38 replies to this topic

[Qu@rk]

Hello PLi-Team,

some users and me discovered a tiny problem in the A/V-Setting "Scaler Sharpness" since PLi 2.1 which has a huge effect on picture quality.
Since PLi 2.1 some users discovered a remarkable degradation in SD picture quality. You can correct this completely by setting "Scaler Sharpness" in the standard A/V-Settings to zero.

Please, could you check this:

When the A/V-Setting "Scaler Sharpness" is set to zero (slider completely on the left side), the picture quality improves a lot.

From what I was able to explore, the so called "Scaler Sharpness" in the standard AV-Settings does not write the "Scaler Sharpness" parameter in the chipset, but writes the "Sharpness" parameter, which is deactivated with a value of zero.

Maybe this affects also other boxes.

Please could you check this and correct it. Thanks a lot.

[pieterg]

do you have any idea how this could suddenly have occurred in 2.1 (and apparently not in 2.0)?
We are not aware of any related e2 changes, this should have been the same for a long time in 2.0 as well.

[Qu@rk]

Sorry, but I don't know the/your development process.

My tests and research showed the following behaviour and information:

The chipset parameter Scaler Sharpness will have no effect, if no scaling is done (SD to SD and HD to HD as used with AutoResolution plugin).
The paremeter in the AV-Settings does affect the picture quality regardless of input/output resolution, which is only the case with the "Sharpness" parameter of the chipset.
The chipset parameter "Sharpness" will be disabled if you set it to zero. The chipset parameter "Scaler Sharpness" has its neutral setting at 128.

If you set chipset "Sharpness" to 128 you will get no neutral picture processing, only with zero.

If you test the different settings with "Scaler Sharpness" in the A/V-Settings you will see, that the chipset parameter "Sharpnes" has been used.

In my humble opinion there could be two ways to solve this:
1. Rename "Scaler Sharpness" in the A/V-Settings to "Sharpness" and use the default value of zero (disabled).
2. Keep the name "Scaler Sharpness" in the A/V-Settings but write the correct chipset paremeter, with a default value of 128 (neutral).

Would this be possible?

Another important possiblity:
Maybe there was no change in the E2 sources but the new ET9000 drivers (shortly released after PLi 2.1) swapped the output devices of Sharpness and Scaler Sharpness. So you write to the correct device in PLi 2.1 but it reaches the wrong chipset parameter. Odd but possible and worth to check, if possible.

[pieterg]

ok, thanks for your report, we'll have a look

[pieterg]

The chipset parameter Scaler Sharpness will have no effect, if no scaling is done (SD to SD and HD to HD as used with AutoResolution plugin).
The paremeter in the AV-Settings does affect the picture quality regardless of input/output resolution, which is only the case with the "Sharpness" parameter of the chipset.
The chipset parameter "Sharpness" will be disabled if you set it to zero. The chipset parameter "Scaler Sharpness" has its neutral setting at 128.

Actually, the 'sharpness' has a 0..256 range (with a default of 0), whereas the 'scaler sharpness' has a 0..26 range (with a default of 13)

config.pep.sharpness = ConfigSlider(default=0, limits=(0,256))
config.av.scaler_sharpness = ConfigSlider(default=13, limits=(0,26))

Both 'sharpness' and 'scaler sharpness' are available in the videoenhancement plugin, but only 'scaler sharpness' is available in the AV settings.

If you set chipset "Sharpness" to 128 you will get no neutral picture processing, only with zero.

ok, but since the default is 0, that should be fine then.

If you test the different settings with "Scaler Sharpness" in the A/V-Settings you will see, that the chipset parameter "Sharpnes" has been used.

I've just checked, the AV settings 'scaler sharpness' does in fact control "/proc/stb/vmpeg/0/pep_scaler_sharpness".
So that should be correct I guess.
But could it be that this setting also works on the deinterlacer sharpness?
That might be the reason why you see effect on non-scaled content as well?
Or do you even see effect on 720p content being output at 720p?

[Erik Slagter]

On 18/08/11 15:30, pieterg wrote:

I've just checked, the AV settings 'scaler sharpness' does in fact control "/proc/stb/vmpeg/0/pep_scaler_sharpness".
So that should be correct I guess.
But could it be that this setting also works on the deinterlacer sharpness?
That might be the reason why you see effect on non-scaled content as well?
Or do you even see effect on 720p content being output at 720p?

I think I know this issue and has always been there.

A nice example is BBC HD, they broadcast at 1440x1080i, this is
different from the 1920x1080i the std is supposed to output, so it uses
the scaler, no problem with that. BUT it also uses the "scaler
sharpness" value, which I have quite aggressive for SD content, NOT for
HD content, that gets aggressively sharpened while it shouldn't be...

I suppose the same counts for 720p <-> 1080i etc.

[Qu@rk]

The chipset parameter Scaler Sharpness will have no effect, if no scaling is done (SD to SD and HD to HD as used with AutoResolution plugin).
The paremeter in the AV-Settings does affect the picture quality regardless of input/output resolution, which is only the case with the "Sharpness" parameter of the chipset.
The chipset parameter "Sharpness" will be disabled if you set it to zero. The chipset parameter "Scaler Sharpness" has its neutral setting at 128.

