Pro-L 2 by FabFilter

[rocolin]

Here are some screenshots.
This is the cosine @ 22050Hz. I made this in Matlab. I bounced it in Bitwig.

originalFs_2.png

This is Limiter6 GE with a random preset, Good Morning ME.

L6gmMe.png

This is Pro-L2, in transparent mode.

FFTransparent.png

I don't know how many pictures I can load per post. I think I have to make another post for more pics.



 

[rocolin]

This is Limitless in tight mode.

Limitless_tight.png

Barricade with compressor, limiter, and peak mode.

Barricade4.png

The tail ended abruptly , just like the head, so the limiters reacted to it as well. L6 and Pro-L2 look somewhat similar to the head, but Limiteless had a different tail. It kind of just faded slowly.

Limitless_tight_tail.png

[rocolin]

When I mixed sines of 56Hz and the 22050Hz:

This is where the higher frequency signal ends and the free 56Hz sine continues to run.
Limiter6 GE:

L6Composite.png

Fabfilter Pro-L2:

FFcomp.png

Limitless:

limitlessComp.png

[rocolin]

DC8C2, oversampled 8x, in limiter mode looks similar but a little weird.

DC8C2.png

I was just interested to see how limiters react, especifically their envelopes in different modes. I have basic knowledge of DSP, but not really an expert. I'm interested in the technical side of effects.

[plexuss]

What are we looking at here? How should these be interpreted?

[mutantdog]

For reference, this is what the above random L6GE preset looks like:

Some deep compression, soft clipping and multiband limiting there. I wouldn't expect it to be at all transparent.

[rocolin]

The first picture is the unprocessed signal, 22050 Hz at 0dBFS. When I passed the signal through a limiter with ISP, "true peak limiting"; then bounced it,that's how the limiters reacted. From that I assume that some limiters use oversampled side-chain while others oversample the actual signal. TDR and Fabfilter seemed to oversample the signal; while DMG, Toneboosters, and Izotope (not in the pics) might use oversampled sidechain to limit the signal.

I thought that it might be possible that some of the developers discarded the decimated samples outside the boundaries of the original signal. But when you look at the signal with mixed sinusoidals, you can see how the oversampling affects the signal somewhere middle. So therefore I conclude they are not oversampling the actual signal you hear.

I tested pro-L2 without oversampling and without True Peak limiting, and the results look just like other limiters which don't seem to OS the signals.

I only posted the pics because someone might find it interesting. I think it's interesting how different limiters use different envelopes when they process the signal.

Different setting will affect the signal in different ways ( attack, release, envelop shape, etc). So I wanted to see how the signal looks after fiddling with different parameters.

[rocolin]

You can make your own tests. I normally like to use an oscillator. I use MOscillator from Melda. Then I like using LFOTool to create some beat-like sidechaining, just to give it some shape. For this test I used just straight off/on (basically a reactangle) shape in LFOTool. I wanted the limiters to see "discontinuities" to see how they handled it. I also wanted to add a couple of other controlled signals in different frequency ranges (low, mid, hi). I wanted to see how they distort the signal, and how far you can push them.

I like using a very high-frequency sine with to see their envelops easily, if I zoom out.

I did the same with compressors a while ago, to see how they worked in side-chain mode. It helps me see how compressors have different curves (attack and release) for side chaining purposes and how they distort in such cases. I found LFOTool better than compressors, you have more control (you can see it) and it is also cleaner (fewer harmonics). Just keep sharp corners out of the envelop and you are good.

[xx JPRacer xx]

Update from FabFilter:

Apple's AAC RoundTrip as used for Mastered for iTunes etc), simply seems to use the example detection filtering (4x oversampling) as described in the ITU-R BS.1770-4 specs. This method is generally considered to be sub-optimal, and can give quite significant under- and overreads. As the spec already implies, "Higher sampling rates and over-sampling ratios are preferred".

So RoundTrip do an "ok" job but FF method is better.

