Understanding DSP NR
Understanding DSP NR
Is there a layman's explanation of the DSP noise reduction in Thetis? Specifically what delay, gain, leak, NPE method all mean and how changes to any of the values affect the noise reduction? The Thetis manual briefly touches some but not all of these.
Thanks
Ron
N4XD
Thanks
Ron
N4XD
Re: Understanding DSP NR
Ron,
There is some information available starting on page 50 of the WDSP Guide:
https://github.com/TAPR/OpenHPSDR-wdsp
Also, at 13 minutes into this presentation:
Finally, this presentation, while not discussion NR2, is nevertheless also interesting. It concentrates on later developments like PureSignal 2.0, etc.:
73,
Scott
Edited 20 May 2022 to add the two noise reduction papers Warren references in his brief as attachments to this post.
There is some information available starting on page 50 of the WDSP Guide:
https://github.com/TAPR/OpenHPSDR-wdsp
Also, at 13 minutes into this presentation:
Finally, this presentation, while not discussion NR2, is nevertheless also interesting. It concentrates on later developments like PureSignal 2.0, etc.:
73,
Scott
Edited 20 May 2022 to add the two noise reduction papers Warren references in his brief as attachments to this post.
- Attachments
-
- Ephraim_Speech_Enhancement_Minimum_ASSP85.pdf
- (165.63 KiB) Downloaded 352 times
-
- MMSE_Speech_Enhancement_Under_Speech_Presence_Unce.pdf
- (164.01 KiB) Downloaded 334 times
Re: Understanding DSP NR
These videos were fantastic! Thanks for sharing
I compare my FTDX101D to the Anan constantly and yes the Anan NR2 when its a big signal, sounds amazing, but for weaker signals the 101D really outshines it when it comes to NR (only). Are there any suggestions how to better "tune" the NR parameters? OR any suggestions how to improve the NR2 for weaker signals?
I have already enabled the attenuator and AGC methods with no real improvements to copy weaker signals in the noise.
Thanks!
Mike
WA2COP
I compare my FTDX101D to the Anan constantly and yes the Anan NR2 when its a big signal, sounds amazing, but for weaker signals the 101D really outshines it when it comes to NR (only). Are there any suggestions how to better "tune" the NR parameters? OR any suggestions how to improve the NR2 for weaker signals?
I have already enabled the attenuator and AGC methods with no real improvements to copy weaker signals in the noise.
Thanks!
Mike
WA2COP
w-u-2-o wrote:Ron,
There is some information available starting on page 50 of the WDSP Guide:
https://github.com/TAPR/OpenHPSDR-wdsp
Also, at 13 minutes into this presentation:
Finally, this presentation, while not discussion NR2, is nevertheless also interesting. It concentrates on later developments like PureSignal 2.0, etc.:
73,
Scott
Re: Understanding DSP NR
IMHO, NR2 and the AGC in Thetis are superior to any other receiver available, so I have little to offer you in terms of how to make it better. I can copy signals with Thetis that are barely above the noise floor, I've never been able to do that as well with any other radio from any other manufacturer.
Perhaps if you like what the FTDX101D is doing better on weak signals then I'd suggest switching Thetis to NR mode instead of NR2, since the LMS NR algorithm is pretty much what is used in nearly all other Japanese radios. Consider also switching to a narrower passband filter, and maybe trying a little bit of receive EQ to boost the higher passband frequency content.
Perhaps if you like what the FTDX101D is doing better on weak signals then I'd suggest switching Thetis to NR mode instead of NR2, since the LMS NR algorithm is pretty much what is used in nearly all other Japanese radios. Consider also switching to a narrower passband filter, and maybe trying a little bit of receive EQ to boost the higher passband frequency content.
Re: Understanding DSP NR
Scott, thanks for posting the videos
Quick question. With the AGC Gain set to auto (which works very, very well!) do you still "need" to set the AGC slope? If so, for weak DX, especially on 160, what would be your suggestion?
Thanks
Ron
N4XD
Quick question. With the AGC Gain set to auto (which works very, very well!) do you still "need" to set the AGC slope? If so, for weak DX, especially on 160, what would be your suggestion?
Thanks
Ron
N4XD
Re: Understanding DSP NR
I personally do not care for auto AGC gain. The original AGC algorithm normally takes into account the attenuator setting, but that feature is lost when auto AGC gain is used. This has the effect of losing any additional quieting associated with increased attenuator settings. I suppose many people will not notice this, but I do.
