHz bandwith and now it sounds much better!
Thank you, Frank ! |
Title: Re:NR slightly too strong
Post by: DF8OE on 12. March 2019, 19:38:12
The first what I try to get better readability of a weak or interferred signal is reducing reception bandwidth. If this does not satisfy, I switch on NR.... So I never tried NR using larger abndwidths - and now I tried. OK - using NR with higher bandwidths is without any sense 8)
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: SP9SOY on 12. March 2019, 20:35:51
Hallo !
Im Gegensatz zu DF5LI und seinem wertvollen Beitrag zur Diskussion denke ich, dass die Rauschunterdrückung des UHSDR sehr gut ist, und NR des IC-7300 ist eigentlich nicht so gut (verringert den Lautstärkepegel, ist aber nicht so effizient bei der Reduzierung von Rauschen).
Die einzige Verbesserung, die ich in UHSDR vorschlagen würde, ist, das Anfangs-NR-Pegel zu verringern. Ich denke, dass der tatsächliche NR-Pegel auf der Einstellung 5 (Minimum) zu hoch ist.
Wie ich bereits geschrieben habe, versuche ich, dies zu umgehen, indem ich die Menüeinstellung "NR asnr" auf ein Minimum reduziere. Es macht die NR-Aktion weicher. Aber diese Einstellung ist flüchtig, jedes Mal muss sie nach jedem Ausschalten neu eingestellt werden.
Danke für diese kleine Verbesserung, wenn es geht.
Übersetzt mit www.DeepL.com/Translator
Hello !
Contrary to DF5LI and his valuable input to the discussion, I think that the UHSDR's Noise Reduction is very good, and NR of IC-7300 is actually not that good (decreases volume level, but is not so efficient in reducing noise).
The only improvement that I would suggest in UHSDR is to decrease the starting NR level. I think thectual NR level on the setting 5 (minimum) is too high.
As I wrote before, I'm trying to bypass this, by decreasing the Menu Setting "NR asnr" to minimum. It makes NR action weaker. But this setting is volatile, each time it has to be set again after every switch off.
Thanks for this little improvement, if it can be done. |
Title: Re:NR slightly too strong
Post by: OE3HKC on 12. March 2019, 20:56:46
Hallo allerseits,
um es für mich einfacher machen zu schreiben, tue ich es in GERMAN-LANGUAGE...
sollte jemand interessiert daran sein, es in ENGLISH zu translaten, mögen er oder sie es bitte über Google berwerkstelligen...
ich habe zwar vor 50 Jahren in Englisch maturiert, kann mich aber leider in besagter Sprache bei weitem nicht so ausdrücken, wir mir der Schnabel gewachsen ist....
aber nun zum eigentlichen Thema:
die von Frank implementierte noise reduction ist eine der besten features, die die FW anzubieten hat !!
ich verwende bei SSB eine maximale BW von 2,7 kHz und bei dieser BW funktioniert die NR noch immer bestens...
gleich wie Andreas mit setting "50", egal wie stark die Signale sind, die Verbesserung der Sprachqualität ist mehr als überzeugend...
es ist aber nicht nur mein persönlicher Eindruck...
viele meiner befreundeten OM's aus meinem ADL haben mir bei Besuchen dies bestätigt...
einige davon verfügen auch über teure kommerzielle Amateurfunkgeräte, die ich mir gar nicht leisten könnte...
und diese OM's bestätigten mir, dass ihre "teuren" Geräte (ich möchte hier keine Markennamen erwähnen) auch nicht besser "klingen"...
ich persönlich besitze neben meinem MiniTRX noch einige kommerzielle Funkgeräte, einen DX 70 von Alinco, der rauscht im Vergleich wie ein herbstlicher Blätterwald, und mein früheres (vorm MiniTRX) Yaesu-Schlachtschiff FT890, der es leider mit der UHSDR-FW und meinem MiniTRX auch nicht aufnehmen kann... tzotz nachgerüsteter Quarzfilter...