Actually, the 'sharpness' has a 0..256 range (with a default of 0), whereas the 'scaler sharpness' has a 0..26 range (with a default of 13)

config.pep.sharpness = ConfigSlider(default=0, limits=(0,256)) 
config.av.scaler_sharpness = ConfigSlider(default=13, limits=(0,26)) 

Both 'sharpness' and 'scaler sharpness' are available in the videoenhancement plugin, but only 'scaler sharpness' is available in the AV settings.

If you set chipset "Sharpness" to 128 you will get no neutral picture processing, only with zero.

ok, but since the default is 0, that should be fine then.

If you test the different settings with "Scaler Sharpness" in the A/V-Settings you will see, that the chipset parameter "Sharpnes" has been used.

I've just checked, the AV settings 'scaler sharpness' does in fact control "/proc/stb/vmpeg/0/pep_scaler_sharpness".
So that should be correct I guess.
But could it be that this setting also works on the deinterlacer sharpness?
That might be the reason why you see effect on non-scaled content as well?
Or do you even see effect on 720p content being output at 720p?

Thanks a lot for your investigation. So, from what you could see in the E2 source, everything is fine?

Sorry, of course, your are correct. The "Scaler Sharpness" has a range of 0..26 and a neutral position of 13 (like expected).

I have to check this using the VideoEnhancement plugin again on my ET9000. At present I do not use ths plugin (any more).
I was also looking into the sources of the VideoEnhancement plugin to see, which values are written to which chipset parameters.
Everything is fine here also.

I see definitely an effect at 720p content being output at 720p - that was the point, why I became skeptical and started my investigation.

With AV-Setting "Scaler Sharpness = default (13)" I get a much softer image than with zero (also 720p). So I became more skeptical.

Now I am puzzled. The AV-Setting "Scaler Sharpness=0" reacts like Sharpness=0.

Some time ago there was a problem with VU boxes, where Scaler Sharpness was swapped with Sharpness.

I'll have to check the VideoEnhancement plugin again and see, what happens exactly.

I am sorry for any additional effort I might have produced.
I'll have to check this even more deeply. Maybe I can see, that there is a driver problem or not.
Or maybe the chipset or the driver does not react as expected.

Please, to make testing easier:
Is there an easy way to write the chipset parameters e.g. /proc/stb/vmpeg/0/pep_scaler_sharpness clean and correctly using a shell command or script?
Testing would be much easier and to see what happens in the picture.
Is it possible to read from e.g. /proc/stb/vmpeg/0/pep_scaler_sharpness to get the current value or are these write-only register?

Thanks again.

[pieterg]

to get the current value:

cat /proc/stb/vmpeg/0/pep_scaler_sharpness

(in hex)

echo D > /proc/stb/vmpeg/0/pep_scaler_sharpness

to set the default 13

echo 0 > /proc/stb/vmpeg/0/pep_scaler_sharpness

to set it to zero

though reading back what we just wrote does not seem to work.


I'm not too sure about the pep_sharpness value:

myval = int(config.value*256)

so what we have to write seems 0..256*256
(also in hex, so 00000000..00010000)

[Qu@rk]

So, there is a significant difference in sharpness and PQ at 0 and 13 (0x0D).
Tested with 576i-576i (very strong effect, very good PQ with 0), 720p-720p (good to notice, better PQ with 0)

At 1080i the effect is reversed:
1080i-1080i (0 = a little bit softer, 1 = very soft, 13 = a little bit sharper than 0).

Strange behavior at 1080i.

In my opinion "0" at Scaler Sharpness produces the more neutral and natural picture than with 13 at all resolutions, esp. with 576i or 720p.

Maybe there is something wrong within the drivers or the chipset reacts not like expected.
We would need a driver developer to investigate this to make sure, that this behavior is no bug.

I can live with it, to set it to "0" manually. But maybe, it is a good reason to check the drivers, if someone thinks it could be necessary.
We have such a nice box, we should try to get the best out of it.

Thanks again for your support and help.

[hemertje]

please post your comments and investigation to ET...

[pieterg]

if 0 is better than 13 for other hardware as well (vu, dmm), I think we should just change our default.
However, if the interpretation of the scaler sharpness value is different between the various brands, we simply cannot pick a common default value I'm afraid.

[Qu@rk]

Oh, I didn't think about that. If there is a difference in behavior between the chipsets then we have a problem. Understood.

If you decide to set "0" as the default value for "Scaler Sharpness" maybe it would be very helpful to name it "off" in the AV-Settings instead of an empty slider bar.
Maybe it is not technically 100% correct but more user friendly. Also because 1 is very soft and 26 is very sharp the user could be puzzled about the behavior.

Perfect would be to have an on/off toggle.
If you switch it on, the default value of scaler sharpness is 13 and you see the slider bar.
If you switch if off, the value will be set to "0" and the slider bar is grayed out or disapears.

Would it be possible to show the value besides the slider bar? Or is this a skin problem?
Maybe also in the VideoEnhancement plugin? (if you can convince the author or how can we reach him?)