More to your point, AGC gain has nothing to do with AGC slope. For weak signal work, the default (no slope) works pretty well. I prefer to soften things a little bit with a 3dB slope for less demanding phone work, but that's a very subjective setting.
More to your point, AGC gain has nothing to do with AGC slope. For weak signal work, the default (no slope) works pretty well. I prefer to soften things a little bit with a 3dB slope for less demanding phone work, but that's a very subjective setting.
Re: Understanding DSP NR
I see your point.
I'm fortunate to have space for receive antennas I use on 160, 80 and 40. Their output is low enough so the attenuator is never needed when using them. I rarely listen on the transmit antenna on those bands. Which, I'm guessing, are the bands where you're using the attenuator.
Thanks
Ron
I'm fortunate to have space for receive antennas I use on 160, 80 and 40. Their output is low enough so the attenuator is never needed when using them. I rarely listen on the transmit antenna on those bands. Which, I'm guessing, are the bands where you're using the attenuator.
Thanks
Ron
Re: Understanding DSP NR
w-u-2-o wrote:Ron,
There is some information available starting on page 50 of the WDSP Guide:
https://github.com/TAPR/OpenHPSDR-wdsp
Also, at 13 minutes into this presentation:
Finally, this presentation, while not discussion NR2, is nevertheless also interesting. It concentrates on later developments like PureSignal 2.0, etc.:
73,
Scott
Thanks, Scott for sharing these videos. Thanks to Dr Warran NR0V for the in-depth description in a very easy to understand way.
73
Bhanu
VU2MB
Dell Alienware R10 AMD Ryzen 9 3950 PC-16 core @3.5 Ghz,
32GB Dual Channel HyperX(TM) F URY DDR4 XMP at 2666MHz
AMD Radeon RX 5700 XT 8GB GDDR 6
ANAN 8000DLE ANAN 200D ANAN 100D
Amplifier Mercury IIIS, Antenna Tuner: Palstar AT2K
32GB Dual Channel HyperX(TM) F URY DDR4 XMP at 2666MHz
AMD Radeon RX 5700 XT 8GB GDDR 6
ANAN 8000DLE ANAN 200D ANAN 100D
Amplifier Mercury IIIS, Antenna Tuner: Palstar AT2K
Re: Understanding DSP NR
Great sharing, but the CW and speech samples aren't audible at all
.
73 de NI2M
![Sad :(](./images/smilies/icon_e_sad.gif)
73 de NI2M
PC: 8 Core i7-10700 CPU @ 2.90GHz, NVMe SK Hynix 512 GB SSD, 32GB RAM
Windows 10 Home, Version 22H2
Thetis v2.10.4.3 x64
Protocol 2 v2.1.18
Windows 10 Home, Version 22H2
Thetis v2.10.4.3 x64
Protocol 2 v2.1.18
Re: Understanding DSP NR
A quick question about NR / NR2.
I noticed Scott mentioned in a NB topic that he operates with NR2 engaged all the time.
I’m not much of an FT8 operator, mostly just working DX that isn’t on SSB, CW or RTTY.
Does FT8 decode properly using NR2 or is that an exception?
Thanks,
George / W7GES
I noticed Scott mentioned in a NB topic that he operates with NR2 engaged all the time.
I’m not much of an FT8 operator, mostly just working DX that isn’t on SSB, CW or RTTY.
Does FT8 decode properly using NR2 or is that an exception?
Thanks,
George / W7GES
Re: Understanding DSP NR
Allow me to clarify: NR2 engaged 100% of the time for phone modes only. I never use noise reduction or noise blanking when running digital modes.
Re: Understanding DSP NR
I just want to say that I use NR2 ALL the time for FT8!
I find an 8 to 10db receive improvement on WSJT-x on this Awesome Radio!
I basically have Thetis in basic default mode since I don't know too much about the filtering numbers in Settings
Also my AGC slider is set for the line to be under the noise floor.
I keep doing NR2 to off/on and see really great improvement!
More -20 to-24db stations from other countries pop up a whole lot more with NR2 turned on!! (then with nothing or NR).
Problem is now that I got to run 10db more power for those stations to hear me!
My antenna is only up 17 feet off the ground! A 190 foot doublet shaped in a letter C.... Living in the North West Burbs of Chicago!
All my programs are running Virtually on the computer. No external audio connections.
With both boxes checked in Forced mode on Digital VAC1 mode.
Ok I checked..
On digital mode:
Buffer RX/TX 128
Filter RX/TX 4096
Linear Phase
Filter windows BH-7
(I don't understand these numbers... Like which is better, Larger number or smaller number..)