am ehesten an die Audio Qualität meines SDR kommen noch mein alter SB100 von Heathkit und mein alter SX101A von Hallicrafters, das neuere "Klumpert" des Billig-Kommerz hat dagegen auch keine Chance...
war mir ein großes Anliegen, an dieser Stelle meine Hochachtung an die geniale Leistung UNSERER Softworker auszusprechen....
ich als reiner Hardwarer bewundere euch .....
vy 73,
Helmut aus dem Wienerwald |
Title: Re:NR slightly too strong
Post by: DF8OE on 13. March 2019, 07:14:18
Hi Alex,
I have introduced DSP NR minimum value to "1" and stepping to "1" if value is below "11". Does it the job for you?
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: SP9SOY on 13. March 2019, 16:38:21
Great job, dear Andreas! Now the NR is a bit less, as it starts at level "1" !
It would be also nice, if you could - in a free while - change "NR asur" (Debug Settings) to "5" or "10", because at current default setting "30" there is lot of Artifacts. I noticed that the default setting "30" is the maximum setting! A better compromise between noise reduction / artifacts generation is about "10", but maybe you have found another value..
Best wishes, Alex SP9SOY |
Title: Re:NR slightly too strong
Post by: DF8OE on 13. March 2019, 16:45:55
Hmmmmm..... Sorry - I cannot confirm "lot of artifacts". I am running 2.5KHz bandwidth, NR level set to "50" and all other values to default. If I reduce "NR asur" noise suppression is going worth. The values in debug menu are the ones which are tested out during a longer test period of some users... I do not want to reduce it - otherwise I always must restore them to "30" every time I want to use NR...
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: SP9SOY on 13. March 2019, 19:09:53
Dear Andreas,
yes you're right, there aren't "lot of artifacts", just some specific background audible at low signal levels. When "NR asur" is set to "30" the sound is "smoother" with some "underwater" touch and the noise is well cancelled, while lower "NRasur" gives a clearer and sharper sound, which is a better option for listening to weak signals.
But one can’t please everyone. I understand.
It's a pity that "NR asur" is not stored in Eeprom. In this way everyone could configure Noise Reduction to its preferences..
Bests, Alex |
Title: Re:NR slightly too strong
Post by: OE3HKC on 13. March 2019, 20:48:23
I think, some are hearing grass growing ;D
73, Helmut |
Title: Re:NR slightly too strong
Post by: DF8OE on 14. March 2019, 06:42:32
Configuration of UHSDR already is a mess to someone who joins mcHF or OVI40. There are tons of possibilities to "disconfigure" something. So we are very restrictive if we add something to menu or not. Of course every new configuration possibility enhances the power for someone who knows what he is doing. But at the other side it is complicating usage for not so powerful users. And every variable which is misconfigured during runtime and is on its default value at new start will not harm users and lead to a well running device.
At this state all will remain as it is. Later values of the debug menu will change their postition or vanish completely. We will see.
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: SP9SOY on 14. March 2019, 13:17:55
Hi Andreas,
OK, I understand. However, I think the "NR asnr" value should be possible to be set and saved, or - a better idea - removed from the Menu and automatically correlated with "NR level". In this way, if NR level is set low be the user, the sound will be even cleraer and sharper, and if NR level is set high, the sound would be smooth and damped as it is now. This is my idea, and believe me, I know very well and I own different NR reduction systems from major manufactueres, as well as from specialized companies, like BHI (UK).
Nonetheless I must emphasise here, that I consider the UHSDR noise reduction as a top performance implementation.!
Alex
|
Title: Re:NR slightly too strong
Post by: DF8OE on 14. March 2019, 15:02:35
Hi Alex,
as I already wrote: Settings in "Debug" are temporarily and will be removed at a given time. We will think about the "how".
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: DD4WH on 15. March 2019, 20:29:07
Hi Alex,
thanks for your suggestions on the NR level settings! You seem to be one of the very few people who have ever reported on their prefered settings of these DEBUG NR settings.