[dangayle]

Update from FabFilter:

Apple's AAC RoundTrip as used for Mastered for iTunes etc), simply seems to use the example detection filtering (4x oversampling) as described in the ITU-R BS.1770-4 specs. This method is generally considered to be sub-optimal, and can give quite significant under- and overreads. As the spec already implies, "Higher sampling rates and over-sampling ratios are preferred".

So RoundTrip do an "ok" job but FF method is better.
[/quote]

So yeah, RoundTrip does follow the spec to the letter. But Limiter 2 follows the spec to the intent.

FF Limiter 2 shows overs in RoundTrip, but that’s due to RoundTrip using a metering system that can throw false positives and misses. Which means you shouldn’t take it’s readings as gospel.

Essentially, true peak limiting isn’t as binary as some would like it to be, it’s more of a tolerance range.

[10bd01]

So yeah, RoundTrip does follow the spec to the letter. But Limiter 2 follows the spec to the intent.

Well put. As a person who believes in following the spirit of the law rather than the letter of it, this perspective makes a lot of sense.

but that’s due to RoundTrip using a metering system that can throw false positives and misses.

This is why I had asked plexuss if he heard the overs, because if the measuring tool is throwing false positives and your biological measuring tool can't detect whether or not the measuring tool is accurate then the whole criticism becomes specious. Everything then becomes an argument from authority and then it's all about what authority who trusts.

[plexuss]

Update from FabFilter:

Apple's AAC RoundTrip as used for Mastered for iTunes etc), simply seems to use the example detection filtering (4x oversampling) as described in the ITU-R BS.1770-4 specs. This method is generally considered to be sub-optimal, and can give quite significant under- and overreads. As the spec already implies, "Higher sampling rates and over-sampling ratios are preferred".

So RoundTrip do an "ok" job but FF method is better.

So yeah, RoundTrip does follow the spec to the letter. But Limiter 2 follows the spec to the intent.

FF Limiter 2 shows overs in RoundTrip, but that’s due to RoundTrip using a metering system that can throw false positives and misses. Which means you shouldn’t take it’s readings as gospel.

Essentially, true peak limiting isn’t as binary as some would like it to be, it’s more of a tolerance range.

There is no evidence to suggest RoundTrip throws false positives. I have not seen anyone compare the results of RoundTrip to the actual analogue signal showing a correlation between the two. That would be the only way to prove/disprove. Just saying so doesn't make it real.

Use whatever tools you want of course. RoundTrip is my tool of choice because there is a good chance it adheres to the ITU spec but more importantly its the only tool that meters for clipping velocity which is the only thing that has a real impact to audio quality. If RoundTrip detects more potential for clipping then that is preferable to me because it means its doing a more conservative job which enables me to mitigate clipping better than if i used a tool that was more leanient.

Until someone shows a correlation between the detection of a given ISP meter and the clipping that occurs in the analogue domain, there is no proof that any tool is "better". And even if this research existed, clipping is dependent on the DAC so a variety of DACs would need to be tested.

If in fact, say, FF or TDL tools show ISP clipping more than RoundTrip all we can is "FF and TDR show clipping less than RoundTrip". Nothing more can be said without evidence.

They are all going to useful tools of course, for determining the risk of ISP clipping. Use what you have to the best of your abilities.

[bmanic]

@rocolin: Just a friendly tip that it's not all that trivial to try to get an insight into how a heavily program dependent processor works with a few rather simple combined test tones. It takes quite a bit more. So what you are seeing doesn't necessarily correlate all that well with actual musical program material.

Been testing both analogue and digital processes for a long time now and I've yet to come to a good conclusion of how to truly test and compare severely program dependent stuff accurately so that you get some kind of correlation to what you hear.

[bmanic]

This is why I had asked plexuss if he heard the overs, because if the measuring tool is throwing false positives and your biological measuring tool can't detect whether or not the measuring tool is accurate then the whole criticism becomes specious. Everything then becomes an argument from authority and then it's all about what authority who trusts.

[woodsdenis]

I upgraded today, couldn't be happier.