That would help a lot to reproduce settings or to write them down instead of count remote control clicks up or down.

Thanks again for taking care of this little detail.

[Qu@rk]

I have got the following information from a forum member in the CT forum.

The scaler in E2 is a software solution and E2 does not use the hardware directly.
This fact is well known (I did not know this).

So this would explain the difference between the chipset specs. and the current behavior.

He described the behavior of the scaler solution very good and it seems that he is a very competent user.

It is well known (for all E2 boxes) that setting "Scaler Sharpness" to "0" produces the best picture quality because the post processing of the software scaler is disabled.

Here the post. Sorry, German lang.:

Hubert6 wrote:

Es gibt einen Unterschied bei ursprünlichem interlaced Bildmaterial.
Bei Scaler Schärfe Mittelstellung bei diesem Ausgestrahlem Bildmaterial 50i Flag werden oft nur noch 288p Zeilen in 1080i dargestellt.
Es ist schon länger so, nicht erst seit PLI2.1

Bei in 50i gesendetem ursprünglichem Filmaterial von 24-25p sind aber keine Unterschiede erkennbar ob auf Null oder Mittelstellung steht, hier werden die 576p in die 1080i korrekt übertragen.
Interlaced Studiokameramaterial wird mit Scaler Mittelstellung als Grobkörniger empfunden, ist schlichtweg nur ein Programierfehler des Scalers.
Es kommt in der Mittelstellung bei Studiomaterial (nicht Spielfilmen) häufig zu starker Treppenbildung im Bild.

Die Scaler Schärfe ist auch keine echte Hardware Signalverarbeitung, sondern nur per Software obendraufgesetzt, ist daher immer nur eine Qualitätsverschlechterndes fast unützes Spielzeug.
Es gibt besser Möglichkeiten das alles per Hardware zu machen, aber diese nutzt Enigma2 ja nicht.

Daher einfach Scaler Schärfe auf Null stellen, war schon immer das beste Bild.

[littlesat]

We also coud make this default setting depended on the "hardware" platform. In Compenents/About.py we have now a good function for that.

[pieterg]

We also coud make this default setting depended on the "hardware" platform. In Compenents/About.py we have now a good function for that.

not if this is the same for all hardware

Note that all hardware is using the same chipset families, which I'm quite sure implement the same scaler sharpness setting.
So it would not surprise me if they all have the same behaviour of 0=off

[pieterg]

btw, the setting definitely controls the (hardware) scaling settings. There is no way to do such things from software.

[Qu@rk]

btw, there are ways to do this in software. I wrote a rudimental scaler engine some years ago in C and Assembler for a signal processor but it would need a lot of processor power, especially today with HD images. I was suprised, when I read this from Hubert6 but I know E2 not good enough to rate such information.

So I leave it to you and your team to decide how to work this out. Is it ok that way?

If you decide to implement it, maybe you could do it the way I suggested (see post with ON/OFF toggle etc.).
That would be very user friendly, in my opinion.

Thanks again.

[pieterg]

If you decide to implement it, maybe you could do it the way I suggested (see post with ON/OFF toggle etc.).
That would be very user friendly, in my opinion.

yes, that sounds like the best option.
Can you explain one more thing, am I right that:

0 = 0ff
1...26 is a sliding scale, where 13 is the most neutral position?
Or would you say 1 is the most neutral for example?

And have you compared this to other hardware, for example the vu and/or dmm boxes?
If the behavour is indeed like this, and the same for all broadcom chips, we should just make an on/off switch for the scaler sharpness.

[Qu@rk]

Yes you are right:

0 = OFF
1 ... 26 is a sliding scale
where 1 is super soft
and 26 is very sharp,
13 is mid position and neutral (as far as possible).

No, I am sorry, I only have a ET9000 box here.

During my research, I found at least one post about a VU user, who has recommended the same setting (0 at scaler sharpness) because of artifacts with other settings.

In DMM forums there were some discussions but no definite result regarding our topic. Some users recommend a setting of 15 or 16 to get a bit more crisp image but complained some artifacts. But, unfortunately, I found no reports of users who tested "0" in DMM fourms.

I can understand, this is hard to decide for you whether to make it a common change or not. I do not envy you for this decision.

[pieterg]

Yes you are right:

0 = OFF
1 ... 26 is a sliding scale
where 1 is super soft
and 26 is very sharp,
13 is mid position and neutral (as far as possible).

No, I am sorry, I only have a ET9000 box here.

During my research, I found at least one post about a VU user, who has recommended the same setting (0 at scaler sharpness) because of artifacts with other settings.

In DMM forums there were some discussions but no definite result regarding our topic. Some users recommend a setting of 15 or 16 to get a bit more crisp image but complained some artifacts. But, unfortunately, I found no reports of users who tested "0" in DMM fourms.

I can understand, this is hard to decide for you whether to make it a common change or not. I do not envy you for this decision.

yes, and it is hard to judge the exact effect a setting like this actually has.
You seem to have a very good eye for this.

Most users either see no difference, or they claim to see a big difference between two images (which are using the same settings)