Don't matter to me as my CPU is around 8%
I find an 8 to 10db receive improvement on WSJT-x on this Awesome Radio!
I basically have Thetis in basic default mode since I don't know too much about the filtering numbers in Settings
Also my AGC slider is set for the line to be under the noise floor.
I keep doing NR2 to off/on and see really great improvement!
More -20 to-24db stations from other countries pop up a whole lot more with NR2 turned on!! (then with nothing or NR).
Problem is now that I got to run 10db more power for those stations to hear me!
My antenna is only up 17 feet off the ground! A 190 foot doublet shaped in a letter C.... Living in the North West Burbs of Chicago!
All my programs are running Virtually on the computer. No external audio connections.
With both boxes checked in Forced mode on Digital VAC1 mode.
Ok I checked..
On digital mode:
Buffer RX/TX 128
Filter RX/TX 4096
Linear Phase
Filter windows BH-7
(I don't understand these numbers... Like which is better, Larger number or smaller number..)
Don't matter to me as my CPU is around 8%
W7GES wrote:A quick question about NR / NR2.
I noticed Scott mentioned in a NB topic that he operates with NR2 engaged all the time.
I’m not much of an FT8 operator, mostly just working DX that isn’t on SSB, CW or RTTY.
Does FT8 decode properly using NR2 or is that an exception?
Thanks,
George / W7GES
Re: Understanding DSP NR
Well it won't hurt to try NR2 on and compare for myself... I had terrible Electrical RFI noise here until Dec 3, 2022.
I was able to get in touch with the RFI person from our electric company and he was able to fix an arcing wire on a transformer just NE of my house.
It's much easier to work people when you can hear them / print them...
As for the NW burbs of Chicago.... I lived in WI for most of my life and we would take trips to Chicago to visit relatives pretty regularly.
My last trip was May 2022 to visit cousins in Shorewood.
73,
George / W7GES
I was able to get in touch with the RFI person from our electric company and he was able to fix an arcing wire on a transformer just NE of my house.
It's much easier to work people when you can hear them / print them...
As for the NW burbs of Chicago.... I lived in WI for most of my life and we would take trips to Chicago to visit relatives pretty regularly.
My last trip was May 2022 to visit cousins in Shorewood.
73,
George / W7GES
Re: Understanding DSP NR
George,
Not to jack this thread but I use Diversity Receive to eliminate any single point source of electrical RFI regardless of whether the interference is wideband or birdie.
Here's a rather extreme example of wideband interference caused by a single point source, and how effective Diversity mode can be in eliminating the interference.
A storm came through Thursday night and the result the next morning was as shown below. RX2 shows severe wideband interference effectively rendering the entire band unusable for any amateur activity. The upper receiver shows the interference completely eliminated, at least at the frequency of interest (7.235).
Note that the reduction is not total when viewed across the entire band (the interference is still visible at both band edges), but is tunable for max effectiveness at any desired frequency.
Mark
Not to jack this thread but I use Diversity Receive to eliminate any single point source of electrical RFI regardless of whether the interference is wideband or birdie.
Here's a rather extreme example of wideband interference caused by a single point source, and how effective Diversity mode can be in eliminating the interference.
A storm came through Thursday night and the result the next morning was as shown below. RX2 shows severe wideband interference effectively rendering the entire band unusable for any amateur activity. The upper receiver shows the interference completely eliminated, at least at the frequency of interest (7.235).
Note that the reduction is not total when viewed across the entire band (the interference is still visible at both band edges), but is tunable for max effectiveness at any desired frequency.
Mark
Re: Understanding DSP NR
Just an FYI regarding WA6TMJ's post above... Sorry - no name given.. .
I don't know how you're seeing that large of an improvement using NR2 on FT8... I suspect you might have quite substantial local man-made noise or at least that is the best I can come up with. I've tried NR2 in the past and its never been an improvement for use on FT8 or digital modes (JT65/Q65). As a more definitive test I ran it yesterday - looking at FT8 decodes on both WSJT and JTDX (I run both simultaneously) and saw a pretty substantial difference w/o and w/ NR2 on with noticeably fewer decodes with it on. I mention this not to be a contrarian but to point out to others reading this that 'your mileage may vary'.
Seeing a "slight" increase in performance might make some sense but seeing the HUGE increase you're suggesting says there's other factors here - the best I can surmise is that it is substantial local noise.