At the moment, the NR strength setting only sets one parameter:
NR alpha, which is 0.799 + NR strength / 1000, so it can go from 0.804 to 0.999
Other NR settings (in DEBUG menu) are:
NR beta 0.7 to 0.999 (default 0.96)
NR asnr 2 to 30 (default 30)
NR smooth width 1 to 5 (default 4)
NR smooth threshold 10 to 100 (default 40)
So, what I would like to find out is:
What are your favorite settings of these parameters for different situations (eg. high SNR, intermediate SNR, low SNR, extreme DX)?
And your suggestion of automatic setting of asnr with lower settings of NR strength is very interesting! What would you suggest? How should we set asnr when NR strength is from 1 to 10? from 11 to 50? from 51 to 100? >100 ?
Would be very nice of you could report your preferences here! I would like to integrate them into the NR strength adjustment.
My goal is to have one "NR strength" adjustment, that takes care of all other relevant internal parametres (asnr, beta etc.)
All the best 73s
Frank DD4WH
|
Title: Re:NR slightly too strong
Post by: SP9SOY on 16. March 2019, 12:32:28
Dear Frank,
Thank you for your interest in my findings.
I actually noticed that the operation of NR is (in my opinion) too strong at minimum NR setting (5 or even 1). The sound is a bit damped and has and "underwater" touch. Of course NR works really fine, but it makes an effect that one could expect at a higher NR setting.
However, if the "NR asnr" is lowered, the sound becomes more clear and sharp. This is what could better correspond to NR level "1".
I observed, that for strong and readable signals as well NR as NR_asnr could be easily set higher, because a strong NR action does not degrade a powerful audio. However, for weak stations, lower settings ought be used in order to protect readibility of vulnerable signals.
So I propose to combine the "NR asnr" parameter with NR setting. Perhaps a simple formula could be first tested : NR_asnr=a+NR*3/20, where NR=NR strength set by user and a=minimum N_asnr value (I don't know wheter it could be 0, 1 or 2?).
Thanks for developing such an advanced and clever software transceiver,
Bests,
Alex SP9SOY |
Title: Re:NR slightly too strong
Post by: SP9SOY on 16. March 2019, 14:41:34
an Addendum
It seems to me that Andreas DF8EO may not like low NR_asnr levels.
Therefore, I propose here another scheme, where reduced NR_asnr level is set only on the low end of the NR Strength scale:
NR Strength 1-5 -> NR_asnr =2 (or less if applicable)
NR Strength 6-15 -> NR_asnr =5
NR Strength 20-45 -> NR_asnr =10
NR Strength >50 -> NR_asnr =30
Bests, Alex SP9SOY |
Title: Re:NR slightly too strong
Post by: DF8OE on 16. March 2019, 17:17:03
Hi Alex,
I will check the result. Because of I never have had used NR with < 50 all settings are best "as they are". But the settings would not be changed if we use your approach. I will listen the bands using your suggestions.
Many thanks for your efforts!
vy 73
Andreas |
Title: Re:NR slightly too strong
Post by: SP9SOY on 20. March 2019, 17:09:41
Hello,
I did more tests last days, and I agree with Andreas that the best Noise Elimination's result is at the highest NR_asnr settings (30). The NR Strength was set high too. The resulting sound is smooth, pleasant and the noise background attenuated.
When the NR Strength is set low, however, the sound would be clearer if the NR_asnr parameter is set low too.
This is what I found.
All the bests to the developers and all readers,
Alex
|
Title: Re:NR slightly too strong
Post by: DF8OE on 20. March 2019, 17:21:58
I will do some observations too and I think we will reduce NR_asnr correspondent to NR strength if strength is lower than 15... (number may vary as a result of my observations).
But that is not my normal working condition. Normally first iI try to let my brain do noise reduction and if that becomes difficult I switch on NR - with setting 50. That helps very much.
vy 73
Andreas |
Diskussions- und Newsboard des DARC-Ortsverbandes I40 | Powered by YaBB SE
© 2001-2003, YaBB SE Dev Team. All Rights Reserved.
|