As a side note re noise reduction, NB also mentioned in this thread: I find the NB to work extremely well on 6M DXing. I use it all the time. I use a value of 22 for the threshold on HF but can push it to 13 on 6M where, for example, going from a value of 16 to 13 often gains another db reduction in the noise floor and through all of my testing no degradation in the programs ability to decode the digital signal. I do have some local noise - worse at times - and turning NB on will often reduce the noise floor by a good 8-10 db. The downside of course is that it falls appear with strong signals. But often that isn't the case while dxing on FT8 on 6M (and especially when using Q65 on EME on 6M).
Gary
K9RX
I don't know how you're seeing that large of an improvement using NR2 on FT8... I suspect you might have quite substantial local man-made noise or at least that is the best I can come up with. I've tried NR2 in the past and its never been an improvement for use on FT8 or digital modes (JT65/Q65). As a more definitive test I ran it yesterday - looking at FT8 decodes on both WSJT and JTDX (I run both simultaneously) and saw a pretty substantial difference w/o and w/ NR2 on with noticeably fewer decodes with it on. I mention this not to be a contrarian but to point out to others reading this that 'your mileage may vary'.
Seeing a "slight" increase in performance might make some sense but seeing the HUGE increase you're suggesting says there's other factors here - the best I can surmise is that it is substantial local noise.
As a side note re noise reduction, NB also mentioned in this thread: I find the NB to work extremely well on 6M DXing. I use it all the time. I use a value of 22 for the threshold on HF but can push it to 13 on 6M where, for example, going from a value of 16 to 13 often gains another db reduction in the noise floor and through all of my testing no degradation in the programs ability to decode the digital signal. I do have some local noise - worse at times - and turning NB on will often reduce the noise floor by a good 8-10 db. The downside of course is that it falls appear with strong signals. But often that isn't the case while dxing on FT8 on 6M (and especially when using Q65 on EME on 6M).
Gary
K9RX
Re: Understanding DSP NR
Hi,
I just stumbled across this video about RM Noise.
https://www.youtube.com/watch?v=dweN2-pSmz0
(Thanks Kurt, ON7OFF)
73s Ernst
I just stumbled across this video about RM Noise.
https://www.youtube.com/watch?v=dweN2-pSmz0
(Thanks Kurt, ON7OFF)
73s Ernst
Re: Understanding DSP NR
Thanks for that link, Ernst.
Simon Brown has had a virtually identical AI NR option imbedded in the most recent Beta builds of his SDR Console.
Unfortunately, IMO both versions I've seen in operation (Simon's and the online version in Kurt's video) introduce some serious latency into the final audio.
With that said, I'm very much hoping to someday see this technology incorporated into Thetis.
Mark
Simon Brown has had a virtually identical AI NR option imbedded in the most recent Beta builds of his SDR Console.
Unfortunately, IMO both versions I've seen in operation (Simon's and the online version in Kurt's video) introduce some serious latency into the final audio.
With that said, I'm very much hoping to someday see this technology incorporated into Thetis.
Mark
Re: Understanding DSP NR
Another problem with RM Noise is that it requires an active internet connection because the AI analysis and training appears to be done on a remote server. That's a non-starter for me.
Coincidentally, I just tried an interesting VST plugin for my DAW, which can be found here: https://goyo.app/
With NR2 it is unnecessary, but for those who are running "regular" radios, Goyo represents an excellent, and currently free, option that IMHO is superior to anything else other than NR2.
Coincidentally, I just tried an interesting VST plugin for my DAW, which can be found here: https://goyo.app/
With NR2 it is unnecessary, but for those who are running "regular" radios, Goyo represents an excellent, and currently free, option that IMHO is superior to anything else other than NR2.
Re: Understanding DSP NR
oe3ide wrote:Hi,
I just stumbled across this video about RM Noise.
https://www.youtube.com/watch?v=dweN2-pSmz0
(Thanks Kurt, ON7OFF)
73s Ernst
Thanks Ernst!
I've tested this filter today and it is really amazing!, much, much better than NR2.
It will be perfect as a standalone app, but looks like the developer doesn't plan to release it
New call sign EA3CL
-
- Posts: 115
- Joined: Fri Apr 21, 2017 4:49 pm
Re: Understanding DSP NR
Hy Guys,
trie tghe lingua NF-Dsp, I have 2, one for the left and one for the right Channel.
I think its slightly better as the the Nr 2, becorse weak Station are comming up and are clear here to work
73 Erwin
trie tghe lingua NF-Dsp, I have 2, one for the left and one for the right Channel.
I think its slightly better as the the Nr 2, becorse weak Station are comming up and are clear here to work
73 Erwin
Anan 7000DLE MK2 black, P.2 v2.1.18, WIN 10, 10.0.18362 (1903), i7-7700 @3.60 Ghz, 2x Monitor 24"@144 Hz and 1x 32" Monitor @120 FPS for Thetis
Re: Understanding DSP NR
Is the Lingua made anymore?
Re: Understanding DSP NR
w-u-2-o wrote:Another problem with RM Noise is that it requires an active internet connection because the AI analysis and training appears to be done on a remote server. That's a non-starter for me.
Coincidentally, I just tried an interesting VST plugin for my DAW, which can be found here: https://goyo.app/
With NR2 it is unnecessary, but for those who are running "regular" radios, Goyo represents an excellent, and currently free, option that IMHO is superior to anything else other than NR2.
Scott, since this is a plug-in ( Goyo) , what kind of software is needed to use it? Is it something that will work with Voicemeeter Potato? Or something more involved. I would like to try it with my Flex as it might be easier to use instead of depending on off-site audio processing like RM Noise does. Noise Reduction is one of my major beefs with SmartSDR. ( plus the poor working Automatic Notch Filter)
James
WD5GWY
Re: Understanding DSP NR
Goyo is a VST plug-in. That is a digital audio industry standard. VST plug-ins have to be hosted by some sort of digital audio workstation software. Reaper, for example.
There is a relatively lightweight way of doing this with Voicemeeter. If you Google "voicemeeter potato vst plugins" you'll get links to a great many videos that show you exactly how to do this.
There is a relatively lightweight way of doing this with Voicemeeter. If you Google "voicemeeter potato vst plugins" you'll get links to a great many videos that show you exactly how to do this.
Re: Understanding DSP NR
w-u-2-o wrote:Goyo is a VST plug-in. That is a digital audio industry standard. VST plug-ins have to be hosted by some sort of digital audio workstation software. Reaper, for example.
There is a relatively lightweight way of doing this with Voicemeeter. If you Google "voicemeeter potato vst plugins" you'll get links to a great many videos that show you exactly how to do this.
Thank you Scott, that is what I needed to find out.
James
WD5GWY
Re: Understanding DSP NR
I followed the instructions on a couple of videos on setting up Voicemeeter Banana and the Goyo plug-in. I had to also install a host program for the plug-in. I tried every setting I could and still could not get it to pass the audio ( DAX audio in SmartSDR for the Flex radios) from my radio to VMB.
I gave up on it for the evening. I then tried RM Noise with my Flex 6600M. I was surprised at how well it worked compared to the noise reduction in SmartSDR. Delay was in the 100ms range. Once the filter started working, it removed all the background hiss from the audio.
The downside is,it uses fixed filter widths. 2.7Khz for SSB along with the processing, makes the audio harsh. It isn't as good as NR2 in Thetis. But, compared to the noise reduction in SmartSDR, it is light years better. If all the processing was done on the client instead of over the internet, and the filter bandwidth was adjustable, this application would be a great add on for radios like the Flex.
Not knowing the developer's background, it is quite impressive.
Tomorrow I will tackle Goyo again.
James
WD5GWY
I gave up on it for the evening. I then tried RM Noise with my Flex 6600M. I was surprised at how well it worked compared to the noise reduction in SmartSDR. Delay was in the 100ms range. Once the filter started working, it removed all the background hiss from the audio.
The downside is,it uses fixed filter widths. 2.7Khz for SSB along with the processing, makes the audio harsh. It isn't as good as NR2 in Thetis. But, compared to the noise reduction in SmartSDR, it is light years better. If all the processing was done on the client instead of over the internet, and the filter bandwidth was adjustable, this application would be a great add on for radios like the Flex.
Not knowing the developer's background, it is quite impressive.
Tomorrow I will tackle Goyo again.
James
WD5GWY
-
- Posts: 115
- Joined: Fri Apr 21, 2017 4:49 pm
Re: Understanding DSP NR
w-u-2-o wrote:Is the Lingua made anymore?
Scott,
you are right,
no more made now, sometimes Linguas are offered on E-Bay.
This Speakers/Moduls from BHI are also good:
https://www.wimo.com/de/zubehoer/geraet ... _brand=bhi
I m very happy to 2 Linguas!!!
73 Erwin
Anan 7000DLE MK2 black, P.2 v2.1.18, WIN 10, 10.0.18362 (1903), i7-7700 @3.60 Ghz, 2x Monitor 24"@144 Hz and 1x 32" Monitor @120 FPS for Thetis